public inbox for [email protected]
help / color / mirror / Atom feed[PATCH v4 3/3] Allow recovery via loadable modules.
5+ messages / 3 participants
[nested] [flat]
* [PATCH v4 3/3] Allow recovery via loadable modules.
@ 2022-12-10 03:40 Nathan Bossart <[email protected]>
0 siblings, 0 replies; 5+ messages in thread
From: Nathan Bossart @ 2022-12-10 03:40 UTC (permalink / raw)
This adds the restore_library parameter to allow archive recovery
via a loadable module, rather than running shell commands.
---
contrib/basic_archive/Makefile | 4 +-
contrib/basic_archive/basic_archive.c | 67 ++++++-
contrib/basic_archive/meson.build | 7 +-
contrib/basic_archive/t/001_restore.pl | 44 +++++
doc/src/sgml/archive-modules.sgml | 168 ++++++++++++++++--
doc/src/sgml/backup.sgml | 43 ++++-
doc/src/sgml/basic-archive.sgml | 33 ++--
doc/src/sgml/config.sgml | 54 +++++-
doc/src/sgml/high-availability.sgml | 23 ++-
src/backend/access/transam/shell_restore.c | 21 ++-
src/backend/access/transam/xlog.c | 13 +-
src/backend/access/transam/xlogarchive.c | 70 +++++++-
src/backend/access/transam/xlogrecovery.c | 26 ++-
src/backend/postmaster/checkpointer.c | 26 +++
src/backend/postmaster/pgarch.c | 7 +-
src/backend/postmaster/startup.c | 23 ++-
src/backend/utils/misc/guc.c | 14 ++
src/backend/utils/misc/guc_tables.c | 10 ++
src/backend/utils/misc/postgresql.conf.sample | 1 +
src/include/access/xlog_internal.h | 1 +
src/include/access/xlogarchive.h | 44 ++++-
src/include/access/xlogrecovery.h | 1 +
src/include/utils/guc.h | 2 +
23 files changed, 618 insertions(+), 84 deletions(-)
create mode 100644 contrib/basic_archive/t/001_restore.pl
diff --git a/contrib/basic_archive/Makefile b/contrib/basic_archive/Makefile
index 55d299d650..487dc563f3 100644
--- a/contrib/basic_archive/Makefile
+++ b/contrib/basic_archive/Makefile
@@ -1,7 +1,7 @@
# contrib/basic_archive/Makefile
MODULES = basic_archive
-PGFILEDESC = "basic_archive - basic archive module"
+PGFILEDESC = "basic_archive - basic archive and recovery module"
REGRESS = basic_archive
REGRESS_OPTS = --temp-config $(top_srcdir)/contrib/basic_archive/basic_archive.conf
@@ -9,6 +9,8 @@ REGRESS_OPTS = --temp-config $(top_srcdir)/contrib/basic_archive/basic_archive.c
# which typical installcheck users do not have (e.g. buildfarm clients).
NO_INSTALLCHECK = 1
+TAP_TESTS = 1
+
ifdef USE_PGXS
PG_CONFIG = pg_config
PGXS := $(shell $(PG_CONFIG) --pgxs)
diff --git a/contrib/basic_archive/basic_archive.c b/contrib/basic_archive/basic_archive.c
index 28cbb6cce0..8c333c8f99 100644
--- a/contrib/basic_archive/basic_archive.c
+++ b/contrib/basic_archive/basic_archive.c
@@ -17,6 +17,11 @@
* a file is successfully archived and then the system crashes before
* a durable record of the success has been made.
*
+ * This file also demonstrates a basic restore library implementation that
+ * is roughly equivalent to the following shell command:
+ *
+ * cp /path/to/archivedir/%f %p
+ *
* Copyright (c) 2022-2023, PostgreSQL Global Development Group
*
* IDENTIFICATION
@@ -30,6 +35,7 @@
#include <sys/time.h>
#include <unistd.h>
+#include "access/xlogarchive.h"
#include "common/int.h"
#include "miscadmin.h"
#include "postmaster/pgarch.h"
@@ -48,6 +54,8 @@ static bool basic_archive_file(const char *file, const char *path);
static void basic_archive_file_internal(const char *file, const char *path);
static bool check_archive_directory(char **newval, void **extra, GucSource source);
static bool compare_files(const char *file1, const char *file2);
+static bool basic_restore_file(const char *file, const char *path,
+ const char *lastRestartPointFileName);
/*
* _PG_init
@@ -87,6 +95,19 @@ _PG_archive_module_init(ArchiveModuleCallbacks *cb)
cb->archive_file_cb = basic_archive_file;
}
+/*
+ * _PG_recovery_module_init
+ *
+ * Returns the module's restore callback.
+ */
+void
+_PG_recovery_module_init(RecoveryModuleCallbacks *cb)
+{
+ AssertVariableIsOfType(&_PG_recovery_module_init, RecoveryModuleInit);
+
+ cb->restore_cb = basic_restore_file;
+}
+
/*
* check_archive_directory
*
@@ -99,8 +120,8 @@ check_archive_directory(char **newval, void **extra, GucSource source)
/*
* The default value is an empty string, so we have to accept that value.
- * Our check_configured callback also checks for this and prevents
- * archiving from proceeding if it is still empty.
+ * Our check_configured and restore callbacks also check for this and
+ * prevent archiving or recovery from proceeding if it is still empty.
*/
if (*newval == NULL || *newval[0] == '\0')
return true;
@@ -368,3 +389,45 @@ compare_files(const char *file1, const char *file2)
return ret;
}
+
+/*
+ * basic_restore_file
+ *
+ * Retrieves one file from the WAL archives.
+ */
+static bool
+basic_restore_file(const char *file, const char *path,
+ const char *lastRestartPointFileName)
+{
+ char source[MAXPGPATH];
+ struct stat st;
+
+ ereport(DEBUG1,
+ (errmsg("restoring \"%s\" via basic_archive", file)));
+
+ if (archive_directory == NULL || archive_directory[0] == '\0')
+ ereport(ERROR,
+ (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
+ errmsg("\"basic_archive.archive_directory\" is not set")));
+
+ /*
+ * Check whether the file exists. If not, we return false to indicate that
+ * there are no more files to restore.
+ */
+ snprintf(source, MAXPGPATH, "%s/%s", archive_directory, file);
+ if (stat(source, &st) != 0)
+ {
+ int elevel = (errno == ENOENT) ? DEBUG1 : ERROR;
+
+ ereport(elevel,
+ (errcode_for_file_access(),
+ errmsg("could not stat file \"%s\": %m", source)));
+ return false;
+ }
+
+ copy_file(source, unconstify(char *, path));
+
+ ereport(DEBUG1,
+ (errmsg("restored \"%s\" via basic_archive", file)));
+ return true;
+}
diff --git a/contrib/basic_archive/meson.build b/contrib/basic_archive/meson.build
index bc1380e6f6..af4580dea9 100644
--- a/contrib/basic_archive/meson.build
+++ b/contrib/basic_archive/meson.build
@@ -7,7 +7,7 @@ basic_archive_sources = files(
if host_system == 'windows'
basic_archive_sources += rc_lib_gen.process(win32ver_rc, extra_args: [
'--NAME', 'basic_archive',
- '--FILEDESC', 'basic_archive - basic archive module',])
+ '--FILEDESC', 'basic_archive - basic archive and recovery module',])
endif
basic_archive = shared_module('basic_archive',
@@ -31,4 +31,9 @@ tests += {
# which typical runningcheck users do not have (e.g. buildfarm clients).
'runningcheck': false,
},
+ 'tap': {
+ 'tests': [
+ 't/001_restore.pl',
+ ],
+ },
}
diff --git a/contrib/basic_archive/t/001_restore.pl b/contrib/basic_archive/t/001_restore.pl
new file mode 100644
index 0000000000..ec8767d740
--- /dev/null
+++ b/contrib/basic_archive/t/001_restore.pl
@@ -0,0 +1,44 @@
+
+# Copyright (c) 2022, PostgreSQL Global Development Group
+
+use strict;
+use warnings;
+use PostgreSQL::Test::Cluster;
+use PostgreSQL::Test::Utils;
+use Test::More;
+
+# start a node
+my $node = PostgreSQL::Test::Cluster->new('node');
+$node->init(has_archiving => 1, allows_streaming => 1);
+my $archive_dir = $node->archive_dir;
+$archive_dir =~ s!\\!/!g if $PostgreSQL::Test::Utils::windows_os;
+$node->append_conf('postgresql.conf', "archive_command = ''");
+$node->append_conf('postgresql.conf', "archive_library = 'basic_archive'");
+$node->append_conf('postgresql.conf', "basic_archive.archive_directory = '$archive_dir'");
+$node->start;
+
+# backup the node
+my $backup = 'backup';
+$node->backup($backup);
+
+# generate some new WAL files
+$node->safe_psql('postgres', "CREATE TABLE test (a INT);");
+$node->safe_psql('postgres', "SELECT pg_switch_wal();");
+$node->safe_psql('postgres', "INSERT INTO test VALUES (1);");
+
+# shut down the node (this should archive all WAL files)
+$node->stop;
+
+# restore from the backup
+my $restore = PostgreSQL::Test::Cluster->new('restore');
+$restore->init_from_backup($node, $backup, has_restoring => 1, standby => 0);
+$restore->append_conf('postgresql.conf', "restore_command = ''");
+$restore->append_conf('postgresql.conf', "restore_library = 'basic_archive'");
+$restore->append_conf('postgresql.conf', "basic_archive.archive_directory = '$archive_dir'");
+$restore->start;
+
+# ensure post-backup WAL was replayed
+my $result = $restore->safe_psql("postgres", "SELECT count(*) FROM test;");
+is($result, "1", "check restore content");
+
+done_testing();
diff --git a/doc/src/sgml/archive-modules.sgml b/doc/src/sgml/archive-modules.sgml
index ef02051f7f..53e657040b 100644
--- a/doc/src/sgml/archive-modules.sgml
+++ b/doc/src/sgml/archive-modules.sgml
@@ -1,34 +1,40 @@
<!-- doc/src/sgml/archive-modules.sgml -->
<chapter id="archive-modules">
- <title>Archive Modules</title>
+ <title>Archive and Recovery Modules</title>
<indexterm zone="archive-modules">
- <primary>Archive Modules</primary>
+ <primary>Archive and Recovery Modules</primary>
</indexterm>
<para>
PostgreSQL provides infrastructure to create custom modules for continuous
- archiving (see <xref linkend="continuous-archiving"/>). While archiving via
- a shell command (i.e., <xref linkend="guc-archive-command"/>) is much
- simpler, a custom archive module will often be considerably more robust and
- performant.
