public inbox for [email protected]
help / color / mirror / Atom feed[PATCH 4/4] Additional test for new GUC setting.
55+ messages / 7 participants
[nested] [flat]
* [PATCH 4/4] Additional test for new GUC setting.
@ 2019-08-28 05:12 Kyotaro Horiguchi <[email protected]>
0 siblings, 0 replies; 55+ messages in thread
From: Kyotaro Horiguchi @ 2019-08-28 05:12 UTC (permalink / raw)
This patchset adds new GUC variable effective_io_block_size that
controls wheter WAL-skipped tables are finally WAL-logged or
fcync'ed. All of the TAP test performs WAL-logging so this adds an
item that performs file sync.
---
src/test/recovery/t/018_wal_optimize.pl | 18 +++++++++++++++++-
1 file changed, 17 insertions(+), 1 deletion(-)
diff --git a/src/test/recovery/t/018_wal_optimize.pl b/src/test/recovery/t/018_wal_optimize.pl
index b041121745..95063ab131 100644
--- a/src/test/recovery/t/018_wal_optimize.pl
+++ b/src/test/recovery/t/018_wal_optimize.pl
@@ -11,7 +11,7 @@ use warnings;
use PostgresNode;
use TestLib;
-use Test::More tests => 26;
+use Test::More tests => 28;
sub check_orphan_relfilenodes
{
@@ -102,7 +102,23 @@ max_prepared_transactions = 1
$result = $node->safe_psql('postgres', "SELECT count(*) FROM test2a;");
is($result, qq(1),
"wal_level = $wal_level, optimized truncation with prepared transaction");
+ # Same for file sync mode
+ # Tuples inserted after the truncation should be seen.
+ $node->safe_psql('postgres', "
+ SET effective_io_block_size to 0;
+ BEGIN;
+ CREATE TABLE test2b (id serial PRIMARY KEY);
+ INSERT INTO test2b VALUES (DEFAULT);
+ TRUNCATE test2b;
+ INSERT INTO test2b VALUES (DEFAULT);
+ COMMIT;");
+ $node->stop('immediate');
+ $node->start;
+
+ $result = $node->safe_psql('postgres', "SELECT count(*) FROM test2b;");
+ is($result, qq(1),
+ "wal_level = $wal_level, optimized truncation with file-sync");
# Data file for COPY query in follow-up tests.
my $basedir = $node->basedir;
--
2.16.3
----Next_Part(Wed_Aug_28_15_42_10_2019_037)----
^ permalink raw reply [nested|flat] 55+ messages in thread
* [PATCH v23 5/5] Additional test for new GUC setting.
@ 2019-08-28 05:12 Kyotaro Horiguchi <[email protected]>
0 siblings, 0 replies; 55+ messages in thread
From: Kyotaro Horiguchi @ 2019-08-28 05:12 UTC (permalink / raw)
This patchset adds new GUC variable effective_io_block_size that
controls wheter WAL-skipped tables are finally WAL-logged or
fcync'ed. All of the TAP test performs WAL-logging so this adds an
item that performs file sync.
---
src/test/recovery/t/018_wal_optimize.pl | 38 ++++++++++++++++++++++++-
1 file changed, 37 insertions(+), 1 deletion(-)
diff --git a/src/test/recovery/t/018_wal_optimize.pl b/src/test/recovery/t/018_wal_optimize.pl
index ac62c77a42..470d4f048c 100644
--- a/src/test/recovery/t/018_wal_optimize.pl
+++ b/src/test/recovery/t/018_wal_optimize.pl
@@ -18,7 +18,7 @@ use warnings;
use PostgresNode;
use TestLib;
-use Test::More tests => 26;
+use Test::More tests => 32;
# Make sure no orphan relfilenode files exist.
sub check_orphan_relfilenodes
@@ -52,6 +52,8 @@ sub run_wal_optimize
$node->append_conf('postgresql.conf', qq(
wal_level = $wal_level
max_prepared_transactions = 1
+wal_log_hints = on
+wal_skip_threshold = 0
));
$node->start;
@@ -111,7 +113,23 @@ max_prepared_transactions = 1
$result = $node->safe_psql('postgres', "SELECT count(*) FROM test2a;");
is($result, qq(1),
"wal_level = $wal_level, optimized truncation with prepared transaction");
+ # Same for file sync mode
+ # Tuples inserted after the truncation should be seen.
+ $node->safe_psql('postgres', "
+ SET wal_skip_threshold to 0;
+ BEGIN;
+ CREATE TABLE test2b (id serial PRIMARY KEY);
+ INSERT INTO test2b VALUES (DEFAULT);
+ TRUNCATE test2b;
+ INSERT INTO test2b VALUES (DEFAULT);
+ COMMIT;");
+
+ $node->stop('immediate');
+ $node->start;
+ $result = $node->safe_psql('postgres', "SELECT count(*) FROM test2b;");
+ is($result, qq(1),
+ "wal_level = $wal_level, optimized truncation with file-sync");
# Data file for COPY query in follow-up tests.
my $basedir = $node->basedir;
@@ -187,6 +205,24 @@ max_prepared_transactions = 1
is($result, qq(3),
"wal_level = $wal_level, SET TABLESPACE in subtransaction");
+ $node->safe_psql('postgres', "
+ BEGIN;
+ CREATE TABLE test3a5 (c int PRIMARY KEY);
+ SAVEPOINT q; INSERT INTO test3a5 VALUES (1); ROLLBACK TO q;
+ CHECKPOINT;
+ INSERT INTO test3a5 VALUES (1); -- set index hint bit
+ INSERT INTO test3a5 VALUES (2);
+ COMMIT;");
+ $node->stop('immediate');
+ $node->start;
+ $result = $node->psql('postgres', );
+ my($ret, $stdout, $stderr) = $node->psql(
+ 'postgres', "INSERT INTO test3a5 VALUES (2);");
+ is($ret, qq(3),
+ "wal_level = $wal_level, unique index LP_DEAD");
+ like($stderr, qr/violates unique/,
+ "wal_level = $wal_level, unique index LP_DEAD message");
+
# UPDATE touches two buffers; one is BufferNeedsWAL(); the other is not.
$node->safe_psql('postgres', "
BEGIN;
--
2.23.0
----Next_Part(Wed_Nov__6_17_29_27_2019_575)----
^ permalink raw reply [nested|flat] 55+ messages in thread
* [PATCH v23 5/5] Additional test for new GUC setting.
@ 2019-08-28 05:12 Kyotaro Horiguchi <[email protected]>
0 siblings, 0 replies; 55+ messages in thread
From: Kyotaro Horiguchi @ 2019-08-28 05:12 UTC (permalink / raw)
This patchset adds new GUC variable effective_io_block_size that
controls wheter WAL-skipped tables are finally WAL-logged or
fcync'ed. All of the TAP test performs WAL-logging so this adds an
item that performs file sync.
---
src/test/recovery/t/018_wal_optimize.pl | 38 ++++++++++++++++++++++++-
1 file changed, 37 insertions(+), 1 deletion(-)
diff --git a/src/test/recovery/t/018_wal_optimize.pl b/src/test/recovery/t/018_wal_optimize.pl
index ac62c77a42..470d4f048c 100644
--- a/src/test/recovery/t/018_wal_optimize.pl
+++ b/src/test/recovery/t/018_wal_optimize.pl
@@ -18,7 +18,7 @@ use warnings;
use PostgresNode;
use TestLib;
-use Test::More tests => 26;
+use Test::More tests => 32;
# Make sure no orphan relfilenode files exist.
sub check_orphan_relfilenodes
@@ -52,6 +52,8 @@ sub run_wal_optimize
$node->append_conf('postgresql.conf', qq(
wal_level = $wal_level
max_prepared_transactions = 1
+wal_log_hints = on
+wal_skip_threshold = 0
));
$node->start;
@@ -111,7 +113,23 @@ max_prepared_transactions = 1
$result = $node->safe_psql('postgres', "SELECT count(*) FROM test2a;");
is($result, qq(1),
"wal_level = $wal_level, optimized truncation with prepared transaction");
+ # Same for file sync mode
+ # Tuples inserted after the truncation should be seen.
+ $node->safe_psql('postgres', "
+ SET wal_skip_threshold to 0;
+ BEGIN;
+ CREATE TABLE test2b (id serial PRIMARY KEY);
+ INSERT INTO test2b VALUES (DEFAULT);
+ TRUNCATE test2b;
+ INSERT INTO test2b VALUES (DEFAULT);
+ COMMIT;");
+
+ $node->stop('immediate');
+ $node->start;
+ $result = $node->safe_psql('postgres', "SELECT count(*) FROM test2b;");
+ is($result, qq(1),
+ "wal_level = $wal_level, optimized truncation with file-sync");
# Data file for COPY query in follow-up tests.
my $basedir = $node->basedir;
@@ -187,6 +205,24 @@ max_prepared_transactions = 1
is($result, qq(3),
"wal_level = $wal_level, SET TABLESPACE in subtransaction");
+ $node->safe_psql('postgres', "
+ BEGIN;
+ CREATE TABLE test3a5 (c int PRIMARY KEY);
+ SAVEPOINT q; INSERT INTO test3a5 VALUES (1); ROLLBACK TO q;
+ CHECKPOINT;
+ INSERT INTO test3a5 VALUES (1); -- set index hint bit
+ INSERT INTO test3a5 VALUES (2);
+ COMMIT;");
+ $node->stop('immediate');
+ $node->start;
+ $result = $node->psql('postgres', );
+ my($ret, $stdout, $stderr) = $node->psql(
+ 'postgres', "INSERT INTO test3a5 VALUES (2);");
+ is($ret, qq(3),
+ "wal_level = $wal_level, unique index LP_DEAD");
+ like($stderr, qr/violates unique/,
+ "wal_level = $wal_level, unique index LP_DEAD message");
+
# UPDATE touches two buffers; one is BufferNeedsWAL(); the other is not.
$node->safe_psql('postgres', "
BEGIN;
--
2.23.0
----Next_Part(Wed_Nov__6_17_29_27_2019_575)----
^ permalink raw reply [nested|flat] 55+ messages in thread
* [PATCH v21 5/5] Additional test for new GUC setting.
@ 2019-08-28 05:12 Kyotaro Horiguchi <[email protected]>
0 siblings, 0 replies; 55+ messages in thread
From: Kyotaro Horiguchi @ 2019-08-28 05:12 UTC (permalink / raw)
This patchset adds new GUC variable effective_io_block_size that
controls wheter WAL-skipped tables are finally WAL-logged or
fcync'ed. All of the TAP test performs WAL-logging so this adds an
item that performs file sync.
---
src/test/recovery/t/018_wal_optimize.pl | 38 ++++++++++++++++++++++++-
1 file changed, 37 insertions(+), 1 deletion(-)
diff --git a/src/test/recovery/t/018_wal_optimize.pl b/src/test/recovery/t/018_wal_optimize.pl
index b041121745..ba9185e2ba 100644
--- a/src/test/recovery/t/018_wal_optimize.pl
+++ b/src/test/recovery/t/018_wal_optimize.pl
@@ -11,7 +11,7 @@ use warnings;
use PostgresNode;
use TestLib;
-use Test::More tests => 26;
+use Test::More tests => 32;
sub check_orphan_relfilenodes
{
@@ -43,6 +43,8 @@ sub run_wal_optimize
$node->append_conf('postgresql.conf', qq(
wal_level = $wal_level
max_prepared_transactions = 1
+wal_log_hints = on
+wal_skip_threshold = 0
));
$node->start;
@@ -102,7 +104,23 @@ max_prepared_transactions = 1
$result = $node->safe_psql('postgres', "SELECT count(*) FROM test2a;");
is($result, qq(1),
"wal_level = $wal_level, optimized truncation with prepared transaction");
+ # Same for file sync mode
+ # Tuples inserted after the truncation should be seen.
+ $node->safe_psql('postgres', "
+ SET wal_skip_threshold to 0;
+ BEGIN;
+ CREATE TABLE test2b (id serial PRIMARY KEY);
+ INSERT INTO test2b VALUES (DEFAULT);
+ TRUNCATE test2b;
+ INSERT INTO test2b VALUES (DEFAULT);
+ COMMIT;");
+
+ $node->stop('immediate');
+ $node->start;
+ $result = $node->safe_psql('postgres', "SELECT count(*) FROM test2b;");
+ is($result, qq(1),
+ "wal_level = $wal_level, optimized truncation with file-sync");
# Data file for COPY query in follow-up tests.
my $basedir = $node->basedir;
@@ -178,6 +196,24 @@ max_prepared_transactions = 1
is($result, qq(3),
"wal_level = $wal_level, SET TABLESPACE in subtransaction");
+ $node->safe_psql('postgres', "
+ BEGIN;
+ CREATE TABLE test3a5 (c int PRIMARY KEY);
+ SAVEPOINT q; INSERT INTO test3a5 VALUES (1); ROLLBACK TO q;
+ CHECKPOINT;
+ INSERT INTO test3a5 VALUES (1); -- set index hint bit
+ INSERT INTO test3a5 VALUES (2);
+ COMMIT;");
+ $node->stop('immediate');
+ $node->start;
+ $result = $node->psql('postgres', );
+ my($ret, $stdout, $stderr) = $node->psql(
+ 'postgres', "INSERT INTO test3a5 VALUES (2);");
+ is($ret, qq(3),
+ "wal_level = $wal_level, unique index LP_DEAD");
+ like($stderr, qr/violates unique/,
+ "wal_level = $wal_level, unique index LP_DEAD message");
+
# UPDATE touches two buffers; one is BufferNeedsWAL(); the other is not.
$node->safe_psql('postgres', "
BEGIN;
--
2.23.0
----Next_Part(Fri_Oct_25_13_12_51_2019_465)----
^ permalink raw reply [nested|flat] 55+ messages in thread
* Re: standby recovery fails (tablespace related) (tentative patch and discussion)
@ 2022-04-05 08:18 Kyotaro Horiguchi <[email protected]>
0 siblings, 1 reply; 55+ messages in thread
From: Kyotaro Horiguchi @ 2022-04-05 08:18 UTC (permalink / raw)
To: [email protected]; +Cc: [email protected]; [email protected]; [email protected]; [email protected]; pgsql-hackers
At Tue, 05 Apr 2022 16:38:06 +0900 (JST), Kyotaro Horiguchi <[email protected]> wrote in
> At Tue, 05 Apr 2022 11:16:44 +0900 (JST), Kyotaro Horiguchi <[email protected]> wrote in
> > So, I have the following points in my mind for now.
> >
> > - We create the directory "since we know it is just tentative state".
> >
> > - Then, check that no directory in pg_tblspc when reaching consistency
> > when allow_in_place_tablespaces is false.
> >
> > - Leave the log_invalid_page() mechanism alone as it is always result
> > in a corrpt page if a differential WAL record is applied on a newly
> > created page that should have been exist.
> >
> > However, while working on it, I found that I found that recovery faces
> > missing tablespace directories *after* reaching consistency. I'm
> > examining that further.
>
> Okay, it was my thinko.
>
> This is the first cut of the above.
It had an unused variable for Windows.
regards.
--
Kyotaro Horiguchi
NTT Open Source Software Center
Attachments:
[text/x-patch] v23-0001-Fix-replay-of-create-database-records-on-standby.patch (18.3K, ../../[email protected]/2-v23-0001-Fix-replay-of-create-database-records-on-standby.patch)
download | inline diff:
From 1e7f5e5e10ea504e168d01b3db8be12c2f63b6d6 Mon Sep 17 00:00:00 2001
From: Kyotaro Horiguchi <[email protected]>
Date: Tue, 5 Apr 2022 15:31:45 +0900
Subject: [PATCH v23] Fix replay of create database records on standby
Crash recovery on standby may encounter missing directories when
replaying create database WAL records. Prior to this patch, the
standby would fail to recover in such a case. However, the
directories could be legitimately missing. Consider a sequence of WAL
records as follows:
CREATE DATABASE
DROP DATABASE
DROP TABLESPACE
If, after replaying the last WAL record and removing the tablespace
directory, the standby crashes and has to replay the create database
record again, the crash recovery must be able to move on.
This patch allows missing tablespaces to be created during recovery
before reaching consistency. The tablespaces are created as real
directories that should not exists but will be removed until reaching
consistency. CheckRecoveryConsistency is responsible to make sure they
have disappeared.
Similar to log_invalid_page mechanism, the GUC ignore_invalid_pages
turns into PANIC errors detected by this patch into WARNING, which
allows continueing recovery.
Diagnosed-by: Paul Guo <[email protected]>
Author: Paul Guo <[email protected]>
Author: Kyotaro Horiguchi <[email protected]>
Author: Asim R Praveen <[email protected]>
Discussion: https://postgr.es/m/CAEET0ZGx9AvioViLf7nbR_8tH9-=27DN5xWJ2P9-ROH16e4JUA@mail.gmail.com
---
doc/src/sgml/config.sgml | 5 +-
src/backend/access/transam/xlogrecovery.c | 55 +++++++
src/backend/commands/dbcommands.c | 71 +++++++++
src/backend/commands/tablespace.c | 28 +---
src/backend/utils/misc/guc.c | 8 +-
src/include/access/xlogutils.h | 2 +
src/test/recovery/t/029_replay_tsp_drops.pl | 155 ++++++++++++++++++++
7 files changed, 292 insertions(+), 32 deletions(-)
create mode 100644 src/test/recovery/t/029_replay_tsp_drops.pl
diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml
index 43e4ade83e..c22229468b 100644
--- a/doc/src/sgml/config.sgml
+++ b/doc/src/sgml/config.sgml
@@ -11228,11 +11228,12 @@ LOG: CleanUpLock: deleting: lock(0xb7acd844) id(24688,24696,0,0,0,1)
<listitem>
<para>
If set to <literal>off</literal> (the default), detection of
- WAL records having references to invalid pages during
+ WAL records having references to invalid pages or
+ WAL records resulting in invalid directory operations during
recovery causes <productname>PostgreSQL</productname> to
raise a PANIC-level error, aborting the recovery. Setting
<varname>ignore_invalid_pages</varname> to <literal>on</literal>
- causes the system to ignore invalid page references in WAL records
+ causes the system to ignore invalid actions caused by such WAL records
(but still report a warning), and continue the recovery.
This behavior may <emphasis>cause crashes, data loss,
propagate or hide corruption, or other serious problems</emphasis>.
diff --git a/src/backend/access/transam/xlogrecovery.c b/src/backend/access/transam/xlogrecovery.c
index 8d2395dae2..18dcc452ca 100644
--- a/src/backend/access/transam/xlogrecovery.c
+++ b/src/backend/access/transam/xlogrecovery.c
@@ -1986,6 +1986,53 @@ xlogrecovery_redo(XLogReaderState *record, TimeLineID replayTLI)
}
}
+/*
+ * Makes sure that ./pg_tblspc directory doesn't contain a real directory.
+ *
+ * This is intended to be called after reaching consistency.
+ * ignore_invalid_pages=on turns into the PANIC error into WARNING so that
+ * recovery can continue.
+ *
+ * Note that it is the normal behavior when allow_in_place_tablespaces=on, but
+ * we don't bother caring that case since it is a developer-only setting.
+ */
+static void
+CheckTablespaceDirectory(void)
+{
+ char *tblspc_path = "./pg_tblspc";
+ DIR *dir;
+ struct dirent *de;
+
+ dir = AllocateDir(tblspc_path);
+ while ((de = ReadDir(dir, tblspc_path)) != NULL)
+ {
+ char path[MAXPGPATH];
+ char *p;
+#ifndef WIN32
+ struct stat st;
+#endif
+
+ /* Skip entries of non-oid names */
+ for (p = de->d_name; *p && isdigit(*p); p++);
+ if (*p)
+ continue;
+
+ snprintf(path, MAXPGPATH, "%s/%s", tblspc_path, de->d_name);
+
+#ifndef WIN32
+ if (lstat(path, &st) < 0)
+ ereport(ERROR, errcode_for_file_access(),
+ errmsg("could not stat file \"%s\": %m", path));
+
+ if (!S_ISLNK(st.st_mode))
+#else
+ if (!pgwin32_is_junction(path))
+#endif
+ elog(ignore_invalid_pages ? WARNING : PANIC,
+ "real directory found in pg_tblspc directory: %s", de->d_name);
+ }
+}
+
/*
* Checks if recovery has reached a consistent state. When consistency is
* reached and we have a valid starting standby snapshot, tell postmaster
@@ -2051,6 +2098,14 @@ CheckRecoveryConsistency(void)
ereport(LOG,
(errmsg("consistent recovery state reached at %X/%X",
LSN_FORMAT_ARGS(lastReplayedEndRecPtr))));
+
+ /*
+ * Check that pg_tblspc doesn't contain a real
+ * directory. Database/CREATE_* records may create a tablespace
+ * directory that should have been removed until consistency is
+ * reached.
+ */
+ CheckTablespaceDirectory();
}
/*
diff --git a/src/backend/commands/dbcommands.c b/src/backend/commands/dbcommands.c
index df16533901..910101da01 100644
--- a/src/backend/commands/dbcommands.c
+++ b/src/backend/commands/dbcommands.c
@@ -30,6 +30,7 @@
#include "access/tableam.h"
#include "access/xact.h"
#include "access/xloginsert.h"
+#include "access/xlogrecovery.h"
#include "access/xlogutils.h"
#include "catalog/catalog.h"
#include "catalog/dependency.h"
@@ -47,6 +48,7 @@
#include "commands/defrem.h"
#include "commands/seclabel.h"
#include "commands/tablespace.h"
+#include "common/file_perm.h"
#include "mb/pg_wchar.h"
#include "miscadmin.h"
#include "pgstat.h"
@@ -62,6 +64,7 @@
#include "utils/acl.h"
#include "utils/builtins.h"
#include "utils/fmgroids.h"
+#include "utils/guc.h"
#include "utils/pg_locale.h"
#include "utils/relmapper.h"
#include "utils/snapmgr.h"
@@ -135,6 +138,7 @@ static void CreateDirAndVersionFile(char *dbpath, Oid dbid, Oid tsid,
bool isRedo);
static void CreateDatabaseUsingFileCopy(Oid src_dboid, Oid dboid, Oid src_tsid,
Oid dst_tsid);
+static void maybe_create_directory(char *path);
/*
* Create a new database using the WAL_LOG strategy.
@@ -3003,6 +3007,43 @@ get_database_name(Oid dbid)
return result;
}
+/*
+ * maybe_create_directory()
+ *
+ * During recovery, there's a case where we validly need to recover a missing
+ * tablespace directory so that recovery can continue. This happens when
+ * recovery wants to create a database but the holding tablespace has been
+ * removed before the server stopped. Since we expect that the directory will
+ * be gone before reaching recovery consistency, and we have no knowledge about
+ * the tablespace other than its OID here, we create a real directory under
+ * pg_tblspc here instead of restoring the symlink. ignore_invalid_pages=on
+ * reduces the error level so that recovery can continue.
+ */
+static void
+maybe_create_directory(char *path)
+{
+ struct stat st;
+
+ Assert(RecoveryInProgress());
+
+ if (stat(path, &st) == 0)
+ return;
+
+ /* XXX: Do we make sure that the path is under pg_tblspc? */
+
+ if (reachedConsistency && !ignore_invalid_pages)
+ ereport(PANIC,
+ errmsg("missing directory \"%s\"", path));
+
+ elog(reachedConsistency ? WARNING : DEBUG1,
+ "creating missing directory: %s", path);
+
+ if (pg_mkdir_p(path, pg_dir_create_mode) != 0)
+ ereport(PANIC,
+ errmsg("could not create missing directory \"%s\": %m", path));
+}
+
+
/*
* DATABASE resource manager's routines
*/
@@ -3039,6 +3080,30 @@ dbase_redo(XLogReaderState *record)
dst_path)));
}
+ if (stat(dst_path, &st) < 0)
+ {
+ char *parent_path;
+
+ if (errno != ENOENT)
+ ereport(FATAL,
+ errmsg("could not stat directory \"%s\": %m",
+ dst_path));
+
+ /* create the parent directory if needed and valid */
+ parent_path = pstrdup(dst_path);
+ get_parent_directory(parent_path);
+ maybe_create_directory(parent_path);
+ }
+
+ /*
+ * There's a case where the copy source directory is missing for the
+ * same reason above. Create the emtpy source directory so that
+ * copydir below doesn't fail. The directory will be dropped soon by
+ * recovery.
+ */
+ if (stat(src_path, &st) < 0 && errno == ENOENT)
+ maybe_create_directory(src_path);
+
/*
* Force dirty buffers out to disk, to ensure source database is
* up-to-date for the copy.
@@ -3057,9 +3122,15 @@ dbase_redo(XLogReaderState *record)
xl_dbase_create_wal_log_rec *xlrec =
(xl_dbase_create_wal_log_rec *) XLogRecGetData(record);
char *dbpath;
+ char *parent_path;
dbpath = GetDatabasePath(xlrec->db_id, xlrec->tablespace_id);
+ /* create the parent directory if needed and valid */
+ parent_path = pstrdup(dbpath);
+ get_parent_directory(parent_path);
+ maybe_create_directory(parent_path);
+
/* Create the database directory with the version file. */
CreateDirAndVersionFile(dbpath, xlrec->db_id, xlrec->tablespace_id,
true);
diff --git a/src/backend/commands/tablespace.c b/src/backend/commands/tablespace.c
index 40514ab550..675f578dfe 100644
--- a/src/backend/commands/tablespace.c
+++ b/src/backend/commands/tablespace.c
@@ -155,8 +155,6 @@ TablespaceCreateDbspace(Oid spcNode, Oid dbNode, bool isRedo)
/* Directory creation failed? */
if (MakePGDirectory(dir) < 0)
{
- char *parentdir;
-
/* Failure other than not exists or not in WAL replay? */
if (errno != ENOENT || !isRedo)
ereport(ERROR,
@@ -169,32 +167,8 @@ TablespaceCreateDbspace(Oid spcNode, Oid dbNode, bool isRedo)
* continue by creating simple parent directories rather
* than a symlink.
*/
-
- /* create two parents up if not exist */
- parentdir = pstrdup(dir);
- get_parent_directory(parentdir);
- get_parent_directory(parentdir);
- /* Can't create parent and it doesn't already exist? */
- if (MakePGDirectory(parentdir) < 0 && errno != EEXIST)
- ereport(ERROR,
- (errcode_for_file_access(),
- errmsg("could not create directory \"%s\": %m",
- parentdir)));
- pfree(parentdir);
-
- /* create one parent up if not exist */
- parentdir = pstrdup(dir);
- get_parent_directory(parentdir);
- /* Can't create parent and it doesn't already exist? */
- if (MakePGDirectory(parentdir) < 0 && errno != EEXIST)
- ereport(ERROR,
- (errcode_for_file_access(),
- errmsg("could not create directory \"%s\": %m",
- parentdir)));
- pfree(parentdir);
-
/* Create database directory */
- if (MakePGDirectory(dir) < 0)
+ if (pg_mkdir_p(dir, pg_dir_create_mode) < 0)
ereport(ERROR,
(errcode_for_file_access(),
errmsg("could not create directory \"%s\": %m",
diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c
index 9e8ab1420d..9134a73d3d 100644
--- a/src/backend/utils/misc/guc.c
+++ b/src/backend/utils/misc/guc.c
@@ -42,6 +42,7 @@
#include "access/xact.h"
#include "access/xlog_internal.h"
#include "access/xlogrecovery.h"
+#include "access/xlogutils.h"
#include "catalog/namespace.h"
#include "catalog/objectaccess.h"
#include "catalog/pg_authid.h"
@@ -139,7 +140,6 @@ extern int CommitSiblings;
extern char *default_tablespace;
extern char *temp_tablespaces;
extern bool ignore_checksum_failure;
-extern bool ignore_invalid_pages;
extern bool synchronize_seqscans;
#ifdef TRACE_SYNCSCAN
@@ -1304,10 +1304,12 @@ static struct config_bool ConfigureNamesBool[] =
{"ignore_invalid_pages", PGC_POSTMASTER, DEVELOPER_OPTIONS,
gettext_noop("Continues recovery after an invalid pages failure."),
gettext_noop("Detection of WAL records having references to "
- "invalid pages during recovery causes PostgreSQL to "
+ "invalid pages or WAL records resulting in invalid "
+ "directory operations during "
+ "recovery that cause PostgreSQL"
"raise a PANIC-level error, aborting the recovery. "
"Setting ignore_invalid_pages to true causes "
- "the system to ignore invalid page references "
+ "the system to ignore those inconsistencies "
"in WAL records (but still report a warning), "
"and continue recovery. This behavior may cause "
"crashes, data loss, propagate or hide corruption, "
diff --git a/src/include/access/xlogutils.h b/src/include/access/xlogutils.h
index 64708949db..d88661997f 100644
--- a/src/include/access/xlogutils.h
+++ b/src/include/access/xlogutils.h
@@ -54,6 +54,8 @@ typedef enum
extern HotStandbyState standbyState;
+extern bool ignore_invalid_pages;
+
#define InHotStandby (standbyState >= STANDBY_SNAPSHOT_PENDING)
diff --git a/src/test/recovery/t/029_replay_tsp_drops.pl b/src/test/recovery/t/029_replay_tsp_drops.pl
new file mode 100644
index 0000000000..b401ab8072
--- /dev/null
+++ b/src/test/recovery/t/029_replay_tsp_drops.pl
@@ -0,0 +1,155 @@
+
+# Copyright (c) 2021-2022, PostgreSQL Global Development Group
+
+#
+# Tests relating to PostgreSQL crash recovery and redo
+#
+use strict;
+use warnings;
+use PostgreSQL::Test::Cluster;
+use PostgreSQL::Test::Utils;
+use Test::More;
+
+sub test_tablespace
+{
+ my ($strategy) = @_;
+
+ my $node_primary = PostgreSQL::Test::Cluster->new("primary1_$strategy");
+ $node_primary->init(allows_streaming => 1);
+ $node_primary->start;
+ $node_primary->psql('postgres',
+ qq[
+ SET allow_in_place_tablespaces=on;
+ CREATE TABLESPACE dropme_ts1 LOCATION '';
+ CREATE TABLESPACE dropme_ts2 LOCATION '';
+ CREATE TABLESPACE source_ts LOCATION '';
+ CREATE TABLESPACE target_ts LOCATION '';
+ CREATE DATABASE template_db IS_TEMPLATE = true;
+ ]);
+ my $backup_name = 'my_backup';
+ $node_primary->backup($backup_name);
+
+ my $node_standby = PostgreSQL::Test::Cluster->new("standby2_$strategy");
+ $node_standby->init_from_backup($node_primary, $backup_name, has_streaming => 1);
+ $node_standby->append_conf('postgresql.conf', "ignore_invalid_pages = on");
+ $node_standby->start;
+
+ # Make sure connection is made
+ $node_primary->poll_query_until(
+ 'postgres', 'SELECT count(*) = 1 FROM pg_stat_replication');
+
+ $node_standby->safe_psql('postgres', 'CHECKPOINT');
+
+ # Do immediate shutdown just after a sequence of CREAT DATABASE / DROP
+ # DATABASE / DROP TABLESPACE. This causes CREATE DATABASE WAL records
+ # to be applied to already-removed directories.
+ my $query = q[
+ CREATE DATABASE dropme_db1 WITH TABLESPACE dropme_ts1 STRATEGY=<STRATEGY>;
+ CREATE TABLE t (a int) TABLESPACE dropme_ts2;
+ CREATE DATABASE dropme_db2 WITH TABLESPACE dropme_ts2 STRATEGY=<STRATEGY>;
+ CREATE DATABASE moveme_db TABLESPACE source_ts STRATEGY=<STRATEGY>;
+ ALTER DATABASE moveme_db SET TABLESPACE target_ts;
+ CREATE DATABASE newdb TEMPLATE template_db STRATEGY=<STRATEGY>;
+ ALTER DATABASE template_db IS_TEMPLATE = false;
+ DROP DATABASE dropme_db1;
+ DROP TABLE t;
+ DROP DATABASE dropme_db2; DROP TABLESPACE dropme_ts2;
+ DROP TABLESPACE source_ts;
+ DROP DATABASE template_db;];
+
+ $query =~ s/<STRATEGY>/$strategy/g;
+ $node_primary->safe_psql('postgres', $query);
+ $node_primary->wait_for_catchup($node_standby, 'replay',
+ $node_primary->lsn('replay'));
+
+ # show "create missing directory" log message
+ $node_standby->safe_psql('postgres',
+ "ALTER SYSTEM SET log_min_messages TO debug1;");
+ $node_standby->stop('immediate');
+ # Should restart ignoring directory creation error.
+ is($node_standby->start(fail_ok => 1), 1);
+ $node_standby->stop('immediate');
+}
+
+test_tablespace("FILE_COPY");
+test_tablespace("WAL_LOG");
+
+# Ensure that a missing tablespace directory during create database
+# replay immediately causes panic if the standby has already reached
+# consistent state (archive recovery is in progress). This is
+# effective only for CREATE DATABASE WITH STRATEGY=FILE_COPY.
+
+my $node_primary = PostgreSQL::Test::Cluster->new('primary2');
+$node_primary->init(allows_streaming => 1);
+$node_primary->start;
+
+# Create tablespace
+$node_primary->safe_psql('postgres', q[
+ SET allow_in_place_tablespaces=on;
+ CREATE TABLESPACE ts1 LOCATION '']);
+$node_primary->safe_psql('postgres', "CREATE DATABASE db1 WITH TABLESPACE ts1 STRATEGY=FILE_COPY");
+
+# Take backup
+my $backup_name = 'my_backup';
+$node_primary->backup($backup_name);
+my $node_standby = PostgreSQL::Test::Cluster->new('standby3');
+$node_standby->init_from_backup($node_primary, $backup_name, has_streaming => 1);
+$node_standby->append_conf('postgresql.conf', "ignore_invalid_pages = on");
+$node_standby->start;
+
+# Make sure standby reached consistency and starts accepting connections
+$node_standby->poll_query_until('postgres', 'SELECT 1', '1');
+
+# Remove standby tablespace directory so it will be missing when
+# replay resumes.
+my $tspoid = $node_standby->safe_psql('postgres',
+ "SELECT oid FROM pg_tablespace WHERE spcname = 'ts1';");
+my $tspdir = $node_standby->data_dir . "/pg_tblspc/$tspoid";
+File::Path::rmtree($tspdir);
+
+my $logstart = get_log_size($node_standby);
+
+# Create a database in the tablespace and a table in default tablespace
+$node_primary->safe_psql('postgres',
+ q[CREATE TABLE should_not_replay_insertion(a int);
+ CREATE DATABASE db2 WITH TABLESPACE ts1 STRATEGY=FILE_COPY;
+ INSERT INTO should_not_replay_insertion VALUES (1);]);
+
+# Standby should fail and should not silently skip replaying the wal
+# In this test, PANIC turns into WARNING by ignore_invalid_pages.
+# Check the log messages instead of confirming standby failure.
+my $max_attempts = $PostgreSQL::Test::Utils::timeout_default;
+while ($max_attempts-- >= 0)
+{
+ last if (find_in_log(
+ $node_standby,
+ "WARNING: creating missing directory: pg_tblspc/",
+ $logstart));
+ sleep 1;
+}
+ok($max_attempts > 0, "invalid directory creation is detected");
+
+done_testing();
+
+
+# return the size of logfile of $node in bytes
+sub get_log_size
+{
+ my ($node) = @_;
+
+ return (stat $node->logfile)[7];
+}
+
+# find $pat in logfile of $node after $off-th byte
+sub find_in_log
+{
+ my ($node, $pat, $off) = @_;
+
+ $off = 0 unless defined $off;
+ my $log = PostgreSQL::Test::Utils::slurp_file($node->logfile);
+ return 0 if (length($log) <= $off);
+
+ $log = substr($log, $off);
+
+ return $log =~ m/$pat/;
+}
--
2.27.0
^ permalink raw reply [nested|flat] 55+ messages in thread
* Re: standby recovery fails (tablespace related) (tentative patch and discussion)
@ 2022-07-13 16:43 Alvaro Herrera <[email protected]>
parent: Kyotaro Horiguchi <[email protected]>
0 siblings, 1 reply; 55+ messages in thread
From: Alvaro Herrera @ 2022-07-13 16:43 UTC (permalink / raw)
To: Kyotaro Horiguchi <[email protected]>; +Cc: [email protected]; [email protected]; [email protected]; [email protected]; pgsql-hackers
Not a review, just a preparatory rebase across some trivially
conflicting changes. I also noticed that
src/test/recovery/t/031_recovery_conflict.pl, which was added two days
after v23 was sent, and which uses allow_in_place_tablespaces, bails out
because of the checks introduced by this patch, so I made the check
routine do nothing in that case.
Anyway, here's v24.
--
Álvaro Herrera Breisgau, Deutschland — https://www.EnterpriseDB.com/
"La conclusión que podemos sacar de esos estudios es que
no podemos sacar ninguna conclusión de ellos" (Tanenbaum)
^ permalink raw reply [nested|flat] 55+ messages in thread
* Re: standby recovery fails (tablespace related) (tentative patch and discussion)
@ 2022-07-14 21:47 Alvaro Herrera <[email protected]>
parent: Alvaro Herrera <[email protected]>
0 siblings, 2 replies; 55+ messages in thread
From: Alvaro Herrera @ 2022-07-14 21:47 UTC (permalink / raw)
To: Kyotaro Horiguchi <[email protected]>; +Cc: [email protected]; [email protected]; [email protected]; [email protected]; pgsql-hackers
Here's a couple of fixups. 0001 is the same as before. In 0002 I think
CheckTablespaceDirectory ends up easier to read if we split out the test
for validity of the link. Looking at that again, I think we don't need
to piggyback on ignore_invalid_pages, which is already a stretch, so
let's not -- instead we can use allow_in_place_tablespaces if users need
a workaround. So that's 0003 (this bit needs more than zero docs,
however.)
0004 is straightforward: let's check for bad directories before logging
about consistent state.
After all this, I'm not sure what to think of dbase_redo. At line 3102,
is the directory supposed to exist or not? I'm confused as to what is
the expected state at that point. I rewrote this, but now I think my
rewrite continues to be confusing, so I'll have to think more about it.
Another aspect are the tests. Robert described a scenario where the
previously committed version of this patch created trouble. Do we have
a test case to cover that problematic case? I think we should strive to
cover it, if possible.
--
Álvaro Herrera PostgreSQL Developer — https://www.EnterpriseDB.com/
"The eagle never lost so much time, as
when he submitted to learn of the crow." (William Blake)
^ permalink raw reply [nested|flat] 55+ messages in thread
* Re: standby recovery fails (tablespace related) (tentative patch and discussion)
@ 2022-07-15 07:30 Kyotaro Horiguchi <[email protected]>
parent: Alvaro Herrera <[email protected]>
1 sibling, 2 replies; 55+ messages in thread
From: Kyotaro Horiguchi @ 2022-07-15 07:30 UTC (permalink / raw)
To: [email protected]; +Cc: [email protected]; [email protected]; [email protected]; [email protected]; pgsql-hackers
At Thu, 14 Jul 2022 23:47:40 +0200, Alvaro Herrera <[email protected]> wrote in
> Here's a couple of fixups. 0001 is the same as before. In 0002 I think
Thanks!
+ if (!S_ISLNK(st.st_mode))
+ #else
+ if (!pgwin32_is_junction(path))
+ #endif
+ elog(ignore_invalid_pages ? WARNING : PANIC,
+ "real directory found in pg_tblspc directory: %s", de->d_name);
A regular file with an oid-name also causes this error. Doesn't
something like "unexpected non-(sym)link entry..." work?
> CheckTablespaceDirectory ends up easier to read if we split out the test
> for validity of the link. Looking at that again, I think we don't need
> to piggyback on ignore_invalid_pages, which is already a stretch, so
> let's not -- instead we can use allow_in_place_tablespaces if users need
> a workaround. So that's 0003 (this bit needs more than zero docs,
> however.)
The result of 0003 looks good.
0002:
+is_path_tslink(const char *path)
What the "ts" of tslink stands for? If it stands for tablespace, the
function is not specific for table spaces. We already have
+ errmsg("could not stat file \"%s\": %m", path));
I'm not sure we need such correctness, but what is failing there is
lstat. I found similar codes in two places in backend and one place
in frontend. So couldn't it be moved to /common and have a more
generic name?
- dir = AllocateDir(tblspc_path);
- while ((de = ReadDir(dir, tblspc_path)) != NULL)
+ dir = AllocateDir("pg_tblspc");
+ while ((de = ReadDir(dir, "pg_tblspc")) != NULL)
xlog.c uses the macro XLOGDIR. Why don't we define TBLSPCDIR?
- for (p = de->d_name; *p && isdigit(*p); p++);
- if (*p)
+ if (strspn(de->d_name, "0123456789") != strlen(de->d_name))
continue;
The pattern "strspn != strlen" looks kind of remote, or somewhat
pedantic..
+ char path[MAXPGPATH + 10];
..
- snprintf(path, MAXPGPATH, "%s/%s", tblspc_path, de->d_name);
+ snprintf(path, sizeof(path), "pg_tblspc/%s", de->d_name);
I don't think we need the extra 10 bytes. A bit paranoic, but we can
check the return value to confirm the d_name is fully stored in the
buffer.
> 0004 is straightforward: let's check for bad directories before logging
> about consistent state.
I was about to write a comment to do this when looking 0001.
> After all this, I'm not sure what to think of dbase_redo. At line 3102,
> is the directory supposed to exist or not? I'm confused as to what is
> the expected state at that point. I rewrote this, but now I think my
> rewrite continues to be confusing, so I'll have to think more about it.
I'm not sure l3102 exactly points, but haven't we chosen to create
everything required to keep recovery going, whether it is supposed to
exist or not?
> Another aspect are the tests. Robert described a scenario where the
> previously committed version of this patch created trouble. Do we have
> a test case to cover that problematic case? I think we should strive to
> cover it, if possible.
I counldn't recall that clearly and failed to dig out from the thread,
but doesn't the "creating everything needed" strategy naturally save
that case? We could add that test, but it seems to me a little
cumbersome to confirm the test correctly detect that case..
regards.
--
Kyotaro Horiguchi
NTT Open Source Software Center
^ permalink raw reply [nested|flat] 55+ messages in thread
* Re: standby recovery fails (tablespace related) (tentative patch and discussion)
@ 2022-07-15 07:56 Alvaro Herrera <[email protected]>
parent: Kyotaro Horiguchi <[email protected]>
1 sibling, 0 replies; 55+ messages in thread
From: Alvaro Herrera @ 2022-07-15 07:56 UTC (permalink / raw)
To: Kyotaro Horiguchi <[email protected]>; +Cc: [email protected]; [email protected]; [email protected]; [email protected]; pgsql-hackers
On 2022-Jul-15, Kyotaro Horiguchi wrote:
> At Thu, 14 Jul 2022 23:47:40 +0200, Alvaro Herrera <[email protected]> wrote in
> > Here's a couple of fixups. 0001 is the same as before. In 0002 I think
>
> Thanks!
>
> + if (!S_ISLNK(st.st_mode))
> + #else
> + if (!pgwin32_is_junction(path))
> + #endif
> + elog(ignore_invalid_pages ? WARNING : PANIC,
> + "real directory found in pg_tblspc directory: %s", de->d_name);
>
> A regular file with an oid-name also causes this error. Doesn't
> something like "unexpected non-(sym)link entry..." work?
Hmm, good point. I also wonder if we need to cater for using the term
"junction point" rather than "symlink" when under Windows.
> > CheckTablespaceDirectory ends up easier to read if we split out the test
> > for validity of the link. Looking at that again, I think we don't need
> > to piggyback on ignore_invalid_pages, which is already a stretch, so
> > let's not -- instead we can use allow_in_place_tablespaces if users need
> > a workaround. So that's 0003 (this bit needs more than zero docs,
> > however.)
>
> The result of 0003 looks good.
Great, will merge.
> 0002:
> +is_path_tslink(const char *path)
>
> What the "ts" of tslink stands for? If it stands for tablespace, the
> function is not specific for table spaces.
Oh, of course.
> We already have
>
> + errmsg("could not stat file \"%s\": %m", path));
>
> I'm not sure we need such correctness, but what is failing there is
> lstat.
I'll have a look at what we use for lstat failures in other places.
> I found similar codes in two places in backend and one place
> in frontend. So couldn't it be moved to /common and have a more
> generic name?
I'll have a look at those. I had the same instinct initially ...
> - dir = AllocateDir(tblspc_path);
> - while ((de = ReadDir(dir, tblspc_path)) != NULL)
> + dir = AllocateDir("pg_tblspc");
> + while ((de = ReadDir(dir, "pg_tblspc")) != NULL)
>
> xlog.c uses the macro XLOGDIR. Why don't we define TBLSPCDIR?
Oh yes, let's do that. I'd even backpatch that, to avoid a future
backpatching gotcha.
> - for (p = de->d_name; *p && isdigit(*p); p++);
> - if (*p)
> + if (strspn(de->d_name, "0123456789") != strlen(de->d_name))
> continue;
>
> The pattern "strspn != strlen" looks kind of remote, or somewhat
> pedantic..
>
> + char path[MAXPGPATH + 10];
> ..
> - snprintf(path, MAXPGPATH, "%s/%s", tblspc_path, de->d_name);
> + snprintf(path, sizeof(path), "pg_tblspc/%s", de->d_name);
>
> I don't think we need the extra 10 bytes.
I forgot to mention this, but I just copied these bits from some other
place that processes pg_tblspc entries. It seemed to me that the
bodiless for loop was a bit too suspicious-looking.
> A bit paranoic, but we can check the return value to confirm the
> d_name is fully stored in the buffer.
Hmm ... I don't think we need to care about that in this patch. This
coding pattern is already being used in other places. If we want to
change that, let's do it everywhere, and not in an unrelated
backpatchable bug fix.
> > After all this, I'm not sure what to think of dbase_redo. At line 3102,
> > is the directory supposed to exist or not? I'm confused as to what is
> > the expected state at that point. I rewrote this, but now I think my
> > rewrite continues to be confusing, so I'll have to think more about it.
>
> I'm not sure l3102 exactly points, but haven't we chosen to create
> everything required to keep recovery going, whether it is supposed to
> exist or not?
I mean just after the two stat() calls for the target directory.
> > Another aspect are the tests. Robert described a scenario where the
> > previously committed version of this patch created trouble. Do we have
> > a test case to cover that problematic case? I think we should strive to
> > cover it, if possible.
>
> I counldn't recall that clearly and failed to dig out from the thread,
> but doesn't the "creating everything needed" strategy naturally save
> that case? We could add that test, but it seems to me a little
> cumbersome to confirm the test correctly detect that case..
Well, I *hope* it does ... but hope is no strategy, and I've frequently
been on the wrong side when trusting that untested code does what I
think it does.
Thanks for reviewing,
--
Álvaro Herrera PostgreSQL Developer — https://www.EnterpriseDB.com/
"It takes less than 2 seconds to get to 78% complete; that's a good sign.
A few seconds later it's at 90%, but it seems to have stuck there. Did
somebody make percentages logarithmic while I wasn't looking?"
http://smylers.hates-software.com/2005/09/08/1995c749.html
^ permalink raw reply [nested|flat] 55+ messages in thread
* Re: standby recovery fails (tablespace related) (tentative patch and discussion)
@ 2022-07-15 10:58 Alvaro Herrera <[email protected]>
parent: Kyotaro Horiguchi <[email protected]>
1 sibling, 1 reply; 55+ messages in thread
From: Alvaro Herrera @ 2022-07-15 10:58 UTC (permalink / raw)
To: Kyotaro Horiguchi <[email protected]>; +Cc: [email protected]; [email protected]; [email protected]; [email protected]; pgsql-hackers
On 2022-Jul-15, Kyotaro Horiguchi wrote:
> 0002:
> +is_path_tslink(const char *path)
>
> What the "ts" of tslink stands for? If it stands for tablespace, the
> function is not specific for table spaces. We already have
>
> + errmsg("could not stat file \"%s\": %m", path));
>
> I'm not sure we need such correctness, but what is failing there is
> lstat. I found similar codes in two places in backend and one place
> in frontend. So couldn't it be moved to /common and have a more
> generic name?
I wondered whether it'd be better to check whether get_dirent_type
returns PGFILETYPE_LNK. However, that doesn't deal with junction points
at all, which seems pretty odd ... I mean, isn't it rather useful as an
abstraction if it doesn't abstract away the one platform-dependent point
we have in the area?
However, looking closer I noticed that on Windows we use our own
readdir() implementation, which AFAICT includes everything to handle
reparse points as symlinks correctly in get_dirent_type. Which means
that do_pg_start_backup is wasting its time with the "#ifdef WIN32" bits
to handle junction points separately. We could just do this
diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
index b809a2152c..4966213fde 100644
--- a/src/backend/access/transam/xlog.c
+++ b/src/backend/access/transam/xlog.c
@@ -8302,13 +8302,8 @@ do_pg_backup_start(const char *backupidstr, bool fast, TimeLineID *starttli_p,
* we sometimes use allow_in_place_tablespaces to create
* directories directly under pg_tblspc, which would fail below.
*/
-#ifdef WIN32
- if (!pgwin32_is_junction(fullpath))
- continue;
-#else
if (get_dirent_type(fullpath, de, false, ERROR) != PGFILETYPE_LNK)
continue;
-#endif
#if defined(HAVE_READLINK) || defined(WIN32)
rllen = readlink(fullpath, linkpath, sizeof(linkpath));
And everything should continue to work.
--
Álvaro Herrera PostgreSQL Developer — https://www.EnterpriseDB.com/
^ permalink raw reply [nested|flat] 55+ messages in thread
* Re: standby recovery fails (tablespace related) (tentative patch and discussion)
@ 2022-07-15 12:03 Alvaro Herrera <[email protected]>
parent: Alvaro Herrera <[email protected]>
0 siblings, 0 replies; 55+ messages in thread
From: Alvaro Herrera @ 2022-07-15 12:03 UTC (permalink / raw)
To: Kyotaro Horiguchi <[email protected]>; +Cc: [email protected]; [email protected]; [email protected]; [email protected]; pgsql-hackers
On 2022-Jul-15, Alvaro Herrera wrote:
> However, looking closer I noticed that on Windows we use our own
> readdir() implementation, which AFAICT includes everything to handle
> reparse points as symlinks correctly in get_dirent_type. Which means
> that do_pg_start_backup is wasting its time with the "#ifdef WIN32" bits
> to handle junction points separately. We could just do this
>
> diff --git a/src/backend/access/transam/xlog.c b/src/backend/access/transam/xlog.c
> index b809a2152c..4966213fde 100644
> --- a/src/backend/access/transam/xlog.c
> +++ b/src/backend/access/transam/xlog.c
> @@ -8302,13 +8302,8 @@ do_pg_backup_start(const char *backupidstr, bool fast, TimeLineID *starttli_p,
> * we sometimes use allow_in_place_tablespaces to create
> * directories directly under pg_tblspc, which would fail below.
> */
> -#ifdef WIN32
> - if (!pgwin32_is_junction(fullpath))
> - continue;
> -#else
> if (get_dirent_type(fullpath, de, false, ERROR) != PGFILETYPE_LNK)
> continue;
> -#endif
>
> #if defined(HAVE_READLINK) || defined(WIN32)
> rllen = readlink(fullpath, linkpath, sizeof(linkpath));
>
> And everything should continue to work.
Hmm, but it does not:
https://cirrus-ci.com/build/4824963784900608
--
Álvaro Herrera PostgreSQL Developer — https://www.EnterpriseDB.com/
^ permalink raw reply [nested|flat] 55+ messages in thread
* Re: standby recovery fails (tablespace related) (tentative patch and discussion)
@ 2022-07-20 10:50 Alvaro Herrera <[email protected]>
parent: Alvaro Herrera <[email protected]>
1 sibling, 3 replies; 55+ messages in thread
From: Alvaro Herrera @ 2022-07-20 10:50 UTC (permalink / raw)
To: Kyotaro Horiguchi <[email protected]>; +Cc: [email protected]; [email protected]; [email protected]; [email protected]; pgsql-hackers
v26 here. I spent some time fighting the readdir() stuff for
Windows (so that get_dirent_type returns LNK for junction points)
but couldn't make it to work and was unable to figure out why.
So I ended up doing what do_pg_backup_start is already doing:
an #ifdef to call pgwin32_is_junction instead. I remove the
newly added path_is_symlink function, because I realized that
it would mean an extra syscall everywhere other than Windows.
So if somebody wants to fix get_dirent_type() so that it works properly
on Windows, we can change all these places together.
I also change the use of allow_invalid_pages to
allow_in_place_tablespaces. We could add a
separate GUC for this, but it seems overengineering.
--
Álvaro Herrera PostgreSQL Developer — https://www.EnterpriseDB.com/
"Most hackers will be perfectly comfortable conceptualizing users as entropy
sources, so let's move on." (Nathaniel Smith)
^ permalink raw reply [nested|flat] 55+ messages in thread
* Re: standby recovery fails (tablespace related) (tentative patch and discussion)
@ 2022-07-20 15:25 Alvaro Herrera <[email protected]>
parent: Alvaro Herrera <[email protected]>
2 siblings, 2 replies; 55+ messages in thread
From: Alvaro Herrera @ 2022-07-20 15:25 UTC (permalink / raw)
To: Kyotaro Horiguchi <[email protected]>; +Cc: [email protected]; [email protected]; [email protected]; [email protected]; pgsql-hackers
On 2022-Jul-20, Alvaro Herrera wrote:
> I also change the use of allow_invalid_pages to
> allow_in_place_tablespaces. We could add a
> separate GUC for this, but it seems overengineering.
Oh, but allow_in_place_tablespaces doesn't exist in versions 14 and
older, so this strategy doesn't really work.
I see the following alternatives:
1. not backpatch this fix to 14 and older
2. use a different GUC; either allow_invalid_pages as previously
suggested, or create a new one just for this purpose
3. not provide any overriding mechanism in versions 14 and older
--
Álvaro Herrera PostgreSQL Developer — https://www.EnterpriseDB.com/
"Always assume the user will do much worse than the stupidest thing
you can imagine." (Julien PUYDT)
^ permalink raw reply [nested|flat] 55+ messages in thread
* Re: standby recovery fails (tablespace related) (tentative patch and discussion)
@ 2022-07-20 16:34 Alvaro Herrera <[email protected]>
parent: Alvaro Herrera <[email protected]>
1 sibling, 0 replies; 55+ messages in thread
From: Alvaro Herrera @ 2022-07-20 16:34 UTC (permalink / raw)
To: Kyotaro Horiguchi <[email protected]>; +Cc: [email protected]; [email protected]; [email protected]; [email protected]; pgsql-hackers
On 2022-Jul-20, Alvaro Herrera wrote:
> On 2022-Jul-20, Alvaro Herrera wrote:
>
> > I also change the use of allow_invalid_pages to
> > allow_in_place_tablespaces. We could add a
> > separate GUC for this, but it seems overengineering.
>
> Oh, but allow_in_place_tablespaces doesn't exist in versions 14 and
> older, so this strategy doesn't really work.
... and get_dirent_type is new in 14, so that'll be one more hurdle.
--
Álvaro Herrera Breisgau, Deutschland — https://www.EnterpriseDB.com/
"Cuando no hay humildad las personas se degradan" (A. Christie)
^ permalink raw reply [nested|flat] 55+ messages in thread
* Re: standby recovery fails (tablespace related) (tentative patch and discussion)
@ 2022-07-21 11:01 Alvaro Herrera <[email protected]>
parent: Alvaro Herrera <[email protected]>
1 sibling, 1 reply; 55+ messages in thread
From: Alvaro Herrera @ 2022-07-21 11:01 UTC (permalink / raw)
To: Kyotaro Horiguchi <[email protected]>; +Cc: [email protected]; [email protected]; [email protected]; [email protected]; pgsql-hackers
On 2022-Jul-20, Alvaro Herrera wrote:
> I see the following alternatives:
>
> 1. not backpatch this fix to 14 and older
> 2. use a different GUC; either allow_invalid_pages as previously
> suggested, or create a new one just for this purpose
> 3. not provide any overriding mechanism in versions 14 and older
I've got no opinions on this. I don't like either 1 or 3, so I'm going
to add and backpatch a new GUC allow_recovery_tablespaces as the
override mechanism.
If others disagree with this choice, please speak up.
--
Álvaro Herrera Breisgau, Deutschland — https://www.EnterpriseDB.com/
^ permalink raw reply [nested|flat] 55+ messages in thread
* Re: standby recovery fails (tablespace related) (tentative patch and discussion)
@ 2022-07-21 11:11 Thomas Munro <[email protected]>
parent: Alvaro Herrera <[email protected]>
2 siblings, 1 reply; 55+ messages in thread
From: Thomas Munro @ 2022-07-21 11:11 UTC (permalink / raw)
To: Alvaro Herrera <[email protected]>; +Cc: Kyotaro Horiguchi <[email protected]>; Dilip Kumar <[email protected]>; Robert Haas <[email protected]>; Michael Paquier <[email protected]>; Julien Rouhaud <[email protected]>; pgsql-hackers
On Wed, Jul 20, 2022 at 10:51 PM Alvaro Herrera <[email protected]> wrote:
> v26 here. I spent some time fighting the readdir() stuff for
> Windows (so that get_dirent_type returns LNK for junction points)
> but couldn't make it to work and was unable to figure out why.
Was it because of this?
https://www.postgresql.org/message-id/CA%2BhUKGKv%2B736Pc8kSj3%3DDijDGd1eC79-uT3Vi16n7jYkcc_raw%40ma...
^ permalink raw reply [nested|flat] 55+ messages in thread
* Re: standby recovery fails (tablespace related) (tentative patch and discussion)
@ 2022-07-21 11:14 Thomas Munro <[email protected]>
parent: Alvaro Herrera <[email protected]>
0 siblings, 2 replies; 55+ messages in thread
From: Thomas Munro @ 2022-07-21 11:14 UTC (permalink / raw)
To: Alvaro Herrera <[email protected]>; +Cc: Kyotaro Horiguchi <[email protected]>; Dilip Kumar <[email protected]>; Robert Haas <[email protected]>; Michael Paquier <[email protected]>; Julien Rouhaud <[email protected]>; pgsql-hackers
On Thu, Jul 21, 2022 at 11:01 PM Alvaro Herrera <[email protected]> wrote:
> On 2022-Jul-20, Alvaro Herrera wrote:
> > I see the following alternatives:
> >
> > 1. not backpatch this fix to 14 and older
> > 2. use a different GUC; either allow_invalid_pages as previously
> > suggested, or create a new one just for this purpose
> > 3. not provide any overriding mechanism in versions 14 and older
>
> I've got no opinions on this. I don't like either 1 or 3, so I'm going
> to add and backpatch a new GUC allow_recovery_tablespaces as the
> override mechanism.
>
> If others disagree with this choice, please speak up.
Would it help if we back-patched the allow_in_place_tablespaces stuff?
I'm not sure how hard/destabilising that would be, but I could take a
look tomorrow.
^ permalink raw reply [nested|flat] 55+ messages in thread
* Re: standby recovery fails (tablespace related) (tentative patch and discussion)
@ 2022-07-21 11:17 Alvaro Herrera <[email protected]>
parent: Thomas Munro <[email protected]>
0 siblings, 0 replies; 55+ messages in thread
From: Alvaro Herrera @ 2022-07-21 11:17 UTC (permalink / raw)
To: Thomas Munro <[email protected]>; +Cc: Kyotaro Horiguchi <[email protected]>; Dilip Kumar <[email protected]>; Robert Haas <[email protected]>; Michael Paquier <[email protected]>; Julien Rouhaud <[email protected]>; pgsql-hackers
On 2022-Jul-21, Thomas Munro wrote:
> On Wed, Jul 20, 2022 at 10:51 PM Alvaro Herrera <[email protected]> wrote:
> > v26 here. I spent some time fighting the readdir() stuff for
> > Windows (so that get_dirent_type returns LNK for junction points)
> > but couldn't make it to work and was unable to figure out why.
>
> Was it because of this?
>
> https://www.postgresql.org/message-id/CA%2BhUKGKv%2B736Pc8kSj3%3DDijDGd1eC79-uT3Vi16n7jYkcc_raw%40ma...
Oh, that sounds very likely, yeah. I didn't think of testing the
FILE_ATTRIBUTE_DIRECTORY bit for junction points.
I +1 pushing both of these patches to 14. Then this patch becomes a
couple of lines shorter.
--
Álvaro Herrera PostgreSQL Developer — https://www.EnterpriseDB.com/
"Before you were born your parents weren't as boring as they are now. They
got that way paying your bills, cleaning up your room and listening to you
tell them how idealistic you are." -- Charles J. Sykes' advice to teenagers
^ permalink raw reply [nested|flat] 55+ messages in thread
* Re: standby recovery fails (tablespace related) (tentative patch and discussion)
@ 2022-07-21 11:20 Alvaro Herrera <[email protected]>
parent: Thomas Munro <[email protected]>
1 sibling, 1 reply; 55+ messages in thread
From: Alvaro Herrera @ 2022-07-21 11:20 UTC (permalink / raw)
To: Thomas Munro <[email protected]>; +Cc: Kyotaro Horiguchi <[email protected]>; Dilip Kumar <[email protected]>; Robert Haas <[email protected]>; Michael Paquier <[email protected]>; Julien Rouhaud <[email protected]>; pgsql-hackers
On 2022-Jul-21, Thomas Munro wrote:
> On Thu, Jul 21, 2022 at 11:01 PM Alvaro Herrera <[email protected]> wrote:
> > I've got no opinions on this. I don't like either 1 or 3, so I'm going
> > to add and backpatch a new GUC allow_recovery_tablespaces as the
> > override mechanism.
> >
> > If others disagree with this choice, please speak up.
>
> Would it help if we back-patched the allow_in_place_tablespaces stuff?
> I'm not sure how hard/destabilising that would be, but I could take a
> look tomorrow.
Yeah, I think that would reduce cruft. I'm not sure this is more
against backpatching policy or less, compared to adding a separate
GUC just for this bugfix.
--
Álvaro Herrera Breisgau, Deutschland — https://www.EnterpriseDB.com/
"The problem with the facetime model is not just that it's demoralizing, but
that the people pretending to work interrupt the ones actually working."
(Paul Graham)
^ permalink raw reply [nested|flat] 55+ messages in thread
* Re: standby recovery fails (tablespace related) (tentative patch and discussion)
@ 2022-07-21 11:25 Alvaro Herrera <[email protected]>
parent: Alvaro Herrera <[email protected]>
0 siblings, 1 reply; 55+ messages in thread
From: Alvaro Herrera @ 2022-07-21 11:25 UTC (permalink / raw)
To: Thomas Munro <[email protected]>; +Cc: Kyotaro Horiguchi <[email protected]>; Dilip Kumar <[email protected]>; Robert Haas <[email protected]>; Michael Paquier <[email protected]>; Julien Rouhaud <[email protected]>; pgsql-hackers
On 2022-Jul-21, Alvaro Herrera wrote:
> Yeah, I think that would reduce cruft. I'm not sure this is more
> against backpatching policy or less, compared to adding a separate
> GUC just for this bugfix.
cruft:
{
{"allow_recovery_tablespaces", PG_POSTMASTER, WAL_RECOVERY,
gettext_noop("Continues recovery after finding invalid database directories."),
gettext_noop("It is possible for tablespace drop to interfere with database creation "
"so that WAL replay is forced to create fake database directories. "
"These should have been dropped by the time recovery ends; "
"but in case they aren't, this option lets recovery continue if they "
"are present. Note that these directories must be removed manually afterwards."),
GUC_NOT_IN_SAMPLE
},
&allow_recovery_tablespaces,
false,
NULL, NULL, NULL
},
This is not a very good explanation, but I don't know how to make it
better.
--
Álvaro Herrera PostgreSQL Developer — https://www.EnterpriseDB.com/
"I think my standards have lowered enough that now I think 'good design'
is when the page doesn't irritate the living f*ck out of me." (JWZ)
^ permalink raw reply [nested|flat] 55+ messages in thread
* Re: standby recovery fails (tablespace related) (tentative patch and discussion)
@ 2022-07-22 00:20 Kyotaro Horiguchi <[email protected]>
parent: Thomas Munro <[email protected]>
1 sibling, 1 reply; 55+ messages in thread
From: Kyotaro Horiguchi @ 2022-07-22 00:20 UTC (permalink / raw)
To: [email protected]; +Cc: [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; pgsql-hackers
At Thu, 21 Jul 2022 23:14:57 +1200, Thomas Munro <[email protected]> wrote in
> On Thu, Jul 21, 2022 at 11:01 PM Alvaro Herrera <[email protected]> wrote:
> > On 2022-Jul-20, Alvaro Herrera wrote:
> > > I see the following alternatives:
> > >
> > > 1. not backpatch this fix to 14 and older
> > > 2. use a different GUC; either allow_invalid_pages as previously
> > > suggested, or create a new one just for this purpose
> > > 3. not provide any overriding mechanism in versions 14 and older
> >
> > I've got no opinions on this. I don't like either 1 or 3, so I'm going
> > to add and backpatch a new GUC allow_recovery_tablespaces as the
> > override mechanism.
> >
> > If others disagree with this choice, please speak up.
>
> Would it help if we back-patched the allow_in_place_tablespaces stuff?
> I'm not sure how hard/destabilising that would be, but I could take a
> look tomorrow.
+1. Addiotional reason for me is it is a developer option.
regards.
