agora inbox for [email protected]help / color / mirror / Atom feed
[PATCH v1 2/2] WIP: Fix EEOP_JSON_CONSTRUCTOR and EEOP_JSONEXPR size. 90+ messages / 3 participants [nested] [flat]
* [PATCH v1 2/2] WIP: Fix EEOP_JSON_CONSTRUCTOR and EEOP_JSONEXPR size. @ 2022-06-17 01:33 Andres Freund <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Andres Freund @ 2022-06-17 01:33 UTC (permalink / raw) Author: Reviewed-By: Discussion: https://postgr.es/m/ Backpatch: --- src/include/executor/execExpr.h | 2 ++ 1 file changed, 2 insertions(+) diff --git a/src/include/executor/execExpr.h b/src/include/executor/execExpr.h index 2fba1450c65..5bfa7e9bdd4 100644 --- a/src/include/executor/execExpr.h +++ b/src/include/executor/execExpr.h @@ -22,6 +22,8 @@ struct ExprEvalStep; struct SubscriptingRefState; struct ScalarArrayOpExprHashTable; struct JsonbValue; +struct JsonExprState; +struct JsonConstructorExprState; /* Bits in ExprState->flags (see also execnodes.h for public flag bits): */ /* expression's interpreter has been initialized */ -- 2.35.1.677.gabf474a5dd --rftf6si7y32exfgc-- ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH 1/6] Add REPACK command @ 2026-01-27 10:48 Antonin Houska <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Antonin Houska @ 2026-01-27 10:48 UTC (permalink / raw) REPACK absorbs the functionality of VACUUM FULL and CLUSTER in a single command. Because this functionality is completely different from regular VACUUM, having it separate from VACUUM makes it easier for users to understand; as for CLUSTER, the term is heavily overloaded in the IT world and even in Postgres itself, so it's good that we can avoid it. This also adds pg_repackdb, a new utility that can invoke the new commands. This is heavily based on vacuumdb. Author: Antonin Houska <[email protected]> Co-authored-by: =C3=81lvaro Herrera <[email protected]> Reviewed-by: Mihail Nikalayeu <[email protected]> Reviewed-by: Robert Treat <[email protected]> Reviewed-by: Euler Taveira <[email protected]> Reviewed-by: Matheus Alcantara <[email protected]> Reviewed-by: Junwang Zhao <[email protected]> Reviewed-by: jian he <[email protected]> Discussion: https://postgr.es/m/82651.1720540558@antos Discussion: https://postgr.es/m/[email protected] --- doc/src/sgml/monitoring.sgml | 223 +++++- doc/src/sgml/ref/allfiles.sgml | 2 + doc/src/sgml/ref/cluster.sgml | 97 +-- doc/src/sgml/ref/clusterdb.sgml | 5 + doc/src/sgml/ref/pg_repackdb.sgml | 488 +++++++++++++ doc/src/sgml/ref/repack.sgml | 328 +++++++++ doc/src/sgml/ref/vacuum.sgml | 33 +- doc/src/sgml/reference.sgml | 2 + src/backend/access/heap/heapam_handler.c | 32 +- src/backend/catalog/index.c | 2 +- src/backend/catalog/system_views.sql | 29 +- src/backend/commands/cluster.c | 842 +++++++++++++++-------- src/backend/commands/vacuum.c | 6 +- src/backend/parser/gram.y | 86 ++- src/backend/tcop/utility.c | 23 +- src/backend/utils/adt/pgstatfuncs.c | 4 +- src/bin/psql/tab-complete.in.c | 42 +- src/bin/scripts/Makefile | 3 + src/bin/scripts/meson.build | 2 + src/bin/scripts/pg_repackdb.c | 240 +++++++ src/bin/scripts/t/103_repackdb.pl | 47 ++ src/bin/scripts/vacuuming.c | 102 ++- src/bin/scripts/vacuuming.h | 3 + src/include/commands/cluster.h | 8 +- src/include/commands/progress.h | 50 +- src/include/nodes/parsenodes.h | 35 +- src/include/parser/kwlist.h | 1 + src/include/tcop/cmdtaglist.h | 1 + src/include/utils/backend_progress.h | 2 +- src/test/regress/expected/cluster.out | 134 +++- src/test/regress/expected/rules.out | 72 +- src/test/regress/sql/cluster.sql | 70 +- src/tools/pgindent/typedefs.list | 2 + 33 files changed, 2480 insertions(+), 536 deletions(-) create mode 100644 doc/src/sgml/ref/pg_repackdb.sgml create mode 100644 doc/src/sgml/ref/repack.sgml create mode 100644 src/bin/scripts/pg_repackdb.c create mode 100644 src/bin/scripts/t/103_repackdb.pl diff --git a/doc/src/sgml/monitoring.sgml b/doc/src/sgml/monitoring.sgml index b77d189a500..71c92ed53ef 100644 --- a/doc/src/sgml/monitoring.sgml +++ b/doc/src/sgml/monitoring.sgml @@ -405,6 +405,14 @@ postgres 27093 0.0 0.0 30096 2752 ? Ss = 11:34 0:00 postgres: ser </entry> </row> =20 + <row> + <entry><structname>pg_stat_progress_repack</structname><indexterm><p= rimary>pg_stat_progress_repack</primary></indexterm></entry> + <entry>One row for each backend running + <command>REPACK</command>, showing current progress. See + <xref linkend=3D"repack-progress-reporting"/>. + </entry> + </row> + <row> <entry><structname>pg_stat_progress_basebackup</structname><indexter= m><primary>pg_stat_progress_basebackup</primary></indexterm></entry> <entry>One row for each WAL sender process streaming a base backup, @@ -5646,7 +5654,8 @@ FROM pg_stat_get_backend_idset() AS backendid; certain commands during command execution. Currently, the only commands which support progress reporting are <command>ANALYZE</command>, <command>CLUSTER</command>, - <command>CREATE INDEX</command>, <command>VACUUM</command>, + <command>CREATE INDEX</command>, <command>REPACK</command>, + <command>VACUUM</command>, <command>COPY</command>, and <xref linkend=3D"protocol-replication-base-backup"/> (i.e., replica= tion command that <xref linkend=3D"app-pgbasebackup"/> issues to take @@ -6130,6 +6139,218 @@ FROM pg_stat_get_backend_idset() AS backendid; </table> </sect2> =20 + <sect2 id=3D"repack-progress-reporting"> + <title>REPACK Progress Reporting</title> + + <indexterm> + <primary>pg_stat_progress_repack</primary> + </indexterm> + + <para> + Whenever <command>REPACK</command> is running, + the <structname>pg_stat_progress_repack</structname> view will contain a + row for each backend that is currently running the command. The tables + below describe the information that will be reported and provide + information about how to interpret it. + </para> + + <table id=3D"pg-stat-progress-repack-view" xreflabel=3D"pg_stat_progress= _repack"> + <title><structname>pg_stat_progress_repack</structname> View</title> + <tgroup cols=3D"1"> + <thead> + <row> + <entry role=3D"catalog_table_entry"><para role=3D"column_definition"> + Column Type + </para> + <para> + Description + </para></entry> + </row> + </thead> + + <tbody> + <row> + <entry role=3D"catalog_table_entry"><para role=3D"column_definition"> + <structfield>pid</structfield> <type>integer</type> + </para> + <para> + Process ID of backend. + </para></entry> + </row> + + <row> + <entry role=3D"catalog_table_entry"><para role=3D"column_definition"> + <structfield>datid</structfield> <type>oid</type> + </para> + <para> + OID of the database to which this backend is connected. + </para></entry> + </row> + + <row> + <entry role=3D"catalog_table_entry"><para role=3D"column_definition"> + <structfield>datname</structfield> <type>name</type> + </para> + <para> + Name of the database to which this backend is connected. + </para></entry> + </row> + + <row> + <entry role=3D"catalog_table_entry"><para role=3D"column_definition"> + <structfield>relid</structfield> <type>oid</type> + </para> + <para> + OID of the table being repacked. + </para></entry> + </row> + + <row> + <entry role=3D"catalog_table_entry"><para role=3D"column_definition"> + <structfield>phase</structfield> <type>text</type> + </para> + <para> + Current processing phase. See <xref linkend=3D"repack-phases"/>. + </para></entry> + </row> + + <row> + <entry role=3D"catalog_table_entry"><para role=3D"column_definition"> + <structfield>repack_index_relid</structfield> <type>oid</type> + </para> + <para> + If the table is being scanned using an index, this is the OID of the + index being used; otherwise, it is zero. + </para></entry> + </row> + + <row> + <entry role=3D"catalog_table_entry"><para role=3D"column_definition"> + <structfield>heap_tuples_scanned</structfield> <type>bigint</type> + </para> + <para> + Number of heap tuples scanned. + This counter only advances when the phase is + <literal>seq scanning heap</literal>, + <literal>index scanning heap</literal> + or <literal>writing new heap</literal>. + </para></entry> + </row> + + <row> + <entry role=3D"catalog_table_entry"><para role=3D"column_definition"> + <structfield>heap_tuples_written</structfield> <type>bigint</type> + </para> + <para> + Number of heap tuples written. + This counter only advances when the phase is + <literal>seq scanning heap</literal>, + <literal>index scanning heap</literal> + or <literal>writing new heap</literal>. + </para></entry> + </row> + + <row> + <entry role=3D"catalog_table_entry"><para role=3D"column_definition"> + <structfield>heap_blks_total</structfield> <type>bigint</type> + </para> + <para> + Total number of heap blocks in the table. This number is reported + as of the beginning of <literal>seq scanning heap</literal>. + </para></entry> + </row> + + <row> + <entry role=3D"catalog_table_entry"><para role=3D"column_definition"> + <structfield>heap_blks_scanned</structfield> <type>bigint</type> + </para> + <para> + Number of heap blocks scanned. This counter only advances when the + phase is <literal>seq scanning heap</literal>. + </para></entry> + </row> + + <row> + <entry role=3D"catalog_table_entry"><para role=3D"column_definition"> + <structfield>index_rebuild_count</structfield> <type>bigint</type> + </para> + <para> + Number of indexes rebuilt. This counter only advances when the pha= se + is <literal>rebuilding index</literal>. + </para></entry> + </row> + </tbody> + </tgroup> + </table> + + <table id=3D"repack-phases"> + <title>REPACK Phases</title> + <tgroup cols=3D"2"> + <colspec colname=3D"col1" colwidth=3D"1*"/> + <colspec colname=3D"col2" colwidth=3D"2*"/> + <thead> + <row> + <entry>Phase</entry> + <entry>Description</entry> + </row> + </thead> + + <tbody> + <row> + <entry><literal>initializing</literal></entry> + <entry> + The command is preparing to begin scanning the heap. This phase is + expected to be very brief. + </entry> + </row> + <row> + <entry><literal>seq scanning heap</literal></entry> + <entry> + The command is currently scanning the table using a sequential scan. + </entry> + </row> + <row> + <entry><literal>index scanning heap</literal></entry> + <entry> + <command>REPACK</command> is currently scanning the table using an = index scan. + </entry> + </row> + <row> + <entry><literal>sorting tuples</literal></entry> + <entry> + <command>REPACK</command> is currently sorting tuples. + </entry> + </row> + <row> + <entry><literal>writing new heap</literal></entry> + <entry> + <command>REPACK</command> is currently writing the new heap. + </entry> + </row> + <row> + <entry><literal>swapping relation files</literal></entry> + <entry> + The command is currently swapping newly-built files into place. + </entry> + </row> + <row> + <entry><literal>rebuilding index</literal></entry> + <entry> + The command is currently rebuilding an index. + </entry> + </row> + <row> + <entry><literal>performing final cleanup</literal></entry> + <entry> + The command is performing final cleanup. When this phase is + completed, <command>REPACK</command> will end. + </entry> + </row> + </tbody> + </tgroup> + </table> + </sect2> + <sect2 id=3D"copy-progress-reporting"> <title>COPY Progress Reporting</title> =20 diff --git a/doc/src/sgml/ref/allfiles.sgml b/doc/src/sgml/ref/allfiles.sgml index e167406c744..5df944d13ca 100644 --- a/doc/src/sgml/ref/allfiles.sgml +++ b/doc/src/sgml/ref/allfiles.sgml @@ -167,6 +167,7 @@ Complete list of usable sgml source files in this direc= tory. <!ENTITY refreshMaterializedView SYSTEM "refresh_materialized_view.sgml"> <!ENTITY reindex SYSTEM "reindex.sgml"> <!ENTITY releaseSavepoint SYSTEM "release_savepoint.sgml"> +<!ENTITY repack SYSTEM "repack.sgml"> <!ENTITY reset SYSTEM "reset.sgml"> <!ENTITY revoke SYSTEM "revoke.sgml"> <!ENTITY rollback SYSTEM "rollback.sgml"> @@ -213,6 +214,7 @@ Complete list of usable sgml source files in this direc= tory. <!ENTITY pgIsready SYSTEM "pg_isready.sgml"> <!ENTITY pgReceivewal SYSTEM "pg_receivewal.sgml"> <!ENTITY pgRecvlogical SYSTEM "pg_recvlogical.sgml"> +<!ENTITY pgRepackdb SYSTEM "pg_repackdb.sgml"> <!ENTITY pgResetwal SYSTEM "pg_resetwal.sgml"> <!ENTITY pgRestore SYSTEM "pg_restore.sgml"> <!ENTITY pgRewind SYSTEM "pg_rewind.sgml"> diff --git a/doc/src/sgml/ref/cluster.sgml b/doc/src/sgml/ref/cluster.sgml index 0b47460080b..2cda711bc9f 100644 --- a/doc/src/sgml/ref/cluster.sgml +++ b/doc/src/sgml/ref/cluster.sgml @@ -33,51 +33,13 @@ CLUSTER [ ( <replaceable class=3D"parameter">option</re= placeable> [, ...] ) ] [ <r <title>Description</title> =20 <para> - <command>CLUSTER</command> instructs <productname>PostgreSQL</productna= me> - to cluster the table specified - by <replaceable class=3D"parameter">table_name</replaceable> - based on the index specified by - <replaceable class=3D"parameter">index_name</replaceable>. The index mu= st - already have been defined on - <replaceable class=3D"parameter">table_name</replaceable>. + The <command>CLUSTER</command> command is equivalent to + <xref linkend=3D"sql-repack"/> with an <literal>USING INDEX</literal> + clause. See there for more details. </para> =20 - <para> - When a table is clustered, it is physically reordered - based on the index information. Clustering is a one-time operation: - when the table is subsequently updated, the changes are - not clustered. That is, no attempt is made to store new or - updated rows according to their index order. (If one wishes, one can - periodically recluster by issuing the command again. Also, setting - the table's <literal>fillfactor</literal> storage parameter to less than - 100% can aid in preserving cluster ordering during updates, since updat= ed - rows are kept on the same page if enough space is available there.) - </para> - - <para> - When a table is clustered, <productname>PostgreSQL</productname> - remembers which index it was clustered by. The form - <command>CLUSTER <replaceable class=3D"parameter">table_name</replaceab= le></command> - reclusters the table using the same index as before. You can also - use the <literal>CLUSTER</literal> or <literal>SET WITHOUT CLUSTER</lit= eral> - forms of <link linkend=3D"sql-altertable"><command>ALTER TABLE</command= ></link> to set the index to be used for - future cluster operations, or to clear any previous setting. - </para> - - <para> - <command>CLUSTER</command> without a - <replaceable class=3D"parameter">table_name</replaceable> reclusters al= l the - previously-clustered tables in the current database that the calling us= er - has privileges for. This form of <command>CLUSTER</command> cannot be - executed inside a transaction block. - </para> +<!-- Do we need to describe exactly which options map to what? They seem = obvious to me. --> =20 - <para> - When a table is being clustered, an <literal>ACCESS - EXCLUSIVE</literal> lock is acquired on it. This prevents any other - database operations (both reads and writes) from operating on the - table until the <command>CLUSTER</command> is finished. - </para> </refsect1> =20 <refsect1> @@ -136,63 +98,12 @@ CLUSTER [ ( <replaceable class=3D"parameter">option</r= eplaceable> [, ...] ) ] [ <r on the table. </para> =20 - <para> - In cases where you are accessing single rows randomly - within a table, the actual order of the data in the - table is unimportant. However, if you tend to access some - data more than others, and there is an index that groups - them together, you will benefit from using <command>CLUSTER</command>. - If you are requesting a range of indexed values from a table, or a - single indexed value that has multiple rows that match, - <command>CLUSTER</command> will help because once the index identifies= the - table page for the first row that matches, all other rows - that match are probably already on the same table page, - and so you save disk accesses and speed up the query. - </para> - - <para> - <command>CLUSTER</command> can re-sort the table using either an index= scan - on the specified index, or (if the index is a b-tree) a sequential - scan followed by sorting. It will attempt to choose the method that - will be faster, based on planner cost parameters and available statist= ical - information. - </para> - <para> While <command>CLUSTER</command> is running, the <xref linkend=3D"guc-search-path"/> is temporarily changed to <literal>pg_ca= talog, pg_temp</literal>. </para> =20 - <para> - When an index scan is used, a temporary copy of the table is created t= hat - contains the table data in the index order. Temporary copies of each - index on the table are created as well. Therefore, you need free spac= e on - disk at least equal to the sum of the table size and the index sizes. - </para> - - <para> - When a sequential scan and sort is used, a temporary sort file is - also created, so that the peak temporary space requirement is as much - as double the table size, plus the index sizes. This method is often - faster than the index scan method, but if the disk space requirement is - intolerable, you can disable this choice by temporarily setting <xref - linkend=3D"guc-enable-sort"/> to <literal>off</literal>. - </para> - - <para> - It is advisable to set <xref linkend=3D"guc-maintenance-work-mem"/> to - a reasonably large value (but not more than the amount of RAM you can - dedicate to the <command>CLUSTER</command> operation) before clusterin= g. - </para> - - <para> - Because the planner records statistics about the ordering of - tables, it is advisable to run <link linkend=3D"sql-analyze"><command>= ANALYZE</command></link> - on the newly clustered table. - Otherwise, the planner might make poor choices of query plans. - </para> - <para> Because <command>CLUSTER</command> remembers which indexes are cluster= ed, one can cluster the tables one wants clustered manually the first time, diff --git a/doc/src/sgml/ref/clusterdb.sgml b/doc/src/sgml/ref/clusterdb.s= gml index 0d2051bf6f1..b50c9581a98 100644 --- a/doc/src/sgml/ref/clusterdb.sgml +++ b/doc/src/sgml/ref/clusterdb.sgml @@ -64,6 +64,11 @@ PostgreSQL documentation this utility and via other methods for accessing the server. </para> =20 + <para> + <application>clusterdb</application> has been superseded by + <application>pg_repackdb</application>. + </para> + </refsect1> =20 =20 diff --git a/doc/src/sgml/ref/pg_repackdb.sgml b/doc/src/sgml/ref/pg_repack= db.sgml new file mode 100644 index 00000000000..a5ea513c6b1 --- /dev/null +++ b/doc/src/sgml/ref/pg_repackdb.sgml @@ -0,0 +1,488 @@ +<!