+ archiving and recovery (see <xref linkend="continuous-archiving"/>). While
+ a shell command (e.g., <xref linkend="guc-archive-command"/>,
+ <xref linkend="guc-restore-command"/>) is much simpler, a custom module will
+ often be considerably more robust and performant.
</para>
<para>
When a custom <xref linkend="guc-archive-library"/> is configured, PostgreSQL
will submit completed WAL files to the module, and the server will avoid
recycling or removing these WAL files until the module indicates that the files
- were successfully archived. It is ultimately up to the module to decide what
- to do with each WAL file, but many recommendations are listed at
- <xref linkend="backup-archiving-wal"/>.
+ were successfully archived. When a custom
+ <xref linkend="guc-restore-library"/> is configured, PostgreSQL will use the
+ module for recovery actions. It is ultimately up to the module to decide how
+ to accomplish each task, but some recommendations are listed at
+ <xref linkend="backup-archiving-wal"/> and
+ <xref linkend="backup-pitr-recovery"/>.
</para>
<para>
- Archiving modules must at least consist of an initialization function (see
- <xref linkend="archive-module-init"/>) and the required callbacks (see
- <xref linkend="archive-module-callbacks"/>). However, archive modules are
- also permitted to do much more (e.g., declare GUCs and register background
- workers).
+ Archive and recovery modules must at least consist of an initialization
+ function (see <xref linkend="archive-module-init"/> and
+ <xref linkend="recovery-module-init"/>) and the required callbacks (see
+ <xref linkend="archive-module-callbacks"/> and
+ <xref linkend="recovery-module-callbacks"/>). However, archive and recovery
+ modules are also permitted to do much more (e.g., declare GUCs and register
+ background workers). A module may be used for both
+ <varname>archive_library</varname> and <varname>restore_library</varname>.
</para>
<para>
@@ -37,7 +43,7 @@
</para>
<sect1 id="archive-module-init">
- <title>Initialization Functions</title>
+ <title>Archive Module Initialization Functions</title>
<indexterm zone="archive-module-init">
<primary>_PG_archive_module_init</primary>
</indexterm>
@@ -64,6 +70,12 @@ typedef void (*ArchiveModuleInit) (struct ArchiveModuleCallbacks *cb);
Only the <function>archive_file_cb</function> callback is required. The
others are optional.
</para>
+
+ <note>
+ <para>
+ <varname>archive_library</varname> is only loaded in the archiver process.
+ </para>
+ </note>
</sect1>
<sect1 id="archive-module-callbacks">
@@ -129,6 +141,132 @@ typedef bool (*ArchiveFileCB) (const char *file, const char *path);
<programlisting>
typedef void (*ArchiveShutdownCB) (void);
+</programlisting>
+ </para>
+ </sect2>
+ </sect1>
+
+ <sect1 id="recovery-module-init">
+ <title>Recovery Module Initialization Functions</title>
+ <indexterm zone="recovery-module-init">
+ <primary>_PG_recovery_module_init</primary>
+ </indexterm>
+ <para>
+ A recovery library is loaded by dynamically loading a shared library with the
+ <xref linkend="guc-restore-library"/> as the library base name. The normal
+ library search path is used to locate the library. To provide the required
+ recovery module callbacks and to indicate that the library is actually a
+ recovery module, it needs to provide a function named
+ <function>_PG_recovery_module_init</function>. This function is passed a
+ struct that needs to be filled with the callback function pointers for
+ individual actions.
+
+<programlisting>
+typedef struct RecoveryModuleCallbacks
+{
+ RecoveryRestoreCB restore_cb;
+ RecoveryArchiveCleanupCB archive_cleanup_cb;
+ RecoveryEndCB recovery_end_cb;
+ RecoveryShutdownCB shutdown_cb;
+} RecoveryModuleCallbacks;
+typedef void (*RecoveryModuleInit) (struct RecoveryModuleCallbacks *cb);
+</programlisting>
+
+ The <function>restore_cb</function> callback is required for archive
+ recovery, but it is optional for streaming replication. The others are
+ always optional.
+ </para>
+
+ <note>
+ <para>
+ <varname>restore_library</varname> is only loaded in the startup and
+ checkpointer processes and in single-user mode.
+ </para>
+ </note>
+ </sect1>
+
+ <sect1 id="recovery-module-callbacks">
+ <title>Recovery Module Callbacks</title>
+ <para>
+ The recovery callbacks define the actual behavior of the module. The server
+ will call them as required to execute recovery actions.
+ </para>
+
+ <sect2 id="recovery-module-restore">
+ <title>Restore Callback</title>
+ <para>
+ The <function>restore_cb</function> callback is called to retrieve a single
+ archived segment of the WAL file series for archive recovery or streaming
+ replication.
+
+<programlisting>
+typedef bool (*RecoveryRestoreCB) (const char *file, const char *path, const char *lastRestartPointFileName);
+</programlisting>
+
+ This callback must return <literal>true</literal> only if the file was
+ successfully retrieved. If the file is not available in the archives, the
+ callback must return <literal>false</literal>.
+ <replaceable>file</replaceable> will contain just the file name
+ of the WAL file to retrieve, while <replaceable>path</replaceable> contains
+ the destination's relative path (including the file name).
+ <replaceable>lastRestartPointFileName</replaceable> will contain the name
+ of the file containing the last valid restart point. That is the earliest
+ file that must be kept to allow a restore to be restartable, so this
+ information can be used to truncate the archive to just the minimum
+ required to support restarting from the current restore.
+ <replaceable>lastRestartPointFileName</replaceable> is typically only used
+ by warm-standby configurations (see <xref linkend="warm-standby"/>). Note
+ that if multiple standby servers are restoring from the same archive
+ directory, you will need to ensure that you do not delete WAL files until
+ they are no longer needed by any of the servers.
+ </para>
+ </sect2>
+
+ <sect2 id="recovery-module-archive-cleanup">
+ <title>Archive Cleanup Callback</title>
+ <para>
+ The <function>archive_cleanup_cb</function> callback is called at every
+ restart point and is intended to provide a mechanism for cleaning up old
+ archived WAL files that are no longer needed by the standby server.
+
+<programlisting>
+typedef void (*RecoveryArchiveCleanupCB) (const char *lastRestartPointFileName);
+</programlisting>
+
+ <replaceable>lastRestartPointFileName</replaceable> will contain the name
+ of the file containing the last valid restart point, like in
+ <link linkend="recovery-module-restore"><function>restore_cb</function></link>.
+ </para>
+ </sect2>
+
+ <sect2 id="recovery-module-end">
+ <title>Recovery End Callback</title>
+ <para>
+ The <function>recovery_end_cb</function> callback is called once at the end
+ of recovery and is intended to provide a mechanism for cleanup following
+ replication or recovery.
+
+<programlisting>
+typedef void (*RecoveryEndCB) (const char *lastRestartPointFileName);
+</programlisting>
+
+ <replaceable>lastRestartPointFileName</replaceable> will contain the name
+ of the file containing the last valid restart point, like in
+ <link linkend="recovery-module-restore"><function>restore_cb</function></link>.
+ </para>
+ </sect2>
+
+ <sect2 id="recovery-module-shutdown">
+ <title>Shutdown Callback</title>
+ <para>
+ The <function>shutdown_cb</function> callback is called when a process that
+ has loaded the recovery module exits (e.g., after an error) or the value of
+ <xref linkend="guc-restore-library"/> changes. If no
+ <function>shutdown_cb</function> is defined, no special action is taken in
+ these situations.
+
+<programlisting>
+typedef void (*RecoveryShutdownCB) (void);
</programlisting>
</para>
</sect2>
diff --git a/doc/src/sgml/backup.sgml b/doc/src/sgml/backup.sgml
index 8bab521718..d6d34078fc 100644
--- a/doc/src/sgml/backup.sgml
+++ b/doc/src/sgml/backup.sgml
@@ -1181,9 +1181,27 @@ SELECT * FROM pg_backup_stop(wait_for_archive => true);
<para>
The key part of all this is to set up a recovery configuration that
describes how you want to recover and how far the recovery should
- run. The one thing that you absolutely must specify is the <varname>restore_command</varname>,
- which tells <productname>PostgreSQL</productname> how to retrieve archived
- WAL file segments. Like the <varname>archive_command</varname>, this is
+ run. The one thing that you absolutely must specify is either
+ <varname>restore_command</varname> or a <varname>restore_library</varname>
+ that defines a restore callback, which tells
+ <productname>PostgreSQL</productname> how to retrieve archived WAL file
+ segments.
+ </para>
+
+ <para>
+ Like the <varname>archive_library</varname> parameter,
+ <varname>restore_library</varname> is a shared library. Since such
+ libraries are written in <literal>C</literal>, creating your own may
+ require considerably more effort than writing a shell command. However,
+ recovery modules can be more performant than restoring via shell, and they
+ will have access to many useful server resources. For more information
+ about creating a <varname>restore_library</varname>, see
+ <xref linkend="archive-modules"/>.
+ </para>
+
+ <para>
+ Like the <varname>archive_command</varname>,
+ <varname>restore_command</varname> is
a shell command string. It can contain <literal>%f</literal>, which is
replaced by the name of the desired WAL file, and <literal>%p</literal>,
which is replaced by the path name to copy the WAL file to.
@@ -1202,14 +1220,20 @@ restore_command = 'cp /mnt/server/archivedir/%f %p'
</para>
<para>
- It is important that the command return nonzero exit status on failure.
- The command <emphasis>will</emphasis> be called requesting files that are not
- present in the archive; it must return nonzero when so asked. This is not
- an error condition. An exception is that if the command was terminated by
+ It is important that the <varname>restore_command</varname> return nonzero
+ exit status on failure, or, if you are using a
+ <varname>restore_library</varname>, that the restore function returns
+ <literal>false</literal> on failure. The command or library
+ <emphasis>will</emphasis> be called requesting files that are not
+ present in the archive; it must fail when so asked. This is not
+ an error condition. An exception is that if the
+ <varname>restore_command</varname> was terminated by
a signal (other than <systemitem>SIGTERM</systemitem>, which is used as
part of a database server shutdown) or an error by the shell (such as
command not found), then recovery will abort and the server will not start
- up.
+ up. Likewise, if the restore function provided by the
+ <varname>restore_library</varname> emits an <literal>ERROR</literal> or
+ <literal>FATAL</literal>, recovery will abort and the server won't start.