--
Kyotaro Horiguchi
NTT Open Source Software Center
^ permalink raw reply [nested|flat] 55+ messages in thread
* Re: standby recovery fails (tablespace related) (tentative patch and discussion)
@ 2022-07-22 01:02 Kyotaro Horiguchi <[email protected]>
parent: Alvaro Herrera <[email protected]>
0 siblings, 0 replies; 55+ messages in thread
From: Kyotaro Horiguchi @ 2022-07-22 01:02 UTC (permalink / raw)
To: [email protected]; +Cc: [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; pgsql-hackers
At Thu, 21 Jul 2022 13:25:05 +0200, Alvaro Herrera <[email protected]> wrote in
> On 2022-Jul-21, Alvaro Herrera wrote:
>
> > Yeah, I think that would reduce cruft. I'm not sure this is more
> > against backpatching policy or less, compared to adding a separate
> > GUC just for this bugfix.
>
> cruft:
>
> {
> {"allow_recovery_tablespaces", PG_POSTMASTER, WAL_RECOVERY,
> gettext_noop("Continues recovery after finding invalid database directories."),
> gettext_noop("It is possible for tablespace drop to interfere with database creation "
> "so that WAL replay is forced to create fake database directories. "
> "These should have been dropped by the time recovery ends; "
> "but in case they aren't, this option lets recovery continue if they "
> "are present. Note that these directories must be removed manually afterwards."),
> GUC_NOT_IN_SAMPLE
> },
> &allow_recovery_tablespaces,
> false,
> NULL, NULL, NULL
> },
>
> This is not a very good explanation, but I don't know how to make it
> better.
It looks a bit too detailed. I crafted the following..
Recovery can create tentative in-place tablespace directories under
pg_tblspc/. They are assumed to be removed until reaching recovery
consistency, but otherwise PostgreSQL raises a PANIC-level error,
aborting the recovery. Setting allow_recovery_tablespaces to true
causes the system to allow such directories during normal
operation. In case those directories are left after reaching
consistency, that implies data loss and metadata inconsistency and may
cause failure of future tablespace creation.
Though, after writing this, I became to think that piggy-backing on
allow_in_place_tablespaces might be a bit different..
regards.
--
Kyotaro Horiguchi
NTT Open Source Software Center
^ permalink raw reply [nested|flat] 55+ messages in thread
* Re: standby recovery fails (tablespace related) (tentative patch and discussion)
@ 2022-07-22 08:18 Alvaro Herrera <[email protected]>
parent: Kyotaro Horiguchi <[email protected]>
0 siblings, 2 replies; 55+ messages in thread
From: Alvaro Herrera @ 2022-07-22 08:18 UTC (permalink / raw)
To: Kyotaro Horiguchi <[email protected]>; +Cc: [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; pgsql-hackers
On 2022-Jul-22, Kyotaro Horiguchi wrote:
> At Thu, 21 Jul 2022 23:14:57 +1200, Thomas Munro <[email protected]> wrote in
> > Would it help if we back-patched the allow_in_place_tablespaces stuff?
> > I'm not sure how hard/destabilising that would be, but I could take a
> > look tomorrow.
>
> +1. Addiotional reason for me is it is a developer option.
OK, I'll wait for allow_in_place_tablespaces to be backpatched then.
I would like to get this fix pushed before the next set of minors, so if
you won't have time for the backpatches early enough, maybe I can work
on getting it done.
Which commits would we consider?
7170f2159fb2 Allow "in place" tablespaces.
f6f0db4d6240 Fix pg_tablespace_location() with in-place tablespaces
--
Álvaro Herrera 48°01'N 7°57'E — https://www.EnterpriseDB.com/
"Most hackers will be perfectly comfortable conceptualizing users as entropy
sources, so let's move on." (Nathaniel Smith)
^ permalink raw reply [nested|flat] 55+ messages in thread
* Re: standby recovery fails (tablespace related) (tentative patch and discussion)
@ 2022-07-22 08:49 Kyotaro Horiguchi <[email protected]>
parent: Alvaro Herrera <[email protected]>
1 sibling, 1 reply; 55+ messages in thread
From: Kyotaro Horiguchi @ 2022-07-22 08:49 UTC (permalink / raw)
To: [email protected]; +Cc: [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; pgsql-hackers
At Fri, 22 Jul 2022 10:18:58 +0200, Alvaro Herrera <[email protected]> wrote in
> OK, I'll wait for allow_in_place_tablespaces to be backpatched then.
>
> I would like to get this fix pushed before the next set of minors, so if
> you won't have time for the backpatches early enough, maybe I can work
> on getting it done.
>
> Which commits would we consider?
>
> 7170f2159fb2 Allow "in place" tablespaces.
> f6f0db4d6240 Fix pg_tablespace_location() with in-place tablespaces
The second one is just to make the function work with in-place
tablespaces. Without it the function yeilds the following error.
> ERROR: could not read symbolic link "pg_tblspc/16407": Invalid argument
This looks actually odd but I think no need of back-patching because
there's no actual user of the feature is not seen in our test suite.
If we have a test that needs the feature in future, it would be enough
to back-patch it then.
So I think only the first one is needed for now.
regards.
--
Kyotaro Horiguchi
NTT Open Source Software Center
^ permalink raw reply [nested|flat] 55+ messages in thread
* Re: standby recovery fails (tablespace related) (tentative patch and discussion)
@ 2022-07-22 08:53 Thomas Munro <[email protected]>
parent: Alvaro Herrera <[email protected]>
1 sibling, 0 replies; 55+ messages in thread
From: Thomas Munro @ 2022-07-22 08:53 UTC (permalink / raw)
To: Alvaro Herrera <[email protected]>; +Cc: Kyotaro Horiguchi <[email protected]>; Dilip Kumar <[email protected]>; Robert Haas <[email protected]>; Michael Paquier <[email protected]>; Julien Rouhaud <[email protected]>; pgsql-hackers
On Fri, Jul 22, 2022 at 8:19 PM Alvaro Herrera <[email protected]> wrote:
> On 2022-Jul-22, Kyotaro Horiguchi wrote:
> > At Thu, 21 Jul 2022 23:14:57 +1200, Thomas Munro <[email protected]> wrote in
> > > Would it help if we back-patched the allow_in_place_tablespaces stuff?
> > > I'm not sure how hard/destabilising that would be, but I could take a
> > > look tomorrow.
> >
> > +1. Addiotional reason for me is it is a developer option.
>
> OK, I'll wait for allow_in_place_tablespaces to be backpatched then.
>
> I would like to get this fix pushed before the next set of minors, so if
> you won't have time for the backpatches early enough, maybe I can work
> on getting it done.
>
> Which commits would we consider?
I wonder how crazy it would be to back-patch
src/test/recovery/t/027_stream_regress.pl too.
^ permalink raw reply [nested|flat] 55+ messages in thread
* Re: standby recovery fails (tablespace related) (tentative patch and discussion)
@ 2022-07-27 06:07 Alvaro Herrera <[email protected]>
parent: Kyotaro Horiguchi <[email protected]>
0 siblings, 0 replies; 55+ messages in thread
From: Alvaro Herrera @ 2022-07-27 06:07 UTC (permalink / raw)
To: Kyotaro Horiguchi <[email protected]>; +Cc: [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; pgsql-hackers
On 2022-Jul-22, Kyotaro Horiguchi wrote:
> At Fri, 22 Jul 2022 10:18:58 +0200, Alvaro Herrera <[email protected]> wrote in
> > Which commits would we consider?
> >
> > 7170f2159fb2 Allow "in place" tablespaces.
> > f6f0db4d6240 Fix pg_tablespace_location() with in-place tablespaces
>
> The second one is just to make the function work with in-place
> tablespaces. Without it the function yeilds the following error.
>
> > ERROR: could not read symbolic link "pg_tblspc/16407": Invalid argument
>
> This looks actually odd but I think no need of back-patching because
> there's no actual user of the feature is not seen in our test suite.
> If we have a test that needs the feature in future, it would be enough
> to back-patch it then.
Actually, I found that the new test added by the fix in this thread does
depend on this being fixed, so I included an even larger set, which I
think makes this more complete:
7170f2159fb2 Allow "in place" tablespaces.
c6f2f01611d4 Fix pg_basebackup with in-place tablespaces.
f6f0db4d6240 Fix pg_tablespace_location() with in-place tablespaces
7a7cd84893e0 doc: Remove mention to in-place tablespaces for pg_tablespace_location()
5344723755bd Remove unnecessary Windows-specific basebackup code.
I didn't include any of the test changes for now. I don't intend to do
so, unless we see another reason for that; I think the new tests that
are going to be added by the recovery bugfix should be sufficient
coverage.
--
Álvaro Herrera PostgreSQL Developer — https://www.EnterpriseDB.com/
"La fuerza no está en los medios físicos
sino que reside en una voluntad indomable" (Gandhi)
^ permalink raw reply [nested|flat] 55+ messages in thread
* Re: standby recovery fails (tablespace related) (tentative patch and discussion)
@ 2022-07-27 18:54 Alvaro Herrera <[email protected]>
parent: Alvaro Herrera <[email protected]>
2 siblings, 1 reply; 55+ messages in thread
From: Alvaro Herrera @ 2022-07-27 18:54 UTC (permalink / raw)
To: Kyotaro Horiguchi <[email protected]>; +Cc: [email protected]; [email protected]; [email protected]; [email protected]; pgsql-hackers
Okay, I think I'm done with this. Here's v27 for the master branch,
where I fixed some comments as well as thinkos in the test script.
The ones on older branches aren't materially different, they just have
tonnes of conflicts resolved. I'll get this pushed tomorrow morning.
I have run it through CI and it seems ... not completely broken, at
least, but I have no working recipes for Windows on branches 14 and
older, so it doesn't really work fully. If anybody does, please share.
You can see mine here
https://github.com/alvherre/postgres/commits/REL_11_STABLE [etc]
https://cirrus-ci.com/build/5320904228995072
https://cirrus-ci.com/github/alvherre/postgres
--
Álvaro Herrera PostgreSQL Developer — https://www.EnterpriseDB.com/
"Every machine is a smoke machine if you operate it wrong enough."
https://twitter.com/libseybieda/status/1541673325781196801
^ permalink raw reply [nested|flat] 55+ messages in thread
* Re: standby recovery fails (tablespace related) (tentative patch and discussion)
@ 2022-07-28 18:04 Matthias van de Meent <[email protected]>
parent: Alvaro Herrera <[email protected]>
0 siblings, 1 reply; 55+ messages in thread
From: Matthias van de Meent @ 2022-07-28 18:04 UTC (permalink / raw)
To: Alvaro Herrera <[email protected]>; +Cc: pgsql-hackers
On Wed, 27 Jul 2022 at 20:55, Alvaro Herrera <[email protected]> wrote:
>
> Okay, I think I'm done with this. Here's v27 for the master branch,
> where I fixed some comments as well as thinkos in the test script.
> The ones on older branches aren't materially different, they just have
> tonnes of conflicts resolved. I'll get this pushed tomorrow morning.
>
> I have run it through CI and it seems ... not completely broken, at
> least, but I have no working recipes for Windows on branches 14 and
> older, so it doesn't really work fully. If anybody does, please share.
> You can see mine here
> https://github.com/alvherre/postgres/commits/REL_11_STABLE [etc]
> https://cirrus-ci.com/build/5320904228995072
> https://cirrus-ci.com/github/alvherre/postgres
I'd like to bring to your attention that the test that was introduced
with 9e4f914b seem to be flaky in FreeBSD 13 in the CFBot builds: it
sometimes times out while waiting for the secondary to catch up. Or,
at least I think it does, and I'm not too familiar with TAP failure
outputs: it returns with error code 29 and logs that I'd expect when
the timeout is reached.
See bottom for examples (all 3 builds for different patches).
Kind regards,
Matthias van de Meent.
[1] https://cirrus-ci.com/task/4960990331666432?logs=test_world#L2631-L2662
[2] https://cirrus-ci.com/task/5012678384025600?logs=test_world#L2631-L2662
[3] https://cirrus-ci.com/task/5147001137397760?logs=test_world#L2631-L2662
^ permalink raw reply [nested|flat] 55+ messages in thread
* Re: standby recovery fails (tablespace related) (tentative patch and discussion)
@ 2022-07-28 21:57 Tom Lane <[email protected]>
parent: Matthias van de Meent <[email protected]>
0 siblings, 1 reply; 55+ messages in thread
From: Tom Lane @ 2022-07-28 21:57 UTC (permalink / raw)
To: Matthias van de Meent <[email protected]>; +Cc: Alvaro Herrera <[email protected]>; pgsql-hackers
Matthias van de Meent <[email protected]> writes:
> I'd like to bring to your attention that the test that was introduced
> with 9e4f914b seem to be flaky in FreeBSD 13 in the CFBot builds: it
> sometimes times out while waiting for the secondary to catch up. Or,
> at least I think it does, and I'm not too familiar with TAP failure
> outputs: it returns with error code 29 and logs that I'd expect when
> the timeout is reached.
It's also failing in the buildfarm, eg
https://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=conchuela&dt=2022-07-28%2020%3A57%3A50
Looks like only conchuela so far, reinforcing the idea that we're
only seeing it on FreeBSD. I'd tentatively bet on a timing problem
that requires some FreeBSD scheduling quirk to manifest; we've seen
such quirks before.
regards, tom lane
^ permalink raw reply [nested|flat] 55+ messages in thread
* Re: standby recovery fails (tablespace related) (tentative patch and discussion)
@ 2022-07-28 23:27 Thomas Munro <[email protected]>
parent: Tom Lane <[email protected]>
0 siblings, 1 reply; 55+ messages in thread
From: Thomas Munro @ 2022-07-28 23:27 UTC (permalink / raw)
To: Tom Lane <[email protected]>; +Cc: Matthias van de Meent <[email protected]>; Alvaro Herrera <[email protected]>; pgsql-hackers
On Fri, Jul 29, 2022 at 9:57 AM Tom Lane <[email protected]> wrote:
> Matthias van de Meent <[email protected]> writes:
> > I'd like to bring to your attention that the test that was introduced
> > with 9e4f914b seem to be flaky in FreeBSD 13 in the CFBot builds: it
> > sometimes times out while waiting for the secondary to catch up. Or,
> > at least I think it does, and I'm not too familiar with TAP failure
> > outputs: it returns with error code 29 and logs that I'd expect when
> > the timeout is reached.
>
> It's also failing in the buildfarm, eg
>
> https://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=conchuela&dt=2022-07-28%2020%3A57%3A50
>
> Looks like only conchuela so far, reinforcing the idea that we're
> only seeing it on FreeBSD. I'd tentatively bet on a timing problem
> that requires some FreeBSD scheduling quirk to manifest; we've seen
> such quirks before.
Maybe it just needs a replication slot? I see:
ERROR: requested WAL segment 000000010000000000000003 has already been removed
^ permalink raw reply [nested|flat] 55+ messages in thread
* Re: standby recovery fails (tablespace related) (tentative patch and discussion)
@ 2022-07-29 05:20 Kyotaro Horiguchi <[email protected]>
parent: Thomas Munro <[email protected]>
0 siblings, 1 reply; 55+ messages in thread
From: Kyotaro Horiguchi @ 2022-07-29 05:20 UTC (permalink / raw)
To: [email protected]; +Cc: [email protected]; [email protected]; [email protected]; pgsql-hackers
At Fri, 29 Jul 2022 11:27:01 +1200, Thomas Munro <[email protected]> wrote in
> Maybe it just needs a replication slot? I see:
>
> ERROR: requested WAL segment 000000010000000000000003 has already been removed
Agreed, I see the same. The same failure can be surely reproducible
by inserting wal-switch+checkpoint after taking backup [1]. And it is
fixed by the attached.
regards.
--
Kyotaro Horiguchi
NTT Open Source Software Center
[1]:
--- a/src/test/recovery/t/033_replay_tsp_drops.pl
+++ b/src/test/recovery/t/033_replay_tsp_drops.pl
@@ -30,6 +30,13 @@ sub test_tablespace
my $backup_name = 'my_backup';
$node_primary->backup($backup_name);
+ $node_primary->psql(
+ 'postgres',
+ qq[
+ CREATE TABLE t(); DROP TABLE t; SELECT pg_switch_wal();
+ CHECKPOINT;
+ ]);
+
my $node_standby = PostgreSQL::Test::Cluster->new("standby2_$strategy");
$node_standby->init_from_backup($node_primary, $backup_name,
has_streaming => 1);
Attachments:
[text/x-patch] fix_tsp_drop_test_error.diff (516B, ../../[email protected]/2-fix_tsp_drop_test_error.diff)
download | inline diff:
diff --git a/src/test/recovery/t/033_replay_tsp_drops.pl b/src/test/recovery/t/033_replay_tsp_drops.pl
index 9b74cb09ac..0756ca6c87 100644
--- a/src/test/recovery/t/033_replay_tsp_drops.pl
+++ b/src/test/recovery/t/033_replay_tsp_drops.pl
@@ -20,6 +20,7 @@ sub test_tablespace
$node_primary->psql(
'postgres',
qq[
+ SELECT pg_create_physical_replication_slot('slot1', true);
SET allow_in_place_tablespaces=on;
CREATE TABLESPACE dropme_ts1 LOCATION '';
CREATE TABLESPACE dropme_ts2 LOCATION '';
^ permalink raw reply [nested|flat] 55+ messages in thread
* Re: standby recovery fails (tablespace related) (tentative patch and discussion)
@ 2022-07-29 10:59 Alvaro Herrera <[email protected]>
parent: Kyotaro Horiguchi <[email protected]>
0 siblings, 1 reply; 55+ messages in thread
From: Alvaro Herrera @ 2022-07-29 10:59 UTC (permalink / raw)
To: Kyotaro Horiguchi <[email protected]>; +Cc: [email protected]; [email protected]; [email protected]; pgsql-hackers
On 2022-Jul-29, Kyotaro Horiguchi wrote:
> At Fri, 29 Jul 2022 11:27:01 +1200, Thomas Munro <[email protected]> wrote in
> > Maybe it just needs a replication slot? I see:
> >
> > ERROR: requested WAL segment 000000010000000000000003 has already been removed
>
> Agreed, I see the same. The same failure can be surely reproducible
> by inserting wal-switch+checkpoint after taking backup [1]. And it is
> fixed by the attached.
WFM, pushed that way. I added a slot drop after the pg_stat_replication
count check to be a little less intrusive. Thanks Matthias for
reporting. (Note that the Cirrus page has a download link for the
complete logs as artifacts).
--
Álvaro Herrera 48°01'N 7°57'E — https://www.EnterpriseDB.com/
"I'm always right, but sometimes I'm more right than other times."
(Linus Torvalds)
^ permalink raw reply [nested|flat] 55+ messages in thread
* Re: standby recovery fails (tablespace related) (tentative patch and discussion)
@ 2022-07-30 14:37 Tom Lane <[email protected]>
parent: Alvaro Herrera <[email protected]>
0 siblings, 2 replies; 55+ messages in thread
From: Tom Lane @ 2022-07-30 14:37 UTC (permalink / raw)
To: Alvaro Herrera <[email protected]>; +Cc: Kyotaro Horiguchi <[email protected]>; [email protected]; [email protected]; pgsql-hackers
Alvaro Herrera <[email protected]> writes:
> WFM, pushed that way.
Looks like conchuela is still intermittently unhappy.
https://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=conchuela&dt=2022-07-30%2004%3A57%3A51
regards, tom lane
^ permalink raw reply [nested|flat] 55+ messages in thread
* Re: standby recovery fails (tablespace related) (tentative patch and discussion)
@ 2022-07-30 16:51 Tom Lane <[email protected]>
parent: Tom Lane <[email protected]>
1 sibling, 1 reply; 55+ messages in thread
From: Tom Lane @ 2022-07-30 16:51 UTC (permalink / raw)
To: Alvaro Herrera <[email protected]>; +Cc: Kyotaro Horiguchi <[email protected]>; [email protected]; [email protected]; pgsql-hackers
I wrote:
> Looks like conchuela is still intermittently unhappy.
BTW, quite aside from stability, is it really necessary for this test to
be so freakin' slow? florican for instance reports
[12:43:38] t/025_stuck_on_old_timeline.pl ....... ok 49010 ms ( 0.00 usr 0.00 sys + 3.64 cusr 2.49 csys = 6.13 CPU)
[12:44:12] t/026_overwrite_contrecord.pl ........ ok 34751 ms ( 0.01 usr 0.00 sys + 3.14 cusr 1.76 csys = 4.91 CPU)
[12:49:00] t/027_stream_regress.pl .............. ok 287278 ms ( 0.00 usr 0.00 sys + 9.66 cusr 6.95 csys = 16.60 CPU)
[12:50:04] t/028_pitr_timelines.pl .............. ok 64543 ms ( 0.00 usr 0.00 sys + 3.59 cusr 3.20 csys = 6.78 CPU)
[12:50:17] t/029_stats_restart.pl ............... ok 12505 ms ( 0.02 usr 0.00 sys + 3.16 cusr 1.40 csys = 4.57 CPU)
[12:50:51] t/030_stats_cleanup_replica.pl ....... ok 33933 ms ( 0.01 usr 0.01 sys + 3.55 cusr 2.46 csys = 6.03 CPU)
[12:51:25] t/031_recovery_conflict.pl ........... ok 34249 ms ( 0.00 usr 0.00 sys + 3.37 cusr 2.20 csys = 5.57 CPU)
[12:52:09] t/032_relfilenode_reuse.pl ........... ok 44274 ms ( 0.01 usr 0.00 sys + 3.21 cusr 2.05 csys = 5.27 CPU)
[12:54:07] t/033_replay_tsp_drops.pl ............ ok 117840 ms ( 0.01 usr 0.00 sys + 8.72 cusr 5.41 csys = 14.14 CPU)
027 is so bloated because it runs the core regression tests YA time,
which I'm not very happy about either; but that's no excuse for
every new test to contribute an additional couple of minutes.
regards, tom lane
^ permalink raw reply [nested|flat] 55+ messages in thread
* Re: standby recovery fails (tablespace related) (tentative patch and discussion)
@ 2022-07-30 23:08 Thomas Munro <[email protected]>
parent: Tom Lane <[email protected]>
0 siblings, 1 reply; 55+ messages in thread
From: Thomas Munro @ 2022-07-30 23:08 UTC (permalink / raw)
To: Tom Lane <[email protected]>; +Cc: Alvaro Herrera <[email protected]>; Kyotaro Horiguchi <[email protected]>; Matthias van de Meent <[email protected]>; pgsql-hackers
On Sun, Jul 31, 2022 at 4:51 AM Tom Lane <[email protected]> wrote:
> BTW, quite aside from stability, is it really necessary for this test to
> be so freakin' slow? florican for instance reports
>
> [12:43:38] t/025_stuck_on_old_timeline.pl ....... ok 49010 ms ( 0.00 usr 0.00 sys + 3.64 cusr 2.49 csys = 6.13 CPU)
> [12:44:12] t/026_overwrite_contrecord.pl ........ ok 34751 ms ( 0.01 usr 0.00 sys + 3.14 cusr 1.76 csys = 4.91 CPU)
> [12:49:00] t/027_stream_regress.pl .............. ok 287278 ms ( 0.00 usr 0.00 sys + 9.66 cusr 6.95 csys = 16.60 CPU)
> [12:50:04] t/028_pitr_timelines.pl .............. ok 64543 ms ( 0.00 usr 0.00 sys + 3.59 cusr 3.20 csys = 6.78 CPU)
> [12:50:17] t/029_stats_restart.pl ............... ok 12505 ms ( 0.02 usr 0.00 sys + 3.16 cusr 1.40 csys = 4.57 CPU)
> [12:50:51] t/030_stats_cleanup_replica.pl ....... ok 33933 ms ( 0.01 usr 0.01 sys + 3.55 cusr 2.46 csys = 6.03 CPU)
> [12:51:25] t/031_recovery_conflict.pl ........... ok 34249 ms ( 0.00 usr 0.00 sys + 3.37 cusr 2.20 csys = 5.57 CPU)
> [12:52:09] t/032_relfilenode_reuse.pl ........... ok 44274 ms ( 0.01 usr 0.00 sys + 3.21 cusr 2.05 csys = 5.27 CPU)
> [12:54:07] t/033_replay_tsp_drops.pl ............ ok 117840 ms ( 0.01 usr 0.00 sys + 8.72 cusr 5.41 csys = 14.14 CPU)
>
> 027 is so bloated because it runs the core regression tests YA time,
> which I'm not very happy about either; but that's no excuse for
> every new test to contribute an additional couple of minutes.
Complaints about 027 noted, I'm thinking about what we could do about that.
As for 033, I worried that it might be the new ProcSignalBarrier stuff
around tablespaces, but thankfully the DEBUG logging I added there
recently shows those all completing in single digit milliseconds. I
also confirmed there are no unexpected fsync'd being produced here.
That is quite a lot of CPU, but it's a huge amount of total runtime.
It runs in 5-8 seconds on various modern systems, 19 seconds on my
Linux RPi4, and 50 seconds on my Celeron-powered NAS box with spinning
disks.
I noticed this is a 32 bit FBSD system. Is it running on UFS, perhaps
on slow storage? Are soft updates enabled (visible as options in
output of "mount")? Without soft updates, a lot more file system ops
perform synchronous I/O, which really slows down our tests. In
general, UFS isn't as good as modern file systems at avoiding I/O for
short-lived files, and we set up and tear down a lot of them in our
testing. Another thing that makes a difference is to use a filesystem
with 8KB block size. This has been a subject of investigation for
speeding up CI (see src/tools/ci/gcp_freebsd_repartition.sh), but
several mysteries remain unsolved...
^ permalink raw reply [nested|flat] 55+ messages in thread
* Re: standby recovery fails (tablespace related) (tentative patch and discussion)
@ 2022-07-30 23:17 Tom Lane <[email protected]>
parent: Thomas Munro <[email protected]>
0 siblings, 0 replies; 55+ messages in thread
From: Tom Lane @ 2022-07-30 23:17 UTC (permalink / raw)
To: Thomas Munro <[email protected]>; +Cc: Alvaro Herrera <[email protected]>; Kyotaro Horiguchi <[email protected]>; Matthias van de Meent <[email protected]>; pgsql-hackers
Thomas Munro <[email protected]> writes:
> I noticed this is a 32 bit FBSD system. Is it running on UFS, perhaps
> on slow storage? Are soft updates enabled (visible as options in
> output of "mount")?
It's an ancient (2006) mac mini with 5400RPM spinning rust.
"mount" says
/dev/ada0s2a on / (ufs, local, soft-updates, journaled soft-updates)
devfs on /dev (devfs)
regards, tom lane
^ permalink raw reply [nested|flat] 55+ messages in thread
* Re: standby recovery fails (tablespace related) (tentative patch and discussion)
@ 2022-07-31 03:46 Thomas Munro <[email protected]>
parent: Tom Lane <[email protected]>
1 sibling, 0 replies; 55+ messages in thread
From: Thomas Munro @ 2022-07-31 03:46 UTC (permalink / raw)
To: Tom Lane <[email protected]>; +Cc: Alvaro Herrera <[email protected]>; Kyotaro Horiguchi <[email protected]>; Matthias van de Meent <[email protected]>; pgsql-hackers
On Sun, Jul 31, 2022 at 2:37 AM Tom Lane <[email protected]> wrote:
> Alvaro Herrera <[email protected]> writes:
> > WFM, pushed that way.
>
> Looks like conchuela is still intermittently unhappy.
>
> https://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=conchuela&dt=2022-07-30%2004%3A57%3A51
And here's one from CI that failed on Linux (this was a cfbot run with
an unrelated patch, parent commit b998196 so a few commits after "Fix
test instability"):
https://cirrus-ci.com/task/5282155000496128
https://api.cirrus-ci.com/v1/artifact/task/5282155000496128/log/src/test/recovery/tmp_check/log/033_...
It looks like this sequence is racy and we need to wait for more than
just "connection is made" before dropping the slot?
$node_standby->start;
# Make sure connection is made
$node_primary->poll_query_until('postgres',
'SELECT count(*) = 1 FROM pg_stat_replication');
$node_primary->safe_psql('postgres', "SELECT
pg_drop_replication_slot('slot')");
Why not set the replication slot name so that the standby uses it
"properly", like in other tests?
^ permalink raw reply [nested|flat] 55+ messages in thread
* [PATCH 5/5] Use background worker to do logical decoding.
@ 2025-12-09 18:44 Antonin Houska <[email protected]>
0 siblings, 0 replies; 55+ messages in thread
From: Antonin Houska @ 2025-12-09 18:44 UTC (permalink / raw)
If the backend performing REPACK (CONCURRENTLY) does both data copying and
logical decoding, it has to "travel in time" back and forth and therefore it
has to invalidate system caches quite a few times. (The copying and the
decoding work with different catalog snapshots.) As the decoding worker has
separate caches, the switching is not necessary.
Without the worker, it'd also be difficult to switch between potentiallly long
running tasks like index build and WAL decoding. (No decoding during that time
at all can suspend archiving / recycling of WAL segments for some time, which
in turn may result in full disk.)
Another problem is that, after having acquired AccessExclusiveLock (in order
to swap the files), the backend needs to both decode and apply the data
changes that took place while it was waiting for the lock. With the decoding
worker, the decoding runs all the time, so the backend only needs to apply the
changes. This can reduce the time the exclusive lock is held for.
Note that the code added in order to handle ERRORs in the background worker
almost duplicates the existing code that does the same for other types of
workers (See ProcessParallelMessages() and
ProcessParallelApplyMessages()). Refactoring of the existing code might be
useful, to reduce the duplication.
---
src/backend/access/heap/heapam_handler.c | 44 -
src/backend/commands/cluster.c | 1139 +++++++++++++----
src/backend/libpq/pqmq.c | 5 +
src/backend/postmaster/bgworker.c | 4 +
src/backend/replication/logical/logical.c | 6 +-
.../pgoutput_repack/pgoutput_repack.c | 54 +-
src/backend/storage/ipc/procsignal.c | 4 +
src/backend/tcop/postgres.c | 4 +
.../utils/activity/wait_event_names.txt | 2 +
src/include/access/tableam.h | 7 +-
src/include/commands/cluster.h | 68 +-
src/include/storage/procsignal.h | 1 +
src/tools/pgindent/typedefs.list | 4 +-
13 files changed, 932 insertions(+), 410 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index 01be29eb405..e6d630fa2f7 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -33,7 +33,6 @@
#include "catalog/index.h"
#include "catalog/storage.h"
#include "catalog/storage_xlog.h"
-#include "commands/cluster.h"
#include "commands/progress.h"
#include "executor/executor.h"
#include "miscadmin.h"
@@ -688,7 +687,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
Relation OldIndex, bool use_sort,
TransactionId OldestXmin,
Snapshot snapshot,
- LogicalDecodingContext *decoding_ctx,
TransactionId *xid_cutoff,
MultiXactId *multi_cutoff,
double *num_tuples,
@@ -710,7 +708,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
BufferHeapTupleTableSlot *hslot;
BlockNumber prev_cblock = InvalidBlockNumber;
bool concurrent = snapshot != NULL;
- XLogRecPtr end_of_wal_prev = GetFlushRecPtr(NULL);
/* Remember if it's a system catalog */
is_system_catalog = IsSystemRelation(OldHeap);
@@ -957,31 +954,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
ct_val[1] = *num_tuples;
pgstat_progress_update_multi_param(2, ct_index, ct_val);
}
-
- /*
- * Process the WAL produced by the load, as well as by other
- * transactions, so that the replication slot can advance and WAL does
- * not pile up. Use wal_segment_size as a threshold so that we do not
- * introduce the decoding overhead too often.
- *
- * Of course, we must not apply the changes until the initial load has
- * completed.
- *
- * Note that our insertions into the new table should not be decoded
- * as we (intentionally) do not write the logical decoding specific
- * information to WAL.
- */
- if (concurrent)
- {
- XLogRecPtr end_of_wal;
-
- end_of_wal = GetFlushRecPtr(NULL);
- if ((end_of_wal - end_of_wal_prev) > wal_segment_size)
- {
- repack_decode_concurrent_changes(decoding_ctx, end_of_wal);
- end_of_wal_prev = end_of_wal;
- }
- }
}
if (indexScan != NULL)
@@ -1027,22 +999,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
-
- /*
- * Try to keep the amount of not-yet-decoded WAL small, like
- * above.
- */
- if (concurrent)
- {
- XLogRecPtr end_of_wal;
-
- end_of_wal = GetFlushRecPtr(NULL);
- if ((end_of_wal - end_of_wal_prev) > wal_segment_size)
- {
- repack_decode_concurrent_changes(decoding_ctx, end_of_wal);
- end_of_wal_prev = end_of_wal;
- }
- }
}
tuplesort_end(tuplesort);
diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c
index 501bd36c23e..f2a2ec6d3e5 100644
--- a/src/backend/commands/cluster.c
+++ b/src/backend/commands/cluster.c
@@ -46,6 +46,8 @@
#include "commands/tablecmds.h"
#include "commands/vacuum.h"
#include "executor/executor.h"
+#include "libpq/pqformat.h"
+#include "libpq/pqmq.h"
#include "miscadmin.h"
#include "optimizer/optimizer.h"
#include "pgstat.h"
@@ -56,6 +58,8 @@
#include "storage/ipc.h"
#include "storage/lmgr.h"
#include "storage/predicate.h"
+#include "storage/procsignal.h"
+#include "tcop/tcopprot.h"
#include "utils/acl.h"
#include "utils/fmgroids.h"
#include "utils/guc.h"
@@ -123,6 +127,103 @@ typedef struct ChangeDest
IndexInsertState *iistate;
} ChangeDest;
+/*
+ * Layout of shared memory used for communication between backend and the
+ * worker that performs logical decoding of data changes
+ */
+typedef struct DecodingWorkerShared
+{
+ /*
+ * Once the worker has reached this LSN, it should close the current
+ * output file and either create a new one or exit, according to the field
+ * 'done'. If the value is InvalidXLogRecPtr, the worker should decode all
+ * the WAL available and keep checking this field. It is ok if the worker
+ * had already decoded records whose LSN is >= lsn_upto before this field
+ * has been set.
+ */
+ XLogRecPtr lsn_upto;
+
+ /* Exit after closing the current file? */
+ bool done;
+
+ /* The output is stored here. */
+ SharedFileSet sfs;
+
+ /* Can backend read the file contents? */
+ bool sfs_valid;
+
+ /* Number of the last file exported by the worker. */
+ int last_exported;
+
+ /* Synchronize access to the fields above. */
+ slock_t mutex;
+
+ /* Database to connect to. */
+ Oid dbid;
+
+ /* Role to connect as. */
+ Oid roleid;
+
+ /* Decode data changes of this relation. */
+ Oid relid;
+
+ /* The backend uses this to wait for the worker. */
+ ConditionVariable cv;
+
+ /* Info to signal the backend. */
+ PGPROC *backend_proc;
+ pid_t backend_pid;
+ ProcNumber backend_proc_number;
+
+ /* Error queue. */
+ shm_mq *error_mq;
+
+ /*
+ * Memory the queue is located int.
+ *
+ * For considerations on the value see the comments of
+ * PARALLEL_ERROR_QUEUE_SIZE.
+ */
+#define REPACK_ERROR_QUEUE_SIZE 16384
+ char error_queue[FLEXIBLE_ARRAY_MEMBER];
+} DecodingWorkerShared;
+
+/*
+ * Generate output file name. If relations of the same 'relid' happen to be
+ * processed at the same time, they must be from different databases and
+ * therefore different backends must be involved. (PID is already present in
+ * the fileset name.)
+ */
+static inline void
+DecodingWorkerFileName(char *fname, Oid relid, uint32 seq)
+{
+ snprintf(fname, MAXPGPATH, "%u-%u", relid, seq);
+}
+
+/*
+ * Backend-local information to control the decoding worker.
+ */
+typedef struct DecodingWorker
+{
+ /* The worker. */
+ BackgroundWorkerHandle *handle;
+
+ /* DecodingWorkerShared is in this segment. */
+ dsm_segment *seg;
+
+ /* Handle of the error queue. */
+ shm_mq_handle *error_mqh;
+} DecodingWorker;
+
+/* Pointer to currently running decoding worker. */
+static DecodingWorker *decoding_worker = NULL;
+
+/*
+ * Is there a message sent by a repack worker that the backend needs to
+ * receive?
+ */
+volatile sig_atomic_t RepackMessagePending = false;
+
static bool cluster_rel_recheck(RepackCommand cmd, Relation OldHeap,
Oid indexOid, Oid userid, LOCKMODE lmode,
int options);
@@ -130,7 +231,7 @@ static void check_repack_concurrently_requirements(Relation rel);
static void rebuild_relation(Relation OldHeap, Relation index, bool verbose,
bool concurrent);
static void copy_table_data(Relation NewHeap, Relation OldHeap, Relation OldIndex,
- Snapshot snapshot, LogicalDecodingContext *decoding_ctx,
+ Snapshot snapshot,
bool verbose,
bool *pSwapToastByContent,
TransactionId *pFreezeXid,
@@ -143,12 +244,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd,
static bool cluster_is_permitted_for_relation(RepackCommand cmd,
Oid relid, Oid userid);
-static void begin_concurrent_repack(Relation rel);
-static void end_concurrent_repack(void);
static LogicalDecodingContext *setup_logical_decoding(Oid relid);
-static HeapTuple get_changed_tuple(char *change);
-static void apply_concurrent_changes(RepackDecodingState *dstate,
- ChangeDest *dest);
+static bool decode_concurrent_changes(LogicalDecodingContext *ctx,
+ DecodingWorkerShared *shared);
+static void apply_concurrent_changes(BufFile *file, ChangeDest *dest);
static void apply_concurrent_insert(Relation rel, HeapTuple tup,
IndexInsertState *iistate,
TupleTableSlot *index_slot);
@@ -160,9 +259,9 @@ static void apply_concurrent_delete(Relation rel, HeapTuple tup_target);
static HeapTuple find_target_tuple(Relation rel, ChangeDest *dest,
HeapTuple tup_key,
TupleTableSlot *ident_slot);
-static void process_concurrent_changes(LogicalDecodingContext *decoding_ctx,
- XLogRecPtr end_of_wal,
- ChangeDest *dest);
+static void process_concurrent_changes(XLogRecPtr end_of_wal,
+ ChangeDest *dest,
+ bool done);
static IndexInsertState *get_index_insert_state(Relation relation,
Oid ident_index_id,
Relation *ident_index_p);
@@ -172,7 +271,6 @@ static void free_index_insert_state(IndexInsertState *iistate);
static void cleanup_logical_decoding(LogicalDecodingContext *ctx);
static void rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
Relation cl_index,
- LogicalDecodingContext *decoding_ctx,
TransactionId frozenXid,
MultiXactId cutoffMulti);
static List *build_new_indexes(Relation NewHeap, Relation OldHeap, List *OldIndexes);
@@ -182,6 +280,13 @@ static Relation process_single_relation(RepackStmt *stmt,
ClusterParams *params);
static Oid determine_clustered_index(Relation rel, bool usingindex,
const char *indexname);
+static void start_decoding_worker(Oid relid);
+static void stop_decoding_worker(void);
+static void repack_worker_internal(dsm_segment *seg);
+static void export_initial_snapshot(Snapshot snapshot,
+ DecodingWorkerShared *shared);
+static Snapshot get_initial_snapshot(DecodingWorker *worker);
+static void ProcessRepackMessage(StringInfo msg);
static const char *RepackCommandAsString(RepackCommand cmd);
@@ -604,20 +709,20 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid,
/* rebuild_relation does all the dirty work */
PG_TRY();
{
- /*
- * For concurrent processing, make sure that our logical decoding
- * ignores data changes of other tables than the one we are
- * processing.
- */
- if (concurrent)
- begin_concurrent_repack(OldHeap);
-
rebuild_relation(OldHeap, index, verbose, concurrent);
}
PG_FINALLY();
{
if (concurrent)
- end_concurrent_repack();
+ {
+ /*
+ * Since during normal operation the worker was already asked to
+ * exit, stopping it explicitly is especially important on ERROR.
+ * However it still seems a good practice to make sure that the
+ * worker never survives the REPACK command.
+ */
+ stop_decoding_worker();
+ }
}
PG_END_TRY();
@@ -914,7 +1019,6 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
bool swap_toast_by_content;
TransactionId frozenXid;
MultiXactId cutoffMulti;
- LogicalDecodingContext *decoding_ctx = NULL;
Snapshot snapshot = NULL;
#if USE_ASSERT_CHECKING
LOCKMODE lmode;
@@ -928,19 +1032,36 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
if (concurrent)
{
/*
- * Prepare to capture the concurrent data changes.
+ * The worker needs to be member of the locking group we're the leader
+ * of. We ought to become the leader before the worker starts. The
+ * worker will join the group as soon as it starts.
+ *
+ * This is to make sure that the deadlock described below is
+ * detectable by deadlock.c: if the worker waits for a transaction to
+ * complete and we are waiting for the worker output, then effectively
+ * we (i.e. this backend) are waiting for that transaction.
+ */
+ BecomeLockGroupLeader();
+
+ /*
+ * Start the worker that decodes data changes applied while we're
+ * copying the table contents.
*
- * Note that this call waits for all transactions with XID already
- * assigned to finish. If some of those transactions is waiting for a
- * lock conflicting with ShareUpdateExclusiveLock on our table (e.g.
- * it runs CREATE INDEX), we can end up in a deadlock. Not sure this
- * risk is worth unlocking/locking the table (and its clustering
- * index) and checking again if its still eligible for REPACK
- * CONCURRENTLY.
+ * Note that the worker has to wait for all transactions with XID
+ * already assigned to finish. If some of those transactions is
+ * waiting for a lock conflicting with ShareUpdateExclusiveLock on our
+ * table (e.g. it runs CREATE INDEX), we can end up in a deadlock.
+ * Not sure this risk is worth unlocking/locking the table (and its
+ * clustering index) and checking again if its still eligible for
+ * REPACK CONCURRENTLY.
+ */
+ start_decoding_worker(tableOid);
+
+ /*
+ * Wait until the worker has the initial snapshot and retrieve it.
*/
- decoding_ctx = setup_logical_decoding(tableOid);
+ snapshot = get_initial_snapshot(decoding_worker);
- snapshot = SnapBuildInitialSnapshotForRepack(decoding_ctx->snapshot_builder);
PushActiveSnapshot(snapshot);
}
@@ -965,7 +1086,7 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
NewHeap = table_open(OIDNewHeap, NoLock);
/* Copy the heap data into the new table in the desired order */
- copy_table_data(NewHeap, OldHeap, index, snapshot, decoding_ctx, verbose,
+ copy_table_data(NewHeap, OldHeap, index, snapshot, verbose,
&swap_toast_by_content, &frozenXid, &cutoffMulti);
/* The historic snapshot won't be needed anymore. */
@@ -989,15 +1110,11 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
Assert(!swap_toast_by_content);
rebuild_relation_finish_concurrent(NewHeap, OldHeap, index,
- decoding_ctx,
frozenXid, cutoffMulti);
PopActiveSnapshot();
pgstat_progress_update_param(PROGRESS_REPACK_PHASE,
PROGRESS_REPACK_PHASE_FINAL_CLEANUP);
-
- /* Done with decoding. */
- cleanup_logical_decoding(decoding_ctx);
}
else
{
@@ -1168,8 +1285,7 @@ make_new_heap(Oid OIDOldHeap, Oid NewTableSpace, Oid NewAccessMethod,
*/
static void
copy_table_data(Relation NewHeap, Relation OldHeap, Relation OldIndex,
- Snapshot snapshot, LogicalDecodingContext *decoding_ctx,
- bool verbose, bool *pSwapToastByContent,
+ Snapshot snapshot, bool verbose, bool *pSwapToastByContent,
TransactionId *pFreezeXid, MultiXactId *pCutoffMulti)
{
Relation relRelation;
@@ -1330,7 +1446,6 @@ copy_table_data(Relation NewHeap, Relation OldHeap, Relation OldIndex,
*/
table_relation_copy_for_cluster(OldHeap, NewHeap, OldIndex, use_sort,
cutoffs.OldestXmin, snapshot,
- decoding_ctx,
&cutoffs.FreezeLimit,
&cutoffs.MultiXactCutoff,
&num_tuples, &tups_vacuumed,
@@ -2363,59 +2478,6 @@ RepackCommandAsString(RepackCommand cmd)
return "???";
}
-
-/*
- * Call this function before REPACK CONCURRENTLY starts to setup logical
- * decoding. It makes sure that other users of the table put enough
- * information into WAL.
- *
- * The point is that at various places we expect that the table we're
- * processing is treated like a system catalog. For example, we need to be
- * able to scan it using a "historic snapshot" anytime during the processing
- * (as opposed to scanning only at the start point of the decoding, as logical
- * replication does during initial table synchronization), in order to apply
- * concurrent UPDATE / DELETE commands.
- *
- * Note that TOAST table needs no attention here as it's not scanned using
- * historic snapshot.
- */
-static void
-begin_concurrent_repack(Relation rel)
-{
- Oid toastrelid;
-
- /*
- * Avoid logical decoding of other relations by this backend. The lock we
- * have guarantees that the actual locator cannot be changed concurrently:
- * TRUNCATE needs AccessExclusiveLock.
- */
- Assert(CheckRelationLockedByMe(rel, ShareUpdateExclusiveLock, false));
- repacked_rel_locator = rel->rd_locator;
- toastrelid = rel->rd_rel->reltoastrelid;
- if (OidIsValid(toastrelid))
- {
- Relation toastrel;
-
- /* Avoid logical decoding of other TOAST relations. */
- toastrel = table_open(toastrelid, AccessShareLock);
- repacked_rel_toast_locator = toastrel->rd_locator;
- table_close(toastrel, AccessShareLock);
- }
-}
-
-/*
- * Call this when done with REPACK CONCURRENTLY.
- */
-static void
-end_concurrent_repack(void)
-{
- /*
- * Restore normal function of (future) logical decoding for this backend.
- */
- repacked_rel_locator.relNumber = InvalidOid;
- repacked_rel_toast_locator.relNumber = InvalidOid;
-}
-
/*
* This function is much like pg_create_logical_replication_slot() except that
* the new slot is neither released (if anyone else could read changes from
@@ -2427,9 +2489,10 @@ static LogicalDecodingContext *
setup_logical_decoding(Oid relid)
{
Relation rel;
- TupleDesc tupdesc;
+ Oid toastrelid;
LogicalDecodingContext *ctx;
- RepackDecodingState *dstate = palloc0_object(RepackDecodingState);
+ NameData slotname;
+ RepackDecodingState *dstate;
/*
* REPACK CONCURRENTLY is not allowed in a transaction block, so this
@@ -2437,21 +2500,21 @@ setup_logical_decoding(Oid relid)
*/
Assert(!TransactionIdIsValid(GetTopTransactionIdIfAny()));
- /*
- * A single backend should not execute multiple REPACK commands at a time,
- * so use PID to make the slot unique.
- */
- snprintf(NameStr(dstate->slotname), NAMEDATALEN, "repack_%d", MyProcPid);
-
/*
* Check if we can use logical decoding.
*/
CheckSlotPermissions();
CheckLogicalDecodingRequirements();
- /* RS_TEMPORARY so that the slot gets cleaned up on ERROR. */
- ReplicationSlotCreate(NameStr(dstate->slotname), true, RS_TEMPORARY,
- false, false, false);
+ /*
+ * A single backend should not execute multiple REPACK commands at a time,
+ * so use PID to make the slot unique.
+ *
+ * RS_TEMPORARY so that the slot gets cleaned up on ERROR.
+ */
+ snprintf(NameStr(slotname), NAMEDATALEN, "repack_%d", MyProcPid);
+ ReplicationSlotCreate(NameStr(slotname), true, RS_TEMPORARY, false, false,
+ false);
/*
* Neither prepare_write nor do_write callback nor update_progress is
@@ -2473,197 +2536,238 @@ setup_logical_decoding(Oid relid)
DecodingContextFindStartpoint(ctx);
+ /*
+ * decode_concurrent_changes() needs non-blocking callback.
+ */
+ ctx->reader->routine.page_read = read_local_xlog_page_no_wait;
+
+ /*
+ * read_local_xlog_page_no_wait() needs to be able to indicate the end of
+ * WAL.
+ */
+ ctx->reader->private_data = MemoryContextAllocZero(ctx->context,
+ sizeof(ReadLocalXLogPageNoWaitPrivate));
+
+
/* Some WAL records should have been read. */
Assert(ctx->reader->EndRecPtr != InvalidXLogRecPtr);
+ /*
+ * Initialize repack_current_segment so that we can notice WAL segment
+ * boundaries.
+ */
XLByteToSeg(ctx->reader->EndRecPtr, repack_current_segment,
wal_segment_size);
- /*
- * Setup structures to store decoded changes.
- */
+ dstate = palloc0_object(RepackDecodingState);
dstate->relid = relid;
- dstate->tstore = tuplestore_begin_heap(false, false,
- maintenance_work_mem);
- /* Caller should already have the table locked. */
- rel = table_open(relid, NoLock);
- tupdesc = CreateTupleDescCopy(RelationGetDescr(rel));
- dstate->tupdesc = tupdesc;
- table_close(rel, NoLock);
+ /*
+ * Tuple descriptor may be needed to flatten a tuple before we write it to
+ * a file. A copy is needed because the decoding worker invalidates system
+ * caches before it starts to do the actual work.
+ */
+ rel = table_open(relid, AccessShareLock);
+ dstate->tupdesc = CreateTupleDescCopy(RelationGetDescr(rel));
- /* Initialize the descriptor to store the changes ... */
- dstate->tupdesc_change = CreateTemplateTupleDesc(1);
+ /* Avoid logical decoding of other relations. */
+ repacked_rel_locator = rel->rd_locator;
+ toastrelid = rel->rd_rel->reltoastrelid;
+ if (OidIsValid(toastrelid))
+ {
+ Relation toastrel;
- TupleDescInitEntry(dstate->tupdesc_change, 1, NULL, BYTEAOID, -1, 0);
- /* ... as well as the corresponding slot. */
- dstate->tsslot = MakeSingleTupleTableSlot(dstate->tupdesc_change,
- &TTSOpsMinimalTuple);
+ /* Avoid logical decoding of other TOAST relations. */
+ toastrel = table_open(toastrelid, AccessShareLock);
+ repacked_rel_toast_locator = toastrel->rd_locator;
+ table_close(toastrel, AccessShareLock);
+ }
+ table_close(rel, AccessShareLock);
- dstate->resowner = ResourceOwnerCreate(CurrentResourceOwner,
- "logical decoding");
+ /* The file will be set as soon as we have it opened. */
+ dstate->file = NULL;
ctx->output_writer_private = dstate;
+
return ctx;
}
/*
- * Retrieve tuple from ConcurrentChange structure.
+ * Decode logical changes from the WAL sequence and store them to a file.
*
- * The input data starts with the structure but it might not be appropriately
- * aligned.
+ * If true is returned, there is no more work for the worker.
*/
-static HeapTuple
-get_changed_tuple(char *change)
+static bool
+decode_concurrent_changes(LogicalDecodingContext *ctx,
+ DecodingWorkerShared *shared)
{
- HeapTupleData tup_data;
- HeapTuple result;
- char *src;
+ RepackDecodingState *dstate;
+ XLogRecPtr lsn_upto;
+ bool done;
+ char fname[MAXPGPATH];
- /*
- * Ensure alignment before accessing the fields. (This is why we can't use
- * heap_copytuple() instead of this function.)
- */
- src = change + offsetof(ConcurrentChange, tup_data);
- memcpy(&tup_data, src, sizeof(HeapTupleData));
+ dstate = (RepackDecodingState *) ctx->output_writer_private;
- result = (HeapTuple) palloc(HEAPTUPLESIZE + tup_data.t_len);
- memcpy(result, &tup_data, sizeof(HeapTupleData));
- result->t_data = (HeapTupleHeader) ((char *) result + HEAPTUPLESIZE);
- src = change + SizeOfConcurrentChange;
- memcpy(result->t_data, src, result->t_len);
+ /* Open the output file. */
+ DecodingWorkerFileName(fname, shared->relid, shared->last_exported + 1);
+ dstate->file = BufFileCreateFileSet(&shared->sfs.fs, fname);
- return result;
-}
+ SpinLockAcquire(&shared->mutex);
+ lsn_upto = shared->lsn_upto;
+ done = shared->done;
+ SpinLockRelease(&shared->mutex);
-/*
- * Decode logical changes from the WAL sequence up to end_of_wal.
- */
-void
-repack_decode_concurrent_changes(LogicalDecodingContext *ctx,
- XLogRecPtr end_of_wal)
-{
- RepackDecodingState *dstate;
- ResourceOwner resowner_old;
-
- /*
- * Invalidate the "present" cache before moving to "(recent) history".
- */
- InvalidateSystemCaches();
+ while (XLogRecPtrIsInvalid(lsn_upto) || ctx->reader->EndRecPtr < lsn_upto)
+ {
+ XLogRecord *record;
+ XLogSegNo segno_new;
+ char *errm = NULL;
+ XLogRecPtr end_lsn;
- dstate = (RepackDecodingState *) ctx->output_writer_private;
- resowner_old = CurrentResourceOwner;
- CurrentResourceOwner = dstate->resowner;
+ CHECK_FOR_INTERRUPTS();
- PG_TRY();
- {
- while (ctx->reader->EndRecPtr < end_of_wal)
+ record = XLogReadRecord(ctx->reader, &errm);
+ if (record == NULL)
{
- XLogRecord *record;
- XLogSegNo segno_new;
- char *errm = NULL;
- XLogRecPtr end_lsn;
+ ReadLocalXLogPageNoWaitPrivate *priv;
- record = XLogReadRecord(ctx->reader, &errm);
if (errm)
- elog(ERROR, "%s", errm);
-
- if (record != NULL)
- LogicalDecodingProcessRecord(ctx, ctx->reader);
+ ereport(ERROR, (errmsg("%s", errm)));
/*
- * If WAL segment boundary has been crossed, inform the decoding
- * system that the catalog_xmin can advance. (We can confirm more
- * often, but a filling a single WAL segment should not take much
- * time.)
+ * In the decoding loop we do not want to get blocked when there
+ * is no more WAL available, otherwise the loop would become
+ * uninterruptible. The point is that the worker is only useful if
+ * it starts decoding before lsn_upto is set. Thus it can reach
+ * the end of WAL and find out later that it did not have to go
+ * that far.
*/
- end_lsn = ctx->reader->EndRecPtr;
- XLByteToSeg(end_lsn, segno_new, wal_segment_size);
- if (segno_new != repack_current_segment)
+ priv = (ReadLocalXLogPageNoWaitPrivate *)
+ ctx->reader->private_data;
+ if (priv->end_of_wal)
{
- LogicalConfirmReceivedLocation(end_lsn);
- elog(DEBUG1, "REPACK: confirmed receive location %X/%X",
- (uint32) (end_lsn >> 32), (uint32) end_lsn);
- repack_current_segment = segno_new;
+ priv->end_of_wal = false;
+
+ /* Do we know how far we should get? */
+ if (XLogRecPtrIsInvalid(lsn_upto))
+ {
+ SpinLockAcquire(&shared->mutex);
+ lsn_upto = shared->lsn_upto;
+ /* 'done' should be set at the same time as 'lsn_upto' */
+ done = shared->done;
+ SpinLockRelease(&shared->mutex);
+
+ /* Check if the work happens to be complete. */
+ continue;
+ }
+
+ /* Wait a bit before we retry reading WAL. */
+ (void) WaitLatch(MyLatch,
+ WL_LATCH_SET | WL_TIMEOUT | WL_EXIT_ON_PM_DEATH,
+ 1000L,
+ WAIT_EVENT_REPACK_WORKER_MAIN);
+
+ continue;
}
+ else
+ ereport(ERROR, (errmsg("could not read WAL record")));
+ }
- CHECK_FOR_INTERRUPTS();
+ LogicalDecodingProcessRecord(ctx, ctx->reader);
+
+ /*
+ * If WAL segment boundary has been crossed, inform the decoding
+ * system that the catalog_xmin can advance.
+ *
+ * TODO Does it make sense to confirm more often? Segment size seems
+ * appropriate for restart_lsn (because less than a segment cannot be
+ * recycled anyway), however more frequent checks might be beneficial
+ * for catalog_xmin.
+ */
+ end_lsn = ctx->reader->EndRecPtr;
+ XLByteToSeg(end_lsn, segno_new, wal_segment_size);
+ if (segno_new != repack_current_segment)
+ {
+ LogicalConfirmReceivedLocation(end_lsn);
+ elog(DEBUG1, "REPACK: confirmed receive location %X/%X",
+ (uint32) (end_lsn >> 32), (uint32) end_lsn);
+ repack_current_segment = segno_new;
+ }
+
+ /* Keep checking if 'lsn_upto' was specified. */
+ if (XLogRecPtrIsInvalid(lsn_upto))
+ {
+ SpinLockAcquire(&shared->mutex);
+ lsn_upto = shared->lsn_upto;
+ /* 'done' should be set at the same time as 'lsn_upto' */
+ done = shared->done;
+ SpinLockRelease(&shared->mutex);
}
- InvalidateSystemCaches();
- CurrentResourceOwner = resowner_old;
- }
- PG_CATCH();
- {
- /* clear all timetravel entries */
- InvalidateSystemCaches();
- CurrentResourceOwner = resowner_old;
- PG_RE_THROW();
}
- PG_END_TRY();
+
+ /*
+ * Close the file and make it available to the backend.
+ */
+ BufFileClose(dstate->file);
+ dstate->file = NULL;
+ SpinLockAcquire(&shared->mutex);
+ shared->lsn_upto = InvalidXLogRecPtr;
+ shared->sfs_valid = true;
+ shared->last_exported++;
+ SpinLockRelease(&shared->mutex);
+ ConditionVariableSignal(&shared->cv);
+
+ return done;
}
/*
- * Apply changes that happened during the initial load.
- *
- * Scan key is passed by caller, so it does not have to be constructed
- * multiple times. Key entries have all fields initialized, except for
- * sk_argument.
+ * Apply changes stored in 'file'.
*/
static void
-apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
+apply_concurrent_changes(BufFile *file, ChangeDest *dest)
{
+ char kind;
+ uint32 t_len;
Relation rel = dest->rel;
TupleTableSlot *index_slot,
*ident_slot;
HeapTuple tup_old = NULL;
- if (dstate->nchanges == 0)
- return;
-
/* TupleTableSlot is needed to pass the tuple to ExecInsertIndexTuples(). */
- index_slot = MakeSingleTupleTableSlot(dstate->tupdesc, &TTSOpsHeapTuple);
+ index_slot = MakeSingleTupleTableSlot(RelationGetDescr(rel),
+ &TTSOpsHeapTuple);
/* A slot to fetch tuples from identity index. */
ident_slot = table_slot_create(rel, NULL);
- while (tuplestore_gettupleslot(dstate->tstore, true, false,
- dstate->tsslot))
+ while (true)
{
- bool shouldFree;
- HeapTuple tup_change,
- tup,
+ size_t nread;
+ HeapTuple tup,
tup_exist;
- char *change_raw,
- *src;
- ConcurrentChange change;
- bool isnull[1];
- Datum values[1];
CHECK_FOR_INTERRUPTS();
- /* Get the change from the single-column tuple. */
- tup_change = ExecFetchSlotHeapTuple(dstate->tsslot, false, &shouldFree);
- heap_deform_tuple(tup_change, dstate->tupdesc_change, values, isnull);
- Assert(!isnull[0]);
-
- /* Make sure we access aligned data. */
- change_raw = (char *) DatumGetByteaP(values[0]);
- src = (char *) VARDATA(change_raw);
- memcpy(&change, src, SizeOfConcurrentChange);
+ nread = BufFileReadMaybeEOF(file, &kind, 1, true);
+ /* Are we done with the file? */
+ if (nread == 0)
+ break;
- /*
- * Extract the tuple from the change. The tuple is copied here because
- * it might be assigned to 'tup_old', in which case it needs to
- * survive into the next iteration.
- */
- tup = get_changed_tuple(src);
+ /* Read the tuple. */
+ BufFileReadExact(file, &t_len, sizeof(t_len));
+ tup = (HeapTuple) palloc(HEAPTUPLESIZE + t_len);
+ tup->t_data = (HeapTupleHeader) ((char *) tup + HEAPTUPLESIZE);
+ BufFileReadExact(file, tup->t_data, t_len);
+ tup->t_len = t_len;
+ ItemPointerSetInvalid(&tup->t_self);
+ tup->t_tableOid = RelationGetRelid(dest->rel);
- if (change.kind == CHANGE_UPDATE_OLD)
+ if (kind == CHANGE_UPDATE_OLD)
{
Assert(tup_old == NULL);
tup_old = tup;
}
- else if (change.kind == CHANGE_INSERT)
+ else if (kind == CHANGE_INSERT)
{
Assert(tup_old == NULL);
@@ -2671,12 +2775,11 @@ apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
pfree(tup);
}
- else if (change.kind == CHANGE_UPDATE_NEW ||
- change.kind == CHANGE_DELETE)
+ else if (kind == CHANGE_UPDATE_NEW || kind == CHANGE_DELETE)
{
HeapTuple tup_key;
- if (change.kind == CHANGE_UPDATE_NEW)
+ if (kind == CHANGE_UPDATE_NEW)
{
tup_key = tup_old != NULL ? tup_old : tup;
}
@@ -2693,7 +2796,7 @@ apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
if (tup_exist == NULL)
elog(ERROR, "failed to find target tuple");
- if (change.kind == CHANGE_UPDATE_NEW)
+ if (kind == CHANGE_UPDATE_NEW)
apply_concurrent_update(rel, tup, tup_exist, dest->iistate,
index_slot);
else
@@ -2708,26 +2811,19 @@ apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
pfree(tup);
}
else
- elog(ERROR, "unrecognized kind of change: %d", change.kind);
+ elog(ERROR, "unrecognized kind of change: %d", kind);
/*
* If a change was applied now, increment CID for next writes and
* update the snapshot so it sees the changes we've applied so far.
*/
- if (change.kind != CHANGE_UPDATE_OLD)
+ if (kind != CHANGE_UPDATE_OLD)
{
CommandCounterIncrement();
UpdateActiveSnapshotCommandId();
}
-
- /* TTSOpsMinimalTuple has .get_heap_tuple==NULL. */
- Assert(shouldFree);
- pfree(tup_change);
}
- tuplestore_clear(dstate->tstore);
- dstate->nchanges = 0;
-
/* Cleanup. */
ExecDropSingleTupleTableSlot(index_slot);
ExecDropSingleTupleTableSlot(ident_slot);
@@ -2865,6 +2961,12 @@ find_target_tuple(Relation rel, ChangeDest *dest, HeapTuple tup_key,
/* XXX no instrumentation for now */
scan = index_beginscan(rel, ident_index, GetActiveSnapshot(),
NULL, dest->ident_key_nentries, 0);
+
+ /*
+ * Scan key is passed by caller, so it does not have to be constructed
+ * multiple times. Key entries have all fields initialized, except for
+ * sk_argument.
+ */
index_rescan(scan, dest->ident_key, dest->ident_key_nentries, NULL, 0);
/* Info needed to retrieve key values from heap tuple. */
@@ -2900,25 +3002,58 @@ find_target_tuple(Relation rel, ChangeDest *dest, HeapTuple tup_key,
}
/*
- * Decode and apply concurrent changes.
+ * Decode and apply concurrent changes, up to (and including) the record whose
+ * LSN is 'end_of_wal'.
*/
static void
-process_concurrent_changes(LogicalDecodingContext *decoding_ctx,
- XLogRecPtr end_of_wal, ChangeDest *dest)
+process_concurrent_changes(XLogRecPtr end_of_wal, ChangeDest *dest, bool done)
{
- RepackDecodingState *dstate;
+ DecodingWorkerShared *shared;
+ char fname[MAXPGPATH];
+ BufFile *file;
pgstat_progress_update_param(PROGRESS_REPACK_PHASE,
PROGRESS_REPACK_PHASE_CATCH_UP);
- dstate = (RepackDecodingState *) decoding_ctx->output_writer_private;
+ /* Ask the worker for the file. */
+ shared = (DecodingWorkerShared *) dsm_segment_address(decoding_worker->seg);
+ SpinLockAcquire(&shared->mutex);
+ shared->lsn_upto = end_of_wal;
+ shared->done = done;
+ SpinLockRelease(&shared->mutex);
- repack_decode_concurrent_changes(decoding_ctx, end_of_wal);
+ /*
+ * The worker needs to finish processing of the current WAL record. Even
+ * if it's idle, it'll need to close the output file. Thus we're likely to
+ * wait, so prepare for sleep.