-- +doc/src/sgml/ref/pg_repackdb.sgml +PostgreSQL documentation +--> + +<refentry id=3D"app-pgrepackdb"> + <indexterm zone=3D"app-pgrepackdb"> + <primary>pg_repackdb</primary> + </indexterm> + + <refmeta> + <refentrytitle><application>pg_repackdb</application></refentrytitle> + <manvolnum>1</manvolnum> + <refmiscinfo>Application</refmiscinfo> + </refmeta> + + <refnamediv> + <refname>pg_repackdb</refname> + <refpurpose>repack and analyze a <productname>PostgreSQL</productname> + database</refpurpose> + </refnamediv> + + <refsynopsisdiv> + <cmdsynopsis> + <command>pg_repackdb</command> + <arg rep=3D"repeat"><replaceable>connection-option</replaceable></arg> + <arg rep=3D"repeat"><replaceable>option</replaceable></arg> + + <arg choice=3D"plain" rep=3D"repeat"> + <arg choice=3D"opt"> + <group choice=3D"plain"> + <arg choice=3D"plain"><option>-t</option></arg> + <arg choice=3D"plain"><option>--table</option></arg> + </group> + <replaceable>table</replaceable> + <arg choice=3D"opt">( <replaceable class=3D"parameter">column</replac= eable> [,...] )</arg> + </arg> + </arg> + + <arg choice=3D"opt"> + <group choice=3D"plain"> + <arg choice=3D"plain"><replaceable>dbname</replaceable></arg> + <arg choice=3D"plain"><option>-a</option></arg> + <arg choice=3D"plain"><option>--all</option></arg> + </group> + </arg> + </cmdsynopsis> + + <cmdsynopsis> + <command>pg_repackdb</command> + <arg rep=3D"repeat"><replaceable>connection-option</replaceable></arg> + <arg rep=3D"repeat"><replaceable>option</replaceable></arg> + + <arg choice=3D"plain" rep=3D"repeat"> + <arg choice=3D"opt"> + <group choice=3D"plain"> + <arg choice=3D"plain"><option>-n</option></arg> + <arg choice=3D"plain"><option>--schema</option></arg> + </group> + <replaceable>schema</replaceable> + </arg> + </arg> + + <arg choice=3D"opt"> + <group choice=3D"plain"> + <arg choice=3D"plain"><replaceable>dbname</replaceable></arg> + <arg choice=3D"plain"><option>-a</option></arg> + <arg choice=3D"plain"><option>--all</option></arg> + </group> + </arg> + </cmdsynopsis> + + <cmdsynopsis> + <command>pg_repackdb</command> + <arg rep=3D"repeat"><replaceable>connection-option</replaceable></arg> + <arg rep=3D"repeat"><replaceable>option</replaceable></arg> + + <arg choice=3D"plain" rep=3D"repeat"> + <arg choice=3D"opt"> + <group choice=3D"plain"> + <arg choice=3D"plain"><option>-N</option></arg> + <arg choice=3D"plain"><option>--exclude-schema</option></arg> + </group> + <replaceable>schema</replaceable> + </arg> + </arg> + + <arg choice=3D"opt"> + <group choice=3D"plain"> + <arg choice=3D"plain"><replaceable>dbname</replaceable></arg> + <arg choice=3D"plain"><option>-a</option></arg> + <arg choice=3D"plain"><option>--all</option></arg> + </group> + </arg> + </cmdsynopsis> + </refsynopsisdiv> + + <refsect1> + <title>Description</title> + + <para> + <application>pg_repackdb</application> is a utility for repacking a + <productname>PostgreSQL</productname> database. + <application>pg_repackdb</application> will also generate internal + statistics used by the <productname>PostgreSQL</productname> query + optimizer. + </para> + + <para> + <application>pg_repackdb</application> is a wrapper around the SQL + command <link linkend=3D"sql-repack"><command>REPACK</command></link>. = There + is no effective difference between repacking and analyzing databases via + this utility and via other methods for accessing the server. + </para> + </refsect1> + + <refsect1> + <title>Options</title> + + <para> + <application>pg_repackdb</application> accepts the following command-l= ine arguments: + <variablelist> + <varlistentry> + <term><option>-a</option></term> + <term><option>--all</option></term> + <listitem> + <para> + Repack all databases. + </para> + </listitem> + </varlistentry> + + <varlistentry> + <term><option><optional>-d</optional> <replaceable class=3D"paramete= r">dbname</replaceable></option></term> + <term><option><optional>--dbname=3D</optional><replaceable class=3D"= parameter">dbname</replaceable></option></term> + <listitem> + <para> + Specifies the name of the database to be repacked or analyzed, + when <option>-a</option>/<option>--all</option> is not used. If t= his + is not specified, the database name is read from the environment + variable <envar>PGDATABASE</envar>. If that is not set, the user = name + specified for the connection is used. + The <replaceable>dbname</replaceable> can be + a <link linkend=3D"libpq-connstring">connection string</link>. If= so, + connection string parameters will override any conflicting command + line options. + </para> + </listitem> + </varlistentry> + + <varlistentry> + <term><option>-e</option></term> + <term><option>--echo</option></term> + <listitem> + <para> + Echo the commands that <application>pg_repackdb</application> + generates and sends to the server. + </para> + </listitem> + </varlistentry> + + <varlistentry> + <term><option>--index<optional>=3D<replaceable class=3D"parameter">i= ndex_name</replaceable></optional></option></term> + <listitem> + <para> + Pass the <literal>USING INDEX</literal> clause to <literal>REPACK<= /literal>, + and optionally the index name to specify. + </para> + </listitem> + </varlistentry> + + <varlistentry> + <term><option>-j <replaceable class=3D"parameter">njobs</replaceable= ></option></term> + <term><option>--jobs=3D<replaceable class=3D"parameter">njobs</repla= ceable></option></term> + <listitem> + <para> + Execute the repack or analyze commands in parallel by running + <replaceable class=3D"parameter">njobs</replaceable> + commands simultaneously. This option may reduce the processing ti= me + but it also increases the load on the database server. + </para> + <para> + <application>pg_repackdb</application> will open + <replaceable class=3D"parameter">njobs</replaceable> connections t= o the + database, so make sure your <xref linkend=3D"guc-max-connections"/> + setting is high enough to accommodate all connections. + </para> + <para> + Note that using this mode might cause deadlock failures if certain + system catalogs are processed in parallel. + </para> + </listitem> + </varlistentry> + + <varlistentry> + <term><option>-n <replaceable class=3D"parameter">schema</replaceabl= e></option></term> + <term><option>--schema=3D<replaceable class=3D"parameter">schema</re= placeable></option></term> + <listitem> + <para> + Repack or analyze all tables in + <replaceable class=3D"parameter">schema</replaceable> only. Multi= ple + schemas can be repacked by writing multiple <option>-n</option> + switches. + </para> + </listitem> + </varlistentry> + + <varlistentry> + <term><option>-N <replaceable class=3D"parameter">schema</replaceabl= e></option></term> + <term><option>--exclude-schema=3D<replaceable class=3D"parameter">sc= hema</replaceable></option></term> + <listitem> + <para> + Do not repack or analyze any tables in + <replaceable class=3D"parameter">schema</replaceable>. Multiple s= chemas + can be excluded by writing multiple <option>-N</option> switches. + </para> + </listitem> + </varlistentry> + + <varlistentry> + <term><option>-q</option></term> + <term><option>--quiet</option></term> + <listitem> + <para> + Do not display progress messages. + </para> + </listitem> + </varlistentry> + + <varlistentry> + <term><option>-t <replaceable class=3D"parameter">table</replaceable= > [ (<replaceable class=3D"parameter">column</replaceable> [,...]) ]</optio= n></term> + <term><option>--table=3D<replaceable class=3D"parameter">table</repl= aceable> [ (<replaceable class=3D"parameter">column</replaceable> [,...]) ]= </option></term> + <listitem> + <para> + Repack or analyze <replaceable class=3D"parameter">table</replacea= ble> + only. Column names can be specified only in conjunction with + the <option>--analyze</option> option. Multiple tables can be + repacked by writing multiple + <option>-t</option> switches. + </para> + <tip> + <para> + If you specify columns, you probably have to escape the parenthes= es + from the shell. (See examples below.) + </para> + </tip> + </listitem> + </varlistentry> + + <varlistentry> + <term><option>-v</option></term> + <term><option>--verbose</option></term> + <listitem> + <para> + Print detailed information during processing. + </para> + </listitem> + </varlistentry> + + <varlistentry> + <term><option>-V</option></term> + <term><option>--version</option></term> + <listitem> + <para> + Print the <application>pg_repackdb</application> version and exit. + </para> + </listitem> + </varlistentry> + + <varlistentry> + <term><option>-z</option></term> + <term><option>--analyze</option></term> + <listitem> + <para> + Also calculate statistics for use by the optimizer. If a column n= ame + list is given, only compute statistics for those columns. + </para> + </listitem> + </varlistentry> + + <varlistentry> + <term><option>-?</option></term> + <term><option>--help</option></term> + <listitem> + <para> + Show help about <application>pg_repackdb</application> command line + arguments, and exit. + </para> + </listitem> + </varlistentry> + + </variablelist> + </para> + + <para> + <application>pg_repackdb</application> also accepts + the following command-line arguments for connection parameters: + <variablelist> + <varlistentry> + <term><option>-h <replaceable class=3D"parameter">host</replaceable>= </option></term> + <term><option>--host=3D<replaceable class=3D"parameter">host</replac= eable></option></term> + <listitem> + <para> + Specifies the host name of the machine on which the server + is running. If the value begins with a slash, it is used + as the directory for the Unix domain socket. + </para> + </listitem> + </varlistentry> + + <varlistentry> + <term><option>-p <replaceable class=3D"parameter">port</replaceable>= </option></term> + <term><option>--port=3D<replaceable class=3D"parameter">port</replac= eable></option></term> + <listitem> + <para> + Specifies the TCP port or local Unix domain socket file + extension on which the server + is listening for connections. + </para> + </listitem> + </varlistentry> + + <varlistentry> + <term><option>-U <replaceable class=3D"parameter">username</replacea= ble></option></term> + <term><option>--username=3D<replaceable class=3D"parameter">username= </replaceable></option></term> + <listitem> + <para> + User name to connect as. + </para> + </listitem> + </varlistentry> + + <varlistentry> + <term><option>-w</option></term> + <term><option>--no-password</option></term> + <listitem> + <para> + Never issue a password prompt. If the server requires + password authentication and a password is not available by + other means such as a <filename>.pgpass</filename> file, the + connection attempt will fail. This option can be useful in + batch jobs and scripts where no user is present to enter a + password. + </para> + </listitem> + </varlistentry> + + <varlistentry> + <term><option>-W</option></term> + <term><option>--password</option></term> + <listitem> + <para> + Force <application>pg_repackdb</application> to prompt for a + password before connecting to a database. + </para> + + <para> + This option is never essential, since + <application>pg_repackdb</application> will automatically prompt + for a password if the server demands password authentication. + However, <application>pg_repackdb</application> will waste a + connection attempt finding out that the server wants a password. + In some cases it is worth typing <option>-W</option> to avoid the = extra + connection attempt. + </para> + </listitem> + </varlistentry> + + <varlistentry> + <term><option>--maintenance-db=3D<replaceable class=3D"parameter">db= name</replaceable></option></term> + <listitem> + <para> + When the <option>-a</option>/<option>--all</option> is used, conne= ct + to this database to gather the list of databases to repack. + If not specified, the <literal>postgres</literal> database will be= used, + or if that does not exist, <literal>template1</literal> will be us= ed. + This can be a <link linkend=3D"libpq-connstring">connection + string</link>. If so, connection string parameters will override = any + conflicting command line options. Also, connection string paramet= ers + other than the database name itself will be re-used when connecting + to other databases. + </para> + </listitem> + </varlistentry> + </variablelist> + </para> + </refsect1> + + + <refsect1> + <title>Environment</title> + + <variablelist> + <varlistentry> + <term><envar>PGDATABASE</envar></term> + <term><envar>PGHOST</envar></term> + <term><envar>PGPORT</envar></term> + <term><envar>PGUSER</envar></term> + + <listitem> + <para> + Default connection parameters + </para> + </listitem> + </varlistentry> + + <varlistentry> + <term><envar>PG_COLOR</envar></term> + <listitem> + <para> + Specifies whether to use color in diagnostic messages. Possible valu= es + are <literal>always</literal>, <literal>auto</literal> and + <literal>never</literal>. + </para> + </listitem> + </varlistentry> + </variablelist> + + <para> + This utility, like most other <productname>PostgreSQL</productname> uti= lities, + also uses the environment variables supported by <application>libpq</ap= plication> + (see <xref linkend=3D"libpq-envars"/>). + </para> + + </refsect1> + + + <refsect1> + <title>Diagnostics</title> + + <para> + In case of difficulty, see + <xref linkend=3D"sql-repack"/> and <xref linkend=3D"app-psql"/> for + discussions of potential problems and error messages. + The database server must be running at the + targeted host. Also, any default connection settings and environment + variables used by the <application>libpq</application> front-end + library will apply. + </para> + + </refsect1> + + <refsect1> + <title>Examples</title> + + <para> + To repack the database <literal>test</literal>: +<screen> +<prompt>$ </prompt><userinput>pg_repackdb test</userinput> +</screen> + </para> + + <para> + To repack and analyze for the optimizer a database named + <literal>bigdb</literal>: +<screen> +<prompt>$ </prompt><userinput>pg_repackdb --analyze bigdb</userinput> +</screen> + </para> + + <para> + To repack a single table + <literal>foo</literal> in a database named + <literal>xyzzy</literal>, and analyze a single column + <literal>bar</literal> of the table for the optimizer: +<screen> +<prompt>$ </prompt><userinput>pg_repackdb --analyze --verbose --table=3D'f= oo(bar)' xyzzy</userinput> +</screen></para> + + <para> + To repack all tables in the <literal>foo</literal> and <literal>bar</l= iteral> schemas + in a database named <literal>xyzzy</literal>: +<screen> +<prompt>$ </prompt><userinput>pg_repackdb --schema=3D'foo' --schema=3D'bar= ' xyzzy</userinput> +</screen></para> + + + </refsect1> + + <refsect1> + <title>See Also</title> + + <simplelist type=3D"inline"> + <member><xref linkend=3D"sql-repack"/></member> + </simplelist> + </refsect1> + +</refentry> diff --git a/doc/src/sgml/ref/repack.sgml b/doc/src/sgml/ref/repack.sgml new file mode 100644 index 00000000000..61d5c2cdef1 --- /dev/null +++ b/doc/src/sgml/ref/repack.sgml @@ -0,0 +1,328 @@ +<!-- +doc/src/sgml/ref/repack.sgml +PostgreSQL documentation +--> + +<refentry id=3D"sql-repack"> + <indexterm zone=3D"sql-repack"> + <primary>REPACK</primary> + </indexterm> + + <refmeta> + <refentrytitle>REPACK</refentrytitle> + <manvolnum>7</manvolnum> + <refmiscinfo>SQL - Language Statements</refmiscinfo> + </refmeta> + + <refnamediv> + <refname>REPACK</refname> + <refpurpose>rewrite a table to reclaim disk space</refpurpose> + </refnamediv> + + <refsynopsisdiv> +<synopsis> +REPACK [ ( <replaceable class=3D"parameter">option</replaceable> [, ...] )= ] [ <replaceable class=3D"parameter">table_and_columns</replaceable> [ USI= NG INDEX [ <replaceable class=3D"parameter">index_name</replaceable> ] ] ] +REPACK [ ( <replaceable class=3D"parameter">option</replaceable> [, ...] )= ] USING INDEX + +<phrase>where <replaceable class=3D"parameter">option</replaceable> can be= one of:</phrase> + + VERBOSE [ <replaceable class=3D"parameter">boolean</replaceable> ] + ANALYZE [ <replaceable class=3D"parameter">boolean</replaceable> ] + +<phrase>and <replaceable class=3D"parameter">table_and_columns</replaceabl= e> is:</phrase> + + <replaceable class=3D"parameter">table_name</replaceable> [ ( <replace= able class=3D"parameter">column_name</replaceable> [, ...] ) ] +</synopsis> + </refsynopsisdiv> + + <refsect1> + <title>Description</title> + + <para> + <command>REPACK</command> reclaims storage occupied by dead + tuples. Unlike <command>VACUUM</command>, it does so by rewriting the + entire contents of the table specified + by <replaceable class=3D"parameter">table_name</replaceable> into a new= disk + file with no extra space (except for the space guaranteed by + the <literal>fillfactor</literal> storage parameter), allowing unused s= pace + to be returned to the operating system. + </para> + + <para> + Without + a <replaceable class=3D"parameter">table_name</replaceable>, <command>R= EPACK</command> + processes every table and materialized view in the current database that + the current user has the <literal>MAINTAIN</literal> privilege on. This + form of <command>REPACK</command> cannot be executed inside a transacti= on + block. + </para> + + <para> + If a <literal>USING INDEX</literal> clause is specified, the rows are + physically reordered based on information from an index. Please see the + notes on clustering below. + </para> + + <para> + When a table is being repacked, an <literal>ACCESS EXCLUSIVE</literal> = lock + is acquired on it. This prevents any other database operations (both re= ads + and writes) from operating on the table until the <command>REPACK</comm= and> + is finished. + </para> + + <refsect2 id=3D"sql-repack-notes-on-clustering" xreflabel=3D"Notes on Cl= ustering"> + <title>Notes on Clustering</title> + + <para> + If the <literal>USING INDEX</literal> clause is specified, the rows in + the table are physically reordered following an index: if an index name + is specified in the command, then that index is used; if no index name + is specified, then the index that has been configured as the index to + cluster on. If no index has been configured in this way, an error is + thrown. The index given in the <literal>USING INDEX</literal> clause + is configured as the index to cluster on, as well as an index given + to the <command>CLUSTER</command> command. An index can be set + manually using <command>ALTER TABLE ... CLUSTER ON</command>, and reset + with <command>ALTER TABLE ... SET WITHOUT CLUSTER</command>. + </para> + + <para> + If no table name is specified in <command>REPACK USING INDEX</command>, + all tables which have a clustering index defined and which the calling + user has privileges for are processed. + </para> + + <para> + Clustering is a one-time operation: when the table is + subsequently updated, the changes are not clustered. That is, no atte= mpt + is made to store new or updated rows according to their index order. = (If + one wishes, one can periodically recluster by issuing the command agai= n. + Also, setting the table's <literal>fillfactor</literal> storage parame= ter + to less than 100% can aid in preserving cluster ordering during update= s, + since updated rows are kept on the same page if enough space is availa= ble + there.) + </para> + + <para> + In cases where you are accessing single rows randomly within a table, = the + actual order of the data in the table is unimportant. However, if you = tend + to access some data more than others, and there is an index that groups + them together, you will benefit from using clustering. If + you are requesting a range of indexed values from a table, or a single + indexed value that has multiple rows that match, + <command>REPACK</command> will help because once the index identifies = the + table page for the first row that matches, all other rows that match a= re + probably already on the same table page, and so you save disk accesses= and + speed up the query. + </para> + + <para> + <command>REPACK</command> can re-sort the table using either an index = scan + on the specified index (if the index is a b-tree), or a sequential scan + followed by sorting. It will attempt to choose the method that will be + faster, based on planner cost parameters and available statistical + information. + </para> + + <para> + Because the planner records statistics about the ordering of tables, i= t is + advisable to + run <link linkend=3D"sql-analyze"><command>ANALYZE</command></link> on= the + newly repacked table. Otherwise, the planner might make poor choices = of + query plans. + </para> + </refsect2> + + <refsect2 id=3D"sql-repack-notes-on-resources" xreflabel=3D"Notes on Res= ources"> + <title>Notes on Resources</title> + + <para> + When an index scan or a sequential scan without sort is used, a tempor= ary + copy of the table is created that contains the table data in the index + order. Temporary copies of each index on the table are created as wel= l. + Therefore, you need free space on disk at least equal to the sum of the + table size and the index sizes. + </para> + + <para> + When a sequential scan and sort is used, a temporary sort file is also + created, so that the peak temporary space requirement is as much as do= uble + the table size, plus the index sizes. This method is often faster than + the index scan method, but if the disk space requirement is intolerabl= e, + you can disable this choice by temporarily setting + <xref linkend=3D"guc-enable-sort"/> to <literal>off</literal>. + </para> + + <para> + It is advisable to set <xref linkend=3D"guc-maintenance-work-mem"/> to= a + reasonably large value (but not more than the amount of RAM you can + dedicate to the <command>REPACK</command> operation) before repacking. + </para> + </refsect2> + + </refsect1> + + <refsect1> + <title>Parameters</title> + + <variablelist> + <varlistentry> + <term><replaceable class=3D"parameter">table_name</replaceable></term> + <listitem> + <para> + The name (possibly schema-qualified) of a table. + </para> + </listitem> + </varlistentry> + + <varlistentry> + <term><replaceable class=3D"parameter">column_name</replaceable></term> + <listitem> + <para> + The name of a specific column to analyze. Defaults to all columns. + If a column list is specific, <literal>ANALYZE</literal> must also + be specified. + </para> + </listitem> + </varlistentry> + + <varlistentry> + <term><replaceable