</para>
<para>
@@ -1233,7 +1257,8 @@ restore_command = 'cp /mnt/server/archivedir/%f %p'
close as possible given the available WAL segments). Therefore, a normal
recovery will end with a <quote>file not found</quote> message, the exact text
of the error message depending upon your choice of
- <varname>restore_command</varname>. You may also see an error message
+ <varname>restore_command</varname> or <varname>restore_library</varname>.
+ You may also see an error message
at the start of recovery for a file named something like
<filename>00000001.history</filename>. This is also normal and does not
indicate a problem in simple recovery situations; see
diff --git a/doc/src/sgml/basic-archive.sgml b/doc/src/sgml/basic-archive.sgml
index 0b650f17a8..ac7cd9b967 100644
--- a/doc/src/sgml/basic-archive.sgml
+++ b/doc/src/sgml/basic-archive.sgml
@@ -8,17 +8,20 @@
</indexterm>
<para>
- <filename>basic_archive</filename> is an example of an archive module. This
- module copies completed WAL segment files to the specified directory. This
- may not be especially useful, but it can serve as a starting point for
- developing your own archive module. For more information about archive
- modules, see <xref linkend="archive-modules"/>.
+ <filename>basic_archive</filename> is an example of an archive and recovery
+ module. This module copies completed WAL segment files to or from the
+ specified directory. This may not be especially useful, but it can serve as
+ a starting point for developing your own archive and recovery modules. For
+ more information about archive and recovery modules, see
+ see <xref linkend="archive-modules"/>.
</para>
<para>
- In order to function, this module must be loaded via
+ For use as an archive module, this module must be loaded via
<xref linkend="guc-archive-library"/>, and <xref linkend="guc-archive-mode"/>
- must be enabled.
+ must be enabled. For use as a recovery module, this module must be loaded
+ via <xref linkend="guc-restore-library"/>, and recovery must be enabled (see
+ <xref linkend="runtime-config-wal-archive-recovery"/>).
</para>
<sect2>
@@ -34,11 +37,12 @@
</term>
<listitem>
<para>
- The directory where the server should copy WAL segment files. This
- directory must already exist. The default is an empty string, which
- effectively halts WAL archiving, but if <xref linkend="guc-archive-mode"/>
- is enabled, the server will accumulate WAL segment files in the
- expectation that a value will soon be provided.
+ The directory where the server should copy WAL segment files to or from.
+ This directory must already exist. The default is an empty string,
+ which, when used for archiving, effectively halts WAL archival, but if
+ <xref linkend="guc-archive-mode"/> is enabled, the server will accumulate
+ WAL segment files in the expectation that a value will soon be provided.
+ When an empty string is used for recovery, restore will fail.
</para>
</listitem>
</varlistentry>
@@ -46,7 +50,7 @@
<para>
These parameters must be set in <filename>postgresql.conf</filename>.
- Typical usage might be:
+ Typical usage as an archive module might be:
</para>
<programlisting>
@@ -61,7 +65,8 @@ basic_archive.archive_directory = '/path/to/archive/directory'
<title>Notes</title>
<para>
- Server crashes may leave temporary files with the prefix
+ When <filename>basic_archive</filename> is used as an archive module, server
+ crashes may leave temporary files with the prefix
<filename>archtemp</filename> in the archive directory. It is recommended to
delete such files before restarting the server after a crash. It is safe to
remove such files while the server is running as long as they are unrelated
diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml
index 05b3862d09..899d2b5fdb 100644
--- a/doc/src/sgml/config.sgml
+++ b/doc/src/sgml/config.sgml
@@ -3773,7 +3773,8 @@ include_dir 'conf.d'
recovery when the end of archived WAL is reached, but will keep trying to
continue recovery by connecting to the sending server as specified by the
<varname>primary_conninfo</varname> setting and/or by fetching new WAL
- segments using <varname>restore_command</varname>. For this mode, the
+ segments using <varname>restore_command</varname> or
+ <varname>restore_library</varname>. For this mode, the
parameters from this section and <xref
linkend="runtime-config-replication-standby"/> are of interest.
Parameters from <xref linkend="runtime-config-wal-recovery-target"/> will
@@ -3801,7 +3802,8 @@ include_dir 'conf.d'
<listitem>
<para>
The local shell command to execute to retrieve an archived segment of
- the WAL file series. This parameter is required for archive recovery,
+ the WAL file series. Either <varname>restore_command</varname> or
+ <xref linkend="guc-restore-library"/> is required for archive recovery,
but optional for streaming replication.
Any <literal>%f</literal> in the string is
replaced by the name of the file to retrieve from the archive,
@@ -3836,7 +3838,42 @@ restore_command = 'copy "C:\\server\\archivedir\\%f" "%p"' # Windows
<para>
This parameter can only be set in the <filename>postgresql.conf</filename>
- file or on the server command line.
+ file or on the server command line. It is only used if
+ <varname>restore_library</varname> is set to an empty string. If both
+ <varname>restore_command</varname> and
+ <varname>restore_library</varname> are set, an error will be raised.
+ </para>
+ </listitem>
+ </varlistentry>
+
+ <varlistentry id="guc-restore-library" xreflabel="restore_library">
+ <term><varname>restore_library</varname> (<type>string</type>)
+ <indexterm>
+ <primary><varname>restore_library</varname> configuration parameter</primary>
+ </indexterm>
+ </term>
+ <listitem>
+ <para>
+ The library to use for recovery actions, including retrieving archived
+ segments of the WAL file series and executing tasks at restartpoints
+ and at recovery end. Either <xref linkend="guc-restore-command"/> or
+ <varname>restore_library</varname> is required for archive recovery,
+ but optional for streaming replication. If this parameter is set to an
+ empty string (the default), restoring via shell is enabled, and
+ <varname>restore_command</varname>,
+ <varname>archive_cleanup_command</varname> and
+ <varname>recovery_end_command</varname> are used. If both
+ <varname>restore_library</varname> and any of
+ <varname>restore_command</varname>,
+ <varname>archive_cleanup_command</varname> or
+ <varname>recovery_end_command</varname> are set, an error will be
+ raised. Otherwise, the specified shared library is used for recovery.
+ For more information, see <xref linkend="archive-modules"/>.
+ </para>
+
+ <para>
+ This parameter can only be set in the
+ <filename>postgresql.conf</filename> file or on the server command line.
</para>
</listitem>
</varlistentry>
@@ -3881,7 +3918,10 @@ restore_command = 'copy "C:\\server\\archivedir\\%f" "%p"' # Windows
</para>
<para>
This parameter can only be set in the <filename>postgresql.conf</filename>
- file or on the server command line.
+ file or on the server command line. It is only used if
+ <varname>restore_library</varname> is set to an empty string. If both
+ <varname>archive_cleanup_command</varname> and
+ <varname>restore_library</varname> are set, an error will be raised.
</para>
</listitem>
</varlistentry>
@@ -3910,11 +3950,13 @@ restore_command = 'copy "C:\\server\\archivedir\\%f" "%p"' # Windows
</para>
<para>
This parameter can only be set in the <filename>postgresql.conf</filename>
- file or on the server command line.
+ file or on the server command line. It is only used if
+ <varname>restore_library</varname> is set to an empty string. If both
+ <varname>recovery_end_command</varname> and
+ <varname>restore_library</varname> are set, an error will be raised.
</para>
</listitem>
</varlistentry>
-
</variablelist>
</sect2>
diff --git a/doc/src/sgml/high-availability.sgml b/doc/src/sgml/high-availability.sgml
index f180607528..6266e2df7f 100644
--- a/doc/src/sgml/high-availability.sgml
+++ b/doc/src/sgml/high-availability.sgml
@@ -627,7 +627,8 @@ protocol to make nodes agree on a serializable transactional order.
<para>
In standby mode, the server continuously applies WAL received from the
primary server. The standby server can read WAL from a WAL archive
- (see <xref linkend="guc-restore-command"/>) or directly from the primary
+ (see <xref linkend="guc-restore-command"/> and
+ <xref linkend="guc-restore-library"/>) or directly from the primary
over a TCP connection (streaming replication). The standby server will
also attempt to restore any WAL found in the standby cluster's
<filename>pg_wal</filename> directory. That typically happens after a server
@@ -638,9 +639,11 @@ protocol to make nodes agree on a serializable transactional order.
<para>
At startup, the standby begins by restoring all WAL available in the
- archive location, calling <varname>restore_command</varname>. Once it
- reaches the end of WAL available there and <varname>restore_command</varname>
- fails, it tries to restore any WAL available in the <filename>pg_wal</filename> directory.
+ archive location, either by calling <varname>restore_command</varname> or
+ by executing the <varname>restore_library</varname>'s restore callback.
+ Once it reaches the end of WAL available there and
+ <varname>restore_command</varname> or the restore callback fails, it tries
+ to restore any WAL available in the <filename>pg_wal</filename> directory.
If that fails, and streaming replication has been configured, the
standby tries to connect to the primary server and start streaming WAL
from the last valid record found in archive or <filename>pg_wal</filename>. If that fails
@@ -698,7 +701,8 @@ protocol to make nodes agree on a serializable transactional order.
server (see <xref linkend="backup-pitr-recovery"/>). Create a file
<link linkend="file-standby-signal"><filename>standby.signal</filename></link><indexterm><primary>standby.signal</primary></indexterm>
in the standby's cluster data
- directory. Set <xref linkend="guc-restore-command"/> to a simple command to copy files from
+ directory. Set <xref linkend="guc-restore-command"/> or
+ <xref linkend="guc-restore-library"/> to copy files from
the WAL archive. If you plan to have multiple standby servers for high
availability purposes, make sure that <varname>recovery_target_timeline</varname> is set to
<literal>latest</literal> (the default), to make the standby server follow the timeline change
@@ -707,7 +711,8 @@ protocol to make nodes agree on a serializable transactional order.
<note>
<para>
- <xref linkend="guc-restore-command"/> should return immediately
+ <xref linkend="guc-restore-command"/> and restore callbacks provided by
+ <xref linkend="guc-restore-library"/> should return immediately
if the file does not exist; the server will retry the command again if
necessary.
</para>
@@ -731,8 +736,10 @@ protocol to make nodes agree on a serializable transactional order.
<para>
If you're using a WAL archive, its size can be minimized using the <xref
- linkend="guc-archive-cleanup-command"/> parameter to remove files that are no
- longer required by the standby server.
+ linkend="guc-archive-cleanup-command"/> parameter or the
+ <xref linkend="guc-restore-library"/>'s
+ <function>archive_cleanup_cb</function> callback function to remove files
+ that are no longer required by the standby server.