+ */
+ ConditionVariablePrepareToSleep(&shared->cv);
+ for (;;)
+ {
+ bool valid;
- if (dstate->nchanges == 0)
- return;
+ SpinLockAcquire(&shared->mutex);
+ valid = shared->sfs_valid;
+ SpinLockRelease(&shared->mutex);
+
+ if (valid)
+ break;
+
+ ConditionVariableSleep(&shared->cv, WAIT_EVENT_REPACK_WORKER_EXPORT);
+ }
+ ConditionVariableCancelSleep();
- apply_concurrent_changes(dstate, dest);
+ /* Open the file. */
+ DecodingWorkerFileName(fname, shared->relid, shared->last_exported);
+ file = BufFileOpenFileSet(&shared->sfs.fs, fname, O_RDONLY, false);
+ apply_concurrent_changes(file, dest);
+
+ /* No file is exported until the worker exports the next one. */
+ SpinLockAcquire(&shared->mutex);
+ shared->sfs_valid = false;
+ SpinLockRelease(&shared->mutex);
+
+ BufFileClose(file);
}
/*
@@ -3044,15 +3179,10 @@ cleanup_logical_decoding(LogicalDecodingContext *ctx)
dstate = (RepackDecodingState *) ctx->output_writer_private;
- ExecDropSingleTupleTableSlot(dstate->tsslot);
- FreeTupleDesc(dstate->tupdesc_change);
FreeTupleDesc(dstate->tupdesc);
- tuplestore_end(dstate->tstore);
-
FreeDecodingContext(ctx);
- ReplicationSlotRelease();
- ReplicationSlotDrop(NameStr(dstate->slotname), false);
+ ReplicationSlotDropAcquired();
pfree(dstate);
}
@@ -3067,7 +3197,6 @@ cleanup_logical_decoding(LogicalDecodingContext *ctx)
static void
rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
Relation cl_index,
- LogicalDecodingContext *decoding_ctx,
TransactionId frozenXid,
MultiXactId cutoffMulti)
{
@@ -3170,7 +3299,7 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
* hold AccessExclusiveLock. (Quite some amount of WAL could have been
* written during the data copying and index creation.)
*/
- process_concurrent_changes(decoding_ctx, end_of_wal, &chgdst);
+ process_concurrent_changes(end_of_wal, &chgdst, false);
/*
* Acquire AccessExclusiveLock on the table, its TOAST relation (if there
@@ -3266,8 +3395,11 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
XLogFlush(wal_insert_ptr);
end_of_wal = GetFlushRecPtr(NULL);
- /* Apply the concurrent changes again. */
- process_concurrent_changes(decoding_ctx, end_of_wal, &chgdst);
+ /*
+ * Apply the concurrent changes again. Indicate that the decoding worker
+ * won't be needed anymore.
+ */
+ process_concurrent_changes(end_of_wal, &chgdst, true);
/* Remember info about rel before closing OldHeap */
relpersistence = OldHeap->rd_rel->relpersistence;
@@ -3380,3 +3512,488 @@ build_new_indexes(Relation NewHeap, Relation OldHeap, List *OldIndexes)
return result;
}
+
+/*
+ * Try to start a background worker to perform logical decoding of data
+ * changes applied to relation while REPACK CONCURRENTLY is copying its
+ * contents to a new table.
+ */
+static void
+start_decoding_worker(Oid relid)
+{
+ Size size;
+ dsm_segment *seg;
+ DecodingWorkerShared *shared;
+ shm_mq *mq;
+ shm_mq_handle *mqh;
+ BackgroundWorker bgw;
+
+ /* Setup shared memory. */
+ size = BUFFERALIGN(offsetof(DecodingWorkerShared, error_queue)) +
+ BUFFERALIGN(REPACK_ERROR_QUEUE_SIZE);
+ seg = dsm_create(size, 0);
+ shared = (DecodingWorkerShared *) dsm_segment_address(seg);
+ shared->lsn_upto = InvalidXLogRecPtr;
+ shared->done = false;
+ SharedFileSetInit(&shared->sfs, seg);
+ shared->sfs_valid = false;
+ shared->last_exported = -1;
+ SpinLockInit(&shared->mutex);
+ shared->dbid = MyDatabaseId;
+
+ /*
+ * This is the UserId set in cluster_rel(). Security context shouldn't be
+ * needed for decoding worker.
+ */
+ shared->roleid = GetUserId();
+ shared->relid = relid;
+ ConditionVariableInit(&shared->cv);
+ shared->backend_proc = MyProc;
+ shared->backend_pid = MyProcPid;
+ shared->backend_proc_number = MyProcNumber;
+
+ mq = shm_mq_create((char *) BUFFERALIGN(shared->error_queue),
+ REPACK_ERROR_QUEUE_SIZE);
+ shm_mq_set_receiver(mq, MyProc);
+ mqh = shm_mq_attach(mq, seg, NULL);
+
+ memset(&bgw, 0, sizeof(bgw));
+ snprintf(bgw.bgw_name, BGW_MAXLEN,
+ "REPACK decoding worker for relation \"%s\"",
+ get_rel_name(relid));
+ snprintf(bgw.bgw_type, BGW_MAXLEN, "REPACK decoding worker");
+ bgw.bgw_flags = BGWORKER_SHMEM_ACCESS |
+ BGWORKER_BACKEND_DATABASE_CONNECTION;
+ bgw.bgw_start_time = BgWorkerStart_RecoveryFinished;
+ bgw.bgw_restart_time = BGW_NEVER_RESTART;
+ snprintf(bgw.bgw_library_name, MAXPGPATH, "postgres");
+ snprintf(bgw.bgw_function_name, BGW_MAXLEN, "RepackWorkerMain");
+ bgw.bgw_main_arg = UInt32GetDatum(dsm_segment_handle(seg));
+ bgw.bgw_notify_pid = MyProcPid;
+
+ decoding_worker = palloc0_object(DecodingWorker);
+ if (!RegisterDynamicBackgroundWorker(&bgw, &decoding_worker->handle))
+ ereport(ERROR,
+ (errcode(ERRCODE_CONFIGURATION_LIMIT_EXCEEDED),
+ errmsg("out of background worker slots"),
+ errhint("You might need to increase \"%s\".", "max_worker_processes")));
+
+ decoding_worker->seg = seg;
+ decoding_worker->error_mqh = mqh;
+}
+
+/*
+ * Stop the decoding worker and cleanup the related resources.
+ *
+ * The worker stops on its own when it knows there is no more work to do, but
+ * we need to stop it explicitly at least on ERROR in the launching backend.
+ */
+static void
+stop_decoding_worker(void)
+{
+ BgwHandleStatus status;
+
+ /* Haven't reached the worker startup? */
+ if (decoding_worker == NULL)
+ return;
+
+ /* Could not register the worker? */
+ if (decoding_worker->handle == NULL)
+ return;
+
+ TerminateBackgroundWorker(decoding_worker->handle);
+ /* The worker should really exit before the REPACK command does. */
+ HOLD_INTERRUPTS();
+ status = WaitForBackgroundWorkerShutdown(decoding_worker->handle);
+ RESUME_INTERRUPTS();
+
+ if (status == BGWH_POSTMASTER_DIED)
+ ereport(FATAL,
+ (errcode(ERRCODE_ADMIN_SHUTDOWN),
+ errmsg("postmaster exited during REPACK command")));
+
+ shm_mq_detach(decoding_worker->error_mqh);
+
+ /*
+ * If we could not cancel the current sleep due to ERROR, do that before
+ * we detach from the shared memory the condition variable is located in.
+ * If we did not, the bgworker ERROR handling code would try and fail
+ * badly.
+ */
+ ConditionVariableCancelSleep();
+
+ dsm_detach(decoding_worker->seg);
+ pfree(decoding_worker);
+ decoding_worker = NULL;
+}
+
+/* Is this process a REPACK worker? */
+static bool is_repack_worker = false;
+
+static pid_t backend_pid;
+static ProcNumber backend_proc_number;
+
+/*
+ * See ParallelWorkerShutdown for details.
+ */
+static void
+RepackWorkerShutdown(int code, Datum arg)
+{
+ SendProcSignal(backend_pid,
+ PROCSIG_REPACK_MESSAGE,
+ backend_proc_number);
+
+ dsm_detach((dsm_segment *) DatumGetPointer(arg));
+}
+
+/* REPACK decoding worker entry point */
+void
+RepackWorkerMain(Datum main_arg)
+{
+ dsm_segment *seg;
+ DecodingWorkerShared *shared;
+ shm_mq *mq;
+ shm_mq_handle *mqh;
+
+ is_repack_worker = true;
+
+ /*
+ * Override the default bgworker_die() with die() so we can use
+ * CHECK_FOR_INTERRUPTS().
+ */
+ pqsignal(SIGTERM, die);
+ BackgroundWorkerUnblockSignals();
+
+ seg = dsm_attach(DatumGetUInt32(main_arg));
+ if (seg == NULL)
+ ereport(ERROR,
+ (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
+ errmsg("could not map dynamic shared memory segment")));
+
+ shared = (DecodingWorkerShared *) dsm_segment_address(seg);
+
+ /* Arrange to signal the leader if we exit. */
+ backend_pid = shared->backend_pid;
+ backend_proc_number = shared->backend_proc_number;
+ before_shmem_exit(RepackWorkerShutdown, PointerGetDatum(seg));
+
+ /*
+ * Join locking group - see the comments around the call of
+ * start_decoding_worker().
+ */
+ if (!BecomeLockGroupMember(shared->backend_proc, backend_pid))
+ /* The leader is not running anymore. */
+ return;
+
+ /*
+ * Setup a queue to send error messages to the backend that launched this
+ * worker.
+ */
+ mq = (shm_mq *) (char *) BUFFERALIGN(shared->error_queue);
+ shm_mq_set_sender(mq, MyProc);
+ mqh = shm_mq_attach(mq, seg, NULL);
+ pq_redirect_to_shm_mq(seg, mqh);
+ pq_set_parallel_leader(shared->backend_pid,
+ shared->backend_proc_number);
+
+ /* Connect to the database. */
+ BackgroundWorkerInitializeConnectionByOid(shared->dbid, shared->roleid, 0);
+
+ repack_worker_internal(seg);
+}
+
+static void
+repack_worker_internal(dsm_segment *seg)
+{
+ DecodingWorkerShared *shared;
+ LogicalDecodingContext *decoding_ctx;
+ SharedFileSet *sfs;
+ Snapshot snapshot;
+
+ /*
+ * Transaction is needed to open relation, and it also provides us with a
+ * resource owner.
+ */
+ StartTransactionCommand();
+
+ shared = (DecodingWorkerShared *) dsm_segment_address(seg);
+
+ /*
+ * Not sure the spinlock is needed here - the backend should not change
+ * anything in the shared memory until we have serialized the snapshot.
+ */
+ SpinLockAcquire(&shared->mutex);
+ Assert(XLogRecPtrIsInvalid(shared->lsn_upto));
+ Assert(!shared->sfs_valid);
+ sfs = &shared->sfs;
+ SpinLockRelease(&shared->mutex);
+
+ SharedFileSetAttach(sfs, seg);
+
+ /*
+ * Prepare to capture the concurrent data changes ourselves.
+ */
+ decoding_ctx = setup_logical_decoding(shared->relid);
+
+ /* Build the initial snapshot and export it. */
+ snapshot = SnapBuildInitialSnapshotForRepack(decoding_ctx->snapshot_builder);
+ export_initial_snapshot(snapshot, shared);
+
+ /*
+ * The worker already had to access some system catalogs during startup,
+ * and we even had to open the relation we are processing. Now that we're
+ * going to work with historic snapshots, the system caches must be
+ * invalidated.
+ */
+ InvalidateSystemCaches();
+
+ while (!decode_concurrent_changes(decoding_ctx, shared))
+ ;
+
+ /* Cleanup. */
+ cleanup_logical_decoding(decoding_ctx);
+ CommitTransactionCommand();
+}
+
+/*
+ * Make snapshot available to the backend that launched the decoding worker.
+ */
+static void
+export_initial_snapshot(Snapshot snapshot, DecodingWorkerShared *shared)
+{
+ char fname[MAXPGPATH];
+ BufFile *file;
+ Size snap_size;
+ char *snap_space;
+
+ snap_size = EstimateSnapshotSpace(snapshot);
+ snap_space = (char *) palloc(snap_size);
+ SerializeSnapshot(snapshot, snap_space);
+ FreeSnapshot(snapshot);
+
+ DecodingWorkerFileName(fname, shared->relid, shared->last_exported + 1);
+ file = BufFileCreateFileSet(&shared->sfs.fs, fname);
+ /* To make restoration easier, write the snapshot size first. */
+ BufFileWrite(file, &snap_size, sizeof(snap_size));
+ BufFileWrite(file, snap_space, snap_size);
+ pfree(snap_space);
+ BufFileClose(file);
+
+ /* Tell the backend that the file is available. */
+ SpinLockAcquire(&shared->mutex);
+ shared->sfs_valid = true;
+ shared->last_exported++;
+ SpinLockRelease(&shared->mutex);
+ ConditionVariableSignal(&shared->cv);
+}
+
+/*
+ * Get the initial snapshot from the decoding worker.
+ */
+static Snapshot
+get_initial_snapshot(DecodingWorker *worker)
+{
+ DecodingWorkerShared *shared;
+ char fname[MAXPGPATH];
+ BufFile *file;
+ Size snap_size;
+ char *snap_space;
+ Snapshot snapshot;
+
+ shared = (DecodingWorkerShared *) dsm_segment_address(worker->seg);
+
+ /*
+ * The worker needs to initialize the logical decoding, which usually
+ * takes some time. Therefore it makes sense to prepare for the sleep
+ * first.
+ */
+ ConditionVariablePrepareToSleep(&shared->cv);
+ for (;;)
+ {
+ bool valid;
+
+ SpinLockAcquire(&shared->mutex);
+ valid = shared->sfs_valid;
+ SpinLockRelease(&shared->mutex);
+
+ if (valid)
+ break;
+
+ ConditionVariableSleep(&shared->cv, WAIT_EVENT_REPACK_WORKER_EXPORT);
+ }
+ ConditionVariableCancelSleep();
+
+ /* Read the snapshot from a file. */
+ DecodingWorkerFileName(fname, shared->relid, shared->last_exported);
+ file = BufFileOpenFileSet(&shared->sfs.fs, fname, O_RDONLY, false);
+ BufFileReadExact(file, &snap_size, sizeof(snap_size));
+ snap_space = (char *) palloc(snap_size);
+ BufFileReadExact(file, snap_space, snap_size);
+ BufFileClose(file);
+
+ SpinLockAcquire(&shared->mutex);
+ shared->sfs_valid = false;
+ SpinLockRelease(&shared->mutex);
+
+ /* Restore it. */
+ snapshot = RestoreSnapshot(snap_space);
+ pfree(snap_space);
+
+ return snapshot;
+}
+
+bool
+IsRepackWorker(void)
+{
+ return is_repack_worker;
+}
+
+/*
+ * Handle receipt of an interrupt indicating a repack worker message.
+ *
+ * Note: this is called within a signal handler! All we can do is set
+ * a flag that will cause the next CHECK_FOR_INTERRUPTS() to invoke
+ * ProcessRepackMessages().
+ */
+void
+HandleRepackMessageInterrupt(void)
+{
+ InterruptPending = true;
+ RepackMessagePending = true;
+ SetLatch(MyLatch);
+}
+
+/*
+ * Process any queued protocol messages received from parallel workers.
+ */
+void
+ProcessRepackMessages(void)
+{
+ MemoryContext oldcontext;
+
+ static MemoryContext hpm_context = NULL;
+
+ /*
+ * Nothing to do if we haven't launched the worker yet or have already
+ * terminated it.
+ */
+ if (decoding_worker == NULL)
+ return;
+
+ /*
+ * This is invoked from ProcessInterrupts(), and since some of the
+ * functions it calls contain CHECK_FOR_INTERRUPTS(), there is a potential
+ * for recursive calls if more signals are received while this runs. It's
+ * unclear that recursive entry would be safe, and it doesn't seem useful
+ * even if it is safe, so let's block interrupts until done.
+ */
+ HOLD_INTERRUPTS();
+
+ /*
+ * Moreover, CurrentMemoryContext might be pointing almost anywhere. We
+ * don't want to risk leaking data into long-lived contexts, so let's do
+ * our work here in a private context that we can reset on each use.
+ */
+ if (hpm_context == NULL) /* first time through? */
+ hpm_context = AllocSetContextCreate(TopMemoryContext,
+ "ProcessParallelMessages",
+ ALLOCSET_DEFAULT_SIZES);
+ else
+ MemoryContextReset(hpm_context);
+
+ oldcontext = MemoryContextSwitchTo(hpm_context);
+
+ /* OK to process messages. Reset the flag saying there are more to do. */
+ RepackMessagePending = false;
+
+ /*
+ * Read as many messages as we can from each worker, but stop when no more
+ * messages can be read from the worker without blocking.
+ */
+ while (true)
+ {
+ shm_mq_result res;
+ Size nbytes;
+ void *data;
+
+ res = shm_mq_receive(decoding_worker->error_mqh, &nbytes,
+ &data, true);
+ if (res == SHM_MQ_WOULD_BLOCK)
+ break;
+ else if (res == SHM_MQ_SUCCESS)
+ {
+ StringInfoData msg;
+
+ initStringInfo(&msg);
+ appendBinaryStringInfo(&msg, data, nbytes);
+ ProcessRepackMessage(&msg);
+ pfree(msg.data);
+ }
+ else
+ {
+ /*
+ * The decoding worker is special in that it exits as soon as it
+ * has its work done. Thus the DETACHED result code is fine.
+ */
+ Assert(res = SHM_MQ_DETACHED);
+
+ break;
+ }
+ }
+
+ MemoryContextSwitchTo(oldcontext);
+
+ /* Might as well clear the context on our way out */
+ MemoryContextReset(hpm_context);
+
+ RESUME_INTERRUPTS();
+}
+
+/*
+ * Process a single protocol message received from a single parallel worker.
+ */
+static void
+ProcessRepackMessage(StringInfo msg)
+{
+ char msgtype;
+
+ msgtype = pq_getmsgbyte(msg);
+
+ switch (msgtype)
+ {
+ case PqMsg_ErrorResponse:
+ case PqMsg_NoticeResponse:
+ {
+ ErrorData edata;
+
+ /* Parse ErrorResponse or NoticeResponse. */
+ pq_parse_errornotice(msg, &edata);
+
+ /* Death of a worker isn't enough justification for suicide. */
+ edata.elevel = Min(edata.elevel, ERROR);
+
+ /*
+ * If desired, add a context line to show that this is a
+ * message propagated from a parallel worker. Otherwise, it
+ * can sometimes be confusing to understand what actually
+ * happened.
+ */
+ if (edata.context)
+ edata.context = psprintf("%s\n%s", edata.context,
+ _("decoding worker"));
+ else
+ edata.context = pstrdup(_("decoding worker"));
+
+ /* Rethrow error or print notice. */
+ ThrowErrorData(&edata);
+
+ break;
+ }
+
+ default:
+ {
+ elog(ERROR, "unrecognized message type received from decoding worker: %c (message length %d bytes)",
+ msgtype, msg->len);
+ }
+ }
+}
diff --git a/src/backend/libpq/pqmq.c b/src/backend/libpq/pqmq.c
index 2b75de0ddef..28a5a400fb1 100644
--- a/src/backend/libpq/pqmq.c
+++ b/src/backend/libpq/pqmq.c
@@ -14,6 +14,7 @@
#include "postgres.h"
#include "access/parallel.h"
+#include "commands/cluster.h"
#include "libpq/libpq.h"
#include "libpq/pqformat.h"
#include "libpq/pqmq.h"
@@ -175,6 +176,10 @@ mq_putmessage(char msgtype, const char *s, size_t len)
SendProcSignal(pq_mq_parallel_leader_pid,
PROCSIG_PARALLEL_APPLY_MESSAGE,
pq_mq_parallel_leader_proc_number);
+ else if (IsRepackWorker())
+ SendProcSignal(pq_mq_parallel_leader_pid,
+ PROCSIG_REPACK_MESSAGE,
+ pq_mq_parallel_leader_proc_number);
else
{
Assert(IsParallelWorker());
diff --git a/src/backend/postmaster/bgworker.c b/src/backend/postmaster/bgworker.c
index 142a02eb5e9..b368990e90b 100644
--- a/src/backend/postmaster/bgworker.c
+++ b/src/backend/postmaster/bgworker.c
@@ -13,6 +13,7 @@
#include "postgres.h"
#include "access/parallel.h"
+#include "commands/cluster.h"
#include "libpq/pqsignal.h"
#include "miscadmin.h"
#include "pgstat.h"
@@ -135,6 +136,9 @@ static const struct
},
{
"SequenceSyncWorkerMain", SequenceSyncWorkerMain
+ },
+ {
+ "RepackWorkerMain", RepackWorkerMain
}
};
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index 866f92cf799..85323ba61e2 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -205,7 +205,11 @@ StartupDecodingContext(List *output_plugin_options,
ctx->slot = slot;
- ctx->reader = XLogReaderAllocate(wal_segment_size, NULL, xl_routine, ctx);
+ /*
+ * TODO A separate patch for PG core, unless there's really a reason to
+ * pass ctx for private_data (May extensions expect ctx?).
+ */
+ ctx->reader = XLogReaderAllocate(wal_segment_size, NULL, xl_routine, NULL);
if (!ctx->reader)
ereport(ERROR,
(errcode(ERRCODE_OUT_OF_MEMORY),
diff --git a/src/backend/replication/pgoutput_repack/pgoutput_repack.c b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
index c8930640a0d..fb9956d392d 100644
--- a/src/backend/replication/pgoutput_repack/pgoutput_repack.c
+++ b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
@@ -168,17 +168,13 @@ store_change(LogicalDecodingContext *ctx, ConcurrentChangeKind kind,
HeapTuple tuple)
{
RepackDecodingState *dstate;
- char *change_raw;
- ConcurrentChange change;
+ char kind_byte = (char) kind;
bool flattened = false;
- Size size;
- Datum values[1];
- bool isnull[1];
- char *dst;
dstate = (RepackDecodingState *) ctx->output_writer_private;
- size = VARHDRSZ + SizeOfConcurrentChange;
+ /* Store the change kind. */
+ BufFileWrite(dstate->file, &kind_byte, 1);
/*
* ReorderBufferCommit() stores the TOAST chunks in its private memory
@@ -195,46 +191,12 @@ store_change(LogicalDecodingContext *ctx, ConcurrentChangeKind kind,
tuple = toast_flatten_tuple(tuple, dstate->tupdesc);
flattened = true;
}
+ /* Store the tuple size ... */
+ BufFileWrite(dstate->file, &tuple->t_len, sizeof(tuple->t_len));
+ /* ... and the tuple itself. */
+ BufFileWrite(dstate->file, tuple->t_data, tuple->t_len);
- size += tuple->t_len;
- if (size >= MaxAllocSize)
- elog(ERROR, "Change is too big.");
-
- /* Construct the change. */
- change_raw = (char *) palloc0(size);
- SET_VARSIZE(change_raw, size);
-
- /*
- * Since the varlena alignment might not be sufficient for the structure,
- * set the fields in a local instance and remember where it should
- * eventually be copied.
- */
- change.kind = kind;
- dst = (char *) VARDATA(change_raw);
-
- /*
- * Copy the tuple.
- *
- * Note: change->tup_data.t_data must be fixed on retrieval!
- */
- memcpy(&change.tup_data, tuple, sizeof(HeapTupleData));
- memcpy(dst, &change, SizeOfConcurrentChange);
- dst += SizeOfConcurrentChange;
- memcpy(dst, tuple->t_data, tuple->t_len);
-
- /* The data has been copied. */
+ /* Free the flat copy if created above. */
if (flattened)
pfree(tuple);
-
- /* Store as tuple of 1 bytea column. */
- values[0] = PointerGetDatum(change_raw);
- isnull[0] = false;
- tuplestore_putvalues(dstate->tstore, dstate->tupdesc_change,
- values, isnull);
-
- /* Accounting. */
- dstate->nchanges++;
-
- /* Cleanup. */
- pfree(change_raw);
}
diff --git a/src/backend/storage/ipc/procsignal.c b/src/backend/storage/ipc/procsignal.c
index 087821311cc..af12144795b 100644
--- a/src/backend/storage/ipc/procsignal.c
+++ b/src/backend/storage/ipc/procsignal.c
@@ -19,6 +19,7 @@
#include "access/parallel.h"
#include "commands/async.h"
+#include "commands/cluster.h"
#include "miscadmin.h"
#include "pgstat.h"
#include "port/pg_bitutils.h"
@@ -694,6 +695,9 @@ procsignal_sigusr1_handler(SIGNAL_ARGS)
if (CheckProcSignal(PROCSIG_PARALLEL_APPLY_MESSAGE))
HandleParallelApplyMessageInterrupt();
+ if (CheckProcSignal(PROCSIG_REPACK_MESSAGE))
+ HandleRepackMessageInterrupt();
+
if (CheckProcSignal(PROCSIG_RECOVERY_CONFLICT_DATABASE))
HandleRecoveryConflictInterrupt(PROCSIG_RECOVERY_CONFLICT_DATABASE);
diff --git a/src/backend/tcop/postgres.c b/src/backend/tcop/postgres.c
index 7dd75a490aa..4a4858882f0 100644
--- a/src/backend/tcop/postgres.c
+++ b/src/backend/tcop/postgres.c
@@ -36,6 +36,7 @@
#include "access/xact.h"
#include "catalog/pg_type.h"
#include "commands/async.h"
+#include "commands/cluster.h"
#include "commands/event_trigger.h"
#include "commands/explain_state.h"
#include "commands/prepare.h"
@@ -3541,6 +3542,9 @@ ProcessInterrupts(void)
if (ParallelApplyMessagePending)
ProcessParallelApplyMessages();
+
+ if (RepackMessagePending)
+ ProcessRepackMessages();
}
/*
diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt
index f39830dbb34..cbcc8550960 100644
--- a/src/backend/utils/activity/wait_event_names.txt
+++ b/src/backend/utils/activity/wait_event_names.txt
@@ -62,6 +62,7 @@ LOGICAL_APPLY_MAIN "Waiting in main loop of logical replication apply process."
LOGICAL_LAUNCHER_MAIN "Waiting in main loop of logical replication launcher process."
LOGICAL_PARALLEL_APPLY_MAIN "Waiting in main loop of logical replication parallel apply process."
RECOVERY_WAL_STREAM "Waiting in main loop of startup process for WAL to arrive, during streaming recovery."
+REPACK_WORKER_MAIN "Waiting in main loop of REPACK decoding worker process."
REPLICATION_SLOTSYNC_MAIN "Waiting in main loop of slot sync worker."
REPLICATION_SLOTSYNC_SHUTDOWN "Waiting for slot sync worker to shut down."
SYSLOGGER_MAIN "Waiting in main loop of syslogger process."
@@ -153,6 +154,7 @@ RECOVERY_CONFLICT_SNAPSHOT "Waiting for recovery conflict resolution for a vacuu
RECOVERY_CONFLICT_TABLESPACE "Waiting for recovery conflict resolution for dropping a tablespace."
RECOVERY_END_COMMAND "Waiting for <xref linkend="guc-recovery-end-command"/> to complete."
RECOVERY_PAUSE "Waiting for recovery to be resumed."
+REPACK_WORKER_EXPORT "Waiting for decoding worker to export a new output file."
REPLICATION_ORIGIN_DROP "Waiting for a replication origin to become inactive so it can be dropped."
REPLICATION_SLOT_DROP "Waiting for a replication slot to become inactive so it can be dropped."
RESTORE_COMMAND "Waiting for <xref linkend="guc-restore-command"/> to complete."
diff --git a/src/include/access/tableam.h b/src/include/access/tableam.h
index d8f76d325f9..2b983abce3e 100644
--- a/src/include/access/tableam.h
+++ b/src/include/access/tableam.h
@@ -22,7 +22,6 @@
#include "access/xact.h"
#include "commands/vacuum.h"
#include "executor/tuptable.h"
-#include "replication/logical.h"
#include "storage/read_stream.h"
#include "utils/rel.h"
#include "utils/snapshot.h"
@@ -631,7 +630,6 @@ typedef struct TableAmRoutine
bool use_sort,
TransactionId OldestXmin,
Snapshot snapshot,
- LogicalDecodingContext *decoding_ctx,
TransactionId *xid_cutoff,
MultiXactId *multi_cutoff,
double *num_tuples,
@@ -1652,7 +1650,7 @@ table_relation_copy_data(Relation rel, const RelFileLocator *newrlocator)
* - snapshot - if != NULL, ignore data changes done by transactions that this
* (MVCC) snapshot considers still in-progress or in the future.
* - decoding_ctx - logical decoding context, to capture concurrent data
- * changes.
+ * changes. NULL if background worker takes care of the decoding.
*
* Output parameters:
* - *xid_cutoff - rel's new relfrozenxid value, may be invalid
@@ -1666,7 +1664,6 @@ table_relation_copy_for_cluster(Relation OldTable, Relation NewTable,
bool use_sort,
TransactionId OldestXmin,
Snapshot snapshot,
- LogicalDecodingContext *decoding_ctx,
TransactionId *xid_cutoff,
MultiXactId *multi_cutoff,
double *num_tuples,
@@ -1675,7 +1672,7 @@ table_relation_copy_for_cluster(Relation OldTable, Relation NewTable,
{
OldTable->rd_tableam->relation_copy_for_cluster(OldTable, NewTable, OldIndex,
use_sort, OldestXmin,
- snapshot, decoding_ctx,
+ snapshot,
xid_cutoff, multi_cutoff,
num_tuples, tups_vacuumed,
tups_recently_dead);
diff --git a/src/include/commands/cluster.h b/src/include/commands/cluster.h
index b43a1740053..0ac70ec30d7 100644
--- a/src/include/commands/cluster.h
+++ b/src/include/commands/cluster.h
@@ -17,6 +17,7 @@
#include "nodes/parsenodes.h"
#include "parser/parse_node.h"
#include "replication/logical.h"
+#include "storage/buffile.h"
#include "storage/lock.h"
#include "storage/relfilelocator.h"
#include "utils/relcache.h"
@@ -47,6 +48,9 @@ typedef struct ClusterParams
extern RelFileLocator repacked_rel_locator;
extern RelFileLocator repacked_rel_toast_locator;
+/*
+ * Stored as a single byte in the output file.
+ */
typedef enum
{
CHANGE_INSERT,
@@ -55,68 +59,30 @@ typedef enum
CHANGE_DELETE
} ConcurrentChangeKind;
-typedef struct ConcurrentChange
-{
- /* See the enum above. */
- ConcurrentChangeKind kind;
-
- /*
- * The actual tuple.
- *
- * The tuple data follows the ConcurrentChange structure. Before use make
- * sure the tuple is correctly aligned (ConcurrentChange can be stored as
- * bytea) and that tuple->t_data is fixed.
- */
- HeapTupleData tup_data;
-} ConcurrentChange;
-
-#define SizeOfConcurrentChange (offsetof(ConcurrentChange, tup_data) + \
- sizeof(HeapTupleData))
-
/*
* Logical decoding state.
*
- * Here we store the data changes that we decode from WAL while the table
- * contents is being copied to a new storage. Also the necessary metadata
- * needed to apply these changes to the table is stored here.
+ * The output plugin uses it to store the data changes that it decodes from
+ * WAL while the table contents is being copied to a new storage.
*/
typedef struct RepackDecodingState
{
/* The relation whose changes we're decoding. */
Oid relid;
- /* Replication slot name. */
- NameData slotname;
-
- /*
- * Decoded changes are stored here. Although we try to avoid excessive
- * batches, it can happen that the changes need to be stored to disk. The
- * tuplestore does this transparently.
- */
- Tuplestorestate *tstore;
-
- /* The current number of changes in tstore. */
- double nchanges;
-
- /*
- * Descriptor to store the ConcurrentChange structure serialized (bytea).
- * We can't store the tuple directly because tuplestore only supports
- * minimum tuple and we may need to transfer OID system column from the
- * output plugin. Also we need to transfer the change kind, so it's better
- * to put everything in the structure than to use 2 tuplestores "in
- * parallel".
- */
- TupleDesc tupdesc_change;
-
- /* Tuple descriptor needed to update indexes. */
+ /* Tuple descriptor of the relation being processed. */
TupleDesc tupdesc;
- /* Slot to retrieve data from tstore. */
- TupleTableSlot *tsslot;
-
- ResourceOwner resowner;
+ /* The current output file. */
+ BufFile *file;
} RepackDecodingState;
+extern PGDLLIMPORT volatile sig_atomic_t RepackMessagePending;
+
+extern bool IsRepackWorker(void);
+extern void HandleRepackMessageInterrupt(void);
+extern void ProcessRepackMessages(void);
+
extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel);
extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid,
@@ -125,9 +91,6 @@ extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid,
LOCKMODE lockmode);
extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal);
-extern void repack_decode_concurrent_changes(LogicalDecodingContext *ctx,
- XLogRecPtr end_of_wal);
-
extern Oid make_new_heap(Oid OIDOldHeap, Oid NewTableSpace, Oid NewAccessMethod,
char relpersistence, LOCKMODE lockmode);
extern void finish_heap_swap(Oid OIDOldHeap, Oid OIDNewHeap,
@@ -140,4 +103,5 @@ extern void finish_heap_swap(Oid OIDOldHeap, Oid OIDNewHeap,
MultiXactId cutoffMulti,
char newrelpersistence);
+extern void RepackWorkerMain(Datum main_arg);
#endif /* CLUSTER_H */
diff --git a/src/include/storage/procsignal.h b/src/include/storage/procsignal.h
index afeeb1ca019..c0a66516b66 100644
--- a/src/include/storage/procsignal.h
+++ b/src/include/storage/procsignal.h
@@ -36,6 +36,7 @@ typedef enum
PROCSIG_BARRIER, /* global barrier interrupt */
PROCSIG_LOG_MEMORY_CONTEXT, /* ask backend to log the memory contexts */
PROCSIG_PARALLEL_APPLY_MESSAGE, /* Message from parallel apply workers */
+ PROCSIG_REPACK_MESSAGE, /* Message from repack worker */
/* Recovery conflict reasons */
PROCSIG_RECOVERY_CONFLICT_FIRST,
diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list
index 3139b14e85f..35344910f65 100644
--- a/src/tools/pgindent/typedefs.list
+++ b/src/tools/pgindent/typedefs.list
@@ -488,7 +488,6 @@ CompressFileHandle
CompressionLocation
CompressorState
ComputeXidHorizonsResult
-ConcurrentChange
ConcurrentChangeKind
ConditionVariable
ConditionVariableMinimallyPadded
@@ -629,6 +628,9 @@ DeclareCursorStmt
DecodedBkpBlock
DecodedXLogRecord
DecodingOutputState
+DecodingWorker
+DecodingWorkerShared
+DecodingWorkerState
DefElem
DefElemAction
DefaultACLInfo
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 55+ messages in thread
* [PATCH 5/5] Use background worker to do logical decoding.
@ 2025-12-09 18:44 Antonin Houska <[email protected]>
0 siblings, 0 replies; 55+ messages in thread
From: Antonin Houska @ 2025-12-09 18:44 UTC (permalink / raw)
If the backend performing REPACK (CONCURRENTLY) does both data copying and
logical decoding, it has to "travel in time" back and forth and therefore it
has to invalidate system caches quite a few times. (The copying and the
decoding work with different catalog snapshots.) As the decoding worker has
separate caches, the switching is not necessary.
Without the worker, it'd also be difficult to switch between potentiallly long
running tasks like index build and WAL decoding. (No decoding during that time
at all can suspend archiving / recycling of WAL segments for some time, which
in turn may result in full disk.)
Another problem is that, after having acquired AccessExclusiveLock (in order
to swap the files), the backend needs to both decode and apply the data
changes that took place while it was waiting for the lock. With the decoding
worker, the decoding runs all the time, so the backend only needs to apply the
changes. This can reduce the time the exclusive lock is held for.
Note that the code added in order to handle ERRORs in the background worker
almost duplicates the existing code that does the same for other types of
workers (See ProcessParallelMessages() and
ProcessParallelApplyMessages()). Refactoring of the existing code might be
useful, to reduce the duplication.
---
src/backend/access/heap/heapam_handler.c | 44 -
src/backend/commands/cluster.c | 1139 +++++++++++++----
src/backend/libpq/pqmq.c | 5 +
src/backend/postmaster/bgworker.c | 4 +
src/backend/replication/logical/logical.c | 6 +-
.../pgoutput_repack/pgoutput_repack.c | 54 +-
src/backend/storage/ipc/procsignal.c | 4 +
src/backend/tcop/postgres.c | 4 +
.../utils/activity/wait_event_names.txt | 2 +
src/include/access/tableam.h | 7 +-
src/include/commands/cluster.h | 68 +-
src/include/storage/procsignal.h | 1 +
src/tools/pgindent/typedefs.list | 4 +-
13 files changed, 932 insertions(+), 410 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index 01be29eb405..e6d630fa2f7 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -33,7 +33,6 @@
#include "catalog/index.h"
#include "catalog/storage.h"
#include "catalog/storage_xlog.h"
-#include "commands/cluster.h"
#include "commands/progress.h"
#include "executor/executor.h"
#include "miscadmin.h"
@@ -688,7 +687,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
Relation OldIndex, bool use_sort,
TransactionId OldestXmin,
Snapshot snapshot,
- LogicalDecodingContext *decoding_ctx,
TransactionId *xid_cutoff,
MultiXactId *multi_cutoff,
double *num_tuples,
@@ -710,7 +708,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
BufferHeapTupleTableSlot *hslot;
BlockNumber prev_cblock = InvalidBlockNumber;
bool concurrent = snapshot != NULL;
- XLogRecPtr end_of_wal_prev = GetFlushRecPtr(NULL);
/* Remember if it's a system catalog */
is_system_catalog = IsSystemRelation(OldHeap);
@@ -957,31 +954,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
ct_val[1] = *num_tuples;
pgstat_progress_update_multi_param(2, ct_index, ct_val);
}
-
- /*
- * Process the WAL produced by the load, as well as by other
- * transactions, so that the replication slot can advance and WAL does
- * not pile up. Use wal_segment_size as a threshold so that we do not
- * introduce the decoding overhead too often.
- *
- * Of course, we must not apply the changes until the initial load has
- * completed.
- *
- * Note that our insertions into the new table should not be decoded
- * as we (intentionally) do not write the logical decoding specific
- * information to WAL.
- */
- if (concurrent)
- {
- XLogRecPtr end_of_wal;
-
- end_of_wal = GetFlushRecPtr(NULL);
- if ((end_of_wal - end_of_wal_prev) > wal_segment_size)
- {
- repack_decode_concurrent_changes(decoding_ctx, end_of_wal);
- end_of_wal_prev = end_of_wal;
- }
- }
}
if (indexScan != NULL)
@@ -1027,22 +999,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
-
- /*
- * Try to keep the amount of not-yet-decoded WAL small, like
- * above.
- */
- if (concurrent)
- {
- XLogRecPtr end_of_wal;
-
- end_of_wal = GetFlushRecPtr(NULL);
- if ((end_of_wal - end_of_wal_prev) > wal_segment_size)
- {
- repack_decode_concurrent_changes(decoding_ctx, end_of_wal);
- end_of_wal_prev = end_of_wal;
- }
- }
}
tuplesort_end(tuplesort);
diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c
index 501bd36c23e..f2a2ec6d3e5 100644
--- a/src/backend/commands/cluster.c
+++ b/src/backend/commands/cluster.c
@@ -46,6 +46,8 @@
#include "commands/tablecmds.h"
#include "commands/vacuum.h"
#include "executor/executor.h"
+#include "libpq/pqformat.h"
+#include "libpq/pqmq.h"
#include "miscadmin.h"
#include "optimizer/optimizer.h"
#include "pgstat.h"
@@ -56,6 +58,8 @@
#include "storage/ipc.h"
#include "storage/lmgr.h"
#include "storage/predicate.h"
+#include "storage/procsignal.h"
+#include "tcop/tcopprot.h"
#include "utils/acl.h"
#include "utils/fmgroids.h"
#include "utils/guc.h"
@@ -123,6 +127,103 @@ typedef struct ChangeDest
IndexInsertState *iistate;
} ChangeDest;
+/*
+ * Layout of shared memory used for communication between backend and the
+ * worker that performs logical decoding of data changes
+ */
+typedef struct DecodingWorkerShared
+{
+ /*
+ * Once the worker has reached this LSN, it should close the current
+ * output file and either create a new one or exit, according to the field
+ * 'done'. If the value is InvalidXLogRecPtr, the worker should decode all
+ * the WAL available and keep checking this field. It is ok if the worker
+ * had already decoded records whose LSN is >= lsn_upto before this field
+ * has been set.
+ */
+ XLogRecPtr lsn_upto;
+
+ /* Exit after closing the current file? */
+ bool done;
+
+ /* The output is stored here. */
+ SharedFileSet sfs;
+
+ /* Can backend read the file contents? */
+ bool sfs_valid;
+
+ /* Number of the last file exported by the worker. */
+ int last_exported;
+
+ /* Synchronize access to the fields above. */
+ slock_t mutex;
+
+ /* Database to connect to. */
+ Oid dbid;
+
+ /* Role to connect as. */
+ Oid roleid;
+
+ /* Decode data changes of this relation. */
+ Oid relid;
+
+ /* The backend uses this to wait for the worker. */
+ ConditionVariable cv;
+
+ /* Info to signal the backend. */
+ PGPROC *backend_proc;
+ pid_t backend_pid;
+ ProcNumber backend_proc_number;
+
+ /* Error queue. */
+ shm_mq *error_mq;
+
+ /*
+ * Memory the queue is located int.
+ *
+ * For considerations on the value see the comments of
+ * PARALLEL_ERROR_QUEUE_SIZE.
+ */
+#define REPACK_ERROR_QUEUE_SIZE 16384
+ char error_queue[FLEXIBLE_ARRAY_MEMBER];
+} DecodingWorkerShared;
+
+/*
+ * Generate output file name. If relations of the same 'relid' happen to be
+ * processed at the same time, they must be from different databases and
+ * therefore different backends must be involved. (PID is already present in
+ * the fileset name.)
+ */
+static inline void
+DecodingWorkerFileName(char *fname, Oid relid, uint32 seq)
+{
+ snprintf(fname, MAXPGPATH, "%u-%u", relid, seq);
+}
+
+/*
+ * Backend-local information to control the decoding worker.
+ */
+typedef struct DecodingWorker
+{
+ /* The worker. */
+ BackgroundWorkerHandle *handle;
+
+ /* DecodingWorkerShared is in this segment. */
+ dsm_segment *seg;
+
+ /* Handle of the error queue. */
+ shm_mq_handle *error_mqh;
+} DecodingWorker;
+
+/* Pointer to currently running decoding worker. */
+static DecodingWorker *decoding_worker = NULL;
+
+/*
+ * Is there a message sent by a repack worker that the backend needs to
+ * receive?
+ */
+volatile sig_atomic_t RepackMessagePending = false;
+
static bool cluster_rel_recheck(RepackCommand cmd, Relation OldHeap,
Oid indexOid, Oid userid, LOCKMODE lmode,
int options);
@@ -130,7 +231,7 @@ static void check_repack_concurrently_requirements(Relation rel);
static void rebuild_relation(Relation OldHeap, Relation index, bool verbose,
bool concurrent);
static void copy_table_data(Relation NewHeap, Relation OldHeap, Relation OldIndex,
- Snapshot snapshot, LogicalDecodingContext *decoding_ctx,
+ Snapshot snapshot,
bool verbose,
bool *pSwapToastByContent,
TransactionId *pFreezeXid,
@@ -143,12 +244,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd,
static bool cluster_is_permitted_for_relation(RepackCommand cmd,
Oid relid, Oid userid);
-static void begin_concurrent_repack(Relation rel);
-static void end_concurrent_repack(void);
static LogicalDecodingContext *setup_logical_decoding(Oid relid);
-static HeapTuple get_changed_tuple(char *change);
-static void apply_concurrent_changes(RepackDecodingState *dstate,
- ChangeDest *dest);
+static bool decode_concurrent_changes(LogicalDecodingContext *ctx,
+ DecodingWorkerShared *shared);
+static void apply_concurrent_changes(BufFile *file, ChangeDest *dest);
static void apply_concurrent_insert(Relation rel, HeapTuple tup,
IndexInsertState *iistate,
TupleTableSlot *index_slot);
@@ -160,9 +259,9 @@ static void apply_concurrent_delete(Relation rel, HeapTuple tup_target);
static HeapTuple find_target_tuple(Relation rel, ChangeDest *dest,
HeapTuple tup_key,
TupleTableSlot *ident_slot);
-static void process_concurrent_changes(LogicalDecodingContext *decoding_ctx,
- XLogRecPtr end_of_wal,
- ChangeDest *dest);
+static void process_concurrent_changes(XLogRecPtr end_of_wal,
+ ChangeDest *dest,
+ bool done);
static IndexInsertState *get_index_insert_state(Relation relation,
Oid ident_index_id,
Relation *ident_index_p);
@@ -172,7 +271,6 @@ static void free_index_insert_state(IndexInsertState *iistate);
static void cleanup_logical_decoding(LogicalDecodingContext *ctx);
static void rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
Relation cl_index,
- LogicalDecodingContext *decoding_ctx,
TransactionId frozenXid,
MultiXactId cutoffMulti);
static List *build_new_indexes(Relation NewHeap, Relation OldHeap, List *OldIndexes);
@@ -182,6 +280,13 @@ static Relation process_single_relation(RepackStmt *stmt,
ClusterParams *params);
static Oid determine_clustered_index(Relation rel, bool usingindex,
const char *indexname);
+static void start_decoding_worker(Oid relid);
+static void stop_decoding_worker(void);
+static void repack_worker_internal(dsm_segment *seg);
+static void export_initial_snapshot(Snapshot snapshot,
+ DecodingWorkerShared *shared);
+static Snapshot get_initial_snapshot(DecodingWorker *worker);
+static void ProcessRepackMessage(StringInfo msg);
static const char *RepackCommandAsString(RepackCommand cmd);
@@ -604,20 +709,20 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid,
/* rebuild_relation does all the dirty work */
PG_TRY();
{
- /*
- * For concurrent processing, make sure that our logical decoding
- * ignores data changes of other tables than the one we are
- * processing.
- */
- if (concurrent)
- begin_concurrent_repack(OldHeap);
-
rebuild_relation(OldHeap, index, verbose, concurrent);
}
PG_FINALLY();
{
if (concurrent)
- end_concurrent_repack();
+ {
+ /*
+ * Since during normal operation the worker was already asked to
+ * exit, stopping it explicitly is especially important on ERROR.
+ * However it still seems a good practice to make sure that the
+ * worker never survives the REPACK command.
+ */
+ stop_decoding_worker();
+ }
}
PG_END_TRY();
@@ -914,7 +1019,6 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
bool swap_toast_by_content;
TransactionId frozenXid;
MultiXactId cutoffMulti;
- LogicalDecodingContext *decoding_ctx = NULL;
Snapshot snapshot = NULL;
#if USE_ASSERT_CHECKING
LOCKMODE lmode;
@@ -928,19 +1032,36 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
if (concurrent)
{
/*
- * Prepare to capture the concurrent data changes.
+ * The worker needs to be member of the locking group we're the leader
+ * of. We ought to become the leader before the worker starts. The
+ * worker will join the group as soon as it starts.
+ *
+ * This is to make sure that the deadlock described below is
+ * detectable by deadlock.c: if the worker waits for a transaction to
+ * complete and we are waiting for the worker output, then effectively
+ * we (i.e. this backend) are waiting for that transaction.
+ */
+ BecomeLockGroupLeader();
+
+ /*
+ * Start the worker that decodes data changes applied while we're
+ * copying the table contents.
*
- * Note that this call waits for all transactions with XID already
- * assigned to finish. If some of those transactions is waiting for a
- * lock conflicting with ShareUpdateExclusiveLock on our table (e.g.
- * it runs CREATE INDEX), we can end up in a deadlock. Not sure this
- * risk is worth unlocking/locking the table (and its clustering
- * index) and checking again if its still eligible for REPACK
- * CONCURRENTLY.
+ * Note that the worker has to wait for all transactions with XID
+ * already assigned to finish. If some of those transactions is
+ * waiting for a lock conflicting with ShareUpdateExclusiveLock on our
+ * table (e.g. it runs CREATE INDEX), we can end up in a deadlock.
+ * Not sure this risk is worth unlocking/locking the table (and its
+ * clustering index) and checking again if its still eligible for
+ * REPACK CONCURRENTLY.
+ */
+ start_decoding_worker(tableOid);
+
+ /*
+ * Wait until the worker has the initial snapshot and retrieve it.
*/
- decoding_ctx = setup_logical_decoding(tableOid);
+ snapshot = get_initial_snapshot(decoding_worker);
- snapshot = SnapBuildInitialSnapshotForRepack(decoding_ctx->snapshot_builder);
PushActiveSnapshot(snapshot);
}
@@ -965,7 +1086,7 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
NewHeap = table_open(OIDNewHeap, NoLock);
/* Copy the heap data into the new table in the desired order */
- copy_table_data(NewHeap, OldHeap, index, snapshot, decoding_ctx, verbose,
+ copy_table_data(NewHeap, OldHeap, index, snapshot, verbose,
&swap_toast_by_content, &frozenXid, &cutoffMulti);
/* The historic snapshot won't be needed anymore. */
@@ -989,15 +1110,11 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
Assert(!swap_toast_by_content);
rebuild_relation_finish_concurrent(NewHeap, OldHeap, index,
- decoding_ctx,
frozenXid, cutoffMulti);
PopActiveSnapshot();
pgstat_progress_update_param(PROGRESS_REPACK_PHASE,
PROGRESS_REPACK_PHASE_FINAL_CLEANUP);
-
- /* Done with decoding. */
- cleanup_logical_decoding(decoding_ctx);
}
else
{
@@ -1168,8 +1285,7 @@ make_new_heap(Oid OIDOldHeap, Oid NewTableSpace, Oid NewAccessMethod,
*/
static void
copy_table_data(Relation NewHeap, Relation OldHeap, Relation OldIndex,
- Snapshot snapshot, LogicalDecodingContext *decoding_ctx,
- bool verbose, bool *pSwapToastByContent,
+ Snapshot snapshot, bool verbose, bool *pSwapToastByContent,
TransactionId *pFreezeXid, MultiXactId *pCutoffMulti)
{
Relation relRelation;
@@ -1330,7 +1446,6 @@ copy_table_data(Relation NewHeap, Relation OldHeap, Relation OldIndex,
*/
table_relation_copy_for_cluster(OldHeap, NewHeap, OldIndex, use_sort,
cutoffs.OldestXmin, snapshot,
- decoding_ctx,
&cutoffs.FreezeLimit,
&cutoffs.MultiXactCutoff,
&num_tuples, &tups_vacuumed,
@@ -2363,59 +2478,6 @@ RepackCommandAsString(RepackCommand cmd)
return "???";
}
-
-/*
- * Call this function before REPACK CONCURRENTLY starts to setup logical
- * decoding. It makes sure that other users of the table put enough
- * information into WAL.
- *
- * The point is that at various places we expect that the table we're
- * processing is treated like a system catalog. For example, we need to be
- * able to scan it using a "historic snapshot" anytime during the processing
- * (as opposed to scanning only at the start point of the decoding, as logical
- * replication does during initial table synchronization), in order to apply
- * concurrent UPDATE / DELETE commands.
- *
- * Note that TOAST table needs no attention here as it's not scanned using
- * historic snapshot.
- */
-static void
-begin_concurrent_repack(Relation rel)
-{
- Oid toastrelid;
-
- /*
- * Avoid logical decoding of other relations by this backend. The lock we
- * have guarantees that the actual locator cannot be changed concurrently:
- * TRUNCATE needs AccessExclusiveLock.
- */
- Assert(CheckRelationLockedByMe(rel, ShareUpdateExclusiveLock, false));
- repacked_rel_locator = rel->rd_locator;
- toastrelid = rel->rd_rel->reltoastrelid;
- if (OidIsValid(toastrelid))
- {
- Relation toastrel;
-
- /* Avoid logical decoding of other TOAST relations. */
- toastrel = table_open(toastrelid, AccessShareLock);
- repacked_rel_toast_locator = toastrel->rd_locator;
- table_close(toastrel, AccessShareLock);
- }
-}
-
-/*
- * Call this when done with REPACK CONCURRENTLY.
- */
-static void
-end_concurrent_repack(void)
-{
- /*
- * Restore normal function of (future) logical decoding for this backend.
- */
- repacked_rel_locator.relNumber = InvalidOid;
- repacked_rel_toast_locator.relNumber = InvalidOid;
-}
-
/*
* This function is much like pg_create_logical_replication_slot() except that
* the new slot is neither released (if anyone else could read changes from
@@ -2427,9 +2489,10 @@ static LogicalDecodingContext *
setup_logical_decoding(Oid relid)
{
Relation rel;
- TupleDesc tupdesc;
+ Oid toastrelid;
LogicalDecodingContext *ctx;
- RepackDecodingState *dstate = palloc0_object(RepackDecodingState);
+ NameData slotname;
+ RepackDecodingState *dstate;
/*
* REPACK CONCURRENTLY is not allowed in a transaction block, so this
@@ -2437,21 +2500,21 @@ setup_logical_decoding(Oid relid)
*/
Assert(!TransactionIdIsValid(GetTopTransactionIdIfAny()));
- /*
- * A single backend should not execute multiple REPACK commands at a time,
- * so use PID to make the slot unique.
- */
- snprintf(NameStr(dstate->slotname), NAMEDATALEN, "repack_%d", MyProcPid);
-
/*
* Check if we can use logical decoding.
*/
CheckSlotPermissions();
CheckLogicalDecodingRequirements();
- /* RS_TEMPORARY so that the slot gets cleaned up on ERROR. */
- ReplicationSlotCreate(NameStr(dstate->slotname), true, RS_TEMPORARY,
- false, false, false);
+ /*
+ * A single backend should not execute multiple REPACK commands at a time,
+ * so use PID to make the slot unique.
+ *
+ * RS_TEMPORARY so that the slot gets cleaned up on ERROR.
+ */
+ snprintf(NameStr(slotname), NAMEDATALEN, "repack_%d", MyProcPid);
+ ReplicationSlotCreate(NameStr(slotname), true, RS_TEMPORARY, false, false,
+ false);
/*
* Neither prepare_write nor do_write callback nor update_progress is
@@ -2473,197 +2536,238 @@ setup_logical_decoding(Oid relid)
DecodingContextFindStartpoint(ctx);
+ /*
+ * decode_concurrent_changes() needs non-blocking callback.
+ */
+ ctx->reader->routine.page_read = read_local_xlog_page_no_wait;
+
+ /*
+ * read_local_xlog_page_no_wait() needs to be able to indicate the end of
+ * WAL.
+ */
+ ctx->reader->private_data = MemoryContextAllocZero(ctx->context,
+ sizeof(ReadLocalXLogPageNoWaitPrivate));
+
+
/* Some WAL records should have been read. */
Assert(ctx->reader->EndRecPtr != InvalidXLogRecPtr);
+ /*
+ * Initialize repack_current_segment so that we can notice WAL segment
+ * boundaries.
+ */
XLByteToSeg(ctx->reader->EndRecPtr, repack_current_segment,
wal_segment_size);
- /*
- * Setup structures to store decoded changes.
- */
+ dstate = palloc0_object(RepackDecodingState);
dstate->relid = relid;
- dstate->tstore = tuplestore_begin_heap(false, false,
- maintenance_work_mem);
- /* Caller should already have the table locked. */
- rel = table_open(relid, NoLock);
- tupdesc = CreateTupleDescCopy(RelationGetDescr(rel));
- dstate->tupdesc = tupdesc;
- table_close(rel, NoLock);
+ /*
+ * Tuple descriptor may be needed to flatten a tuple before we write it to
+ * a file. A copy is needed because the decoding worker invalidates system
+ * caches before it starts to do the actual work.
+ */
+ rel = table_open(relid, AccessShareLock);
+ dstate->tupdesc = CreateTupleDescCopy(RelationGetDescr(rel));
- /* Initialize the descriptor to store the changes ... */
- dstate->tupdesc_change = CreateTemplateTupleDesc(1);
+ /* Avoid logical decoding of other relations. */
+ repacked_rel_locator = rel->rd_locator;
+ toastrelid = rel->rd_rel->reltoastrelid;
+ if (OidIsValid(toastrelid))
+ {
+ Relation toastrel;
- TupleDescInitEntry(dstate->tupdesc_change, 1, NULL, BYTEAOID, -1, 0);
- /* ... as well as the corresponding slot. */
- dstate->tsslot = MakeSingleTupleTableSlot(dstate->tupdesc_change,
- &TTSOpsMinimalTuple);
+ /* Avoid logical decoding of other TOAST relations. */
+ toastrel = table_open(toastrelid, AccessShareLock);
+ repacked_rel_toast_locator = toastrel->rd_locator;
+ table_close(toastrel, AccessShareLock);
+ }
+ table_close(rel, AccessShareLock);
- dstate->resowner = ResourceOwnerCreate(CurrentResourceOwner,
- "logical decoding");
+ /* The file will be set as soon as we have it opened. */
+ dstate->file = NULL;
ctx->output_writer_private = dstate;
+
return ctx;
}
/*
- * Retrieve tuple from ConcurrentChange structure.
+ * Decode logical changes from the WAL sequence and store them to a file.
*
- * The input data starts with the structure but it might not be appropriately
- * aligned.
+ * If true is returned, there is no more work for the worker.
*/
-static HeapTuple
-get_changed_tuple(char *change)
+static bool
+decode_concurrent_changes(LogicalDecodingContext *ctx,
+ DecodingWorkerShared *shared)
{
- HeapTupleData tup_data;
- HeapTuple result;
- char *src;
+ RepackDecodingState *dstate;
+ XLogRecPtr lsn_upto;
+ bool done;
+ char fname[MAXPGPATH];
- /*
- * Ensure alignment before accessing the fields. (This is why we can't use
- * heap_copytuple() instead of this function.)
- */
- src = change + offsetof(ConcurrentChange, tup_data);
- memcpy(&tup_data, src, sizeof(HeapTupleData));
+ dstate = (RepackDecodingState *) ctx->output_writer_private;
- result = (HeapTuple) palloc(HEAPTUPLESIZE + tup_data.t_len);
- memcpy(result, &tup_data, sizeof(HeapTupleData));
- result->t_data = (HeapTupleHeader) ((char *) result + HEAPTUPLESIZE);
- src = change + SizeOfConcurrentChange;
- memcpy(result->t_data, src, result->t_len);
+ /* Open the output file. */
+ DecodingWorkerFileName(fname, shared->relid, shared->last_exported + 1);
+ dstate->file = BufFileCreateFileSet(&shared->sfs.fs, fname);
- return result;
-}
+ SpinLockAcquire(&shared->mutex);
+ lsn_upto = shared->lsn_upto;
+ done = shared->done;
+ SpinLockRelease(&shared->mutex);
-/*
- * Decode logical changes from the WAL sequence up to end_of_wal.
- */
-void
-repack_decode_concurrent_changes(LogicalDecodingContext *ctx,
- XLogRecPtr end_of_wal)
-{
- RepackDecodingState *dstate;
- ResourceOwner resowner_old;
-
- /*
- * Invalidate the "present" cache before moving to "(recent) history".
- */
- InvalidateSystemCaches();
+ while (XLogRecPtrIsInvalid(lsn_upto) || ctx->reader->EndRecPtr < lsn_upto)
+ {
+ XLogRecord *record;
+ XLogSegNo segno_new;
+ char *errm = NULL;
+ XLogRecPtr end_lsn;
- dstate = (RepackDecodingState *) ctx->output_writer_private;
- resowner_old = CurrentResourceOwner;
- CurrentResourceOwner = dstate->resowner;
+ CHECK_FOR_INTERRUPTS();
- PG_TRY();
- {
- while (ctx->reader->EndRecPtr < end_of_wal)
+ record = XLogReadRecord(ctx->reader, &errm);
+ if (record == NULL)
{
- XLogRecord *record;
- XLogSegNo segno_new;
- char *errm = NULL;
- XLogRecPtr end_lsn;
+ ReadLocalXLogPageNoWaitPrivate *priv;
- record = XLogReadRecord(ctx->reader, &errm);
if (errm)
- elog(ERROR, "%s", errm);
-
- if (record != NULL)
- LogicalDecodingProcessRecord(ctx, ctx->reader);
+ ereport(ERROR, (errmsg("%s", errm)));
/*
- * If WAL segment boundary has been crossed, inform the decoding
- * system that the catalog_xmin can advance. (We can confirm more
- * often, but a filling a single WAL segment should not take much
- * time.)
+ * In the decoding loop we do not want to get blocked when there
+ * is no more WAL available, otherwise the loop would become
+ * uninterruptible. The point is that the worker is only useful if
+ * it starts decoding before lsn_upto is set. Thus it can reach
+ * the end of WAL and find out later that it did not have to go
+ * that far.
*/
- end_lsn = ctx->reader->EndRecPtr;
- XLByteToSeg(end_lsn, segno_new, wal_segment_size);
- if (segno_new != repack_current_segment)
+ priv = (ReadLocalXLogPageNoWaitPrivate *)
+ ctx->reader->private_data;
+ if (priv->end_of_wal)
{
- LogicalConfirmReceivedLocation(end_lsn);
- elog(DEBUG1, "REPACK: confirmed receive location %X/%X",
- (uint32) (end_lsn >> 32), (uint32) end_lsn);
- repack_current_segment = segno_new;
+ priv->end_of_wal = false;
+
+ /* Do we know how far we should get? */
+ if (XLogRecPtrIsInvalid(lsn_upto))
+ {
+ SpinLockAcquire(&shared->mutex);
+ lsn_upto = shared->lsn_upto;
+ /* 'done' should be set at the same time as 'lsn_upto' */
+ done = shared->done;
+ SpinLockRelease(&shared->mutex);
+
+ /* Check if the work happens to be complete. */
+ continue;
+ }
+
+ /* Wait a bit before we retry reading WAL. */
+ (void) WaitLatch(MyLatch,
+ WL_LATCH_SET | WL_TIMEOUT | WL_EXIT_ON_PM_DEATH,
+ 1000L,
+ WAIT_EVENT_REPACK_WORKER_MAIN);
+
+ continue;
}
+ else
+ ereport(ERROR, (errmsg("could not read WAL record")));
+ }
- CHECK_FOR_INTERRUPTS();
+ LogicalDecodingProcessRecord(ctx, ctx->reader);
+
+ /*
+ * If WAL segment boundary has been crossed, inform the decoding
+ * system that the catalog_xmin can advance.
+ *
+ * TODO Does it make sense to confirm more often? Segment size seems
+ * appropriate for restart_lsn (because less than a segment cannot be
+ * recycled anyway), however more frequent checks might be beneficial
+ * for catalog_xmin.
+ */
+ end_lsn = ctx->reader->EndRecPtr;
+ XLByteToSeg(end_lsn, segno_new, wal_segment_size);
+ if (segno_new != repack_current_segment)
+ {
+ LogicalConfirmReceivedLocation(end_lsn);
+ elog(DEBUG1, "REPACK: confirmed receive location %X/%X",
+ (uint32) (end_lsn >> 32), (uint32) end_lsn);
+ repack_current_segment = segno_new;
+ }
+
+ /* Keep checking if 'lsn_upto' was specified. */
+ if (XLogRecPtrIsInvalid(lsn_upto))
+ {
+ SpinLockAcquire(&shared->mutex);
+ lsn_upto = shared->lsn_upto;
+ /* 'done' should be set at the same time as 'lsn_upto' */
+ done = shared->done;
+ SpinLockRelease(&shared->mutex);
}
- InvalidateSystemCaches();
- CurrentResourceOwner = resowner_old;
- }
- PG_CATCH();
- {
- /* clear all timetravel entries */
- InvalidateSystemCaches();
- CurrentResourceOwner = resowner_old;
- PG_RE_THROW();
}
- PG_END_TRY();
+
+ /*
+ * Close the file and make it available to the backend.
+ */
+ BufFileClose(dstate->file);
+ dstate->file = NULL;
+ SpinLockAcquire(&shared->mutex);
+ shared->lsn_upto = InvalidXLogRecPtr;
+ shared->sfs_valid = true;
+ shared->last_exported++;
+ SpinLockRelease(&shared->mutex);
+ ConditionVariableSignal(&shared->cv);
+
+ return done;
}
/*
- * Apply changes that happened during the initial load.
- *
- * Scan key is passed by caller, so it does not have to be constructed
- * multiple times. Key entries have all fields initialized, except for
- * sk_argument.
+ * Apply changes stored in 'file'.
*/
static void
-apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
+apply_concurrent_changes(BufFile *file, ChangeDest *dest)
{
+ char kind;
+ uint32 t_len;
Relation rel = dest->rel;
TupleTableSlot *index_slot,
*ident_slot;
HeapTuple tup_old = NULL;
- if (dstate->nchanges == 0)
- return;
-
/* TupleTableSlot is needed to pass the tuple to ExecInsertIndexTuples(). */
- index_slot = MakeSingleTupleTableSlot(dstate->tupdesc, &TTSOpsHeapTuple);
+ index_slot = MakeSingleTupleTableSlot(RelationGetDescr(rel),
+ &TTSOpsHeapTuple);
/* A slot to fetch tuples from identity index. */
ident_slot = table_slot_create(rel, NULL);
- while (tuplestore_gettupleslot(dstate->tstore, true, false,
- dstate->tsslot))
+ while (true)
{
- bool shouldFree;
- HeapTuple tup_change,
- tup,
+ size_t nread;
+ HeapTuple tup,
tup_exist;
- char *change_raw,
- *src;
- ConcurrentChange change;
- bool isnull[1];
- Datum values[1];
CHECK_FOR_INTERRUPTS();
- /* Get the change from the single-column tuple. */
- tup_change = ExecFetchSlotHeapTuple(dstate->tsslot, false, &shouldFree);
- heap_deform_tuple(tup_change, dstate->tupdesc_change, values, isnull);
- Assert(!isnull[0]);
-
- /* Make sure we access aligned data. */
- change_raw = (char *) DatumGetByteaP(values[0]);
- src = (char *) VARDATA(change_raw);
- memcpy(&change, src, SizeOfConcurrentChange);
+ nread = BufFileReadMaybeEOF(file, &kind, 1, true);
+ /* Are we done with the file? */
+ if (nread == 0)
+ break;
- /*
- * Extract the tuple from the change. The tuple is copied here because
- * it might be assigned to 'tup_old', in which case it needs to
- * survive into the next iteration.