class=3D"parameter">index_name</replaceable></term> + <listitem> + <para> + The name of an index. + </para> + </listitem> + </varlistentry> + + <varlistentry> + <term><literal>VERBOSE</literal></term> + <listitem> + <para> + Prints a progress report as each table is repacked + at <literal>INFO</literal> level. + </para> + </listitem> + </varlistentry> + + <varlistentry> + <term><literal>ANALYZE</literal></term> + <term><literal>ANALYSE</literal></term> + <listitem> + <para> + Applies <xref linkend=3D"sql-analyze"/> on the table after repacking= . This is + currently only supported when a single (non-partitioned) table is sp= ecified. + </para> + </listitem> + </varlistentry> + + <varlistentry> + <term><replaceable class=3D"parameter">boolean</replaceable></term> + <listitem> + <para> + Specifies whether the selected option should be turned on or off. + You can write <literal>TRUE</literal>, <literal>ON</literal>, or + <literal>1</literal> to enable the option, and <literal>FALSE</liter= al>, + <literal>OFF</literal>, or <literal>0</literal> to disable it. The + <replaceable class=3D"parameter">boolean</replaceable> value can also + be omitted, in which case <literal>TRUE</literal> is assumed. + </para> + </listitem> + </varlistentry> + </variablelist> + </refsect1> + + <refsect1> + <title>Notes</title> + + <para> + To repack a table, one must have the <literal>MAINTAIN</literal> privi= lege + on the table. + </para> + + <para> + While <command>REPACK</command> is running, the <xref + linkend=3D"guc-search-path"/> is temporarily changed to <literal>pg_ca= talog, + pg_temp</literal>. + </para> + + <para> + Each backend running <command>REPACK</command> will report its progress + in the <structname>pg_stat_progress_repack</structname> view. See + <xref linkend=3D"repack-progress-reporting"/> for details. + </para> + + <para> + Repacking a partitioned table repacks each of its partitions. If an in= dex + is specified, each partition is repacked using the partition of that + index. <command>REPACK</command> on a partitioned table cannot be exec= uted + inside a transaction block. + </para> + + </refsect1> + + <refsect1> + <title>Examples</title> + + <para> + Repack the table <literal>employees</literal>: +<programlisting> +REPACK employees; +</programlisting> + </para> + + <para> + Repack the table <literal>employees</literal> on the basis of its + index <literal>employees_ind</literal> (Since index is used here, this = is + effectively clustering): +<programlisting> +REPACK employees USING INDEX employees_ind; +</programlisting> + </para> + + <para> + Repack the table <literal>cases</literal> on physical ordering, + running an <command>ANALYZE</command> on the given columns once + repacking is done, showing informational messages: +<programlisting> +REPACK (ANALYZE, VERBOSE) cases (district, case_nr); +</programlisting> + </para> + + <para> + Repack all tables in the database on which you have + the <literal>MAINTAIN</literal> privilege: +<programlisting> +REPACK; +</programlisting> + </para> + + <para> + Repack all tables for which a clustering index has previously been + configured on which you have the <literal>MAINTAIN</literal> privilege, + showing informational messages: +<programlisting> +REPACK (VERBOSE) USING INDEX; +</programlisting> + </para> + + </refsect1> + + <refsect1> + <title>Compatibility</title> + + <para> + There is no <command>REPACK</command> statement in the SQL standard. + </para> + </refsect1> + + <refsect1> + <title>See Also</title> + + <simplelist type=3D"inline"> + <member><xref linkend=3D"app-pgrepackdb"/></member> + <member><xref linkend=3D"repack-progress-reporting"/></member> + </simplelist> + </refsect1> + +</refentry> diff --git a/doc/src/sgml/ref/vacuum.sgml b/doc/src/sgml/ref/vacuum.sgml index 6d0fdd43cfb..ac5d083d468 100644 --- a/doc/src/sgml/ref/vacuum.sgml +++ b/doc/src/sgml/ref/vacuum.sgml @@ -25,7 +25,6 @@ VACUUM [ ( <replaceable class=3D"parameter">option</repla= ceable> [, ...] ) ] [ <re =20 <phrase>where <replaceable class=3D"parameter">option</replaceable> can be= one of:</phrase> =20 - FULL [ <replaceable class=3D"parameter">boolean</replaceable> ] FREEZE [ <replaceable class=3D"parameter">boolean</replaceable> ] VERBOSE [ <replaceable class=3D"parameter">boolean</replaceable> ] ANALYZE [ <replaceable class=3D"parameter">boolean</replaceable> ] @@ -39,6 +38,7 @@ VACUUM [ ( <replaceable class=3D"parameter">option</repla= ceable> [, ...] ) ] [ <re SKIP_DATABASE_STATS [ <replaceable class=3D"parameter">boolean</replac= eable> ] ONLY_DATABASE_STATS [ <replaceable class=3D"parameter">boolean</replac= eable> ] BUFFER_USAGE_LIMIT <replaceable class=3D"parameter">size</replaceable> + FULL [ <replaceable class=3D"parameter">boolean</replaceable> ] =20 <phrase>and <replaceable class=3D"parameter">table_and_columns</replaceabl= e> is:</phrase> =20 @@ -95,20 +95,6 @@ VACUUM [ ( <replaceable class=3D"parameter">option</repl= aceable> [, ...] ) ] [ <re <title>Parameters</title> =20 <variablelist> - <varlistentry> - <term><literal>FULL</literal></term> - <listitem> - <para> - Selects <quote>full</quote> vacuum, which can reclaim more - space, but takes much longer and exclusively locks the table. - This method also requires extra disk space, since it writes a - new copy of the table and doesn't release the old copy until - the operation is complete. Usually this should only be used when a - significant amount of space needs to be reclaimed from within the ta= ble. - </para> - </listitem> - </varlistentry> - <varlistentry> <term><literal>FREEZE</literal></term> <listitem> @@ -362,6 +348,23 @@ VACUUM [ ( <replaceable class=3D"parameter">option</re= placeable> [, ...] ) ] [ <re </listitem> </varlistentry> =20 + <varlistentry> + <term><literal>FULL</literal></term> + <listitem> + <para> + This option, which is deprecated, makes <command>VACUUM</command> + behave like <command>REPACK</command> without a + <literal>USING INDEX</literal> clause. + This method of compacting the table takes much longer than + <command>VACUUM</command> and exclusively locks the table. + This method also requires extra disk space, since it writes a + new copy of the table and doesn't release the old copy until + the operation is complete. Usually this should only be used when a + significant amount of space needs to be reclaimed from within the ta= ble. + </para> + </listitem> + </varlistentry> + <varlistentry> <term><replaceable class=3D"parameter">boolean</replaceable></term> <listitem> diff --git a/doc/src/sgml/reference.sgml b/doc/src/sgml/reference.sgml index 2cf02c37b17..5d9a8a25a02 100644 --- a/doc/src/sgml/reference.sgml +++ b/doc/src/sgml/reference.sgml @@ -195,6 +195,7 @@ &refreshMaterializedView; &reindex; &releaseSavepoint; + &repack; &reset; &revoke; &rollback; @@ -258,6 +259,7 @@ &pgIsready; &pgReceivewal; &pgRecvlogical; + &pgRepackdb; &pgRestore; &pgVerifyBackup; &psqlRef; diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/= heap/heapam_handler.c index cbef73e5d4b..7d4b48e5a97 100644 --- a/src/backend/access/heap/heapam_handler.c +++ b/src/backend/access/heap/heapam_handler.c @@ -741,13 +741,13 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Re= lation NewHeap, if (OldIndex !=3D NULL && !use_sort) { const int ci_index[] =3D { - PROGRESS_CLUSTER_PHASE, - PROGRESS_CLUSTER_INDEX_RELID + PROGRESS_REPACK_PHASE, + PROGRESS_REPACK_INDEX_RELID }; int64 ci_val[2]; =20 /* Set phase and OIDOldIndex to columns */ - ci_val[0] =3D PROGRESS_CLUSTER_PHASE_INDEX_SCAN_HEAP; + ci_val[0] =3D PROGRESS_REPACK_PHASE_INDEX_SCAN_HEAP; ci_val[1] =3D RelationGetRelid(OldIndex); pgstat_progress_update_multi_param(2, ci_index, ci_val); =20 @@ -759,15 +759,15 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Re= lation NewHeap, else { /* In scan-and-sort mode and also VACUUM FULL, set phase */ - pgstat_progress_update_param(PROGRESS_CLUSTER_PHASE, - PROGRESS_CLUSTER_PHASE_SEQ_SCAN_HEAP); + pgstat_progress_update_param(PROGRESS_REPACK_PHASE, + PROGRESS_REPACK_PHASE_SEQ_SCAN_HEAP); =20 tableScan =3D table_beginscan(OldHeap, SnapshotAny, 0, (ScanKey) NULL); heapScan =3D (HeapScanDesc) tableScan; indexScan =3D NULL; =20 /* Set total heap blocks */ - pgstat_progress_update_param(PROGRESS_CLUSTER_TOTAL_HEAP_BLKS, + pgstat_progress_update_param(PROGRESS_REPACK_TOTAL_HEAP_BLKS, heapScan->rs_nblocks); } =20 @@ -809,7 +809,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Rela= tion NewHeap, * is manually updated to the correct value when the table * scan finishes. */ - pgstat_progress_update_param(PROGRESS_CLUSTER_HEAP_BLKS_SCANNED, + pgstat_progress_update_param(PROGRESS_REPACK_HEAP_BLKS_SCANNED, heapScan->rs_nblocks); break; } @@ -825,7 +825,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Rela= tion NewHeap, */ if (prev_cblock !=3D heapScan->rs_cblock) { - pgstat_progress_update_param(PROGRESS_CLUSTER_HEAP_BLKS_SCANNED, + pgstat_progress_update_param(PROGRESS_REPACK_HEAP_BLKS_SCANNED, (heapScan->rs_cblock + heapScan->rs_nblocks - heapScan->rs_startblock @@ -926,14 +926,14 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Re= lation NewHeap, * In scan-and-sort mode, report increase in number of tuples * scanned */ - pgstat_progress_update_param(PROGRESS_CLUSTER_HEAP_TUPLES_SCANNED, + pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_SCANNED, *num_tuples); } else { const int ct_index[] =3D { - PROGRESS_CLUSTER_HEAP_TUPLES_SCANNED, - PROGRESS_CLUSTER_HEAP_TUPLES_WRITTEN + PROGRESS_REPACK_HEAP_TUPLES_SCANNED, + PROGRESS_REPACK_HEAP_TUPLES_WRITTEN }; int64 ct_val[2]; =20 @@ -966,14 +966,14 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Re= lation NewHeap, double n_tuples =3D 0; =20 /* Report that we are now sorting tuples */ - pgstat_progress_update_param(PROGRESS_CLUSTER_PHASE, - PROGRESS_CLUSTER_PHASE_SORT_TUPLES); + pgstat_progress_update_param(PROGRESS_REPACK_PHASE, + PROGRESS_REPACK_PHASE_SORT_TUPLES); =20 tuplesort_performsort(tuplesort); =20 /* Report that we are now writing new heap */ - pgstat_progress_update_param(PROGRESS_CLUSTER_PHASE, - PROGRESS_CLUSTER_PHASE_WRITE_NEW_HEAP); + pgstat_progress_update_param(PROGRESS_REPACK_PHASE, + PROGRESS_REPACK_PHASE_WRITE_NEW_HEAP); =20 for (;;) { @@ -991,7 +991,7 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Rela= tion NewHeap, values, isnull, rwstate); /* Report n_tuples */ - pgstat_progress_update_param(PROGRESS_CLUSTER_HEAP_TUPLES_WRITTEN, + pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_WRITTEN, n_tuples); } =20 diff --git a/src/backend/catalog/index.c b/src/backend/catalog/index.c index 43de42ce39e..5ee6389d39c 100644 --- a/src/backend/catalog/index.c +++ b/src/backend/catalog/index.c @@ -4077,7 +4077,7 @@ reindex_relation(const ReindexStmt *stmt, Oid relid, = int flags, Assert(!ReindexIsProcessingIndex(indexOid)); =20 /* Set index rebuild count */ - pgstat_progress_update_param(PROGRESS_CLUSTER_INDEX_REBUILD_COUNT, + pgstat_progress_update_param(PROGRESS_REPACK_INDEX_REBUILD_COUNT, i); i++; } diff --git a/src/backend/catalog/system_views.sql b/src/backend/catalog/sys= tem_views.sql index 7553f31fef0..3f05ba3083a 100644 --- a/src/backend/catalog/system_views.sql +++ b/src/backend/catalog/system_views.sql @@ -1283,14 +1283,15 @@ CREATE VIEW pg_stat_progress_vacuum AS FROM pg_stat_get_progress_info('VACUUM') AS S LEFT JOIN pg_database D ON S.datid =3D D.oid; =20 -CREATE VIEW pg_stat_progress_cluster AS +CREATE VIEW pg_stat_progress_repack AS SELECT S.pid AS pid, S.datid AS datid, D.datname AS datname, S.relid AS relid, CASE S.param1 WHEN 1 THEN 'CLUSTER' - WHEN 2 THEN 'VACUUM FULL' + WHEN 2 THEN 'REPACK' + WHEN 3 THEN 'VACUUM FULL' END AS command, CASE S.param2 WHEN 0 THEN 'initializing' WHEN 1 THEN 'seq scanning heap' @@ -1301,15 +1302,35 @@ CREATE VIEW pg_stat_progress_cluster AS WHEN 6 THEN 'rebuilding index' WHEN 7 THEN 'performing final cleanup' END AS phase, - CAST(S.param3 AS oid) AS cluster_index_relid, + CAST(S.param3 AS oid) AS repack_index_relid, S.param4 AS heap_tuples_scanned, S.param5 AS heap_tuples_written, S.param6 AS heap_blks_total, S.param7 AS heap_blks_scanned, S.param8 AS index_rebuild_count - FROM pg_stat_get_progress_info('CLUSTER') AS S + FROM pg_stat_get_progress_info('REPACK') AS S LEFT JOIN pg_database D ON S.datid =3D D.oid; =20 +-- This view is as the one above, except for renaming a column and avoiding +-- 'REPACK' as a command name to report. +CREATE VIEW pg_stat_progress_cluster AS + SELECT + pid, + datid, + datname, + relid, + CASE WHEN command IN ('CLUSTER', 'VACUUM FULL') THEN command + WHEN repack_index_relid =3D 0 THEN 'VACUUM FULL' + ELSE 'CLUSTER' END AS command, + phase, + repack_index_relid AS cluster_index_relid, + heap_tuples_scanned, + heap_tuples_written, + heap_blks_total, + heap_blks_scanned, + index_rebuild_count + FROM pg_stat_progress_repack; + CREATE VIEW pg_stat_progress_create_index AS SELECT S.pid AS pid, S.datid AS datid, D.datname AS datname, diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c index 60a4617a585..e19675a6d05 100644 --- a/src/backend/commands/cluster.c +++ b/src/backend/commands/cluster.c @@ -1,7 +1,8 @@ /*------------------------------------------------------------------------- * * cluster.c - * CLUSTER a table on an index. This is now also used for VACUUM FULL. + * CLUSTER a table on an index. This is now also used for VACUUM FULL a= nd + * REPACK. * * There is hardly anything left of Paul Brown's original implementation... * @@ -67,27 +68,35 @@ typedef struct Oid indexOid; } RelToCluster; =20 - -static void cluster_multiple_rels(List *rtcs, ClusterParams *params); +static bool cluster_rel_recheck(RepackCommand cmd, Relation OldHeap, + Oid indexOid, Oid userid, int options); static void rebuild_relation(Relation OldHeap, Relation index, bool verbos= e); static void copy_table_data(Relation NewHeap, Relation OldHeap, Relation O= ldIndex, bool verbose, bool *pSwapToastByContent, TransactionId *pFreezeXid, MultiXactId *pCutoffMulti); -static List *get_tables_to_cluster(MemoryContext cluster_context); -static List *get_tables_to_cluster_partitioned(MemoryContext cluster_conte= xt, - Oid indexOid); -static bool cluster_is_permitted_for_relation(Oid relid, Oid userid); +static List *get_tables_to_repack(RepackCommand cmd, bool usingindex, + MemoryContext permcxt); +static List *get_tables_to_repack_partitioned(RepackCommand cmd, + Oid relid, bool rel_is_index, + MemoryContext permcxt); +static bool cluster_is_permitted_for_relation(RepackCommand cmd, + Oid relid, Oid userid); +static Relation process_single_relation(RepackStmt *stmt, + ClusterParams *params); +static Oid determine_clustered_index(Relation rel, bool usingindex, + const char *indexname); +static const char *RepackCommandAsString(RepackCommand cmd); =20 =20 -/*------------------------------------------------------------------------= --- - * This cluster code allows for clustering multiple tables at once. Because +/* + * The repack code allows for processing multiple tables at once. Because * of this, we cannot just run everything on a single transaction, or we * would be forced to acquire exclusive locks on all the tables being * clustered, simultaneously --- very likely leading to deadlock. * - * To solve this we follow a similar strategy to VACUUM code, - * clustering each relation in a separate transaction. For this to work, - * we need to: + * To solve this we follow a similar strategy to VACUUM code, processing e= ach + * relation in a separate transaction. For this to work, we need to: + * * - provide a separate memory context so that we can pass information in * a way that survives across transactions * - start a new transaction every time a new relation is clustered @@ -98,197 +107,166 @@ static bool cluster_is_permitted_for_relation(Oid re= lid, Oid userid); * * The single-relation case does not have any such overhead. * - * We also allow a relation to be specified without index. In that case, - * the indisclustered bit will be looked up, and an ERROR will be thrown - * if there is no index with the bit set. - *------------------------------------------------------------------------= --- + * We also allow a relation to be repacked following an index, but without + * naming a specific one. In that case, the indisclustered bit will be + * looked up, and an ERROR will be thrown if no so-marked index is found. */ void -cluster(ParseState *pstate, ClusterStmt *stmt, bool isTopLevel) +ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel) { - ListCell *lc; ClusterParams params =3D {0}; - bool verbose =3D false; Relation rel =3D NULL; - Oid indexOid =3D InvalidOid; - MemoryContext cluster_context; + MemoryContext repack_context; List *rtcs; =20 /* Parse option list */ - foreach(lc, stmt->params) + foreach_node(DefElem, opt, stmt->params) { - DefElem *opt =3D (DefElem *) lfirst(lc); - if (strcmp(opt->defname, "verbose") =3D=3D 0) - verbose =3D defGetBoolean(opt); + params.options |=3D defGetBoolean(opt) ? CLUOPT_VERBOSE : 0; + else if (strcmp(opt->defname, "analyze") =3D=3D 0 || + strcmp(opt->defname, "analyse") =3D=3D 0) + params.options |=3D defGetBoolean(opt) ? CLUOPT_ANALYZE : 0; else ereport(ERROR, - (errcode(ERRCODE_SYNTAX_ERROR), - errmsg("unrecognized %s option \"%s\"", - "CLUSTER", opt->defname), - parser_errposition(pstate, opt->location))); + errcode(ERRCODE_SYNTAX_ERROR), + errmsg("unrecognized %s option \"%s\"", + RepackCommandAsString(stmt->command), + opt->defname), + parser_errposition(pstate, opt->location)); } =20 - params.options =3D (verbose ? CLUOPT_VERBOSE : 0); - + /* + * If a single relation is specified, process it and we're done ... unless + * the relation is a partitioned table, in which case we fall through. + */ if (stmt->relation !=3D NULL) { - /* This is the single-relation case. */ - Oid tableOid; - - /* - * Find, lock, and check permissions on the table. We obtain - * AccessExclusiveLock right away to avoid lock-upgrade hazard in the - * single-transaction case. - */ - tableOid =3D RangeVarGetRelidExtended(stmt->relation, - AccessExclusiveLock, - 0, - RangeVarCallbackMaintainsTable, - NULL); - rel =3D table_open(tableOid, NoLock); - - /* - * Reject clustering a remote temp table ... their local buffer - * manager is not going to cope. - */ - if (RELATION_IS_OTHER_TEMP(rel)) - ereport(ERROR, - (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), - errmsg("cannot cluster temporary tables of other sessions"))); - - if (stmt->indexname =3D=3D NULL) - { - ListCell *index; - - /* We need to find the index that has indisclustered set. */ - foreach(index, RelationGetIndexList(rel)) - { - indexOid =3D lfirst_oid(index); - if (get_index_isclustered(indexOid)) - break; - indexOid =3D InvalidOid; - } - - if (!OidIsValid(indexOid)) - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("there is no previously clustered index for table \"%s\"", - stmt->relation->relname))); - } - else - { - /* - * The index is expected to be in the same namespace as the - * relation. - */ - indexOid =3D get_relname_relid(stmt->indexname, - rel->rd_rel->relnamespace); - if (!OidIsValid(indexOid)) - ereport(ERROR, - (errcode(ERRCODE_UNDEFINED_OBJECT), - errmsg("index \"%s\" for table \"%s\" does not exist", - stmt->indexname, stmt->relation->relname))); - } - - /* For non-partitioned tables, do what we came here to do. */ - if (rel->rd_rel->relkind !=3D RELKIND_PARTITIONED_TABLE) - { - cluster_rel(rel, indexOid, ¶ms); - /* cluster_rel closes the relation, but keeps lock */ - - return; - } + rel =3D process_single_relation(stmt, ¶ms); + if (rel =3D=3D NULL) + return; /* all done */ } =20 + /* + * Don't allow ANALYZE in the multiple-relation case for now. Maybe we + * can add support for this later. + */ + if (params.options & CLUOPT_ANALYZE) + ereport(ERROR, + errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("cannot %s multiple tables", "REPACK (ANALYZE)")); + /* * By here, we know we are in a multi-table situation. In order to avoid * holding locks for too long, we want to process each table in its own * transaction. This forces us to disallow running inside a user * transaction block. */ - PreventInTransactionBlock(isTopLevel, "CLUSTER"); + PreventInTransactionBlock(isTopLevel, RepackCommandAsString(stmt->command= )); =20 /* Also, we need a memory context to hold our list of relations */ - cluster_context =3D AllocSetContextCreate(PortalContext, - "Cluster", - ALLOCSET_DEFAULT_SIZES); + repack_context =3D AllocSetContextCreate(PortalContext, + "Repack", + ALLOCSET_DEFAULT_SIZES); + + params.options |=3D CLUOPT_RECHECK; =20 /* - * Either we're processing a partitioned table, or we were not given any - * table name at all. In either case, obtain a list of relations to - * process. - * - * In the former case, an index name must have been given, so we don't - * need to recheck its "indisclustered" bit, but we have to check that it - * is an index that we can cluster on. In the latter case, we set the - * option bit to have indisclustered verified. - * - * Rechecking the relation itself is necessary here in all cases. + * If we don't have a relation yet, determine a relation list. If we do, + * then it must be a partitioned table, and we want to process its + * partitions. */ - params.options |=3D CLUOPT_RECHECK; - if (rel !