The <application>pg_archivecleanup</application> utility is designed specifically to
be used with <varname>archive_cleanup_command</varname> in typical single-standby
configurations, see <xref linkend="pgarchivecleanup"/>.
diff --git a/src/backend/access/transam/shell_restore.c b/src/backend/access/transam/shell_restore.c
index 073e709e06..124a0bbdb6 100644
--- a/src/backend/access/transam/shell_restore.c
+++ b/src/backend/access/transam/shell_restore.c
@@ -3,7 +3,8 @@
* shell_restore.c
*
* These recovery functions use a user-specified shell command (e.g., the
- * restore_command GUC).
+ * restore_command GUC). It is used as the default, but other modules may
+ * define their own recovery logic.
*
* Copyright (c) 2022, PostgreSQL Global Development Group
*
@@ -22,6 +23,10 @@
#include "storage/ipc.h"
#include "utils/wait_event.h"
+static bool shell_restore(const char *file, const char *path,
+ const char *lastRestartPointFileName);
+static void shell_archive_cleanup(const char *lastRestartPointFileName);
+static void shell_recovery_end(const char *lastRestartPointFileName);
static char *BuildCleanupCommand(const char *command,
const char *lastRestartPointFileName);
static bool ExecuteRecoveryCommand(const char *command,
@@ -29,6 +34,16 @@ static bool ExecuteRecoveryCommand(const char *command,
bool exitOnSigterm, uint32 wait_event_info,
int fail_elevel);
+void
+shell_restore_init(RecoveryModuleCallbacks *cb)
+{
+ AssertVariableIsOfType(&shell_restore_init, RecoveryModuleInit);
+
+ cb->restore_cb = shell_restore;
+ cb->archive_cleanup_cb = shell_archive_cleanup;
+ cb->recovery_end_cb = shell_recovery_end;
+}
+
bool
shell_restore(const char *file, const char *path,
const char *lastRestartPointFileName)
@@ -73,7 +88,7 @@ shell_restore(const char *file, const char *path,
return ret;
}
-void
+static void
shell_archive_cleanup(const char *lastRestartPointFileName)
{
char *cmd = BuildCleanupCommand(archiveCleanupCommand,
@@ -84,7 +99,7 @@ shell_archive_cleanup(const char *lastRestartPointFileName)
pfree(cmd);
}
-void
+static void
shell_recovery_end(const char *lastRestartPointFileName)
{
char *cmd = BuildCleanupCommand(recoveryEndCommand,
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index fdce12614a..c4ad571327 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -4883,10 +4883,11 @@ static void
CleanupAfterArchiveRecovery(TimeLineID EndOfLogTLI, XLogRecPtr EndOfLog,
TimeLineID newTLI)
{
+
/*
- * Execute the recovery_end_command, if any.
+ * Execute the recovery-end callback, if any.
*/
- if (recoveryEndCommand && strcmp(recoveryEndCommand, "") != 0)
+ if (RecoveryContext.recovery_end_cb)
{
char lastRestartPointFname[MAXPGPATH];
XLogSegNo restartSegNo;
@@ -4903,7 +4904,7 @@ CleanupAfterArchiveRecovery(TimeLineID EndOfLogTLI, XLogRecPtr EndOfLog,
XLogFileName(lastRestartPointFname, restartTli, restartSegNo,
wal_segment_size);
- shell_recovery_end(lastRestartPointFname);
+ RecoveryContext.recovery_end_cb(lastRestartPointFname);
}
/*
@@ -7318,9 +7319,9 @@ CreateRestartPoint(int flags)
timestamptz_to_str(xtime)) : 0));
/*
- * Finally, execute archive_cleanup_command, if any.
+ * Execute the archive-cleanup callback, if any.
*/
- if (archiveCleanupCommand && strcmp(archiveCleanupCommand, "") != 0)
+ if (RecoveryContext.archive_cleanup_cb)
{
char lastRestartPointFname[MAXPGPATH];
XLogSegNo restartSegNo;
@@ -7337,7 +7338,7 @@ CreateRestartPoint(int flags)
XLogFileName(lastRestartPointFname, restartTli, restartSegNo,
wal_segment_size);
- shell_archive_cleanup(lastRestartPointFname);
+ RecoveryContext.archive_cleanup_cb(lastRestartPointFname);
}
return true;
diff --git a/src/backend/access/transam/xlogarchive.c b/src/backend/access/transam/xlogarchive.c
index b5cb060d55..fdab7dad43 100644
--- a/src/backend/access/transam/xlogarchive.c
+++ b/src/backend/access/transam/xlogarchive.c
@@ -22,7 +22,9 @@
#include "access/xlog.h"
#include "access/xlog_internal.h"
#include "access/xlogarchive.h"
+#include "access/xlogrecovery.h"
#include "common/archive.h"
+#include "fmgr.h"
#include "miscadmin.h"
#include "pgstat.h"
#include "postmaster/startup.h"
@@ -32,6 +34,11 @@
#include "storage/ipc.h"
#include "storage/lwlock.h"
+/*
+ * Global context for recovery-related callbacks.
+ */
+RecoveryModuleCallbacks RecoveryContext;
+
/*
* Attempt to retrieve the specified file from off-line archival storage.
* If successful, fill "path" with its complete path (note that this will be
@@ -71,7 +78,7 @@ RestoreArchivedFile(char *path, const char *xlogfname,
goto not_available;
/* In standby mode, restore_command might not be supplied */
- if (recoveryRestoreCommand == NULL || strcmp(recoveryRestoreCommand, "") == 0)
+ if (RecoveryContext.restore_cb == NULL)
goto not_available;
/*
@@ -149,14 +156,15 @@ RestoreArchivedFile(char *path, const char *xlogfname,
XLogFileName(lastRestartPointFname, 0, 0L, wal_segment_size);
/*
- * Check signals before restore command and reset afterwards.
+ * Check signals before restore callback and reset afterwards.
*/
PreRestoreCommand();
/*
* Copy xlog from archival storage to XLOGDIR
*/
- ret = shell_restore(xlogfname, xlogpath, lastRestartPointFname);
+ ret = RecoveryContext.restore_cb(xlogfname, xlogpath,
+ lastRestartPointFname);
PostRestoreCommand();
@@ -603,3 +611,59 @@ XLogArchiveCleanup(const char *xlog)
unlink(archiveStatusPath);
/* should we complain about failure? */
}
+
+/*
+ * Loads all the recovery callbacks into our global RecoveryContext. The
+ * caller is responsible for validating the combination of library/command
+ * parameters that are set (e.g., restore_command and restore_library cannot
+ * both be set).
+ */
+void
+LoadRecoveryCallbacks(void)
+{
+ RecoveryModuleInit init;
+
+ /*
+ * If the shell command is enabled, use our special initialization
+ * function. Otherwise, load the library and call its
+ * _PG_recovery_module_init().
+ */
+ if (restoreLibrary[0] == '\0')
+ init = shell_restore_init;
+ else
+ init = (RecoveryModuleInit)
+ load_external_function(restoreLibrary, "_PG_recovery_module_init",
+ false, NULL);
+
+ if (init == NULL)
+ ereport(ERROR,
+ (errmsg("recovery modules have to define the symbol "
+ "_PG_recovery_module_init")));
+
+ memset(&RecoveryContext, 0, sizeof(RecoveryModuleCallbacks));
+ (*init) (&RecoveryContext);
+
+ /*
+ * If using shell commands, remove callbacks for any commands that are not
+ * set.
+ */
+ if (restoreLibrary[0] == '\0')
+ {
+ if (recoveryRestoreCommand[0] == '\0')
+ RecoveryContext.restore_cb = NULL;
+ if (archiveCleanupCommand[0] == '\0')
+ RecoveryContext.archive_cleanup_cb = NULL;
+ if (recoveryEndCommand[0] == '\0')
+ RecoveryContext.recovery_end_cb = NULL;
+ }
+}
+
+/*
+ * Call the shutdown callback of the loaded recovery module, if defined.
+ */
+void
+call_recovery_module_shutdown_cb(int code, Datum arg)
+{
+ if (RecoveryContext.shutdown_cb)
+ RecoveryContext.shutdown_cb();
+}
diff --git a/src/backend/access/transam/xlogrecovery.c b/src/backend/access/transam/xlogrecovery.c
index bc3c3eb3e7..95bcc3a0b7 100644
--- a/src/backend/access/transam/xlogrecovery.c
+++ b/src/backend/access/transam/xlogrecovery.c
@@ -80,6 +80,7 @@ const struct config_enum_entry recovery_target_action_options[] = {
/* options formerly taken from recovery.conf for archive recovery */
char *recoveryRestoreCommand = NULL;
+char *restoreLibrary = NULL;
char *recoveryEndCommand = NULL;
char *archiveCleanupCommand = NULL;
RecoveryTargetType recoveryTarget = RECOVERY_TARGET_UNSET;
@@ -1053,24 +1054,37 @@ validateRecoveryParameters(void)
if (!ArchiveRecoveryRequested)
return;
+ /*
+ * Check for invalid combinations of the command/library parameters and
+ * load the callbacks.
+ */
+ CheckMutuallyExclusiveGUCs(restoreLibrary, "restore_library",
+ recoveryRestoreCommand, "restore_command");
+ CheckMutuallyExclusiveGUCs(restoreLibrary, "restore_library",
+ recoveryEndCommand, "recovery_end_command");
+ before_shmem_exit(call_recovery_module_shutdown_cb, 0);
+ LoadRecoveryCallbacks();
+
/*
* Check for compulsory parameters
*/
if (StandbyModeRequested)
{
if ((PrimaryConnInfo == NULL || strcmp(PrimaryConnInfo, "") == 0) &&
- (recoveryRestoreCommand == NULL || strcmp(recoveryRestoreCommand, "") == 0))
+ RecoveryContext.restore_cb == NULL)
ereport(WARNING,
- (errmsg("specified neither primary_conninfo nor restore_command"),
- errhint("The database server will regularly poll the pg_wal subdirectory to check for files placed there.")));
+ (errmsg("specified neither primary_conninfo nor restore_command "
+ "nor a restore_library that defines a restore callback"),
+ errhint("The database server will regularly poll the pg_wal "
+ "subdirectory to check for files placed there.")));
}
else
{
- if (recoveryRestoreCommand == NULL ||
- strcmp(recoveryRestoreCommand, "") == 0)
+ if (RecoveryContext.restore_cb == NULL)
ereport(FATAL,
(errcode(ERRCODE_INVALID_PARAMETER_VALUE),
- errmsg("must specify restore_command when standby mode is not enabled")));
+ errmsg("must specify restore_command or a restore_library that defines "
+ "a restore callback when standby mode is not enabled")));
}
/*
diff --git a/src/backend/postmaster/checkpointer.c b/src/backend/postmaster/checkpointer.c
index de0bbbfa79..6350fd0b83 100644
--- a/src/backend/postmaster/checkpointer.c
+++ b/src/backend/postmaster/checkpointer.c
@@ -38,6 +38,7 @@
#include "access/xlog.h"
#include "access/xlog_internal.h"
+#include "access/xlogarchive.h"
#include "access/xlogrecovery.h"
#include "libpq/pqsignal.h"
#include "miscadmin.h"
@@ -222,6 +223,16 @@ CheckpointerMain(void)
*/
before_shmem_exit(pgstat_before_server_shutdown, 0);
+ /*
+ * Check for invalid combinations of the command/library parameters and
+ * load the callbacks. We do this before setting up the exception handler
+ * so that any problems result in a server crash shortly after startup.