- */
- tup = get_changed_tuple(src);
+ /* Read the tuple. */
+ BufFileReadExact(file, &t_len, sizeof(t_len));
+ tup = (HeapTuple) palloc(HEAPTUPLESIZE + t_len);
+ tup->t_data = (HeapTupleHeader) ((char *) tup + HEAPTUPLESIZE);
+ BufFileReadExact(file, tup->t_data, t_len);
+ tup->t_len = t_len;
+ ItemPointerSetInvalid(&tup->t_self);
+ tup->t_tableOid = RelationGetRelid(dest->rel);
- if (change.kind == CHANGE_UPDATE_OLD)
+ if (kind == CHANGE_UPDATE_OLD)
{
Assert(tup_old == NULL);
tup_old = tup;
}
- else if (change.kind == CHANGE_INSERT)
+ else if (kind == CHANGE_INSERT)
{
Assert(tup_old == NULL);
@@ -2671,12 +2775,11 @@ apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
pfree(tup);
}
- else if (change.kind == CHANGE_UPDATE_NEW ||
- change.kind == CHANGE_DELETE)
+ else if (kind == CHANGE_UPDATE_NEW || kind == CHANGE_DELETE)
{
HeapTuple tup_key;
- if (change.kind == CHANGE_UPDATE_NEW)
+ if (kind == CHANGE_UPDATE_NEW)
{
tup_key = tup_old != NULL ? tup_old : tup;
}
@@ -2693,7 +2796,7 @@ apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
if (tup_exist == NULL)
elog(ERROR, "failed to find target tuple");
- if (change.kind == CHANGE_UPDATE_NEW)
+ if (kind == CHANGE_UPDATE_NEW)
apply_concurrent_update(rel, tup, tup_exist, dest->iistate,
index_slot);
else
@@ -2708,26 +2811,19 @@ apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
pfree(tup);
}
else
- elog(ERROR, "unrecognized kind of change: %d", change.kind);
+ elog(ERROR, "unrecognized kind of change: %d", kind);
/*
* If a change was applied now, increment CID for next writes and
* update the snapshot so it sees the changes we've applied so far.
*/
- if (change.kind != CHANGE_UPDATE_OLD)
+ if (kind != CHANGE_UPDATE_OLD)
{
CommandCounterIncrement();
UpdateActiveSnapshotCommandId();
}
-
- /* TTSOpsMinimalTuple has .get_heap_tuple==NULL. */
- Assert(shouldFree);
- pfree(tup_change);
}
- tuplestore_clear(dstate->tstore);
- dstate->nchanges = 0;
-
/* Cleanup. */
ExecDropSingleTupleTableSlot(index_slot);
ExecDropSingleTupleTableSlot(ident_slot);
@@ -2865,6 +2961,12 @@ find_target_tuple(Relation rel, ChangeDest *dest, HeapTuple tup_key,
/* XXX no instrumentation for now */
scan = index_beginscan(rel, ident_index, GetActiveSnapshot(),
NULL, dest->ident_key_nentries, 0);
+
+ /*
+ * Scan key is passed by caller, so it does not have to be constructed
+ * multiple times. Key entries have all fields initialized, except for
+ * sk_argument.
+ */
index_rescan(scan, dest->ident_key, dest->ident_key_nentries, NULL, 0);
/* Info needed to retrieve key values from heap tuple. */
@@ -2900,25 +3002,58 @@ find_target_tuple(Relation rel, ChangeDest *dest, HeapTuple tup_key,
}
/*
- * Decode and apply concurrent changes.
+ * Decode and apply concurrent changes, up to (and including) the record whose
+ * LSN is 'end_of_wal'.
*/
static void
-process_concurrent_changes(LogicalDecodingContext *decoding_ctx,
- XLogRecPtr end_of_wal, ChangeDest *dest)
+process_concurrent_changes(XLogRecPtr end_of_wal, ChangeDest *dest, bool done)
{
- RepackDecodingState *dstate;
+ DecodingWorkerShared *shared;
+ char fname[MAXPGPATH];
+ BufFile *file;
pgstat_progress_update_param(PROGRESS_REPACK_PHASE,
PROGRESS_REPACK_PHASE_CATCH_UP);
- dstate = (RepackDecodingState *) decoding_ctx->output_writer_private;
+ /* Ask the worker for the file. */
+ shared = (DecodingWorkerShared *) dsm_segment_address(decoding_worker->seg);
+ SpinLockAcquire(&shared->mutex);
+ shared->lsn_upto = end_of_wal;
+ shared->done = done;
+ SpinLockRelease(&shared->mutex);
- repack_decode_concurrent_changes(decoding_ctx, end_of_wal);
+ /*
+ * The worker needs to finish processing of the current WAL record. Even
+ * if it's idle, it'll need to close the output file. Thus we're likely to
+ * wait, so prepare for sleep.
+ */
+ ConditionVariablePrepareToSleep(&shared->cv);
+ for (;;)
+ {
+ bool valid;
- if (dstate->nchanges == 0)
- return;
+ SpinLockAcquire(&shared->mutex);
+ valid = shared->sfs_valid;
+ SpinLockRelease(&shared->mutex);
+
+ if (valid)
+ break;
+
+ ConditionVariableSleep(&shared->cv, WAIT_EVENT_REPACK_WORKER_EXPORT);
+ }
+ ConditionVariableCancelSleep();
- apply_concurrent_changes(dstate, dest);
+ /* Open the file. */
+ DecodingWorkerFileName(fname, shared->relid, shared->last_exported);
+ file = BufFileOpenFileSet(&shared->sfs.fs, fname, O_RDONLY, false);
+ apply_concurrent_changes(file, dest);
+
+ /* No file is exported until the worker exports the next one. */
+ SpinLockAcquire(&shared->mutex);
+ shared->sfs_valid = false;
+ SpinLockRelease(&shared->mutex);
+
+ BufFileClose(file);
}
/*
@@ -3044,15 +3179,10 @@ cleanup_logical_decoding(LogicalDecodingContext *ctx)
dstate = (RepackDecodingState *) ctx->output_writer_private;
- ExecDropSingleTupleTableSlot(dstate->tsslot);
- FreeTupleDesc(dstate->tupdesc_change);
FreeTupleDesc(dstate->tupdesc);
- tuplestore_end(dstate->tstore);
-
FreeDecodingContext(ctx);
- ReplicationSlotRelease();
- ReplicationSlotDrop(NameStr(dstate->slotname), false);
+ ReplicationSlotDropAcquired();
pfree(dstate);
}
@@ -3067,7 +3197,6 @@ cleanup_logical_decoding(LogicalDecodingContext *ctx)
static void
rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
Relation cl_index,
- LogicalDecodingContext *decoding_ctx,
TransactionId frozenXid,
MultiXactId cutoffMulti)
{
@@ -3170,7 +3299,7 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
* hold AccessExclusiveLock. (Quite some amount of WAL could have been
* written during the data copying and index creation.)
*/
- process_concurrent_changes(decoding_ctx, end_of_wal, &chgdst);
+ process_concurrent_changes(end_of_wal, &chgdst, false);
/*
* Acquire AccessExclusiveLock on the table, its TOAST relation (if there
@@ -3266,8 +3395,11 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
XLogFlush(wal_insert_ptr);
end_of_wal = GetFlushRecPtr(NULL);
- /* Apply the concurrent changes again. */
- process_concurrent_changes(decoding_ctx, end_of_wal, &chgdst);
+ /*
+ * Apply the concurrent changes again. Indicate that the decoding worker
+ * won't be needed anymore.
+ */
+ process_concurrent_changes(end_of_wal, &chgdst, true);
/* Remember info about rel before closing OldHeap */
relpersistence = OldHeap->rd_rel->relpersistence;
@@ -3380,3 +3512,488 @@ build_new_indexes(Relation NewHeap, Relation OldHeap, List *OldIndexes)
return result;
}
+
+/*
+ * Try to start a background worker to perform logical decoding of data
+ * changes applied to relation while REPACK CONCURRENTLY is copying its
+ * contents to a new table.
+ */
+static void
+start_decoding_worker(Oid relid)
+{
+ Size size;
+ dsm_segment *seg;
+ DecodingWorkerShared *shared;
+ shm_mq *mq;
+ shm_mq_handle *mqh;
+ BackgroundWorker bgw;
+
+ /* Setup shared memory. */
+ size = BUFFERALIGN(offsetof(DecodingWorkerShared, error_queue)) +
+ BUFFERALIGN(REPACK_ERROR_QUEUE_SIZE);
+ seg = dsm_create(size, 0);
+ shared = (DecodingWorkerShared *) dsm_segment_address(seg);
+ shared->lsn_upto = InvalidXLogRecPtr;
+ shared->done = false;
+ SharedFileSetInit(&shared->sfs, seg);
+ shared->sfs_valid = false;
+ shared->last_exported = -1;
+ SpinLockInit(&shared->mutex);
+ shared->dbid = MyDatabaseId;
+
+ /*
+ * This is the UserId set in cluster_rel(). Security context shouldn't be
+ * needed for decoding worker.
+ */
+ shared->roleid = GetUserId();
+ shared->relid = relid;
+ ConditionVariableInit(&shared->cv);
+ shared->backend_proc = MyProc;
+ shared->backend_pid = MyProcPid;
+ shared->backend_proc_number = MyProcNumber;
+
+ mq = shm_mq_create((char *) BUFFERALIGN(shared->error_queue),
+ REPACK_ERROR_QUEUE_SIZE);
+ shm_mq_set_receiver(mq, MyProc);
+ mqh = shm_mq_attach(mq, seg, NULL);
+
+ memset(&bgw, 0, sizeof(bgw));
+ snprintf(bgw.bgw_name, BGW_MAXLEN,
+ "REPACK decoding worker for relation \"%s\"",
+ get_rel_name(relid));
+ snprintf(bgw.bgw_type, BGW_MAXLEN, "REPACK decoding worker");
+ bgw.bgw_flags = BGWORKER_SHMEM_ACCESS |
+ BGWORKER_BACKEND_DATABASE_CONNECTION;
+ bgw.bgw_start_time = BgWorkerStart_RecoveryFinished;
+ bgw.bgw_restart_time = BGW_NEVER_RESTART;
+ snprintf(bgw.bgw_library_name, MAXPGPATH, "postgres");
+ snprintf(bgw.bgw_function_name, BGW_MAXLEN, "RepackWorkerMain");
+ bgw.bgw_main_arg = UInt32GetDatum(dsm_segment_handle(seg));
+ bgw.bgw_notify_pid = MyProcPid;
+
+ decoding_worker = palloc0_object(DecodingWorker);
+ if (!RegisterDynamicBackgroundWorker(&bgw, &decoding_worker->handle))
+ ereport(ERROR,
+ (errcode(ERRCODE_CONFIGURATION_LIMIT_EXCEEDED),
+ errmsg("out of background worker slots"),
+ errhint("You might need to increase \"%s\".", "max_worker_processes")));
+
+ decoding_worker->seg = seg;
+ decoding_worker->error_mqh = mqh;
+}
+
+/*
+ * Stop the decoding worker and cleanup the related resources.
+ *
+ * The worker stops on its own when it knows there is no more work to do, but
+ * we need to stop it explicitly at least on ERROR in the launching backend.
+ */
+static void
+stop_decoding_worker(void)
+{
+ BgwHandleStatus status;
+
+ /* Haven't reached the worker startup? */
+ if (decoding_worker == NULL)
+ return;
+
+ /* Could not register the worker? */
+ if (decoding_worker->handle == NULL)
+ return;
+
+ TerminateBackgroundWorker(decoding_worker->handle);
+ /* The worker should really exit before the REPACK command does. */
+ HOLD_INTERRUPTS();
+ status = WaitForBackgroundWorkerShutdown(decoding_worker->handle);
+ RESUME_INTERRUPTS();
+
+ if (status == BGWH_POSTMASTER_DIED)
+ ereport(FATAL,
+ (errcode(ERRCODE_ADMIN_SHUTDOWN),
+ errmsg("postmaster exited during REPACK command")));
+
+ shm_mq_detach(decoding_worker->error_mqh);
+
+ /*
+ * If we could not cancel the current sleep due to ERROR, do that before
+ * we detach from the shared memory the condition variable is located in.
+ * If we did not, the bgworker ERROR handling code would try and fail
+ * badly.
+ */
+ ConditionVariableCancelSleep();
+
+ dsm_detach(decoding_worker->seg);
+ pfree(decoding_worker);
+ decoding_worker = NULL;
+}
+
+/* Is this process a REPACK worker? */
+static bool is_repack_worker = false;
+
+static pid_t backend_pid;
+static ProcNumber backend_proc_number;
+
+/*
+ * See ParallelWorkerShutdown for details.
+ */
+static void
+RepackWorkerShutdown(int code, Datum arg)
+{
+ SendProcSignal(backend_pid,
+ PROCSIG_REPACK_MESSAGE,
+ backend_proc_number);
+
+ dsm_detach((dsm_segment *) DatumGetPointer(arg));
+}
+
+/* REPACK decoding worker entry point */
+void
+RepackWorkerMain(Datum main_arg)
+{
+ dsm_segment *seg;
+ DecodingWorkerShared *shared;
+ shm_mq *mq;
+ shm_mq_handle *mqh;
+
+ is_repack_worker = true;
+
+ /*
+ * Override the default bgworker_die() with die() so we can use
+ * CHECK_FOR_INTERRUPTS().
+ */
+ pqsignal(SIGTERM, die);
+ BackgroundWorkerUnblockSignals();
+
+ seg = dsm_attach(DatumGetUInt32(main_arg));
+ if (seg == NULL)
+ ereport(ERROR,
+ (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
+ errmsg("could not map dynamic shared memory segment")));
+
+ shared = (DecodingWorkerShared *) dsm_segment_address(seg);
+
+ /* Arrange to signal the leader if we exit. */
+ backend_pid = shared->backend_pid;
+ backend_proc_number = shared->backend_proc_number;
+ before_shmem_exit(RepackWorkerShutdown, PointerGetDatum(seg));
+
+ /*
+ * Join locking group - see the comments around the call of
+ * start_decoding_worker().
+ */
+ if (!BecomeLockGroupMember(shared->backend_proc, backend_pid))
+ /* The leader is not running anymore. */
+ return;
+
+ /*
+ * Setup a queue to send error messages to the backend that launched this
+ * worker.
+ */
+ mq = (shm_mq *) (char *) BUFFERALIGN(shared->error_queue);
+ shm_mq_set_sender(mq, MyProc);
+ mqh = shm_mq_attach(mq, seg, NULL);
+ pq_redirect_to_shm_mq(seg, mqh);
+ pq_set_parallel_leader(shared->backend_pid,
+ shared->backend_proc_number);
+
+ /* Connect to the database. */
+ BackgroundWorkerInitializeConnectionByOid(shared->dbid, shared->roleid, 0);
+
+ repack_worker_internal(seg);
+}
+
+static void
+repack_worker_internal(dsm_segment *seg)
+{
+ DecodingWorkerShared *shared;
+ LogicalDecodingContext *decoding_ctx;
+ SharedFileSet *sfs;
+ Snapshot snapshot;
+
+ /*
+ * Transaction is needed to open relation, and it also provides us with a
+ * resource owner.
+ */
+ StartTransactionCommand();
+
+ shared = (DecodingWorkerShared *) dsm_segment_address(seg);
+
+ /*
+ * Not sure the spinlock is needed here - the backend should not change
+ * anything in the shared memory until we have serialized the snapshot.
+ */
+ SpinLockAcquire(&shared->mutex);
+ Assert(XLogRecPtrIsInvalid(shared->lsn_upto));
+ Assert(!shared->sfs_valid);
+ sfs = &shared->sfs;
+ SpinLockRelease(&shared->mutex);
+
+ SharedFileSetAttach(sfs, seg);
+
+ /*
+ * Prepare to capture the concurrent data changes ourselves.
+ */
+ decoding_ctx = setup_logical_decoding(shared->relid);
+
+ /* Build the initial snapshot and export it. */
+ snapshot = SnapBuildInitialSnapshotForRepack(decoding_ctx->snapshot_builder);
+ export_initial_snapshot(snapshot, shared);
+
+ /*
+ * The worker already had to access some system catalogs during startup,
+ * and we even had to open the relation we are processing. Now that we're
+ * going to work with historic snapshots, the system caches must be
+ * invalidated.
+ */
+ InvalidateSystemCaches();
+
+ while (!decode_concurrent_changes(decoding_ctx, shared))
+ ;
+
+ /* Cleanup. */
+ cleanup_logical_decoding(decoding_ctx);
+ CommitTransactionCommand();
+}
+
+/*
+ * Make snapshot available to the backend that launched the decoding worker.
+ */
+static void
+export_initial_snapshot(Snapshot snapshot, DecodingWorkerShared *shared)
+{
+ char fname[MAXPGPATH];
+ BufFile *file;
+ Size snap_size;
+ char *snap_space;
+
+ snap_size = EstimateSnapshotSpace(snapshot);
+ snap_space = (char *) palloc(snap_size);
+ SerializeSnapshot(snapshot, snap_space);
+ FreeSnapshot(snapshot);
+
+ DecodingWorkerFileName(fname, shared->relid, shared->last_exported + 1);
+ file = BufFileCreateFileSet(&shared->sfs.fs, fname);
+ /* To make restoration easier, write the snapshot size first. */
+ BufFileWrite(file, &snap_size, sizeof(snap_size));
+ BufFileWrite(file, snap_space, snap_size);
+ pfree(snap_space);
+ BufFileClose(file);
+
+ /* Tell the backend that the file is available. */
+ SpinLockAcquire(&shared->mutex);
+ shared->sfs_valid = true;
+ shared->last_exported++;
+ SpinLockRelease(&shared->mutex);
+ ConditionVariableSignal(&shared->cv);
+}
+
+/*
+ * Get the initial snapshot from the decoding worker.
+ */
+static Snapshot
+get_initial_snapshot(DecodingWorker *worker)
+{
+ DecodingWorkerShared *shared;
+ char fname[MAXPGPATH];
+ BufFile *file;
+ Size snap_size;
+ char *snap_space;
+ Snapshot snapshot;
+
+ shared = (DecodingWorkerShared *) dsm_segment_address(worker->seg);
+
+ /*
+ * The worker needs to initialize the logical decoding, which usually
+ * takes some time. Therefore it makes sense to prepare for the sleep
+ * first.
+ */
+ ConditionVariablePrepareToSleep(&shared->cv);
+ for (;;)
+ {
+ bool valid;
+
+ SpinLockAcquire(&shared->mutex);
+ valid = shared->sfs_valid;
+ SpinLockRelease(&shared->mutex);
+
+ if (valid)
+ break;
+
+ ConditionVariableSleep(&shared->cv, WAIT_EVENT_REPACK_WORKER_EXPORT);
+ }
+ ConditionVariableCancelSleep();
+
+ /* Read the snapshot from a file. */
+ DecodingWorkerFileName(fname, shared->relid, shared->last_exported);
+ file = BufFileOpenFileSet(&shared->sfs.fs, fname, O_RDONLY, false);
+ BufFileReadExact(file, &snap_size, sizeof(snap_size));
+ snap_space = (char *) palloc(snap_size);
+ BufFileReadExact(file, snap_space, snap_size);
+ BufFileClose(file);
+
+ SpinLockAcquire(&shared->mutex);
+ shared->sfs_valid = false;
+ SpinLockRelease(&shared->mutex);
+
+ /* Restore it. */
+ snapshot = RestoreSnapshot(snap_space);
+ pfree(snap_space);
+
+ return snapshot;
+}
+
+bool
+IsRepackWorker(void)
+{
+ return is_repack_worker;
+}
+
+/*
+ * Handle receipt of an interrupt indicating a repack worker message.
+ *
+ * Note: this is called within a signal handler! All we can do is set
+ * a flag that will cause the next CHECK_FOR_INTERRUPTS() to invoke
+ * ProcessRepackMessages().
+ */
+void
+HandleRepackMessageInterrupt(void)
+{
+ InterruptPending = true;
+ RepackMessagePending = true;
+ SetLatch(MyLatch);
+}
+
+/*
+ * Process any queued protocol messages received from parallel workers.
+ */
+void
+ProcessRepackMessages(void)
+{
+ MemoryContext oldcontext;
+
+ static MemoryContext hpm_context = NULL;
+
+ /*
+ * Nothing to do if we haven't launched the worker yet or have already
+ * terminated it.
+ */
+ if (decoding_worker == NULL)
+ return;
+
+ /*
+ * This is invoked from ProcessInterrupts(), and since some of the
+ * functions it calls contain CHECK_FOR_INTERRUPTS(), there is a potential
+ * for recursive calls if more signals are received while this runs. It's
+ * unclear that recursive entry would be safe, and it doesn't seem useful
+ * even if it is safe, so let's block interrupts until done.
+ */
+ HOLD_INTERRUPTS();
+
+ /*
+ * Moreover, CurrentMemoryContext might be pointing almost anywhere. We
+ * don't want to risk leaking data into long-lived contexts, so let's do
+ * our work here in a private context that we can reset on each use.
+ */
+ if (hpm_context == NULL) /* first time through? */
+ hpm_context = AllocSetContextCreate(TopMemoryContext,
+ "ProcessParallelMessages",
+ ALLOCSET_DEFAULT_SIZES);
+ else
+ MemoryContextReset(hpm_context);
+
+ oldcontext = MemoryContextSwitchTo(hpm_context);
+
+ /* OK to process messages. Reset the flag saying there are more to do. */
+ RepackMessagePending = false;
+
+ /*
+ * Read as many messages as we can from each worker, but stop when no more
+ * messages can be read from the worker without blocking.
+ */
+ while (true)
+ {
+ shm_mq_result res;
+ Size nbytes;
+ void *data;
+
+ res = shm_mq_receive(decoding_worker->error_mqh, &nbytes,
+ &data, true);
+ if (res == SHM_MQ_WOULD_BLOCK)
+ break;
+ else if (res == SHM_MQ_SUCCESS)
+ {
+ StringInfoData msg;
+
+ initStringInfo(&msg);
+ appendBinaryStringInfo(&msg, data, nbytes);
+ ProcessRepackMessage(&msg);
+ pfree(msg.data);
+ }
+ else
+ {
+ /*
+ * The decoding worker is special in that it exits as soon as it
+ * has its work done. Thus the DETACHED result code is fine.
+ */
+ Assert(res = SHM_MQ_DETACHED);
+
+ break;
+ }
+ }
+
+ MemoryContextSwitchTo(oldcontext);
+
+ /* Might as well clear the context on our way out */
+ MemoryContextReset(hpm_context);
+
+ RESUME_INTERRUPTS();
+}
+
+/*
+ * Process a single protocol message received from a single parallel worker.
+ */
+static void
+ProcessRepackMessage(StringInfo msg)
+{
+ char msgtype;
+
+ msgtype = pq_getmsgbyte(msg);
+
+ switch (msgtype)
+ {
+ case PqMsg_ErrorResponse:
+ case PqMsg_NoticeResponse:
+ {
+ ErrorData edata;
+
+ /* Parse ErrorResponse or NoticeResponse. */
+ pq_parse_errornotice(msg, &edata);
+
+ /* Death of a worker isn't enough justification for suicide. */
+ edata.elevel = Min(edata.elevel, ERROR);
+
+ /*
+ * If desired, add a context line to show that this is a
+ * message propagated from a parallel worker. Otherwise, it
+ * can sometimes be confusing to understand what actually
+ * happened.
+ */
+ if (edata.context)
+ edata.context = psprintf("%s\n%s", edata.context,
+ _("decoding worker"));
+ else
+ edata.context = pstrdup(_("decoding worker"));
+
+ /* Rethrow error or print notice. */
+ ThrowErrorData(&edata);
+
+ break;
+ }
+
+ default:
+ {
+ elog(ERROR, "unrecognized message type received from decoding worker: %c (message length %d bytes)",
+ msgtype, msg->len);
+ }
+ }
+}
diff --git a/src/backend/libpq/pqmq.c b/src/backend/libpq/pqmq.c
index 2b75de0ddef..28a5a400fb1 100644
--- a/src/backend/libpq/pqmq.c
+++ b/src/backend/libpq/pqmq.c
@@ -14,6 +14,7 @@
#include "postgres.h"
#include "access/parallel.h"
+#include "commands/cluster.h"
#include "libpq/libpq.h"
#include "libpq/pqformat.h"
#include "libpq/pqmq.h"
@@ -175,6 +176,10 @@ mq_putmessage(char msgtype, const char *s, size_t len)
SendProcSignal(pq_mq_parallel_leader_pid,
PROCSIG_PARALLEL_APPLY_MESSAGE,
pq_mq_parallel_leader_proc_number);
+ else if (IsRepackWorker())
+ SendProcSignal(pq_mq_parallel_leader_pid,
+ PROCSIG_REPACK_MESSAGE,
+ pq_mq_parallel_leader_proc_number);
else
{
Assert(IsParallelWorker());
diff --git a/src/backend/postmaster/bgworker.c b/src/backend/postmaster/bgworker.c
index 142a02eb5e9..b368990e90b 100644
--- a/src/backend/postmaster/bgworker.c
+++ b/src/backend/postmaster/bgworker.c
@@ -13,6 +13,7 @@
#include "postgres.h"
#include "access/parallel.h"
+#include "commands/cluster.h"
#include "libpq/pqsignal.h"
#include "miscadmin.h"
#include "pgstat.h"
@@ -135,6 +136,9 @@ static const struct
},
{
"SequenceSyncWorkerMain", SequenceSyncWorkerMain
+ },
+ {
+ "RepackWorkerMain", RepackWorkerMain
}
};
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index 866f92cf799..85323ba61e2 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -205,7 +205,11 @@ StartupDecodingContext(List *output_plugin_options,
ctx->slot = slot;
- ctx->reader = XLogReaderAllocate(wal_segment_size, NULL, xl_routine, ctx);
+ /*
+ * TODO A separate patch for PG core, unless there's really a reason to
+ * pass ctx for private_data (May extensions expect ctx?).
+ */
+ ctx->reader = XLogReaderAllocate(wal_segment_size, NULL, xl_routine, NULL);
if (!ctx->reader)
ereport(ERROR,
(errcode(ERRCODE_OUT_OF_MEMORY),
diff --git a/src/backend/replication/pgoutput_repack/pgoutput_repack.c b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
index c8930640a0d..fb9956d392d 100644
--- a/src/backend/replication/pgoutput_repack/pgoutput_repack.c
+++ b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
@@ -168,17 +168,13 @@ store_change(LogicalDecodingContext *ctx, ConcurrentChangeKind kind,
HeapTuple tuple)
{
RepackDecodingState *dstate;
- char *change_raw;
- ConcurrentChange change;
+ char kind_byte = (char) kind;
bool flattened = false;
- Size size;
- Datum values[1];
- bool isnull[1];
- char *dst;
dstate = (RepackDecodingState *) ctx->output_writer_private;
- size = VARHDRSZ + SizeOfConcurrentChange;
+ /* Store the change kind. */
+ BufFileWrite(dstate->file, &kind_byte, 1);
/*
* ReorderBufferCommit() stores the TOAST chunks in its private memory
@@ -195,46 +191,12 @@ store_change(LogicalDecodingContext *ctx, ConcurrentChangeKind kind,
tuple = toast_flatten_tuple(tuple, dstate->tupdesc);
flattened = true;
}
+ /* Store the tuple size ... */
+ BufFileWrite(dstate->file, &tuple->t_len, sizeof(tuple->t_len));
+ /* ... and the tuple itself. */
+ BufFileWrite(dstate->file, tuple->t_data, tuple->t_len);
- size += tuple->t_len;
- if (size >= MaxAllocSize)
- elog(ERROR, "Change is too big.");
-
- /* Construct the change. */
- change_raw = (char *) palloc0(size);
- SET_VARSIZE(change_raw, size);
-
- /*
- * Since the varlena alignment might not be sufficient for the structure,
- * set the fields in a local instance and remember where it should
- * eventually be copied.
- */
- change.kind = kind;
- dst = (char *) VARDATA(change_raw);
-
- /*
- * Copy the tuple.
- *
- * Note: change->tup_data.t_data must be fixed on retrieval!
- */
- memcpy(&change.tup_data, tuple, sizeof(HeapTupleData));
- memcpy(dst, &change, SizeOfConcurrentChange);
- dst += SizeOfConcurrentChange;
- memcpy(dst, tuple->t_data, tuple->t_len);
-
- /* The data has been copied. */
+ /* Free the flat copy if created above. */
if (flattened)
pfree(tuple);
-
- /* Store as tuple of 1 bytea column. */
- values[0] = PointerGetDatum(change_raw);
- isnull[0] = false;
- tuplestore_putvalues(dstate->tstore, dstate->tupdesc_change,
- values, isnull);
-
- /* Accounting. */
- dstate->nchanges++;
-
- /* Cleanup. */
- pfree(change_raw);
}
diff --git a/src/backend/storage/ipc/procsignal.c b/src/backend/storage/ipc/procsignal.c
index 087821311cc..af12144795b 100644
--- a/src/backend/storage/ipc/procsignal.c
+++ b/src/backend/storage/ipc/procsignal.c
@@ -19,6 +19,7 @@
#include "access/parallel.h"
#include "commands/async.h"
+#include "commands/cluster.h"
#include "miscadmin.h"
#include "pgstat.h"
#include "port/pg_bitutils.h"
@@ -694,6 +695,9 @@ procsignal_sigusr1_handler(SIGNAL_ARGS)
if (CheckProcSignal(PROCSIG_PARALLEL_APPLY_MESSAGE))
HandleParallelApplyMessageInterrupt();
+ if (CheckProcSignal(PROCSIG_REPACK_MESSAGE))
+ HandleRepackMessageInterrupt();
+
if (CheckProcSignal(PROCSIG_RECOVERY_CONFLICT_DATABASE))
HandleRecoveryConflictInterrupt(PROCSIG_RECOVERY_CONFLICT_DATABASE);
diff --git a/src/backend/tcop/postgres.c b/src/backend/tcop/postgres.c
index 7dd75a490aa..4a4858882f0 100644
--- a/src/backend/tcop/postgres.c
+++ b/src/backend/tcop/postgres.c
@@ -36,6 +36,7 @@
#include "access/xact.h"
#include "catalog/pg_type.h"
#include "commands/async.h"
+#include "commands/cluster.h"
#include "commands/event_trigger.h"
#include "commands/explain_state.h"
#include "commands/prepare.h"
@@ -3541,6 +3542,9 @@ ProcessInterrupts(void)
if (ParallelApplyMessagePending)
ProcessParallelApplyMessages();
+
+ if (RepackMessagePending)
+ ProcessRepackMessages();
}
/*
diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt
index f39830dbb34..cbcc8550960 100644
--- a/src/backend/utils/activity/wait_event_names.txt
+++ b/src/backend/utils/activity/wait_event_names.txt
@@ -62,6 +62,7 @@ LOGICAL_APPLY_MAIN "Waiting in main loop of logical replication apply process."
LOGICAL_LAUNCHER_MAIN "Waiting in main loop of logical replication launcher process."
LOGICAL_PARALLEL_APPLY_MAIN "Waiting in main loop of logical replication parallel apply process."
RECOVERY_WAL_STREAM "Waiting in main loop of startup process for WAL to arrive, during streaming recovery."
+REPACK_WORKER_MAIN "Waiting in main loop of REPACK decoding worker process."
REPLICATION_SLOTSYNC_MAIN "Waiting in main loop of slot sync worker."
REPLICATION_SLOTSYNC_SHUTDOWN "Waiting for slot sync worker to shut down."
SYSLOGGER_MAIN "Waiting in main loop of syslogger process."
@@ -153,6 +154,7 @@ RECOVERY_CONFLICT_SNAPSHOT "Waiting for recovery conflict resolution for a vacuu
RECOVERY_CONFLICT_TABLESPACE "Waiting for recovery conflict resolution for dropping a tablespace."
RECOVERY_END_COMMAND "Waiting for <xref linkend="guc-recovery-end-command"/> to complete."
RECOVERY_PAUSE "Waiting for recovery to be resumed."
+REPACK_WORKER_EXPORT "Waiting for decoding worker to export a new output file."
REPLICATION_ORIGIN_DROP "Waiting for a replication origin to become inactive so it can be dropped."
REPLICATION_SLOT_DROP "Waiting for a replication slot to become inactive so it can be dropped."
RESTORE_COMMAND "Waiting for <xref linkend="guc-restore-command"/> to complete."
diff --git a/src/include/access/tableam.h b/src/include/access/tableam.h
index d8f76d325f9..2b983abce3e 100644
--- a/src/include/access/tableam.h
+++ b/src/include/access/tableam.h
@@ -22,7 +22,6 @@
#include "access/xact.h"
#include "commands/vacuum.h"
#include "executor/tuptable.h"
-#include "replication/logical.h"
#include "storage/read_stream.h"
#include "utils/rel.h"
#include "utils/snapshot.h"
@@ -631,7 +630,6 @@ typedef struct TableAmRoutine
bool use_sort,
TransactionId OldestXmin,
Snapshot snapshot,
- LogicalDecodingContext *decoding_ctx,
TransactionId *xid_cutoff,
MultiXactId *multi_cutoff,
double *num_tuples,
@@ -1652,7 +1650,7 @@ table_relation_copy_data(Relation rel, const RelFileLocator *newrlocator)
* - snapshot - if != NULL, ignore data changes done by transactions that this
* (MVCC) snapshot considers still in-progress or in the future.
* - decoding_ctx - logical decoding context, to capture concurrent data
- * changes.
+ * changes. NULL if background worker takes care of the decoding.
*
* Output parameters:
* - *xid_cutoff - rel's new relfrozenxid value, may be invalid
@@ -1666,7 +1664,6 @@ table_relation_copy_for_cluster(Relation OldTable, Relation NewTable,
bool use_sort,
TransactionId OldestXmin,
Snapshot snapshot,
- LogicalDecodingContext *decoding_ctx,
TransactionId *xid_cutoff,
MultiXactId *multi_cutoff,
double *num_tuples,
@@ -1675,7 +1672,7 @@ table_relation_copy_for_cluster(Relation OldTable, Relation NewTable,
{
OldTable->rd_tableam->relation_copy_for_cluster(OldTable, NewTable, OldIndex,
use_sort, OldestXmin,
- snapshot, decoding_ctx,
+ snapshot,
xid_cutoff, multi_cutoff,
num_tuples, tups_vacuumed,
tups_recently_dead);
diff --git a/src/include/commands/cluster.h b/src/include/commands/cluster.h
index b43a1740053..0ac70ec30d7 100644
--- a/src/include/commands/cluster.h
+++ b/src/include/commands/cluster.h
@@ -17,6 +17,7 @@
#include "nodes/parsenodes.h"
#include "parser/parse_node.h"
#include "replication/logical.h"
+#include "storage/buffile.h"
#include "storage/lock.h"
#include "storage/relfilelocator.h"
#include "utils/relcache.h"
@@ -47,6 +48,9 @@ typedef struct ClusterParams
extern RelFileLocator repacked_rel_locator;
extern RelFileLocator repacked_rel_toast_locator;
+/*
+ * Stored as a single byte in the output file.
+ */
typedef enum
{
CHANGE_INSERT,
@@ -55,68 +59,30 @@ typedef enum
CHANGE_DELETE
} ConcurrentChangeKind;
-typedef struct ConcurrentChange
-{
- /* See the enum above. */
- ConcurrentChangeKind kind;
-
- /*
- * The actual tuple.
- *
- * The tuple data follows the ConcurrentChange structure. Before use make
- * sure the tuple is correctly aligned (ConcurrentChange can be stored as
- * bytea) and that tuple->t_data is fixed.
- */
- HeapTupleData tup_data;
-} ConcurrentChange;
-
-#define SizeOfConcurrentChange (offsetof(ConcurrentChange, tup_data) + \
- sizeof(HeapTupleData))
-
/*
* Logical decoding state.
*
- * Here we store the data changes that we decode from WAL while the table
- * contents is being copied to a new storage. Also the necessary metadata
- * needed to apply these changes to the table is stored here.
+ * The output plugin uses it to store the data changes that it decodes from
+ * WAL while the table contents is being copied to a new storage.
*/
typedef struct RepackDecodingState
{
/* The relation whose changes we're decoding. */
Oid relid;
- /* Replication slot name. */
- NameData slotname;
-
- /*
- * Decoded changes are stored here. Although we try to avoid excessive
- * batches, it can happen that the changes need to be stored to disk. The
- * tuplestore does this transparently.
- */
- Tuplestorestate *tstore;
-
- /* The current number of changes in tstore. */
- double nchanges;
-
- /*
- * Descriptor to store the ConcurrentChange structure serialized (bytea).
- * We can't store the tuple directly because tuplestore only supports
- * minimum tuple and we may need to transfer OID system column from the
- * output plugin. Also we need to transfer the change kind, so it's better
- * to put everything in the structure than to use 2 tuplestores "in
- * parallel".
- */
- TupleDesc tupdesc_change;
-
- /* Tuple descriptor needed to update indexes. */
+ /* Tuple descriptor of the relation being processed. */
TupleDesc tupdesc;
- /* Slot to retrieve data from tstore. */
- TupleTableSlot *tsslot;
-
- ResourceOwner resowner;
+ /* The current output file. */
+ BufFile *file;
} RepackDecodingState;
+extern PGDLLIMPORT volatile sig_atomic_t RepackMessagePending;
+
+extern bool IsRepackWorker(void);
+extern void HandleRepackMessageInterrupt(void);
+extern void ProcessRepackMessages(void);
+
extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel);
extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid,
@@ -125,9 +91,6 @@ extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid,
LOCKMODE lockmode);
extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal);
-extern void repack_decode_concurrent_changes(LogicalDecodingContext *ctx,
- XLogRecPtr end_of_wal);
-
extern Oid make_new_heap(Oid OIDOldHeap, Oid NewTableSpace, Oid NewAccessMethod,
char relpersistence, LOCKMODE lockmode);
extern void finish_heap_swap(Oid OIDOldHeap, Oid OIDNewHeap,
@@ -140,4 +103,5 @@ extern void finish_heap_swap(Oid OIDOldHeap, Oid OIDNewHeap,
MultiXactId cutoffMulti,
char newrelpersistence);
+extern void RepackWorkerMain(Datum main_arg);
#endif /* CLUSTER_H */
diff --git a/src/include/storage/procsignal.h b/src/include/storage/procsignal.h
index afeeb1ca019..c0a66516b66 100644
--- a/src/include/storage/procsignal.h
+++ b/src/include/storage/procsignal.h
@@ -36,6 +36,7 @@ typedef enum
PROCSIG_BARRIER, /* global barrier interrupt */
PROCSIG_LOG_MEMORY_CONTEXT, /* ask backend to log the memory contexts */
PROCSIG_PARALLEL_APPLY_MESSAGE, /* Message from parallel apply workers */
+ PROCSIG_REPACK_MESSAGE, /* Message from repack worker */
/* Recovery conflict reasons */
PROCSIG_RECOVERY_CONFLICT_FIRST,
diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list
index 3139b14e85f..35344910f65 100644
--- a/src/tools/pgindent/typedefs.list
+++ b/src/tools/pgindent/typedefs.list
@@ -488,7 +488,6 @@ CompressFileHandle
CompressionLocation
CompressorState
ComputeXidHorizonsResult
-ConcurrentChange
ConcurrentChangeKind
ConditionVariable
ConditionVariableMinimallyPadded
@@ -629,6 +628,9 @@ DeclareCursorStmt
DecodedBkpBlock
DecodedXLogRecord
DecodingOutputState
+DecodingWorker
+DecodingWorkerShared
+DecodingWorkerState
DefElem
DefElemAction
DefaultACLInfo
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 55+ messages in thread
* [PATCH 5/6] Use background worker to do logical decoding.
@ 2025-12-13 18:27 Antonin Houska <[email protected]>
0 siblings, 0 replies; 55+ messages in thread
From: Antonin Houska @ 2025-12-13 18:27 UTC (permalink / raw)
If the backend performing REPACK (CONCURRENTLY) does both data copying and
logical decoding, it has to "travel in time" back and forth and therefore it
has to invalidate system caches quite a few times. (The copying and the
decoding work with different catalog snapshots.) As the decoding worker has
separate caches, the switching is not necessary.
Without the worker, it'd also be difficult to switch between potentiallly long
running tasks like index build and WAL decoding. (No decoding during that time
at all can suspend archiving / recycling of WAL segments for some time, which
in turn may result in full disk.)
Another problem is that, after having acquired AccessExclusiveLock (in order
to swap the files), the backend needs to both decode and apply the data
changes that took place while it was waiting for the lock. With the decoding
worker, the decoding runs all the time, so the backend only needs to apply the
changes. This can reduce the time the exclusive lock is held for.
Note that the code added in order to handle ERRORs in the background worker
almost duplicates the existing code that does the same for other types of
workers (See ProcessParallelMessages() and
ProcessParallelApplyMessages()). Refactoring of the existing code might be
useful, to reduce the duplication.
---
src/backend/access/heap/heapam_handler.c | 44 -
src/backend/commands/cluster.c | 1144 +++++++++++++----
src/backend/libpq/pqmq.c | 5 +
src/backend/postmaster/bgworker.c | 4 +
src/backend/replication/logical/logical.c | 6 +-
.../pgoutput_repack/pgoutput_repack.c | 54 +-
src/backend/storage/ipc/procsignal.c | 4 +
src/backend/tcop/postgres.c | 4 +
.../utils/activity/wait_event_names.txt | 2 +
src/include/access/tableam.h | 7 +-
src/include/commands/cluster.h | 68 +-
src/include/storage/procsignal.h | 1 +
src/tools/pgindent/typedefs.list | 4 +-
13 files changed, 946 insertions(+), 401 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index 6063e037edb..cb09e6fd1dc 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -33,7 +33,6 @@
#include "catalog/index.h"
#include "catalog/storage.h"
#include "catalog/storage_xlog.h"
-#include "commands/cluster.h"
#include "commands/progress.h"
#include "executor/executor.h"
#include "miscadmin.h"
@@ -688,7 +687,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
Relation OldIndex, bool use_sort,
TransactionId OldestXmin,
Snapshot snapshot,
- LogicalDecodingContext *decoding_ctx,
TransactionId *xid_cutoff,
MultiXactId *multi_cutoff,
double *num_tuples,
@@ -710,7 +708,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
BufferHeapTupleTableSlot *hslot;
BlockNumber prev_cblock = InvalidBlockNumber;
bool concurrent = snapshot != NULL;
- XLogRecPtr end_of_wal_prev = GetFlushRecPtr(NULL);
/* Remember if it's a system catalog */
is_system_catalog = IsSystemRelation(OldHeap);
@@ -957,31 +954,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
ct_val[1] = *num_tuples;
pgstat_progress_update_multi_param(2, ct_index, ct_val);
}
-
- /*
- * Process the WAL produced by the load, as well as by other
- * transactions, so that the replication slot can advance and WAL does
- * not pile up. Use wal_segment_size as a threshold so that we do not
- * introduce the decoding overhead too often.
- *
- * Of course, we must not apply the changes until the initial load has
- * completed.
- *
- * Note that our insertions into the new table should not be decoded
- * as we (intentionally) do not write the logical decoding specific
- * information to WAL.
- */
- if (concurrent)
- {
- XLogRecPtr end_of_wal;
-
- end_of_wal = GetFlushRecPtr(NULL);
- if ((end_of_wal - end_of_wal_prev) > wal_segment_size)
- {
- repack_decode_concurrent_changes(decoding_ctx, end_of_wal);
- end_of_wal_prev = end_of_wal;
- }
- }
}
if (indexScan != NULL)
@@ -1027,22 +999,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
-
- /*
- * Try to keep the amount of not-yet-decoded WAL small, like
- * above.
- */
- if (concurrent)
- {
- XLogRecPtr end_of_wal;
-
- end_of_wal = GetFlushRecPtr(NULL);
- if ((end_of_wal - end_of_wal_prev) > wal_segment_size)
- {
- repack_decode_concurrent_changes(decoding_ctx, end_of_wal);
- end_of_wal_prev = end_of_wal;
- }
- }
}
tuplesort_end(tuplesort);
diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c
index 569705abc81..b0383c1375f 100644
--- a/src/backend/commands/cluster.c
+++ b/src/backend/commands/cluster.c
@@ -12,12 +12,13 @@
* In concurrent mode, we lock the table with only ShareUpdateExclusiveLock,
* then do an initial copy as above. However, while the tuples are being
* copied, concurrent transactions could modify the table. To cope with those
- * changes, we rely on logical decoding to obtain them from WAL. The changes
- * are accumulated in a tuplestore. Once the initial copy is complete, we
- * read the changes from the tuplestore and re-apply them on the new heap.
- * Then we upgrade our ShareUpdateExclusiveLock to AccessExclusiveLock and
- * swap the relfilenodes. This way, the time we hold a strong lock on the
- * table is much reduced, and the bloat is eliminated.
+ * changes, we rely on logical decoding to obtain them from WAL. A bgworker
+ * consumes WAL while the initial copy is ongoing (to prevent excessive WAL
+ * from being reserved), and accumulates the changes in a file. Once the
+ * initial copy is complete, we read the changes from the file and re-apply
+ * them on the new heap. Then we upgrade our ShareUpdateExclusiveLock to
+ * AccessExclusiveLock and swap the relfilenodes. This way, the time we hold
+ * a strong lock on the table is much reduced, and the bloat is eliminated.
*
* There is hardly anything left of Paul Brown's original implementation...
*
@@ -61,6 +62,8 @@
#include "commands/tablecmds.h"
#include "commands/vacuum.h"
#include "executor/executor.h"
+#include "libpq/pqformat.h"
+#include "libpq/pqmq.h"
#include "miscadmin.h"
#include "optimizer/optimizer.h"
#include "pgstat.h"
@@ -71,6 +74,8 @@
#include "storage/ipc.h"
#include "storage/lmgr.h"
#include "storage/predicate.h"
+#include "storage/procsignal.h"
+#include "tcop/tcopprot.h"
#include "utils/acl.h"
#include "utils/fmgroids.h"
#include "utils/guc.h"
@@ -138,6 +143,106 @@ typedef struct ChangeDest
IndexInsertState *iistate;
} ChangeDest;
+/*
+ * Layout of shared memory used for communication between backend and the
+ * worker that performs logical decoding of data changes
+ */
+typedef struct DecodingWorkerShared
+{
+ /* Is the decoding initialized? */
+ bool initialized;
+
+ /*
+ * Once the worker has reached this LSN, it should close the current
+ * output file and either create a new one or exit, according to the field
+ * 'done'. If the value is InvalidXLogRecPtr, the worker should decode all
+ * the WAL available and keep checking this field. It is ok if the worker
+ * had already decoded records whose LSN is >= lsn_upto before this field
+ * has been set.
+ */
+ XLogRecPtr lsn_upto;
+
+ /* Exit after closing the current file? */
+ bool done;
+
+ /* The output is stored here. */
+ SharedFileSet sfs;
+
+ /* Can backend read the file contents? */
+ bool sfs_valid;
+
+ /* Number of the last file exported by the worker. */
+ int last_exported;
+
+ /* Synchronize access to the fields above. */
+ slock_t mutex;
+
+ /* Database to connect to. */
+ Oid dbid;
+
+ /* Role to connect as. */
+ Oid roleid;
+
+ /* Decode data changes of this relation. */
+ Oid relid;
+
+ /* The backend uses this to wait for the worker. */
+ ConditionVariable cv;
+
+ /* Info to signal the backend. */
+ PGPROC *backend_proc;
+ pid_t backend_pid;
+ ProcNumber backend_proc_number;
+
+ /* Error queue. */
+ shm_mq *error_mq;
+
+ /*
+ * Memory the queue is located int.
+ *
+ * For considerations on the value see the comments of
+ * PARALLEL_ERROR_QUEUE_SIZE.
+ */
+#define REPACK_ERROR_QUEUE_SIZE 16384
+ char error_queue[FLEXIBLE_ARRAY_MEMBER];
+} DecodingWorkerShared;
+
+/*
+ * Generate worker's output file name. If relations of the same 'relid' happen
+ * to be processed at the same time, they must be from different databases and
+ * therefore different backends must be involved. (PID is already present in
+ * the fileset name.)
+ */
+static inline void
+DecodingWorkerFileName(char *fname, Oid relid, uint32 seq)
+{
+ snprintf(fname, MAXPGPATH, "%u-%u", relid, seq);
+}
+
+/*
+ * Backend-local information to control the decoding worker.
+ */
+typedef struct DecodingWorker
+{
+ /* The worker. */
+ BackgroundWorkerHandle *handle;
+
+ /* DecodingWorkerShared is in this segment. */
+ dsm_segment *seg;
+
+ /* Handle of the error queue. */
+ shm_mq_handle *error_mqh;
+} DecodingWorker;
+
+/* Pointer to currently running decoding worker. */
+static DecodingWorker *decoding_worker = NULL;
+
+/*
+ * Is there a message sent by a repack worker that the backend needs to
+ * receive?
+ */
+volatile sig_atomic_t RepackMessagePending = false;
+
static bool cluster_rel_recheck(RepackCommand cmd, Relation OldHeap,
Oid indexOid, Oid userid, LOCKMODE lmode,
int options);
@@ -145,7 +250,7 @@ static void check_repack_concurrently_requirements(Relation rel);
static void rebuild_relation(Relation OldHeap, Relation index, bool verbose,
bool concurrent);
static void copy_table_data(Relation NewHeap, Relation OldHeap, Relation OldIndex,
- Snapshot snapshot, LogicalDecodingContext *decoding_ctx,
+ Snapshot snapshot,
bool verbose,
bool *pSwapToastByContent,
TransactionId *pFreezeXid,
@@ -158,12 +263,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd,
static bool cluster_is_permitted_for_relation(RepackCommand cmd,
Oid relid, Oid userid);
-static void begin_concurrent_repack(Relation rel);
-static void end_concurrent_repack(void);
static LogicalDecodingContext *setup_logical_decoding(Oid relid);
-static HeapTuple get_changed_tuple(char *change);
-static void apply_concurrent_changes(RepackDecodingState *dstate,
- ChangeDest *dest);
+static bool decode_concurrent_changes(LogicalDecodingContext *ctx,
+ DecodingWorkerShared *shared);
+static void apply_concurrent_changes(BufFile *file, ChangeDest *dest);
static void apply_concurrent_insert(Relation rel, HeapTuple tup,
IndexInsertState *iistate,
TupleTableSlot *index_slot);
@@ -175,9 +278,9 @@ static void apply_concurrent_delete(Relation rel, HeapTuple tup_target);
static HeapTuple find_target_tuple(Relation rel, ChangeDest *dest,
HeapTuple tup_key,
TupleTableSlot *ident_slot);
-static void process_concurrent_changes(LogicalDecodingContext *decoding_ctx,
- XLogRecPtr end_of_wal,
- ChangeDest *dest);
+static void process_concurrent_changes(XLogRecPtr end_of_wal,
+ ChangeDest *dest,
+ bool done);
static IndexInsertState *get_index_insert_state(Relation relation,
Oid ident_index_id,
Relation *ident_index_p);
@@ -187,7 +290,6 @@ static void free_index_insert_state(IndexInsertState *iistate);
static void cleanup_logical_decoding(LogicalDecodingContext *ctx);
static void rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
Relation cl_index,
- LogicalDecodingContext *decoding_ctx,
TransactionId frozenXid,
MultiXactId cutoffMulti);
static List *build_new_indexes(Relation NewHeap, Relation OldHeap, List *OldIndexes);
@@ -197,6 +299,13 @@ static Relation process_single_relation(RepackStmt *stmt,
ClusterParams *params);
static Oid determine_clustered_index(Relation rel, bool usingindex,
const char *indexname);
+static void start_decoding_worker(Oid relid);
+static void stop_decoding_worker(void);
+static void repack_worker_internal(dsm_segment *seg);
+static void export_initial_snapshot(Snapshot snapshot,
+ DecodingWorkerShared *shared);
+static Snapshot get_initial_snapshot(DecodingWorker *worker);
+static void ProcessRepackMessage(StringInfo msg);
static const char *RepackCommandAsString(RepackCommand cmd);
@@ -619,20 +728,20 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid,
/* rebuild_relation does all the dirty work */
PG_TRY();
{
- /*
- * For concurrent processing, make sure that our logical decoding
- * ignores data changes of other tables than the one we are
- * processing.
- */
- if (concurrent)
- begin_concurrent_repack(OldHeap);
-
rebuild_relation(OldHeap, index, verbose, concurrent);
}
PG_FINALLY();
{
if (concurrent)
- end_concurrent_repack();
+ {
+ /*
+ * Since during normal operation the worker was already asked to
+ * exit, stopping it explicitly is especially important on ERROR.
+ * However it still seems a good practice to make sure that the
+ * worker never survives the REPACK command.
+ */
+ stop_decoding_worker();
+ }
}
PG_END_TRY();
@@ -929,7 +1038,6 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
bool swap_toast_by_content;
TransactionId frozenXid;
MultiXactId cutoffMulti;
- LogicalDecodingContext *decoding_ctx = NULL;
Snapshot snapshot = NULL;
#if USE_ASSERT_CHECKING
LOCKMODE lmode;
@@ -943,19 +1051,36 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
if (concurrent)
{
/*
- * Prepare to capture the concurrent data changes.
+ * The worker needs to be member of the locking group we're the leader
+ * of. We ought to become the leader before the worker starts. The
+ * worker will join the group as soon as it starts.
+ *
+ * This is to make sure that the deadlock described below is
+ * detectable by deadlock.c: if the worker waits for a transaction to
+ * complete and we are waiting for the worker output, then effectively
+ * we (i.e. this backend) are waiting for that transaction.
+ */
+ BecomeLockGroupLeader();
+
+ /*
+ * Start the worker that decodes data changes applied while we're
+ * copying the table contents.
*
- * Note that this call waits for all transactions with XID already
- * assigned to finish. If some of those transactions is waiting for a
- * lock conflicting with ShareUpdateExclusiveLock on our table (e.g.
- * it runs CREATE INDEX), we can end up in a deadlock. Not sure this
- * risk is worth unlocking/locking the table (and its clustering
- * index) and checking again if its still eligible for REPACK
- * CONCURRENTLY.
+ * Note that the worker has to wait for all transactions with XID
+ * already assigned to finish. If some of those transactions is
+ * waiting for a lock conflicting with ShareUpdateExclusiveLock on our
+ * table (e.g. it runs CREATE INDEX), we can end up in a deadlock.
+ * Not sure this risk is worth unlocking/locking the table (and its
+ * clustering index) and checking again if its still eligible for
+ * REPACK CONCURRENTLY.
+ */
+ start_decoding_worker(tableOid);
+
+ /*
+ * Wait until the worker has the initial snapshot and retrieve it.
*/
- decoding_ctx = setup_logical_decoding(tableOid);
+ snapshot = get_initial_snapshot(decoding_worker);
- snapshot = SnapBuildInitialSnapshotForRepack(decoding_ctx->snapshot_builder);
PushActiveSnapshot(snapshot);
}
@@ -980,7 +1105,7 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
NewHeap = table_open(OIDNewHeap, NoLock);
/* Copy the heap data into the new table in the desired order */
- copy_table_data(NewHeap, OldHeap, index, snapshot, decoding_ctx, verbose,
+ copy_table_data(NewHeap, OldHeap, index, snapshot, verbose,
&swap_toast_by_content, &frozenXid, &cutoffMulti);
/* The historic snapshot won't be needed anymore. */
@@ -994,14 +1119,10 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
{
Assert(!swap_toast_by_content);
rebuild_relation_finish_concurrent(NewHeap, OldHeap, index,
- decoding_ctx,
frozenXid, cutoffMulti);
pgstat_progress_update_param(PROGRESS_REPACK_PHASE,
PROGRESS_REPACK_PHASE_FINAL_CLEANUP);
-
- /* Done with decoding. */
- cleanup_logical_decoding(decoding_ctx);
}
else
{
@@ -1172,8 +1293,7 @@ make_new_heap(Oid OIDOldHeap, Oid NewTableSpace, Oid NewAccessMethod,
*/
static void
copy_table_data(Relation NewHeap, Relation OldHeap, Relation OldIndex,
- Snapshot snapshot, LogicalDecodingContext *decoding_ctx,
- bool verbose, bool *pSwapToastByContent,
+ Snapshot snapshot, bool verbose, bool *pSwapToastByContent,
TransactionId *pFreezeXid, MultiXactId *pCutoffMulti)
{
Relation relRelation;
@@ -1334,7 +1454,6 @@ copy_table_data(Relation NewHeap, Relation OldHeap, Relation OldIndex,
*/
table_relation_copy_for_cluster(OldHeap, NewHeap, OldIndex, use_sort,
cutoffs.OldestXmin, snapshot,
- decoding_ctx,
&cutoffs.FreezeLimit,
&cutoffs.MultiXactCutoff,
&num_tuples, &tups_vacuumed,
@@ -2367,59 +2486,6 @@ RepackCommandAsString(RepackCommand cmd)
return "???";
}
-
-/*
- * Call this function before REPACK CONCURRENTLY starts to setup logical
- * decoding. It makes sure that other users of the table put enough
- * information into WAL.
- *
- * The point is that at various places we expect that the table we're
- * processing is treated like a system catalog. For example, we need to be
- * able to scan it using a "historic snapshot" anytime during the processing
- * (as opposed to scanning only at the start point of the decoding, as logical
- * replication does during initial table synchronization), in order to apply
- * concurrent UPDATE / DELETE commands.
- *
- * Note that TOAST table needs no attention here as it's not scanned using
- * historic snapshot.
- */
-static void
-begin_concurrent_repack(Relation rel)
-{
- Oid toastrelid;
-
- /*
- * Avoid logical decoding of other relations by this backend. The lock we
- * have guarantees that the actual locator cannot be changed concurrently:
- * TRUNCATE needs AccessExclusiveLock.
- */
- Assert(CheckRelationLockedByMe(rel, ShareUpdateExclusiveLock, false));
- repacked_rel_locator = rel->rd_locator;
- toastrelid = rel->rd_rel->reltoastrelid;
- if (OidIsValid(toastrelid))
- {
- Relation toastrel;
-
- /* Avoid logical decoding of other TOAST relations. */
- toastrel = table_open(toastrelid, AccessShareLock);
- repacked_rel_toast_locator = toastrel->rd_locator;
- table_close(toastrel, AccessShareLock);
- }
-}
-
-/*
- * Call this when done with REPACK CONCURRENTLY.
- */
-static void
-end_concurrent_repack(void)
-{
- /*
- * Restore normal function of (future) logical decoding for this backend.
- */
- repacked_rel_locator.relNumber = InvalidOid;
- repacked_rel_toast_locator.relNumber = InvalidOid;
-}
-
/*
* This function is much like pg_create_logical_replication_slot() except that
* the new slot is neither released (if anyone else could read changes from
@@ -2431,9 +2497,10 @@ static LogicalDecodingContext *
setup_logical_decoding(Oid relid)
{
Relation rel;
- TupleDesc tupdesc;
+ Oid toastrelid;
LogicalDecodingContext *ctx;
- RepackDecodingState *dstate = palloc0_object(RepackDecodingState);
+ NameData slotname;
+ RepackDecodingState *dstate;
/*
* REPACK CONCURRENTLY is not allowed in a transaction block, so this
@@ -2441,21 +2508,21 @@ setup_logical_decoding(Oid relid)
*/
Assert(!TransactionIdIsValid(GetTopTransactionIdIfAny()));
- /*
- * A single backend should not execute multiple REPACK commands at a time,
- * so use PID to make the slot unique.
- */
- snprintf(NameStr(dstate->slotname), NAMEDATALEN, "repack_%d", MyProcPid);
-
/*
* Check if we can use logical decoding.
*/
CheckSlotPermissions();
CheckLogicalDecodingRequirements();
- /* RS_TEMPORARY so that the slot gets cleaned up on ERROR. */
- ReplicationSlotCreate(NameStr(dstate->slotname), true, RS_TEMPORARY,
- false, false, false);
+ /*
+ * A single backend should not execute multiple REPACK commands at a time,
+ * so use PID to make the slot unique.
+ *
+ * RS_TEMPORARY so that the slot gets cleaned up on ERROR.
+ */
+ snprintf(NameStr(slotname), NAMEDATALEN, "repack_%d", MyProcPid);
+ ReplicationSlotCreate(NameStr(slotname), true, RS_TEMPORARY, false, false,
+ false);
/*
* Neither prepare_write nor do_write callback nor update_progress is
@@ -2477,104 +2544,109 @@ setup_logical_decoding(Oid relid)
DecodingContextFindStartpoint(ctx);
+ /*
+ * decode_concurrent_changes() needs non-blocking callback.
+ */
+ ctx->reader->routine.page_read = read_local_xlog_page_no_wait;
+
+ /*
+ * read_local_xlog_page_no_wait() needs to be able to indicate the end of
+ * WAL.
+ */
+ ctx->reader->private_data = MemoryContextAllocZero(ctx->context,
+ sizeof(ReadLocalXLogPageNoWaitPrivate));
+
+
/* Some WAL records should have been read. */
Assert(ctx->reader->EndRecPtr != InvalidXLogRecPtr);
+ /*
+ * Initialize repack_current_segment so that we can notice WAL segment
+ * boundaries.
+ */
XLByteToSeg(ctx->reader->EndRecPtr, repack_current_segment,
wal_segment_size);
- /*
- * Setup structures to store decoded changes.
- */
+ dstate = palloc0_object(RepackDecodingState);
dstate->relid = relid;
- dstate->tstore = tuplestore_begin_heap(false, false,
- maintenance_work_mem);
- /* Caller should already have the table locked. */
- rel = table_open(relid, NoLock);
- tupdesc = CreateTupleDescCopy(RelationGetDescr(rel));
- dstate->tupdesc = tupdesc;
- table_close(rel, NoLock);
+ /*
+ * Tuple descriptor may be needed to flatten a tuple before we write it to
+ * a file. A copy is needed because the decoding worker invalidates system
+ * caches before it starts to do the actual work.
+ */
+ rel = table_open(relid, AccessShareLock);
+ dstate->tupdesc = CreateTupleDescCopy(RelationGetDescr(rel));
- /* Initialize the descriptor to store the changes ... */
- dstate->tupdesc_change = CreateTemplateTupleDesc(1);
+ /* Avoid logical decoding of other relations. */
+ repacked_rel_locator = rel->rd_locator;
+ toastrelid = rel->rd_rel->reltoastrelid;
+ if (OidIsValid(toastrelid))
+ {
+ Relation toastrel;
- TupleDescInitEntry(dstate->tupdesc_change, 1, NULL, BYTEAOID, -1, 0);
- /* ... as well as the corresponding slot. */
- dstate->tsslot = MakeSingleTupleTableSlot(dstate->tupdesc_change,
- &TTSOpsMinimalTuple);
+ /* Avoid logical decoding of other TOAST relations. */
+ toastrel = table_open(toastrelid, AccessShareLock);
+ repacked_rel_toast_locator = toastrel->rd_locator;
+ table_close(toastrel, AccessShareLock);
+ }
+ table_close(rel, AccessShareLock);
- dstate->resowner = ResourceOwnerCreate(CurrentResourceOwner,
- "logical decoding");
+ /* The file will be set as soon as we have it opened. */
+ dstate->file = NULL;
ctx->output_writer_private = dstate;
+
return ctx;
}
/*
- * Retrieve tuple from ConcurrentChange structure.
+ * Decode logical changes from the WAL sequence and store them to a file.
*
- * The input data starts with the structure but it might not be appropriately
- * aligned.
+ * If true is returned, there is no more work for the worker.
*/
-static HeapTuple
-get_changed_tuple(char *change)
-{
- HeapTupleData tup_data;
- HeapTuple result;
- char *src;
-
- /*
- * Ensure alignment before accessing the fields. (This is why we can't use
- * heap_copytuple() instead of this function.)
- */
- src = change + offsetof(ConcurrentChange, tup_data);
- memcpy(&tup_data, src, sizeof(HeapTupleData));
-
- result = (HeapTuple) palloc(HEAPTUPLESIZE + tup_data.t_len);
- memcpy(result, &tup_data, sizeof(HeapTupleData));
- result->t_data = (HeapTupleHeader) ((char *) result + HEAPTUPLESIZE);
- src = change + SizeOfConcurrentChange;
- memcpy(result->t_data, src, result->t_len);
-
- return result;
-}
-
-/*
- * Decode logical changes from the WAL sequence up to end_of_wal.