=3D NULL) + if (rel =3D=3D NULL) { - Assert(rel->rd_rel->relkind =3D=3D RELKIND_PARTITIONED_TABLE); - check_index_is_clusterable(rel, indexOid, AccessShareLock); - rtcs =3D get_tables_to_cluster_partitioned(cluster_context, indexOid); - - /* close relation, releasing lock on parent table */ - table_close(rel, AccessExclusiveLock); + Assert(stmt->indexname =3D=3D NULL); + rtcs =3D get_tables_to_repack(stmt->command, stmt->usingindex, + repack_context); + params.options |=3D CLUOPT_RECHECK_ISCLUSTERED; } else { - rtcs =3D get_tables_to_cluster(cluster_context); - params.options |=3D CLUOPT_RECHECK_ISCLUSTERED; - } + Oid relid; + bool rel_is_index; =20 - /* Do the job. */ - cluster_multiple_rels(rtcs, ¶ms); + Assert(rel->rd_rel->relkind =3D=3D RELKIND_PARTITIONED_TABLE); =20 - /* Start a new transaction for the cleanup work. */ - StartTransactionCommand(); + /* + * If USING INDEX was specified, resolve the index name now and pass + * it down. + */ + if (stmt->usingindex) + { + /* + * If no index name was specified when repacking a partitioned + * table, punt for now. Maybe we can improve this later. + */ + if (!stmt->indexname) + ereport(ERROR, + errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), + errmsg("there is no previously clustered index for table \"%s\"", + RelationGetRelationName(rel))); + + relid =3D determine_clustered_index(rel, stmt->usingindex, + stmt->indexname); + if (!OidIsValid(relid)) + elog(ERROR, "unable to determine index to cluster on"); + /* XXX is this the right place for this check? */ + check_index_is_clusterable(rel, relid, AccessExclusiveLock); + rel_is_index =3D true; + } + else + { + relid =3D RelationGetRelid(rel); + rel_is_index =3D false; + } =20 - /* Clean up working storage */ - MemoryContextDelete(cluster_context); -} + rtcs =3D get_tables_to_repack_partitioned(stmt->command, + relid, rel_is_index, + repack_context); =20 -/* - * Given a list of relations to cluster, process each of them in a separate - * transaction. - * - * We expect to be in a transaction at start, but there isn't one when we - * return. - */ -static void -cluster_multiple_rels(List *rtcs, ClusterParams *params) -{ - ListCell *lc; + /* close parent relation, releasing lock on it */ + table_close(rel, AccessExclusiveLock); + rel =3D NULL; + } =20 /* Commit to get out of starting transaction */ PopActiveSnapshot(); CommitTransactionCommand(); =20 /* Cluster the tables, each in a separate transaction */ - foreach(lc, rtcs) + Assert(rel =3D=3D NULL); + foreach_ptr(RelToCluster, rtc, rtcs) { - RelToCluster *rtc =3D (RelToCluster *) lfirst(lc); - Relation rel; - /* Start a new transaction for each relation. */ StartTransactionCommand(); =20 + /* + * Open the target table, coping with the case where it has been + * dropped. + */ + rel =3D try_table_open(rtc->tableOid, AccessExclusiveLock); + if (rel =3D=3D NULL) + { + CommitTransactionCommand(); + continue; + } + /* functions in indexes may want a snapshot set */ PushActiveSnapshot(GetTransactionSnapshot()); =20 - rel =3D table_open(rtc->tableOid, AccessExclusiveLock); - /* Process this table */ - cluster_rel(rel, rtc->indexOid, params); + cluster_rel(stmt->command, rel, rtc->indexOid, ¶ms); /* cluster_rel closes the relation, but keeps lock */ =20 PopActiveSnapshot(); CommitTransactionCommand(); } + + /* Start a new transaction for the cleanup work. */ + StartTransactionCommand(); + + /* Clean up working storage */ + MemoryContextDelete(repack_context); } =20 /* @@ -304,11 +282,14 @@ cluster_multiple_rels(List *rtcs, ClusterParams *para= ms) * them incrementally while we load the table. * * If indexOid is InvalidOid, the table will be rewritten in physical order - * instead of index order. This is the new implementation of VACUUM FULL, - * and error messages should refer to the operation as VACUUM not CLUSTER. + * instead of index order. + * + * 'cmd' indicates which command is being executed, to be used for error + * messages. */ void -cluster_rel(Relation OldHeap, Oid indexOid, ClusterParams *params) +cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid, + ClusterParams *params) { Oid tableOid =3D RelationGetRelid(OldHeap); Oid save_userid; @@ -323,13 +304,8 @@ cluster_rel(Relation OldHeap, Oid indexOid, ClusterPar= ams *params) /* Check for user-requested abort. */ CHECK_FOR_INTERRUPTS(); =20 - pgstat_progress_start_command(PROGRESS_COMMAND_CLUSTER, tableOid); - if (OidIsValid(indexOid)) - pgstat_progress_update_param(PROGRESS_CLUSTER_COMMAND, - PROGRESS_CLUSTER_COMMAND_CLUSTER); - else - pgstat_progress_update_param(PROGRESS_CLUSTER_COMMAND, - PROGRESS_CLUSTER_COMMAND_VACUUM_FULL); + pgstat_progress_start_command(PROGRESS_COMMAND_REPACK, tableOid); + pgstat_progress_update_param(PROGRESS_REPACK_COMMAND, cmd); =20 /* * Switch to the table owner's userid, so that any index functions are run @@ -350,86 +326,38 @@ cluster_rel(Relation OldHeap, Oid indexOid, ClusterPa= rams *params) * *must* skip the one on indisclustered since it would reject an attempt * to cluster a not-previously-clustered index. */ - if (recheck) - { - /* Check that the user still has privileges for the relation */ - if (!cluster_is_permitted_for_relation(tableOid, save_userid)) - { - relation_close(OldHeap, AccessExclusiveLock); - goto out; - } - - /* - * Silently skip a temp table for a remote session. Only doing this - * check in the "recheck" case is appropriate (which currently means - * somebody is executing a database-wide CLUSTER or on a partitioned - * table), because there is another check in cluster() which will stop - * any attempt to cluster remote temp tables by name. There is - * another check in cluster_rel which is redundant, but we leave it - * for extra safety. - */ - if (RELATION_IS_OTHER_TEMP(OldHeap)) - { - relation_close(OldHeap, AccessExclusiveLock); - goto out; - } - - if (OidIsValid(indexOid)) - { - /* - * Check that the index still exists - */ - if (!SearchSysCacheExists1(RELOID, ObjectIdGetDatum(indexOid))) - { - relation_close(OldHeap, AccessExclusiveLock); - goto out; - } - - /* - * Check that the index is still the one with indisclustered set, - * if needed. - */ - if ((params->options & CLUOPT_RECHECK_ISCLUSTERED) !=3D 0 && - !get_index_isclustered(indexOid)) - { - relation_close(OldHeap, AccessExclusiveLock); - goto out; - } - } - } + if (recheck && + !cluster_rel_recheck(cmd, OldHeap, indexOid, save_userid, + params->options)) + goto out; =20 /* - * We allow VACUUM FULL, but not CLUSTER, on shared catalogs. CLUSTER - * would work in most respects, but the index would only get marked as - * indisclustered in the current database, leading to unexpected behavior - * if CLUSTER were later invoked in another database. + * We allow repacking shared catalogs only when not using an index. It + * would work to use an index in most respects, but the index would only + * get marked as indisclustered in the current database, leading to + * unexpected behavior if CLUSTER were later invoked in another database. */ if (OidIsValid(indexOid) && OldHeap->rd_rel->relisshared) ereport(ERROR, - (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), - errmsg("cannot cluster a shared catalog"))); + errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("cannot run %s on a shared catalog", + RepackCommandAsString(cmd))); =20 /* * Don't process temp tables of other backends ... their local buffer * manager is not going to cope. */ if (RELATION_IS_OTHER_TEMP(OldHeap)) - { - if (OidIsValid(indexOid)) - ereport(ERROR, - (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), - errmsg("cannot cluster temporary tables of other sessions"))); - else - ereport(ERROR, - (errcode(ERRCODE_FEATURE_NOT_SUPPORTED), - errmsg("cannot vacuum temporary tables of other sessions"))); - } + ereport(ERROR, + errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("cannot run %s on temporary tables of other sessions", + RepackCommandAsString(cmd))); =20 /* * Also check for active uses of the relation in the current transaction, * including open scans and pending AFTER trigger events. */ - CheckTableNotInUse(OldHeap, OidIsValid(indexOid) ? "CLUSTER" : "VACUUM"); + CheckTableNotInUse(OldHeap, RepackCommandAsString(cmd)); =20 /* Check heap and index are valid to cluster on */ if (OidIsValid(indexOid)) @@ -442,6 +370,24 @@ cluster_rel(Relation OldHeap, Oid indexOid, ClusterPar= ams *params) else index =3D NULL; =20 + /* + * When allow_system_table_mods is turned off, we disallow repacking a + * catalog on a particular index unless that's already the clustered index + * for that catalog. + * + * XXX We don't check for this in CLUSTER, because it's historically been + * allowed. + */ + if (cmd !=3D REPACK_COMMAND_CLUSTER && + !allowSystemTableMods && OidIsValid(indexOid) && + IsCatalogRelation(OldHeap) && !index->rd_index->indisclustered) + ereport(ERROR, + errcode(ERRCODE_INSUFFICIENT_PRIVILEGE), + errmsg("permission denied: \"%s\" is a system catalog", + RelationGetRelationName(OldHeap)), + errdetail("System catalogs can only be clustered by the index they're = already clustered on, if any, unless \"%s\" is enabled.", + "allow_system_table_mods")); + /* * Quietly ignore the request if this is a materialized view which has not * been populated from its query. No harm is done because there is no data @@ -482,6 +428,63 @@ out: pgstat_progress_end_command(); } =20 +/* + * Check if the table (and its index) still meets the requirements of + * cluster_rel(). + */ +static bool +cluster_rel_recheck(RepackCommand cmd, Relation OldHeap, Oid indexOid, + Oid userid, int options) +{ + Oid tableOid =3D RelationGetRelid(OldHeap); + + /* Check that the user still has privileges for the relation */ + if (!cluster_is_permitted_for_relation(cmd, tableOid, userid)) + { + relation_close(OldHeap, AccessExclusiveLock); + return false; + } + + /* + * Silently skip a temp table for a remote session. Only doing this check + * in the "recheck" case is appropriate (which currently means somebody is + * executing a database-wide CLUSTER or on a partitioned table), because + * there is another check in cluster() which will stop any attempt to + * cluster remote temp tables by name. There is another check in + * cluster_rel which is redundant, but we leave it for extra safety. + */ + if (RELATION_IS_OTHER_TEMP(OldHeap)) + { + relation_close(OldHeap, AccessExclusiveLock); + return false; + } + + if (OidIsValid(indexOid)) + { + /* + * Check that the index still exists + */ + if (!SearchSysCacheExists1(RELOID, ObjectIdGetDatum(indexOid))) + { + relation_close(OldHeap, AccessExclusiveLock); + return false; + } + + /* + * Check that the index is still the one with indisclustered set, if + * needed. + */ + if ((options & CLUOPT_RECHECK_ISCLUSTERED) !=3D 0 && + !get_index_isclustered(indexOid)) + { + relation_close(OldHeap, AccessExclusiveLock); + return false; + } + } + + return true; +} + /* * Verify that the specified heap and index are valid to cluster on * @@ -642,8 +645,8 @@ rebuild_relation(Relation OldHeap, Relation index, bool= verbose) Assert(CheckRelationLockedByMe(OldHeap, AccessExclusiveLock, false) && (index =3D=3D NULL || CheckRelationLockedByMe(index, AccessExclusiveL= ock, false))); =20 - if (index) - /* Mark the correct index as clustered */ + /* for CLUSTER or REPACK USING INDEX, mark the index as the one to use */ + if (index !=3D NULL) mark_index_clustered(OldHeap, RelationGetRelid(index), true); =20 /* Remember info about rel before closing OldHeap */ @@ -958,20 +961,20 @@ copy_table_data(Relation NewHeap, Relation OldHeap, R= elation OldIndex, bool verb /* Log what we're doing */ if (OldIndex !=3D NULL && !use_sort) ereport(elevel, - (errmsg("clustering \"%s.%s\" using index scan on \"%s\"", - nspname, - RelationGetRelationName(OldHeap), - RelationGetRelationName(OldIndex)))); + errmsg("repacking \"%s.%s\" using index scan on \"%s\"", + nspname, + RelationGetRelationName(OldHeap), + RelationGetRelationName(OldIndex))); else if (use_sort) ereport(elevel, - (errmsg("clustering \"%s.%s\" using sequential scan and sort", - nspname, - RelationGetRelationName(OldHeap)))); + errmsg("repacking \"%s.%s\" using sequential scan and sort", + nspname, + RelationGetRelationName(OldHeap))); else ereport(elevel, - (errmsg("vacuuming \"%s.%s\"", - nspname, - RelationGetRelationName(OldHeap)))); + errmsg("repacking \"%s.%s\" in physical order", + nspname, + RelationGetRelationName(OldHeap))); =20 /* * Hand off the actual copying to AM specific function, the generic code @@ -1458,8 +1461,8 @@ finish_heap_swap(Oid OIDOldHeap, Oid OIDNewHeap, int i; =20 /* Report that we are now swapping relation files */ - pgstat_progress_update_param(PROGRESS_CLUSTER_PHASE, - PROGRESS_CLUSTER_PHASE_SWAP_REL_FILES); + pgstat_progress_update_param(PROGRESS_REPACK_PHASE, + PROGRESS_REPACK_PHASE_SWAP_REL_FILES); =20 /* Zero out possible results from swapped_relation_files */ memset(mapped_tables, 0, sizeof(mapped_tables)); @@ -1509,14 +1512,14 @@ finish_heap_swap(Oid OIDOldHeap, Oid OIDNewHeap, reindex_flags |=3D REINDEX_REL_FORCE_INDEXES_PERMANENT; =20 /* Report that we are now reindexing relations */ - pgstat_progress_update_param(PROGRESS_CLUSTER_PHASE, - PROGRESS_CLUSTER_PHASE_REBUILD_INDEX); + pgstat_progress_update_param(PROGRESS_REPACK_PHASE, + PROGRESS_REPACK_PHASE_REBUILD_INDEX); =20 reindex_relation(NULL, OIDOldHeap, reindex_flags, &reindex_params); =20 /* Report that we are now doing clean up */ - pgstat_progress_update_param(PROGRESS_CLUSTER_PHASE, - PROGRESS_CLUSTER_PHASE_FINAL_CLEANUP); + pgstat_progress_update_param(PROGRESS_REPACK_PHASE, + PROGRESS_REPACK_PHASE_FINAL_CLEANUP); =20 /* * If the relation being rebuilt is pg_class, swap_relation_files() @@ -1632,106 +1635,191 @@ finish_heap_swap(Oid OIDOldHeap, Oid OIDNewHeap, } } =20 - /* - * Get a list of tables that the current user has privileges on and - * have indisclustered set. Return the list in a List * of RelToCluster - * (stored in the specified memory context), each one giving the tableOid - * and the indexOid on which the table is already clustered. + * Determine which relations to process, when REPACK/CLUSTER is called + * without specifying a table name. The exact process depends on whether + * USING INDEX was given or not, and in any case we only return tables and + * materialized views that the current user has privileges to repack/clust= er. + * + * If USING INDEX was given, we scan pg_index to find those that have + * indisclustered set; if it was not given, scan pg_class and return all + * tables. + * + * Return it as a list of RelToCluster in the given memory context. */ static List * -get_tables_to_cluster(MemoryContext cluster_context) +get_tables_to_repack(RepackCommand cmd, bool usingindex, MemoryContext per= mcxt) { - Relation indRelation; + Relation catalog; TableScanDesc scan; - ScanKeyData entry; - HeapTuple indexTuple; - Form_pg_index index; - MemoryContext old_context; + HeapTuple tuple; List *rtcs =3D NIL; =20 - /* - * Get all indexes that have indisclustered set and that the current user - * has the appropriate privileges for. - */ - indRelation =3D table_open(IndexRelationId, AccessShareLock); - ScanKeyInit(&entry, - Anum_pg_index_indisclustered, - BTEqualStrategyNumber, F_BOOLEQ, - BoolGetDatum(true)); - scan =3D table_beginscan_catalog(indRelation, 1, &entry); - while ((indexTuple =3D heap_getnext(scan, ForwardScanDirection)) !=3D NUL= L) + if (usingindex) { - RelToCluster *rtc; + ScanKeyData entry; + + catalog =3D table_open(IndexRelationId, AccessShareLock); + ScanKeyInit(&entry, + Anum_pg_index_indisclustered, + BTEqualStrategyNumber, F_BOOLEQ, + BoolGetDatum(true)); + scan =3D table_beginscan_catalog(catalog, 1, &entry); + while ((tuple =3D heap_getnext(scan, ForwardScanDirection)) !=3D NULL) + { + RelToCluster *rtc; + Form_pg_index index; + MemoryContext oldcxt; =20 - index =3D (Form_pg_index) GETSTRUCT(indexTuple); + index =3D (Form_pg_index) GETSTRUCT(tuple); =20 - if (!cluster_is_permitted_for_relation(index->indrelid, GetUserId())) - continue; + /* + * Try to obtain a light lock on the index's table, to ensure it + * doesn't go away while we collect the list. If we cannot, just + * disregard it. + */ + if (!ConditionalLockRelationOid(index->indrelid, AccessShareLock)) + continue; =20 - /* Use a permanent memory context for the result list */ - old_context =3D MemoryContextSwitchTo(cluster_context); + /* Verify that the table still exists */ + if (!SearchSysCacheExists1(RELOID, ObjectIdGetDatum(index->indrelid))) + { + /* Release useless lock */ + UnlockRelationOid(index->indrelid, AccessShareLock); + continue; + } =20 - rtc =3D palloc_object(RelToCluster); - rtc->tableOid =3D index->indrelid; - rtc->indexOid =3D index->indexrelid; - rtcs =3D lappend(rtcs, rtc); + if (!cluster_is_permitted_for_relation(cmd, index->indrelid, + GetUserId())) + continue; + + /* Use a permanent memory context for the result list */ + oldcxt =3D MemoryContextSwitchTo(permcxt); + rtc =3D palloc_object(RelToCluster); + rtc->tableOid =3D index->indrelid; + rtc->indexOid =3D index->indexrelid; + rtcs =3D lappend(rtcs, rtc); + MemoryContextSwitchTo(oldcxt); + } + } + else + { + catalog =3D table_open(RelationRelationId, AccessShareLock); + scan =3D table_beginscan_catalog(catalog, 0, NULL); + + while ((tuple =3D heap_getnext(scan, ForwardScanDirection)) !=3D NULL) + { + RelToCluster *rtc; + Form_pg_class class; + MemoryContext oldcxt; + + class =3D (Form_pg_class) GETSTRUCT(tuple); + + /* + * Try to obtain a light lock on the table, to ensure it doesn't + * go away while we collect the list. If we cannot, just + * disregard the table. + */ + if (!ConditionalLockRelationOid(class->oid, AccessShareLock)) + continue; =20 - MemoryContextSwitchTo(old_context); + /* Verify that the table still exists */ + if (!SearchSysCacheExists1(RELOID, ObjectIdGetDatum(class->oid))) + { + /* Release useless lock */ + UnlockRelationOid(class->oid, AccessShareLock); + continue; + } + + /* Can only process plain tables and matviews */ + if (class->relkind !=3D RELKIND_RELATION && + class->relkind !=3D RELKIND_MATVIEW) + continue; + + /* noisily skip rels which the user can't process */ + if (!cluster_is_permitted_for_relation(cmd, class->oid, + GetUserId())) + continue; + + /* Use a permanent memory context for the result list */ + oldcxt =3D MemoryContextSwitchTo(permcxt); + rtc =3D palloc_object(RelToCluster); + rtc->tableOid =3D class->oid; + rtc->indexOid =3D InvalidOid; + rtcs =3D lappend(rtcs, rtc); + MemoryContextSwitchTo(oldcxt); + } } - table_endscan(scan); =20 - relation_close(indRelation, AccessShareLock); + table_endscan(scan); + relation_close(catalog, AccessShareLock); =20 return rtcs; } =20 /* - * Given an index on a partitioned table, return a list of RelToCluster for + * Given a partitioned table or its index, return a list of RelToCluster f= or * all the children leaves tables/indexes. * * Like expand_vacuum_rel, but here caller must hold AccessExclusiveLock * on the table containing the index. + * + * 'rel_is_index' tells whether 'relid' is that of an index (true) or of t= he + * owning relation. */ static List * -get_tables_to_cluster_partitioned(MemoryContext cluster_context, Oid index= Oid) +get_tables_to_repack_partitioned(RepackCommand cmd, Oid relid, + bool rel_is_index, MemoryContext permcxt) { List *inhoids; - ListCell *lc; List *rtcs =3D NIL; - MemoryContext old_context; - - /* Do not lock the children until they're processed */ - inhoids =3D find_all_inheritors(indexOid, NoLock, NULL); =20 - foreach(lc, inhoids) + /* + * Do not lock the children until they're processed. Note that we do hold + * a lock on the parent partitioned table. + */ + inhoids =3D find_all_inheritors(relid, NoLock, NULL); + foreach_oid(child_oid, inhoids) { - Oid indexrelid =3D lfirst_oid(lc); - Oid relid =3D IndexGetRelation(indexrelid, false); + Oid table_oid, + index_oid; RelToCluster *rtc; + MemoryContext oldcxt; =20 - /* consider only leaf indexes */ - if (get_rel_relkind(indexrelid) !