+ */
+ CheckMutuallyExclusiveGUCs(restoreLibrary, "restore_library",
+ archiveCleanupCommand, "archive_cleanup_command");
+ before_shmem_exit(call_recovery_module_shutdown_cb, 0);
+ LoadRecoveryCallbacks();
+
/*
* Create a memory context that we will do all our work in. We do this so
* that we can reset the context during error recovery and thereby avoid
@@ -548,6 +559,9 @@ HandleCheckpointerInterrupts(void)
if (ConfigReloadPending)
{
+ char *prevRestoreLibrary = pstrdup(restoreLibrary);
+ char *prevArchiveCleanupCommand = pstrdup(archiveCleanupCommand);
+
ConfigReloadPending = false;
ProcessConfigFile(PGC_SIGHUP);
@@ -563,6 +577,18 @@ HandleCheckpointerInterrupts(void)
* because of SIGHUP.
*/
UpdateSharedMemoryConfig();
+
+ CheckMutuallyExclusiveGUCs(restoreLibrary, "restore_library",
+ archiveCleanupCommand, "archive_cleanup_command");
+ if (strcmp(prevRestoreLibrary, restoreLibrary) != 0 ||
+ strcmp(prevArchiveCleanupCommand, archiveCleanupCommand) != 0)
+ {
+ call_recovery_module_shutdown_cb(0, (Datum) 0);
+ LoadRecoveryCallbacks();
+ }
+
+ pfree(prevRestoreLibrary);
+ pfree(prevArchiveCleanupCommand);
}
if (ShutdownRequestPending)
{
diff --git a/src/backend/postmaster/pgarch.c b/src/backend/postmaster/pgarch.c
index 8ecdb9ca23..8e91f2d70f 100644
--- a/src/backend/postmaster/pgarch.c
+++ b/src/backend/postmaster/pgarch.c
@@ -831,11 +831,8 @@ LoadArchiveLibrary(void)
{
ArchiveModuleInit archive_init;
- if (XLogArchiveLibrary[0] != '\0' && XLogArchiveCommand[0] != '\0')
- ereport(ERROR,
- (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
- errmsg("both archive_command and archive_library set"),
- errdetail("Only one of archive_command, archive_library may be set.")));
+ CheckMutuallyExclusiveGUCs(XLogArchiveLibrary, "archive_library",
+ XLogArchiveCommand, "archive_command");
memset(&ArchiveContext, 0, sizeof(ArchiveModuleCallbacks));
diff --git a/src/backend/postmaster/startup.c b/src/backend/postmaster/startup.c
index 8786186898..f9ff2b5583 100644
--- a/src/backend/postmaster/startup.c
+++ b/src/backend/postmaster/startup.c
@@ -20,6 +20,7 @@
#include "postgres.h"
#include "access/xlog.h"
+#include "access/xlogarchive.h"
#include "access/xlogrecovery.h"
#include "access/xlogutils.h"
#include "libpq/pqsignal.h"
@@ -133,13 +134,17 @@ StartupProcShutdownHandler(SIGNAL_ARGS)
* Re-read the config file.
*
* If one of the critical walreceiver options has changed, flag xlog.c
- * to restart it.
+ * to restart it. Also, check for invalid combinations of the command/library
+ * parameters and reload the recovery callbacks if necessary.
*/
static void
StartupRereadConfig(void)
{
char *conninfo = pstrdup(PrimaryConnInfo);
char *slotname = pstrdup(PrimarySlotName);
+ char *prevRestoreLibrary = pstrdup(restoreLibrary);
+ char *prevRestoreCommand = pstrdup(recoveryRestoreCommand);
+ char *prevRecoveryEndCommand = pstrdup(recoveryEndCommand);
bool tempSlot = wal_receiver_create_temp_slot;
bool conninfoChanged;
bool slotnameChanged;
@@ -161,6 +166,22 @@ StartupRereadConfig(void)
if (conninfoChanged || slotnameChanged || tempSlotChanged)
StartupRequestWalReceiverRestart();
+
+ CheckMutuallyExclusiveGUCs(restoreLibrary, "restore_library",
+ recoveryRestoreCommand, "restore_command");
+ CheckMutuallyExclusiveGUCs(restoreLibrary, "restore_library",
+ recoveryEndCommand, "recovery_end_command");
+ if (strcmp(prevRestoreLibrary, restoreLibrary) != 0 ||
+ strcmp(prevRestoreCommand, recoveryRestoreCommand) != 0 ||
+ strcmp(prevRecoveryEndCommand, recoveryEndCommand) != 0)
+ {
+ call_recovery_module_shutdown_cb(0, (Datum) 0);
+ LoadRecoveryCallbacks();
+ }
+
+ pfree(prevRestoreLibrary);
+ pfree(prevRestoreCommand);
+ pfree(prevRecoveryEndCommand);
}
/* Handle various signals that might be sent to the startup process */
diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c
index d52069f446..7858e9a649 100644
--- a/src/backend/utils/misc/guc.c
+++ b/src/backend/utils/misc/guc.c
@@ -6880,3 +6880,17 @@ call_enum_check_hook(struct config_enum *conf, int *newval, void **extra,
return true;
}
+
+/*
+ * ERROR if both parameters are set.
+ */
+void
+CheckMutuallyExclusiveGUCs(const char *p1val, const char *p1name,
+ const char *p2val, const char *p2name)
+{
+ if (p1val[0] != '\0' && p2val[0] != '\0')
+ ereport(ERROR,
+ (errcode(ERRCODE_INVALID_PARAMETER_VALUE),
+ errmsg("both %s and %s set", p1name, p2name),
+ errdetail("Only one of %s, %s may be set.", p1name, p2name)));
+}
diff --git a/src/backend/utils/misc/guc_tables.c b/src/backend/utils/misc/guc_tables.c
index 68328b1402..19746e2489 100644
--- a/src/backend/utils/misc/guc_tables.c
+++ b/src/backend/utils/misc/guc_tables.c
@@ -3764,6 +3764,16 @@ struct config_string ConfigureNamesString[] =
NULL, NULL, NULL
},
+ {
+ {"restore_library", PGC_SIGHUP, WAL_ARCHIVE_RECOVERY,
+ gettext_noop("Sets the library that will be called for recovery actions."),
+ NULL
+ },
+ &restoreLibrary,
+ "",
+ NULL, NULL, NULL
+ },
+
{
{"archive_cleanup_command", PGC_SIGHUP, WAL_ARCHIVE_RECOVERY,
gettext_noop("Sets the shell command that will be executed at every restart point."),
diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample
index 5afdeb04de..e71e79271a 100644
--- a/src/backend/utils/misc/postgresql.conf.sample
+++ b/src/backend/utils/misc/postgresql.conf.sample
@@ -269,6 +269,7 @@
# placeholders: %p = path of file to restore
# %f = file name only
# e.g. 'cp /mnt/server/archivedir/%f %p'
+#restore_library = '' # library to use for recovery actions
#archive_cleanup_command = '' # command to execute at every restartpoint
#recovery_end_command = '' # command to execute at completion of recovery
diff --git a/src/include/access/xlog_internal.h b/src/include/access/xlog_internal.h
index 59fc7bc105..756f0898b5 100644
--- a/src/include/access/xlog_internal.h
+++ b/src/include/access/xlog_internal.h
@@ -400,5 +400,6 @@ extern PGDLLIMPORT bool ArchiveRecoveryRequested;
extern PGDLLIMPORT bool InArchiveRecovery;
extern PGDLLIMPORT bool StandbyMode;
extern PGDLLIMPORT char *recoveryRestoreCommand;
+extern PGDLLIMPORT char *restoreLibrary;
#endif /* XLOG_INTERNAL_H */
diff --git a/src/include/access/xlogarchive.h b/src/include/access/xlogarchive.h
index 299304703e..71c9b88165 100644
--- a/src/include/access/xlogarchive.h
+++ b/src/include/access/xlogarchive.h
@@ -30,9 +30,45 @@ extern bool XLogArchiveIsReady(const char *xlog);
extern bool XLogArchiveIsReadyOrDone(const char *xlog);
extern void XLogArchiveCleanup(const char *xlog);
-extern bool shell_restore(const char *file, const char *path,
- const char *lastRestartPointFileName);
-extern void shell_archive_cleanup(const char *lastRestartPointFileName);
-extern void shell_recovery_end(const char *lastRestartPointFileName);
+/*
+ * Recovery module callbacks
+ *
+ * These callback functions should be defined by recovery libraries and
+ * returned via _PG_recovery_module_init(). For more information about the
+ * purpose of each callback, refer to the recovery modules documentation.
+ */
+typedef bool (*RecoveryRestoreCB) (const char *file, const char *path,
+ const char *lastRestartPointFileName);
+typedef void (*RecoveryArchiveCleanupCB) (const char *lastRestartPointFileName);
+typedef void (*RecoveryEndCB) (const char *lastRestartPointFileName);
+typedef void (*RecoveryShutdownCB) (void);
+
+typedef struct RecoveryModuleCallbacks
+{
+ RecoveryRestoreCB restore_cb;
+ RecoveryArchiveCleanupCB archive_cleanup_cb;
+ RecoveryEndCB recovery_end_cb;
+ RecoveryShutdownCB shutdown_cb;
+} RecoveryModuleCallbacks;
+
+extern RecoveryModuleCallbacks RecoveryContext;
+
+/*
+ * Type of the shared library symbol _PG_recovery_module_init that is looked up
+ * when loading a recovery library.