- */
-void
-repack_decode_concurrent_changes(LogicalDecodingContext *ctx,
- XLogRecPtr end_of_wal)
+static bool
+decode_concurrent_changes(LogicalDecodingContext *ctx,
+ DecodingWorkerShared *shared)
{
RepackDecodingState *dstate;
- ResourceOwner resowner_old;
+ XLogRecPtr lsn_upto;
+ bool done;
+ char fname[MAXPGPATH];
dstate = (RepackDecodingState *) ctx->output_writer_private;
- resowner_old = CurrentResourceOwner;
- CurrentResourceOwner = dstate->resowner;
- PG_TRY();
+ /* Open the output file. */
+ DecodingWorkerFileName(fname, shared->relid, shared->last_exported + 1);
+ dstate->file = BufFileCreateFileSet(&shared->sfs.fs, fname);
+
+ SpinLockAcquire(&shared->mutex);
+ lsn_upto = shared->lsn_upto;
+ done = shared->done;
+ SpinLockRelease(&shared->mutex);
+
+ while (true)
{
- while (ctx->reader->EndRecPtr < end_of_wal)
- {
- XLogRecord *record;
- XLogSegNo segno_new;
- char *errm = NULL;
- XLogRecPtr end_lsn;
+ XLogRecord *record;
+ XLogSegNo segno_new;
+ char *errm = NULL;
+ XLogRecPtr end_lsn;
- record = XLogReadRecord(ctx->reader, &errm);
- if (errm)
- elog(ERROR, "%s", errm);
+ CHECK_FOR_INTERRUPTS();
- if (record != NULL)
- LogicalDecodingProcessRecord(ctx, ctx->reader);
+ record = XLogReadRecord(ctx->reader, &errm);
+ if (record)
+ {
+ LogicalDecodingProcessRecord(ctx, ctx->reader);
/*
* If WAL segment boundary has been crossed, inform the decoding
- * system that the catalog_xmin can advance. (We can confirm more
- * often, but a filling a single WAL segment should not take much
- * time.)
+ * system that the catalog_xmin can advance.
+ *
+ * TODO Does it make sense to confirm more often? Segment size
+ * seems appropriate for restart_lsn (because less than a segment
+ * cannot be recycled anyway), however more frequent checks might
+ * be beneficial for catalog_xmin.
*/
end_lsn = ctx->reader->EndRecPtr;
XLByteToSeg(end_lsn, segno_new, wal_segment_size);
@@ -2585,80 +2657,117 @@ repack_decode_concurrent_changes(LogicalDecodingContext *ctx,
(uint32) (end_lsn >> 32), (uint32) end_lsn);
repack_current_segment = segno_new;
}
+ }
+ else
+ {
+ ReadLocalXLogPageNoWaitPrivate *priv;
+
+ if (errm)
+ ereport(ERROR, (errmsg("%s", errm)));
- CHECK_FOR_INTERRUPTS();
+ /*
+ * In the decoding loop we do not want to get blocked when there
+ * is no more WAL available, otherwise the loop would become
+ * uninterruptible. The point is that the worker is only useful if
+ * it starts decoding before lsn_upto is set. Thus it can reach
+ * the end of WAL and find out later that it did not have to go
+ * that far.
+ */
+ priv = (ReadLocalXLogPageNoWaitPrivate *)
+ ctx->reader->private_data;
+ if (priv->end_of_wal)
+ priv->end_of_wal = false;
+ else
+ ereport(ERROR, (errmsg("could not read WAL record")));
}
- InvalidateSystemCaches();
- CurrentResourceOwner = resowner_old;
- }
- PG_CATCH();
- {
- /* clear all timetravel entries */
- InvalidateSystemCaches();
- CurrentResourceOwner = resowner_old;
- PG_RE_THROW();
+
+ /*
+ * Whether we could read new record or not, keep checking if
+ * 'lsn_upto' was specified.
+ */
+ if (XLogRecPtrIsInvalid(lsn_upto))
+ {
+ SpinLockAcquire(&shared->mutex);
+ lsn_upto = shared->lsn_upto;
+ /* 'done' should be set at the same time as 'lsn_upto' */
+ done = shared->done;
+ SpinLockRelease(&shared->mutex);
+ }
+ if (!XLogRecPtrIsInvalid(lsn_upto) &&
+ ctx->reader->EndRecPtr >= lsn_upto)
+ break;
+
+ if (record == NULL)
+ /* Wait a bit before we retry reading WAL. */
+ (void) WaitLatch(MyLatch,
+ WL_LATCH_SET | WL_TIMEOUT | WL_EXIT_ON_PM_DEATH,
+ 100000L, /* XXX Tune the delay. */
+ WAIT_EVENT_REPACK_WORKER_MAIN);
}
- PG_END_TRY();
+
+ /*
+ * Close the file so we can make it available to the backend.
+ */
+ BufFileClose(dstate->file);
+ dstate->file = NULL;
+ SpinLockAcquire(&shared->mutex);
+ shared->lsn_upto = InvalidXLogRecPtr;
+ shared->sfs_valid = true;
+ shared->last_exported++;
+ SpinLockRelease(&shared->mutex);
+ ConditionVariableSignal(&shared->cv);
+
+ return done;
}
/*
* Apply changes stored in 'file'.
*/
static void
-apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
+apply_concurrent_changes(BufFile *file, ChangeDest *dest)
{
+ char kind;
+ uint32 t_len;
Relation rel = dest->rel;
TupleTableSlot *index_slot,
*ident_slot;
HeapTuple tup_old = NULL;
- if (dstate->nchanges == 0)
- return;
-
/* TupleTableSlot is needed to pass the tuple to ExecInsertIndexTuples(). */
- index_slot = MakeSingleTupleTableSlot(dstate->tupdesc, &TTSOpsHeapTuple);
+ index_slot = MakeSingleTupleTableSlot(RelationGetDescr(rel),
+ &TTSOpsHeapTuple);
/* A slot to fetch tuples from identity index. */
ident_slot = table_slot_create(rel, NULL);
- while (tuplestore_gettupleslot(dstate->tstore, true, false,
- dstate->tsslot))
+ while (true)
{
- bool shouldFree;
- HeapTuple tup_change,
- tup,
+ size_t nread;
+ HeapTuple tup,
tup_exist;
- char *change_raw,
- *src;
- ConcurrentChange change;
- bool isnull[1];
- Datum values[1];
CHECK_FOR_INTERRUPTS();
- /* Get the change from the single-column tuple. */
- tup_change = ExecFetchSlotHeapTuple(dstate->tsslot, false, &shouldFree);
- heap_deform_tuple(tup_change, dstate->tupdesc_change, values, isnull);
- Assert(!isnull[0]);
-
- /* Make sure we access aligned data. */
- change_raw = (char *) DatumGetByteaP(values[0]);
- src = (char *) VARDATA(change_raw);
- memcpy(&change, src, SizeOfConcurrentChange);
+ nread = BufFileReadMaybeEOF(file, &kind, 1, true);
+ /* Are we done with the file? */
+ if (nread == 0)
+ break;
- /*
- * Extract the tuple from the change. The tuple is copied here because
- * it might be assigned to 'tup_old', in which case it needs to
- * survive into the next iteration.
- */
- tup = get_changed_tuple(src);
+ /* Read the tuple. */
+ BufFileReadExact(file, &t_len, sizeof(t_len));
+ tup = (HeapTuple) palloc(HEAPTUPLESIZE + t_len);
+ tup->t_data = (HeapTupleHeader) ((char *) tup + HEAPTUPLESIZE);
+ BufFileReadExact(file, tup->t_data, t_len);
+ tup->t_len = t_len;
+ ItemPointerSetInvalid(&tup->t_self);
+ tup->t_tableOid = RelationGetRelid(dest->rel);
- if (change.kind == CHANGE_UPDATE_OLD)
+ if (kind == CHANGE_UPDATE_OLD)
{
Assert(tup_old == NULL);
tup_old = tup;
}
- else if (change.kind == CHANGE_INSERT)
+ else if (kind == CHANGE_INSERT)
{
Assert(tup_old == NULL);
@@ -2666,12 +2775,11 @@ apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
pfree(tup);
}
- else if (change.kind == CHANGE_UPDATE_NEW ||
- change.kind == CHANGE_DELETE)
+ else if (kind == CHANGE_UPDATE_NEW || kind == CHANGE_DELETE)
{
HeapTuple tup_key;
- if (change.kind == CHANGE_UPDATE_NEW)
+ if (kind == CHANGE_UPDATE_NEW)
{
tup_key = tup_old != NULL ? tup_old : tup;
}
@@ -2688,7 +2796,7 @@ apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
if (tup_exist == NULL)
elog(ERROR, "failed to find target tuple");
- if (change.kind == CHANGE_UPDATE_NEW)
+ if (kind == CHANGE_UPDATE_NEW)
apply_concurrent_update(rel, tup, tup_exist, dest->iistate,
index_slot);
else
@@ -2703,26 +2811,19 @@ apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
pfree(tup);
}
else
- elog(ERROR, "unrecognized kind of change: %d", change.kind);
+ elog(ERROR, "unrecognized kind of change: %d", kind);
/*
* If a change was applied now, increment CID for next writes and
* update the snapshot so it sees the changes we've applied so far.
*/
- if (change.kind != CHANGE_UPDATE_OLD)
+ if (kind != CHANGE_UPDATE_OLD)
{
CommandCounterIncrement();
UpdateActiveSnapshotCommandId();
}
-
- /* TTSOpsMinimalTuple has .get_heap_tuple==NULL. */
- Assert(shouldFree);
- pfree(tup_change);
}
- tuplestore_clear(dstate->tstore);
- dstate->nchanges = 0;
-
/* Cleanup. */
ExecDropSingleTupleTableSlot(index_slot);
ExecDropSingleTupleTableSlot(ident_slot);
@@ -2901,25 +3002,58 @@ find_target_tuple(Relation rel, ChangeDest *dest, HeapTuple tup_key,
}
/*
- * Decode and apply concurrent changes.
+ * Decode and apply concurrent changes, up to (and including) the record whose
+ * LSN is 'end_of_wal'.
*/
static void
-process_concurrent_changes(LogicalDecodingContext *decoding_ctx,
- XLogRecPtr end_of_wal, ChangeDest *dest)
+process_concurrent_changes(XLogRecPtr end_of_wal, ChangeDest *dest, bool done)
{
- RepackDecodingState *dstate;
+ DecodingWorkerShared *shared;
+ char fname[MAXPGPATH];
+ BufFile *file;
pgstat_progress_update_param(PROGRESS_REPACK_PHASE,
PROGRESS_REPACK_PHASE_CATCH_UP);
- dstate = (RepackDecodingState *) decoding_ctx->output_writer_private;
+ /* Ask the worker for the file. */
+ shared = (DecodingWorkerShared *) dsm_segment_address(decoding_worker->seg);
+ SpinLockAcquire(&shared->mutex);
+ shared->lsn_upto = end_of_wal;
+ shared->done = done;
+ SpinLockRelease(&shared->mutex);
- repack_decode_concurrent_changes(decoding_ctx, end_of_wal);
+ /*
+ * The worker needs to finish processing of the current WAL record. Even
+ * if it's idle, it'll need to close the output file. Thus we're likely to
+ * wait, so prepare for sleep.
+ */
+ ConditionVariablePrepareToSleep(&shared->cv);
+ for (;;)
+ {
+ bool valid;
- if (dstate->nchanges == 0)
- return;
+ SpinLockAcquire(&shared->mutex);
+ valid = shared->sfs_valid;
+ SpinLockRelease(&shared->mutex);
+
+ if (valid)
+ break;
+
+ ConditionVariableSleep(&shared->cv, WAIT_EVENT_REPACK_WORKER_EXPORT);
+ }
+ ConditionVariableCancelSleep();
- apply_concurrent_changes(dstate, dest);
+ /* Open the file. */
+ DecodingWorkerFileName(fname, shared->relid, shared->last_exported);
+ file = BufFileOpenFileSet(&shared->sfs.fs, fname, O_RDONLY, false);
+ apply_concurrent_changes(file, dest);
+
+ /* No file is exported until the worker exports the next one. */
+ SpinLockAcquire(&shared->mutex);
+ shared->sfs_valid = false;
+ SpinLockRelease(&shared->mutex);
+
+ BufFileClose(file);
}
/*
@@ -3045,15 +3179,10 @@ cleanup_logical_decoding(LogicalDecodingContext *ctx)
dstate = (RepackDecodingState *) ctx->output_writer_private;
- ExecDropSingleTupleTableSlot(dstate->tsslot);
- FreeTupleDesc(dstate->tupdesc_change);
FreeTupleDesc(dstate->tupdesc);
- tuplestore_end(dstate->tstore);
-
FreeDecodingContext(ctx);
- ReplicationSlotRelease();
- ReplicationSlotDrop(NameStr(dstate->slotname), false);
+ ReplicationSlotDropAcquired();
pfree(dstate);
}
@@ -3068,7 +3197,6 @@ cleanup_logical_decoding(LogicalDecodingContext *ctx)
static void
rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
Relation cl_index,
- LogicalDecodingContext *decoding_ctx,
TransactionId frozenXid,
MultiXactId cutoffMulti)
{
@@ -3172,7 +3300,7 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
* hold AccessExclusiveLock. (Quite some amount of WAL could have been
* written during the data copying and index creation.)
*/
- process_concurrent_changes(decoding_ctx, end_of_wal, &chgdst);
+ process_concurrent_changes(end_of_wal, &chgdst, false);
/*
* Acquire AccessExclusiveLock on the table, its TOAST relation (if there
@@ -3268,8 +3396,11 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
XLogFlush(wal_insert_ptr);
end_of_wal = GetFlushRecPtr(NULL);
- /* Apply the concurrent changes again. */
- process_concurrent_changes(decoding_ctx, end_of_wal, &chgdst);
+ /*
+ * Apply the concurrent changes again. Indicate that the decoding worker
+ * won't be needed anymore.
+ */
+ process_concurrent_changes(end_of_wal, &chgdst, true);
/* Remember info about rel before closing OldHeap */
relpersistence = OldHeap->rd_rel->relpersistence;
@@ -3382,3 +3513,514 @@ build_new_indexes(Relation NewHeap, Relation OldHeap, List *OldIndexes)
return result;
}
+
+/*
+ * Try to start a background worker to perform logical decoding of data
+ * changes applied to relation while REPACK CONCURRENTLY is copying its
+ * contents to a new table.
+ */
+static void
+start_decoding_worker(Oid relid)
+{
+ Size size;
+ dsm_segment *seg;
+ DecodingWorkerShared *shared;
+ shm_mq *mq;
+ shm_mq_handle *mqh;
+ BackgroundWorker bgw;
+
+ /* Setup shared memory. */
+ size = BUFFERALIGN(offsetof(DecodingWorkerShared, error_queue)) +
+ BUFFERALIGN(REPACK_ERROR_QUEUE_SIZE);
+ seg = dsm_create(size, 0);
+ shared = (DecodingWorkerShared *) dsm_segment_address(seg);
+ shared->lsn_upto = InvalidXLogRecPtr;
+ shared->done = false;
+ SharedFileSetInit(&shared->sfs, seg);
+ shared->sfs_valid = false;
+ shared->last_exported = -1;
+ SpinLockInit(&shared->mutex);
+ shared->dbid = MyDatabaseId;
+
+ /*
+ * This is the UserId set in cluster_rel(). Security context shouldn't be
+ * needed for decoding worker.
+ */
+ shared->roleid = GetUserId();
+ shared->relid = relid;
+ ConditionVariableInit(&shared->cv);
+ shared->backend_proc = MyProc;
+ shared->backend_pid = MyProcPid;
+ shared->backend_proc_number = MyProcNumber;
+
+ mq = shm_mq_create((char *) BUFFERALIGN(shared->error_queue),
+ REPACK_ERROR_QUEUE_SIZE);
+ shm_mq_set_receiver(mq, MyProc);
+ mqh = shm_mq_attach(mq, seg, NULL);
+
+ memset(&bgw, 0, sizeof(bgw));
+ snprintf(bgw.bgw_name, BGW_MAXLEN,
+ "REPACK decoding worker for relation \"%s\"",
+ get_rel_name(relid));
+ snprintf(bgw.bgw_type, BGW_MAXLEN, "REPACK decoding worker");
+ bgw.bgw_flags = BGWORKER_SHMEM_ACCESS |
+ BGWORKER_BACKEND_DATABASE_CONNECTION;
+ bgw.bgw_start_time = BgWorkerStart_RecoveryFinished;
+ bgw.bgw_restart_time = BGW_NEVER_RESTART;
+ snprintf(bgw.bgw_library_name, MAXPGPATH, "postgres");
+ snprintf(bgw.bgw_function_name, BGW_MAXLEN, "RepackWorkerMain");
+ bgw.bgw_main_arg = UInt32GetDatum(dsm_segment_handle(seg));
+ bgw.bgw_notify_pid = MyProcPid;
+
+ decoding_worker = palloc0_object(DecodingWorker);
+ if (!RegisterDynamicBackgroundWorker(&bgw, &decoding_worker->handle))
+ ereport(ERROR,
+ (errcode(ERRCODE_CONFIGURATION_LIMIT_EXCEEDED),
+ errmsg("out of background worker slots"),
+ errhint("You might need to increase \"%s\".", "max_worker_processes")));
+
+ decoding_worker->seg = seg;
+ decoding_worker->error_mqh = mqh;
+
+ /*
+ * The decoding setup must be done before the caller can have XID assigned
+ * for any reason, otherwise the worker might end up in a deadlock,
+ * waiting for the caller's transaction to end. Therefore wait here until
+ * the worker indicates that it has the logical decoding initialized.
+ */
+ ConditionVariablePrepareToSleep(&shared->cv);
+ for (;;)
+ {
+ int initialized;
+
+ SpinLockAcquire(&shared->mutex);
+ initialized = shared->initialized;
+ SpinLockRelease(&shared->mutex);
+
+ if (initialized)
+ break;
+
+ ConditionVariableSleep(&shared->cv, WAIT_EVENT_REPACK_WORKER_EXPORT);
+ }
+ ConditionVariableCancelSleep();
+}
+
+/*
+ * Stop the decoding worker and cleanup the related resources.
+ *
+ * The worker stops on its own when it knows there is no more work to do, but
+ * we need to stop it explicitly at least on ERROR in the launching backend.
+ */
+static void
+stop_decoding_worker(void)
+{
+ BgwHandleStatus status;
+
+ /* Haven't reached the worker startup? */
+ if (decoding_worker == NULL)
+ return;
+
+ /* Could not register the worker? */
+ if (decoding_worker->handle == NULL)
+ return;
+
+ TerminateBackgroundWorker(decoding_worker->handle);
+ /* The worker should really exit before the REPACK command does. */
+ HOLD_INTERRUPTS();
+ status = WaitForBackgroundWorkerShutdown(decoding_worker->handle);
+ RESUME_INTERRUPTS();
+
+ if (status == BGWH_POSTMASTER_DIED)
+ ereport(FATAL,
+ (errcode(ERRCODE_ADMIN_SHUTDOWN),
+ errmsg("postmaster exited during REPACK command")));
+
+ shm_mq_detach(decoding_worker->error_mqh);
+
+ /*
+ * If we could not cancel the current sleep due to ERROR, do that before
+ * we detach from the shared memory the condition variable is located in.
+ * If we did not, the bgworker ERROR handling code would try and fail
+ * badly.
+ */
+ ConditionVariableCancelSleep();
+
+ dsm_detach(decoding_worker->seg);
+ pfree(decoding_worker);
+ decoding_worker = NULL;
+}
+
+/* Is this process a REPACK worker? */
+static bool is_repack_worker = false;
+
+static pid_t backend_pid;
+static ProcNumber backend_proc_number;
+
+/*
+ * See ParallelWorkerShutdown for details.
+ */
+static void
+RepackWorkerShutdown(int code, Datum arg)
+{
+ SendProcSignal(backend_pid,
+ PROCSIG_REPACK_MESSAGE,
+ backend_proc_number);
+
+ dsm_detach((dsm_segment *) DatumGetPointer(arg));
+}
+
+/* REPACK decoding worker entry point */
+void
+RepackWorkerMain(Datum main_arg)
+{
+ dsm_segment *seg;
+ DecodingWorkerShared *shared;
+ shm_mq *mq;
+ shm_mq_handle *mqh;
+
+ is_repack_worker = true;
+
+ /*
+ * Override the default bgworker_die() with die() so we can use
+ * CHECK_FOR_INTERRUPTS().
+ */
+ pqsignal(SIGTERM, die);
+ BackgroundWorkerUnblockSignals();
+
+ seg = dsm_attach(DatumGetUInt32(main_arg));
+ if (seg == NULL)
+ ereport(ERROR,
+ (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
+ errmsg("could not map dynamic shared memory segment")));
+
+ shared = (DecodingWorkerShared *) dsm_segment_address(seg);
+
+ /* Arrange to signal the leader if we exit. */
+ backend_pid = shared->backend_pid;
+ backend_proc_number = shared->backend_proc_number;
+ before_shmem_exit(RepackWorkerShutdown, PointerGetDatum(seg));
+
+ /*
+ * Join locking group - see the comments around the call of
+ * start_decoding_worker().
+ */
+ if (!BecomeLockGroupMember(shared->backend_proc, backend_pid))
+ /* The leader is not running anymore. */
+ return;
+
+ /*
+ * Setup a queue to send error messages to the backend that launched this
+ * worker.
+ */
+ mq = (shm_mq *) (char *) BUFFERALIGN(shared->error_queue);
+ shm_mq_set_sender(mq, MyProc);
+ mqh = shm_mq_attach(mq, seg, NULL);
+ pq_redirect_to_shm_mq(seg, mqh);
+ pq_set_parallel_leader(shared->backend_pid,
+ shared->backend_proc_number);
+
+ /* Connect to the database. */
+ BackgroundWorkerInitializeConnectionByOid(shared->dbid, shared->roleid, 0);
+
+ repack_worker_internal(seg);
+}
+
+static void
+repack_worker_internal(dsm_segment *seg)
+{
+ DecodingWorkerShared *shared;
+ LogicalDecodingContext *decoding_ctx;
+ SharedFileSet *sfs;
+ Snapshot snapshot;
+
+ /*
+ * Transaction is needed to open relation, and it also provides us with a
+ * resource owner.
+ */
+ StartTransactionCommand();
+
+ shared = (DecodingWorkerShared *) dsm_segment_address(seg);
+
+ /*
+ * Not sure the spinlock is needed here - the backend should not change
+ * anything in the shared memory until we have serialized the snapshot.
+ */
+ SpinLockAcquire(&shared->mutex);
+ Assert(XLogRecPtrIsInvalid(shared->lsn_upto));
+ Assert(!shared->sfs_valid);
+ sfs = &shared->sfs;
+ SpinLockRelease(&shared->mutex);
+
+ SharedFileSetAttach(sfs, seg);
+
+ /*
+ * Prepare to capture the concurrent data changes ourselves.
+ */
+ decoding_ctx = setup_logical_decoding(shared->relid);
+
+ /* Announce that we're ready. */
+ SpinLockAcquire(&shared->mutex);
+ shared->initialized = true;
+ SpinLockRelease(&shared->mutex);
+ ConditionVariableSignal(&shared->cv);
+
+ /* Build the initial snapshot and export it. */
+ snapshot = SnapBuildInitialSnapshotForRepack(decoding_ctx->snapshot_builder);
+ export_initial_snapshot(snapshot, shared);
+
+ /*
+ * Only historic snapshots should be used now. Do not let us restrict the
+ * progress of xmin horizon.
+ */
+ InvalidateCatalogSnapshot();
+
+ while (!decode_concurrent_changes(decoding_ctx, shared))
+ ;
+
+ /* Cleanup. */
+ cleanup_logical_decoding(decoding_ctx);
+ CommitTransactionCommand();
+}
+
+/*
+ * Make snapshot available to the backend that launched the decoding worker.
+ */
+static void
+export_initial_snapshot(Snapshot snapshot, DecodingWorkerShared *shared)
+{
+ char fname[MAXPGPATH];
+ BufFile *file;
+ Size snap_size;
+ char *snap_space;
+
+ snap_size = EstimateSnapshotSpace(snapshot);
+ snap_space = (char *) palloc(snap_size);
+ SerializeSnapshot(snapshot, snap_space);
+ FreeSnapshot(snapshot);
+
+ DecodingWorkerFileName(fname, shared->relid, shared->last_exported + 1);
+ file = BufFileCreateFileSet(&shared->sfs.fs, fname);
+ /* To make restoration easier, write the snapshot size first. */
+ BufFileWrite(file, &snap_size, sizeof(snap_size));
+ BufFileWrite(file, snap_space, snap_size);
+ pfree(snap_space);
+ BufFileClose(file);
+
+ /* Tell the backend that the file is available. */
+ SpinLockAcquire(&shared->mutex);
+ shared->sfs_valid = true;
+ shared->last_exported++;
+ SpinLockRelease(&shared->mutex);
+ ConditionVariableSignal(&shared->cv);
+}
+
+/*
+ * Get the initial snapshot from the decoding worker.
+ */
+static Snapshot
+get_initial_snapshot(DecodingWorker *worker)
+{
+ DecodingWorkerShared *shared;
+ char fname[MAXPGPATH];
+ BufFile *file;
+ Size snap_size;
+ char *snap_space;
+ Snapshot snapshot;
+
+ shared = (DecodingWorkerShared *) dsm_segment_address(worker->seg);
+
+ /*
+ * The worker needs to initialize the logical decoding, which usually
+ * takes some time. Therefore it makes sense to prepare for the sleep
+ * first.
+ */
+ ConditionVariablePrepareToSleep(&shared->cv);
+ for (;;)
+ {
+ bool valid;
+
+ SpinLockAcquire(&shared->mutex);
+ valid = shared->sfs_valid;
+ SpinLockRelease(&shared->mutex);
+
+ if (valid)
+ break;
+
+ ConditionVariableSleep(&shared->cv, WAIT_EVENT_REPACK_WORKER_EXPORT);
+ }
+ ConditionVariableCancelSleep();
+
+ /* Read the snapshot from a file. */
+ DecodingWorkerFileName(fname, shared->relid, shared->last_exported);
+ file = BufFileOpenFileSet(&shared->sfs.fs, fname, O_RDONLY, false);
+ BufFileReadExact(file, &snap_size, sizeof(snap_size));
+ snap_space = (char *) palloc(snap_size);
+ BufFileReadExact(file, snap_space, snap_size);
+ BufFileClose(file);
+
+ SpinLockAcquire(&shared->mutex);
+ shared->sfs_valid = false;
+ SpinLockRelease(&shared->mutex);
+
+ /* Restore it. */
+ snapshot = RestoreSnapshot(snap_space);
+ pfree(snap_space);
+
+ return snapshot;
+}
+
+bool
+IsRepackWorker(void)
+{
+ return is_repack_worker;
+}
+
+/*
+ * Handle receipt of an interrupt indicating a repack worker message.
+ *
+ * Note: this is called within a signal handler! All we can do is set
+ * a flag that will cause the next CHECK_FOR_INTERRUPTS() to invoke
+ * ProcessRepackMessages().
+ */
+void
+HandleRepackMessageInterrupt(void)
+{
+ InterruptPending = true;
+ RepackMessagePending = true;
+ SetLatch(MyLatch);
+}
+
+/*
+ * Process any queued protocol messages received from parallel workers.
+ */
+void
+ProcessRepackMessages(void)
+{
+ MemoryContext oldcontext;
+
+ static MemoryContext hpm_context = NULL;
+
+ /*
+ * Nothing to do if we haven't launched the worker yet or have already
+ * terminated it.
+ */
+ if (decoding_worker == NULL)
+ return;
+
+ /*
+ * This is invoked from ProcessInterrupts(), and since some of the
+ * functions it calls contain CHECK_FOR_INTERRUPTS(), there is a potential
+ * for recursive calls if more signals are received while this runs. It's
+ * unclear that recursive entry would be safe, and it doesn't seem useful
+ * even if it is safe, so let's block interrupts until done.
+ */
+ HOLD_INTERRUPTS();
+
+ /*
+ * Moreover, CurrentMemoryContext might be pointing almost anywhere. We
+ * don't want to risk leaking data into long-lived contexts, so let's do
+ * our work here in a private context that we can reset on each use.
+ */
+ if (hpm_context == NULL) /* first time through? */
+ hpm_context = AllocSetContextCreate(TopMemoryContext,
+ "ProcessParallelMessages",
+ ALLOCSET_DEFAULT_SIZES);
+ else
+ MemoryContextReset(hpm_context);
+
+ oldcontext = MemoryContextSwitchTo(hpm_context);
+
+ /* OK to process messages. Reset the flag saying there are more to do. */
+ RepackMessagePending = false;
+
+ /*
+ * Read as many messages as we can from each worker, but stop when no more
+ * messages can be read from the worker without blocking.
+ */
+ while (true)
+ {
+ shm_mq_result res;
+ Size nbytes;
+ void *data;
+
+ res = shm_mq_receive(decoding_worker->error_mqh, &nbytes,
+ &data, true);
+ if (res == SHM_MQ_WOULD_BLOCK)
+ break;
+ else if (res == SHM_MQ_SUCCESS)
+ {
+ StringInfoData msg;
+
+ initStringInfo(&msg);
+ appendBinaryStringInfo(&msg, data, nbytes);
+ ProcessRepackMessage(&msg);
+ pfree(msg.data);
+ }
+ else
+ {
+ /*
+ * The decoding worker is special in that it exits as soon as it
+ * has its work done. Thus the DETACHED result code is fine.
+ */
+ Assert(res = SHM_MQ_DETACHED);
+
+ break;
+ }
+ }
+
+ MemoryContextSwitchTo(oldcontext);
+
+ /* Might as well clear the context on our way out */
+ MemoryContextReset(hpm_context);
+
+ RESUME_INTERRUPTS();
+}
+
+/*
+ * Process a single protocol message received from a single parallel worker.
+ */
+static void
+ProcessRepackMessage(StringInfo msg)
+{
+ char msgtype;
+
+ msgtype = pq_getmsgbyte(msg);
+
+ switch (msgtype)
+ {
+ case PqMsg_ErrorResponse:
+ case PqMsg_NoticeResponse:
+ {
+ ErrorData edata;
+
+ /* Parse ErrorResponse or NoticeResponse. */
+ pq_parse_errornotice(msg, &edata);
+
+ /* Death of a worker isn't enough justification for suicide. */
+ edata.elevel = Min(edata.elevel, ERROR);
+
+ /*
+ * If desired, add a context line to show that this is a
+ * message propagated from a parallel worker. Otherwise, it
+ * can sometimes be confusing to understand what actually
+ * happened.
+ */
+ if (edata.context)
+ edata.context = psprintf("%s\n%s", edata.context,
+ _("decoding worker"));
+ else
+ edata.context = pstrdup(_("decoding worker"));
+
+ /* Rethrow error or print notice. */
+ ThrowErrorData(&edata);
+
+ break;
+ }
+
+ default:
+ {
+ elog(ERROR, "unrecognized message type received from decoding worker: %c (message length %d bytes)",
+ msgtype, msg->len);
+ }
+ }
+}
diff --git a/src/backend/libpq/pqmq.c b/src/backend/libpq/pqmq.c
index 2b75de0ddef..28a5a400fb1 100644
--- a/src/backend/libpq/pqmq.c
+++ b/src/backend/libpq/pqmq.c
@@ -14,6 +14,7 @@
#include "postgres.h"
#include "access/parallel.h"
+#include "commands/cluster.h"
#include "libpq/libpq.h"
#include "libpq/pqformat.h"
#include "libpq/pqmq.h"
@@ -175,6 +176,10 @@ mq_putmessage(char msgtype, const char *s, size_t len)
SendProcSignal(pq_mq_parallel_leader_pid,
PROCSIG_PARALLEL_APPLY_MESSAGE,
pq_mq_parallel_leader_proc_number);
+ else if (IsRepackWorker())
+ SendProcSignal(pq_mq_parallel_leader_pid,
+ PROCSIG_REPACK_MESSAGE,
+ pq_mq_parallel_leader_proc_number);
else
{
Assert(IsParallelWorker());
diff --git a/src/backend/postmaster/bgworker.c b/src/backend/postmaster/bgworker.c
index 8e1068969ae..2d1ba94a45f 100644
--- a/src/backend/postmaster/bgworker.c
+++ b/src/backend/postmaster/bgworker.c
@@ -13,6 +13,7 @@
#include "postgres.h"
#include "access/parallel.h"
+#include "commands/cluster.h"
#include "libpq/pqsignal.h"
#include "miscadmin.h"
#include "pgstat.h"
@@ -135,6 +136,9 @@ static const struct
},
{
"SequenceSyncWorkerMain", SequenceSyncWorkerMain
+ },
+ {
+ "RepackWorkerMain", RepackWorkerMain
}
};
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index 1b11ed63dc6..b3fd7fec392 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -205,7 +205,11 @@ StartupDecodingContext(List *output_plugin_options,
ctx->slot = slot;
- ctx->reader = XLogReaderAllocate(wal_segment_size, NULL, xl_routine, ctx);
+ /*
+ * TODO A separate patch for PG core, unless there's really a reason to
+ * pass ctx for private_data (May extensions expect ctx?).
+ */
+ ctx->reader = XLogReaderAllocate(wal_segment_size, NULL, xl_routine, NULL);
if (!ctx->reader)
ereport(ERROR,
(errcode(ERRCODE_OUT_OF_MEMORY),
diff --git a/src/backend/replication/pgoutput_repack/pgoutput_repack.c b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
index c8930640a0d..fb9956d392d 100644
--- a/src/backend/replication/pgoutput_repack/pgoutput_repack.c
+++ b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
@@ -168,17 +168,13 @@ store_change(LogicalDecodingContext *ctx, ConcurrentChangeKind kind,
HeapTuple tuple)
{
RepackDecodingState *dstate;
- char *change_raw;
- ConcurrentChange change;
+ char kind_byte = (char) kind;
bool flattened = false;
- Size size;
- Datum values[1];
- bool isnull[1];
- char *dst;
dstate = (RepackDecodingState *) ctx->output_writer_private;
- size = VARHDRSZ + SizeOfConcurrentChange;
+ /* Store the change kind. */
+ BufFileWrite(dstate->file, &kind_byte, 1);
/*
* ReorderBufferCommit() stores the TOAST chunks in its private memory
@@ -195,46 +191,12 @@ store_change(LogicalDecodingContext *ctx, ConcurrentChangeKind kind,
tuple = toast_flatten_tuple(tuple, dstate->tupdesc);
flattened = true;
}
+ /* Store the tuple size ... */
+ BufFileWrite(dstate->file, &tuple->t_len, sizeof(tuple->t_len));
+ /* ... and the tuple itself. */
+ BufFileWrite(dstate->file, tuple->t_data, tuple->t_len);
- size += tuple->t_len;
- if (size >= MaxAllocSize)
- elog(ERROR, "Change is too big.");
-
- /* Construct the change. */
- change_raw = (char *) palloc0(size);
- SET_VARSIZE(change_raw, size);
-
- /*
- * Since the varlena alignment might not be sufficient for the structure,
- * set the fields in a local instance and remember where it should
- * eventually be copied.
- */
- change.kind = kind;
- dst = (char *) VARDATA(change_raw);
-
- /*
- * Copy the tuple.
- *
- * Note: change->tup_data.t_data must be fixed on retrieval!
- */
- memcpy(&change.tup_data, tuple, sizeof(HeapTupleData));
- memcpy(dst, &change, SizeOfConcurrentChange);
- dst += SizeOfConcurrentChange;
- memcpy(dst, tuple->t_data, tuple->t_len);
-
- /* The data has been copied. */
+ /* Free the flat copy if created above. */
if (flattened)
pfree(tuple);
-
- /* Store as tuple of 1 bytea column. */
- values[0] = PointerGetDatum(change_raw);
- isnull[0] = false;
- tuplestore_putvalues(dstate->tstore, dstate->tupdesc_change,
- values, isnull);
-
- /* Accounting. */
- dstate->nchanges++;
-
- /* Cleanup. */
- pfree(change_raw);
}
diff --git a/src/backend/storage/ipc/procsignal.c b/src/backend/storage/ipc/procsignal.c
index 087821311cc..af12144795b 100644
--- a/src/backend/storage/ipc/procsignal.c
+++ b/src/backend/storage/ipc/procsignal.c
@@ -19,6 +19,7 @@
#include "access/parallel.h"
#include "commands/async.h"
+#include "commands/cluster.h"
#include "miscadmin.h"
#include "pgstat.h"
#include "port/pg_bitutils.h"
@@ -694,6 +695,9 @@ procsignal_sigusr1_handler(SIGNAL_ARGS)
if (CheckProcSignal(PROCSIG_PARALLEL_APPLY_MESSAGE))
HandleParallelApplyMessageInterrupt();
+ if (CheckProcSignal(PROCSIG_REPACK_MESSAGE))
+ HandleRepackMessageInterrupt();
+
if (CheckProcSignal(PROCSIG_RECOVERY_CONFLICT_DATABASE))
HandleRecoveryConflictInterrupt(PROCSIG_RECOVERY_CONFLICT_DATABASE);
diff --git a/src/backend/tcop/postgres.c b/src/backend/tcop/postgres.c
index 7dd75a490aa..4a4858882f0 100644
--- a/src/backend/tcop/postgres.c
+++ b/src/backend/tcop/postgres.c
@@ -36,6 +36,7 @@
#include "access/xact.h"
#include "catalog/pg_type.h"
#include "commands/async.h"
+#include "commands/cluster.h"
#include "commands/event_trigger.h"
#include "commands/explain_state.h"
#include "commands/prepare.h"
@@ -3541,6 +3542,9 @@ ProcessInterrupts(void)
if (ParallelApplyMessagePending)
ProcessParallelApplyMessages();
+
+ if (RepackMessagePending)
+ ProcessRepackMessages();
}
/*
diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt
index f39830dbb34..cbcc8550960 100644
--- a/src/backend/utils/activity/wait_event_names.txt
+++ b/src/backend/utils/activity/wait_event_names.txt
@@ -62,6 +62,7 @@ LOGICAL_APPLY_MAIN "Waiting in main loop of logical replication apply process."
LOGICAL_LAUNCHER_MAIN "Waiting in main loop of logical replication launcher process."
LOGICAL_PARALLEL_APPLY_MAIN "Waiting in main loop of logical replication parallel apply process."
RECOVERY_WAL_STREAM "Waiting in main loop of startup process for WAL to arrive, during streaming recovery."
+REPACK_WORKER_MAIN "Waiting in main loop of REPACK decoding worker process."
REPLICATION_SLOTSYNC_MAIN "Waiting in main loop of slot sync worker."
REPLICATION_SLOTSYNC_SHUTDOWN "Waiting for slot sync worker to shut down."
SYSLOGGER_MAIN "Waiting in main loop of syslogger process."
@@ -153,6 +154,7 @@ RECOVERY_CONFLICT_SNAPSHOT "Waiting for recovery conflict resolution for a vacuu
RECOVERY_CONFLICT_TABLESPACE "Waiting for recovery conflict resolution for dropping a tablespace."
RECOVERY_END_COMMAND "Waiting for <xref linkend="guc-recovery-end-command"/> to complete."
RECOVERY_PAUSE "Waiting for recovery to be resumed."
+REPACK_WORKER_EXPORT "Waiting for decoding worker to export a new output file."
REPLICATION_ORIGIN_DROP "Waiting for a replication origin to become inactive so it can be dropped."
REPLICATION_SLOT_DROP "Waiting for a replication slot to become inactive so it can be dropped."
RESTORE_COMMAND "Waiting for <xref linkend="guc-restore-command"/> to complete."
diff --git a/src/include/access/tableam.h b/src/include/access/tableam.h
index d8f76d325f9..321c00682ec 100644
--- a/src/include/access/tableam.h
+++ b/src/include/access/tableam.h
@@ -22,7 +22,6 @@
#include "access/xact.h"
#include "commands/vacuum.h"
#include "executor/tuptable.h"
-#include "replication/logical.h"
#include "storage/read_stream.h"
#include "utils/rel.h"
#include "utils/snapshot.h"
@@ -631,7 +630,6 @@ typedef struct TableAmRoutine
bool use_sort,
TransactionId OldestXmin,
Snapshot snapshot,
- LogicalDecodingContext *decoding_ctx,
TransactionId *xid_cutoff,
MultiXactId *multi_cutoff,
double *num_tuples,
@@ -1651,8 +1649,6 @@ table_relation_copy_data(Relation rel, const RelFileLocator *newrlocator)
* - *multi_cutoff - ditto
* - snapshot - if != NULL, ignore data changes done by transactions that this
* (MVCC) snapshot considers still in-progress or in the future.
- * - decoding_ctx - logical decoding context, to capture concurrent data
- * changes.
*
* Output parameters:
* - *xid_cutoff - rel's new relfrozenxid value, may be invalid
@@ -1666,7 +1662,6 @@ table_relation_copy_for_cluster(Relation OldTable, Relation NewTable,
bool use_sort,
TransactionId OldestXmin,
Snapshot snapshot,
- LogicalDecodingContext *decoding_ctx,
TransactionId *xid_cutoff,
MultiXactId *multi_cutoff,
double *num_tuples,
@@ -1675,7 +1670,7 @@ table_relation_copy_for_cluster(Relation OldTable, Relation NewTable,
{
OldTable->rd_tableam->relation_copy_for_cluster(OldTable, NewTable, OldIndex,
use_sort, OldestXmin,
- snapshot, decoding_ctx,
+ snapshot,
xid_cutoff, multi_cutoff,
num_tuples, tups_vacuumed,
tups_recently_dead);
diff --git a/src/include/commands/cluster.h b/src/include/commands/cluster.h
index b43a1740053..0ac70ec30d7 100644
--- a/src/include/commands/cluster.h
+++ b/src/include/commands/cluster.h
@@ -17,6 +17,7 @@
#include "nodes/parsenodes.h"
#include "parser/parse_node.h"
#include "replication/logical.h"
+#include "storage/buffile.h"
#include "storage/lock.h"
#include "storage/relfilelocator.h"
#include "utils/relcache.h"
@@ -47,6 +48,9 @@ typedef struct ClusterParams
extern RelFileLocator repacked_rel_locator;
extern RelFileLocator repacked_rel_toast_locator;
+/*
+ * Stored as a single byte in the output file.
+ */
typedef enum
{
CHANGE_INSERT,
@@ -55,68 +59,30 @@ typedef enum
CHANGE_DELETE
} ConcurrentChangeKind;
-typedef struct ConcurrentChange
-{
- /* See the enum above. */
- ConcurrentChangeKind kind;
-
- /*
- * The actual tuple.
- *
- * The tuple data follows the ConcurrentChange structure. Before use make
- * sure the tuple is correctly aligned (ConcurrentChange can be stored as
- * bytea) and that tuple->t_data is fixed.
- */
- HeapTupleData tup_data;
-} ConcurrentChange;
-
-#define SizeOfConcurrentChange (offsetof(ConcurrentChange, tup_data) + \
- sizeof(HeapTupleData))
-
/*
* Logical decoding state.
*
- * Here we store the data changes that we decode from WAL while the table
- * contents is being copied to a new storage. Also the necessary metadata
- * needed to apply these changes to the table is stored here.
+ * The output plugin uses it to store the data changes that it decodes from
+ * WAL while the table contents is being copied to a new storage.
*/
typedef struct RepackDecodingState
{
/* The relation whose changes we're decoding. */
Oid relid;
- /* Replication slot name. */
- NameData slotname;
-
- /*
- * Decoded changes are stored here. Although we try to avoid excessive
- * batches, it can happen that the changes need to be stored to disk. The
- * tuplestore does this transparently.
- */
- Tuplestorestate *tstore;
-
- /* The current number of changes in tstore. */
- double nchanges;
-
- /*
- * Descriptor to store the ConcurrentChange structure serialized (bytea).
- * We can't store the tuple directly because tuplestore only supports
- * minimum tuple and we may need to transfer OID system column from the
- * output plugin. Also we need to transfer the change kind, so it's better
- * to put everything in the structure than to use 2 tuplestores "in
- * parallel".
- */
- TupleDesc tupdesc_change;
-
- /* Tuple descriptor needed to update indexes. */
+ /* Tuple descriptor of the relation being processed. */
TupleDesc tupdesc;
- /* Slot to retrieve data from tstore. */
- TupleTableSlot *tsslot;
-
- ResourceOwner resowner;
+ /* The current output file. */
+ BufFile *file;
} RepackDecodingState;
+extern PGDLLIMPORT volatile sig_atomic_t RepackMessagePending;
+
+extern bool IsRepackWorker(void);
+extern void HandleRepackMessageInterrupt(void);
+extern void ProcessRepackMessages(void);
+
extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel);
extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid,
@@ -125,9 +91,6 @@ extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid,
LOCKMODE lockmode);
extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal);
-extern void repack_decode_concurrent_changes(LogicalDecodingContext *ctx,
- XLogRecPtr end_of_wal);
-
extern Oid make_new_heap(Oid OIDOldHeap, Oid NewTableSpace, Oid NewAccessMethod,
char relpersistence, LOCKMODE lockmode);
extern void finish_heap_swap(Oid OIDOldHeap, Oid OIDNewHeap,
@@ -140,4 +103,5 @@ extern void finish_heap_swap(Oid OIDOldHeap, Oid OIDNewHeap,
MultiXactId cutoffMulti,
char newrelpersistence);
+extern void RepackWorkerMain(Datum main_arg);
#endif /* CLUSTER_H */
diff --git a/src/include/storage/procsignal.h b/src/include/storage/procsignal.h
index afeeb1ca019..c0a66516b66 100644
--- a/src/include/storage/procsignal.h
+++ b/src/include/storage/procsignal.h
@@ -36,6 +36,7 @@ typedef enum
PROCSIG_BARRIER, /* global barrier interrupt */
PROCSIG_LOG_MEMORY_CONTEXT, /* ask backend to log the memory contexts */
PROCSIG_PARALLEL_APPLY_MESSAGE, /* Message from parallel apply workers */
+ PROCSIG_REPACK_MESSAGE, /* Message from repack worker */
/* Recovery conflict reasons */
PROCSIG_RECOVERY_CONFLICT_FIRST,
diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list
index 3139b14e85f..35344910f65 100644
--- a/src/tools/pgindent/typedefs.list
+++ b/src/tools/pgindent/typedefs.list
@@ -488,7 +488,6 @@ CompressFileHandle
CompressionLocation
CompressorState
ComputeXidHorizonsResult
-ConcurrentChange
ConcurrentChangeKind
ConditionVariable
ConditionVariableMinimallyPadded
@@ -629,6 +628,9 @@ DeclareCursorStmt
DecodedBkpBlock
DecodedXLogRecord
DecodingOutputState
+DecodingWorker
+DecodingWorkerShared
+DecodingWorkerState
DefElem
DefElemAction
DefaultACLInfo
--
2.47.3
--=-=-=
Content-Type: text/plain
Content-Disposition: attachment;
filename=v28-0006-Use-multiple-snapshots-to-copy-the-data.patch
^ permalink raw reply [nested|flat] 55+ messages in thread
* [PATCH 5/6] Use background worker to do logical decoding.
@ 2025-12-13 18:27 Antonin Houska <[email protected]>
0 siblings, 0 replies; 55+ messages in thread
From: Antonin Houska @ 2025-12-13 18:27 UTC (permalink / raw)
If the backend performing REPACK (CONCURRENTLY) does both data copying and
logical decoding, it has to "travel in time" back and forth and therefore it
has to invalidate system caches quite a few times. (The copying and the
decoding work with different catalog snapshots.) As the decoding worker has
separate caches, the switching is not necessary.
Without the worker, it'd also be difficult to switch between potentiallly long
running tasks like index build and WAL decoding. (No decoding during that time
at all can suspend archiving / recycling of WAL segments for some time, which
in turn may result in full disk.)
Another problem is that, after having acquired AccessExclusiveLock (in order
to swap the files), the backend needs to both decode and apply the data
changes that took place while it was waiting for the lock. With the decoding
worker, the decoding runs all the time, so the backend only needs to apply the
changes. This can reduce the time the exclusive lock is held for.
Note that the code added in order to handle ERRORs in the background worker
almost duplicates the existing code that does the same for other types of
workers (See ProcessParallelMessages() and
ProcessParallelApplyMessages()). Refactoring of the existing code might be
useful, to reduce the duplication.
---
src/backend/access/heap/heapam_handler.c | 44 -
src/backend/commands/cluster.c | 1144 +++++++++++++----
src/backend/libpq/pqmq.c | 5 +
src/backend/postmaster/bgworker.c | 4 +
src/backend/replication/logical/logical.c | 6 +-
.../pgoutput_repack/pgoutput_repack.c | 54 +-
src/backend/storage/ipc/procsignal.c | 4 +
src/backend/tcop/postgres.c | 4 +
.../utils/activity/wait_event_names.txt | 2 +
src/include/access/tableam.h | 7 +-
src/include/commands/cluster.h | 68 +-
src/include/storage/procsignal.h | 1 +
src/tools/pgindent/typedefs.list | 4 +-
13 files changed, 946 insertions(+), 401 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index 6063e037edb..cb09e6fd1dc 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -33,7 +33,6 @@
#include "catalog/index.h"
#include "catalog/storage.h"
#include "catalog/storage_xlog.h"
-#include "commands/cluster.h"
#include "commands/progress.h"
#include "executor/executor.h"
#include "miscadmin.h"
@@ -688,7 +687,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
Relation OldIndex, bool use_sort,
TransactionId OldestXmin,
Snapshot snapshot,
- LogicalDecodingContext *decoding_ctx,
TransactionId *xid_cutoff,
MultiXactId *multi_cutoff,
double *num_tuples,
@@ -710,7 +708,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
BufferHeapTupleTableSlot *hslot;
BlockNumber prev_cblock = InvalidBlockNumber;
bool concurrent = snapshot != NULL;
- XLogRecPtr end_of_wal_prev = GetFlushRecPtr(NULL);
/* Remember if it's a system catalog */
is_system_catalog = IsSystemRelation(OldHeap);
@@ -957,31 +954,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
ct_val[1] = *num_tuples;
pgstat_progress_update_multi_param(2, ct_index, ct_val);
}
-
- /*
- * Process the WAL produced by the load, as well as by other
- * transactions, so that the replication slot can advance and WAL does
- * not pile up. Use wal_segment_size as a threshold so that we do not
- * introduce the decoding overhead too often.
- *
- * Of course, we must not apply the changes until the initial load has
- * completed.
- *
- * Note that our insertions into the new table should not be decoded
- * as we (intentionally) do not write the logical decoding specific
- * information to WAL.
- */
- if (concurrent)
- {
- XLogRecPtr end_of_wal;
-
- end_of_wal = GetFlushRecPtr(NULL);
- if ((end_of_wal - end_of_wal_prev) > wal_segment_size)
- {
- repack_decode_concurrent_changes(decoding_ctx, end_of_wal);
- end_of_wal_prev = end_of_wal;
- }
- }
}
if (indexScan != NULL)
@@ -1027,22 +999,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
-
- /*
- * Try to keep the amount of not-yet-decoded WAL small, like
- * above.
- */
- if (concurrent)
- {
- XLogRecPtr end_of_wal;
-
- end_of_wal = GetFlushRecPtr(NULL);
- if ((end_of_wal - end_of_wal_prev) > wal_segment_size)
- {
- repack_decode_concurrent_changes(decoding_ctx, end_of_wal);
- end_of_wal_prev = end_of_wal;
- }
- }
}
tuplesort_end(tuplesort);
diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c
index 569705abc81..b0383c1375f 100644
--- a/src/backend/commands/cluster.c
+++ b/src/backend/commands/cluster.c
@@ -12,12 +12,13 @@
* In concurrent mode, we lock the table with only ShareUpdateExclusiveLock,
* then do an initial copy as above. However, while the tuples are being
* copied, concurrent transactions could modify the table. To cope with those
- * changes, we rely on logical decoding to obtain them from WAL. The changes
- * are accumulated in a tuplestore. Once the initial copy is complete, we
- * read the changes from the tuplestore and re-apply them on the new heap.
- * Then we upgrade our ShareUpdateExclusiveLock to AccessExclusiveLock and
- * swap the relfilenodes. This way, the time we hold a strong lock on the
- * table is much reduced, and the bloat is eliminated.
+ * changes, we rely on logical decoding to obtain them from WAL. A bgworker
+ * consumes WAL while the initial copy is ongoing (to prevent excessive WAL
+ * from being reserved), and accumulates the changes in a file. Once the
+ * initial copy is complete, we read the changes from the file and re-apply
+ * them on the new heap. Then we upgrade our ShareUpdateExclusiveLock to
+ * AccessExclusiveLock and swap the relfilenodes. This way, the time we hold
+ * a strong lock on the table is much reduced, and the bloat is eliminated.
*
* There is hardly anything left of Paul Brown's original implementation...
*
@@ -61,6 +62,8 @@
#include "commands/tablecmds.h"
#include "commands/vacuum.h"
#include "executor/executor.h"
+#include "libpq/pqformat.h"
+#include "libpq/pqmq.h"
#include "miscadmin.h"
#include "optimizer/optimizer.h"
#include "pgstat.h"
@@ -71,6 +74,8 @@
#include "storage/ipc.h"
#include "storage/lmgr.h"
#include "storage/predicate.h"
+#include "storage/procsignal.h"
+#include "tcop/tcopprot.h"
#include "utils/acl.h"
#include "utils/fmgroids.h"
#include "utils/guc.h"
@@ -138,6 +143,106 @@ typedef struct ChangeDest
IndexInsertState *iistate;
} ChangeDest;
+/*
+ * Layout of shared memory used for communication between backend and the
+ * worker that performs logical decoding of data changes
+ */
+typedef struct DecodingWorkerShared
+{
+ /* Is the decoding initialized? */
+ bool initialized;
+
+ /*
+ * Once the worker has reached this LSN, it should close the current
+ * output file and either create a new one or exit, according to the field
+ * 'done'. If the value is InvalidXLogRecPtr, the worker should decode all
+ * the WAL available and keep checking this field. It is ok if the worker
+ * had already decoded records whose LSN is >= lsn_upto before this field
+ * has been set.
+ */
+ XLogRecPtr lsn_upto;
+
+ /* Exit after closing the current file? */
+ bool done;
+
+ /* The output is stored here. */
+ SharedFileSet sfs;
+
+ /* Can backend read the file contents? */
+ bool sfs_valid;
+
+ /* Number of the last file exported by the worker. */
+ int last_exported;
+
+ /* Synchronize access to the fields above. */
+ slock_t mutex;
+
+ /* Database to connect to. */
+ Oid dbid;
+
+ /* Role to connect as. */
+ Oid roleid;
+
+ /* Decode data changes of this relation. */
+ Oid relid;
+
+ /* The backend uses this to wait for the worker. */
+ ConditionVariable cv;
+
+ /* Info to signal the backend. */
+ PGPROC *backend_proc;
+ pid_t backend_pid;
+ ProcNumber backend_proc_number;
+
+ /* Error queue. */
+ shm_mq *error_mq;
+
+ /*
+ * Memory the queue is located int.
+ *
+ * For considerations on the value see the comments of
+ * PARALLEL_ERROR_QUEUE_SIZE.
+ */
+#define REPACK_ERROR_QUEUE_SIZE 16384
+ char error_queue[FLEXIBLE_ARRAY_MEMBER];
+} DecodingWorkerShared;
+
+/*
+ * Generate worker's output file name. If relations of the same 'relid' happen
+ * to be processed at the same time, they must be from different databases and
+ * therefore different backends must be involved. (PID is already present in
+ * the fileset name.)
+ */
+static inline void
+DecodingWorkerFileName(char *fname, Oid relid, uint32 seq)
+{
+ snprintf(fname, MAXPGPATH, "%u-%u", relid, seq);
+}
+
+/*
+ * Backend-local information to control the decoding worker.
+ */
+typedef struct DecodingWorker
+{
+ /* The worker. */
+ BackgroundWorkerHandle *handle;
+
+ /* DecodingWorkerShared is in this segment. */
+ dsm_segment *seg;
+
+ /* Handle of the error queue. */
+ shm_mq_handle *error_mqh;
+} DecodingWorker;
+
+/* Pointer to currently running decoding worker. */
+static DecodingWorker *decoding_worker = NULL;
+
+/*
+ * Is there a message sent by a repack worker that the backend needs to
+ * receive?
+ */
+volatile sig_atomic_t RepackMessagePending = false;
+
static bool cluster_rel_recheck(RepackCommand cmd, Relation OldHeap,
Oid indexOid, Oid userid, LOCKMODE lmode,
int options);
@@ -145,7 +250,7 @@ static void check_repack_concurrently_requirements(Relation rel);
static void rebuild_relation(Relation OldHeap, Relation index, bool verbose,
bool concurrent);
static void copy_table_data(Relation NewHeap, Relation OldHeap, Relation OldIndex,
- Snapshot snapshot, LogicalDecodingContext *decoding_ctx,
+ Snapshot snapshot,
bool verbose,
bool *pSwapToastByContent,
TransactionId *pFreezeXid,
@@ -158,12 +263,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd,
static bool cluster_is_permitted_for_relation(RepackCommand cmd,
Oid relid, Oid userid);
-static void begin_concurrent_repack(Relation rel);
-static void end_concurrent_repack(void);
static LogicalDecodingContext *setup_logical_decoding(Oid relid);
-static HeapTuple get_changed_tuple(char *change);
-static void apply_concurrent_changes(RepackDecodingState *dstate,
- ChangeDest *dest);
+static bool decode_concurrent_changes(LogicalDecodingContext *ctx,
+ DecodingWorkerShared *shared);
+static void apply_concurrent_changes(BufFile *file, ChangeDest *dest);
static void apply_concurrent_insert(Relation rel, HeapTuple tup,
IndexInsertState *iistate,
TupleTableSlot *index_slot);
@@ -175,9 +278,9 @@ static void apply_concurrent_delete(Relation rel, HeapTuple tup_target);
static HeapTuple find_target_tuple(Relation rel, ChangeDest *dest,
HeapTuple tup_key,
TupleTableSlot *ident_slot);
-static void process_concurrent_changes(LogicalDecodingContext *decoding_ctx,
- XLogRecPtr end_of_wal,
- ChangeDest *dest);
+static void process_concurrent_changes(XLogRecPtr end_of_wal,
+ ChangeDest *dest,
+ bool done);
static IndexInsertState *get_index_insert_state(Relation relation,
Oid ident_index_id,
Relation *ident_index_p);
@@ -187,7 +290,6 @@ static void free_index_insert_state(IndexInsertState *iistate);
static void cleanup_logical_decoding(LogicalDecodingContext *ctx);
static void rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
Relation cl_index,
- LogicalDecodingContext *decoding_ctx,
TransactionId frozenXid,
MultiXactId cutoffMulti);
static List *build_new_indexes(Relation NewHeap, Relation OldHeap, List *OldIndexes);
@@ -197,6 +299,13 @@ static Relation process_single_relation(RepackStmt *stmt,
ClusterParams *params);
static Oid determine_clustered_index(Relation rel, bool usingindex,
const char *indexname);
+static void start_decoding_worker(Oid relid);
+static void stop_decoding_worker(void);
+static void repack_worker_internal(dsm_segment *seg);
+static void export_initial_snapshot(Snapshot snapshot,
+ DecodingWorkerShared *shared);
+static Snapshot get_initial_snapshot(DecodingWorker *worker);
+static void ProcessRepackMessage(StringInfo msg);
static const char *RepackCommandAsString(RepackCommand cmd);
@@ -619,20 +728,20 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid,
/* rebuild_relation does all the dirty work */
PG_TRY();
{
- /*
- * For concurrent processing, make sure that our logical decoding
- * ignores data changes of other tables than the one we are
- * processing.
- */
- if (concurrent)
- begin_concurrent_repack(OldHeap);
-
rebuild_relation(OldHeap, index, verbose, concurrent);
}
PG_FINALLY();
{
if (concurrent)
- end_concurrent_repack();
+ {
+ /*
+ * Since during normal operation the worker was already asked to
+ * exit, stopping it explicitly is especially important on ERROR.
+ * However it still seems a good practice to make sure that the
+ * worker never survives the REPACK command.
+ */
+ stop_decoding_worker();
+ }
}
PG_END_TRY();
@@ -929,7 +1038,6 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
bool swap_toast_by_content;
TransactionId frozenXid;
MultiXactId cutoffMulti;
- LogicalDecodingContext *decoding_ctx = NULL;
Snapshot snapshot = NULL;
#if USE_ASSERT_CHECKING
LOCKMODE lmode;
@@ -943,19 +1051,36 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
if (concurrent)
{
/*
- * Prepare to capture the concurrent data changes.
+ * The worker needs to be member of the locking group we're the leader
+ * of. We ought to become the leader before the worker starts. The
+ * worker will join the group as soon as it starts.
+ *
+ * This is to make sure that the deadlock described below is
+ * detectable by deadlock.c: if the worker waits for a transaction to
+ * complete and we are waiting for the worker output, then effectively
+ * we (i.e. this backend) are waiting for that transaction.
+ */
+ BecomeLockGroupLeader();
+
+ /*
+ * Start the worker that decodes data changes applied while we're
+ * copying the table contents.
*
- * Note that this call waits for all transactions with XID already
- * assigned to finish. If some of those transactions is waiting for a
- * lock conflicting with ShareUpdateExclusiveLock on our table (e.g.
- * it runs CREATE INDEX), we can end up in a deadlock. Not sure this
- * risk is worth unlocking/locking the table (and its clustering
- * index) and checking again if its still eligible for REPACK
- * CONCURRENTLY.
+ * Note that the worker has to wait for all transactions with XID
+ * already assigned to finish. If some of those transactions is
+ * waiting for a lock conflicting with ShareUpdateExclusiveLock on our
+ * table (e.g. it runs CREATE INDEX), we can end up in a deadlock.
+ * Not sure this risk is worth unlocking/locking the table (and its
+ * clustering index) and checking again if its still eligible for
+ * REPACK CONCURRENTLY.
+ */
+ start_decoding_worker(tableOid);
+
+ /*
+ * Wait until the worker has the initial snapshot and retrieve it.
*/
- decoding_ctx = setup_logical_decoding(tableOid);
+ snapshot = get_initial_snapshot(decoding_worker);
- snapshot = SnapBuildInitialSnapshotForRepack(decoding_ctx->snapshot_builder);
PushActiveSnapshot(snapshot);
}
@@ -980,7 +1105,7 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
NewHeap = table_open(OIDNewHeap, NoLock);
/* Copy the heap data into the new table in the desired order */
- copy_table_data(NewHeap, OldHeap, index, snapshot, decoding_ctx, verbose,
+ copy_table_data(NewHeap, OldHeap, index, snapshot, verbose,
&swap_toast_by_content, &frozenXid, &cutoffMulti);
/* The historic snapshot won't be needed anymore. */
@@ -994,14 +1119,10 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
{
Assert(!swap_toast_by_content);
rebuild_relation_finish_concurrent(NewHeap, OldHeap, index,
- decoding_ctx,
frozenXid, cutoffMulti);
pgstat_progress_update_param(PROGRESS_REPACK_PHASE,
PROGRESS_REPACK_PHASE_FINAL_CLEANUP);
-
- /* Done with decoding. */
- cleanup_logical_decoding(decoding_ctx);
}
else
{
@@ -1172,8 +1293,7 @@ make_new_heap(Oid OIDOldHeap, Oid NewTableSpace, Oid NewAccessMethod,
*/
static void
copy_table_data(Relation NewHeap, Relation OldHeap, Relation OldIndex,
- Snapshot snapshot, LogicalDecodingContext *decoding_ctx,
- bool verbose, bool *pSwapToastByContent,
+ Snapshot snapshot, bool verbose, bool *pSwapToastByContent,
TransactionId *pFreezeXid, MultiXactId *pCutoffMulti)
{
Relation relRelation;
@@ -1334,7 +1454,6 @@ copy_table_data(Relation NewHeap, Relation OldHeap, Relation OldIndex,
*/
table_relation_copy_for_cluster(OldHeap, NewHeap, OldIndex, use_sort,
cutoffs.OldestXmin, snapshot,
- decoding_ctx,
&cutoffs.FreezeLimit,
&cutoffs.MultiXactCutoff,
&num_tuples, &tups_vacuumed,
@@ -2367,59 +2486,6 @@ RepackCommandAsString(RepackCommand cmd)
return "???";
}
-
-/*
- * Call this function before REPACK CONCURRENTLY starts to setup logical
- * decoding. It makes sure that other users of the table put enough
- * information into WAL.
- *
- * The point is that at various places we expect that the table we're
- * processing is treated like a system catalog. For example, we need to be
- * able to scan it using a "historic snapshot" anytime during the processing
- * (as opposed to scanning only at the start point of the decoding, as logical
- * replication does during initial table synchronization), in order to apply
- * concurrent UPDATE / DELETE commands.
- *
- * Note that TOAST table needs no attention here as it's not scanned using
- * historic snapshot.
- */
-static void
-begin_concurrent_repack(Relation rel)
-{
- Oid toastrelid;
-
- /*
- * Avoid logical decoding of other relations by this backend. The lock we
- * have guarantees that the actual locator cannot be changed concurrently:
- * TRUNCATE needs AccessExclusiveLock.
- */
- Assert(CheckRelationLockedByMe(rel, ShareUpdateExclusiveLock, false));
- repacked_rel_locator = rel->rd_locator;
- toastrelid = rel->rd_rel->reltoastrelid;
- if (OidIsValid(toastrelid))
- {
- Relation toastrel;
-
- /* Avoid logical decoding of other TOAST relations. */
- toastrel = table_open(toastrelid, AccessShareLock);
- repacked_rel_toast_locator = toastrel->rd_locator;
- table_close(toastrel, AccessShareLock);
- }
-}
-
-/*
- * Call this when done with REPACK CONCURRENTLY.
- */
-static void
-end_concurrent_repack(void)
-{
- /*
- * Restore normal function of (future) logical decoding for this backend.