=3D RELKIND_INDEX) - continue; + if (rel_is_index) + { + /* consider only leaf indexes */ + if (get_rel_relkind(child_oid) !=3D RELKIND_INDEX) + continue; + + table_oid =3D IndexGetRelation(child_oid, false); + index_oid =3D child_oid; + } + else + { + /* consider only leaf relations */ + if (get_rel_relkind(child_oid) !=3D RELKIND_RELATION) + continue; + + table_oid =3D child_oid; + index_oid =3D InvalidOid; + } =20 /* * It's possible that the user does not have privileges to CLUSTER the - * leaf partition despite having such privileges on the partitioned - * table. We skip any partitions which the user is not permitted to - * CLUSTER. + * leaf partition despite having them on the partitioned table. Skip + * if so. */ - if (!cluster_is_permitted_for_relation(relid, GetUserId())) + if (!cluster_is_permitted_for_relation(cmd, table_oid, GetUserId())) continue; =20 /* Use a permanent memory context for the result list */ - old_context =3D MemoryContextSwitchTo(cluster_context); - + oldcxt =3D MemoryContextSwitchTo(permcxt); rtc =3D palloc_object(RelToCluster); - rtc->tableOid =3D relid; - rtc->indexOid =3D indexrelid; + rtc->tableOid =3D table_oid; + rtc->indexOid =3D index_oid; rtcs =3D lappend(rtcs, rtc); - - MemoryContextSwitchTo(old_context); + MemoryContextSwitchTo(oldcxt); } =20 return rtcs; @@ -1742,13 +1830,167 @@ get_tables_to_cluster_partitioned(MemoryContext cl= uster_context, Oid indexOid) * function emits a WARNING. */ static bool -cluster_is_permitted_for_relation(Oid relid, Oid userid) +cluster_is_permitted_for_relation(RepackCommand cmd, Oid relid, Oid userid) { + Assert(cmd =3D=3D REPACK_COMMAND_CLUSTER || cmd =3D=3D REPACK_COMMAND_REP= ACK); + if (pg_class_aclcheck(relid, userid, ACL_MAINTAIN) =3D=3D ACLCHECK_OK) return true; =20 ereport(WARNING, - (errmsg("permission denied to cluster \"%s\", skipping it", - get_rel_name(relid)))); + errmsg("permission denied to execute %s on \"%s\", skipping it", + RepackCommandAsString(cmd), + get_rel_name(relid))); + return false; } + + +/* + * Given a RepackStmt with an indicated relation name, resolve the relation + * name, obtain lock on it, then determine what to do based on the relation + * type: if it's table and not partitioned, repack it as indicated (using = an + * existing clustered index, or following the given one), and return NULL. + * + * On the other hand, if the table is partitioned, do nothing further and + * instead return the opened and locked relcache entry, so that caller can + * process the partitions using the multiple-table handling code. In this + * case, if an index name is given, it's up to the caller to resolve it. + */ +static Relation +process_single_relation(RepackStmt *stmt, ClusterParams *params) +{ + Relation rel; + Oid tableOid; + + Assert(stmt->relation !=3D NULL); + Assert(stmt->command =3D=3D REPACK_COMMAND_CLUSTER || + stmt->command =3D=3D REPACK_COMMAND_REPACK); + + /* + * Find, lock, and check permissions on the table. We obtain + * AccessExclusiveLock right away to avoid lock-upgrade hazard in the + * single-transaction case. + */ + tableOid =3D RangeVarGetRelidExtended(stmt->relation->relation, + AccessExclusiveLock, + 0, + RangeVarCallbackMaintainsTable, + NULL); + rel =3D table_open(tableOid, NoLock); + + /* + * Reject clustering a remote temp table ... their local buffer manager is + * not going to cope. + */ + if (RELATION_IS_OTHER_TEMP(rel)) + ereport(ERROR, + errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("cannot execute %s on temporary tables of other sessions", + RepackCommandAsString(stmt->command))); + + /* + * Make sure ANALYZE is specified if a column list is present. + */ + if ((params->options & CLUOPT_ANALYZE) =3D=3D 0 && stmt->relation->va_col= s !=3D NIL) + ereport(ERROR, + errcode(ERRCODE_FEATURE_NOT_SUPPORTED), + errmsg("ANALYZE option must be specified when a column list is provide= d")); + + /* + * For partitioned tables, let caller handle this. Otherwise, process it + * here and we're done. + */ + if (rel->rd_rel->relkind =3D=3D RELKIND_PARTITIONED_TABLE) + return rel; + else + { + Oid indexOid; + + indexOid =3D determine_clustered_index(rel, stmt->usingindex, + stmt->indexname); + if (OidIsValid(indexOid)) + check_index_is_clusterable(rel, indexOid, AccessExclusiveLock); + cluster_rel(stmt->command, rel, indexOid, params); + + /* Do an analyze, if requested */ + if (params->options & CLUOPT_ANALYZE) + { + VacuumParams vac_params =3D {0}; + + vac_params.options |=3D VACOPT_ANALYZE; + if (params->options & CLUOPT_VERBOSE) + vac_params.options |=3D VACOPT_VERBOSE; + analyze_rel(tableOid, NULL, vac_params, + stmt->relation->va_cols, true, NULL); + } + + return NULL; + } +} + +/* + * Given a relation and the usingindex/indexname options in a + * REPACK USING INDEX or CLUSTER command, return the OID of the + * index to use for clustering the table. + * + * Caller must hold lock on the relation so that the set of indexes + * doesn't change, and must call check_index_is_clusterable. + */ +static Oid +determine_clustered_index(Relation rel, bool usingindex, const char *index= name) +{ + Oid indexOid; + + if (indexname =3D=3D NULL && usingindex) + { + /* + * If USING INDEX with no name is given, find a clustered index, or + * error out if none. + */ + indexOid =3D InvalidOid; + foreach_oid(idxoid, RelationGetIndexList(rel)) + { + if (get_index_isclustered(idxoid)) + { + indexOid =3D idxoid; + break; + } + } + + if (!OidIsValid(indexOid)) + ereport(ERROR, + errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("there is no previously clustered index for table \"%s\"", + RelationGetRelationName(rel))); + } + else if (indexname !=3D NULL) + { + /* An index was specified; obtain its OID. */ + indexOid =3D get_relname_relid(indexname, rel->rd_rel->relnamespace); + if (!OidIsValid(indexOid)) + ereport(ERROR, + errcode(ERRCODE_UNDEFINED_OBJECT), + errmsg("index \"%s\" for table \"%s\" does not exist", + indexname, RelationGetRelationName(rel))); + } + else + indexOid =3D InvalidOid; + + return indexOid; +} + +static const char * +RepackCommandAsString(RepackCommand cmd) +{ + switch (cmd) + { + case REPACK_COMMAND_REPACK: + return "REPACK"; + case REPACK_COMMAND_VACUUMFULL: + return "VACUUM"; + case REPACK_COMMAND_CLUSTER: + return "CLUSTER"; + } + return "???"; /* keep compiler quiet */ +} diff --git a/src/backend/commands/vacuum.c b/src/backend/commands/vacuum.c index 03932f45c8a..aea998260e1 100644 --- a/src/backend/commands/vacuum.c +++ b/src/backend/commands/vacuum.c @@ -351,7 +351,6 @@ ExecVacuum(ParseState *pstate, VacuumStmt *vacstmt, boo= l isTopLevel) } } =20 - /* * Sanity check DISABLE_PAGE_SKIPPING option. */ @@ -2289,8 +2288,9 @@ vacuum_rel(Oid relid, RangeVar *relation, VacuumParam= s params, if ((params.options & VACOPT_VERBOSE) !=3D 0) cluster_params.options |=3D CLUOPT_VERBOSE; =20 - /* VACUUM FULL is now a variant of CLUSTER; see cluster.c */ - cluster_rel(rel, InvalidOid, &cluster_params); + /* VACUUM FULL is a variant of REPACK; see cluster.c */ + cluster_rel(REPACK_COMMAND_VACUUMFULL, rel, InvalidOid, + &cluster_params); /* cluster_rel closes the relation, but keeps lock */ =20 rel =3D NULL; diff --git a/src/backend/parser/gram.y b/src/backend/parser/gram.y index 713ee5c10a2..54d37c10447 100644 --- a/src/backend/parser/gram.y +++ b/src/backend/parser/gram.y @@ -287,7 +287,7 @@ static Node *makeRecursiveViewSelect(char *relname, Lis= t *aliases, Node *query); AlterCompositeTypeStmt AlterUserMappingStmt AlterRoleStmt AlterRoleSetStmt AlterPolicyStmt AlterStatsStmt AlterDefaultPrivilegesStmt DefACLAction - AnalyzeStmt CallStmt ClosePortalStmt ClusterStmt CommentStmt + AnalyzeStmt CallStmt ClosePortalStmt CommentStmt ConstraintsSetStmt CopyStmt CreateAsStmt CreateCastStmt CreateDomainStmt CreateExtensionStmt CreateGroupStmt CreateOpClassStmt CreateOpFamilyStmt AlterOpFamilyStmt CreatePLangStmt @@ -304,7 +304,7 @@ static Node *makeRecursiveViewSelect(char *relname, Lis= t *aliases, Node *query); GrantStmt GrantRoleStmt ImportForeignSchemaStmt IndexStmt InsertStmt ListenStmt LoadStmt LockStmt MergeStmt NotifyStmt ExplainableStmt Prepar= ableStmt CreateFunctionStmt AlterFunctionStmt ReindexStmt RemoveAggrStmt - RemoveFuncStmt RemoveOperStmt RenameStmt ReturnStmt RevokeStmt RevokeRol= eStmt + RemoveFuncStmt RemoveOperStmt RenameStmt RepackStmt ReturnStmt RevokeStm= t RevokeRoleStmt RuleActionStmt RuleActionStmtOrEmpty RuleStmt SecLabelStmt SelectStmt TransactionStmt TransactionStmtLegacy TruncateSt= mt UnlistenStmt UpdateStmt VacuumStmt @@ -323,7 +323,7 @@ static Node *makeRecursiveViewSelect(char *relname, Lis= t *aliases, Node *query); =20 %type <str> opt_single_name %type <list> opt_qualified_name -%type <boolean> opt_concurrently +%type <boolean> opt_concurrently opt_usingindex %type <dbehavior> opt_drop_behavior %type <list> opt_utility_option_list %type <list> opt_wait_with_clause @@ -773,7 +773,7 @@ static Node *makeRecursiveViewSelect(char *relname, Lis= t *aliases, Node *query); QUOTE QUOTES =20 RANGE READ REAL REASSIGN RECURSIVE REF_P REFERENCES REFERENCING - REFRESH REINDEX RELATIVE_P RELEASE RENAME REPEATABLE REPLACE REPLICA + REFRESH REINDEX RELATIVE_P RELEASE RENAME REPACK REPEATABLE REPLACE REPLI= CA RESET RESPECT_P RESTART RESTRICT RETURN RETURNING RETURNS REVOKE RIGHT RO= LE ROLLBACK ROLLUP ROUTINE ROUTINES ROW ROWS RULE =20 @@ -1035,7 +1035,6 @@ stmt: | CallStmt | CheckPointStmt | ClosePortalStmt - | ClusterStmt | CommentStmt | ConstraintsSetStmt | CopyStmt @@ -1109,6 +1108,7 @@ stmt: | RemoveFuncStmt | RemoveOperStmt | RenameStmt + | RepackStmt | RevokeStmt | RevokeRoleStmt | RuleStmt @@ -1146,6 +1146,11 @@ opt_concurrently: | /*EMPTY*/ { $$ =3D false; } ; =20 +opt_usingindex: + USING INDEX { $$ =3D true; } + | /* EMPTY */ { $$ =3D false; } + ; + opt_drop_behavior: CASCADE { $$ =3D DROP_CASCADE; } | RESTRICT { $$ =3D DROP_RESTRICT; } @@ -12036,38 +12041,82 @@ CreateConversionStmt: /*************************************************************************= **** * * QUERY: + * REPACK [ (options) ] [ <qualified_name> [ <name_list> ] [ USING INDE= X <index_name> ] ] + * + * obsolete variants: * CLUSTER (options) [ <qualified_name> [ USING <index_name> ] ] * CLUSTER [VERBOSE] [ <qualified_name> [ USING <index_name> ] ] * CLUSTER [VERBOSE] <index_name> ON <qualified_name> (for pre-8.3) * *************************************************************************= ****/ =20 -ClusterStmt: - CLUSTER '(' utility_option_list ')' qualified_name cluster_index_specif= ication +RepackStmt: + REPACK opt_utility_option_list vacuum_relation USING INDEX name { - ClusterStmt *n =3D makeNode(ClusterStmt); + RepackStmt *n =3D makeNode(RepackStmt); =20 - n->relation =3D $5; + n->command =3D REPACK_COMMAND_REPACK; + n->relation =3D (VacuumRelation *) $3; n->indexname =3D $6; + n->usingindex =3D true; + n->params =3D $2; + $$ =3D (Node *) n; + } + | REPACK opt_utility_option_list vacuum_relation opt_usingindex + { + RepackStmt *n =3D makeNode(RepackStmt); + + n->command =3D REPACK_COMMAND_REPACK; + n->relation =3D (VacuumRelation *) $3; + n->indexname =3D NULL; + n->usingindex =3D $4; + n->params =3D $2; + $$ =3D (Node *) n; + } + | REPACK opt_utility_option_list opt_usingindex + { + RepackStmt *n =3D makeNode(RepackStmt); + + n->command =3D REPACK_COMMAND_REPACK; + n->relation =3D NULL; + n->indexname =3D NULL; + n->usingindex =3D $3; + n->params =3D $2; + $$ =3D (Node *) n; + } + | CLUSTER '(' utility_option_list ')' qualified_name cluster_index_spec= ification + { + RepackStmt *n =3D makeNode(RepackStmt); + + n->command =3D REPACK_COMMAND_CLUSTER; + n->relation =3D makeNode(VacuumRelation); + n->relation->relation =3D $5; + n->indexname =3D $6; + n->usingindex =3D true; n->params =3D $3; $$ =3D (Node *) n; } | CLUSTER opt_utility_option_list { - ClusterStmt *n =3D makeNode(ClusterStmt); + RepackStmt *n =3D makeNode(RepackStmt); =20 + n->command =3D REPACK_COMMAND_CLUSTER; n->relation =3D NULL; n->indexname =3D NULL; + n->usingindex =3D true; n->params =3D $2; $$ =3D (Node *) n; } /* unparenthesized VERBOSE kept for pre-14 compatibility */ | CLUSTER opt_verbose qualified_name cluster_index_specification { - ClusterStmt *n =3D makeNode(ClusterStmt); + RepackStmt *n =3D makeNode(RepackStmt); =20 - n->relation =3D $3; + n->command =3D REPACK_COMMAND_CLUSTER; + n->relation =3D makeNode(VacuumRelation); + n->relation->relation =3D $3; n->indexname =3D $4; + n->usingindex =3D true; if ($2) n->params =3D list_make1(makeDefElem("verbose", NULL, @2)); $$ =3D (Node *) n; @@ -12075,20 +12124,25 @@ ClusterStmt: /* unparenthesized VERBOSE kept for pre-17 compatibility */ | CLUSTER VERBOSE { - ClusterStmt *n =3D makeNode(ClusterStmt); + RepackStmt *n =3D makeNode(RepackStmt); =20 + n->command =3D REPACK_COMMAND_CLUSTER; n->relation =3D NULL; n->indexname =3D NULL; + n->usingindex =3D true; n->params =3D list_make1(makeDefElem("verbose", NULL, @2)); $$ =3D (Node *) n; } /* kept for pre-8.3 compatibility */ | CLUSTER opt_verbose name ON qualified_name { - ClusterStmt *n =3D makeNode(ClusterStmt); + RepackStmt *n =3D makeNode(RepackStmt); =20 - n->relation =3D $5; + n->command =3D REPACK_COMMAND_CLUSTER; + n->relation =3D makeNode(VacuumRelation); + n->relation->relation =3D $5; n->indexname =3D $3; + n->usingindex =3D true; if ($2) n->params =3D list_make1(makeDefElem("verbose", NULL, @2)); $$ =3D (Node *) n; @@ -18127,6 +18181,7 @@ unreserved_keyword: | RELATIVE_P | RELEASE | RENAME + | REPACK | REPEATABLE | REPLACE | REPLICA @@ -18764,6 +18819,7 @@ bare_label_keyword: | RELATIVE_P | RELEASE | RENAME + | REPACK | REPEATABLE | REPLACE | REPLICA diff --git a/src/backend/tcop/utility.c b/src/backend/tcop/utility.c index 34dd6e18df5..ca737b05115 100644 --- a/src/backend/tcop/utility.c +++ b/src/backend/tcop/utility.c @@ -279,9 +279,9 @@ ClassifyUtilityCommandAsReadOnly(Node *parsetree) return COMMAND_OK_IN_RECOVERY | COMMAND_OK_IN_READ_ONLY_TXN; } =20 - case T_ClusterStmt: case T_ReindexStmt: case T_VacuumStmt: + case T_RepackStmt: { /* * These commands write WAL, so they're not strictly @@ -856,14 +856,14 @@ standard_ProcessUtility(PlannedStmt *pstmt, ExecuteCallStmt(castNode(CallStmt, parsetree), params, isAtomicContext,= dest); break; =20 - case T_ClusterStmt: - cluster(pstate, (ClusterStmt *) parsetree, isTopLevel); - break; - case T_VacuumStmt: ExecVacuum(pstate, (VacuumStmt *) parsetree, isTopLevel); break; =20 + case T_RepackStmt: + ExecRepack(pstate, (RepackStmt *) parsetree, isTopLevel); + break; + case T_ExplainStmt: ExplainQuery(pstate, (ExplainStmt *) parsetree, params, dest); break; @@ -2865,10 +2865,6 @@ CreateCommandTag(Node *parsetree) tag =3D CMDTAG_CALL; break; =20 - case T_ClusterStmt: - tag =3D CMDTAG_CLUSTER; - break; - case T_VacuumStmt: if (((VacuumStmt *) parsetree)->is_vacuumcmd) tag =3D CMDTAG_VACUUM; @@ -2876,6 +2872,13 @@ CreateCommandTag(Node *parsetree) tag =3D CMDTAG_ANALYZE; break; =20 + case T_RepackStmt: + if (((RepackStmt *) parsetree)->command =3D=3D REPACK_COMMAND_CLUSTER) + tag =3D CMDTAG_CLUSTER; + else + tag =3D CMDTAG_REPACK; + break; + case T_ExplainStmt: tag =3D CMDTAG_EXPLAIN; break; @@ -3517,7 +3520,7 @@ GetCommandLogLevel(Node *parsetree) lev =3D LOGSTMT_ALL; break; =20 - case T_ClusterStmt: + case T_RepackStmt: lev =3D LOGSTMT_DDL; break; =20 diff --git a/src/backend/utils/adt/pgstatfuncs.c b/src/backend/utils/adt/pg= statfuncs.c index 73ca0bb0b7f..55a69bf681d 100644 --- a/src/backend/utils/adt/pgstatfuncs.c +++ b/src/backend/utils/adt/pgstatfuncs.c @@ -287,8 +287,8 @@ pg_stat_get_progress_info(PG_FUNCTION_ARGS) cmdtype =3D PROGRESS_COMMAND_VACUUM; else if (pg_strcasecmp(cmd, "ANALYZE") =3D=3D 0) cmdtype =3D PROGRESS_COMMAND_ANALYZE; - else if (pg_strcasecmp(cmd, "CLUSTER") =3D=3D 0) - cmdtype =3D PROGRESS_COMMAND_CLUSTER; + else if (pg_strcasecmp(cmd, "REPACK") =3D=3D 0) + cmdtype =3D PROGRESS_COMMAND_REPACK; else if (pg_strcasecmp(cmd, "CREATE INDEX") =3D=3D 0) cmdtype =3D PROGRESS_COMMAND_CREATE_INDEX; else if (pg_strcasecmp(cmd, "BASEBACKUP") =3D=3D 0) diff --git a/src/bin/psql/tab-complete.in.c b/src/bin/psql/tab-complete.in.c index 8b91bc00062..2a1bb47ff03 100644 --- a/src/bin/psql/tab-complete.in.c +++ b/src/bin/psql/tab-complete.in.c @@ -1267,7 +1267,7 @@ static const char *const sql_commands[] =3D { "DELETE FROM", "DISCARD", "DO", "DROP", "END", "EXECUTE", "EXPLAIN", "FETCH", "GRANT", "IMPORT FOREIGN SCHEMA", "INSERT INTO", "LISTEN", "LOAD= ", "LOCK", "MERGE INTO", "MOVE", "NOTIFY", "PREPARE", - "REASSIGN", "REFRESH MATERIALIZED VIEW", "REINDEX", "RELEASE", + "REASSIGN", "REFRESH MATERIALIZED VIEW", "REINDEX", "RELEASE", "REPACK", "RESET", "REVOKE", "ROLLBACK", "SAVEPOINT", "SECURITY LABEL", "SELECT", "SET", "SHOW", "START", "TABLE", "TRUNCATE", "UNLISTEN", "UPDATE", "VACUUM", "VALUES", @@ -5086,6 +5086,46 @@ match_previous_words(int pattern_id, COMPLETE_WITH_QUERY(Query_for_list_of_tablespaces); } =20 +/* REPACK */ + else if (Matches("REPACK")) + COMPLETE_WITH_SCHEMA_QUERY_PLUS(Query_for_list_of_clusterables, + "(", "USING INDEX"); + else if (Matches("REPACK", "(*)")) + COMPLETE_WITH_SCHEMA_QUERY_PLUS(Query_for_list_of_clusterables, + "USING INDEX"); + else if (Matches("REPACK", MatchAnyExcept("("))) + COMPLETE_WITH("USING INDEX"); + else if (Matches("REPACK", "(*)", MatchAnyExcept("("))) + COMPLETE_WITH("USING INDEX"); + else if (Matches("REPACK", MatchAny, "USING", "INDEX") || + Matches("REPACK", "(*)", MatchAny, "USING", "INDEX")) + { + set_completion_reference(prev3_wd); + COMPLETE_WITH_SCHEMA_QUERY(Query_for_index_of_table); + } + /* + * Complete ... [ (*) ] <sth> USING INDEX, with a list of indexes for + * <sth>. + */ + else if (TailMatches(MatchAny, "USING", "INDEX")) + { + set_completion_reference(prev3_wd); + COMPLETE_WITH_SCHEMA_QUERY(Query_for_index_of_table); + } + else if (HeadMatches("REPACK", "(*") && + !HeadMatches("REPACK", "(*)")) + { + /* + * This fires if we're in an unfinished parenthesized option list. + * get_previous_words treats a completed parenthesized option list as + * one word, so the above test is correct. + */ + if (ends_with(prev_wd, '(') || ends_with(prev_wd, ',')) + COMPLETE_WITH("ANALYZE", "VERBOSE"); + else if (TailMatches("ANALYZE", "VERBOSE")) + COMPLETE_WITH("ON", "OFF"); + } + /* SECURITY LABEL */ else if (Matches("SECURITY")) COMPLETE_WITH("LABEL"); diff --git a/src/bin/scripts/Makefile b/src/bin/scripts/Makefile index e6cd9ef4af5..f964a5803ca 100644 --- a/src/bin/scripts/Makefile +++ b/src/bin/scripts/Makefile @@ -23,6 +23,7 @@ PROGRAMS =3D \ dropdb \ dropuser \ pg_isready \ + pg_repackdb \ reindexdb \ vacuumdb =20 @@ -39,6 +40,7 @@ clusterdb: clusterdb.o common.o $(WIN32RES) | submake-lib= pq submake-libpgport su vacuumdb: vacuumdb.o vacuuming.o common.o $(WIN32RES) | submake-libpq subm= ake-libpgport submake-libpgfeutils reindexdb: reindexdb.o common.o $(WIN32RES) | submake-libpq submake-libpgp= ort submake-libpgfeutils pg_isready: pg_isready.o common.o $(WIN32RES) | submake-libpq submake-libp= gport submake-libpgfeutils +pg_repackdb: pg_repackdb.o vacuuming.o common.o $(WIN32RES) | submake-libp= q submake-libpgport submake-libpgfeutils =20 install: all installdirs $(INSTALL_PROGRAM) createdb$(X) '$(DESTDIR)$(bindir)'/createdb$(X) @@ -49,6 +51,7 @@ install: all installdirs $(INSTALL_PROGRAM) vacuumdb$(X) '$(DESTDIR)$(bindir)'/vacuumdb$(X) $(INSTALL_PROGRAM) reindexdb$(X) '$(DESTDIR)$(bindir)'/reindexdb$(X) $(INSTALL_PROGRAM) pg_isready$(X) '$(DESTDIR)$(bindir)'/pg_isready$(X) + $(INSTALL_PROGRAM) pg_repackdb$(X) '$(DESTDIR)$(bindir)'/pg_repackdb$(X) =20 installdirs: $(MKDIR_P) '$(DESTDIR)$(bindir)' diff --git a/src/bin/scripts/meson.build b/src/bin/scripts/meson.build index c083ec38099..1e88bacebba 100644 --- a/src/bin/scripts/meson.build +++ b/src/bin/scripts/meson.build @@ -42,6 +42,7 @@ vacuuming_common =3D static_library('libvacuuming_common', =20 binaries =3D [ 'vacuumdb', + 'pg_repackdb', ] foreach binary : binaries binary_sources =3D files('@[email protected]'.format(binary)) @@ -80,6 +81,7 @@ tests +=3D { 't/100_vacuumdb.pl', 't/101_vacuumdb_all.pl', 't/102_vacuumdb_stages.pl', + 't/103_repackdb.pl', 't/200_connstr.pl', ], }, diff --git a/src/bin/scripts/pg_repackdb.c b/src/bin/scripts/pg_repackdb.c new file mode 100644 index 00000000000..2765d1e97b8 --- /dev/null +++ b/src/bin/scripts/pg_repackdb.c @@ -0,0 +1,240 @@ +/*------------------------------------------------------------------------- + * + * pg_repackdb + * An utility to run REPACK + * + * Portions Copyright (c) 1996-2025, PostgreSQL Global Development Group + * Portions Copyright (c) 1994, Regents of the University of California + * + * FIXME: this is missing a way to specify the index to use to repack one + * table, or whether to pass a WITH INDEX clause when multiple tables are + * used. Something like --index[=3Dindexname]. Adding that bleeds into + * vacuuming.c as well. + * + * src/bin/scripts/pg_repackdb.c + * + *------------------------------------------------------------------------- + */ + +#include "postgres_fe.h" + +#include <limits.h> + +#include "common.h" +#include "common/logging.h" +#include "fe_utils/option_utils.h" +#include "vacuuming.h" + +static void help(const char *progname); +static void check_objfilter(bits32 objfilter); + +int +main(int argc, char *argv[]) +{ + static struct option long_options[] =3D { + {"host", required_argument, NULL, 'h'}, + {"port", required_argument, NULL, 'p'}, + {"username", required_argument, NULL, 'U'}, + {"no-password", no_argument, NULL, 'w'}, + {"password", no_argument, NULL, 'W'}, + {"echo", no_argument, NULL, 'e'}, + {"quiet", no_argument, NULL, 'q'}, + {"dbname", required_argument, NULL, 'd'}, + {"analyze", no_argument, NULL, 'z'}, + {"all", no_argument, NULL, 'a'}, + /* XXX this could be 'i', but optional_arg is messy */ + {"index", optional_argument, NULL, 1}, + {"table", required_argument, NULL, 't'}, + {"verbose", no_argument, NULL, 'v'}, + {"jobs", required_argument, NULL, 'j'}, + {"schema", required_argument, NULL, 'n'}, + {"exclude-schema", required_argument, NULL, 'N'}, + {"maintenance-db", required_argument, NULL, 2}, + {NULL, 0, NULL, 0} + }; + + const char *progname; + int optindex; + int c; + const char *dbname =3D NULL; + const char *maintenance_db =3D NULL; + ConnParams cparams; + vacuumingOptions vacopts; + SimpleStringList objects =3D {NULL, NULL}; + int concurrentCons =3D 1; + int tbl_count =3D 0; + int ret; + + /* initialize options */ + memset(&vacopts, 0, sizeof(vacopts)); + vacopts.mode =3D MODE_REPACK; + + /* the same for connection parameters */ + memset(&cparams, 0, sizeof(cparams)); + cparams.prompt_password =3D TRI_DEFAULT; + + pg_logging_init(argv[0]); + progname =3D get_progname(argv[0]); + set_pglocale_pgservice(argv[0], PG_TEXTDOMAIN("pgscripts")); + + handle_help_version_opts(argc, argv, progname, help); + + while ((c =3D getopt_long(argc, argv, "ad:eh:j:n:N:p:qt:U:vwWz", + long_options, &optindex)) !