+ */
+typedef void (*RecoveryModuleInit) (RecoveryModuleCallbacks *cb);
+
+extern PGDLLEXPORT void _PG_recovery_module_init(RecoveryModuleCallbacks *cb);
+
+extern void LoadRecoveryCallbacks(void);
+extern void call_recovery_module_shutdown_cb(int code, Datum arg);
+
+/*
+ * Since the logic for recovery via a shell command is in the core server and
+ * does not need to be loaded via a shared library, it has a special
+ * initialization function.
+ */
+extern void shell_restore_init(RecoveryModuleCallbacks *cb);
#endif /* XLOG_ARCHIVE_H */
diff --git a/src/include/access/xlogrecovery.h b/src/include/access/xlogrecovery.h
index 47c29350f5..35d1d09374 100644
--- a/src/include/access/xlogrecovery.h
+++ b/src/include/access/xlogrecovery.h
@@ -55,6 +55,7 @@ extern PGDLLIMPORT int recovery_min_apply_delay;
extern PGDLLIMPORT char *PrimaryConnInfo;
extern PGDLLIMPORT char *PrimarySlotName;
extern PGDLLIMPORT char *recoveryRestoreCommand;
+extern PGDLLIMPORT char *restoreLibrary;
extern PGDLLIMPORT char *recoveryEndCommand;
extern PGDLLIMPORT char *archiveCleanupCommand;
diff --git a/src/include/utils/guc.h b/src/include/utils/guc.h
index ba89d013e6..947597247f 100644
--- a/src/include/utils/guc.h
+++ b/src/include/utils/guc.h
@@ -404,6 +404,8 @@ extern void *guc_malloc(int elevel, size_t size);
extern pg_nodiscard void *guc_realloc(int elevel, void *old, size_t size);
extern char *guc_strdup(int elevel, const char *src);
extern void guc_free(void *ptr);
+extern void CheckMutuallyExclusiveGUCs(const char *p1val, const char *p1name,
+ const char *p2val, const char *p2name);
#ifdef EXEC_BACKEND
extern void write_nondefault_variables(GucContext context);
--
2.25.1
--Kj7319i9nmIyA2yE--
^ permalink raw reply [nested|flat] 5+ messages in thread
* Re: vectorized CRC on ARM64
@ 2026-01-12 09:27 John Naylor <[email protected]>
2026-06-29 08:52 ` Re: vectorized CRC on ARM64 Peter Eisentraut <[email protected]>
0 siblings, 1 reply; 5+ messages in thread
From: John Naylor @ 2026-01-12 09:27 UTC (permalink / raw)
To: PostgreSQL Hackers <[email protected]>
On Wed, May 14, 2025 I wrote:
>
> We did something similar for x86 for v18, and here is some progress
> towards Arm support.
Coming back to this, since there's been recent interest in Arm support.
v2 is a rebase, with a few changes.
- I simplified it by leaving out the inlining for "assume CRC" builds,
since I wanted to avoid alignment considerations if I can. I think
always indirecting through a pointer will have less risk of
regressions in a realistic setting than for x86 since Arm chips
typically have low latency for carryless multiplication instructions.
With just a bit of code we can still use the direct call for small
constant inputs, so I did that to avoid regressions under WAL insert
lock.
- One coding idiom for a vector literal in the generated code was
giving pgindent indigestion, I so rewrote it using Neon intrinsics and
verified it in Godbolt.
> 0002: Like 3c6e8c12389 and in fact uses the same program to generate
> the code, by specifying Neon instructions with the Arm "crypto"
> extension instead. There are some interesting differences from x86
> here as well:
> - The upstream implementation chose to use inline assembly instead of
> intrinsics for some reason. I initially thought that was a way to get
> broader compiler support, but it turns out you still need to pass the
> relevant flags to get the assembly to link.
To follow-up for curiosity's sake, [1] says that Apple chips can issue
PMULL + EOR as a single uop if they are next to each other in the
instruction stream.
> - I only have Meson support for now, since I used MacOS on CI to test.
> That OS and compiler combination apparently targets the CRC extension,
> but the PMULL instruction runtime check uses Linux-only headers, I
> believe, so previously I hacked the choose function to return true for
> testing. The choose function in 0002 is untested in this form.
This is still true, but now the CI hack lives in a separate
not-for-commit patch for clarity.
autoconf support is a WIP, and I will share that after I do some
testing on an Arm Linux instance.
[1] https://dougallj.github.io/applecpu/firestorm.html
--
John Naylor
Amazon Web Services
Attachments:
[text/x-patch] v2-0001-Compute-CRC32C-on-ARM-using-the-Crypto-Extension-.patch (9.5K, ../../CANWCAZYAKWTFW_1byMq44t57urk6YKoDh1z1Zp6fk5-4dOPy-w@mail.gmail.com/2-v2-0001-Compute-CRC32C-on-ARM-using-the-Crypto-Extension-.patch)
download | inline diff:
From 5057eb8ac0b0156de8b9371415a7dfc852cfd0c7 Mon Sep 17 00:00:00 2001
From: John Naylor <[email protected]>
Date: Fri, 9 May 2025 19:48:26 +0700
Subject: [PATCH v2 1/2] Compute CRC32C on ARM using the Crypto Extension where
available
---
meson.build | 33 ++++++++
src/include/port/pg_crc32c.h | 22 ++++--
src/port/meson.build | 1 +
src/port/pg_crc32c_armv8.c | 124 ++++++++++++++++++++++++++++++
src/port/pg_crc32c_armv8_choose.c | 34 ++++++++
5 files changed, 209 insertions(+), 5 deletions(-)
diff --git a/meson.build b/meson.build
index 2064d1b0a8d..6401895a5da 100644
--- a/meson.build
+++ b/meson.build
@@ -2548,6 +2548,39 @@ int main(void)
have_optimized_crc = true
endif
+ # Check if the compiler supports ARMv8 CRYPTO carryless multiplication
+ # and exclusive-or inline assembly instructions used for computing CRC.
+ # Check __crc32cd here as well, since the full implementation relies on
+ # 8-byte CRC instructions.
+ prog = '''
+#include <arm_acle.h>
+#include <arm_neon.h>
+uint64x2_t a;
+uint64x2_t b;
+uint64x2_t c;
+
+int main(void)
+{
+ uint64x2_t r;
+ uint64x2_t r2;
+
+__asm("pmull %0.1q, %2.1d, %3.1d\neor %0.16b, %0.16b, %1.16b\n":"=w"(r), "+w"(c):"w"(a), "w"(b));
+__asm("pmull2 %0.1q, %2.2d, %3.2d\neor %0.16b, %0.16b, %1.16b\n":"=w"(r2), "+w"(c):"w"(a), "w"(b));
+
+ /* return computed value, to prevent the above being optimized away */
+ r = veorq_u64(r, r2);
+ return __crc32cd(0, vgetq_lane_u64(r, 0));
+}
+'''
+
+ if cc.links(prog,
+ name: 'PMULL CRC32C',
+ args: test_c_args + ['-march=armv8-a+crc+simd+crypto'])
+ # Use ARM CRYPTO Extension, with runtime check
+ cflags_crc += '-march=armv8-a+crc+simd+crypto'
+ cdata.set('USE_PMULL_CRC32C_WITH_RUNTIME_CHECK', 1)
+ endif
+
elif host_cpu == 'loongarch64'
prog = '''
diff --git a/src/include/port/pg_crc32c.h b/src/include/port/pg_crc32c.h
index 9ac619aec3e..bbfc68f6dd5 100644
--- a/src/include/port/pg_crc32c.h
+++ b/src/include/port/pg_crc32c.h
@@ -111,13 +111,22 @@ extern pg_crc32c pg_comp_crc32c_avx512(pg_crc32c crc, const void *data, size_t l
#endif
#elif defined(USE_ARMV8_CRC32C)
-/* Use ARMv8 CRC Extension instructions. */
-
+/*
+ * Use either ARMv8 CRC Extension or CRYPTO Extension (PMULL) instructions.
+ * We don't need a runtime check for CRC, so for small constant inputs,
+ * we can avoid an indirect function call.
+ */
#define COMP_CRC32C(crc, data, len) \
- ((crc) = pg_comp_crc32c_armv8((crc), (data), (len)))
+ ((crc) = __builtin_constant_p(len) && len < 32 ? \
+ pg_comp_crc32c_armv8((crc), (data), (len)) : \
+ pg_comp_crc32c((crc), (data), (len)))
#define FIN_CRC32C(crc) ((crc) ^= 0xFFFFFFFF)
+extern pg_crc32c (*pg_comp_crc32c) (pg_crc32c crc, const void *data, size_t len);
extern pg_crc32c pg_comp_crc32c_armv8(pg_crc32c crc, const void *data, size_t len);
+#ifdef USE_PMULL_CRC32C_WITH_RUNTIME_CHECK
+extern pg_crc32c pg_comp_crc32c_pmull(pg_crc32c crc, const void *data, size_t len);
+#endif
#elif defined(USE_LOONGARCH_CRC32C)
/* Use LoongArch CRCC instructions. */
@@ -131,8 +140,8 @@ extern pg_crc32c pg_comp_crc32c_loongarch(pg_crc32c crc, const void *data, size_
#elif defined(USE_ARMV8_CRC32C_WITH_RUNTIME_CHECK)
/*
- * Use ARMv8 instructions, but perform a runtime check first
- * to check that they are available.
+ * Use either ARMv8 CRC Extension or CRYPTO Extension (PMULL) instructions,
+ * but perform a runtime check first to check that they are available.