- */
- repacked_rel_locator.relNumber = InvalidOid;
- repacked_rel_toast_locator.relNumber = InvalidOid;
-}
-
/*
* This function is much like pg_create_logical_replication_slot() except that
* the new slot is neither released (if anyone else could read changes from
@@ -2431,9 +2497,10 @@ static LogicalDecodingContext *
setup_logical_decoding(Oid relid)
{
Relation rel;
- TupleDesc tupdesc;
+ Oid toastrelid;
LogicalDecodingContext *ctx;
- RepackDecodingState *dstate = palloc0_object(RepackDecodingState);
+ NameData slotname;
+ RepackDecodingState *dstate;
/*
* REPACK CONCURRENTLY is not allowed in a transaction block, so this
@@ -2441,21 +2508,21 @@ setup_logical_decoding(Oid relid)
*/
Assert(!TransactionIdIsValid(GetTopTransactionIdIfAny()));
- /*
- * A single backend should not execute multiple REPACK commands at a time,
- * so use PID to make the slot unique.
- */
- snprintf(NameStr(dstate->slotname), NAMEDATALEN, "repack_%d", MyProcPid);
-
/*
* Check if we can use logical decoding.
*/
CheckSlotPermissions();
CheckLogicalDecodingRequirements();
- /* RS_TEMPORARY so that the slot gets cleaned up on ERROR. */
- ReplicationSlotCreate(NameStr(dstate->slotname), true, RS_TEMPORARY,
- false, false, false);
+ /*
+ * A single backend should not execute multiple REPACK commands at a time,
+ * so use PID to make the slot unique.
+ *
+ * RS_TEMPORARY so that the slot gets cleaned up on ERROR.
+ */
+ snprintf(NameStr(slotname), NAMEDATALEN, "repack_%d", MyProcPid);
+ ReplicationSlotCreate(NameStr(slotname), true, RS_TEMPORARY, false, false,
+ false);
/*
* Neither prepare_write nor do_write callback nor update_progress is
@@ -2477,104 +2544,109 @@ setup_logical_decoding(Oid relid)
DecodingContextFindStartpoint(ctx);
+ /*
+ * decode_concurrent_changes() needs non-blocking callback.
+ */
+ ctx->reader->routine.page_read = read_local_xlog_page_no_wait;
+
+ /*
+ * read_local_xlog_page_no_wait() needs to be able to indicate the end of
+ * WAL.
+ */
+ ctx->reader->private_data = MemoryContextAllocZero(ctx->context,
+ sizeof(ReadLocalXLogPageNoWaitPrivate));
+
+
/* Some WAL records should have been read. */
Assert(ctx->reader->EndRecPtr != InvalidXLogRecPtr);
+ /*
+ * Initialize repack_current_segment so that we can notice WAL segment
+ * boundaries.
+ */
XLByteToSeg(ctx->reader->EndRecPtr, repack_current_segment,
wal_segment_size);
- /*
- * Setup structures to store decoded changes.
- */
+ dstate = palloc0_object(RepackDecodingState);
dstate->relid = relid;
- dstate->tstore = tuplestore_begin_heap(false, false,
- maintenance_work_mem);
- /* Caller should already have the table locked. */
- rel = table_open(relid, NoLock);
- tupdesc = CreateTupleDescCopy(RelationGetDescr(rel));
- dstate->tupdesc = tupdesc;
- table_close(rel, NoLock);
+ /*
+ * Tuple descriptor may be needed to flatten a tuple before we write it to
+ * a file. A copy is needed because the decoding worker invalidates system
+ * caches before it starts to do the actual work.
+ */
+ rel = table_open(relid, AccessShareLock);
+ dstate->tupdesc = CreateTupleDescCopy(RelationGetDescr(rel));
- /* Initialize the descriptor to store the changes ... */
- dstate->tupdesc_change = CreateTemplateTupleDesc(1);
+ /* Avoid logical decoding of other relations. */
+ repacked_rel_locator = rel->rd_locator;
+ toastrelid = rel->rd_rel->reltoastrelid;
+ if (OidIsValid(toastrelid))
+ {
+ Relation toastrel;
- TupleDescInitEntry(dstate->tupdesc_change, 1, NULL, BYTEAOID, -1, 0);
- /* ... as well as the corresponding slot. */
- dstate->tsslot = MakeSingleTupleTableSlot(dstate->tupdesc_change,
- &TTSOpsMinimalTuple);
+ /* Avoid logical decoding of other TOAST relations. */
+ toastrel = table_open(toastrelid, AccessShareLock);
+ repacked_rel_toast_locator = toastrel->rd_locator;
+ table_close(toastrel, AccessShareLock);
+ }
+ table_close(rel, AccessShareLock);
- dstate->resowner = ResourceOwnerCreate(CurrentResourceOwner,
- "logical decoding");
+ /* The file will be set as soon as we have it opened. */
+ dstate->file = NULL;
ctx->output_writer_private = dstate;
+
return ctx;
}
/*
- * Retrieve tuple from ConcurrentChange structure.
+ * Decode logical changes from the WAL sequence and store them to a file.
*
- * The input data starts with the structure but it might not be appropriately
- * aligned.
+ * If true is returned, there is no more work for the worker.
*/
-static HeapTuple
-get_changed_tuple(char *change)
-{
- HeapTupleData tup_data;
- HeapTuple result;
- char *src;
-
- /*
- * Ensure alignment before accessing the fields. (This is why we can't use
- * heap_copytuple() instead of this function.)
- */
- src = change + offsetof(ConcurrentChange, tup_data);
- memcpy(&tup_data, src, sizeof(HeapTupleData));
-
- result = (HeapTuple) palloc(HEAPTUPLESIZE + tup_data.t_len);
- memcpy(result, &tup_data, sizeof(HeapTupleData));
- result->t_data = (HeapTupleHeader) ((char *) result + HEAPTUPLESIZE);
- src = change + SizeOfConcurrentChange;
- memcpy(result->t_data, src, result->t_len);
-
- return result;
-}
-
-/*
- * Decode logical changes from the WAL sequence up to end_of_wal.
- */
-void
-repack_decode_concurrent_changes(LogicalDecodingContext *ctx,
- XLogRecPtr end_of_wal)
+static bool
+decode_concurrent_changes(LogicalDecodingContext *ctx,
+ DecodingWorkerShared *shared)
{
RepackDecodingState *dstate;
- ResourceOwner resowner_old;
+ XLogRecPtr lsn_upto;
+ bool done;
+ char fname[MAXPGPATH];
dstate = (RepackDecodingState *) ctx->output_writer_private;
- resowner_old = CurrentResourceOwner;
- CurrentResourceOwner = dstate->resowner;
- PG_TRY();
+ /* Open the output file. */
+ DecodingWorkerFileName(fname, shared->relid, shared->last_exported + 1);
+ dstate->file = BufFileCreateFileSet(&shared->sfs.fs, fname);
+
+ SpinLockAcquire(&shared->mutex);
+ lsn_upto = shared->lsn_upto;
+ done = shared->done;
+ SpinLockRelease(&shared->mutex);
+
+ while (true)
{
- while (ctx->reader->EndRecPtr < end_of_wal)
- {
- XLogRecord *record;
- XLogSegNo segno_new;
- char *errm = NULL;
- XLogRecPtr end_lsn;
+ XLogRecord *record;
+ XLogSegNo segno_new;
+ char *errm = NULL;
+ XLogRecPtr end_lsn;
- record = XLogReadRecord(ctx->reader, &errm);
- if (errm)
- elog(ERROR, "%s", errm);
+ CHECK_FOR_INTERRUPTS();
- if (record != NULL)
- LogicalDecodingProcessRecord(ctx, ctx->reader);
+ record = XLogReadRecord(ctx->reader, &errm);
+ if (record)
+ {
+ LogicalDecodingProcessRecord(ctx, ctx->reader);
/*
* If WAL segment boundary has been crossed, inform the decoding
- * system that the catalog_xmin can advance. (We can confirm more
- * often, but a filling a single WAL segment should not take much
- * time.)
+ * system that the catalog_xmin can advance.
+ *
+ * TODO Does it make sense to confirm more often? Segment size
+ * seems appropriate for restart_lsn (because less than a segment
+ * cannot be recycled anyway), however more frequent checks might
+ * be beneficial for catalog_xmin.
*/
end_lsn = ctx->reader->EndRecPtr;
XLByteToSeg(end_lsn, segno_new, wal_segment_size);
@@ -2585,80 +2657,117 @@ repack_decode_concurrent_changes(LogicalDecodingContext *ctx,
(uint32) (end_lsn >> 32), (uint32) end_lsn);
repack_current_segment = segno_new;
}
+ }
+ else
+ {
+ ReadLocalXLogPageNoWaitPrivate *priv;
+
+ if (errm)
+ ereport(ERROR, (errmsg("%s", errm)));
- CHECK_FOR_INTERRUPTS();
+ /*
+ * In the decoding loop we do not want to get blocked when there
+ * is no more WAL available, otherwise the loop would become
+ * uninterruptible. The point is that the worker is only useful if
+ * it starts decoding before lsn_upto is set. Thus it can reach
+ * the end of WAL and find out later that it did not have to go
+ * that far.
+ */
+ priv = (ReadLocalXLogPageNoWaitPrivate *)
+ ctx->reader->private_data;
+ if (priv->end_of_wal)
+ priv->end_of_wal = false;
+ else
+ ereport(ERROR, (errmsg("could not read WAL record")));
}
- InvalidateSystemCaches();
- CurrentResourceOwner = resowner_old;
- }
- PG_CATCH();
- {
- /* clear all timetravel entries */
- InvalidateSystemCaches();
- CurrentResourceOwner = resowner_old;
- PG_RE_THROW();
+
+ /*
+ * Whether we could read new record or not, keep checking if
+ * 'lsn_upto' was specified.
+ */
+ if (XLogRecPtrIsInvalid(lsn_upto))
+ {
+ SpinLockAcquire(&shared->mutex);
+ lsn_upto = shared->lsn_upto;
+ /* 'done' should be set at the same time as 'lsn_upto' */
+ done = shared->done;
+ SpinLockRelease(&shared->mutex);
+ }
+ if (!XLogRecPtrIsInvalid(lsn_upto) &&
+ ctx->reader->EndRecPtr >= lsn_upto)
+ break;
+
+ if (record == NULL)
+ /* Wait a bit before we retry reading WAL. */
+ (void) WaitLatch(MyLatch,
+ WL_LATCH_SET | WL_TIMEOUT | WL_EXIT_ON_PM_DEATH,
+ 100000L, /* XXX Tune the delay. */
+ WAIT_EVENT_REPACK_WORKER_MAIN);
}
- PG_END_TRY();
+
+ /*
+ * Close the file so we can make it available to the backend.
+ */
+ BufFileClose(dstate->file);
+ dstate->file = NULL;
+ SpinLockAcquire(&shared->mutex);
+ shared->lsn_upto = InvalidXLogRecPtr;
+ shared->sfs_valid = true;
+ shared->last_exported++;
+ SpinLockRelease(&shared->mutex);
+ ConditionVariableSignal(&shared->cv);
+
+ return done;
}
/*
* Apply changes stored in 'file'.
*/
static void
-apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
+apply_concurrent_changes(BufFile *file, ChangeDest *dest)
{
+ char kind;
+ uint32 t_len;
Relation rel = dest->rel;
TupleTableSlot *index_slot,
*ident_slot;
HeapTuple tup_old = NULL;
- if (dstate->nchanges == 0)
- return;
-
/* TupleTableSlot is needed to pass the tuple to ExecInsertIndexTuples(). */
- index_slot = MakeSingleTupleTableSlot(dstate->tupdesc, &TTSOpsHeapTuple);
+ index_slot = MakeSingleTupleTableSlot(RelationGetDescr(rel),
+ &TTSOpsHeapTuple);
/* A slot to fetch tuples from identity index. */
ident_slot = table_slot_create(rel, NULL);
- while (tuplestore_gettupleslot(dstate->tstore, true, false,
- dstate->tsslot))
+ while (true)
{
- bool shouldFree;
- HeapTuple tup_change,
- tup,
+ size_t nread;
+ HeapTuple tup,
tup_exist;
- char *change_raw,
- *src;
- ConcurrentChange change;
- bool isnull[1];
- Datum values[1];
CHECK_FOR_INTERRUPTS();
- /* Get the change from the single-column tuple. */
- tup_change = ExecFetchSlotHeapTuple(dstate->tsslot, false, &shouldFree);
- heap_deform_tuple(tup_change, dstate->tupdesc_change, values, isnull);
- Assert(!isnull[0]);
-
- /* Make sure we access aligned data. */
- change_raw = (char *) DatumGetByteaP(values[0]);
- src = (char *) VARDATA(change_raw);
- memcpy(&change, src, SizeOfConcurrentChange);
+ nread = BufFileReadMaybeEOF(file, &kind, 1, true);
+ /* Are we done with the file? */
+ if (nread == 0)
+ break;
- /*
- * Extract the tuple from the change. The tuple is copied here because
- * it might be assigned to 'tup_old', in which case it needs to
- * survive into the next iteration.
- */
- tup = get_changed_tuple(src);
+ /* Read the tuple. */
+ BufFileReadExact(file, &t_len, sizeof(t_len));
+ tup = (HeapTuple) palloc(HEAPTUPLESIZE + t_len);
+ tup->t_data = (HeapTupleHeader) ((char *) tup + HEAPTUPLESIZE);
+ BufFileReadExact(file, tup->t_data, t_len);
+ tup->t_len = t_len;
+ ItemPointerSetInvalid(&tup->t_self);
+ tup->t_tableOid = RelationGetRelid(dest->rel);
- if (change.kind == CHANGE_UPDATE_OLD)
+ if (kind == CHANGE_UPDATE_OLD)
{
Assert(tup_old == NULL);
tup_old = tup;
}
- else if (change.kind == CHANGE_INSERT)
+ else if (kind == CHANGE_INSERT)
{
Assert(tup_old == NULL);
@@ -2666,12 +2775,11 @@ apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
pfree(tup);
}
- else if (change.kind == CHANGE_UPDATE_NEW ||
- change.kind == CHANGE_DELETE)
+ else if (kind == CHANGE_UPDATE_NEW || kind == CHANGE_DELETE)
{
HeapTuple tup_key;
- if (change.kind == CHANGE_UPDATE_NEW)
+ if (kind == CHANGE_UPDATE_NEW)
{
tup_key = tup_old != NULL ? tup_old : tup;
}
@@ -2688,7 +2796,7 @@ apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
if (tup_exist == NULL)
elog(ERROR, "failed to find target tuple");
- if (change.kind == CHANGE_UPDATE_NEW)
+ if (kind == CHANGE_UPDATE_NEW)
apply_concurrent_update(rel, tup, tup_exist, dest->iistate,
index_slot);
else
@@ -2703,26 +2811,19 @@ apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
pfree(tup);
}
else
- elog(ERROR, "unrecognized kind of change: %d", change.kind);
+ elog(ERROR, "unrecognized kind of change: %d", kind);
/*
* If a change was applied now, increment CID for next writes and
* update the snapshot so it sees the changes we've applied so far.
*/
- if (change.kind != CHANGE_UPDATE_OLD)
+ if (kind != CHANGE_UPDATE_OLD)
{
CommandCounterIncrement();
UpdateActiveSnapshotCommandId();
}
-
- /* TTSOpsMinimalTuple has .get_heap_tuple==NULL. */
- Assert(shouldFree);
- pfree(tup_change);
}
- tuplestore_clear(dstate->tstore);
- dstate->nchanges = 0;
-
/* Cleanup. */
ExecDropSingleTupleTableSlot(index_slot);
ExecDropSingleTupleTableSlot(ident_slot);
@@ -2901,25 +3002,58 @@ find_target_tuple(Relation rel, ChangeDest *dest, HeapTuple tup_key,
}
/*
- * Decode and apply concurrent changes.
+ * Decode and apply concurrent changes, up to (and including) the record whose
+ * LSN is 'end_of_wal'.
*/
static void
-process_concurrent_changes(LogicalDecodingContext *decoding_ctx,
- XLogRecPtr end_of_wal, ChangeDest *dest)
+process_concurrent_changes(XLogRecPtr end_of_wal, ChangeDest *dest, bool done)
{
- RepackDecodingState *dstate;
+ DecodingWorkerShared *shared;
+ char fname[MAXPGPATH];
+ BufFile *file;
pgstat_progress_update_param(PROGRESS_REPACK_PHASE,
PROGRESS_REPACK_PHASE_CATCH_UP);
- dstate = (RepackDecodingState *) decoding_ctx->output_writer_private;
+ /* Ask the worker for the file. */
+ shared = (DecodingWorkerShared *) dsm_segment_address(decoding_worker->seg);
+ SpinLockAcquire(&shared->mutex);
+ shared->lsn_upto = end_of_wal;
+ shared->done = done;
+ SpinLockRelease(&shared->mutex);
- repack_decode_concurrent_changes(decoding_ctx, end_of_wal);
+ /*
+ * The worker needs to finish processing of the current WAL record. Even
+ * if it's idle, it'll need to close the output file. Thus we're likely to
+ * wait, so prepare for sleep.
+ */
+ ConditionVariablePrepareToSleep(&shared->cv);
+ for (;;)
+ {
+ bool valid;
- if (dstate->nchanges == 0)
- return;
+ SpinLockAcquire(&shared->mutex);
+ valid = shared->sfs_valid;
+ SpinLockRelease(&shared->mutex);
+
+ if (valid)
+ break;
+
+ ConditionVariableSleep(&shared->cv, WAIT_EVENT_REPACK_WORKER_EXPORT);
+ }
+ ConditionVariableCancelSleep();
- apply_concurrent_changes(dstate, dest);
+ /* Open the file. */
+ DecodingWorkerFileName(fname, shared->relid, shared->last_exported);
+ file = BufFileOpenFileSet(&shared->sfs.fs, fname, O_RDONLY, false);
+ apply_concurrent_changes(file, dest);
+
+ /* No file is exported until the worker exports the next one. */
+ SpinLockAcquire(&shared->mutex);
+ shared->sfs_valid = false;
+ SpinLockRelease(&shared->mutex);
+
+ BufFileClose(file);
}
/*
@@ -3045,15 +3179,10 @@ cleanup_logical_decoding(LogicalDecodingContext *ctx)
dstate = (RepackDecodingState *) ctx->output_writer_private;
- ExecDropSingleTupleTableSlot(dstate->tsslot);
- FreeTupleDesc(dstate->tupdesc_change);
FreeTupleDesc(dstate->tupdesc);
- tuplestore_end(dstate->tstore);
-
FreeDecodingContext(ctx);
- ReplicationSlotRelease();
- ReplicationSlotDrop(NameStr(dstate->slotname), false);
+ ReplicationSlotDropAcquired();
pfree(dstate);
}
@@ -3068,7 +3197,6 @@ cleanup_logical_decoding(LogicalDecodingContext *ctx)
static void
rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
Relation cl_index,
- LogicalDecodingContext *decoding_ctx,
TransactionId frozenXid,
MultiXactId cutoffMulti)
{
@@ -3172,7 +3300,7 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
* hold AccessExclusiveLock. (Quite some amount of WAL could have been
* written during the data copying and index creation.)
*/
- process_concurrent_changes(decoding_ctx, end_of_wal, &chgdst);
+ process_concurrent_changes(end_of_wal, &chgdst, false);
/*
* Acquire AccessExclusiveLock on the table, its TOAST relation (if there
@@ -3268,8 +3396,11 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
XLogFlush(wal_insert_ptr);
end_of_wal = GetFlushRecPtr(NULL);
- /* Apply the concurrent changes again. */
- process_concurrent_changes(decoding_ctx, end_of_wal, &chgdst);
+ /*
+ * Apply the concurrent changes again. Indicate that the decoding worker
+ * won't be needed anymore.
+ */
+ process_concurrent_changes(end_of_wal, &chgdst, true);
/* Remember info about rel before closing OldHeap */
relpersistence = OldHeap->rd_rel->relpersistence;
@@ -3382,3 +3513,514 @@ build_new_indexes(Relation NewHeap, Relation OldHeap, List *OldIndexes)
return result;
}
+
+/*
+ * Try to start a background worker to perform logical decoding of data
+ * changes applied to relation while REPACK CONCURRENTLY is copying its
+ * contents to a new table.
+ */
+static void
+start_decoding_worker(Oid relid)
+{
+ Size size;
+ dsm_segment *seg;
+ DecodingWorkerShared *shared;
+ shm_mq *mq;
+ shm_mq_handle *mqh;
+ BackgroundWorker bgw;
+
+ /* Setup shared memory. */
+ size = BUFFERALIGN(offsetof(DecodingWorkerShared, error_queue)) +
+ BUFFERALIGN(REPACK_ERROR_QUEUE_SIZE);
+ seg = dsm_create(size, 0);
+ shared = (DecodingWorkerShared *) dsm_segment_address(seg);
+ shared->lsn_upto = InvalidXLogRecPtr;
+ shared->done = false;
+ SharedFileSetInit(&shared->sfs, seg);
+ shared->sfs_valid = false;
+ shared->last_exported = -1;
+ SpinLockInit(&shared->mutex);
+ shared->dbid = MyDatabaseId;
+
+ /*
+ * This is the UserId set in cluster_rel(). Security context shouldn't be
+ * needed for decoding worker.
+ */
+ shared->roleid = GetUserId();
+ shared->relid = relid;
+ ConditionVariableInit(&shared->cv);
+ shared->backend_proc = MyProc;
+ shared->backend_pid = MyProcPid;
+ shared->backend_proc_number = MyProcNumber;
+
+ mq = shm_mq_create((char *) BUFFERALIGN(shared->error_queue),
+ REPACK_ERROR_QUEUE_SIZE);
+ shm_mq_set_receiver(mq, MyProc);
+ mqh = shm_mq_attach(mq, seg, NULL);
+
+ memset(&bgw, 0, sizeof(bgw));
+ snprintf(bgw.bgw_name, BGW_MAXLEN,
+ "REPACK decoding worker for relation \"%s\"",
+ get_rel_name(relid));
+ snprintf(bgw.bgw_type, BGW_MAXLEN, "REPACK decoding worker");
+ bgw.bgw_flags = BGWORKER_SHMEM_ACCESS |
+ BGWORKER_BACKEND_DATABASE_CONNECTION;
+ bgw.bgw_start_time = BgWorkerStart_RecoveryFinished;
+ bgw.bgw_restart_time = BGW_NEVER_RESTART;
+ snprintf(bgw.bgw_library_name, MAXPGPATH, "postgres");
+ snprintf(bgw.bgw_function_name, BGW_MAXLEN, "RepackWorkerMain");
+ bgw.bgw_main_arg = UInt32GetDatum(dsm_segment_handle(seg));
+ bgw.bgw_notify_pid = MyProcPid;
+
+ decoding_worker = palloc0_object(DecodingWorker);
+ if (!RegisterDynamicBackgroundWorker(&bgw, &decoding_worker->handle))
+ ereport(ERROR,
+ (errcode(ERRCODE_CONFIGURATION_LIMIT_EXCEEDED),
+ errmsg("out of background worker slots"),
+ errhint("You might need to increase \"%s\".", "max_worker_processes")));
+
+ decoding_worker->seg = seg;
+ decoding_worker->error_mqh = mqh;
+
+ /*
+ * The decoding setup must be done before the caller can have XID assigned
+ * for any reason, otherwise the worker might end up in a deadlock,
+ * waiting for the caller's transaction to end. Therefore wait here until
+ * the worker indicates that it has the logical decoding initialized.
+ */
+ ConditionVariablePrepareToSleep(&shared->cv);
+ for (;;)
+ {
+ int initialized;
+
+ SpinLockAcquire(&shared->mutex);
+ initialized = shared->initialized;
+ SpinLockRelease(&shared->mutex);
+
+ if (initialized)
+ break;
+
+ ConditionVariableSleep(&shared->cv, WAIT_EVENT_REPACK_WORKER_EXPORT);
+ }
+ ConditionVariableCancelSleep();
+}
+
+/*
+ * Stop the decoding worker and cleanup the related resources.
+ *
+ * The worker stops on its own when it knows there is no more work to do, but
+ * we need to stop it explicitly at least on ERROR in the launching backend.
+ */
+static void
+stop_decoding_worker(void)
+{
+ BgwHandleStatus status;
+
+ /* Haven't reached the worker startup? */
+ if (decoding_worker == NULL)
+ return;
+
+ /* Could not register the worker? */
+ if (decoding_worker->handle == NULL)
+ return;
+
+ TerminateBackgroundWorker(decoding_worker->handle);
+ /* The worker should really exit before the REPACK command does. */
+ HOLD_INTERRUPTS();
+ status = WaitForBackgroundWorkerShutdown(decoding_worker->handle);
+ RESUME_INTERRUPTS();
+
+ if (status == BGWH_POSTMASTER_DIED)
+ ereport(FATAL,
+ (errcode(ERRCODE_ADMIN_SHUTDOWN),
+ errmsg("postmaster exited during REPACK command")));
+
+ shm_mq_detach(decoding_worker->error_mqh);
+
+ /*
+ * If we could not cancel the current sleep due to ERROR, do that before
+ * we detach from the shared memory the condition variable is located in.
+ * If we did not, the bgworker ERROR handling code would try and fail
+ * badly.
+ */
+ ConditionVariableCancelSleep();
+
+ dsm_detach(decoding_worker->seg);
+ pfree(decoding_worker);
+ decoding_worker = NULL;
+}
+
+/* Is this process a REPACK worker? */
+static bool is_repack_worker = false;
+
+static pid_t backend_pid;
+static ProcNumber backend_proc_number;
+
+/*
+ * See ParallelWorkerShutdown for details.
+ */
+static void
+RepackWorkerShutdown(int code, Datum arg)
+{
+ SendProcSignal(backend_pid,
+ PROCSIG_REPACK_MESSAGE,
+ backend_proc_number);
+
+ dsm_detach((dsm_segment *) DatumGetPointer(arg));
+}
+
+/* REPACK decoding worker entry point */
+void
+RepackWorkerMain(Datum main_arg)
+{
+ dsm_segment *seg;
+ DecodingWorkerShared *shared;
+ shm_mq *mq;
+ shm_mq_handle *mqh;
+
+ is_repack_worker = true;
+
+ /*
+ * Override the default bgworker_die() with die() so we can use
+ * CHECK_FOR_INTERRUPTS().
+ */
+ pqsignal(SIGTERM, die);
+ BackgroundWorkerUnblockSignals();
+
+ seg = dsm_attach(DatumGetUInt32(main_arg));
+ if (seg == NULL)
+ ereport(ERROR,
+ (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
+ errmsg("could not map dynamic shared memory segment")));
+
+ shared = (DecodingWorkerShared *) dsm_segment_address(seg);
+
+ /* Arrange to signal the leader if we exit. */
+ backend_pid = shared->backend_pid;
+ backend_proc_number = shared->backend_proc_number;
+ before_shmem_exit(RepackWorkerShutdown, PointerGetDatum(seg));
+
+ /*
+ * Join locking group - see the comments around the call of
+ * start_decoding_worker().
+ */
+ if (!BecomeLockGroupMember(shared->backend_proc, backend_pid))
+ /* The leader is not running anymore. */
+ return;
+
+ /*
+ * Setup a queue to send error messages to the backend that launched this
+ * worker.
+ */
+ mq = (shm_mq *) (char *) BUFFERALIGN(shared->error_queue);
+ shm_mq_set_sender(mq, MyProc);
+ mqh = shm_mq_attach(mq, seg, NULL);
+ pq_redirect_to_shm_mq(seg, mqh);
+ pq_set_parallel_leader(shared->backend_pid,
+ shared->backend_proc_number);
+
+ /* Connect to the database. */
+ BackgroundWorkerInitializeConnectionByOid(shared->dbid, shared->roleid, 0);
+
+ repack_worker_internal(seg);
+}
+
+static void
+repack_worker_internal(dsm_segment *seg)
+{
+ DecodingWorkerShared *shared;
+ LogicalDecodingContext *decoding_ctx;
+ SharedFileSet *sfs;
+ Snapshot snapshot;
+
+ /*
+ * Transaction is needed to open relation, and it also provides us with a
+ * resource owner.
+ */
+ StartTransactionCommand();
+
+ shared = (DecodingWorkerShared *) dsm_segment_address(seg);
+
+ /*
+ * Not sure the spinlock is needed here - the backend should not change
+ * anything in the shared memory until we have serialized the snapshot.
+ */
+ SpinLockAcquire(&shared->mutex);
+ Assert(XLogRecPtrIsInvalid(shared->lsn_upto));
+ Assert(!shared->sfs_valid);
+ sfs = &shared->sfs;
+ SpinLockRelease(&shared->mutex);
+
+ SharedFileSetAttach(sfs, seg);
+
+ /*
+ * Prepare to capture the concurrent data changes ourselves.
+ */
+ decoding_ctx = setup_logical_decoding(shared->relid);
+
+ /* Announce that we're ready. */
+ SpinLockAcquire(&shared->mutex);
+ shared->initialized = true;
+ SpinLockRelease(&shared->mutex);
+ ConditionVariableSignal(&shared->cv);
+
+ /* Build the initial snapshot and export it. */
+ snapshot = SnapBuildInitialSnapshotForRepack(decoding_ctx->snapshot_builder);
+ export_initial_snapshot(snapshot, shared);
+
+ /*
+ * Only historic snapshots should be used now. Do not let us restrict the
+ * progress of xmin horizon.
+ */
+ InvalidateCatalogSnapshot();
+
+ while (!decode_concurrent_changes(decoding_ctx, shared))
+ ;
+
+ /* Cleanup. */
+ cleanup_logical_decoding(decoding_ctx);
+ CommitTransactionCommand();
+}
+
+/*
+ * Make snapshot available to the backend that launched the decoding worker.
+ */
+static void
+export_initial_snapshot(Snapshot snapshot, DecodingWorkerShared *shared)
+{
+ char fname[MAXPGPATH];
+ BufFile *file;
+ Size snap_size;
+ char *snap_space;
+
+ snap_size = EstimateSnapshotSpace(snapshot);
+ snap_space = (char *) palloc(snap_size);
+ SerializeSnapshot(snapshot, snap_space);
+ FreeSnapshot(snapshot);
+
+ DecodingWorkerFileName(fname, shared->relid, shared->last_exported + 1);
+ file = BufFileCreateFileSet(&shared->sfs.fs, fname);
+ /* To make restoration easier, write the snapshot size first. */
+ BufFileWrite(file, &snap_size, sizeof(snap_size));
+ BufFileWrite(file, snap_space, snap_size);
+ pfree(snap_space);
+ BufFileClose(file);
+
+ /* Tell the backend that the file is available. */
+ SpinLockAcquire(&shared->mutex);
+ shared->sfs_valid = true;
+ shared->last_exported++;
+ SpinLockRelease(&shared->mutex);
+ ConditionVariableSignal(&shared->cv);
+}
+
+/*
+ * Get the initial snapshot from the decoding worker.
+ */
+static Snapshot
+get_initial_snapshot(DecodingWorker *worker)
+{
+ DecodingWorkerShared *shared;
+ char fname[MAXPGPATH];
+ BufFile *file;
+ Size snap_size;
+ char *snap_space;
+ Snapshot snapshot;
+
+ shared = (DecodingWorkerShared *) dsm_segment_address(worker->seg);
+
+ /*
+ * The worker needs to initialize the logical decoding, which usually
+ * takes some time. Therefore it makes sense to prepare for the sleep
+ * first.
+ */
+ ConditionVariablePrepareToSleep(&shared->cv);
+ for (;;)
+ {
+ bool valid;
+
+ SpinLockAcquire(&shared->mutex);
+ valid = shared->sfs_valid;
+ SpinLockRelease(&shared->mutex);
+
+ if (valid)
+ break;
+
+ ConditionVariableSleep(&shared->cv, WAIT_EVENT_REPACK_WORKER_EXPORT);
+ }
+ ConditionVariableCancelSleep();
+
+ /* Read the snapshot from a file. */
+ DecodingWorkerFileName(fname, shared->relid, shared->last_exported);
+ file = BufFileOpenFileSet(&shared->sfs.fs, fname, O_RDONLY, false);
+ BufFileReadExact(file, &snap_size, sizeof(snap_size));
+ snap_space = (char *) palloc(snap_size);
+ BufFileReadExact(file, snap_space, snap_size);
+ BufFileClose(file);
+
+ SpinLockAcquire(&shared->mutex);
+ shared->sfs_valid = false;
+ SpinLockRelease(&shared->mutex);
+
+ /* Restore it. */
+ snapshot = RestoreSnapshot(snap_space);
+ pfree(snap_space);
+
+ return snapshot;
+}
+
+bool
+IsRepackWorker(void)
+{
+ return is_repack_worker;
+}
+
+/*
+ * Handle receipt of an interrupt indicating a repack worker message.
+ *
+ * Note: this is called within a signal handler! All we can do is set
+ * a flag that will cause the next CHECK_FOR_INTERRUPTS() to invoke
+ * ProcessRepackMessages().
+ */
+void
+HandleRepackMessageInterrupt(void)
+{
+ InterruptPending = true;
+ RepackMessagePending = true;
+ SetLatch(MyLatch);
+}
+
+/*
+ * Process any queued protocol messages received from parallel workers.
+ */
+void
+ProcessRepackMessages(void)
+{
+ MemoryContext oldcontext;
+
+ static MemoryContext hpm_context = NULL;
+
+ /*
+ * Nothing to do if we haven't launched the worker yet or have already
+ * terminated it.
+ */
+ if (decoding_worker == NULL)
+ return;
+
+ /*
+ * This is invoked from ProcessInterrupts(), and since some of the
+ * functions it calls contain CHECK_FOR_INTERRUPTS(), there is a potential
+ * for recursive calls if more signals are received while this runs. It's
+ * unclear that recursive entry would be safe, and it doesn't seem useful
+ * even if it is safe, so let's block interrupts until done.
+ */
+ HOLD_INTERRUPTS();
+
+ /*
+ * Moreover, CurrentMemoryContext might be pointing almost anywhere. We
+ * don't want to risk leaking data into long-lived contexts, so let's do
+ * our work here in a private context that we can reset on each use.
+ */
+ if (hpm_context == NULL) /* first time through? */
+ hpm_context = AllocSetContextCreate(TopMemoryContext,
+ "ProcessParallelMessages",
+ ALLOCSET_DEFAULT_SIZES);
+ else
+ MemoryContextReset(hpm_context);
+
+ oldcontext = MemoryContextSwitchTo(hpm_context);
+
+ /* OK to process messages. Reset the flag saying there are more to do. */
+ RepackMessagePending = false;
+
+ /*
+ * Read as many messages as we can from each worker, but stop when no more
+ * messages can be read from the worker without blocking.
+ */
+ while (true)
+ {
+ shm_mq_result res;
+ Size nbytes;
+ void *data;
+
+ res = shm_mq_receive(decoding_worker->error_mqh, &nbytes,
+ &data, true);
+ if (res == SHM_MQ_WOULD_BLOCK)
+ break;
+ else if (res == SHM_MQ_SUCCESS)
+ {
+ StringInfoData msg;
+
+ initStringInfo(&msg);
+ appendBinaryStringInfo(&msg, data, nbytes);
+ ProcessRepackMessage(&msg);
+ pfree(msg.data);
+ }
+ else
+ {
+ /*
+ * The decoding worker is special in that it exits as soon as it
+ * has its work done. Thus the DETACHED result code is fine.
+ */
+ Assert(res = SHM_MQ_DETACHED);
+
+ break;
+ }
+ }
+
+ MemoryContextSwitchTo(oldcontext);
+
+ /* Might as well clear the context on our way out */
+ MemoryContextReset(hpm_context);
+
+ RESUME_INTERRUPTS();
+}
+
+/*
+ * Process a single protocol message received from a single parallel worker.
+ */
+static void
+ProcessRepackMessage(StringInfo msg)
+{
+ char msgtype;
+
+ msgtype = pq_getmsgbyte(msg);
+
+ switch (msgtype)
+ {
+ case PqMsg_ErrorResponse:
+ case PqMsg_NoticeResponse:
+ {
+ ErrorData edata;
+
+ /* Parse ErrorResponse or NoticeResponse. */
+ pq_parse_errornotice(msg, &edata);
+
+ /* Death of a worker isn't enough justification for suicide. */
+ edata.elevel = Min(edata.elevel, ERROR);
+
+ /*
+ * If desired, add a context line to show that this is a
+ * message propagated from a parallel worker. Otherwise, it
+ * can sometimes be confusing to understand what actually
+ * happened.
+ */
+ if (edata.context)
+ edata.context = psprintf("%s\n%s", edata.context,
+ _("decoding worker"));
+ else
+ edata.context = pstrdup(_("decoding worker"));
+
+ /* Rethrow error or print notice. */
+ ThrowErrorData(&edata);
+
+ break;
+ }
+
+ default:
+ {
+ elog(ERROR, "unrecognized message type received from decoding worker: %c (message length %d bytes)",
+ msgtype, msg->len);
+ }
+ }
+}
diff --git a/src/backend/libpq/pqmq.c b/src/backend/libpq/pqmq.c
index 2b75de0ddef..28a5a400fb1 100644
--- a/src/backend/libpq/pqmq.c
+++ b/src/backend/libpq/pqmq.c
@@ -14,6 +14,7 @@
#include "postgres.h"
#include "access/parallel.h"
+#include "commands/cluster.h"
#include "libpq/libpq.h"
#include "libpq/pqformat.h"
#include "libpq/pqmq.h"
@@ -175,6 +176,10 @@ mq_putmessage(char msgtype, const char *s, size_t len)
SendProcSignal(pq_mq_parallel_leader_pid,
PROCSIG_PARALLEL_APPLY_MESSAGE,
pq_mq_parallel_leader_proc_number);
+ else if (IsRepackWorker())
+ SendProcSignal(pq_mq_parallel_leader_pid,
+ PROCSIG_REPACK_MESSAGE,
+ pq_mq_parallel_leader_proc_number);
else
{
Assert(IsParallelWorker());
diff --git a/src/backend/postmaster/bgworker.c b/src/backend/postmaster/bgworker.c
index 8e1068969ae..2d1ba94a45f 100644
--- a/src/backend/postmaster/bgworker.c
+++ b/src/backend/postmaster/bgworker.c
@@ -13,6 +13,7 @@
#include "postgres.h"
#include "access/parallel.h"
+#include "commands/cluster.h"
#include "libpq/pqsignal.h"
#include "miscadmin.h"
#include "pgstat.h"
@@ -135,6 +136,9 @@ static const struct
},
{
"SequenceSyncWorkerMain", SequenceSyncWorkerMain
+ },
+ {
+ "RepackWorkerMain", RepackWorkerMain
}
};
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index 1b11ed63dc6..b3fd7fec392 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -205,7 +205,11 @@ StartupDecodingContext(List *output_plugin_options,
ctx->slot = slot;
- ctx->reader = XLogReaderAllocate(wal_segment_size, NULL, xl_routine, ctx);
+ /*
+ * TODO A separate patch for PG core, unless there's really a reason to
+ * pass ctx for private_data (May extensions expect ctx?).
+ */
+ ctx->reader = XLogReaderAllocate(wal_segment_size, NULL, xl_routine, NULL);
if (!ctx->reader)
ereport(ERROR,
(errcode(ERRCODE_OUT_OF_MEMORY),
diff --git a/src/backend/replication/pgoutput_repack/pgoutput_repack.c b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
index c8930640a0d..fb9956d392d 100644
--- a/src/backend/replication/pgoutput_repack/pgoutput_repack.c
+++ b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
@@ -168,17 +168,13 @@ store_change(LogicalDecodingContext *ctx, ConcurrentChangeKind kind,
HeapTuple tuple)
{
RepackDecodingState *dstate;
- char *change_raw;
- ConcurrentChange change;
+ char kind_byte = (char) kind;
bool flattened = false;
- Size size;
- Datum values[1];
- bool isnull[1];
- char *dst;
dstate = (RepackDecodingState *) ctx->output_writer_private;
- size = VARHDRSZ + SizeOfConcurrentChange;
+ /* Store the change kind. */
+ BufFileWrite(dstate->file, &kind_byte, 1);
/*
* ReorderBufferCommit() stores the TOAST chunks in its private memory
@@ -195,46 +191,12 @@ store_change(LogicalDecodingContext *ctx, ConcurrentChangeKind kind,
tuple = toast_flatten_tuple(tuple, dstate->tupdesc);
flattened = true;
}
+ /* Store the tuple size ... */
+ BufFileWrite(dstate->file, &tuple->t_len, sizeof(tuple->t_len));
+ /* ... and the tuple itself. */
+ BufFileWrite(dstate->file, tuple->t_data, tuple->t_len);
- size += tuple->t_len;
- if (size >= MaxAllocSize)
- elog(ERROR, "Change is too big.");
-
- /* Construct the change. */
- change_raw = (char *) palloc0(size);
- SET_VARSIZE(change_raw, size);
-
- /*
- * Since the varlena alignment might not be sufficient for the structure,
- * set the fields in a local instance and remember where it should
- * eventually be copied.
- */
- change.kind = kind;
- dst = (char *) VARDATA(change_raw);
-
- /*
- * Copy the tuple.
- *
- * Note: change->tup_data.t_data must be fixed on retrieval!
- */
- memcpy(&change.tup_data, tuple, sizeof(HeapTupleData));
- memcpy(dst, &change, SizeOfConcurrentChange);
- dst += SizeOfConcurrentChange;
- memcpy(dst, tuple->t_data, tuple->t_len);
-
- /* The data has been copied. */
+ /* Free the flat copy if created above. */
if (flattened)
pfree(tuple);
-
- /* Store as tuple of 1 bytea column. */
- values[0] = PointerGetDatum(change_raw);
- isnull[0] = false;
- tuplestore_putvalues(dstate->tstore, dstate->tupdesc_change,
- values, isnull);
-
- /* Accounting. */
- dstate->nchanges++;
-
- /* Cleanup. */
- pfree(change_raw);
}
diff --git a/src/backend/storage/ipc/procsignal.c b/src/backend/storage/ipc/procsignal.c
index 087821311cc..af12144795b 100644
--- a/src/backend/storage/ipc/procsignal.c
+++ b/src/backend/storage/ipc/procsignal.c
@@ -19,6 +19,7 @@
#include "access/parallel.h"
#include "commands/async.h"
+#include "commands/cluster.h"
#include "miscadmin.h"
#include "pgstat.h"
#include "port/pg_bitutils.h"
@@ -694,6 +695,9 @@ procsignal_sigusr1_handler(SIGNAL_ARGS)
if (CheckProcSignal(PROCSIG_PARALLEL_APPLY_MESSAGE))
HandleParallelApplyMessageInterrupt();
+ if (CheckProcSignal(PROCSIG_REPACK_MESSAGE))
+ HandleRepackMessageInterrupt();
+
if (CheckProcSignal(PROCSIG_RECOVERY_CONFLICT_DATABASE))
HandleRecoveryConflictInterrupt(PROCSIG_RECOVERY_CONFLICT_DATABASE);
diff --git a/src/backend/tcop/postgres.c b/src/backend/tcop/postgres.c
index 7dd75a490aa..4a4858882f0 100644
--- a/src/backend/tcop/postgres.c
+++ b/src/backend/tcop/postgres.c
@@ -36,6 +36,7 @@
#include "access/xact.h"
#include "catalog/pg_type.h"
#include "commands/async.h"
+#include "commands/cluster.h"
#include "commands/event_trigger.h"
#include "commands/explain_state.h"
#include "commands/prepare.h"
@@ -3541,6 +3542,9 @@ ProcessInterrupts(void)
if (ParallelApplyMessagePending)
ProcessParallelApplyMessages();
+
+ if (RepackMessagePending)
+ ProcessRepackMessages();
}
/*
diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt
index f39830dbb34..cbcc8550960 100644
--- a/src/backend/utils/activity/wait_event_names.txt
+++ b/src/backend/utils/activity/wait_event_names.txt
@@ -62,6 +62,7 @@ LOGICAL_APPLY_MAIN "Waiting in main loop of logical replication apply process."
LOGICAL_LAUNCHER_MAIN "Waiting in main loop of logical replication launcher process."
LOGICAL_PARALLEL_APPLY_MAIN "Waiting in main loop of logical replication parallel apply process."
RECOVERY_WAL_STREAM "Waiting in main loop of startup process for WAL to arrive, during streaming recovery."
+REPACK_WORKER_MAIN "Waiting in main loop of REPACK decoding worker process."
REPLICATION_SLOTSYNC_MAIN "Waiting in main loop of slot sync worker."
REPLICATION_SLOTSYNC_SHUTDOWN "Waiting for slot sync worker to shut down."
SYSLOGGER_MAIN "Waiting in main loop of syslogger process."
@@ -153,6 +154,7 @@ RECOVERY_CONFLICT_SNAPSHOT "Waiting for recovery conflict resolution for a vacuu
RECOVERY_CONFLICT_TABLESPACE "Waiting for recovery conflict resolution for dropping a tablespace."
RECOVERY_END_COMMAND "Waiting for <xref linkend="guc-recovery-end-command"/> to complete."
RECOVERY_PAUSE "Waiting for recovery to be resumed."
+REPACK_WORKER_EXPORT "Waiting for decoding worker to export a new output file."
REPLICATION_ORIGIN_DROP "Waiting for a replication origin to become inactive so it can be dropped."
REPLICATION_SLOT_DROP "Waiting for a replication slot to become inactive so it can be dropped."
RESTORE_COMMAND "Waiting for <xref linkend="guc-restore-command"/> to complete."
diff --git a/src/include/access/tableam.h b/src/include/access/tableam.h
index d8f76d325f9..321c00682ec 100644
--- a/src/include/access/tableam.h
+++ b/src/include/access/tableam.h
@@ -22,7 +22,6 @@
#include "access/xact.h"
#include "commands/vacuum.h"
#include "executor/tuptable.h"
-#include "replication/logical.h"
#include "storage/read_stream.h"
#include "utils/rel.h"
#include "utils/snapshot.h"
@@ -631,7 +630,6 @@ typedef struct TableAmRoutine
bool use_sort,
TransactionId OldestXmin,
Snapshot snapshot,
- LogicalDecodingContext *decoding_ctx,
TransactionId *xid_cutoff,
MultiXactId *multi_cutoff,
double *num_tuples,
@@ -1651,8 +1649,6 @@ table_relation_copy_data(Relation rel, const RelFileLocator *newrlocator)
* - *multi_cutoff - ditto
* - snapshot - if != NULL, ignore data changes done by transactions that this
* (MVCC) snapshot considers still in-progress or in the future.
- * - decoding_ctx - logical decoding context, to capture concurrent data
- * changes.
*
* Output parameters:
* - *xid_cutoff - rel's new relfrozenxid value, may be invalid
@@ -1666,7 +1662,6 @@ table_relation_copy_for_cluster(Relation OldTable, Relation NewTable,
bool use_sort,
TransactionId OldestXmin,
Snapshot snapshot,
- LogicalDecodingContext *decoding_ctx,
TransactionId *xid_cutoff,
MultiXactId *multi_cutoff,
double *num_tuples,
@@ -1675,7 +1670,7 @@ table_relation_copy_for_cluster(Relation OldTable, Relation NewTable,
{
OldTable->rd_tableam->relation_copy_for_cluster(OldTable, NewTable, OldIndex,
use_sort, OldestXmin,
- snapshot, decoding_ctx,
+ snapshot,
xid_cutoff, multi_cutoff,
num_tuples, tups_vacuumed,
tups_recently_dead);
diff --git a/src/include/commands/cluster.h b/src/include/commands/cluster.h
index b43a1740053..0ac70ec30d7 100644
--- a/src/include/commands/cluster.h
+++ b/src/include/commands/cluster.h
@@ -17,6 +17,7 @@
#include "nodes/parsenodes.h"
#include "parser/parse_node.h"
#include "replication/logical.h"
+#include "storage/buffile.h"
#include "storage/lock.h"
#include "storage/relfilelocator.h"
#include "utils/relcache.h"
@@ -47,6 +48,9 @@ typedef struct ClusterParams
extern RelFileLocator repacked_rel_locator;
extern RelFileLocator repacked_rel_toast_locator;
+/*
+ * Stored as a single byte in the output file.
+ */
typedef enum
{
CHANGE_INSERT,
@@ -55,68 +59,30 @@ typedef enum
CHANGE_DELETE
} ConcurrentChangeKind;
-typedef struct ConcurrentChange
-{
- /* See the enum above. */
- ConcurrentChangeKind kind;
-
- /*
- * The actual tuple.
- *
- * The tuple data follows the ConcurrentChange structure. Before use make
- * sure the tuple is correctly aligned (ConcurrentChange can be stored as
- * bytea) and that tuple->t_data is fixed.
- */
- HeapTupleData tup_data;
-} ConcurrentChange;
-
-#define SizeOfConcurrentChange (offsetof(ConcurrentChange, tup_data) + \
- sizeof(HeapTupleData))
-
/*
* Logical decoding state.
*
- * Here we store the data changes that we decode from WAL while the table
- * contents is being copied to a new storage. Also the necessary metadata
- * needed to apply these changes to the table is stored here.
+ * The output plugin uses it to store the data changes that it decodes from
+ * WAL while the table contents is being copied to a new storage.
*/
typedef struct RepackDecodingState
{
/* The relation whose changes we're decoding. */
Oid relid;
- /* Replication slot name. */
- NameData slotname;
-
- /*
- * Decoded changes are stored here. Although we try to avoid excessive
- * batches, it can happen that the changes need to be stored to disk. The
- * tuplestore does this transparently.
- */
- Tuplestorestate *tstore;
-
- /* The current number of changes in tstore. */
- double nchanges;
-
- /*
- * Descriptor to store the ConcurrentChange structure serialized (bytea).
- * We can't store the tuple directly because tuplestore only supports
- * minimum tuple and we may need to transfer OID system column from the
- * output plugin. Also we need to transfer the change kind, so it's better
- * to put everything in the structure than to use 2 tuplestores "in
- * parallel".
- */
- TupleDesc tupdesc_change;
-
- /* Tuple descriptor needed to update indexes. */
+ /* Tuple descriptor of the relation being processed. */
TupleDesc tupdesc;
- /* Slot to retrieve data from tstore. */
- TupleTableSlot *tsslot;
-
- ResourceOwner resowner;
+ /* The current output file. */
+ BufFile *file;
} RepackDecodingState;
+extern PGDLLIMPORT volatile sig_atomic_t RepackMessagePending;
+
+extern bool IsRepackWorker(void);
+extern void HandleRepackMessageInterrupt(void);
+extern void ProcessRepackMessages(void);
+
extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel);
extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid,
@@ -125,9 +91,6 @@ extern void check_index_is_clusterable(Relation OldHeap, Oid indexOid,
LOCKMODE lockmode);
extern void mark_index_clustered(Relation rel, Oid indexOid, bool is_internal);
-extern void repack_decode_concurrent_changes(LogicalDecodingContext *ctx,
- XLogRecPtr end_of_wal);
-
extern Oid make_new_heap(Oid OIDOldHeap, Oid NewTableSpace, Oid NewAccessMethod,
char relpersistence, LOCKMODE lockmode);
extern void finish_heap_swap(Oid OIDOldHeap, Oid OIDNewHeap,
@@ -140,4 +103,5 @@ extern void finish_heap_swap(Oid OIDOldHeap, Oid OIDNewHeap,
MultiXactId cutoffMulti,
char newrelpersistence);
+extern void RepackWorkerMain(Datum main_arg);
#endif /* CLUSTER_H */
diff --git a/src/include/storage/procsignal.h b/src/include/storage/procsignal.h
index afeeb1ca019..c0a66516b66 100644
--- a/src/include/storage/procsignal.h
+++ b/src/include/storage/procsignal.h
@@ -36,6 +36,7 @@ typedef enum
PROCSIG_BARRIER, /* global barrier interrupt */
PROCSIG_LOG_MEMORY_CONTEXT, /* ask backend to log the memory contexts */
PROCSIG_PARALLEL_APPLY_MESSAGE, /* Message from parallel apply workers */
+ PROCSIG_REPACK_MESSAGE, /* Message from repack worker */
/* Recovery conflict reasons */
PROCSIG_RECOVERY_CONFLICT_FIRST,
diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list
index 3139b14e85f..35344910f65 100644
--- a/src/tools/pgindent/typedefs.list
+++ b/src/tools/pgindent/typedefs.list
@@ -488,7 +488,6 @@ CompressFileHandle
CompressionLocation
CompressorState
ComputeXidHorizonsResult
-ConcurrentChange
ConcurrentChangeKind
ConditionVariable
ConditionVariableMinimallyPadded
@@ -629,6 +628,9 @@ DeclareCursorStmt
DecodedBkpBlock
DecodedXLogRecord
DecodingOutputState
+DecodingWorker
+DecodingWorkerShared
+DecodingWorkerState
DefElem
DefElemAction
DefaultACLInfo
--
2.47.3
--=-=-=
Content-Type: text/plain
Content-Disposition: attachment;
filename=v28-0006-Use-multiple-snapshots-to-copy-the-data.patch
^ permalink raw reply [nested|flat] 55+ messages in thread
* [PATCH 5/6] Use background worker to do logical decoding.
@ 2026-01-08 16:47 Antonin Houska <[email protected]>
0 siblings, 0 replies; 55+ messages in thread
From: Antonin Houska @ 2026-01-08 16:47 UTC (permalink / raw)
If the backend performing REPACK (CONCURRENTLY) does both data copying and
logical decoding, it has to "travel in time" back and forth and therefore it
has to invalidate system caches quite a few times. (The copying and the
decoding work with different catalog snapshots.) As the decoding worker has
separate caches, the switching is not necessary.
Without the worker, it'd also be difficult to switch between potentiallly long
running tasks like index build and WAL decoding. (No decoding during that time
at all can suspend archiving / recycling of WAL segments for some time, which
in turn may result in full disk.)
Another problem is that, after having acquired AccessExclusiveLock (in order
to swap the files), the backend needs to both decode and apply the data
changes that took place while it was waiting for the lock. With the decoding
worker, the decoding runs all the time, so the backend only needs to apply the
changes. This can reduce the time the exclusive lock is held for.
Note that the code added in order to handle ERRORs in the background worker
almost duplicates the existing code that does the same for other types of
workers (See ProcessParallelMessages() and
ProcessParallelApplyMessages()). Refactoring of the existing code might be
useful, to reduce the duplication.
---
src/backend/access/heap/heapam_handler.c | 44 -
src/backend/commands/cluster.c | 1174 +++++++++++++----
src/backend/libpq/pqmq.c | 5 +
src/backend/postmaster/bgworker.c | 4 +
src/backend/replication/logical/logical.c | 6 +-
.../pgoutput_repack/pgoutput_repack.c | 54 +-
src/backend/storage/ipc/procsignal.c | 4 +
src/backend/tcop/postgres.c | 4 +
.../utils/activity/wait_event_names.txt | 2 +
src/include/access/tableam.h | 7 +-
src/include/commands/cluster.h | 71 +-
src/include/storage/procsignal.h | 1 +
src/tools/pgindent/typedefs.list | 4 +-
13 files changed, 979 insertions(+), 401 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index 3526b6adcb5..475c536ce43 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -33,7 +33,6 @@
#include "catalog/index.h"
#include "catalog/storage.h"
#include "catalog/storage_xlog.h"
-#include "commands/cluster.h"
#include "commands/progress.h"
#include "executor/executor.h"
#include "miscadmin.h"
@@ -688,7 +687,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
Relation OldIndex, bool use_sort,
TransactionId OldestXmin,
Snapshot snapshot,
- LogicalDecodingContext *decoding_ctx,
TransactionId *xid_cutoff,
MultiXactId *multi_cutoff,
double *num_tuples,
@@ -710,7 +708,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
BufferHeapTupleTableSlot *hslot;
BlockNumber prev_cblock = InvalidBlockNumber;
bool concurrent = snapshot != NULL;
- XLogRecPtr end_of_wal_prev = GetFlushRecPtr(NULL);
/* Remember if it's a system catalog */
is_system_catalog = IsSystemRelation(OldHeap);
@@ -957,31 +954,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
ct_val[1] = *num_tuples;
pgstat_progress_update_multi_param(2, ct_index, ct_val);
}
-
- /*
- * Process the WAL produced by the load, as well as by other
- * transactions, so that the replication slot can advance and WAL does
- * not pile up. Use wal_segment_size as a threshold so that we do not
- * introduce the decoding overhead too often.
- *
- * Of course, we must not apply the changes until the initial load has
- * completed.
- *
- * Note that our insertions into the new table should not be decoded
- * as we (intentionally) do not write the logical decoding specific
- * information to WAL.
- */
- if (concurrent)
- {
- XLogRecPtr end_of_wal;
-
- end_of_wal = GetFlushRecPtr(NULL);
- if ((end_of_wal - end_of_wal_prev) > wal_segment_size)
- {
- repack_decode_concurrent_changes(decoding_ctx, end_of_wal);
- end_of_wal_prev = end_of_wal;
- }
- }
}
if (indexScan != NULL)
@@ -1027,22 +999,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
-
- /*
- * Try to keep the amount of not-yet-decoded WAL small, like
- * above.
- */
- if (concurrent)
- {
- XLogRecPtr end_of_wal;
-
- end_of_wal = GetFlushRecPtr(NULL);
- if ((end_of_wal - end_of_wal_prev) > wal_segment_size)
- {
- repack_decode_concurrent_changes(decoding_ctx, end_of_wal);
- end_of_wal_prev = end_of_wal;
- }
- }
}
tuplesort_end(tuplesort);
diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c
index c3feb0c3de4..5232fbfb57d 100644
--- a/src/backend/commands/cluster.c
+++ b/src/backend/commands/cluster.c
@@ -12,12 +12,13 @@
* In concurrent mode, we lock the table with only ShareUpdateExclusiveLock,
* then do an initial copy as above. However, while the tuples are being
* copied, concurrent transactions could modify the table. To cope with those
- * changes, we rely on logical decoding to obtain them from WAL. The changes
- * are accumulated in a tuplestore. Once the initial copy is complete, we
- * read the changes from the tuplestore and re-apply them on the new heap.
- * Then we upgrade our ShareUpdateExclusiveLock to AccessExclusiveLock and
- * swap the relfilenodes. This way, the time we hold a strong lock on the
- * table is much reduced, and the bloat is eliminated.
+ * changes, we rely on logical decoding to obtain them from WAL. A bgworker
+ * consumes WAL while the initial copy is ongoing (to prevent excessive WAL
+ * from being reserved), and accumulates the changes in a file. Once the
+ * initial copy is complete, we read the changes from the file and re-apply
+ * them on the new heap. Then we upgrade our ShareUpdateExclusiveLock to
+ * AccessExclusiveLock and swap the relfilenodes. This way, the time we hold
+ * a strong lock on the table is much reduced, and the bloat is eliminated.
*
* There is hardly anything left of Paul Brown's original implementation...
*
@@ -45,6 +46,7 @@
#include "access/xlog_internal.h"
#include "access/xloginsert.h"
#include "access/xlogutils.h"
+#include "access/xlogwait.h"
#include "catalog/catalog.h"
#include "catalog/dependency.h"
#include "catalog/heap.h"
@@ -61,6 +63,8 @@
#include "commands/tablecmds.h"
#include "commands/vacuum.h"
#include "executor/executor.h"
+#include "libpq/pqformat.h"
+#include "libpq/pqmq.h"
#include "miscadmin.h"
#include "optimizer/optimizer.h"
#include "pgstat.h"
@@ -71,6 +75,8 @@
#include "storage/ipc.h"
#include "storage/lmgr.h"
#include "storage/predicate.h"
+#include "storage/procsignal.h"
+#include "tcop/tcopprot.h"
#include "utils/acl.h"
#include "utils/fmgroids.h"
#include "utils/guc.h"
@@ -117,6 +123,12 @@ typedef struct IndexInsertState
/* The WAL segment being decoded. */
static XLogSegNo repack_current_segment = 0;
+/*
+ * The first file exported by the decoding worker must contain a snapshot, the
+ * following ones contain the data changes.
+ */
+#define WORKER_FILE_SNAPSHOT 0
+
/*
* Information needed to apply concurrent data changes.
*/
@@ -136,8 +148,113 @@ typedef struct ChangeDest
/* Needed to update indexes of rel_dst. */
IndexInsertState *iistate;
+
+ /*
+ * Sequential number of the file containing the changes.
+ *
+ * TODO This field makes the structure name less descriptive. Should we
+ * rename it, e.g. to ChangeApplyInfo?
+ */
+ int file_seq;
} ChangeDest;
+/*
+ * Layout of shared memory used for communication between backend and the
+ * worker that performs logical decoding of data changes
+ */
+typedef struct DecodingWorkerShared
+{
+ /* Is the decoding initialized? */
+ bool initialized;
+
+ /*
+ * Once the worker has reached this LSN, it should close the current
+ * output file and either create a new one or exit, according to the field
+ * 'done'. If the value is InvalidXLogRecPtr, the worker should decode all
+ * the WAL available and keep checking this field. It is ok if the worker
+ * had already decoded records whose LSN is >= lsn_upto before this field
+ * has been set.
+ */
+ XLogRecPtr lsn_upto;
+
+ /* Exit after closing the current file? */
+ bool done;
+
+ /* The output is stored here. */
+ SharedFileSet sfs;
+
+ /* Number of the last file exported by the worker. */
+ int last_exported;
+
+ /* Synchronize access to the fields above. */
+ slock_t mutex;
+
+ /* Database to connect to. */
+ Oid dbid;
+
+ /* Role to connect as. */
+ Oid roleid;
+
+ /* Decode data changes of this relation. */
+ Oid relid;
+
+ /* The backend uses this to wait for the worker. */
+ ConditionVariable cv;
+
+ /* Info to signal the backend. */
+ PGPROC *backend_proc;
+ pid_t backend_pid;
+ ProcNumber backend_proc_number;
+
+ /* Error queue. */
+ shm_mq *error_mq;
+
+ /*
+ * Memory the queue is located int.
+ *
+ * For considerations on the value see the comments of
+ * PARALLEL_ERROR_QUEUE_SIZE.
+ */
+#define REPACK_ERROR_QUEUE_SIZE 16384
+ char error_queue[FLEXIBLE_ARRAY_MEMBER];
+} DecodingWorkerShared;
+
+/*
+ * Generate worker's output file name. If relations of the same 'relid' happen
+ * to be processed at the same time, they must be from different databases and
+ * therefore different backends must be involved. (PID is already present in
+ * the fileset name.)
+ */
+static inline void
+DecodingWorkerFileName(char *fname, Oid relid, uint32 seq)
+{
+ snprintf(fname, MAXPGPATH, "%u-%u", relid, seq);
+}
+
+/*
+ * Backend-local information to control the decoding worker.
+ */
+typedef struct DecodingWorker
+{
+ /* The worker. */
+ BackgroundWorkerHandle *handle;
+
+ /* DecodingWorkerShared is in this segment. */
+ dsm_segment *seg;
+
+ /* Handle of the error queue. */
+ shm_mq_handle *error_mqh;
+} DecodingWorker;
+
+/* Pointer to currently running decoding worker. */
+static DecodingWorker *decoding_worker = NULL;
+
+/*
+ * Is there a message sent by a repack worker that the backend needs to
+ * receive?
+ */
+volatile sig_atomic_t RepackMessagePending = false;
+
static bool cluster_rel_recheck(RepackCommand cmd, Relation OldHeap,
Oid indexOid, Oid userid, LOCKMODE lmode,
int options);
@@ -145,7 +262,7 @@ static void check_repack_concurrently_requirements(Relation rel);
static void rebuild_relation(Relation OldHeap, Relation index, bool verbose,
bool concurrent);
static void copy_table_data(Relation NewHeap, Relation OldHeap, Relation OldIndex,
- Snapshot snapshot, LogicalDecodingContext *decoding_ctx,
+ Snapshot snapshot,
bool verbose,
bool *pSwapToastByContent,
TransactionId *pFreezeXid,
@@ -158,12 +275,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd,
static bool cluster_is_permitted_for_relation(RepackCommand cmd,
Oid relid, Oid userid);
-static void begin_concurrent_repack(Relation rel);
-static void end_concurrent_repack(void);
static LogicalDecodingContext *setup_logical_decoding(Oid relid);
-static HeapTuple get_changed_tuple(char *change);
-static void apply_concurrent_changes(RepackDecodingState *dstate,
- ChangeDest *dest);
+static bool decode_concurrent_changes(LogicalDecodingContext *ctx,
+ DecodingWorkerShared *shared);
+static void apply_concurrent_changes(BufFile *file, ChangeDest *dest);
static void apply_concurrent_insert(Relation rel, HeapTuple tup,
IndexInsertState *iistate,
TupleTableSlot *index_slot);
@@ -175,9 +290,9 @@ static void apply_concurrent_delete(Relation rel, HeapTuple tup_target);
static HeapTuple find_target_tuple(Relation rel, ChangeDest *dest,
HeapTuple tup_key,
TupleTableSlot *ident_slot);
-static void process_concurrent_changes(LogicalDecodingContext *decoding_ctx,
- XLogRecPtr end_of_wal,
- ChangeDest *dest);
+static void process_concurrent_changes(XLogRecPtr end_of_wal,
+ ChangeDest *dest,
+ bool done);
static IndexInsertState *get_index_insert_state(Relation relation,
Oid ident_index_id,
Relation *ident_index_p);
@@ -187,7 +302,6 @@ static void free_index_insert_state(IndexInsertState *iistate);
static void cleanup_logical_decoding(LogicalDecodingContext *ctx);
static void rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
Relation cl_index,
- LogicalDecodingContext *decoding_ctx,
TransactionId frozenXid,
MultiXactId cutoffMulti);
static List *build_new_indexes(Relation NewHeap, Relation OldHeap, List *OldIndexes);
@@ -197,6 +311,13 @@ static Relation process_single_relation(RepackStmt *stmt,
ClusterParams *params);
static Oid determine_clustered_index(Relation rel, bool usingindex,
const char *indexname);
+static void start_decoding_worker(Oid relid);
+static void stop_decoding_worker(void);
+static void repack_worker_internal(dsm_segment *seg);
+static void export_initial_snapshot(Snapshot snapshot,
+ DecodingWorkerShared *shared);
+static Snapshot get_initial_snapshot(DecodingWorker *worker);
+static void ProcessRepackMessage(StringInfo msg);
static const char *RepackCommandAsString(RepackCommand cmd);
@@ -619,20 +740,20 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid,
/* rebuild_relation does all the dirty work */
PG_TRY();
{
- /*
- * For concurrent processing, make sure that our logical decoding
- * ignores data changes of other tables than the one we are
- * processing.