=3D -1) + { + switch (c) + { + case 'a': + vacopts.objfilter |=3D OBJFILTER_ALL_DBS; + break; + case 'd': + vacopts.objfilter |=3D OBJFILTER_DATABASE; + dbname =3D pg_strdup(optarg); + break; + case 'e': + vacopts.echo =3D true; + break; + case 'h': + cparams.pghost =3D pg_strdup(optarg); + break; + case 'j': + if (!option_parse_int(optarg, "-j/--jobs", 1, INT_MAX, + &concurrentCons)) + exit(1); + break; + case 'n': + vacopts.objfilter |=3D OBJFILTER_SCHEMA; + simple_string_list_append(&objects, optarg); + break; + case 'N': + vacopts.objfilter |=3D OBJFILTER_SCHEMA_EXCLUDE; + simple_string_list_append(&objects, optarg); + break; + case 'p': + cparams.pgport =3D pg_strdup(optarg); + break; + case 'q': + vacopts.quiet =3D true; + break; + case 't': + vacopts.objfilter |=3D OBJFILTER_TABLE; + simple_string_list_append(&objects, optarg); + tbl_count++; + break; + case 'U': + cparams.pguser =3D pg_strdup(optarg); + break; + case 'v': + vacopts.verbose =3D true; + break; + case 'w': + cparams.prompt_password =3D TRI_NO; + break; + case 'W': + cparams.prompt_password =3D TRI_YES; + break; + case 'z': + vacopts.and_analyze =3D true; + break; + case 1: + vacopts.using_index =3D true; + if (optarg) + vacopts.indexname =3D pg_strdup(optarg); + else + vacopts.indexname =3D NULL; + break; + case 2: + maintenance_db =3D pg_strdup(optarg); + break; + default: + /* getopt_long already emitted a complaint */ + pg_log_error_hint("Try \"%s --help\" for more information.", progname); + exit(1); + } + } + + /* + * Non-option argument specifies database name as long as it wasn't + * already specified with -d / --dbname + */ + if (optind < argc && dbname =3D=3D NULL) + { + vacopts.objfilter |=3D OBJFILTER_DATABASE; + dbname =3D argv[optind]; + optind++; + } + + if (optind < argc) + { + pg_log_error("too many command-line arguments (first is \"%s\")", + argv[optind]); + pg_log_error_hint("Try \"%s --help\" for more information.", progname); + exit(1); + } + + /* + * Validate the combination of filters specified in the command-line + * options. + */ + check_objfilter(vacopts.objfilter); + + ret =3D vacuuming_main(&cparams, dbname, maintenance_db, &vacopts, + &objects, tbl_count, concurrentCons, + progname); + exit(ret); +} + +/* + * Verify that the filters used at command line are compatible. + */ +void +check_objfilter(bits32 objfilter) +{ + if ((objfilter & OBJFILTER_ALL_DBS) && + (objfilter & OBJFILTER_DATABASE)) + pg_fatal("cannot repack all databases and a specific one at the same tim= e"); + + if ((objfilter & OBJFILTER_TABLE) && + (objfilter & OBJFILTER_SCHEMA)) + pg_fatal("cannot repack all tables in schema(s) and specific table(s) at= the same time"); + + if ((objfilter & OBJFILTER_TABLE) && + (objfilter & OBJFILTER_SCHEMA_EXCLUDE)) + pg_fatal("cannot repack specific table(s) and exclude schema(s) at the s= ame time"); + + if ((objfilter & OBJFILTER_SCHEMA) && + (objfilter & OBJFILTER_SCHEMA_EXCLUDE)) + pg_fatal("cannot repack all tables in schema(s) and exclude schema(s) at= the same time"); +} + +static void +help(const char *progname) +{ + printf(_("%s repacks a PostgreSQL database.\n\n"), progname); + printf(_("Usage:\n")); + printf(_(" %s [OPTION]... [DBNAME]\n"), progname); + printf(_("\nOptions:\n")); + printf(_(" -a, --all repack all databases\n")); + printf(_(" -d, --dbname=3DDBNAME database to repack\n")); + printf(_(" -e, --echo show the commands being sent = to the server\n")); + printf(_(" --index[=3DINDEX] repack following an index\n= ")); + printf(_(" -j, --jobs=3DNUM use this many concurrent co= nnections to repack\n")); + printf(_(" -n, --schema=3DSCHEMA repack tables in the specif= ied schema(s) only\n")); + printf(_(" -N, --exclude-schema=3DSCHEMA do not repack tables in the= specified schema(s)\n")); + printf(_(" -q, --quiet don't write any messages\n")); + printf(_(" -t, --table=3D'TABLE[(COLUMNS)]' repack specific table(s) on= ly\n")); + printf(_(" -v, --verbose write a lot of output\n")); + printf(_(" -V, --version output version information, t= hen exit\n")); + printf(_(" -z, --analyze update optimizer statistics\n= ")); + printf(_(" -?, --help show this help, then exit\n")= ); + printf(_("\nConnection options:\n")); + printf(_(" -h, --host=3DHOSTNAME database server host or socket di= rectory\n")); + printf(_(" -p, --port=3DPORT database server port\n")); + printf(_(" -U, --username=3DUSERNAME user name to connect as\n")); + printf(_(" -w, --no-password never prompt for password\n")); + printf(_(" -W, --password force password prompt\n")); + printf(_(" --maintenance-db=3DDBNAME alternate maintenance database\n"= )); + printf(_("\nRead the description of the SQL command REPACK for details.\n= ")); + printf(_("\nReport bugs to <%s>.\n"), PACKAGE_BUGREPORT); + printf(_("%s home page: <%s>\n"), PACKAGE_NAME, PACKAGE_URL); +} diff --git a/src/bin/scripts/t/103_repackdb.pl b/src/bin/scripts/t/103_repa= ckdb.pl new file mode 100644 index 00000000000..44113367a61 --- /dev/null +++ b/src/bin/scripts/t/103_repackdb.pl @@ -0,0 +1,47 @@ +# Copyright (c) 2021-2026, PostgreSQL Global Development Group + +use strict; +use warnings FATAL =3D> 'all'; + +use PostgreSQL::Test::Cluster; +use PostgreSQL::Test::Utils; +use Test::More; + +program_help_ok('pg_repackdb'); +program_version_ok('pg_repackdb'); +program_options_handling_ok('pg_repackdb'); + +my $node =3D PostgreSQL::Test::Cluster->new('main'); +$node->init; +$node->start; + +$node->issues_sql_like( + [ 'pg_repackdb', 'postgres', '-t', 'pg_class'], + qr/statement: REPACK.*pg_class;/, + 'pg_repackdb processes a single table'); + +$node->safe_psql('postgres', 'CREATE USER testusr; + GRANT CREATE ON SCHEMA public TO testusr'); +$node->safe_psql('postgres', + 'CREATE TABLE cluster_1 (a int primary key); + ALTER TABLE cluster_1 CLUSTER ON cluster_1_pkey; + CREATE TABLE cluster_2 (a int unique); + ALTER TABLE cluster_2 CLUSTER ON cluster_2_a_key;', + extra_params =3D> ['-U' =3D> 'testusr']); + +$node->issues_sql_like( + [ 'pg_repackdb', 'postgres', '-U', 'testusr' ], + qr/statement: REPACK.*;/, + 'SQL REPACK run'); + +$node->issues_sql_like( + [ 'pg_repackdb', 'postgres', '--index'], + qr/statement: REPACK.*cluster_1 USING INDEX.*statement: REPACK.*cluster_2= USING INDEX/ms, + 'pg_repackdb --index chooses multiple tables'); + +$node->issues_sql_like( + [ 'pg_repackdb', 'postgres', '--analyze', '-t', 'cluster_1'], + qr/statement: REPACK \(ANALYZE\) public.cluster_1/, + 'pg_repackdb --analyze works'); + +done_testing(); diff --git a/src/bin/scripts/vacuuming.c b/src/bin/scripts/vacuuming.c index faac9089a01..0fa7be3c0c3 100644 --- a/src/bin/scripts/vacuuming.c +++ b/src/bin/scripts/vacuuming.c @@ -1,6 +1,6 @@ /*------------------------------------------------------------------------- * vacuuming.c - * Helper routines for vacuumdb + * Helper routines for vacuumdb and pg_repackdb * * Portions Copyright (c) 1996-2026, PostgreSQL Global Development Group * Portions Copyright (c) 1994, Regents of the University of California @@ -194,6 +194,14 @@ vacuum_one_database(ConnParams *cparams, =20 conn =3D connectDatabase(cparams, progname, vacopts->echo, false, true); =20 + if (vacopts->mode =3D=3D MODE_REPACK && PQserverVersion(conn) < 190000) + { + /* XXX arguably, here we should use VACUUM FULL instead of failing */ + PQfinish(conn); + pg_fatal("cannot use the \"%s\" command on server versions older than Po= stgreSQL %s", + "REPACK", "19"); + } + if (vacopts->disable_page_skipping && PQserverVersion(conn) < 90600) { PQfinish(conn); @@ -286,9 +294,18 @@ vacuum_one_database(ConnParams *cparams, if (vacopts->mode =3D=3D MODE_ANALYZE_IN_STAGES) printf(_("%s: processing database \"%s\": %s\n"), progname, PQdb(conn), _(stage_messages[stage])); - else + else if (vacopts->mode =3D=3D MODE_ANALYZE) + printf(_("%s: analyzing database \"%s\"\n"), + progname, PQdb(conn)); + else if (vacopts->mode =3D=3D MODE_VACUUM) printf(_("%s: vacuuming database \"%s\"\n"), progname, PQdb(conn)); + else + { + Assert(vacopts->mode =3D=3D MODE_REPACK); + printf(_("%s: repacking database \"%s\"\n"), + progname, PQdb(conn)); + } fflush(stdout); } =20 @@ -640,6 +657,35 @@ retrieve_objects(PGconn *conn, vacuumingOptions *vacop= ts, " AND listed_objects.object_oid IS NOT NULL\n"); } =20 + /* + * In REPACK mode, if the 'using_index' option was given but no index + * name, filter only tables that have an index with indisclustered set. + * (If an index name is given, we trust the user to pass a reasonable list + * of tables.) + * + * XXX it may be worth printing an error if an index name is given with no + * list of tables. + */ + if (vacopts->mode =3D=3D MODE_REPACK && + vacopts->using_index && !vacopts->indexname) + { + appendPQExpBufferStr(&catalog_query, + " AND EXISTS (SELECT 1 FROM pg_catalog.pg_index\n" + " WHERE indrelid =3D c.oid AND indisclustered)\n"); + } + + /* + * In REPACK mode, only consider the tables that the current user has + * MAINTAIN privileges on. XXX maybe we should do this in all cases, not + * just REPACK. The vacuumdb output is too noisy for no reason. + */ + if (vacopts->mode =3D=3D MODE_REPACK) + { + appendPQExpBufferStr(&catalog_query, + " AND pg_catalog.has_table_privilege(current_user, " + "c.oid, 'MAINTAIN')\n"); + } + /* * If no tables were listed, filter for the relevant relation types. If * tables were given via --table, don't bother filtering by relation type. @@ -878,8 +924,10 @@ prepare_vacuum_command(PGconn *conn, PQExpBuffer sql, if (vacopts->verbose) appendPQExpBufferStr(sql, " VERBOSE"); } + + appendPQExpBuffer(sql, " %s", table); } - else + else if (vacopts->mode =3D=3D MODE_VACUUM) { appendPQExpBufferStr(sql, "VACUUM"); =20 @@ -993,9 +1041,39 @@ prepare_vacuum_command(PGconn *conn, PQExpBuffer sql, if (vacopts->and_analyze) appendPQExpBufferStr(sql, " ANALYZE"); } + + appendPQExpBuffer(sql, " %s", table); } + else if (vacopts->mode =3D=3D MODE_REPACK) + { + appendPQExpBufferStr(sql, "REPACK"); =20 - appendPQExpBuffer(sql, " %s;", table); + if (vacopts->verbose) + { + appendPQExpBuffer(sql, "%sVERBOSE", sep); + sep =3D comma; + } + if (vacopts->and_analyze) + { + appendPQExpBuffer(sql, "%sANALYZE", sep); + sep =3D comma; + } + + if (sep !=3D paren) + appendPQExpBufferChar(sql, ')'); + + appendPQExpBuffer(sql, " %s", table); + + if (vacopts->using_index) + { + appendPQExpBuffer(sql, " USING INDEX"); + if (vacopts->indexname) + appendPQExpBuffer(sql, " %s", fmtIdEnc(vacopts->indexname, + PQclientEncoding(conn))); + } + } + + appendPQExpBufferChar(sql, ';'); } =20 /* @@ -1024,13 +1102,21 @@ run_vacuum_command(ParallelSlot *free_slot, vacuumi= ngOptions *vacopts, { if (table) { - pg_log_error("vacuuming of table \"%s\" in database \"%s\" failed: %s", - table, PQdb(conn), PQerrorMessage(conn)); + if (vacopts->mode =3D=3D MODE_VACUUM) + pg_log_error("vacuuming of table \"%s\" in database \"%s\" failed: %s", + table, PQdb(conn), PQerrorMessage(conn)); + else + pg_log_error("repacking of table \"%s\" in database \"%s\" failed: %s", + table, PQdb(conn), PQerrorMessage(conn)); } else { - pg_log_error("vacuuming of database \"%s\" failed: %s", - PQdb(conn), PQerrorMessage(conn)); + if (vacopts->mode =3D=3D MODE_VACUUM) + pg_log_error("vacuuming of database \"%s\" failed: %s", + PQdb(conn), PQerrorMessage(conn)); + else + pg_log_error("repacking of database \"%s\" failed: %s", + PQdb(conn), PQerrorMessage(conn)); } } } diff --git a/src/bin/scripts/vacuuming.h b/src/bin/scripts/vacuuming.h index be4a75ef2f4..6e8abe00a4f 100644 --- a/src/bin/scripts/vacuuming.h +++ b/src/bin/scripts/vacuuming.h @@ -20,6 +20,7 @@ typedef enum { MODE_VACUUM, + MODE_REPACK, MODE_ANALYZE, MODE_ANALYZE_IN_STAGES } RunMode; @@ -37,6 +38,8 @@ typedef struct vacuumingOptions bool and_analyze; bool full; bool freeze; + bool using_index; + char *indexname; bool disable_page_skipping; bool skip_locked; int min_xid_age; diff --git a/src/include/commands/cluster.h b/src/include/commands/cluster.h index 8ea81622f9d..28741988478 100644 --- a/src/include/commands/cluster.h +++ b/src/include/commands/cluster.h @@ -24,6 +24,7 @@ #define CLUOPT_RECHECK 0x02 /* recheck relation state */ #define CLUOPT_RECHECK_ISCLUSTERED 0x04 /* recheck relation state for * indisclustered */ +#define CLUOPT_ANALYZE 0x08 /* do an ANALYZE */ =20 /* options for CLUSTER */ typedef struct ClusterParams @@ -31,8 +32,11 @@ typedef struct ClusterParams bits32 options; /* bitmask of CLUOPT_* */ } ClusterParams; =20 -extern void cluster(ParseState *pstate, ClusterStmt *stmt, bool isTopLevel= ); -extern void cluster_rel(Relation OldHeap, Oid indexOid, ClusterParams *par= ams); + +extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLev= el); + +extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid index= Oid, + ClusterParams *params); extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid, LOCKMODE lockmode); extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_inter= nal); diff --git a/src/include/commands/progress.h b/src/include/commands/progres= s.h index 359221dc296..f00e39b937d 100644 --- a/src/include/commands/progress.h +++ b/src/include/commands/progress.h @@ -73,28 +73,34 @@ #define PROGRESS_ANALYZE_STARTED_BY_MANUAL 1 #define PROGRESS_ANALYZE_STARTED_BY_AUTOVACUUM 2 =20 -/* Progress parameters for cluster */ -#define PROGRESS_CLUSTER_COMMAND 0 -#define PROGRESS_CLUSTER_PHASE 1 -#define PROGRESS_CLUSTER_INDEX_RELID 2 -#define PROGRESS_CLUSTER_HEAP_TUPLES_SCANNED 3 -#define PROGRESS_CLUSTER_HEAP_TUPLES_WRITTEN 4 -#define PROGRESS_CLUSTER_TOTAL_HEAP_BLKS 5 -#define PROGRESS_CLUSTER_HEAP_BLKS_SCANNED 6 -#define PROGRESS_CLUSTER_INDEX_REBUILD_COUNT 7 - -/* Phases of cluster (as advertised via PROGRESS_CLUSTER_PHASE) */ -#define PROGRESS_CLUSTER_PHASE_SEQ_SCAN_HEAP 1 -#define PROGRESS_CLUSTER_PHASE_INDEX_SCAN_HEAP 2 -#define PROGRESS_CLUSTER_PHASE_SORT_TUPLES 3 -#define PROGRESS_CLUSTER_PHASE_WRITE_NEW_HEAP 4 -#define PROGRESS_CLUSTER_PHASE_SWAP_REL_FILES 5 -#define PROGRESS_CLUSTER_PHASE_REBUILD_INDEX 6 -#define PROGRESS_CLUSTER_PHASE_FINAL_CLEANUP 7 - -/* Commands of PROGRESS_CLUSTER */ -#define PROGRESS_CLUSTER_COMMAND_CLUSTER 1 -#define PROGRESS_CLUSTER_COMMAND_VACUUM_FULL 2 +/* + * Progress parameters for REPACK. + * + * Values for PROGRESS_REPACK_COMMAND are defined as in RepackCommand. + * + * Note: Since REPACK shares code with CLUSTER, these values are also + * used by CLUSTER. (CLUSTER is now deprecated, so it makes little sense to + * introduce a separate set of constants.) + */ +#define PROGRESS_REPACK_COMMAND 0 +#define PROGRESS_REPACK_PHASE 1 +#define PROGRESS_REPACK_INDEX_RELID 2 +#define PROGRESS_REPACK_HEAP_TUPLES_SCANNED 3 +#define PROGRESS_REPACK_HEAP_TUPLES_WRITTEN 4 +#define PROGRESS_REPACK_TOTAL_HEAP_BLKS 5 +#define PROGRESS_REPACK_HEAP_BLKS_SCANNED 6 +#define PROGRESS_REPACK_INDEX_REBUILD_COUNT 7 + +/* + * Phases of repack (as advertised via PROGRESS_REPACK_PHASE). + */ +#define PROGRESS_REPACK_PHASE_SEQ_SCAN_HEAP 1 +#define PROGRESS_REPACK_PHASE_INDEX_SCAN_HEAP 2 +#define PROGRESS_REPACK_PHASE_SORT_TUPLES 3 +#define PROGRESS_REPACK_PHASE_WRITE_NEW_HEAP 4 +#define PROGRESS_REPACK_PHASE_SWAP_REL_FILES 5 +#define PROGRESS_REPACK_PHASE_REBUILD_INDEX 6 +#define PROGRESS_REPACK_PHASE_FINAL_CLEANUP 7 =20 /* Progress parameters for CREATE INDEX */ /* 3, 4 and 5 reserved for "waitfor" metrics */ diff --git a/src/include/nodes/parsenodes.h b/src/include/nodes/parsenodes.h index 646d6ced763..c15ba5e6f29 100644 --- a/src/include/nodes/parsenodes.h +++ b/src/include/nodes/parsenodes.h @@ -3980,18 +3980,6 @@ typedef struct AlterSystemStmt VariableSetStmt *setstmt; /* SET subcommand */ } AlterSystemStmt; =20 -/* ---------------------- - * Cluster Statement (support pbrown's cluster index implementation) - * ---------------------- - */ -typedef struct ClusterStmt -{ - NodeTag type; - RangeVar *relation; /* relation being indexed, or NULL if all */ - char *indexname; /* original index defined */ - List *params; /* list of DefElem nodes */ -} ClusterStmt; - /* ---------------------- * Vacuum and Analyze Statements * @@ -4004,7 +3992,7 @@ typedef struct VacuumStmt NodeTag type; List *options; /* list of DefElem nodes */ List *rels; /* list of VacuumRelation, or NIL for all */ - bool is_vacuumcmd; /* true for VACUUM, false for ANALYZE */ + bool is_vacuumcmd; /* true for VACUUM, false otherwise */ } VacuumStmt; =20 /* @@ -4022,6 +4010,27 @@ typedef struct VacuumRelation List *va_cols; /* list of column names, or NIL for all */ } VacuumRelation; =20 +/* ---------------------- + * Repack Statement + * ---------------------- + */ +typedef enum RepackCommand +{ + REPACK_COMMAND_CLUSTER =3D 1, + REPACK_COMMAND_REPACK, + REPACK_COMMAND_VACUUMFULL, +} RepackCommand; + +typedef struct RepackStmt +{ + NodeTag type; + RepackCommand command; /* type of command being run */ + VacuumRelation *relation; /* relation being repacked */ + char *indexname; /* order tuples by this index */ + bool usingindex; /* whether USING INDEX is specified */ + List *params; /* list of DefElem nodes */ +} RepackStmt; + /* ---------------------- * Explain Statement * diff --git a/src/include/parser/kwlist.h b/src/include/parser/kwlist.h index f7753c5c8a8..6f74a8c05c7 100644 --- a/src/include/parser/kwlist.h +++ b/src/include/parser/kwlist.h @@ -377,6 +377,7 @@ PG_KEYWORD("reindex", REINDEX, UNRESERVED_KEYWORD, BARE= _LABEL) PG_KEYWORD("relative", RELATIVE_P, UNRESERVED_KEYWORD, BARE_LABEL) PG_KEYWORD("release", RELEASE, UNRESERVED_KEYWORD, BARE_LABEL) PG_KEYWORD("rename", RENAME, UNRESERVED_KEYWORD, BARE_LABEL) +PG_KEYWORD("repack", REPACK, UNRESERVED_KEYWORD, BARE_LABEL) PG_KEYWORD("repeatable", REPEATABLE, UNRESERVED_KEYWORD, BARE_LABEL) PG_KEYWORD("replace", REPLACE, UNRESERVED_KEYWORD, BARE_LABEL) PG_KEYWORD("replica", REPLICA, UNRESERVED_KEYWORD, BARE_LABEL) diff --git a/src/include/tcop/cmdtaglist.h b/src/include/tcop/cmdtaglist.h index 1290c9bab68..652dc61b834 100644 --- a/src/include/tcop/cmdtaglist.h +++ b/src/include/tcop/cmdtaglist.h @@ -196,6 +196,7 @@ PG_CMDTAG(CMDTAG_REASSIGN_OWNED, "REASSIGN OWNED", fals= e, false, false) PG_CMDTAG(CMDTAG_REFRESH_MATERIALIZED_VIEW, "REFRESH MATERIALIZED VIEW", t= rue, false, false) PG_CMDTAG(CMDTAG_REINDEX, "REINDEX", true, false, false) PG_CMDTAG(CMDTAG_RELEASE, "RELEASE", false, false, false) +PG_CMDTAG(CMDTAG_REPACK, "REPACK", false, false, false) PG_CMDTAG(CMDTAG_RESET, "RESET", false, false, false) PG_CMDTAG(CMDTAG_REVOKE, "REVOKE", true, false, false) PG_CMDTAG(CMDTAG_REVOKE_ROLE, "REVOKE ROLE", false, false, false) diff --git a/src/include/utils/backend_progress.h b/src/include/utils/backe= nd_progress.h index 19f63b41431..6300dbd15d5 100644 --- a/src/include/utils/backend_progress.h +++ b/src/include/utils/backend_progress.h @@ -24,10 +24,10 @@ typedef enum ProgressCommandType PROGRESS_COMMAND_INVALID, PROGRESS_COMMAND_VACUUM, PROGRESS_COMMAND_ANALYZE, - PROGRESS_COMMAND_CLUSTER, PROGRESS_COMMAND_CREATE_INDEX, PROGRESS_COMMAND_BASEBACKUP, PROGRESS_COMMAND_COPY, + PROGRESS_COMMAND_REPACK, } ProgressCommandType; =20 #define PGSTAT_NUM_PROGRESS_PARAM 20 diff --git a/src/test/regress/expected/cluster.out b/src/test/regress/expec= ted/cluster.out index 4d40a6809ab..277854418fa 100644 --- a/src/test/regress/expected/cluster.out +++ b/src/test/regress/expected/cluster.out @@ -495,6 +495,43 @@ ALTER TABLE clstrpart SET WITHOUT CLUSTER; ERROR: cannot mark index clustered in partitioned table ALTER TABLE clstrpart CLUSTER ON clstrpart_idx; ERROR: cannot mark index clustered in partitioned table +-- Check that REPACK sets new relfilenodes: it should process exactly the = same +-- tables as CLUSTER did. +DROP TABLE old_cluster_info; +DROP TABLE new_cluster_info; +CREATE TEMP TABLE old_cluster_info AS SELECT relname, level, relfilenode, = relkind FROM pg_partition_tree('clstrpart'::regclass) AS tree JOIN pg_class= c ON c.oid=3Dtree.relid ; +REPACK clstrpart USING INDEX clstrpart_idx; +CREATE TEMP TABLE new_cluster_info AS SELECT relname, level, relfilenode, = relkind FROM pg_partition_tree('clstrpart'::regclass) AS tree JOIN pg_class= c ON c.oid=3Dtree.relid ; +SELECT relname, old.level, old.relkind, old.relfilenode =3D new.relfilenod= e FROM old_cluster_info AS old JOIN new_cluster_info AS new USING (relname)= ORDER BY relname COLLATE "C"; + relname | level | relkind | ?column?