*/
#define COMP_CRC32C(crc, data, len) \
((crc) = pg_comp_crc32c((crc), (data), (len)))
@@ -141,6 +150,9 @@ extern pg_crc32c pg_comp_crc32c_loongarch(pg_crc32c crc, const void *data, size_
extern pg_crc32c pg_comp_crc32c_sb8(pg_crc32c crc, const void *data, size_t len);
extern pg_crc32c (*pg_comp_crc32c) (pg_crc32c crc, const void *data, size_t len);
extern pg_crc32c pg_comp_crc32c_armv8(pg_crc32c crc, const void *data, size_t len);
+#ifdef USE_PMULL_CRC32C_WITH_RUNTIME_CHECK
+extern pg_crc32c pg_comp_crc32c_pmull(pg_crc32c crc, const void *data, size_t len);
+#endif
#else
/*
diff --git a/src/port/meson.build b/src/port/meson.build
index 28655142ebe..9faafbbe8bf 100644
--- a/src/port/meson.build
+++ b/src/port/meson.build
@@ -93,6 +93,7 @@ replace_funcs_pos = [
# arm / aarch64
['pg_crc32c_armv8', 'USE_ARMV8_CRC32C'],
['pg_crc32c_armv8', 'USE_ARMV8_CRC32C_WITH_RUNTIME_CHECK', 'crc'],
+ ['pg_crc32c_armv8_choose', 'USE_ARMV8_CRC32C'],
['pg_crc32c_armv8_choose', 'USE_ARMV8_CRC32C_WITH_RUNTIME_CHECK'],
['pg_crc32c_sb8', 'USE_ARMV8_CRC32C_WITH_RUNTIME_CHECK'],
diff --git a/src/port/pg_crc32c_armv8.c b/src/port/pg_crc32c_armv8.c
index 039986c7b33..7d70ad055cd 100644
--- a/src/port/pg_crc32c_armv8.c
+++ b/src/port/pg_crc32c_armv8.c
@@ -20,6 +20,10 @@
#include <arm_acle.h>
#endif
+#ifdef USE_PMULL_CRC32C_WITH_RUNTIME_CHECK
+#include <arm_neon.h>
+#endif
+
#include "port/pg_crc32c.h"
pg_crc32c
@@ -77,3 +81,123 @@ pg_comp_crc32c_armv8(pg_crc32c crc, const void *data, size_t len)
return crc;
}
+
+#ifdef USE_PMULL_CRC32C_WITH_RUNTIME_CHECK
+
+/*
+ * Note: There is no copyright notice in the following generated code.
+ *
+ * We have modified the output to
+ * - match our function declaration
+ * - match whitespace to our project style
+ * - be more friendly for pgindent
+ */
+
+/* Generated by https://github.com/corsix/fast-crc32/ using: */
+/* ./generate -i neon -p crc32c -a v4e */
+/* MIT licensed */
+
+static inline
+uint64x2_t
+clmul_lo_e(uint64x2_t a, uint64x2_t b, uint64x2_t c)
+{
+ uint64x2_t r;
+
+__asm("pmull %0.1q, %2.1d, %3.1d\neor %0.16b, %0.16b, %1.16b\n":"=w"(r), "+w"(c):"w"(a), "w"(b));
+ return r;
+}
+
+static inline
+uint64x2_t
+clmul_hi_e(uint64x2_t a, uint64x2_t b, uint64x2_t c)
+{
+ uint64x2_t r;
+
+__asm("pmull2 %0.1q, %2.2d, %3.2d\neor %0.16b, %0.16b, %1.16b\n":"=w"(r), "+w"(c):"w"(a), "w"(b));
+ return r;
+}
+
+pg_crc32c
+pg_comp_crc32c_pmull(pg_crc32c crc, const void *data, size_t len)
+{
+ /* adjust names to match generated code */
+ pg_crc32c crc0 = crc;
+ const char *buf = data;
+
+ /* align to 16 bytes */
+ for (; len && ((uintptr_t) buf & 7); --len)
+ {
+ crc0 = __crc32cb(crc0, *buf++);
+ }
+ if (((uintptr_t) buf & 8) && len >= 8)
+ {
+ crc0 = __crc32cd(crc0, *(const uint64_t *) buf);
+ buf += 8;
+ len -= 8;
+ }
+
+ if (len >= 64)
+ {
+ const char *end = buf + len;
+ const char *limit = buf + len - 64;
+
+ /* First vector chunk. */
+ uint64x2_t x0 = vld1q_u64((const uint64_t *) buf),
+ y0;
+ uint64x2_t x1 = vld1q_u64((const uint64_t *) (buf + 16)),
+ y1;
+ uint64x2_t x2 = vld1q_u64((const uint64_t *) (buf + 32)),
+ y2;
+ uint64x2_t x3 = vld1q_u64((const uint64_t *) (buf + 48)),
+ y3;
+ uint64x2_t k;
+
+ {
+ static const uint64_t pg_attribute_aligned(16) k_[] = {0x740eef02, 0x9e4addf8};
+
+ k = vld1q_u64(k_);
+ }
+
+ /*
+ * pgindent complained of unmatched parens upstream:
+ *
+ * x0 = veorq_u64((uint64x2_t) {crc0, 0}, x0);
+ */
+ x0 = veorq_u64((uint64x2_t) vsetq_lane_u64(crc0, vdupq_n_u64(0), 0), x0);
+ buf += 64;
+
+ /* Main loop. */
+ while (buf <= limit)
+ {
+ y0 = clmul_lo_e(x0, k, vld1q_u64((const uint64_t *) buf)), x0 = clmul_hi_e(x0, k, y0);
+ y1 = clmul_lo_e(x1, k, vld1q_u64((const uint64_t *) (buf + 16))), x1 = clmul_hi_e(x1, k, y1);
+ y2 = clmul_lo_e(x2, k, vld1q_u64((const uint64_t *) (buf + 32))), x2 = clmul_hi_e(x2, k, y2);
+ y3 = clmul_lo_e(x3, k, vld1q_u64((const uint64_t *) (buf + 48))), x3 = clmul_hi_e(x3, k, y3);
+ buf += 64;
+ }
+
+ /* Reduce x0 ... x3 to just x0. */
+ {
+ static const uint64_t pg_attribute_aligned(16) k_[] = {0xf20c0dfe, 0x493c7d27};
+
+ k = vld1q_u64(k_);
+ }
+ y0 = clmul_lo_e(x0, k, x1), x0 = clmul_hi_e(x0, k, y0);
+ y2 = clmul_lo_e(x2, k, x3), x2 = clmul_hi_e(x2, k, y2);
+ {
+ static const uint64_t pg_attribute_aligned(16) k_[] = {0x3da6d0cb, 0xba4fc28e};
+
+ k = vld1q_u64(k_);
+ }
+ y0 = clmul_lo_e(x0, k, x2), x0 = clmul_hi_e(x0, k, y0);
+
+ /* Reduce 128 bits to 32 bits, and multiply by x^32. */
+ crc0 = __crc32cd(0, vgetq_lane_u64(x0, 0));
+ crc0 = __crc32cd(crc0, vgetq_lane_u64(x0, 1));
+ len = end - buf;
+ }
+
+ return pg_comp_crc32c_armv8(crc0, buf, len);
+}
+
+#endif
diff --git a/src/port/pg_crc32c_armv8_choose.c b/src/port/pg_crc32c_armv8_choose.c
index a1f0e540c6b..ac6e2862e8e 100644
--- a/src/port/pg_crc32c_armv8_choose.c
+++ b/src/port/pg_crc32c_armv8_choose.c
@@ -108,6 +108,27 @@ pg_crc32c_armv8_available(void)
#endif
}
+static inline bool
+pg_pmull_available(void)
+{
+#ifdef __aarch64__
+
+#ifdef HAVE_ELF_AUX_INFO
+ unsigned long value;
+
+ return elf_aux_info(AT_HWCAP, &value, sizeof(value)) == 0 &&
+ (value & HWCAP_PMULL) != 0;
+#elif defined(HAVE_GETAUXVAL)
+ return (getauxval(AT_HWCAP) & HWCAP_PMULL) != 0;
+#else
+ return false;
+#endif
+
+#else
+ return false;
+#endif
+}
+
/*
* This gets called on the first call. It replaces the function pointer
* so that subsequent calls are routed directly to the chosen implementation.
@@ -115,10 +136,23 @@ pg_crc32c_armv8_available(void)
static pg_crc32c
pg_comp_crc32c_choose(pg_crc32c crc, const void *data, size_t len)
{
+#if defined(USE_ARMV8_CRC32C_WITH_RUNTIME_CHECK)
if (pg_crc32c_armv8_available())
pg_comp_crc32c = pg_comp_crc32c_armv8;
else
pg_comp_crc32c = pg_comp_crc32c_sb8;
+#elif defined(USE_ARMV8_CRC32C)
+ /*
+ * We still set the function pointer as a fallback for the PMULL
+ * implementation.
+ */
+ pg_comp_crc32c = pg_comp_crc32c_armv8;
+#endif
+
+#ifdef USE_PMULL_CRC32C_WITH_RUNTIME_CHECK
+ if (pg_pmull_available())
+ pg_comp_crc32c = pg_comp_crc32c_pmull;
+#endif
return pg_comp_crc32c(crc, data, len);
}
--
2.52.0
[text/x-patch] v2-0002-Force-testing-on-MacOS-CI-XXX-not-for-commit.patch (790B, ../../CANWCAZYAKWTFW_1byMq44t57urk6YKoDh1z1Zp6fk5-4dOPy-w@mail.gmail.com/3-v2-0002-Force-testing-on-MacOS-CI-XXX-not-for-commit.patch)
download | inline diff:
From c871c612297d1c0526520b5bf4b5fef0713072c9 Mon Sep 17 00:00:00 2001
From: John Naylor <[email protected]>
Date: Mon, 12 Jan 2026 15:35:07 +0700
Subject: [PATCH v2 2/2] Force testing on MacOS CI XXX not for commit
---
src/port/pg_crc32c_armv8_choose.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/src/port/pg_crc32c_armv8_choose.c b/src/port/pg_crc32c_armv8_choose.c
index ac6e2862e8e..1ed66b2fb76 100644
--- a/src/port/pg_crc32c_armv8_choose.c
+++ b/src/port/pg_crc32c_armv8_choose.c
@@ -150,7 +150,7 @@ pg_comp_crc32c_choose(pg_crc32c crc, const void *data, size_t len)
#endif
#ifdef USE_PMULL_CRC32C_WITH_RUNTIME_CHECK
- if (pg_pmull_available())
+ if (true || pg_pmull_available())
pg_comp_crc32c = pg_comp_crc32c_pmull;
#endif
--
2.52.0
^ permalink raw reply [nested|flat] 5+ messages in thread
* Re: vectorized CRC on ARM64
2026-01-12 09:27 Re: vectorized CRC on ARM64 John Naylor <[email protected]>
@ 2026-06-29 08:52 ` Peter Eisentraut <[email protected]>
2026-06-29 09:35 ` Re: vectorized CRC on ARM64 John Naylor <[email protected]>
0 siblings, 1 reply; 5+ messages in thread
From: Peter Eisentraut @ 2026-06-29 08:52 UTC (permalink / raw)
To: John Naylor <[email protected]>; Nathan Bossart <[email protected]>; +Cc: PostgreSQL Hackers <[email protected]>
It looks like this patch was developed concurrently with the switch from
pg_attribute_aligned to C11 standard alignas, and it ended up being
committed with the "old" style. I propose the attached patch to update
that. It's only a stylistic change, but good for consistency and as an
example for future code.