- */
- if (concurrent)
- begin_concurrent_repack(OldHeap);
-
rebuild_relation(OldHeap, index, verbose, concurrent);
}
PG_FINALLY();
{
if (concurrent)
- end_concurrent_repack();
+ {
+ /*
+ * Since during normal operation the worker was already asked to
+ * exit, stopping it explicitly is especially important on ERROR.
+ * However it still seems a good practice to make sure that the
+ * worker never survives the REPACK command.
+ */
+ stop_decoding_worker();
+ }
}
PG_END_TRY();
@@ -929,7 +1050,6 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
bool swap_toast_by_content;
TransactionId frozenXid;
MultiXactId cutoffMulti;
- LogicalDecodingContext *decoding_ctx = NULL;
Snapshot snapshot = NULL;
#if USE_ASSERT_CHECKING
LOCKMODE lmode;
@@ -943,19 +1063,36 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
if (concurrent)
{
/*
- * Prepare to capture the concurrent data changes.
+ * The worker needs to be member of the locking group we're the leader
+ * of. We ought to become the leader before the worker starts. The
+ * worker will join the group as soon as it starts.
*
- * Note that this call waits for all transactions with XID already
- * assigned to finish. If some of those transactions is waiting for a
- * lock conflicting with ShareUpdateExclusiveLock on our table (e.g.
- * it runs CREATE INDEX), we can end up in a deadlock. Not sure this
- * risk is worth unlocking/locking the table (and its clustering
- * index) and checking again if its still eligible for REPACK
- * CONCURRENTLY.
+ * This is to make sure that the deadlock described below is
+ * detectable by deadlock.c: if the worker waits for a transaction to
+ * complete and we are waiting for the worker output, then effectively
+ * we (i.e. this backend) are waiting for that transaction.
*/
- decoding_ctx = setup_logical_decoding(tableOid);
+ BecomeLockGroupLeader();
+
+ /*
+ * Start the worker that decodes data changes applied while we're
+ * copying the table contents.
+ *
+ * Note that the worker has to wait for all transactions with XID
+ * already assigned to finish. If some of those transactions is
+ * waiting for a lock conflicting with ShareUpdateExclusiveLock on our
+ * table (e.g. it runs CREATE INDEX), we can end up in a deadlock.
+ * Not sure this risk is worth unlocking/locking the table (and its
+ * clustering index) and checking again if its still eligible for
+ * REPACK CONCURRENTLY.
+ */
+ start_decoding_worker(tableOid);
+
+ /*
+ * Wait until the worker has the initial snapshot and retrieve it.
+ */
+ snapshot = get_initial_snapshot(decoding_worker);
- snapshot = SnapBuildInitialSnapshotForRepack(decoding_ctx->snapshot_builder);
PushActiveSnapshot(snapshot);
}
@@ -980,7 +1117,7 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
NewHeap = table_open(OIDNewHeap, NoLock);
/* Copy the heap data into the new table in the desired order */
- copy_table_data(NewHeap, OldHeap, index, snapshot, decoding_ctx, verbose,
+ copy_table_data(NewHeap, OldHeap, index, snapshot, verbose,
&swap_toast_by_content, &frozenXid, &cutoffMulti);
/* The historic snapshot won't be needed anymore. */
@@ -994,14 +1131,10 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
{
Assert(!swap_toast_by_content);
rebuild_relation_finish_concurrent(NewHeap, OldHeap, index,
- decoding_ctx,
frozenXid, cutoffMulti);
pgstat_progress_update_param(PROGRESS_REPACK_PHASE,
PROGRESS_REPACK_PHASE_FINAL_CLEANUP);
-
- /* Done with decoding. */
- cleanup_logical_decoding(decoding_ctx);
}
else
{
@@ -1172,8 +1305,7 @@ make_new_heap(Oid OIDOldHeap, Oid NewTableSpace, Oid NewAccessMethod,
*/
static void
copy_table_data(Relation NewHeap, Relation OldHeap, Relation OldIndex,
- Snapshot snapshot, LogicalDecodingContext *decoding_ctx,
- bool verbose, bool *pSwapToastByContent,
+ Snapshot snapshot, bool verbose, bool *pSwapToastByContent,
TransactionId *pFreezeXid, MultiXactId *pCutoffMulti)
{
Relation relRelation;
@@ -1334,7 +1466,6 @@ copy_table_data(Relation NewHeap, Relation OldHeap, Relation OldIndex,
*/
table_relation_copy_for_cluster(OldHeap, NewHeap, OldIndex, use_sort,
cutoffs.OldestXmin, snapshot,
- decoding_ctx,
&cutoffs.FreezeLimit,
&cutoffs.MultiXactCutoff,
&num_tuples, &tups_vacuumed,
@@ -2367,59 +2498,6 @@ RepackCommandAsString(RepackCommand cmd)
return "???";
}
-
-/*
- * Call this function before REPACK CONCURRENTLY starts to setup logical
- * decoding. It makes sure that other users of the table put enough
- * information into WAL.
- *
- * The point is that at various places we expect that the table we're
- * processing is treated like a system catalog. For example, we need to be
- * able to scan it using a "historic snapshot" anytime during the processing
- * (as opposed to scanning only at the start point of the decoding, as logical
- * replication does during initial table synchronization), in order to apply
- * concurrent UPDATE / DELETE commands.
- *
- * Note that TOAST table needs no attention here as it's not scanned using
- * historic snapshot.
- */
-static void
-begin_concurrent_repack(Relation rel)
-{
- Oid toastrelid;
-
- /*
- * Avoid logical decoding of other relations by this backend. The lock we
- * have guarantees that the actual locator cannot be changed concurrently:
- * TRUNCATE needs AccessExclusiveLock.
- */
- Assert(CheckRelationLockedByMe(rel, ShareUpdateExclusiveLock, false));
- repacked_rel_locator = rel->rd_locator;
- toastrelid = rel->rd_rel->reltoastrelid;
- if (OidIsValid(toastrelid))
- {
- Relation toastrel;
-
- /* Avoid logical decoding of other TOAST relations. */
- toastrel = table_open(toastrelid, AccessShareLock);
- repacked_rel_toast_locator = toastrel->rd_locator;
- table_close(toastrel, AccessShareLock);
- }
-}
-
-/*
- * Call this when done with REPACK CONCURRENTLY.
- */
-static void
-end_concurrent_repack(void)
-{
- /*
- * Restore normal function of (future) logical decoding for this backend.
- */
- repacked_rel_locator.relNumber = InvalidOid;
- repacked_rel_toast_locator.relNumber = InvalidOid;
-}
-
/*
* Is this backend performing logical decoding on behalf of REPACK
* (CONCURRENTLY) ?
@@ -2484,9 +2562,10 @@ static LogicalDecodingContext *
setup_logical_decoding(Oid relid)
{
Relation rel;
- TupleDesc tupdesc;
+ Oid toastrelid;
LogicalDecodingContext *ctx;
- RepackDecodingState *dstate = palloc0_object(RepackDecodingState);
+ NameData slotname;
+ RepackDecodingState *dstate;
/*
* REPACK CONCURRENTLY is not allowed in a transaction block, so this
@@ -2494,21 +2573,21 @@ setup_logical_decoding(Oid relid)
*/
Assert(!TransactionIdIsValid(GetTopTransactionIdIfAny()));
- /*
- * A single backend should not execute multiple REPACK commands at a time,
- * so use PID to make the slot unique.
- */
- snprintf(NameStr(dstate->slotname), NAMEDATALEN, "repack_%d", MyProcPid);
-
/*
* Check if we can use logical decoding.
*/
CheckSlotPermissions();
CheckLogicalDecodingRequirements();
- /* RS_TEMPORARY so that the slot gets cleaned up on ERROR. */
- ReplicationSlotCreate(NameStr(dstate->slotname), true, RS_TEMPORARY,
- false, false, false);
+ /*
+ * A single backend should not execute multiple REPACK commands at a time,
+ * so use PID to make the slot unique.
+ *
+ * RS_TEMPORARY so that the slot gets cleaned up on ERROR.
+ */
+ snprintf(NameStr(slotname), NAMEDATALEN, "repack_%d", MyProcPid);
+ ReplicationSlotCreate(NameStr(slotname), true, RS_TEMPORARY, false, false,
+ false);
/*
* Neither prepare_write nor do_write callback nor update_progress is
@@ -2530,104 +2609,109 @@ setup_logical_decoding(Oid relid)
DecodingContextFindStartpoint(ctx);
+ /*
+ * decode_concurrent_changes() needs non-blocking callback.
+ */
+ ctx->reader->routine.page_read = read_local_xlog_page_no_wait;
+
+ /*
+ * read_local_xlog_page_no_wait() needs to be able to indicate the end of
+ * WAL.
+ */
+ ctx->reader->private_data = MemoryContextAllocZero(ctx->context,
+ sizeof(ReadLocalXLogPageNoWaitPrivate));
+
+
/* Some WAL records should have been read. */
Assert(ctx->reader->EndRecPtr != InvalidXLogRecPtr);
+ /*
+ * Initialize repack_current_segment so that we can notice WAL segment
+ * boundaries.
+ */
XLByteToSeg(ctx->reader->EndRecPtr, repack_current_segment,
wal_segment_size);
- /*
- * Setup structures to store decoded changes.
- */
+ dstate = palloc0_object(RepackDecodingState);
dstate->relid = relid;
- dstate->tstore = tuplestore_begin_heap(false, false,
- maintenance_work_mem);
- /* Caller should already have the table locked. */
- rel = table_open(relid, NoLock);
- tupdesc = CreateTupleDescCopy(RelationGetDescr(rel));
- dstate->tupdesc = tupdesc;
- table_close(rel, NoLock);
+ /*
+ * Tuple descriptor may be needed to flatten a tuple before we write it to
+ * a file. A copy is needed because the decoding worker invalidates system
+ * caches before it starts to do the actual work.
+ */
+ rel = table_open(relid, AccessShareLock);
+ dstate->tupdesc = CreateTupleDescCopy(RelationGetDescr(rel));
- /* Initialize the descriptor to store the changes ... */
- dstate->tupdesc_change = CreateTemplateTupleDesc(1);
+ /* Avoid logical decoding of other relations. */
+ repacked_rel_locator = rel->rd_locator;
+ toastrelid = rel->rd_rel->reltoastrelid;
+ if (OidIsValid(toastrelid))
+ {
+ Relation toastrel;
- TupleDescInitEntry(dstate->tupdesc_change, 1, NULL, BYTEAOID, -1, 0);
- /* ... as well as the corresponding slot. */
- dstate->tsslot = MakeSingleTupleTableSlot(dstate->tupdesc_change,
- &TTSOpsMinimalTuple);
+ /* Avoid logical decoding of other TOAST relations. */
+ toastrel = table_open(toastrelid, AccessShareLock);
+ repacked_rel_toast_locator = toastrel->rd_locator;
+ table_close(toastrel, AccessShareLock);
+ }
+ table_close(rel, AccessShareLock);
- dstate->resowner = ResourceOwnerCreate(CurrentResourceOwner,
- "logical decoding");
+ /* The file will be set as soon as we have it opened. */
+ dstate->file = NULL;
ctx->output_writer_private = dstate;
+
return ctx;
}
/*
- * Retrieve tuple from ConcurrentChange structure.
+ * Decode logical changes from the WAL sequence and store them to a file.
*
- * The input data starts with the structure but it might not be appropriately
- * aligned.
- */
-static HeapTuple
-get_changed_tuple(char *change)
-{
- HeapTupleData tup_data;
- HeapTuple result;
- char *src;
-
- /*
- * Ensure alignment before accessing the fields. (This is why we can't use
- * heap_copytuple() instead of this function.)
- */
- src = change + offsetof(ConcurrentChange, tup_data);
- memcpy(&tup_data, src, sizeof(HeapTupleData));
-
- result = (HeapTuple) palloc(HEAPTUPLESIZE + tup_data.t_len);
- memcpy(result, &tup_data, sizeof(HeapTupleData));
- result->t_data = (HeapTupleHeader) ((char *) result + HEAPTUPLESIZE);
- src = change + SizeOfConcurrentChange;
- memcpy(result->t_data, src, result->t_len);
-
- return result;
-}
-
-/*
- * Decode logical changes from the WAL sequence up to end_of_wal.
+ * If true is returned, there is no more work for the worker.
*/
-void
-repack_decode_concurrent_changes(LogicalDecodingContext *ctx,
- XLogRecPtr end_of_wal)
+static bool
+decode_concurrent_changes(LogicalDecodingContext *ctx,
+ DecodingWorkerShared *shared)
{
RepackDecodingState *dstate;
- ResourceOwner resowner_old;
+ XLogRecPtr lsn_upto;
+ bool done;
+ char fname[MAXPGPATH];
dstate = (RepackDecodingState *) ctx->output_writer_private;
- resowner_old = CurrentResourceOwner;
- CurrentResourceOwner = dstate->resowner;
- PG_TRY();
+ /* Open the output file. */
+ DecodingWorkerFileName(fname, shared->relid, shared->last_exported + 1);
+ dstate->file = BufFileCreateFileSet(&shared->sfs.fs, fname);
+
+ SpinLockAcquire(&shared->mutex);
+ lsn_upto = shared->lsn_upto;
+ done = shared->done;
+ SpinLockRelease(&shared->mutex);
+
+ while (true)
{
- while (ctx->reader->EndRecPtr < end_of_wal)
- {
- XLogRecord *record;
- XLogSegNo segno_new;
- char *errm = NULL;
- XLogRecPtr end_lsn;
+ XLogRecord *record;
+ XLogSegNo segno_new;
+ char *errm = NULL;
+ XLogRecPtr end_lsn;
- record = XLogReadRecord(ctx->reader, &errm);
- if (errm)
- elog(ERROR, "%s", errm);
+ CHECK_FOR_INTERRUPTS();
- if (record != NULL)
- LogicalDecodingProcessRecord(ctx, ctx->reader);
+ record = XLogReadRecord(ctx->reader, &errm);
+ if (record)
+ {
+ LogicalDecodingProcessRecord(ctx, ctx->reader);
/*
* If WAL segment boundary has been crossed, inform the decoding
- * system that the catalog_xmin can advance. (We can confirm more
- * often, but a filling a single WAL segment should not take much
- * time.)
+ * system that the catalog_xmin can advance.
+ *
+ * TODO Does it make sense to confirm more often? Segment size
+ * seems appropriate for restart_lsn (because less than a segment
+ * cannot be recycled anyway), however more frequent checks might
+ * be beneficial for catalog_xmin.
*/
end_lsn = ctx->reader->EndRecPtr;
XLByteToSeg(end_lsn, segno_new, wal_segment_size);
@@ -2638,80 +2722,137 @@ repack_decode_concurrent_changes(LogicalDecodingContext *ctx,
(uint32) (end_lsn >> 32), (uint32) end_lsn);
repack_current_segment = segno_new;
}
+ }
+ else
+ {
+ ReadLocalXLogPageNoWaitPrivate *priv;
- CHECK_FOR_INTERRUPTS();
+ if (errm)
+ ereport(ERROR, (errmsg("%s", errm)));
+
+ /*
+ * In the decoding loop we do not want to get blocked when there
+ * is no more WAL available, otherwise the loop would become
+ * uninterruptible.
+ */
+ priv = (ReadLocalXLogPageNoWaitPrivate *)
+ ctx->reader->private_data;
+ if (priv->end_of_wal)
+ /* Do not miss the end of WAL condition next time. */
+ priv->end_of_wal = false;
+ else
+ ereport(ERROR, (errmsg("could not read WAL record")));
+ }
+
+ /*
+ * Whether we could read new record or not, keep checking if
+ * 'lsn_upto' was specified.
+ */
+ if (XLogRecPtrIsInvalid(lsn_upto))
+ {
+ SpinLockAcquire(&shared->mutex);
+ lsn_upto = shared->lsn_upto;
+ /* 'done' should be set at the same time as 'lsn_upto' */
+ done = shared->done;
+ SpinLockRelease(&shared->mutex);
+ }
+ if (!XLogRecPtrIsInvalid(lsn_upto) &&
+ ctx->reader->EndRecPtr >= lsn_upto)
+ break;
+
+ if (record == NULL)
+ {
+ int64 timeout = 0;
+ WaitLSNResult res;
+
+ /*
+ * Before we retry reading, wait until new WAL is flushed.
+ *
+ * There is a race condition such that the backend executing
+ * REPACK determines 'lsn_upto', but before it sets the shared
+ * variable, we reach the end of WAL. In that case we'd need to
+ * wait until the next WAL flush (unrelated to REPACK). Although
+ * that should not be a problem in a busy system, it might be
+ * noticeable in other cases, including regression tests (which
+ * are not necessarily executed in parallel). Therefore it makes
+ * sense to use timeout.
+ *
+ * If lsn_upto is valid, WAL records having LSN lower than that
+ * should already have been flushed to disk.
+ */
+ if (XLogRecPtrIsInvalid(lsn_upto))
+ timeout = 100L;
+ res = WaitForLSN(WAIT_LSN_TYPE_PRIMARY_FLUSH,
+ ctx->reader->EndRecPtr + 1,
+ timeout);
+ if (res != WAIT_LSN_RESULT_SUCCESS &&
+ res != WAIT_LSN_RESULT_TIMEOUT)
+ ereport(ERROR, (errmsg("waiting for WAL failed")));
}
- InvalidateSystemCaches();
- CurrentResourceOwner = resowner_old;
- }
- PG_CATCH();
- {
- /* clear all timetravel entries */
- InvalidateSystemCaches();
- CurrentResourceOwner = resowner_old;
- PG_RE_THROW();
}
- PG_END_TRY();
+
+ /*
+ * Close the file so we can make it available to the backend.
+ */
+ BufFileClose(dstate->file);
+ dstate->file = NULL;
+ SpinLockAcquire(&shared->mutex);
+ shared->lsn_upto = InvalidXLogRecPtr;
+ shared->last_exported++;
+ SpinLockRelease(&shared->mutex);
+ ConditionVariableSignal(&shared->cv);
+
+ return done;
}
/*
* Apply changes stored in 'file'.
*/
static void
-apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
+apply_concurrent_changes(BufFile *file, ChangeDest *dest)
{
+ char kind;
+ uint32 t_len;
Relation rel = dest->rel;
TupleTableSlot *index_slot,
*ident_slot;
HeapTuple tup_old = NULL;
- if (dstate->nchanges == 0)
- return;
-
/* TupleTableSlot is needed to pass the tuple to ExecInsertIndexTuples(). */
- index_slot = MakeSingleTupleTableSlot(dstate->tupdesc, &TTSOpsHeapTuple);
+ index_slot = MakeSingleTupleTableSlot(RelationGetDescr(rel),
+ &TTSOpsHeapTuple);
/* A slot to fetch tuples from identity index. */
ident_slot = table_slot_create(rel, NULL);
- while (tuplestore_gettupleslot(dstate->tstore, true, false,
- dstate->tsslot))
+ while (true)
{
- bool shouldFree;
- HeapTuple tup_change,
- tup,
+ size_t nread;
+ HeapTuple tup,
tup_exist;
- char *change_raw,
- *src;
- ConcurrentChange change;
- bool isnull[1];
- Datum values[1];
CHECK_FOR_INTERRUPTS();
- /* Get the change from the single-column tuple. */
- tup_change = ExecFetchSlotHeapTuple(dstate->tsslot, false, &shouldFree);
- heap_deform_tuple(tup_change, dstate->tupdesc_change, values, isnull);
- Assert(!isnull[0]);
-
- /* Make sure we access aligned data. */
- change_raw = (char *) DatumGetByteaP(values[0]);
- src = (char *) VARDATA(change_raw);
- memcpy(&change, src, SizeOfConcurrentChange);
+ nread = BufFileReadMaybeEOF(file, &kind, 1, true);
+ /* Are we done with the file? */
+ if (nread == 0)
+ break;
- /*
- * Extract the tuple from the change. The tuple is copied here because
- * it might be assigned to 'tup_old', in which case it needs to
- * survive into the next iteration.
- */
- tup = get_changed_tuple(src);
+ /* Read the tuple. */
+ BufFileReadExact(file, &t_len, sizeof(t_len));
+ tup = (HeapTuple) palloc(HEAPTUPLESIZE + t_len);
+ tup->t_data = (HeapTupleHeader) ((char *) tup + HEAPTUPLESIZE);
+ BufFileReadExact(file, tup->t_data, t_len);
+ tup->t_len = t_len;
+ ItemPointerSetInvalid(&tup->t_self);
+ tup->t_tableOid = RelationGetRelid(dest->rel);
- if (change.kind == CHANGE_UPDATE_OLD)
+ if (kind == CHANGE_UPDATE_OLD)
{
Assert(tup_old == NULL);
tup_old = tup;
}
- else if (change.kind == CHANGE_INSERT)
+ else if (kind == CHANGE_INSERT)
{
Assert(tup_old == NULL);
@@ -2719,12 +2860,11 @@ apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
pfree(tup);
}
- else if (change.kind == CHANGE_UPDATE_NEW ||
- change.kind == CHANGE_DELETE)
+ else if (kind == CHANGE_UPDATE_NEW || kind == CHANGE_DELETE)
{
HeapTuple tup_key;
- if (change.kind == CHANGE_UPDATE_NEW)
+ if (kind == CHANGE_UPDATE_NEW)
{
tup_key = tup_old != NULL ? tup_old : tup;
}
@@ -2741,7 +2881,7 @@ apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
if (tup_exist == NULL)
elog(ERROR, "failed to find target tuple");
- if (change.kind == CHANGE_UPDATE_NEW)
+ if (kind == CHANGE_UPDATE_NEW)
apply_concurrent_update(rel, tup, tup_exist, dest->iistate,
index_slot);
else
@@ -2756,26 +2896,19 @@ apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
pfree(tup);
}
else
- elog(ERROR, "unrecognized kind of change: %d", change.kind);
+ elog(ERROR, "unrecognized kind of change: %d", kind);
/*
* If a change was applied now, increment CID for next writes and
* update the snapshot so it sees the changes we've applied so far.
*/
- if (change.kind != CHANGE_UPDATE_OLD)
+ if (kind != CHANGE_UPDATE_OLD)
{
CommandCounterIncrement();
UpdateActiveSnapshotCommandId();
}
-
- /* TTSOpsMinimalTuple has .get_heap_tuple==NULL. */
- Assert(shouldFree);
- pfree(tup_change);
}
- tuplestore_clear(dstate->tstore);
- dstate->nchanges = 0;
-
/* Cleanup. */
ExecDropSingleTupleTableSlot(index_slot);
ExecDropSingleTupleTableSlot(ident_slot);
@@ -2954,25 +3087,59 @@ find_target_tuple(Relation rel, ChangeDest *dest, HeapTuple tup_key,
}
/*
- * Decode and apply concurrent changes.
+ * Decode and apply concurrent changes, up to (and including) the record whose
+ * LSN is 'end_of_wal'.
*/
static void
-process_concurrent_changes(LogicalDecodingContext *decoding_ctx,
- XLogRecPtr end_of_wal, ChangeDest *dest)
+process_concurrent_changes(XLogRecPtr end_of_wal, ChangeDest *dest, bool done)
{
- RepackDecodingState *dstate;
+ DecodingWorkerShared *shared;
+ char fname[MAXPGPATH];
+ BufFile *file;
pgstat_progress_update_param(PROGRESS_REPACK_PHASE,
PROGRESS_REPACK_PHASE_CATCH_UP);
- dstate = (RepackDecodingState *) decoding_ctx->output_writer_private;
+ /* Ask the worker for the file. */
+ shared = (DecodingWorkerShared *) dsm_segment_address(decoding_worker->seg);
+ SpinLockAcquire(&shared->mutex);
+ shared->lsn_upto = end_of_wal;
+ shared->done = done;
+ SpinLockRelease(&shared->mutex);
- repack_decode_concurrent_changes(decoding_ctx, end_of_wal);
+ /*
+ * The worker needs to finish processing of the current WAL record. Even
+ * if it's idle, it'll need to close the output file. Thus we're likely to
+ * wait, so prepare for sleep.
+ */
+ ConditionVariablePrepareToSleep(&shared->cv);
+ for (;;)
+ {
+ int last_exported;
- if (dstate->nchanges == 0)
- return;
+ SpinLockAcquire(&shared->mutex);
+ last_exported = shared->last_exported;
+ SpinLockRelease(&shared->mutex);
+
+ /*
+ * Has the worker exported the file we are waiting for?
+ */
+ if (last_exported == dest->file_seq)
+ break;
+
+ ConditionVariableSleep(&shared->cv, WAIT_EVENT_REPACK_WORKER_EXPORT);
+ }
+ ConditionVariableCancelSleep();
- apply_concurrent_changes(dstate, dest);
+ /* Open the file. */
+ DecodingWorkerFileName(fname, shared->relid, dest->file_seq);
+ file = BufFileOpenFileSet(&shared->sfs.fs, fname, O_RDONLY, false);
+ apply_concurrent_changes(file, dest);
+
+ BufFileClose(file);
+
+ /* Get ready for the next file. */
+ dest->file_seq++;
}
/*
@@ -3098,15 +3265,10 @@ cleanup_logical_decoding(LogicalDecodingContext *ctx)
dstate = (RepackDecodingState *) ctx->output_writer_private;
- ExecDropSingleTupleTableSlot(dstate->tsslot);
- FreeTupleDesc(dstate->tupdesc_change);
FreeTupleDesc(dstate->tupdesc);
- tuplestore_end(dstate->tstore);
-
FreeDecodingContext(ctx);
- ReplicationSlotRelease();
- ReplicationSlotDrop(NameStr(dstate->slotname), false);
+ ReplicationSlotDropAcquired();
pfree(dstate);
}
@@ -3121,7 +3283,6 @@ cleanup_logical_decoding(LogicalDecodingContext *ctx)
static void
rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
Relation cl_index,
- LogicalDecodingContext *decoding_ctx,
TransactionId frozenXid,
MultiXactId cutoffMulti)
{
@@ -3204,6 +3365,7 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
&chgdst.ident_index);
chgdst.ident_key = build_identity_key(ident_idx_new, OldHeap,
&chgdst.ident_key_nentries);
+ chgdst.file_seq = WORKER_FILE_SNAPSHOT + 1;
/*
* During testing, wait for another backend to perform concurrent data
@@ -3225,7 +3387,7 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
* hold AccessExclusiveLock. (Quite some amount of WAL could have been
* written during the data copying and index creation.)
*/
- process_concurrent_changes(decoding_ctx, end_of_wal, &chgdst);
+ process_concurrent_changes(end_of_wal, &chgdst, false);
/*
* Acquire AccessExclusiveLock on the table, its TOAST relation (if there
@@ -3319,8 +3481,11 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
XLogFlush(wal_insert_ptr);
end_of_wal = GetFlushRecPtr(NULL);
- /* Apply the concurrent changes again. */
- process_concurrent_changes(decoding_ctx, end_of_wal, &chgdst);
+ /*
+ * Apply the concurrent changes again. Indicate that the decoding worker
+ * won't be needed anymore.
+ */
+ process_concurrent_changes(end_of_wal, &chgdst, true);
/* Remember info about rel before closing OldHeap */
relpersistence = OldHeap->rd_rel->relpersistence;
@@ -3430,3 +3595,510 @@ build_new_indexes(Relation NewHeap, Relation OldHeap, List *OldIndexes)
return result;
}
+
+/*
+ * Try to start a background worker to perform logical decoding of data
+ * changes applied to relation while REPACK CONCURRENTLY is copying its
+ * contents to a new table.
+ */
+static void
+start_decoding_worker(Oid relid)
+{
+ Size size;
+ dsm_segment *seg;
+ DecodingWorkerShared *shared;
+ shm_mq *mq;
+ shm_mq_handle *mqh;
+ BackgroundWorker bgw;
+
+ /* Setup shared memory. */
+ size = BUFFERALIGN(offsetof(DecodingWorkerShared, error_queue)) +
+ BUFFERALIGN(REPACK_ERROR_QUEUE_SIZE);
+ seg = dsm_create(size, 0);
+ shared = (DecodingWorkerShared *) dsm_segment_address(seg);
+ shared->lsn_upto = InvalidXLogRecPtr;
+ shared->done = false;
+ SharedFileSetInit(&shared->sfs, seg);
+ shared->last_exported = -1;
+ SpinLockInit(&shared->mutex);
+ shared->dbid = MyDatabaseId;
+
+ /*
+ * This is the UserId set in cluster_rel(). Security context shouldn't be
+ * needed for decoding worker.
+ */
+ shared->roleid = GetUserId();
+ shared->relid = relid;
+ ConditionVariableInit(&shared->cv);
+ shared->backend_proc = MyProc;
+ shared->backend_pid = MyProcPid;
+ shared->backend_proc_number = MyProcNumber;
+
+ mq = shm_mq_create((char *) BUFFERALIGN(shared->error_queue),
+ REPACK_ERROR_QUEUE_SIZE);
+ shm_mq_set_receiver(mq, MyProc);
+ mqh = shm_mq_attach(mq, seg, NULL);
+
+ memset(&bgw, 0, sizeof(bgw));
+ snprintf(bgw.bgw_name, BGW_MAXLEN,
+ "REPACK decoding worker for relation \"%s\"",
+ get_rel_name(relid));
+ snprintf(bgw.bgw_type, BGW_MAXLEN, "REPACK decoding worker");
+ bgw.bgw_flags = BGWORKER_SHMEM_ACCESS |
+ BGWORKER_BACKEND_DATABASE_CONNECTION;
+ bgw.bgw_start_time = BgWorkerStart_RecoveryFinished;
+ bgw.bgw_restart_time = BGW_NEVER_RESTART;
+ snprintf(bgw.bgw_library_name, MAXPGPATH, "postgres");
+ snprintf(bgw.bgw_function_name, BGW_MAXLEN, "RepackWorkerMain");
+ bgw.bgw_main_arg = UInt32GetDatum(dsm_segment_handle(seg));
+ bgw.bgw_notify_pid = MyProcPid;
+
+ decoding_worker = palloc0_object(DecodingWorker);
+ if (!RegisterDynamicBackgroundWorker(&bgw, &decoding_worker->handle))
+ ereport(ERROR,
+ (errcode(ERRCODE_CONFIGURATION_LIMIT_EXCEEDED),
+ errmsg("out of background worker slots"),
+ errhint("You might need to increase \"%s\".", "max_worker_processes")));
+
+ decoding_worker->seg = seg;
+ decoding_worker->error_mqh = mqh;
+
+ /*
+ * The decoding setup must be done before the caller can have XID assigned
+ * for any reason, otherwise the worker might end up in a deadlock,
+ * waiting for the caller's transaction to end. Therefore wait here until
+ * the worker indicates that it has the logical decoding initialized.
+ */
+ ConditionVariablePrepareToSleep(&shared->cv);
+ for (;;)
+ {
+ int initialized;
+
+ SpinLockAcquire(&shared->mutex);
+ initialized = shared->initialized;
+ SpinLockRelease(&shared->mutex);
+
+ if (initialized)
+ break;
+
+ ConditionVariableSleep(&shared->cv, WAIT_EVENT_REPACK_WORKER_EXPORT);
+ }
+ ConditionVariableCancelSleep();
+}
+
+/*
+ * Stop the decoding worker and cleanup the related resources.
+ *
+ * The worker stops on its own when it knows there is no more work to do, but
+ * we need to stop it explicitly at least on ERROR in the launching backend.
+ */
+static void
+stop_decoding_worker(void)
+{
+ BgwHandleStatus status;
+
+ /* Haven't reached the worker startup? */
+ if (decoding_worker == NULL)
+ return;
+
+ /* Could not register the worker? */
+ if (decoding_worker->handle == NULL)
+ return;
+
+ TerminateBackgroundWorker(decoding_worker->handle);
+ /* The worker should really exit before the REPACK command does. */
+ HOLD_INTERRUPTS();
+ status = WaitForBackgroundWorkerShutdown(decoding_worker->handle);
+ RESUME_INTERRUPTS();
+
+ if (status == BGWH_POSTMASTER_DIED)
+ ereport(FATAL,
+ (errcode(ERRCODE_ADMIN_SHUTDOWN),
+ errmsg("postmaster exited during REPACK command")));
+
+ shm_mq_detach(decoding_worker->error_mqh);
+
+ /*
+ * If we could not cancel the current sleep due to ERROR, do that before
+ * we detach from the shared memory the condition variable is located in.
+ * If we did not, the bgworker ERROR handling code would try and fail
+ * badly.
+ */
+ ConditionVariableCancelSleep();
+
+ dsm_detach(decoding_worker->seg);
+ pfree(decoding_worker);
+ decoding_worker = NULL;
+}
+
+/* Is this process a REPACK worker? */
+static bool is_repack_worker = false;
+
+static pid_t backend_pid;
+static ProcNumber backend_proc_number;
+
+/*
+ * See ParallelWorkerShutdown for details.
+ */
+static void
+RepackWorkerShutdown(int code, Datum arg)
+{
+ SendProcSignal(backend_pid,
+ PROCSIG_REPACK_MESSAGE,
+ backend_proc_number);
+
+ dsm_detach((dsm_segment *) DatumGetPointer(arg));
+}
+
+/* REPACK decoding worker entry point */
+void
+RepackWorkerMain(Datum main_arg)
+{
+ dsm_segment *seg;
+ DecodingWorkerShared *shared;
+ shm_mq *mq;
+ shm_mq_handle *mqh;
+
+ is_repack_worker = true;
+
+ /*
+ * Override the default bgworker_die() with die() so we can use
+ * CHECK_FOR_INTERRUPTS().
+ */
+ pqsignal(SIGTERM, die);
+ BackgroundWorkerUnblockSignals();
+
+ seg = dsm_attach(DatumGetUInt32(main_arg));
+ if (seg == NULL)
+ ereport(ERROR,
+ (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
+ errmsg("could not map dynamic shared memory segment")));
+
+ shared = (DecodingWorkerShared *) dsm_segment_address(seg);
+
+ /* Arrange to signal the leader if we exit. */
+ backend_pid = shared->backend_pid;
+ backend_proc_number = shared->backend_proc_number;
+ before_shmem_exit(RepackWorkerShutdown, PointerGetDatum(seg));
+
+ /*
+ * Join locking group - see the comments around the call of
+ * start_decoding_worker().
+ */
+ if (!BecomeLockGroupMember(shared->backend_proc, backend_pid))
+ /* The leader is not running anymore. */
+ return;
+
+ /*
+ * Setup a queue to send error messages to the backend that launched this
+ * worker.
+ */
+ mq = (shm_mq *) (char *) BUFFERALIGN(shared->error_queue);
+ shm_mq_set_sender(mq, MyProc);
+ mqh = shm_mq_attach(mq, seg, NULL);
+ pq_redirect_to_shm_mq(seg, mqh);
+ pq_set_parallel_leader(shared->backend_pid,
+ shared->backend_proc_number);
+
+ /* Connect to the database. */
+ BackgroundWorkerInitializeConnectionByOid(shared->dbid, shared->roleid, 0);
+
+ repack_worker_internal(seg);
+}
+
+static void
+repack_worker_internal(dsm_segment *seg)
+{
+ DecodingWorkerShared *shared;
+ LogicalDecodingContext *decoding_ctx;
+ SharedFileSet *sfs;
+ Snapshot snapshot;
+
+ /*
+ * Transaction is needed to open relation, and it also provides us with a
+ * resource owner.
+ */
+ StartTransactionCommand();
+
+ shared = (DecodingWorkerShared *) dsm_segment_address(seg);
+
+ /*
+ * Not sure the spinlock is needed here - the backend should not change
+ * anything in the shared memory until we have serialized the snapshot.
+ */
+ SpinLockAcquire(&shared->mutex);
+ Assert(XLogRecPtrIsInvalid(shared->lsn_upto));
+ sfs = &shared->sfs;
+ SpinLockRelease(&shared->mutex);
+
+ SharedFileSetAttach(sfs, seg);
+
+ /*
+ * Prepare to capture the concurrent data changes ourselves.
+ */
+ decoding_ctx = setup_logical_decoding(shared->relid);
+
+ /* Announce that we're ready. */
+ SpinLockAcquire(&shared->mutex);
+ shared->initialized = true;
+ SpinLockRelease(&shared->mutex);
+ ConditionVariableSignal(&shared->cv);
+
+ /* Build the initial snapshot and export it. */
+ snapshot = SnapBuildInitialSnapshotForRepack(decoding_ctx->snapshot_builder);
+ export_initial_snapshot(snapshot, shared);
+
+ /*
+ * Only historic snapshots should be used now. Do not let us restrict the
+ * progress of xmin horizon.
+ */
+ InvalidateCatalogSnapshot();
+
+ while (!decode_concurrent_changes(decoding_ctx, shared))
+ ;
+
+ /* Cleanup. */
+ cleanup_logical_decoding(decoding_ctx);
+ CommitTransactionCommand();
+}
+
+/*
+ * Make snapshot available to the backend that launched the decoding worker.
+ */
+static void
+export_initial_snapshot(Snapshot snapshot, DecodingWorkerShared *shared)
+{
+ char fname[MAXPGPATH];
+ BufFile *file;
+ Size snap_size;
+ char *snap_space;
+
+ snap_size = EstimateSnapshotSpace(snapshot);
+ snap_space = (char *) palloc(snap_size);
+ SerializeSnapshot(snapshot, snap_space);
+ FreeSnapshot(snapshot);
+
+ DecodingWorkerFileName(fname, shared->relid, shared->last_exported + 1);
+ file = BufFileCreateFileSet(&shared->sfs.fs, fname);
+ /* To make restoration easier, write the snapshot size first. */
+ BufFileWrite(file, &snap_size, sizeof(snap_size));
+ BufFileWrite(file, snap_space, snap_size);
+ pfree(snap_space);
+ BufFileClose(file);
+
+ /* Increase the counter to tell the backend that the file is available. */
+ SpinLockAcquire(&shared->mutex);
+ shared->last_exported++;
+ SpinLockRelease(&shared->mutex);
+ ConditionVariableSignal(&shared->cv);
+}
+
+/*
+ * Get the initial snapshot from the decoding worker.
+ */
+static Snapshot
+get_initial_snapshot(DecodingWorker *worker)
+{
+ DecodingWorkerShared *shared;
+ char fname[MAXPGPATH];
+ BufFile *file;
+ Size snap_size;
+ char *snap_space;
+ Snapshot snapshot;
+
+ shared = (DecodingWorkerShared *) dsm_segment_address(worker->seg);
+
+ /*
+ * The worker needs to initialize the logical decoding, which usually
+ * takes some time. Therefore it makes sense to prepare for the sleep
+ * first.
+ */
+ ConditionVariablePrepareToSleep(&shared->cv);
+ for (;;)
+ {
+ int last_exported;
+
+ SpinLockAcquire(&shared->mutex);
+ last_exported = shared->last_exported;
+ SpinLockRelease(&shared->mutex);
+
+ /*
+ * Has the worker exported the file we are waiting for?
+ */
+ if (last_exported == WORKER_FILE_SNAPSHOT)
+ break;
+
+ ConditionVariableSleep(&shared->cv, WAIT_EVENT_REPACK_WORKER_EXPORT);
+ }
+ ConditionVariableCancelSleep();
+
+ /* Read the snapshot from a file. */
+ DecodingWorkerFileName(fname, shared->relid, WORKER_FILE_SNAPSHOT);
+ file = BufFileOpenFileSet(&shared->sfs.fs, fname, O_RDONLY, false);
+ BufFileReadExact(file, &snap_size, sizeof(snap_size));
+ snap_space = (char *) palloc(snap_size);
+ BufFileReadExact(file, snap_space, snap_size);
+ BufFileClose(file);
+
+ /* Restore it. */
+ snapshot = RestoreSnapshot(snap_space);
+ pfree(snap_space);
+
+ return snapshot;
+}
+
+bool
+IsRepackWorker(void)
+{
+ return is_repack_worker;
+}
+
+/*
+ * Handle receipt of an interrupt indicating a repack worker message.
+ *
+ * Note: this is called within a signal handler! All we can do is set
+ * a flag that will cause the next CHECK_FOR_INTERRUPTS() to invoke
+ * ProcessRepackMessages().
+ */
+void
+HandleRepackMessageInterrupt(void)
+{
+ InterruptPending = true;
+ RepackMessagePending = true;
+ SetLatch(MyLatch);
+}
+
+/*
+ * Process any queued protocol messages received from parallel workers.
+ */
+void
+ProcessRepackMessages(void)
+{
+ MemoryContext oldcontext;
+
+ static MemoryContext hpm_context = NULL;
+
+ /*
+ * Nothing to do if we haven't launched the worker yet or have already
+ * terminated it.
+ */
+ if (decoding_worker == NULL)
+ return;
+
+ /*
+ * This is invoked from ProcessInterrupts(), and since some of the
+ * functions it calls contain CHECK_FOR_INTERRUPTS(), there is a potential
+ * for recursive calls if more signals are received while this runs. It's
+ * unclear that recursive entry would be safe, and it doesn't seem useful
+ * even if it is safe, so let's block interrupts until done.
+ */
+ HOLD_INTERRUPTS();
+
+ /*
+ * Moreover, CurrentMemoryContext might be pointing almost anywhere. We
+ * don't want to risk leaking data into long-lived contexts, so let's do
+ * our work here in a private context that we can reset on each use.
+ */
+ if (hpm_context == NULL) /* first time through? */
+ hpm_context = AllocSetContextCreate(TopMemoryContext,
+ "ProcessRepackMessages",
+ ALLOCSET_DEFAULT_SIZES);
+ else
+ MemoryContextReset(hpm_context);
+
+ oldcontext = MemoryContextSwitchTo(hpm_context);
+
+ /* OK to process messages. Reset the flag saying there are more to do. */
+ RepackMessagePending = false;
+
+ /*
+ * Read as many messages as we can from each worker, but stop when no more
+ * messages can be read from the worker without blocking.
+ */
+ while (true)
+ {
+ shm_mq_result res;
+ Size nbytes;
+ void *data;
+
+ res = shm_mq_receive(decoding_worker->error_mqh, &nbytes,
+ &data, true);
+ if (res == SHM_MQ_WOULD_BLOCK)
+ break;
+ else if (res == SHM_MQ_SUCCESS)
+ {
+ StringInfoData msg;
+
+ initStringInfo(&msg);
+ appendBinaryStringInfo(&msg, data, nbytes);
+ ProcessRepackMessage(&msg);
+ pfree(msg.data);
+ }
+ else
+ {
+ /*
+ * The decoding worker is special in that it exits as soon as it
+ * has its work done. Thus the DETACHED result code is fine.
+ */
+ Assert(res == SHM_MQ_DETACHED);
+
+ break;
+ }
+ }
+
+ MemoryContextSwitchTo(oldcontext);
+
+ /* Might as well clear the context on our way out */
+ MemoryContextReset(hpm_context);
+
+ RESUME_INTERRUPTS();
+}
+
+/*
+ * Process a single protocol message received from a single parallel worker.
+ */
+static void
+ProcessRepackMessage(StringInfo msg)
+{
+ char msgtype;
+
+ msgtype = pq_getmsgbyte(msg);
+
+ switch (msgtype)
+ {
+ case PqMsg_ErrorResponse:
+ case PqMsg_NoticeResponse:
+ {
+ ErrorData edata;
+
+ /* Parse ErrorResponse or NoticeResponse. */
+ pq_parse_errornotice(msg, &edata);
+
+ /* Death of a worker isn't enough justification for suicide. */
+ edata.elevel = Min(edata.elevel, ERROR);
+
+ /*
+ * If desired, add a context line to show that this is a
+ * message propagated from a parallel worker. Otherwise, it
+ * can sometimes be confusing to understand what actually
+ * happened.
+ */
+ if (edata.context)
+ edata.context = psprintf("%s\n%s", edata.context,
+ _("decoding worker"));
+ else
+ edata.context = pstrdup(_("decoding worker"));
+
+ /* Rethrow error or print notice. */
+ ThrowErrorData(&edata);
+
+ break;
+ }
+
+ default:
+ {
+ elog(ERROR, "unrecognized message type received from decoding worker: %c (message length %d bytes)",
+ msgtype, msg->len);
+ }
+ }
+}
diff --git a/src/backend/libpq/pqmq.c b/src/backend/libpq/pqmq.c
index 6e4bbfb5aa1..42f6fa472c5 100644
--- a/src/backend/libpq/pqmq.c
+++ b/src/backend/libpq/pqmq.c
@@ -14,6 +14,7 @@
#include "postgres.h"
#include "access/parallel.h"
+#include "commands/cluster.h"
#include "libpq/libpq.h"
#include "libpq/pqformat.h"
#include "libpq/pqmq.h"
@@ -175,6 +176,10 @@ mq_putmessage(char msgtype, const char *s, size_t len)
SendProcSignal(pq_mq_parallel_leader_pid,
PROCSIG_PARALLEL_APPLY_MESSAGE,
pq_mq_parallel_leader_proc_number);
+ else if (IsRepackWorker())
+ SendProcSignal(pq_mq_parallel_leader_pid,
+ PROCSIG_REPACK_MESSAGE,
+ pq_mq_parallel_leader_proc_number);
else
{
Assert(IsParallelWorker());
diff --git a/src/backend/postmaster/bgworker.c b/src/backend/postmaster/bgworker.c
index 65deabe91a7..334bb708c5b 100644
--- a/src/backend/postmaster/bgworker.c
+++ b/src/backend/postmaster/bgworker.c
@@ -13,6 +13,7 @@
#include "postgres.h"
#include "access/parallel.h"
+#include "commands/cluster.h"
#include "libpq/pqsignal.h"
#include "miscadmin.h"
#include "pgstat.h"
@@ -136,6 +137,9 @@ static const struct
},
{
"SequenceSyncWorkerMain", SequenceSyncWorkerMain
+ },
+ {
+ "RepackWorkerMain", RepackWorkerMain
}
};
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index b0ef1a12520..35a46988285 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -194,7 +194,11 @@ StartupDecodingContext(List *output_plugin_options,
ctx->slot = slot;
- ctx->reader = XLogReaderAllocate(wal_segment_size, NULL, xl_routine, ctx);
+ /*
+ * TODO A separate patch for PG core, unless there's really a reason to
+ * pass ctx for private_data (May extensions expect ctx?).
+ */
+ ctx->reader = XLogReaderAllocate(wal_segment_size, NULL, xl_routine, NULL);
if (!ctx->reader)
ereport(ERROR,
(errcode(ERRCODE_OUT_OF_MEMORY),
diff --git a/src/backend/replication/pgoutput_repack/pgoutput_repack.c b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
index c8930640a0d..fb9956d392d 100644
--- a/src/backend/replication/pgoutput_repack/pgoutput_repack.c
+++ b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
@@ -168,17 +168,13 @@ store_change(LogicalDecodingContext *ctx, ConcurrentChangeKind kind,
HeapTuple tuple)
{
RepackDecodingState *dstate;
- char *change_raw;
- ConcurrentChange change;
+ char kind_byte = (char) kind;
bool flattened = false;
- Size size;
- Datum values[1];
- bool isnull[1];
- char *dst;
dstate = (RepackDecodingState *) ctx->output_writer_private;
- size = VARHDRSZ + SizeOfConcurrentChange;
+ /* Store the change kind. */
+ BufFileWrite(dstate->file, &kind_byte, 1);
/*
* ReorderBufferCommit() stores the TOAST chunks in its private memory
@@ -195,46 +191,12 @@ store_change(LogicalDecodingContext *ctx, ConcurrentChangeKind kind,
tuple = toast_flatten_tuple(tuple, dstate->tupdesc);
flattened = true;
}
+ /* Store the tuple size ... */
+ BufFileWrite(dstate->file, &tuple->t_len, sizeof(tuple->t_len));
+ /* ... and the tuple itself. */
+ BufFileWrite(dstate->file, tuple->t_data, tuple->t_len);
- size += tuple->t_len;
- if (size >= MaxAllocSize)
- elog(ERROR, "Change is too big.");
-
- /* Construct the change. */
- change_raw = (char *) palloc0(size);
- SET_VARSIZE(change_raw, size);
-
- /*
- * Since the varlena alignment might not be sufficient for the structure,
- * set the fields in a local instance and remember where it should
- * eventually be copied.
- */
- change.kind = kind;
- dst = (char *) VARDATA(change_raw);
-
- /*
- * Copy the tuple.
- *
- * Note: change->tup_data.t_data must be fixed on retrieval!
- */
- memcpy(&change.tup_data, tuple, sizeof(HeapTupleData));
- memcpy(dst, &change, SizeOfConcurrentChange);
- dst += SizeOfConcurrentChange;
- memcpy(dst, tuple->t_data, tuple->t_len);
-
- /* The data has been copied. */
+ /* Free the flat copy if created above. */
if (flattened)
pfree(tuple);
-
- /* Store as tuple of 1 bytea column. */
- values[0] = PointerGetDatum(change_raw);
- isnull[0] = false;
- tuplestore_putvalues(dstate->tstore, dstate->tupdesc_change,
- values, isnull);
-
- /* Accounting. */
- dstate->nchanges++;
-
- /* Cleanup. */
- pfree(change_raw);
}
diff --git a/src/backend/storage/ipc/procsignal.c b/src/backend/storage/ipc/procsignal.c
index 8e56922dcea..6f9e7a7aab7 100644
--- a/src/backend/storage/ipc/procsignal.c
+++ b/src/backend/storage/ipc/procsignal.c
@@ -19,6 +19,7 @@
#include "access/parallel.h"
#include "commands/async.h"
+#include "commands/cluster.h"
#include "miscadmin.h"
#include "pgstat.h"
#include "port/pg_bitutils.h"
@@ -697,6 +698,9 @@ procsignal_sigusr1_handler(SIGNAL_ARGS)
if (CheckProcSignal(PROCSIG_PARALLEL_APPLY_MESSAGE))
HandleParallelApplyMessageInterrupt();
+ if (CheckProcSignal(PROCSIG_REPACK_MESSAGE))
+ HandleRepackMessageInterrupt();
+
if (CheckProcSignal(PROCSIG_RECOVERY_CONFLICT_DATABASE))
HandleRecoveryConflictInterrupt(PROCSIG_RECOVERY_CONFLICT_DATABASE);
diff --git a/src/backend/tcop/postgres.c b/src/backend/tcop/postgres.c
index 015c67bbeba..566e5a50c30 100644
--- a/src/backend/tcop/postgres.c
+++ b/src/backend/tcop/postgres.c
@@ -36,6 +36,7 @@
#include "access/xact.h"
#include "catalog/pg_type.h"
#include "commands/async.h"
+#include "commands/cluster.h"
#include "commands/event_trigger.h"
#include "commands/explain_state.h"
#include "commands/prepare.h"
@@ -3541,6 +3542,9 @@ ProcessInterrupts(void)
if (ParallelApplyMessagePending)
ProcessParallelApplyMessages();
+
+ if (RepackMessagePending)
+ ProcessRepackMessages();
}
/*
diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt
index 3299de23bb3..73a3def69bc 100644
--- a/src/backend/utils/activity/wait_event_names.txt
+++ b/src/backend/utils/activity/wait_event_names.txt
@@ -62,6 +62,7 @@ LOGICAL_APPLY_MAIN "Waiting in main loop of logical replication apply process."
LOGICAL_LAUNCHER_MAIN "Waiting in main loop of logical replication launcher process."
LOGICAL_PARALLEL_APPLY_MAIN "Waiting in main loop of logical replication parallel apply process."
RECOVERY_WAL_STREAM "Waiting in main loop of startup process for WAL to arrive, during streaming recovery."
+REPACK_WORKER_MAIN "Waiting in main loop of REPACK decoding worker process."
REPLICATION_SLOTSYNC_MAIN "Waiting in main loop of slot synchronization."
REPLICATION_SLOTSYNC_SHUTDOWN "Waiting for slot sync worker to shut down."
SYSLOGGER_MAIN "Waiting in main loop of syslogger process."
@@ -154,6 +155,7 @@ RECOVERY_CONFLICT_SNAPSHOT "Waiting for recovery conflict resolution for a vacuu
RECOVERY_CONFLICT_TABLESPACE "Waiting for recovery conflict resolution for dropping a tablespace."
RECOVERY_END_COMMAND "Waiting for <xref linkend="guc-recovery-end-command"/> to complete."
RECOVERY_PAUSE "Waiting for recovery to be resumed."
+REPACK_WORKER_EXPORT "Waiting for decoding worker to export a new output file."
REPLICATION_ORIGIN_DROP "Waiting for a replication origin to become inactive so it can be dropped."
REPLICATION_SLOT_DROP "Waiting for a replication slot to become inactive so it can be dropped."
RESTORE_COMMAND "Waiting for <xref linkend="guc-restore-command"/> to complete."
diff --git a/src/include/access/tableam.h b/src/include/access/tableam.h
index 76aa993009a..15760363a1a 100644
--- a/src/include/access/tableam.h
+++ b/src/include/access/tableam.h
@@ -22,7 +22,6 @@
#include "access/xact.h"
#include "commands/vacuum.h"
#include "executor/tuptable.h"
-#include "replication/logical.h"
#include "storage/read_stream.h"
#include "utils/rel.h"
#include "utils/snapshot.h"
@@ -631,7 +630,6 @@ typedef struct TableAmRoutine
bool use_sort,
TransactionId OldestXmin,
Snapshot snapshot,
- LogicalDecodingContext *decoding_ctx,
TransactionId *xid_cutoff,
MultiXactId *multi_cutoff,
double *num_tuples,
@@ -1651,8 +1649,6 @@ table_relation_copy_data(Relation rel, const RelFileLocator *newrlocator)
* - *multi_cutoff - ditto
* - snapshot - if != NULL, ignore data changes done by transactions that this
* (MVCC) snapshot considers still in-progress or in the future.
- * - decoding_ctx - logical decoding context, to capture concurrent data
- * changes.
*
* Output parameters:
* - *xid_cutoff - rel's new relfrozenxid value, may be invalid
@@ -1666,7 +1662,6 @@ table_relation_copy_for_cluster(Relation OldTable, Relation NewTable,
bool use_sort,
TransactionId OldestXmin,
Snapshot snapshot,
- LogicalDecodingContext *decoding_ctx,
TransactionId *xid_cutoff,
MultiXactId *multi_cutoff,
double *num_tuples,
@@ -1675,7 +1670,7 @@ table_relation_copy_for_cluster(Relation OldTable, Relation NewTable,
{
OldTable->rd_tableam->relation_copy_for_cluster(OldTable, NewTable, OldIndex,
use_sort, OldestXmin,
- snapshot, decoding_ctx,
+ snapshot,
xid_cutoff, multi_cutoff,
num_tuples, tups_vacuumed,
tups_recently_dead);
diff --git a/src/include/commands/cluster.h b/src/include/commands/cluster.h
index 6a5c476294a..1b05d5d418b 100644
--- a/src/include/commands/cluster.h
+++ b/src/include/commands/cluster.h
@@ -17,11 +17,13 @@
#include "nodes/parsenodes.h"
#include "parser/parse_node.h"
#include "replication/decode.h"
+#include "postmaster/bgworker.h"
#include "replication/logical.h"
+#include "storage/buffile.h"
#include "storage/lock.h"
+#include "storage/shm_mq.h"
#include "utils/relcache.h"
#include "utils/resowner.h"
-#include "utils/tuplestore.h"
/* flag bits for ClusterParams->options */
@@ -44,6 +46,9 @@ typedef struct ClusterParams
* The following definitions are used by REPACK CONCURRENTLY.
*/
+/*
+ * Stored as a single byte in the output file.
+ */
typedef enum
{
CHANGE_INSERT,
@@ -52,68 +57,30 @@ typedef enum
CHANGE_DELETE
} ConcurrentChangeKind;
-typedef struct ConcurrentChange
-{
- /* See the enum above. */
- ConcurrentChangeKind kind;
-
- /*
- * The actual tuple.
- *
- * The tuple data follows the ConcurrentChange structure. Before use make
- * sure the tuple is correctly aligned (ConcurrentChange can be stored as
- * bytea) and that tuple->t_data is fixed.
- */
- HeapTupleData tup_data;
-} ConcurrentChange;
-
-#define SizeOfConcurrentChange (offsetof(ConcurrentChange, tup_data) + \
- sizeof(HeapTupleData))
-
/*
* Logical decoding state.
*
- * Here we store the data changes that we decode from WAL while the table
- * contents is being copied to a new storage. Also the necessary metadata
- * needed to apply these changes to the table is stored here.
+ * The output plugin uses it to store the data changes that it decodes from
+ * WAL while the table contents is being copied to a new storage.
*/
typedef struct RepackDecodingState
{
/* The relation whose changes we're decoding. */
Oid relid;
- /* Replication slot name. */
- NameData slotname;
-
- /*
- * Decoded changes are stored here. Although we try to avoid excessive
- * batches, it can happen that the changes need to be stored to disk. The
- * tuplestore does this transparently.
- */
- Tuplestorestate *tstore;
-
- /* The current number of changes in tstore. */
- double nchanges;
-
- /*
- * Descriptor to store the ConcurrentChange structure serialized (bytea).
- * We can't store the tuple directly because tuplestore only supports
- * minimum tuple and we may need to transfer OID system column from the
- * output plugin. Also we need to transfer the change kind, so it's better
- * to put everything in the structure than to use 2 tuplestores "in
- * parallel".
- */
- TupleDesc tupdesc_change;
-
- /* Tuple descriptor needed to update indexes. */
+ /* Tuple descriptor of the relation being processed. */
TupleDesc tupdesc;
- /* Slot to retrieve data from tstore. */
- TupleTableSlot *tsslot;
-
- ResourceOwner resowner;
+ /* The current output file. */
+ BufFile *file;
} RepackDecodingState;
+extern PGDLLIMPORT volatile sig_atomic_t RepackMessagePending;
+
+extern bool IsRepackWorker(void);
+extern void HandleRepackMessageInterrupt(void);
+extern void ProcessRepackMessages(void);
+
extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel);
extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid,
@@ -136,6 +103,6 @@ extern void finish_heap_swap(Oid OIDOldHeap, Oid OIDNewHeap,
extern bool am_decoding_for_repack(void);
extern bool change_useless_for_repack(XLogRecordBuffer *buf);
-extern void repack_decode_concurrent_changes(LogicalDecodingContext *ctx,
- XLogRecPtr end_of_wal);
+
+extern void RepackWorkerMain(Datum main_arg);
#endif /* CLUSTER_H */
diff --git a/src/include/storage/procsignal.h b/src/include/storage/procsignal.h
index e52b8eb7697..3ef35ca6b80 100644
--- a/src/include/storage/procsignal.h
+++ b/src/include/storage/procsignal.h
@@ -36,6 +36,7 @@ typedef enum
PROCSIG_BARRIER, /* global barrier interrupt */
PROCSIG_LOG_MEMORY_CONTEXT, /* ask backend to log the memory contexts */
PROCSIG_PARALLEL_APPLY_MESSAGE, /* Message from parallel apply workers */
+ PROCSIG_REPACK_MESSAGE, /* Message from repack worker */
/* Recovery conflict reasons */
PROCSIG_RECOVERY_CONFLICT_FIRST,
diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list
index a0b7b38a5e2..d1a694f9008 100644
--- a/src/tools/pgindent/typedefs.list
+++ b/src/tools/pgindent/typedefs.list
@@ -496,7 +496,6 @@ CompressFileHandle
CompressionLocation
CompressorState
ComputeXidHorizonsResult
-ConcurrentChange
ConcurrentChangeKind
ConditionVariable
ConditionVariableMinimallyPadded
@@ -636,6 +635,9 @@ DeclareCursorStmt
DecodedBkpBlock
DecodedXLogRecord
DecodingOutputState
+DecodingWorker
+DecodingWorkerShared
+DecodingWorkerState
DefElem
DefElemAction
DefaultACLInfo
--
2.47.3
--=-=-=
Content-Type: text/plain
Content-Disposition: attachment;
filename=v29-0006-Use-multiple-snapshots-to-copy-the-data.patch
^ permalink raw reply [nested|flat] 55+ messages in thread
* [PATCH 5/6] Use background worker to do logical decoding.
@ 2026-01-08 16:47 Antonin Houska <[email protected]>
0 siblings, 0 replies; 55+ messages in thread
From: Antonin Houska @ 2026-01-08 16:47 UTC (permalink / raw)
If the backend performing REPACK (CONCURRENTLY) does both data copying and
logical decoding, it has to "travel in time" back and forth and therefore it
has to invalidate system caches quite a few times. (The copying and the
decoding work with different catalog snapshots.) As the decoding worker has
separate caches, the switching is not necessary.
Without the worker, it'd also be difficult to switch between potentiallly long
running tasks like index build and WAL decoding. (No decoding during that time
at all can suspend archiving / recycling of WAL segments for some time, which
in turn may result in full disk.)
Another problem is that, after having acquired AccessExclusiveLock (in order
to swap the files), the backend needs to both decode and apply the data
changes that took place while it was waiting for the lock. With the decoding
worker, the decoding runs all the time, so the backend only needs to apply the
changes. This can reduce the time the exclusive lock is held for.
Note that the code added in order to handle ERRORs in the background worker
almost duplicates the existing code that does the same for other types of
workers (See ProcessParallelMessages() and
ProcessParallelApplyMessages()). Refactoring of the existing code might be
useful, to reduce the duplication.
---
src/backend/access/heap/heapam_handler.c | 44 -
src/backend/commands/cluster.c | 1174 +++++++++++++----
src/backend/libpq/pqmq.c | 5 +
src/backend/postmaster/bgworker.c | 4 +
src/backend/replication/logical/logical.c | 6 +-
.../pgoutput_repack/pgoutput_repack.c | 54 +-
src/backend/storage/ipc/procsignal.c | 4 +
src/backend/tcop/postgres.c | 4 +
.../utils/activity/wait_event_names.txt | 2 +
src/include/access/tableam.h | 7 +-
src/include/commands/cluster.h | 71 +-
src/include/storage/procsignal.h | 1 +
src/tools/pgindent/typedefs.list | 4 +-
13 files changed, 979 insertions(+), 401 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index 3526b6adcb5..475c536ce43 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -33,7 +33,6 @@
#include "catalog/index.h"
#include "catalog/storage.h"
#include "catalog/storage_xlog.h"
-#include "commands/cluster.h"
#include "commands/progress.h"
#include "executor/executor.h"
#include "miscadmin.h"
@@ -688,7 +687,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
Relation OldIndex, bool use_sort,
TransactionId OldestXmin,
Snapshot snapshot,
- LogicalDecodingContext *decoding_ctx,
TransactionId *xid_cutoff,
MultiXactId *multi_cutoff,
double *num_tuples,
@@ -710,7 +708,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
BufferHeapTupleTableSlot *hslot;
BlockNumber prev_cblock = InvalidBlockNumber;
bool concurrent = snapshot != NULL;
- XLogRecPtr end_of_wal_prev = GetFlushRecPtr(NULL);
/* Remember if it's a system catalog */
is_system_catalog = IsSystemRelation(OldHeap);
@@ -957,31 +954,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
ct_val[1] = *num_tuples;
pgstat_progress_update_multi_param(2, ct_index, ct_val);
}
-
- /*
- * Process the WAL produced by the load, as well as by other
- * transactions, so that the replication slot can advance and WAL does
- * not pile up. Use wal_segment_size as a threshold so that we do not
- * introduce the decoding overhead too often.
- *
- * Of course, we must not apply the changes until the initial load has
- * completed.
- *
- * Note that our insertions into the new table should not be decoded
- * as we (intentionally) do not write the logical decoding specific
- * information to WAL.
- */
- if (concurrent)
- {
- XLogRecPtr end_of_wal;
-
- end_of_wal = GetFlushRecPtr(NULL);
- if ((end_of_wal - end_of_wal_prev) > wal_segment_size)
- {
- repack_decode_concurrent_changes(decoding_ctx, end_of_wal);
- end_of_wal_prev = end_of_wal;
- }
- }
}
if (indexScan != NULL)
@@ -1027,22 +999,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
-
- /*
- * Try to keep the amount of not-yet-decoded WAL small, like
- * above.