=20 +-------------+-------+---------+---------- + clstrpart | 0 | p | t + clstrpart1 | 1 | p | t + clstrpart11 | 2 | r | f + clstrpart12 | 2 | p | t + clstrpart2 | 1 | r | f + clstrpart3 | 1 | p | t + clstrpart33 | 2 | r | f +(7 rows) + +-- And finally the same for REPACK w/o index. +DROP TABLE old_cluster_info; +DROP TABLE new_cluster_info; +CREATE TEMP TABLE old_cluster_info AS SELECT relname, level, relfilenode, = relkind FROM pg_partition_tree('clstrpart'::regclass) AS tree JOIN pg_class= c ON c.oid=3Dtree.relid ; +REPACK clstrpart; +CREATE TEMP TABLE new_cluster_info AS SELECT relname, level, relfilenode, = relkind FROM pg_partition_tree('clstrpart'::regclass) AS tree JOIN pg_class= c ON c.oid=3Dtree.relid ; +SELECT relname, old.level, old.relkind, old.relfilenode =3D new.relfilenod= e FROM old_cluster_info AS old JOIN new_cluster_info AS new USING (relname)= ORDER BY relname COLLATE "C"; + relname | level | relkind | ?column?=20 +-------------+-------+---------+---------- + clstrpart | 0 | p | t + clstrpart1 | 1 | p | t + clstrpart11 | 2 | r | f + clstrpart12 | 2 | p | t + clstrpart2 | 1 | r | f + clstrpart3 | 1 | p | t + clstrpart33 | 2 | r | f +(7 rows) + DROP TABLE clstrpart; -- Ownership of partitions is checked CREATE TABLE ptnowner(i int unique) PARTITION BY LIST (i); @@ -513,7 +550,7 @@ CREATE TEMP TABLE ptnowner_oldnodes AS JOIN pg_class AS c ON c.oid=3Dtree.relid; SET SESSION AUTHORIZATION regress_ptnowner; CLUSTER ptnowner USING ptnowner_i_idx; -WARNING: permission denied to cluster "ptnowner2", skipping it +WARNING: permission denied to execute CLUSTER on "ptnowner2", skipping it RESET SESSION AUTHORIZATION; SELECT a.relname, a.relfilenode=3Db.relfilenode FROM pg_class a JOIN ptnowner_oldnodes b USING (oid) ORDER BY a.relname COLLATE "C"; @@ -665,6 +702,101 @@ SELECT * FROM clstr_expression WHERE -a =3D -3 ORDER = BY -a, b; (4 rows) =20 COMMIT; +---------------------------------------------------------------------- +-- +-- REPACK +-- +---------------------------------------------------------------------- +-- REPACK handles individual tables identically to CLUSTER, but it's worth +-- checking if it handles table hierarchies identically as well. +REPACK clstr_tst USING INDEX clstr_tst_c; +-- Verify that inheritance link still works +INSERT INTO clstr_tst_inh VALUES (0, 100, 'in child table 2'); +SELECT a,b,c,substring(d for 30), length(d) from clstr_tst; + a | b | c | substring | length=20 +----+-----+------------------+--------------------------------+-------- + 10 | 14 | catorce | |=20=20=20= =20=20=20=20 + 18 | 5 | cinco | |=20=20=20= =20=20=20=20 + 9 | 4 | cuatro | |=20=20=20= =20=20=20=20 + 26 | 19 | diecinueve | |=20=20=20= =20=20=20=20 + 12 | 18 | dieciocho | |=20=20=20= =20=20=20=20 + 30 | 16 | dieciseis | |=20=20=20= =20=20=20=20 + 24 | 17 | diecisiete | |=20=20=20= =20=20=20=20 + 2 | 10 | diez | |=20=20=20= =20=20=20=20 + 23 | 12 | doce | |=20=20=20= =20=20=20=20 + 11 | 2 | dos | |=20=20=20= =20=20=20=20 + 25 | 9 | nueve | |=20=20=20= =20=20=20=20 + 31 | 8 | ocho | |=20=20=20= =20=20=20=20 + 1 | 11 | once | |=20=20=20= =20=20=20=20 + 28 | 15 | quince | |=20=20=20= =20=20=20=20 + 32 | 6 | seis | xyzzyxyzzyxyzzyxyzzyxyzzyxyzzy | 500000 + 29 | 7 | siete | |=20=20=20= =20=20=20=20 + 15 | 13 | trece | |=20=20=20= =20=20=20=20 + 22 | 30 | treinta | |=20=20=20= =20=20=20=20 + 17 | 32 | treinta y dos | |=20=20=20= =20=20=20=20 + 3 | 31 | treinta y uno | |=20=20=20= =20=20=20=20 + 5 | 3 | tres | |=20=20=20= =20=20=20=20 + 20 | 1 | uno | |=20=20=20= =20=20=20=20 + 6 | 20 | veinte | |=20=20=20= =20=20=20=20 + 14 | 25 | veinticinco | |=20=20=20= =20=20=20=20 + 21 | 24 | veinticuatro | |=20=20=20= =20=20=20=20 + 4 | 22 | veintidos | |=20=20=20= =20=20=20=20 + 19 | 29 | veintinueve | |=20=20=20= =20=20=20=20 + 16 | 28 | veintiocho | |=20=20=20= =20=20=20=20 + 27 | 26 | veintiseis | |=20=20=20= =20=20=20=20 + 13 | 27 | veintisiete | |=20=20=20= =20=20=20=20 + 7 | 23 | veintitres | |=20=20=20= =20=20=20=20 + 8 | 21 | veintiuno | |=20=20=20= =20=20=20=20 + 0 | 100 | in child table | |=20=20=20= =20=20=20=20 + 0 | 100 | in child table 2 | |=20=20=20= =20=20=20=20 +(34 rows) + +-- Verify that foreign key link still works +INSERT INTO clstr_tst (b, c) VALUES (1111, 'this should fail'); +ERROR: insert or update on table "clstr_tst" violates foreign key constra= int "clstr_tst_con" +DETAIL: Key (b)=3D(1111) is not present in table "clstr_tst_s". +SELECT conname FROM pg_constraint WHERE conrelid =3D 'clstr_tst'::regclass +ORDER BY 1; + conname=20=20=20=20=20=20=20=20 +---------------------- + clstr_tst_a_not_null + clstr_tst_con + clstr_tst_pkey +(3 rows) + +-- Verify partial analyze works +REPACK (ANALYZE) clstr_tst (a); +REPACK (ANALYZE) clstr_tst; +REPACK (VERBOSE) clstr_tst (a); +ERROR: ANALYZE option must be specified when a column list is provided +-- REPACK w/o argument performs no ordering, so we can only check which ta= bles +-- have the relfilenode changed. +RESET SESSION AUTHORIZATION; +CREATE TEMP TABLE relnodes_old AS +(SELECT relname, relfilenode +FROM pg_class +WHERE relname IN ('clstr_1', 'clstr_2', 'clstr_3')); +SET SESSION AUTHORIZATION regress_clstr_user; +SET client_min_messages =3D ERROR; -- order of "skipping" warnings may va= ry +REPACK; +RESET client_min_messages; +RESET SESSION AUTHORIZATION; +CREATE TEMP TABLE relnodes_new AS +(SELECT relname, relfilenode +FROM pg_class +WHERE relname IN ('clstr_1', 'clstr_2', 'clstr_3')); +-- Do the actual comparison. Unlike CLUSTER, clstr_3 should have been +-- processed because there is nothing like clustering index here. +SELECT o.relname FROM relnodes_old o +JOIN relnodes_new n ON o.relname =3D n.relname +WHERE o.relfilenode <> n.relfilenode +ORDER BY o.relname; + relname=20 +--------- + clstr_1 + clstr_3 +(2 rows) + -- clean up DROP TABLE clustertest; DROP TABLE clstr_1; diff --git a/src/test/regress/expected/rules.out b/src/test/regress/expecte= d/rules.out index f4ee2bd7459..48461550636 100644 --- a/src/test/regress/expected/rules.out +++ b/src/test/regress/expected/rules.out @@ -2002,34 +2002,23 @@ pg_stat_progress_basebackup| SELECT pid, ELSE NULL::text END AS backup_type FROM pg_stat_get_progress_info('BASEBACKUP'::text) s(pid, datid, relid,= param1, param2, param3, param4, param5, param6, param7, param8, param9, pa= ram10, param11, param12, param13, param14, param15, param16, param17, param= 18, param19, param20); -pg_stat_progress_cluster| SELECT s.pid, - s.datid, - d.datname, - s.relid, - CASE s.param1 - WHEN 1 THEN 'CLUSTER'::text - WHEN 2 THEN 'VACUUM FULL'::text - ELSE NULL::text +pg_stat_progress_cluster| SELECT pid, + datid, + datname, + relid, + CASE + WHEN (command =3D ANY (ARRAY['CLUSTER'::text, 'VACUUM FULL'::t= ext])) THEN command + WHEN (repack_index_relid =3D (0)::oid) THEN 'VACUUM FULL'::text + ELSE 'CLUSTER'::text END AS command, - CASE s.param2 - WHEN 0 THEN 'initializing'::text - WHEN 1 THEN 'seq scanning heap'::text - WHEN 2 THEN 'index scanning heap'::text - WHEN 3 THEN 'sorting tuples'::text - WHEN 4 THEN 'writing new heap'::text - WHEN 5 THEN 'swapping relation files'::text - WHEN 6 THEN 'rebuilding index'::text - WHEN 7 THEN 'performing final cleanup'::text - ELSE NULL::text - END AS phase, - (s.param3)::oid AS cluster_index_relid, - s.param4 AS heap_tuples_scanned, - s.param5 AS heap_tuples_written, - s.param6 AS heap_blks_total, - s.param7 AS heap_blks_scanned, - s.param8 AS index_rebuild_count - FROM (pg_stat_get_progress_info('CLUSTER'::text) s(pid, datid, relid, p= aram1, param2, param3, param4, param5, param6, param7, param8, param9, para= m10, param11, param12, param13, param14, param15, param16, param17, param18= , param19, param20) - LEFT JOIN pg_database d ON ((s.datid =3D d.oid))); + phase, + repack_index_relid AS cluster_index_relid, + heap_tuples_scanned, + heap_tuples_written, + heap_blks_total, + heap_blks_scanned, + index_rebuild_count + FROM pg_stat_progress_repack; pg_stat_progress_copy| SELECT s.pid, s.datid, d.datname, @@ -2089,6 +2078,35 @@ pg_stat_progress_create_index| SELECT s.pid, s.param15 AS partitions_done FROM (pg_stat_get_progress_info('CREATE INDEX'::text) s(pid, datid, rel= id, param1, param2, param3, param4, param5, param6, param7, param8, param9,= param10, param11, param12, param13, param14, param15, param16, param17, pa= ram18, param19, param20) LEFT JOIN pg_database d ON ((s.datid =3D d.oid))); +pg_stat_progress_repack| SELECT s.pid, + s.datid, + d.datname, + s.relid, + CASE s.param1 + WHEN 1 THEN 'CLUSTER'::text + WHEN 2 THEN 'REPACK'::text + WHEN 3 THEN 'VACUUM FULL'::text + ELSE NULL::text + END AS command, + CASE s.param2 + WHEN 0 THEN 'initializing'::text + WHEN 1 THEN 'seq scanning heap'::text + WHEN 2 THEN 'index scanning heap'::text + WHEN 3 THEN 'sorting tuples'::text + WHEN 4 THEN 'writing new heap'::text + WHEN 5 THEN 'swapping relation files'::text + WHEN 6 THEN 'rebuilding index'::text + WHEN 7 THEN 'performing final cleanup'::text + ELSE NULL::text + END AS phase, + (s.param3)::oid AS repack_index_relid, + s.param4 AS heap_tuples_scanned, + s.param5 AS heap_tuples_written, + s.param6 AS heap_blks_total, + s.param7 AS heap_blks_scanned, + s.param8 AS index_rebuild_count + FROM (pg_stat_get_progress_info('REPACK'::text) s(pid, datid, relid, pa= ram1, param2, param3, param4, param5, param6, param7, param8, param9, param= 10, param11, param12, param13, param14, param15, param16, param17, param18,= param19, param20) + LEFT JOIN pg_database d ON ((s.datid =3D d.oid))); pg_stat_progress_vacuum| SELECT s.pid, s.datid, d.datname, diff --git a/src/test/regress/sql/cluster.sql b/src/test/regress/sql/cluste= r.sql index b7115f86104..c976823a3cb 100644 --- a/src/test/regress/sql/cluster.sql +++ b/src/test/regress/sql/cluster.sql @@ -76,7 +76,6 @@ INSERT INTO clstr_tst (b, c) VALUES (1111, 'this should f= ail'); SELECT conname FROM pg_constraint WHERE conrelid =3D 'clstr_tst'::regclass ORDER BY 1; =20 - SELECT relname, relkind, EXISTS(SELECT 1 FROM pg_class WHERE oid =3D c.reltoastrelid) AS hastoa= st FROM pg_class c WHERE relname LIKE 'clstr_tst%' ORDER BY relname; @@ -229,6 +228,24 @@ SELECT relname, old.level, old.relkind, old.relfilenod= e =3D new.relfilenode FROM o CLUSTER clstrpart; ALTER TABLE clstrpart SET WITHOUT CLUSTER; ALTER TABLE clstrpart CLUSTER ON clstrpart_idx; + +-- Check that REPACK sets new relfilenodes: it should process exactly the = same +-- tables as CLUSTER did. +DROP TABLE old_cluster_info; +DROP TABLE new_cluster_info; +CREATE TEMP TABLE old_cluster_info AS SELECT relname, level, relfilenode, = relkind FROM pg_partition_tree('clstrpart'::regclass) AS tree JOIN pg_class= c ON c.oid=3Dtree.relid ; +REPACK clstrpart USING INDEX clstrpart_idx; +CREATE TEMP TABLE new_cluster_info AS SELECT relname, level, relfilenode, = relkind FROM pg_partition_tree('clstrpart'::regclass) AS tree JOIN pg_class= c ON c.oid=3Dtree.relid ; +SELECT relname, old.level, old.relkind, old.relfilenode =3D new.relfilenod= e FROM old_cluster_info AS old JOIN new_cluster_info AS new USING (relname)= ORDER BY relname COLLATE "C"; + +-- And finally the same for REPACK w/o index. +DROP TABLE old_cluster_info; +DROP TABLE new_cluster_info; +CREATE TEMP TABLE old_cluster_info AS SELECT relname, level, relfilenode, = relkind FROM pg_partition_tree('clstrpart'::regclass) AS tree JOIN pg_class= c ON c.oid=3Dtree.relid ; +REPACK clstrpart; +CREATE TEMP TABLE new_cluster_info AS SELECT relname, level, relfilenode, = relkind FROM pg_partition_tree('clstrpart'::regclass) AS tree JOIN pg_class= c ON c.oid=3Dtree.relid ; +SELECT relname, old.level, old.relkind, old.relfilenode =3D new.relfilenod= e FROM old_cluster_info AS old JOIN new_cluster_info AS new USING (relname)= ORDER BY relname COLLATE "C"; + DROP TABLE clstrpart; =20 -- Ownership of partitions is checked @@ -313,6 +330,57 @@ EXPLAIN (COSTS OFF) SELECT * FROM clstr_expression WHE= RE -a =3D -3 ORDER BY -a, b; SELECT * FROM clstr_expression WHERE -a =3D -3 ORDER BY -a, b; COMMIT; =20 +---------------------------------------------------------------------- +-- +-- REPACK +-- +---------------------------------------------------------------------- + +-- REPACK handles individual tables identically to CLUSTER, but it's worth +-- checking if it handles table hierarchies identically as well. +REPACK clstr_tst USING INDEX clstr_tst_c; + +-- Verify that inheritance link still works +INSERT INTO clstr_tst_inh VALUES (0, 100, 'in child table 2'); +SELECT a,b,c,substring(d for 30), length(d) from clstr_tst; + +-- Verify that foreign key link still works +INSERT INTO clstr_tst (b, c) VALUES (1111, 'this should fail'); + +SELECT conname FROM pg_constraint WHERE conrelid =3D 'clstr_tst'::regclass +ORDER BY 1; + +-- Verify partial analyze works +REPACK (ANALYZE) clstr_tst (a); +REPACK (ANALYZE) clstr_tst; +REPACK (VERBOSE) clstr_tst (a); + +-- REPACK w/o argument performs no ordering, so we can only check which ta= bles +-- have the relfilenode changed. +RESET SESSION AUTHORIZATION; +CREATE TEMP TABLE relnodes_old AS +(SELECT relname, relfilenode +FROM pg_class +WHERE relname IN ('clstr_1', 'clstr_2', 'clstr_3')); + +SET SESSION AUTHORIZATION regress_clstr_user; +SET client_min_messages =3D ERROR; -- order of "skipping" warnings may va= ry +REPACK; +RESET client_min_messages; + +RESET SESSION AUTHORIZATION; +CREATE TEMP TABLE relnodes_new AS +(SELECT relname, relfilenode +FROM pg_class +WHERE relname IN ('clstr_1', 'clstr_2', 'clstr_3')); + +-- Do the actual comparison. Unlike CLUSTER, clstr_3 should have been +-- processed because there is nothing like clustering index here. +SELECT o.relname FROM relnodes_old o +JOIN relnodes_new n ON o.relname =3D n.relname +WHERE o.relfilenode <> n.relfilenode +ORDER BY o.relname; + -- clean up DROP TABLE clustertest; DROP TABLE clstr_1; diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs= .list index ddbe4c64971..c0a4c9ed1c4 100644 --- a/src/tools/pgindent/typedefs.list +++ b/src/tools/pgindent/typedefs.list @@ -2576,6 +2576,8 @@ ReorderBufferTupleCidKey ReorderBufferUpdateProgressTxnCB ReorderTuple RepOriginId +RepackCommand +RepackStmt ReparameterizeForeignPathByChild_function ReplaceVarsFromTargetList_context ReplaceVarsNoMatchOption --=20 2.47.3 --=-=-= Content-Type: text/x-diff Content-Disposition: attachment; filename=v32-0002-Refactor-index_concurrently_create_copy-for-use-with.patch ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
* [PATCH v56 1/3] Allow logical replication snapshots to be database-specific @ 2026-04-06 22:07 Álvaro Herrera <[email protected]> 0 siblings, 0 replies; 90+ messages in thread From: Álvaro Herrera @ 2026-04-06 22:07 UTC (permalink / raw) By default, the logical decoding assumes access to shared catalogs, so the snapshot builder needs to consider cluster-wide XIDs during startup. That in turn means that, if any transaction is already running (and has XID assigned), the snapshot builder needs to wait for its completion, as it does not know if that transaction performed catalog changes earlier. A possible problem with this concept is that if REPACK (CONCURRENTLY) is running in some database, backends running the same command in other databases get stuck until the first one has committed. Thus only a single backend in the cluster can run REPACK (CONCURRENTLY) at any time. Likewise, REPACK (CONCURRENTLY) can block walsenders starting on behalf of subscriptions throughout the cluster. This patch adds a new option to logical replication output plugin, to declare that it does not use shared catalogs (i.e. catalogs that can be changed by transactions running in other databases in the cluster). In that case, no snapshot the backend will use during the decoding needs to contain information about transactions running in other databases. Thus the snapshot builder only needs to wait for completion of transactions in the current database. Currently we only use this option in the REPACK background worker. It could possibly be used in the plugin for logical replication too, however that would need thorough analysis of that plugin. Bump WAL version number, due to a new field in xl_running_xacts. Author: Antonin Houska <[email protected]> Reviewed-by: Álvaro Herrera <[email protected]> Discussion: https://postgr.es/m/90475.1775218118@localhost --- contrib/pg_visibility/pg_visibility.c | 4 +- doc/src/sgml/logicaldecoding.sgml | 4 ++ src/backend/access/index/genam.c | 8 +++ src/backend/access/rmgrdesc/standbydesc.c | 2 + src/backend/access/transam/xlog.c | 2 +- src/backend/access/transam/xlogfuncs.c | 2 +- src/backend/postmaster/bgwriter.c | 2 +- src/backend/replication/logical/decode.c | 17 +++++- src/backend/replication/logical/logical.c | 3 + src/backend/replication/logical/snapbuild.c | 63 ++++++++++++++++++++- src/backend/replication/pgrepack/pgrepack.c | 7 +++ src/backend/replication/slot.c | 2 +- src/backend/storage/ipc/procarray.c | 23 +++++++- src/backend/storage/ipc/standby.c | 24 +++++++- src/include/access/genam.h | 1 + src/include/access/xlog_internal.h | 2 +- src/include/miscadmin.h | 9 +++ src/include/replication/output_plugin.h | 1 + src/include/replication/snapbuild.h | 3 +- src/include/storage/procarray.h | 2 +- src/include/storage/standby.h | 3 +- src/include/storage/standbydefs.h | 1 + 22 files changed, 167 insertions(+), 18 deletions(-) diff --git a/contrib/pg_visibility/pg_visibility.c b/contrib/pg_visibility/pg_visibility.c index dfab0b64cf5..d564bd2a00c 100644 --- a/contrib/pg_visibility/pg_visibility.c +++ b/contrib/pg_visibility/pg_visibility.c @@ -621,7 +621,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) else if (rel == NULL || rel->rd_rel->relisshared) { /* Shared relation: take into account all running xids */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestRunningXid; @@ -632,7 +632,7 @@ GetStrictOldestNonRemovableTransactionId(Relation rel) * Normal relation: take into account xids running within the current * database */ - runningTransactions = GetRunningTransactionData(); + runningTransactions = GetRunningTransactionData(InvalidOid); LWLockRelease(ProcArrayLock); LWLockRelease(XidGenLock); return runningTransactions->oldestDatabaseRunningXid; diff --git a/doc/src/sgml/logicaldecoding.sgml b/doc/src/sgml/logicaldecoding.sgml index 6dc49108997..9b1d68d0de6 100644 --- a/doc/src/sgml/logicaldecoding.sgml +++ b/doc/src/sgml/logicaldecoding.sgml @@ -959,6 +959,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; </programlisting> <literal>output_type</literal> has to either be set to @@ -969,6 +970,9 @@ typedef struct OutputPluginOptions also be called for changes made by heap rewrites during certain DDL operations. These are of interest to plugins that handle DDL replication, but they require special handling. + <literal>need_shared_catalogs</literal> can be set to false if you are + certain the plugin functions do not access shared system catalogs. + Doing so can speed up creation of replication slots that use this plugin. </para> <para> diff --git a/src/backend/access/index/genam.c b/src/backend/access/index/genam.c index 1408989c568..97d44b84622 100644 --- a/src/backend/access/index/genam.c +++ b/src/backend/access/index/genam.c @@ -394,6 +394,14 @@ systable_beginscan(Relation heapRelation, SysScanDesc sysscan; Relation irel; + /* + * If this backend promised that it won't access shared catalogs during + * logical decoding, this it the right place to verify. + */ + Assert(!HistoricSnapshotActive() || + accessSharedCatalogsInDecoding || + !heapRelation->rd_rel->relisshared); + if (indexOK && !IgnoreSystemIndexes && !ReindexIsProcessingIndex(indexId)) diff --git a/src/backend/access/rmgrdesc/standbydesc.c b/src/backend/access/rmgrdesc/standbydesc.c index 0a291354ae2..685d1bdb024 100644 --- a/src/backend/access/rmgrdesc/standbydesc.c +++ b/src/backend/access/rmgrdesc/standbydesc.c @@ -41,6 +41,8 @@ standby_desc_running_xacts(StringInfo buf, xl_running_xacts *xlrec) for (i = 0; i < xlrec->subxcnt; i++) appendStringInfo(buf, " %u", xlrec->xids[xlrec->xcnt + i]); } + + appendStringInfo(buf, "; dbid: %u", xlrec->dbid); } void diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c index b82af9a85c0..3f08a832ca6 100644 --- a/src/backend/access/transam/xlog.c +++ b/src/backend/access/transam/xlog.c @@ -7735,7 +7735,7 @@ CreateCheckPoint(int flags) * recovery we don't need to write running xact data. */ if (!shutdown && XLogStandbyInfoActive()) - LogStandbySnapshot(); + LogStandbySnapshot(InvalidOid); START_CRIT_SECTION(); diff --git a/src/backend/access/transam/xlogfuncs.c b/src/backend/access/transam/xlogfuncs.c index 65bbaeda59c..0f5979691e6 100644 --- a/src/backend/access/transam/xlogfuncs.c +++ b/src/backend/access/transam/xlogfuncs.c @@ -245,7 +245,7 @@ pg_log_standby_snapshot(PG_FUNCTION_ARGS) (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), errmsg("pg_log_standby_snapshot() can only be used if \"wal_level\" >= \"replica\""))); - recptr = LogStandbySnapshot(); + recptr = LogStandbySnapshot(InvalidOid); /* * As a convenience, return the WAL location of the last inserted record diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c index 1d8947774a9..a30de4262eb 100644 --- a/src/backend/postmaster/bgwriter.c +++ b/src/backend/postmaster/bgwriter.c @@ -289,7 +289,7 @@ BackgroundWriterMain(const void *startup_data, size_t startup_data_len) if (now >= timeout && last_snapshot_lsn <= GetLastImportantRecPtr()) { - last_snapshot_lsn = LogStandbySnapshot(); + last_snapshot_lsn = LogStandbySnapshot(InvalidOid); last_snapshot_ts = now; } } diff --git a/src/backend/replication/logical/decode.c b/src/backend/replication/logical/decode.c index c9fea8cad28..38c5a4f5540 100644 --- a/src/backend/replication/logical/decode.c +++ b/src/backend/replication/logical/decode.c @@ -382,7 +382,16 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) { xl_running_xacts *running = (xl_running_xacts *) XLogRecGetData(r); - SnapBuildProcessRunningXacts(builder, buf->origptr, running); + /* + * Update this decoder's idea of transactions currently + * running. In doing so we will determine whether we have + * reached consistent status. + * + * If the output plugin doesn't need access to shared + * catalogs, we can ignore transactions in other databases. + */ + SnapBuildProcessRunningXacts(builder, buf->origptr, running, + !ctx->options.need_shared_catalogs); /* * Abort all transactions that we keep track of, that are @@ -392,8 +401,12 @@ standby_decode(LogicalDecodingContext *ctx, XLogRecordBuffer *buf) * all running transactions which includes prepared ones, * while shutdown checkpoints just know that no non-prepared * transactions are in progress. + * + * The database-specific records might work here too, but it's + * not their purpose. */ - ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); + if (!OidIsValid(running->dbid)) + ReorderBufferAbortOld(ctx->reorder, running->oldestRunningXid); } break; case XLOG_STANDBY_LOCK: diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c index f20d0c542f3..8ceaf64d164 100644 --- a/src/backend/replication/logical/logical.c +++ b/src/backend/replication/logical/logical.c @@ -285,6 +285,9 @@ StartupDecodingContext(List *output_plugin_options, ctx->write = do_write; ctx->update_progress = update_progress; + /* Assume shared catalog access. The startup callback can change it. */ + ctx->options.need_shared_catalogs = true; + ctx->output_plugin_options = output_plugin_options; ctx->fast_forward = fast_forward; diff --git a/src/backend/replication/logical/snapbuild.c b/src/backend/replication/logical/snapbuild.c index b4269a3b102..ffb4ab2cf2a 100644 --- a/src/backend/replication/logical/snapbuild.c +++ b/src/backend/replication/logical/snapbuild.c @@ -154,6 +154,14 @@ static ResourceOwner SavedResourceOwnerDuringExport = NULL; static bool ExportInProgress = false; +/* + * If a backend is going to do logical decoding and the output plugin does + * not need to access shared catalogs, setting this variable to false can make + * the decoding startup faster. In particular, the backend will not need to + * wait for completion of already running transactions in other databases. + */ +bool accessSharedCatalogsInDecoding = true; + /* ->committed and ->catchange manipulation */ static void SnapBuildPurgeOlderTxn(SnapBuild *builder); @@ -170,7 +178,8 @@ static inline bool SnapBuildXidHasCatalogChanges(SnapBuild *builder, Transaction uint32 xinfo); /* xlog reading helper functions for SnapBuildProcessRunningXacts */ -static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running); +static bool SnapBuildFindSnapshot(SnapBuild *builder, XLogRecPtr lsn, + xl_running_xacts *running); static void SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff); /* serialization functions */ @@ -226,6 +235,9 @@ AllocateSnapshotBuilder(ReorderBuffer *reorder, MemoryContextSwitchTo(oldcontext); + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + return builder; } @@ -244,6 +256,9 @@ FreeSnapshotBuilder(SnapBuild *builder) builder->snapshot = NULL; } + /* The default is that shared catalog are used. */ + accessSharedCatalogsInDecoding = true; + /* other resources are deallocated via memory context reset */ MemoryContextDelete(context); } @@ -1136,7 +1151,8 @@ SnapBuildXidHasCatalogChanges(SnapBuild *builder, TransactionId xid, * anymore. */ void -SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running) +SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xacts *running, + bool db_specific) { ReorderBufferTXN *txn; TransactionId xmin; @@ -1148,6 +1164,33 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact */ if (builder->state < SNAPBUILD_CONSISTENT) { + /* + * To reduce the potential for unnecessarily waiting for completion of + * unrelated transactions, the caller can declare that only + * transactions of the current database are relevant at this stage. + */ + if (db_specific) + { + /* + * If we must only keep track of transactions running in the + * current database, we need transaction info from exactly that + * database. + */ + if (running->dbid != MyDatabaseId) + { + LogStandbySnapshot(MyDatabaseId); + + return; + } + + /* + * We'd better be able to check during scan if the plugin does not + * lie. + */ + if (accessSharedCatalogsInDecoding) + accessSharedCatalogsInDecoding = false; + } + /* returns false if there's no point in performing cleanup just yet */ if (!SnapBuildFindSnapshot(builder, lsn, running)) return; @@ -1155,6 +1198,16 @@ SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, xl_running_xact else SnapBuildSerialize(builder, lsn); + /* + * Database specific transaction info may exist to reach CONSISTENT state + * faster, however the code below makes no use of it. Moreover, such + * record might cause problems because the following normal (cluster-wide) + * record can have lower value of oldestRunningXid. In that case, let's + * wait with the cleanup for the next regular cluster-wide record. + */ + if (OidIsValid(running->dbid)) + return; + /* * Update range of interesting xids based on the running xacts * information. We don't increase ->xmax using it, because once we are in @@ -1465,7 +1518,11 @@ SnapBuildWaitSnapshot(xl_running_xacts *running, TransactionId cutoff) */ if (!RecoveryInProgress()) { - LogStandbySnapshot(); + /* + * If the last transaction info was about specific database, so needs + * to be the next one - at least until we're in the CONSISTENT state. + */ + LogStandbySnapshot(running->dbid); } } diff --git a/src/backend/replication/pgrepack/pgrepack.c b/src/backend/replication/pgrepack/pgrepack.c index 457b6518a8e..4da66aa0f44 100644 --- a/src/backend/replication/pgrepack/pgrepack.c +++ b/src/backend/replication/pgrepack/pgrepack.c @@ -52,6 +52,13 @@ repack_startup(LogicalDecodingContext *ctx, OutputPluginOptions *opt, /* Probably unnecessary, as we don't use the SQL interface ... */ opt->output_type = OUTPUT_PLUGIN_BINARY_OUTPUT; + /* + * REPACK doesn't need access to shared catalogs, so we can speed up the + * historic snapshot creation by setting this flag. We'll only have to + * wait for transactions in our database. + */ + opt->need_shared_catalogs = false; + if (ctx->output_plugin_options != NIL) { ereport(ERROR, diff --git a/src/backend/replication/slot.c b/src/backend/replication/slot.c index 21a213a0ebf..a1f37e59dbc 100644 --- a/src/backend/replication/slot.c +++ b/src/backend/replication/slot.c @@ -1756,7 +1756,7 @@ ReplicationSlotReserveWal(void) XLogRecPtr flushptr; /* make sure we have enough information to start */ - flushptr = LogStandbySnapshot(); + flushptr = LogStandbySnapshot(InvalidOid); /* and make sure it's fsynced to disk */ XLogFlush(flushptr); diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c index f540bb6b23f..9299bcebbda 100644 --- a/src/backend/storage/ipc/procarray.c +++ b/src/backend/storage/ipc/procarray.c @@ -2623,9 +2623,11 @@ ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc) * * Note that if any transaction has overflowed its cached subtransactions * then there is no real need include any subtransactions. + * + * If 'dbid' is valid, only gather transactions running in that database. */ RunningTransactions -GetRunningTransactionData(void) +GetRunningTransactionData(Oid dbid) { /* result workspace */ static RunningTransactionsData CurrentRunningXactsData; @@ -2700,6 +2702,18 @@ GetRunningTransactionData(void) if (!TransactionIdIsValid(xid)) continue; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid)) + { + int pgprocno = arrayP->pgprocnos[index]; + PGPROC *proc = &allProcs[pgprocno]; + + if (proc->databaseId != dbid) + continue; + } + /* * Be careful not to exclude any xids before calculating the values of * oldestRunningXid and suboverflowed, since these are used to clean @@ -2750,6 +2764,12 @@ GetRunningTransactionData(void) PGPROC *proc = &allProcs[pgprocno]; int nsubxids; + /* + * Filter by database OID if requested. + */ + if (OidIsValid(dbid) && proc->databaseId != dbid) + continue; + /* * Save subtransaction XIDs. Other backends can't add or remove * entries while we're holding XidGenLock. @@ -2783,6 +2803,7 @@ GetRunningTransactionData(void) * increases if slots do. */ + CurrentRunningXacts->dbid = dbid; CurrentRunningXacts->xcnt = count - subcount; CurrentRunningXacts->subxcnt = subcount; CurrentRunningXacts->subxid_status = suboverflowed ? SUBXIDS_IN_SUBTRANS : SUBXIDS_IN_ARRAY; diff --git a/src/backend/storage/ipc/standby.c b/src/backend/storage/ipc/standby.c index de9092fdf5b..29af7733948 100644 --- a/src/backend/storage/ipc/standby.c +++ b/src/backend/storage/ipc/standby.c @@ -1188,6 +1188,14 @@ standby_redo(XLogReaderState *record) xl_running_xacts *xlrec = (xl_running_xacts *) XLogRecGetData(record); RunningTransactionsData running; + /* + * Records issued for specific database are not suitable for physical + * replication because that affects the whole cluster. In particular, + * the list of XID is probably incomplete here. + */ + if (OidIsValid(xlrec->dbid)) + return; + running.xcnt = xlrec->xcnt; running.subxcnt = xlrec->subxcnt; running.subxid_status = xlrec->subxid_overflow ? SUBXIDS_MISSING : SUBXIDS_IN_ARRAY; @@ -1277,11 +1285,22 @@ standby_redo(XLogReaderState *record) * as there's no independent knob to just enable logical decoding. For * details of how this is used, check snapbuild.c's introductory comment. * + * If 'dbid' is valid, only gather transactions running in that + * database. snapbuild.c can use such running xacts information for faster + * startup, but it still needs normal (cluster-wide) during the actual + * decoding - see standby_decode() and SnapBuildProcessRunningXacts() for + * details. Other processes (e.g. checkpointer) issue the cluster-wide records + * whether logical decoding is active or not. + * + * Please be careful about using this argument for other purposes. In + * particular, physical replication *must* ignore the database-specific + * records, exactly because they do not cover the whole cluster - see + * standby_redo(). * * Returns the RecPtr of the last inserted record. */ XLogRecPtr -LogStandbySnapshot(void) +LogStandbySnapshot(Oid dbid) { XLogRecPtr recptr; RunningTransactions running; @@ -1314,7 +1333,7 @@ LogStandbySnapshot(void) * Log details of all in-progress transactions. This should be the last * record we write, because standby will open up when it sees this. */ - running = GetRunningTransactionData(); + running = GetRunningTransactionData(dbid); /* * GetRunningTransactionData() acquired ProcArrayLock, we must release it. @@ -1358,6 +1377,7 @@ LogCurrentRunningXacts(RunningTransactions CurrRunningXacts) xl_running_xacts xlrec; XLogRecPtr recptr; + xlrec.dbid = CurrRunningXacts->dbid; xlrec.xcnt = CurrRunningXacts->xcnt; xlrec.subxcnt = CurrRunningXacts->subxcnt; xlrec.subxid_overflow = (CurrRunningXacts->subxid_status != SUBXIDS_IN_ARRAY); diff --git a/src/include/access/genam.h b/src/include/access/genam.h index b69320a7fc8..de55821e414 100644 --- a/src/include/access/genam.h +++ b/src/include/access/genam.h @@ -136,6 +136,7 @@ typedef struct IndexOrderByDistance bool isnull; } IndexOrderByDistance; + /* * generalized index_ interface routines (in indexam.c) */ diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h index 10c18d39ff8..13ae3ad4fbb 100644 --- a/src/include/access/xlog_internal.h +++ b/src/include/access/xlog_internal.h @@ -32,7 +32,7 @@ /* * Each page of XLOG file has a header like this: */ -#define XLOG_PAGE_MAGIC 0xD11E /* can be used as WAL version indicator */ +#define XLOG_PAGE_MAGIC 0xD11F /* can be used as WAL version indicator */ typedef struct XLogPageHeaderData { diff --git a/src/include/miscadmin.h b/src/include/miscadmin.h index 2e10e3c814d..93b7816c09c 100644 --- a/src/include/miscadmin.h +++ b/src/include/miscadmin.h @@ -310,6 +310,15 @@ extern void PreventCommandIfReadOnly(const char *cmdname); extern void PreventCommandIfParallelMode(const char *cmdname); extern void PreventCommandDuringRecovery(const char *cmdname); +/* in replication/snapbuild.c */ + +/* + * Keep track of whether logical decoding in this backend promised not to + * access shared catalogs, as a safety check. This is checked by genam.c when + * a catalog scan takes place to verify that no shared catalogs are accessed. + */ +extern bool accessSharedCatalogsInDecoding; + /***************************************************************************** * pdir.h -- * * POSTGRES directory path definitions. * diff --git a/src/include/replication/output_plugin.h b/src/include/replication/output_plugin.h index 842fcde67f9..917f3cff232 100644 --- a/src/include/replication/output_plugin.h +++ b/src/include/replication/output_plugin.h @@ -27,6 +27,7 @@ typedef struct OutputPluginOptions { OutputPluginOutputType output_type; bool receive_rewrites; + bool need_shared_catalogs; } OutputPluginOptions; /* diff --git a/src/include/replication/snapbuild.h b/src/include/replication/snapbuild.h index a22a83a2f23..d02530a912a 100644 --- a/src/include/replication/snapbuild.h +++ b/src/include/replication/snapbuild.h @@ -92,7 +92,8 @@ extern void SnapBuildProcessNewCid(SnapBuild *builder, TransactionId xid, XLogRecPtr lsn, xl_heap_new_cid *xlrec); extern void SnapBuildProcessRunningXacts(SnapBuild *builder, XLogRecPtr lsn, - xl_running_xacts *running); + xl_running_xacts *running, + bool db_specific); extern void SnapBuildSerializationPoint(SnapBuild *builder, XLogRecPtr lsn); extern bool SnapBuildSnapshotExists(XLogRecPtr lsn); diff --git a/src/include/storage/procarray.h b/src/include/storage/procarray.h index d718a5b542f..ec89c448220 100644 --- a/src/include/storage/procarray.h +++ b/src/include/storage/procarray.h @@ -47,7 +47,7 @@ extern bool ProcArrayInstallImportedXmin(TransactionId xmin, VirtualTransactionId *sourcevxid); extern bool ProcArrayInstallRestoredXmin(TransactionId xmin, PGPROC *proc); -extern RunningTransactions GetRunningTransactionData(void); +extern RunningTransactions GetRunningTransactionData(Oid dbid); extern bool TransactionIdIsInProgress(TransactionId xid); extern TransactionId GetOldestNonRemovableTransactionId(Relation rel); diff --git a/src/include/storage/standby.h b/src/include/storage/standby.h index 6a314c693cd..8715c08e94f 100644 --- a/src/include/storage/standby.h +++ b/src/include/storage/standby.h @@ -126,6 +126,7 @@ typedef enum typedef struct RunningTransactionsData { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ subxids_array_status subxid_status; @@ -143,7 +144,7 @@ typedef RunningTransactionsData *RunningTransactions; extern void LogAccessExclusiveLock(Oid dbOid, Oid relOid); extern void LogAccessExclusiveLockPrepare(void); -extern XLogRecPtr LogStandbySnapshot(void); +extern XLogRecPtr LogStandbySnapshot(Oid dbid); extern void LogStandbyInvalidations(int nmsgs, SharedInvalidationMessage *msgs, bool relcacheInitFileInval); diff --git a/src/include/storage/standbydefs.h b/src/include/storage/standbydefs.h index 231d251fd51..e75b7078766 100644 --- a/src/include/storage/standbydefs.h +++ b/src/include/storage/standbydefs.h @@ -46,6 +46,7 @@ typedef struct xl_standby_locks */ typedef struct xl_running_xacts { + Oid dbid; /* only track xacts in this database */ int xcnt; /* # of xact ids in xids[] */ int subxcnt; /* # of subxact ids in xids[] */ bool subxid_overflow; /* snapshot overflowed, subxids missing */ -- 2.47.3 --qs77q6fpwolne4ds Content-Type: text/x-diff; charset=utf-8 Content-Disposition: attachment; filename="v56-0002-Reserve-replication-slots-specifically-for-REPAC.patch" ^ permalink raw reply [nested|flat] 90+ messages in thread
end of thread, other threads:[~2026-04-06 22:07 UTC | newest] Thread overview: 90+ messages (download: mbox mbox.gz follow: Atom feed) -- links below jump to the message on this page -- 2022-06-17 01:33 [PATCH v1 2/2] WIP: Fix EEOP_JSON_CONSTRUCTOR and EEOP_JSONEXPR size. Andres Freund <[email protected]> 2026-01-27 10:48 [PATCH 1/6] Add REPACK command Antonin Houska <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[email protected]> 2026-04-06 22:07 [PATCH v56 1/3] Allow logical replication snapshots to be database-specific Álvaro Herrera <[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