On 03.04.26 10:22, John Naylor wrote:
> On Thu, Apr 2, 2026 at 11:17 PM Nathan Bossart <[email protected]> wrote:
>>
>> On Thu, Apr 02, 2026 at 10:53:24AM -0500, Nathan Bossart wrote:
>>> I think the new pg_comp_crc32_choose() is infinitely recursing on macOS
>>> because USE_ARMV8_CRC32C_WITH_RUNTIME_CHECK is not defined but
>>> pg_crc32c_armv8_available() returns false. If I trace through that
>>> function, I see that it's going straight to the
>>>
>>> #else
>>> return false;
>>> #endif
>>>
>>> at the end. And sure enough, both HAVE_ELF_AUX_INFO and HAVE_GETAUXVAL
>
> Ah of course.
>
>>> aren't defined in pg_config.h. I think we might need to use sysctlbyname()
>>> to determine PMULL support on macOS, but at this stage of the development
>>> cycle, I would probably lean towards just compiling in the sb8
>>> implementation.
>>
>> Hm. On second thought, that probably regresses macOS builds because it was
>> presumably using the armv8 path without runtime checks before...
>
> I went with the following for v5, and it passes MacOS on my Github CI:
>
> + /* set fallbacks */
> +#ifdef USE_ARMV8_CRC32C
> + /* On e.g. MacOS, our runtime feature detection doesn't work */
> + pg_comp_crc32c = pg_comp_crc32c_armv8;
> +#else
> + pg_comp_crc32c = pg_comp_crc32c_sb8;
> +#endif
> + [...crc and pmull checks]
>
> That should keep scalar hardware support working, but now it'll only
> use direct calls for constant inputs.
>
> I also did some benchmarking on an ARM Neoverse N1 / gcc 8.3
> (attached). There the vector loop still works well all the way down to
> the minimum input size of 64 bytes, and on long inputs it's almost
> twice as fast as scalar. For reproduceability, I slightly modified the
> benchmark we used last year, to make sure the input is aligned
> (attached but not for CI). In the end, I want to add a length check so
> that inputs smaller than 80 bytes go straight to the scalar path.
> Above 80, after alignment adjustments in the preamble, that still
> guarantees at least one loop iteration in the vector path.
>
> --
> John Naylor
> Amazon Web Services
From de2c64985662e0a788da5a4ed9f4e7358ed544f4 Mon Sep 17 00:00:00 2001
From: Peter Eisentraut <[email protected]>
Date: Mon, 29 Jun 2026 10:46:51 +0200
Subject: [PATCH] Use C11 alignas instead of pg_attribute_aligned
Replace pg_attribute_aligned with C11 alignas, for consistency with
current conventions.
Discussion: https://postgr.es/m/CANWCAZaKhE+RD5KKouUFoxx1EbUNrNhcduM1VQ=DkSDadNEFng@mail.gmail.com
---
src/port/pg_crc32c_armv8.c | 6 +++---
1 file changed, 3 insertions(+), 3 deletions(-)
diff --git a/src/port/pg_crc32c_armv8.c b/src/port/pg_crc32c_armv8.c
index 8cf838a510a..8fad2f83c24 100644
--- a/src/port/pg_crc32c_armv8.c
+++ b/src/port/pg_crc32c_armv8.c
@@ -166,7 +166,7 @@ pg_comp_crc32c_pmull(pg_crc32c crc, const void *data, size_t len)
uint64x2_t k;
{
- static const uint64_t pg_attribute_aligned(16) k_[] = {0x740eef02, 0x9e4addf8};
+ static const alignas(16) uint64_t k_[] = {0x740eef02, 0x9e4addf8};
k = vld1q_u64(k_);
}
@@ -192,14 +192,14 @@ pg_comp_crc32c_pmull(pg_crc32c crc, const void *data, size_t len)
/* Reduce x0 ... x3 to just x0. */
{
- static const uint64_t pg_attribute_aligned(16) k_[] = {0xf20c0dfe, 0x493c7d27};
+ static const alignas(16) uint64_t k_[] = {0xf20c0dfe, 0x493c7d27};
k = vld1q_u64(k_);
}
y0 = clmul_lo_e(x0, k, x1), x0 = clmul_hi_e(x0, k, y0);
y2 = clmul_lo_e(x2, k, x3), x2 = clmul_hi_e(x2, k, y2);
{
- static const uint64_t pg_attribute_aligned(16) k_[] = {0x3da6d0cb, 0xba4fc28e};
+ static const alignas(16) uint64_t k_[] = {0x3da6d0cb, 0xba4fc28e};
k = vld1q_u64(k_);
}
--
2.54.0
Attachments:
[text/plain] 0001-Use-C11-alignas-instead-of-pg_attribute_aligned.patch (1.6K, ../../[email protected]/2-0001-Use-C11-alignas-instead-of-pg_attribute_aligned.patch)
download | inline diff:
From de2c64985662e0a788da5a4ed9f4e7358ed544f4 Mon Sep 17 00:00:00 2001
From: Peter Eisentraut <[email protected]>
Date: Mon, 29 Jun 2026 10:46:51 +0200
Subject: [PATCH] Use C11 alignas instead of pg_attribute_aligned
Replace pg_attribute_aligned with C11 alignas, for consistency with
current conventions.
Discussion: https://postgr.es/m/CANWCAZaKhE+RD5KKouUFoxx1EbUNrNhcduM1VQ=DkSDadNEFng@mail.gmail.com
---
src/port/pg_crc32c_armv8.c | 6 +++---
1 file changed, 3 insertions(+), 3 deletions(-)
diff --git a/src/port/pg_crc32c_armv8.c b/src/port/pg_crc32c_armv8.c
index 8cf838a510a..8fad2f83c24 100644
--- a/src/port/pg_crc32c_armv8.c
+++ b/src/port/pg_crc32c_armv8.c
@@ -166,7 +166,7 @@ pg_comp_crc32c_pmull(pg_crc32c crc, const void *data, size_t len)
uint64x2_t k;
{
- static const uint64_t pg_attribute_aligned(16) k_[] = {0x740eef02, 0x9e4addf8};
+ static const alignas(16) uint64_t k_[] = {0x740eef02, 0x9e4addf8};
k = vld1q_u64(k_);
}
@@ -192,14 +192,14 @@ pg_comp_crc32c_pmull(pg_crc32c crc, const void *data, size_t len)
/* Reduce x0 ... x3 to just x0. */
{
- static const uint64_t pg_attribute_aligned(16) k_[] = {0xf20c0dfe, 0x493c7d27};
+ static const alignas(16) uint64_t k_[] = {0xf20c0dfe, 0x493c7d27};
k = vld1q_u64(k_);
}
y0 = clmul_lo_e(x0, k, x1), x0 = clmul_hi_e(x0, k, y0);
y2 = clmul_lo_e(x2, k, x3), x2 = clmul_hi_e(x2, k, y2);
{
- static const uint64_t pg_attribute_aligned(16) k_[] = {0x3da6d0cb, 0xba4fc28e};
+ static const alignas(16) uint64_t k_[] = {0x3da6d0cb, 0xba4fc28e};
k = vld1q_u64(k_);
}
--
2.54.0
^ permalink raw reply [nested|flat] 5+ messages in thread
* Re: vectorized CRC on ARM64
2026-01-12 09:27 Re: vectorized CRC on ARM64 John Naylor <[email protected]>
2026-06-29 08:52 ` Re: vectorized CRC on ARM64 Peter Eisentraut <[email protected]>
@ 2026-06-29 09:35 ` John Naylor <[email protected]>
2026-07-01 07:07 ` Re: vectorized CRC on ARM64 Peter Eisentraut <[email protected]>
0 siblings, 1 reply; 5+ messages in thread
From: John Naylor @ 2026-06-29 09:35 UTC (permalink / raw)
To: Peter Eisentraut <[email protected]>; +Cc: Nathan Bossart <[email protected]>; PostgreSQL Hackers <[email protected]>
On Mon, Jun 29, 2026 at 3:52 PM Peter Eisentraut <[email protected]> wrote:
>
> It looks like this patch was developed concurrently with the switch from
> pg_attribute_aligned to C11 standard alignas, and it ended up being
> committed with the "old" style. I propose the attached patch to update
> that. It's only a stylistic change, but good for consistency and as an
> example for future code.
Looks good to me.
--
John Naylor
Amazon Web Services
^ permalink raw reply [nested|flat] 5+ messages in thread
* Re: vectorized CRC on ARM64
2026-01-12 09:27 Re: vectorized CRC on ARM64 John Naylor <[email protected]>
2026-06-29 08:52 ` Re: vectorized CRC on ARM64 Peter Eisentraut <[email protected]>
2026-06-29 09:35 ` Re: vectorized CRC on ARM64 John Naylor <[email protected]>
@ 2026-07-01 07:07 ` Peter Eisentraut <[email protected]>
0 siblings, 0 replies; 5+ messages in thread
From: Peter Eisentraut @ 2026-07-01 07:07 UTC (permalink / raw)
To: John Naylor <[email protected]>; +Cc: Nathan Bossart <[email protected]>; PostgreSQL Hackers <[email protected]>
On 29.06.26 11:35, John Naylor wrote:
> On Mon, Jun 29, 2026 at 3:52 PM Peter Eisentraut <[email protected]> wrote:
>>
>> It looks like this patch was developed concurrently with the switch from
>> pg_attribute_aligned to C11 standard alignas, and it ended up being
>> committed with the "old" style. I propose the attached patch to update
>> that. It's only a stylistic change, but good for consistency and as an
>> example for future code.
>
> Looks good to me.
Committed.
^ permalink raw reply [nested|flat] 5+ messages in thread
end of thread, other threads:[~2026-07-01 07:07 UTC | newest]
Thread overview: 5+ messages (download: mbox mbox.gz follow: Atom feed)
-- links below jump to the message on this page --
2022-12-10 03:40 [PATCH v4 3/3] Allow recovery via loadable modules. Nathan Bossart <[email protected]>
2026-01-12 09:27 Re: vectorized CRC on ARM64 John Naylor <[email protected]>
2026-06-29 08:52 ` Re: vectorized CRC on ARM64 Peter Eisentraut <[email protected]>
2026-06-29 09:35 ` Re: vectorized CRC on ARM64 John Naylor <[email protected]>
2026-07-01 07:07 ` Re: vectorized CRC on ARM64 Peter Eisentraut <[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