- */
- if (concurrent)
- {
- XLogRecPtr end_of_wal;
-
- end_of_wal = GetFlushRecPtr(NULL);
- if ((end_of_wal - end_of_wal_prev) > wal_segment_size)
- {
- repack_decode_concurrent_changes(decoding_ctx, end_of_wal);
- end_of_wal_prev = end_of_wal;
- }
- }
}
tuplesort_end(tuplesort);
diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c
index c3feb0c3de4..5232fbfb57d 100644
--- a/src/backend/commands/cluster.c
+++ b/src/backend/commands/cluster.c
@@ -12,12 +12,13 @@
* In concurrent mode, we lock the table with only ShareUpdateExclusiveLock,
* then do an initial copy as above. However, while the tuples are being
* copied, concurrent transactions could modify the table. To cope with those
- * changes, we rely on logical decoding to obtain them from WAL. The changes
- * are accumulated in a tuplestore. Once the initial copy is complete, we
- * read the changes from the tuplestore and re-apply them on the new heap.
- * Then we upgrade our ShareUpdateExclusiveLock to AccessExclusiveLock and
- * swap the relfilenodes. This way, the time we hold a strong lock on the
- * table is much reduced, and the bloat is eliminated.
+ * changes, we rely on logical decoding to obtain them from WAL. A bgworker
+ * consumes WAL while the initial copy is ongoing (to prevent excessive WAL
+ * from being reserved), and accumulates the changes in a file. Once the
+ * initial copy is complete, we read the changes from the file and re-apply
+ * them on the new heap. Then we upgrade our ShareUpdateExclusiveLock to
+ * AccessExclusiveLock and swap the relfilenodes. This way, the time we hold
+ * a strong lock on the table is much reduced, and the bloat is eliminated.
*
* There is hardly anything left of Paul Brown's original implementation...
*
@@ -45,6 +46,7 @@
#include "access/xlog_internal.h"
#include "access/xloginsert.h"
#include "access/xlogutils.h"
+#include "access/xlogwait.h"
#include "catalog/catalog.h"
#include "catalog/dependency.h"
#include "catalog/heap.h"
@@ -61,6 +63,8 @@
#include "commands/tablecmds.h"
#include "commands/vacuum.h"
#include "executor/executor.h"
+#include "libpq/pqformat.h"
+#include "libpq/pqmq.h"
#include "miscadmin.h"
#include "optimizer/optimizer.h"
#include "pgstat.h"
@@ -71,6 +75,8 @@
#include "storage/ipc.h"
#include "storage/lmgr.h"
#include "storage/predicate.h"
+#include "storage/procsignal.h"
+#include "tcop/tcopprot.h"
#include "utils/acl.h"
#include "utils/fmgroids.h"
#include "utils/guc.h"
@@ -117,6 +123,12 @@ typedef struct IndexInsertState
/* The WAL segment being decoded. */
static XLogSegNo repack_current_segment = 0;
+/*
+ * The first file exported by the decoding worker must contain a snapshot, the
+ * following ones contain the data changes.
+ */
+#define WORKER_FILE_SNAPSHOT 0
+
/*
* Information needed to apply concurrent data changes.
*/
@@ -136,8 +148,113 @@ typedef struct ChangeDest
/* Needed to update indexes of rel_dst. */
IndexInsertState *iistate;
+
+ /*
+ * Sequential number of the file containing the changes.
+ *
+ * TODO This field makes the structure name less descriptive. Should we
+ * rename it, e.g. to ChangeApplyInfo?
+ */
+ int file_seq;
} ChangeDest;
+/*
+ * Layout of shared memory used for communication between backend and the
+ * worker that performs logical decoding of data changes
+ */
+typedef struct DecodingWorkerShared
+{
+ /* Is the decoding initialized? */
+ bool initialized;
+
+ /*
+ * Once the worker has reached this LSN, it should close the current
+ * output file and either create a new one or exit, according to the field
+ * 'done'. If the value is InvalidXLogRecPtr, the worker should decode all
+ * the WAL available and keep checking this field. It is ok if the worker
+ * had already decoded records whose LSN is >= lsn_upto before this field
+ * has been set.
+ */
+ XLogRecPtr lsn_upto;
+
+ /* Exit after closing the current file? */
+ bool done;
+
+ /* The output is stored here. */
+ SharedFileSet sfs;
+
+ /* Number of the last file exported by the worker. */
+ int last_exported;
+
+ /* Synchronize access to the fields above. */
+ slock_t mutex;
+
+ /* Database to connect to. */
+ Oid dbid;
+
+ /* Role to connect as. */
+ Oid roleid;
+
+ /* Decode data changes of this relation. */
+ Oid relid;
+
+ /* The backend uses this to wait for the worker. */
+ ConditionVariable cv;
+
+ /* Info to signal the backend. */
+ PGPROC *backend_proc;
+ pid_t backend_pid;
+ ProcNumber backend_proc_number;
+
+ /* Error queue. */
+ shm_mq *error_mq;
+
+ /*
+ * Memory the queue is located int.
+ *
+ * For considerations on the value see the comments of
+ * PARALLEL_ERROR_QUEUE_SIZE.
+ */
+#define REPACK_ERROR_QUEUE_SIZE 16384
+ char error_queue[FLEXIBLE_ARRAY_MEMBER];
+} DecodingWorkerShared;
+
+/*
+ * Generate worker's output file name. If relations of the same 'relid' happen
+ * to be processed at the same time, they must be from different databases and
+ * therefore different backends must be involved. (PID is already present in
+ * the fileset name.)
+ */
+static inline void
+DecodingWorkerFileName(char *fname, Oid relid, uint32 seq)
+{
+ snprintf(fname, MAXPGPATH, "%u-%u", relid, seq);
+}
+
+/*
+ * Backend-local information to control the decoding worker.
+ */
+typedef struct DecodingWorker
+{
+ /* The worker. */
+ BackgroundWorkerHandle *handle;
+
+ /* DecodingWorkerShared is in this segment. */
+ dsm_segment *seg;
+
+ /* Handle of the error queue. */
+ shm_mq_handle *error_mqh;
+} DecodingWorker;
+
+/* Pointer to currently running decoding worker. */
+static DecodingWorker *decoding_worker = NULL;
+
+/*
+ * Is there a message sent by a repack worker that the backend needs to
+ * receive?
+ */
+volatile sig_atomic_t RepackMessagePending = false;
+
static bool cluster_rel_recheck(RepackCommand cmd, Relation OldHeap,
Oid indexOid, Oid userid, LOCKMODE lmode,
int options);
@@ -145,7 +262,7 @@ static void check_repack_concurrently_requirements(Relation rel);
static void rebuild_relation(Relation OldHeap, Relation index, bool verbose,
bool concurrent);
static void copy_table_data(Relation NewHeap, Relation OldHeap, Relation OldIndex,
- Snapshot snapshot, LogicalDecodingContext *decoding_ctx,
+ Snapshot snapshot,
bool verbose,
bool *pSwapToastByContent,
TransactionId *pFreezeXid,
@@ -158,12 +275,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd,
static bool cluster_is_permitted_for_relation(RepackCommand cmd,
Oid relid, Oid userid);
-static void begin_concurrent_repack(Relation rel);
-static void end_concurrent_repack(void);
static LogicalDecodingContext *setup_logical_decoding(Oid relid);
-static HeapTuple get_changed_tuple(char *change);
-static void apply_concurrent_changes(RepackDecodingState *dstate,
- ChangeDest *dest);
+static bool decode_concurrent_changes(LogicalDecodingContext *ctx,
+ DecodingWorkerShared *shared);
+static void apply_concurrent_changes(BufFile *file, ChangeDest *dest);
static void apply_concurrent_insert(Relation rel, HeapTuple tup,
IndexInsertState *iistate,
TupleTableSlot *index_slot);
@@ -175,9 +290,9 @@ static void apply_concurrent_delete(Relation rel, HeapTuple tup_target);
static HeapTuple find_target_tuple(Relation rel, ChangeDest *dest,
HeapTuple tup_key,
TupleTableSlot *ident_slot);
-static void process_concurrent_changes(LogicalDecodingContext *decoding_ctx,
- XLogRecPtr end_of_wal,
- ChangeDest *dest);
+static void process_concurrent_changes(XLogRecPtr end_of_wal,
+ ChangeDest *dest,
+ bool done);
static IndexInsertState *get_index_insert_state(Relation relation,
Oid ident_index_id,
Relation *ident_index_p);
@@ -187,7 +302,6 @@ static void free_index_insert_state(IndexInsertState *iistate);
static void cleanup_logical_decoding(LogicalDecodingContext *ctx);
static void rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
Relation cl_index,
- LogicalDecodingContext *decoding_ctx,
TransactionId frozenXid,
MultiXactId cutoffMulti);
static List *build_new_indexes(Relation NewHeap, Relation OldHeap, List *OldIndexes);
@@ -197,6 +311,13 @@ static Relation process_single_relation(RepackStmt *stmt,
ClusterParams *params);
static Oid determine_clustered_index(Relation rel, bool usingindex,
const char *indexname);
+static void start_decoding_worker(Oid relid);
+static void stop_decoding_worker(void);
+static void repack_worker_internal(dsm_segment *seg);
+static void export_initial_snapshot(Snapshot snapshot,
+ DecodingWorkerShared *shared);
+static Snapshot get_initial_snapshot(DecodingWorker *worker);
+static void ProcessRepackMessage(StringInfo msg);
static const char *RepackCommandAsString(RepackCommand cmd);
@@ -619,20 +740,20 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid,
/* rebuild_relation does all the dirty work */
PG_TRY();
{
- /*
- * For concurrent processing, make sure that our logical decoding
- * ignores data changes of other tables than the one we are
- * processing.
- */
- if (concurrent)
- begin_concurrent_repack(OldHeap);
-
rebuild_relation(OldHeap, index, verbose, concurrent);
}
PG_FINALLY();
{
if (concurrent)
- end_concurrent_repack();
+ {
+ /*
+ * Since during normal operation the worker was already asked to
+ * exit, stopping it explicitly is especially important on ERROR.
+ * However it still seems a good practice to make sure that the
+ * worker never survives the REPACK command.
+ */
+ stop_decoding_worker();
+ }
}
PG_END_TRY();
@@ -929,7 +1050,6 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
bool swap_toast_by_content;
TransactionId frozenXid;
MultiXactId cutoffMulti;
- LogicalDecodingContext *decoding_ctx = NULL;
Snapshot snapshot = NULL;
#if USE_ASSERT_CHECKING
LOCKMODE lmode;
@@ -943,19 +1063,36 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
if (concurrent)
{
/*
- * Prepare to capture the concurrent data changes.
+ * The worker needs to be member of the locking group we're the leader
+ * of. We ought to become the leader before the worker starts. The
+ * worker will join the group as soon as it starts.
*
- * Note that this call waits for all transactions with XID already
- * assigned to finish. If some of those transactions is waiting for a
- * lock conflicting with ShareUpdateExclusiveLock on our table (e.g.
- * it runs CREATE INDEX), we can end up in a deadlock. Not sure this
- * risk is worth unlocking/locking the table (and its clustering
- * index) and checking again if its still eligible for REPACK
- * CONCURRENTLY.
+ * This is to make sure that the deadlock described below is
+ * detectable by deadlock.c: if the worker waits for a transaction to
+ * complete and we are waiting for the worker output, then effectively
+ * we (i.e. this backend) are waiting for that transaction.
*/
- decoding_ctx = setup_logical_decoding(tableOid);
+ BecomeLockGroupLeader();
+
+ /*
+ * Start the worker that decodes data changes applied while we're
+ * copying the table contents.
+ *
+ * Note that the worker has to wait for all transactions with XID
+ * already assigned to finish. If some of those transactions is
+ * waiting for a lock conflicting with ShareUpdateExclusiveLock on our
+ * table (e.g. it runs CREATE INDEX), we can end up in a deadlock.
+ * Not sure this risk is worth unlocking/locking the table (and its
+ * clustering index) and checking again if its still eligible for
+ * REPACK CONCURRENTLY.
+ */
+ start_decoding_worker(tableOid);
+
+ /*
+ * Wait until the worker has the initial snapshot and retrieve it.
+ */
+ snapshot = get_initial_snapshot(decoding_worker);
- snapshot = SnapBuildInitialSnapshotForRepack(decoding_ctx->snapshot_builder);
PushActiveSnapshot(snapshot);
}
@@ -980,7 +1117,7 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
NewHeap = table_open(OIDNewHeap, NoLock);
/* Copy the heap data into the new table in the desired order */
- copy_table_data(NewHeap, OldHeap, index, snapshot, decoding_ctx, verbose,
+ copy_table_data(NewHeap, OldHeap, index, snapshot, verbose,
&swap_toast_by_content, &frozenXid, &cutoffMulti);
/* The historic snapshot won't be needed anymore. */
@@ -994,14 +1131,10 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
{
Assert(!swap_toast_by_content);
rebuild_relation_finish_concurrent(NewHeap, OldHeap, index,
- decoding_ctx,
frozenXid, cutoffMulti);
pgstat_progress_update_param(PROGRESS_REPACK_PHASE,
PROGRESS_REPACK_PHASE_FINAL_CLEANUP);
-
- /* Done with decoding. */
- cleanup_logical_decoding(decoding_ctx);
}
else
{
@@ -1172,8 +1305,7 @@ make_new_heap(Oid OIDOldHeap, Oid NewTableSpace, Oid NewAccessMethod,
*/
static void
copy_table_data(Relation NewHeap, Relation OldHeap, Relation OldIndex,
- Snapshot snapshot, LogicalDecodingContext *decoding_ctx,
- bool verbose, bool *pSwapToastByContent,
+ Snapshot snapshot, bool verbose, bool *pSwapToastByContent,
TransactionId *pFreezeXid, MultiXactId *pCutoffMulti)
{
Relation relRelation;
@@ -1334,7 +1466,6 @@ copy_table_data(Relation NewHeap, Relation OldHeap, Relation OldIndex,
*/
table_relation_copy_for_cluster(OldHeap, NewHeap, OldIndex, use_sort,
cutoffs.OldestXmin, snapshot,
- decoding_ctx,
&cutoffs.FreezeLimit,
&cutoffs.MultiXactCutoff,
&num_tuples, &tups_vacuumed,
@@ -2367,59 +2498,6 @@ RepackCommandAsString(RepackCommand cmd)
return "???";
}
-
-/*
- * Call this function before REPACK CONCURRENTLY starts to setup logical
- * decoding. It makes sure that other users of the table put enough
- * information into WAL.
- *
- * The point is that at various places we expect that the table we're
- * processing is treated like a system catalog. For example, we need to be
- * able to scan it using a "historic snapshot" anytime during the processing
- * (as opposed to scanning only at the start point of the decoding, as logical
- * replication does during initial table synchronization), in order to apply
- * concurrent UPDATE / DELETE commands.
- *
- * Note that TOAST table needs no attention here as it's not scanned using
- * historic snapshot.
- */
-static void
-begin_concurrent_repack(Relation rel)
-{
- Oid toastrelid;
-
- /*
- * Avoid logical decoding of other relations by this backend. The lock we
- * have guarantees that the actual locator cannot be changed concurrently:
- * TRUNCATE needs AccessExclusiveLock.
- */
- Assert(CheckRelationLockedByMe(rel, ShareUpdateExclusiveLock, false));
- repacked_rel_locator = rel->rd_locator;
- toastrelid = rel->rd_rel->reltoastrelid;
- if (OidIsValid(toastrelid))
- {
- Relation toastrel;
-
- /* Avoid logical decoding of other TOAST relations. */
- toastrel = table_open(toastrelid, AccessShareLock);
- repacked_rel_toast_locator = toastrel->rd_locator;
- table_close(toastrel, AccessShareLock);
- }
-}
-
-/*
- * Call this when done with REPACK CONCURRENTLY.
- */
-static void
-end_concurrent_repack(void)
-{
- /*
- * Restore normal function of (future) logical decoding for this backend.
- */
- repacked_rel_locator.relNumber = InvalidOid;
- repacked_rel_toast_locator.relNumber = InvalidOid;
-}
-
/*
* Is this backend performing logical decoding on behalf of REPACK
* (CONCURRENTLY) ?
@@ -2484,9 +2562,10 @@ static LogicalDecodingContext *
setup_logical_decoding(Oid relid)
{
Relation rel;
- TupleDesc tupdesc;
+ Oid toastrelid;
LogicalDecodingContext *ctx;
- RepackDecodingState *dstate = palloc0_object(RepackDecodingState);
+ NameData slotname;
+ RepackDecodingState *dstate;
/*
* REPACK CONCURRENTLY is not allowed in a transaction block, so this
@@ -2494,21 +2573,21 @@ setup_logical_decoding(Oid relid)
*/
Assert(!TransactionIdIsValid(GetTopTransactionIdIfAny()));
- /*
- * A single backend should not execute multiple REPACK commands at a time,
- * so use PID to make the slot unique.
- */
- snprintf(NameStr(dstate->slotname), NAMEDATALEN, "repack_%d", MyProcPid);
-
/*
* Check if we can use logical decoding.
*/
CheckSlotPermissions();
CheckLogicalDecodingRequirements();
- /* RS_TEMPORARY so that the slot gets cleaned up on ERROR. */
- ReplicationSlotCreate(NameStr(dstate->slotname), true, RS_TEMPORARY,
- false, false, false);
+ /*
+ * A single backend should not execute multiple REPACK commands at a time,
+ * so use PID to make the slot unique.
+ *
+ * RS_TEMPORARY so that the slot gets cleaned up on ERROR.
+ */
+ snprintf(NameStr(slotname), NAMEDATALEN, "repack_%d", MyProcPid);
+ ReplicationSlotCreate(NameStr(slotname), true, RS_TEMPORARY, false, false,
+ false);
/*
* Neither prepare_write nor do_write callback nor update_progress is
@@ -2530,104 +2609,109 @@ setup_logical_decoding(Oid relid)
DecodingContextFindStartpoint(ctx);
+ /*
+ * decode_concurrent_changes() needs non-blocking callback.
+ */
+ ctx->reader->routine.page_read = read_local_xlog_page_no_wait;
+
+ /*
+ * read_local_xlog_page_no_wait() needs to be able to indicate the end of
+ * WAL.
+ */
+ ctx->reader->private_data = MemoryContextAllocZero(ctx->context,
+ sizeof(ReadLocalXLogPageNoWaitPrivate));
+
+
/* Some WAL records should have been read. */
Assert(ctx->reader->EndRecPtr != InvalidXLogRecPtr);
+ /*
+ * Initialize repack_current_segment so that we can notice WAL segment
+ * boundaries.
+ */
XLByteToSeg(ctx->reader->EndRecPtr, repack_current_segment,
wal_segment_size);
- /*
- * Setup structures to store decoded changes.
- */
+ dstate = palloc0_object(RepackDecodingState);
dstate->relid = relid;
- dstate->tstore = tuplestore_begin_heap(false, false,
- maintenance_work_mem);
- /* Caller should already have the table locked. */
- rel = table_open(relid, NoLock);
- tupdesc = CreateTupleDescCopy(RelationGetDescr(rel));
- dstate->tupdesc = tupdesc;
- table_close(rel, NoLock);
+ /*
+ * Tuple descriptor may be needed to flatten a tuple before we write it to
+ * a file. A copy is needed because the decoding worker invalidates system
+ * caches before it starts to do the actual work.
+ */
+ rel = table_open(relid, AccessShareLock);
+ dstate->tupdesc = CreateTupleDescCopy(RelationGetDescr(rel));
- /* Initialize the descriptor to store the changes ... */
- dstate->tupdesc_change = CreateTemplateTupleDesc(1);
+ /* Avoid logical decoding of other relations. */
+ repacked_rel_locator = rel->rd_locator;
+ toastrelid = rel->rd_rel->reltoastrelid;
+ if (OidIsValid(toastrelid))
+ {
+ Relation toastrel;
- TupleDescInitEntry(dstate->tupdesc_change, 1, NULL, BYTEAOID, -1, 0);
- /* ... as well as the corresponding slot. */
- dstate->tsslot = MakeSingleTupleTableSlot(dstate->tupdesc_change,
- &TTSOpsMinimalTuple);
+ /* Avoid logical decoding of other TOAST relations. */
+ toastrel = table_open(toastrelid, AccessShareLock);
+ repacked_rel_toast_locator = toastrel->rd_locator;
+ table_close(toastrel, AccessShareLock);
+ }
+ table_close(rel, AccessShareLock);
- dstate->resowner = ResourceOwnerCreate(CurrentResourceOwner,
- "logical decoding");
+ /* The file will be set as soon as we have it opened. */
+ dstate->file = NULL;
ctx->output_writer_private = dstate;
+
return ctx;
}
/*
- * Retrieve tuple from ConcurrentChange structure.
+ * Decode logical changes from the WAL sequence and store them to a file.
*
- * The input data starts with the structure but it might not be appropriately
- * aligned.
- */
-static HeapTuple
-get_changed_tuple(char *change)
-{
- HeapTupleData tup_data;
- HeapTuple result;
- char *src;
-
- /*
- * Ensure alignment before accessing the fields. (This is why we can't use
- * heap_copytuple() instead of this function.)
- */
- src = change + offsetof(ConcurrentChange, tup_data);
- memcpy(&tup_data, src, sizeof(HeapTupleData));
-
- result = (HeapTuple) palloc(HEAPTUPLESIZE + tup_data.t_len);
- memcpy(result, &tup_data, sizeof(HeapTupleData));
- result->t_data = (HeapTupleHeader) ((char *) result + HEAPTUPLESIZE);
- src = change + SizeOfConcurrentChange;
- memcpy(result->t_data, src, result->t_len);
-
- return result;
-}
-
-/*
- * Decode logical changes from the WAL sequence up to end_of_wal.
+ * If true is returned, there is no more work for the worker.
*/
-void
-repack_decode_concurrent_changes(LogicalDecodingContext *ctx,
- XLogRecPtr end_of_wal)
+static bool
+decode_concurrent_changes(LogicalDecodingContext *ctx,
+ DecodingWorkerShared *shared)
{
RepackDecodingState *dstate;
- ResourceOwner resowner_old;
+ XLogRecPtr lsn_upto;
+ bool done;
+ char fname[MAXPGPATH];
dstate = (RepackDecodingState *) ctx->output_writer_private;
- resowner_old = CurrentResourceOwner;
- CurrentResourceOwner = dstate->resowner;
- PG_TRY();
+ /* Open the output file. */
+ DecodingWorkerFileName(fname, shared->relid, shared->last_exported + 1);
+ dstate->file = BufFileCreateFileSet(&shared->sfs.fs, fname);
+
+ SpinLockAcquire(&shared->mutex);
+ lsn_upto = shared->lsn_upto;
+ done = shared->done;
+ SpinLockRelease(&shared->mutex);
+
+ while (true)
{
- while (ctx->reader->EndRecPtr < end_of_wal)
- {
- XLogRecord *record;
- XLogSegNo segno_new;
- char *errm = NULL;
- XLogRecPtr end_lsn;
+ XLogRecord *record;
+ XLogSegNo segno_new;
+ char *errm = NULL;
+ XLogRecPtr end_lsn;
- record = XLogReadRecord(ctx->reader, &errm);
- if (errm)
- elog(ERROR, "%s", errm);
+ CHECK_FOR_INTERRUPTS();
- if (record != NULL)
- LogicalDecodingProcessRecord(ctx, ctx->reader);
+ record = XLogReadRecord(ctx->reader, &errm);
+ if (record)
+ {
+ LogicalDecodingProcessRecord(ctx, ctx->reader);
/*
* If WAL segment boundary has been crossed, inform the decoding
- * system that the catalog_xmin can advance. (We can confirm more
- * often, but a filling a single WAL segment should not take much
- * time.)
+ * system that the catalog_xmin can advance.
+ *
+ * TODO Does it make sense to confirm more often? Segment size
+ * seems appropriate for restart_lsn (because less than a segment
+ * cannot be recycled anyway), however more frequent checks might
+ * be beneficial for catalog_xmin.
*/
end_lsn = ctx->reader->EndRecPtr;
XLByteToSeg(end_lsn, segno_new, wal_segment_size);
@@ -2638,80 +2722,137 @@ repack_decode_concurrent_changes(LogicalDecodingContext *ctx,
(uint32) (end_lsn >> 32), (uint32) end_lsn);
repack_current_segment = segno_new;
}
+ }
+ else
+ {
+ ReadLocalXLogPageNoWaitPrivate *priv;
- CHECK_FOR_INTERRUPTS();
+ if (errm)
+ ereport(ERROR, (errmsg("%s", errm)));
+
+ /*
+ * In the decoding loop we do not want to get blocked when there
+ * is no more WAL available, otherwise the loop would become
+ * uninterruptible.
+ */
+ priv = (ReadLocalXLogPageNoWaitPrivate *)
+ ctx->reader->private_data;
+ if (priv->end_of_wal)
+ /* Do not miss the end of WAL condition next time. */
+ priv->end_of_wal = false;
+ else
+ ereport(ERROR, (errmsg("could not read WAL record")));
+ }
+
+ /*
+ * Whether we could read new record or not, keep checking if
+ * 'lsn_upto' was specified.
+ */
+ if (XLogRecPtrIsInvalid(lsn_upto))
+ {
+ SpinLockAcquire(&shared->mutex);
+ lsn_upto = shared->lsn_upto;
+ /* 'done' should be set at the same time as 'lsn_upto' */
+ done = shared->done;
+ SpinLockRelease(&shared->mutex);
+ }
+ if (!XLogRecPtrIsInvalid(lsn_upto) &&
+ ctx->reader->EndRecPtr >= lsn_upto)
+ break;
+
+ if (record == NULL)
+ {
+ int64 timeout = 0;
+ WaitLSNResult res;
+
+ /*
+ * Before we retry reading, wait until new WAL is flushed.
+ *
+ * There is a race condition such that the backend executing
+ * REPACK determines 'lsn_upto', but before it sets the shared
+ * variable, we reach the end of WAL. In that case we'd need to
+ * wait until the next WAL flush (unrelated to REPACK). Although
+ * that should not be a problem in a busy system, it might be
+ * noticeable in other cases, including regression tests (which
+ * are not necessarily executed in parallel). Therefore it makes
+ * sense to use timeout.
+ *
+ * If lsn_upto is valid, WAL records having LSN lower than that
+ * should already have been flushed to disk.
+ */
+ if (XLogRecPtrIsInvalid(lsn_upto))
+ timeout = 100L;
+ res = WaitForLSN(WAIT_LSN_TYPE_PRIMARY_FLUSH,
+ ctx->reader->EndRecPtr + 1,
+ timeout);
+ if (res != WAIT_LSN_RESULT_SUCCESS &&
+ res != WAIT_LSN_RESULT_TIMEOUT)
+ ereport(ERROR, (errmsg("waiting for WAL failed")));
}
- InvalidateSystemCaches();
- CurrentResourceOwner = resowner_old;
- }
- PG_CATCH();
- {
- /* clear all timetravel entries */
- InvalidateSystemCaches();
- CurrentResourceOwner = resowner_old;
- PG_RE_THROW();
}
- PG_END_TRY();
+
+ /*
+ * Close the file so we can make it available to the backend.
+ */
+ BufFileClose(dstate->file);
+ dstate->file = NULL;
+ SpinLockAcquire(&shared->mutex);
+ shared->lsn_upto = InvalidXLogRecPtr;
+ shared->last_exported++;
+ SpinLockRelease(&shared->mutex);
+ ConditionVariableSignal(&shared->cv);
+
+ return done;
}
/*
* Apply changes stored in 'file'.
*/
static void
-apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
+apply_concurrent_changes(BufFile *file, ChangeDest *dest)
{
+ char kind;
+ uint32 t_len;
Relation rel = dest->rel;
TupleTableSlot *index_slot,
*ident_slot;
HeapTuple tup_old = NULL;
- if (dstate->nchanges == 0)
- return;
-
/* TupleTableSlot is needed to pass the tuple to ExecInsertIndexTuples(). */
- index_slot = MakeSingleTupleTableSlot(dstate->tupdesc, &TTSOpsHeapTuple);
+ index_slot = MakeSingleTupleTableSlot(RelationGetDescr(rel),
+ &TTSOpsHeapTuple);
/* A slot to fetch tuples from identity index. */
ident_slot = table_slot_create(rel, NULL);
- while (tuplestore_gettupleslot(dstate->tstore, true, false,
- dstate->tsslot))
+ while (true)
{
- bool shouldFree;
- HeapTuple tup_change,
- tup,
+ size_t nread;
+ HeapTuple tup,
tup_exist;
- char *change_raw,
- *src;
- ConcurrentChange change;
- bool isnull[1];
- Datum values[1];
CHECK_FOR_INTERRUPTS();
- /* Get the change from the single-column tuple. */
- tup_change = ExecFetchSlotHeapTuple(dstate->tsslot, false, &shouldFree);
- heap_deform_tuple(tup_change, dstate->tupdesc_change, values, isnull);
- Assert(!isnull[0]);
-
- /* Make sure we access aligned data. */
- change_raw = (char *) DatumGetByteaP(values[0]);
- src = (char *) VARDATA(change_raw);
- memcpy(&change, src, SizeOfConcurrentChange);
+ nread = BufFileReadMaybeEOF(file, &kind, 1, true);
+ /* Are we done with the file? */
+ if (nread == 0)
+ break;
- /*
- * Extract the tuple from the change. The tuple is copied here because
- * it might be assigned to 'tup_old', in which case it needs to
- * survive into the next iteration.
- */
- tup = get_changed_tuple(src);
+ /* Read the tuple. */
+ BufFileReadExact(file, &t_len, sizeof(t_len));
+ tup = (HeapTuple) palloc(HEAPTUPLESIZE + t_len);
+ tup->t_data = (HeapTupleHeader) ((char *) tup + HEAPTUPLESIZE);
+ BufFileReadExact(file, tup->t_data, t_len);
+ tup->t_len = t_len;
+ ItemPointerSetInvalid(&tup->t_self);
+ tup->t_tableOid = RelationGetRelid(dest->rel);
- if (change.kind == CHANGE_UPDATE_OLD)
+ if (kind == CHANGE_UPDATE_OLD)
{
Assert(tup_old == NULL);
tup_old = tup;
}
- else if (change.kind == CHANGE_INSERT)
+ else if (kind == CHANGE_INSERT)
{
Assert(tup_old == NULL);
@@ -2719,12 +2860,11 @@ apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
pfree(tup);
}
- else if (change.kind == CHANGE_UPDATE_NEW ||
- change.kind == CHANGE_DELETE)
+ else if (kind == CHANGE_UPDATE_NEW || kind == CHANGE_DELETE)
{
HeapTuple tup_key;
- if (change.kind == CHANGE_UPDATE_NEW)
+ if (kind == CHANGE_UPDATE_NEW)
{
tup_key = tup_old != NULL ? tup_old : tup;
}
@@ -2741,7 +2881,7 @@ apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
if (tup_exist == NULL)
elog(ERROR, "failed to find target tuple");
- if (change.kind == CHANGE_UPDATE_NEW)
+ if (kind == CHANGE_UPDATE_NEW)
apply_concurrent_update(rel, tup, tup_exist, dest->iistate,
index_slot);
else
@@ -2756,26 +2896,19 @@ apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
pfree(tup);
}
else
- elog(ERROR, "unrecognized kind of change: %d", change.kind);
+ elog(ERROR, "unrecognized kind of change: %d", kind);
/*
* If a change was applied now, increment CID for next writes and
* update the snapshot so it sees the changes we've applied so far.
*/
- if (change.kind != CHANGE_UPDATE_OLD)
+ if (kind != CHANGE_UPDATE_OLD)
{
CommandCounterIncrement();
UpdateActiveSnapshotCommandId();
}
-
- /* TTSOpsMinimalTuple has .get_heap_tuple==NULL. */
- Assert(shouldFree);
- pfree(tup_change);
}
- tuplestore_clear(dstate->tstore);
- dstate->nchanges = 0;
-
/* Cleanup. */
ExecDropSingleTupleTableSlot(index_slot);
ExecDropSingleTupleTableSlot(ident_slot);
@@ -2954,25 +3087,59 @@ find_target_tuple(Relation rel, ChangeDest *dest, HeapTuple tup_key,
}
/*
- * Decode and apply concurrent changes.
+ * Decode and apply concurrent changes, up to (and including) the record whose
+ * LSN is 'end_of_wal'.
*/
static void
-process_concurrent_changes(LogicalDecodingContext *decoding_ctx,
- XLogRecPtr end_of_wal, ChangeDest *dest)
+process_concurrent_changes(XLogRecPtr end_of_wal, ChangeDest *dest, bool done)
{
- RepackDecodingState *dstate;
+ DecodingWorkerShared *shared;
+ char fname[MAXPGPATH];
+ BufFile *file;
pgstat_progress_update_param(PROGRESS_REPACK_PHASE,
PROGRESS_REPACK_PHASE_CATCH_UP);
- dstate = (RepackDecodingState *) decoding_ctx->output_writer_private;
+ /* Ask the worker for the file. */
+ shared = (DecodingWorkerShared *) dsm_segment_address(decoding_worker->seg);
+ SpinLockAcquire(&shared->mutex);
+ shared->lsn_upto = end_of_wal;
+ shared->done = done;
+ SpinLockRelease(&shared->mutex);
- repack_decode_concurrent_changes(decoding_ctx, end_of_wal);
+ /*
+ * The worker needs to finish processing of the current WAL record. Even
+ * if it's idle, it'll need to close the output file. Thus we're likely to
+ * wait, so prepare for sleep.
+ */
+ ConditionVariablePrepareToSleep(&shared->cv);
+ for (;;)
+ {
+ int last_exported;
- if (dstate->nchanges == 0)
- return;
+ SpinLockAcquire(&shared->mutex);
+ last_exported = shared->last_exported;
+ SpinLockRelease(&shared->mutex);
+
+ /*
+ * Has the worker exported the file we are waiting for?
+ */
+ if (last_exported == dest->file_seq)
+ break;
+
+ ConditionVariableSleep(&shared->cv, WAIT_EVENT_REPACK_WORKER_EXPORT);
+ }
+ ConditionVariableCancelSleep();
- apply_concurrent_changes(dstate, dest);
+ /* Open the file. */
+ DecodingWorkerFileName(fname, shared->relid, dest->file_seq);
+ file = BufFileOpenFileSet(&shared->sfs.fs, fname, O_RDONLY, false);
+ apply_concurrent_changes(file, dest);
+
+ BufFileClose(file);
+
+ /* Get ready for the next file. */
+ dest->file_seq++;
}
/*
@@ -3098,15 +3265,10 @@ cleanup_logical_decoding(LogicalDecodingContext *ctx)
dstate = (RepackDecodingState *) ctx->output_writer_private;
- ExecDropSingleTupleTableSlot(dstate->tsslot);
- FreeTupleDesc(dstate->tupdesc_change);
FreeTupleDesc(dstate->tupdesc);
- tuplestore_end(dstate->tstore);
-
FreeDecodingContext(ctx);
- ReplicationSlotRelease();
- ReplicationSlotDrop(NameStr(dstate->slotname), false);
+ ReplicationSlotDropAcquired();
pfree(dstate);
}
@@ -3121,7 +3283,6 @@ cleanup_logical_decoding(LogicalDecodingContext *ctx)
static void
rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
Relation cl_index,
- LogicalDecodingContext *decoding_ctx,
TransactionId frozenXid,
MultiXactId cutoffMulti)
{
@@ -3204,6 +3365,7 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
&chgdst.ident_index);
chgdst.ident_key = build_identity_key(ident_idx_new, OldHeap,
&chgdst.ident_key_nentries);
+ chgdst.file_seq = WORKER_FILE_SNAPSHOT + 1;
/*
* During testing, wait for another backend to perform concurrent data
@@ -3225,7 +3387,7 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
* hold AccessExclusiveLock. (Quite some amount of WAL could have been
* written during the data copying and index creation.)
*/
- process_concurrent_changes(decoding_ctx, end_of_wal, &chgdst);
+ process_concurrent_changes(end_of_wal, &chgdst, false);
/*
* Acquire AccessExclusiveLock on the table, its TOAST relation (if there
@@ -3319,8 +3481,11 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
XLogFlush(wal_insert_ptr);
end_of_wal = GetFlushRecPtr(NULL);
- /* Apply the concurrent changes again. */
- process_concurrent_changes(decoding_ctx, end_of_wal, &chgdst);
+ /*
+ * Apply the concurrent changes again. Indicate that the decoding worker
+ * won't be needed anymore.
+ */
+ process_concurrent_changes(end_of_wal, &chgdst, true);
/* Remember info about rel before closing OldHeap */
relpersistence = OldHeap->rd_rel->relpersistence;
@@ -3430,3 +3595,510 @@ build_new_indexes(Relation NewHeap, Relation OldHeap, List *OldIndexes)
return result;
}
+
+/*
+ * Try to start a background worker to perform logical decoding of data
+ * changes applied to relation while REPACK CONCURRENTLY is copying its
+ * contents to a new table.
+ */
+static void
+start_decoding_worker(Oid relid)
+{
+ Size size;
+ dsm_segment *seg;
+ DecodingWorkerShared *shared;
+ shm_mq *mq;
+ shm_mq_handle *mqh;
+ BackgroundWorker bgw;
+
+ /* Setup shared memory. */
+ size = BUFFERALIGN(offsetof(DecodingWorkerShared, error_queue)) +
+ BUFFERALIGN(REPACK_ERROR_QUEUE_SIZE);
+ seg = dsm_create(size, 0);
+ shared = (DecodingWorkerShared *) dsm_segment_address(seg);
+ shared->lsn_upto = InvalidXLogRecPtr;
+ shared->done = false;
+ SharedFileSetInit(&shared->sfs, seg);
+ shared->last_exported = -1;
+ SpinLockInit(&shared->mutex);
+ shared->dbid = MyDatabaseId;
+
+ /*
+ * This is the UserId set in cluster_rel(). Security context shouldn't be
+ * needed for decoding worker.
+ */
+ shared->roleid = GetUserId();
+ shared->relid = relid;
+ ConditionVariableInit(&shared->cv);
+ shared->backend_proc = MyProc;
+ shared->backend_pid = MyProcPid;
+ shared->backend_proc_number = MyProcNumber;
+
+ mq = shm_mq_create((char *) BUFFERALIGN(shared->error_queue),
+ REPACK_ERROR_QUEUE_SIZE);
+ shm_mq_set_receiver(mq, MyProc);
+ mqh = shm_mq_attach(mq, seg, NULL);
+
+ memset(&bgw, 0, sizeof(bgw));
+ snprintf(bgw.bgw_name, BGW_MAXLEN,
+ "REPACK decoding worker for relation \"%s\"",
+ get_rel_name(relid));
+ snprintf(bgw.bgw_type, BGW_MAXLEN, "REPACK decoding worker");
+ bgw.bgw_flags = BGWORKER_SHMEM_ACCESS |
+ BGWORKER_BACKEND_DATABASE_CONNECTION;
+ bgw.bgw_start_time = BgWorkerStart_RecoveryFinished;
+ bgw.bgw_restart_time = BGW_NEVER_RESTART;
+ snprintf(bgw.bgw_library_name, MAXPGPATH, "postgres");
+ snprintf(bgw.bgw_function_name, BGW_MAXLEN, "RepackWorkerMain");
+ bgw.bgw_main_arg = UInt32GetDatum(dsm_segment_handle(seg));
+ bgw.bgw_notify_pid = MyProcPid;
+
+ decoding_worker = palloc0_object(DecodingWorker);
+ if (!RegisterDynamicBackgroundWorker(&bgw, &decoding_worker->handle))
+ ereport(ERROR,
+ (errcode(ERRCODE_CONFIGURATION_LIMIT_EXCEEDED),
+ errmsg("out of background worker slots"),
+ errhint("You might need to increase \"%s\".", "max_worker_processes")));
+
+ decoding_worker->seg = seg;
+ decoding_worker->error_mqh = mqh;
+
+ /*
+ * The decoding setup must be done before the caller can have XID assigned
+ * for any reason, otherwise the worker might end up in a deadlock,
+ * waiting for the caller's transaction to end. Therefore wait here until
+ * the worker indicates that it has the logical decoding initialized.
+ */
+ ConditionVariablePrepareToSleep(&shared->cv);
+ for (;;)
+ {
+ int initialized;
+
+ SpinLockAcquire(&shared->mutex);
+ initialized = shared->initialized;
+ SpinLockRelease(&shared->mutex);
+
+ if (initialized)
+ break;
+
+ ConditionVariableSleep(&shared->cv, WAIT_EVENT_REPACK_WORKER_EXPORT);
+ }
+ ConditionVariableCancelSleep();
+}
+
+/*
+ * Stop the decoding worker and cleanup the related resources.
+ *
+ * The worker stops on its own when it knows there is no more work to do, but
+ * we need to stop it explicitly at least on ERROR in the launching backend.
+ */
+static void
+stop_decoding_worker(void)
+{
+ BgwHandleStatus status;
+
+ /* Haven't reached the worker startup? */
+ if (decoding_worker == NULL)
+ return;
+
+ /* Could not register the worker? */
+ if (decoding_worker->handle == NULL)
+ return;
+
+ TerminateBackgroundWorker(decoding_worker->handle);
+ /* The worker should really exit before the REPACK command does. */
+ HOLD_INTERRUPTS();
+ status = WaitForBackgroundWorkerShutdown(decoding_worker->handle);
+ RESUME_INTERRUPTS();
+
+ if (status == BGWH_POSTMASTER_DIED)
+ ereport(FATAL,
+ (errcode(ERRCODE_ADMIN_SHUTDOWN),
+ errmsg("postmaster exited during REPACK command")));
+
+ shm_mq_detach(decoding_worker->error_mqh);
+
+ /*
+ * If we could not cancel the current sleep due to ERROR, do that before
+ * we detach from the shared memory the condition variable is located in.
+ * If we did not, the bgworker ERROR handling code would try and fail
+ * badly.
+ */
+ ConditionVariableCancelSleep();
+
+ dsm_detach(decoding_worker->seg);
+ pfree(decoding_worker);
+ decoding_worker = NULL;
+}
+
+/* Is this process a REPACK worker? */
+static bool is_repack_worker = false;
+
+static pid_t backend_pid;
+static ProcNumber backend_proc_number;
+
+/*
+ * See ParallelWorkerShutdown for details.
+ */
+static void
+RepackWorkerShutdown(int code, Datum arg)
+{
+ SendProcSignal(backend_pid,
+ PROCSIG_REPACK_MESSAGE,
+ backend_proc_number);
+
+ dsm_detach((dsm_segment *) DatumGetPointer(arg));
+}
+
+/* REPACK decoding worker entry point */
+void
+RepackWorkerMain(Datum main_arg)
+{
+ dsm_segment *seg;
+ DecodingWorkerShared *shared;
+ shm_mq *mq;
+ shm_mq_handle *mqh;
+
+ is_repack_worker = true;
+
+ /*
+ * Override the default bgworker_die() with die() so we can use
+ * CHECK_FOR_INTERRUPTS().
+ */
+ pqsignal(SIGTERM, die);
+ BackgroundWorkerUnblockSignals();
+
+ seg = dsm_attach(DatumGetUInt32(main_arg));
+ if (seg == NULL)
+ ereport(ERROR,
+ (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
+ errmsg("could not map dynamic shared memory segment")));
+
+ shared = (DecodingWorkerShared *) dsm_segment_address(seg);
+
+ /* Arrange to signal the leader if we exit. */
+ backend_pid = shared->backend_pid;
+ backend_proc_number = shared->backend_proc_number;
+ before_shmem_exit(RepackWorkerShutdown, PointerGetDatum(seg));
+
+ /*
+ * Join locking group - see the comments around the call of
+ * start_decoding_worker().
+ */
+ if (!BecomeLockGroupMember(shared->backend_proc, backend_pid))
+ /* The leader is not running anymore. */
+ return;
+
+ /*
+ * Setup a queue to send error messages to the backend that launched this
+ * worker.
+ */
+ mq = (shm_mq *) (char *) BUFFERALIGN(shared->error_queue);
+ shm_mq_set_sender(mq, MyProc);
+ mqh = shm_mq_attach(mq, seg, NULL);
+ pq_redirect_to_shm_mq(seg, mqh);
+ pq_set_parallel_leader(shared->backend_pid,
+ shared->backend_proc_number);
+
+ /* Connect to the database. */
+ BackgroundWorkerInitializeConnectionByOid(shared->dbid, shared->roleid, 0);
+
+ repack_worker_internal(seg);
+}
+
+static void
+repack_worker_internal(dsm_segment *seg)
+{
+ DecodingWorkerShared *shared;
+ LogicalDecodingContext *decoding_ctx;
+ SharedFileSet *sfs;
+ Snapshot snapshot;
+
+ /*
+ * Transaction is needed to open relation, and it also provides us with a
+ * resource owner.
+ */
+ StartTransactionCommand();
+
+ shared = (DecodingWorkerShared *) dsm_segment_address(seg);
+
+ /*
+ * Not sure the spinlock is needed here - the backend should not change
+ * anything in the shared memory until we have serialized the snapshot.
+ */
+ SpinLockAcquire(&shared->mutex);
+ Assert(XLogRecPtrIsInvalid(shared->lsn_upto));
+ sfs = &shared->sfs;
+ SpinLockRelease(&shared->mutex);
+
+ SharedFileSetAttach(sfs, seg);
+
+ /*
+ * Prepare to capture the concurrent data changes ourselves.
+ */
+ decoding_ctx = setup_logical_decoding(shared->relid);
+
+ /* Announce that we're ready. */
+ SpinLockAcquire(&shared->mutex);
+ shared->initialized = true;
+ SpinLockRelease(&shared->mutex);
+ ConditionVariableSignal(&shared->cv);
+
+ /* Build the initial snapshot and export it. */
+ snapshot = SnapBuildInitialSnapshotForRepack(decoding_ctx->snapshot_builder);
+ export_initial_snapshot(snapshot, shared);
+
+ /*
+ * Only historic snapshots should be used now. Do not let us restrict the
+ * progress of xmin horizon.
+ */
+ InvalidateCatalogSnapshot();
+
+ while (!decode_concurrent_changes(decoding_ctx, shared))
+ ;
+
+ /* Cleanup. */
+ cleanup_logical_decoding(decoding_ctx);
+ CommitTransactionCommand();
+}
+
+/*
+ * Make snapshot available to the backend that launched the decoding worker.
+ */
+static void
+export_initial_snapshot(Snapshot snapshot, DecodingWorkerShared *shared)
+{
+ char fname[MAXPGPATH];
+ BufFile *file;
+ Size snap_size;
+ char *snap_space;
+
+ snap_size = EstimateSnapshotSpace(snapshot);
+ snap_space = (char *) palloc(snap_size);
+ SerializeSnapshot(snapshot, snap_space);
+ FreeSnapshot(snapshot);
+
+ DecodingWorkerFileName(fname, shared->relid, shared->last_exported + 1);
+ file = BufFileCreateFileSet(&shared->sfs.fs, fname);
+ /* To make restoration easier, write the snapshot size first. */
+ BufFileWrite(file, &snap_size, sizeof(snap_size));
+ BufFileWrite(file, snap_space, snap_size);
+ pfree(snap_space);
+ BufFileClose(file);
+
+ /* Increase the counter to tell the backend that the file is available. */
+ SpinLockAcquire(&shared->mutex);
+ shared->last_exported++;
+ SpinLockRelease(&shared->mutex);
+ ConditionVariableSignal(&shared->cv);
+}
+
+/*
+ * Get the initial snapshot from the decoding worker.
+ */
+static Snapshot
+get_initial_snapshot(DecodingWorker *worker)
+{
+ DecodingWorkerShared *shared;
+ char fname[MAXPGPATH];
+ BufFile *file;
+ Size snap_size;
+ char *snap_space;
+ Snapshot snapshot;
+
+ shared = (DecodingWorkerShared *) dsm_segment_address(worker->seg);
+
+ /*
+ * The worker needs to initialize the logical decoding, which usually
+ * takes some time. Therefore it makes sense to prepare for the sleep
+ * first.
+ */
+ ConditionVariablePrepareToSleep(&shared->cv);
+ for (;;)
+ {
+ int last_exported;
+
+ SpinLockAcquire(&shared->mutex);
+ last_exported = shared->last_exported;
+ SpinLockRelease(&shared->mutex);
+
+ /*
+ * Has the worker exported the file we are waiting for?
+ */
+ if (last_exported == WORKER_FILE_SNAPSHOT)
+ break;
+
+ ConditionVariableSleep(&shared->cv, WAIT_EVENT_REPACK_WORKER_EXPORT);
+ }
+ ConditionVariableCancelSleep();
+
+ /* Read the snapshot from a file. */
+ DecodingWorkerFileName(fname, shared->relid, WORKER_FILE_SNAPSHOT);
+ file = BufFileOpenFileSet(&shared->sfs.fs, fname, O_RDONLY, false);
+ BufFileReadExact(file, &snap_size, sizeof(snap_size));
+ snap_space = (char *) palloc(snap_size);
+ BufFileReadExact(file, snap_space, snap_size);
+ BufFileClose(file);
+
+ /* Restore it. */
+ snapshot = RestoreSnapshot(snap_space);
+ pfree(snap_space);
+
+ return snapshot;
+}
+
+bool
+IsRepackWorker(void)
+{
+ return is_repack_worker;
+}
+
+/*
+ * Handle receipt of an interrupt indicating a repack worker message.
+ *
+ * Note: this is called within a signal handler! All we can do is set
+ * a flag that will cause the next CHECK_FOR_INTERRUPTS() to invoke
+ * ProcessRepackMessages().
+ */
+void
+HandleRepackMessageInterrupt(void)
+{
+ InterruptPending = true;
+ RepackMessagePending = true;
+ SetLatch(MyLatch);
+}
+
+/*
+ * Process any queued protocol messages received from parallel workers.
+ */
+void
+ProcessRepackMessages(void)
+{
+ MemoryContext oldcontext;
+
+ static MemoryContext hpm_context = NULL;
+
+ /*
+ * Nothing to do if we haven't launched the worker yet or have already
+ * terminated it.
+ */
+ if (decoding_worker == NULL)
+ return;
+
+ /*
+ * This is invoked from ProcessInterrupts(), and since some of the
+ * functions it calls contain CHECK_FOR_INTERRUPTS(), there is a potential
+ * for recursive calls if more signals are received while this runs. It's
+ * unclear that recursive entry would be safe, and it doesn't seem useful
+ * even if it is safe, so let's block interrupts until done.
+ */
+ HOLD_INTERRUPTS();
+
+ /*
+ * Moreover, CurrentMemoryContext might be pointing almost anywhere. We
+ * don't want to risk leaking data into long-lived contexts, so let's do
+ * our work here in a private context that we can reset on each use.
+ */
+ if (hpm_context == NULL) /* first time through? */
+ hpm_context = AllocSetContextCreate(TopMemoryContext,
+ "ProcessRepackMessages",
+ ALLOCSET_DEFAULT_SIZES);
+ else
+ MemoryContextReset(hpm_context);
+
+ oldcontext = MemoryContextSwitchTo(hpm_context);
+
+ /* OK to process messages. Reset the flag saying there are more to do. */
+ RepackMessagePending = false;
+
+ /*
+ * Read as many messages as we can from each worker, but stop when no more
+ * messages can be read from the worker without blocking.
+ */
+ while (true)
+ {
+ shm_mq_result res;
+ Size nbytes;
+ void *data;
+
+ res = shm_mq_receive(decoding_worker->error_mqh, &nbytes,
+ &data, true);
+ if (res == SHM_MQ_WOULD_BLOCK)
+ break;
+ else if (res == SHM_MQ_SUCCESS)
+ {
+ StringInfoData msg;
+
+ initStringInfo(&msg);
+ appendBinaryStringInfo(&msg, data, nbytes);
+ ProcessRepackMessage(&msg);
+ pfree(msg.data);
+ }
+ else
+ {
+ /*
+ * The decoding worker is special in that it exits as soon as it
+ * has its work done. Thus the DETACHED result code is fine.
+ */
+ Assert(res == SHM_MQ_DETACHED);
+
+ break;
+ }
+ }
+
+ MemoryContextSwitchTo(oldcontext);
+
+ /* Might as well clear the context on our way out */
+ MemoryContextReset(hpm_context);
+
+ RESUME_INTERRUPTS();
+}
+
+/*
+ * Process a single protocol message received from a single parallel worker.
+ */
+static void
+ProcessRepackMessage(StringInfo msg)
+{
+ char msgtype;
+
+ msgtype = pq_getmsgbyte(msg);
+
+ switch (msgtype)
+ {
+ case PqMsg_ErrorResponse:
+ case PqMsg_NoticeResponse:
+ {
+ ErrorData edata;
+
+ /* Parse ErrorResponse or NoticeResponse. */
+ pq_parse_errornotice(msg, &edata);
+
+ /* Death of a worker isn't enough justification for suicide. */
+ edata.elevel = Min(edata.elevel, ERROR);
+
+ /*
+ * If desired, add a context line to show that this is a
+ * message propagated from a parallel worker. Otherwise, it
+ * can sometimes be confusing to understand what actually
+ * happened.
+ */
+ if (edata.context)
+ edata.context = psprintf("%s\n%s", edata.context,
+ _("decoding worker"));
+ else
+ edata.context = pstrdup(_("decoding worker"));
+
+ /* Rethrow error or print notice. */
+ ThrowErrorData(&edata);
+
+ break;
+ }
+
+ default:
+ {
+ elog(ERROR, "unrecognized message type received from decoding worker: %c (message length %d bytes)",
+ msgtype, msg->len);
+ }
+ }
+}
diff --git a/src/backend/libpq/pqmq.c b/src/backend/libpq/pqmq.c
index 6e4bbfb5aa1..42f6fa472c5 100644
--- a/src/backend/libpq/pqmq.c
+++ b/src/backend/libpq/pqmq.c
@@ -14,6 +14,7 @@
#include "postgres.h"
#include "access/parallel.h"
+#include "commands/cluster.h"
#include "libpq/libpq.h"
#include "libpq/pqformat.h"
#include "libpq/pqmq.h"
@@ -175,6 +176,10 @@ mq_putmessage(char msgtype, const char *s, size_t len)
SendProcSignal(pq_mq_parallel_leader_pid,
PROCSIG_PARALLEL_APPLY_MESSAGE,
pq_mq_parallel_leader_proc_number);
+ else if (IsRepackWorker())
+ SendProcSignal(pq_mq_parallel_leader_pid,
+ PROCSIG_REPACK_MESSAGE,
+ pq_mq_parallel_leader_proc_number);
else
{
Assert(IsParallelWorker());
diff --git a/src/backend/postmaster/bgworker.c b/src/backend/postmaster/bgworker.c
index 65deabe91a7..334bb708c5b 100644
--- a/src/backend/postmaster/bgworker.c
+++ b/src/backend/postmaster/bgworker.c
@@ -13,6 +13,7 @@
#include "postgres.h"
#include "access/parallel.h"
+#include "commands/cluster.h"
#include "libpq/pqsignal.h"
#include "miscadmin.h"
#include "pgstat.h"
@@ -136,6 +137,9 @@ static const struct
},
{
"SequenceSyncWorkerMain", SequenceSyncWorkerMain
+ },
+ {
+ "RepackWorkerMain", RepackWorkerMain
}
};
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index b0ef1a12520..35a46988285 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -194,7 +194,11 @@ StartupDecodingContext(List *output_plugin_options,
ctx->slot = slot;
- ctx->reader = XLogReaderAllocate(wal_segment_size, NULL, xl_routine, ctx);
+ /*
+ * TODO A separate patch for PG core, unless there's really a reason to
+ * pass ctx for private_data (May extensions expect ctx?).
+ */
+ ctx->reader = XLogReaderAllocate(wal_segment_size, NULL, xl_routine, NULL);
if (!ctx->reader)
ereport(ERROR,
(errcode(ERRCODE_OUT_OF_MEMORY),
diff --git a/src/backend/replication/pgoutput_repack/pgoutput_repack.c b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
index c8930640a0d..fb9956d392d 100644
--- a/src/backend/replication/pgoutput_repack/pgoutput_repack.c
+++ b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
@@ -168,17 +168,13 @@ store_change(LogicalDecodingContext *ctx, ConcurrentChangeKind kind,
HeapTuple tuple)
{
RepackDecodingState *dstate;
- char *change_raw;
- ConcurrentChange change;
+ char kind_byte = (char) kind;
bool flattened = false;
- Size size;
- Datum values[1];
- bool isnull[1];
- char *dst;
dstate = (RepackDecodingState *) ctx->output_writer_private;
- size = VARHDRSZ + SizeOfConcurrentChange;
+ /* Store the change kind. */
+ BufFileWrite(dstate->file, &kind_byte, 1);
/*
* ReorderBufferCommit() stores the TOAST chunks in its private memory
@@ -195,46 +191,12 @@ store_change(LogicalDecodingContext *ctx, ConcurrentChangeKind kind,
tuple = toast_flatten_tuple(tuple, dstate->tupdesc);
flattened = true;
}
+ /* Store the tuple size ... */
+ BufFileWrite(dstate->file, &tuple->t_len, sizeof(tuple->t_len));
+ /* ... and the tuple itself. */
+ BufFileWrite(dstate->file, tuple->t_data, tuple->t_len);
- size += tuple->t_len;
- if (size >= MaxAllocSize)
- elog(ERROR, "Change is too big.");
-
- /* Construct the change. */
- change_raw = (char *) palloc0(size);
- SET_VARSIZE(change_raw, size);
-
- /*
- * Since the varlena alignment might not be sufficient for the structure,
- * set the fields in a local instance and remember where it should
- * eventually be copied.
- */
- change.kind = kind;
- dst = (char *) VARDATA(change_raw);
-
- /*
- * Copy the tuple.
- *
- * Note: change->tup_data.t_data must be fixed on retrieval!
- */
- memcpy(&change.tup_data, tuple, sizeof(HeapTupleData));
- memcpy(dst, &change, SizeOfConcurrentChange);
- dst += SizeOfConcurrentChange;
- memcpy(dst, tuple->t_data, tuple->t_len);
-
- /* The data has been copied. */
+ /* Free the flat copy if created above. */
if (flattened)
pfree(tuple);
-
- /* Store as tuple of 1 bytea column. */
- values[0] = PointerGetDatum(change_raw);
- isnull[0] = false;
- tuplestore_putvalues(dstate->tstore, dstate->tupdesc_change,
- values, isnull);
-
- /* Accounting. */
- dstate->nchanges++;
-
- /* Cleanup. */
- pfree(change_raw);
}
diff --git a/src/backend/storage/ipc/procsignal.c b/src/backend/storage/ipc/procsignal.c
index 8e56922dcea..6f9e7a7aab7 100644
--- a/src/backend/storage/ipc/procsignal.c
+++ b/src/backend/storage/ipc/procsignal.c
@@ -19,6 +19,7 @@
#include "access/parallel.h"
#include "commands/async.h"
+#include "commands/cluster.h"
#include "miscadmin.h"
#include "pgstat.h"
#include "port/pg_bitutils.h"
@@ -697,6 +698,9 @@ procsignal_sigusr1_handler(SIGNAL_ARGS)
if (CheckProcSignal(PROCSIG_PARALLEL_APPLY_MESSAGE))
HandleParallelApplyMessageInterrupt();
+ if (CheckProcSignal(PROCSIG_REPACK_MESSAGE))
+ HandleRepackMessageInterrupt();
+
if (CheckProcSignal(PROCSIG_RECOVERY_CONFLICT_DATABASE))
HandleRecoveryConflictInterrupt(PROCSIG_RECOVERY_CONFLICT_DATABASE);
diff --git a/src/backend/tcop/postgres.c b/src/backend/tcop/postgres.c
index 015c67bbeba..566e5a50c30 100644
--- a/src/backend/tcop/postgres.c
+++ b/src/backend/tcop/postgres.c
@@ -36,6 +36,7 @@
#include "access/xact.h"
#include "catalog/pg_type.h"
#include "commands/async.h"
+#include "commands/cluster.h"
#include "commands/event_trigger.h"
#include "commands/explain_state.h"
#include "commands/prepare.h"
@@ -3541,6 +3542,9 @@ ProcessInterrupts(void)
if (ParallelApplyMessagePending)
ProcessParallelApplyMessages();
+
+ if (RepackMessagePending)
+ ProcessRepackMessages();
}
/*
diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt
index 3299de23bb3..73a3def69bc 100644
--- a/src/backend/utils/activity/wait_event_names.txt
+++ b/src/backend/utils/activity/wait_event_names.txt
@@ -62,6 +62,7 @@ LOGICAL_APPLY_MAIN "Waiting in main loop of logical replication apply process."
LOGICAL_LAUNCHER_MAIN "Waiting in main loop of logical replication launcher process."
LOGICAL_PARALLEL_APPLY_MAIN "Waiting in main loop of logical replication parallel apply process."
RECOVERY_WAL_STREAM "Waiting in main loop of startup process for WAL to arrive, during streaming recovery."
+REPACK_WORKER_MAIN "Waiting in main loop of REPACK decoding worker process."
REPLICATION_SLOTSYNC_MAIN "Waiting in main loop of slot synchronization."
REPLICATION_SLOTSYNC_SHUTDOWN "Waiting for slot sync worker to shut down."
SYSLOGGER_MAIN "Waiting in main loop of syslogger process."
@@ -154,6 +155,7 @@ RECOVERY_CONFLICT_SNAPSHOT "Waiting for recovery conflict resolution for a vacuu
RECOVERY_CONFLICT_TABLESPACE "Waiting for recovery conflict resolution for dropping a tablespace."
RECOVERY_END_COMMAND "Waiting for <xref linkend="guc-recovery-end-command"/> to complete."
RECOVERY_PAUSE "Waiting for recovery to be resumed."
+REPACK_WORKER_EXPORT "Waiting for decoding worker to export a new output file."
REPLICATION_ORIGIN_DROP "Waiting for a replication origin to become inactive so it can be dropped."
REPLICATION_SLOT_DROP "Waiting for a replication slot to become inactive so it can be dropped."
RESTORE_COMMAND "Waiting for <xref linkend="guc-restore-command"/> to complete."
diff --git a/src/include/access/tableam.h b/src/include/access/tableam.h
index 76aa993009a..15760363a1a 100644
--- a/src/include/access/tableam.h
+++ b/src/include/access/tableam.h
@@ -22,7 +22,6 @@
#include "access/xact.h"
#include "commands/vacuum.h"
#include "executor/tuptable.h"
-#include "replication/logical.h"
#include "storage/read_stream.h"
#include "utils/rel.h"
#include "utils/snapshot.h"
@@ -631,7 +630,6 @@ typedef struct TableAmRoutine
bool use_sort,
TransactionId OldestXmin,
Snapshot snapshot,
- LogicalDecodingContext *decoding_ctx,
TransactionId *xid_cutoff,
MultiXactId *multi_cutoff,
double *num_tuples,
@@ -1651,8 +1649,6 @@ table_relation_copy_data(Relation rel, const RelFileLocator *newrlocator)
* - *multi_cutoff - ditto
* - snapshot - if != NULL, ignore data changes done by transactions that this
* (MVCC) snapshot considers still in-progress or in the future.
- * - decoding_ctx - logical decoding context, to capture concurrent data
- * changes.
*
* Output parameters:
* - *xid_cutoff - rel's new relfrozenxid value, may be invalid
@@ -1666,7 +1662,6 @@ table_relation_copy_for_cluster(Relation OldTable, Relation NewTable,
bool use_sort,
TransactionId OldestXmin,
Snapshot snapshot,
- LogicalDecodingContext *decoding_ctx,
TransactionId *xid_cutoff,
MultiXactId *multi_cutoff,
double *num_tuples,
@@ -1675,7 +1670,7 @@ table_relation_copy_for_cluster(Relation OldTable, Relation NewTable,
{
OldTable->rd_tableam->relation_copy_for_cluster(OldTable, NewTable, OldIndex,
use_sort, OldestXmin,
- snapshot, decoding_ctx,
+ snapshot,
xid_cutoff, multi_cutoff,
num_tuples, tups_vacuumed,
tups_recently_dead);
diff --git a/src/include/commands/cluster.h b/src/include/commands/cluster.h
index 6a5c476294a..1b05d5d418b 100644
--- a/src/include/commands/cluster.h
+++ b/src/include/commands/cluster.h
@@ -17,11 +17,13 @@
#include "nodes/parsenodes.h"
#include "parser/parse_node.h"
#include "replication/decode.h"
+#include "postmaster/bgworker.h"
#include "replication/logical.h"
+#include "storage/buffile.h"
#include "storage/lock.h"
+#include "storage/shm_mq.h"
#include "utils/relcache.h"
#include "utils/resowner.h"
-#include "utils/tuplestore.h"
/* flag bits for ClusterParams->options */
@@ -44,6 +46,9 @@ typedef struct ClusterParams
* The following definitions are used by REPACK CONCURRENTLY.
*/
+/*
+ * Stored as a single byte in the output file.
+ */
typedef enum
{
CHANGE_INSERT,
@@ -52,68 +57,30 @@ typedef enum
CHANGE_DELETE
} ConcurrentChangeKind;
-typedef struct ConcurrentChange
-{
- /* See the enum above. */
- ConcurrentChangeKind kind;
-
- /*
- * The actual tuple.
- *
- * The tuple data follows the ConcurrentChange structure. Before use make
- * sure the tuple is correctly aligned (ConcurrentChange can be stored as
- * bytea) and that tuple->t_data is fixed.
- */
- HeapTupleData tup_data;
-} ConcurrentChange;
-
-#define SizeOfConcurrentChange (offsetof(ConcurrentChange, tup_data) + \
- sizeof(HeapTupleData))
-
/*
* Logical decoding state.
*
- * Here we store the data changes that we decode from WAL while the table
- * contents is being copied to a new storage. Also the necessary metadata
- * needed to apply these changes to the table is stored here.
+ * The output plugin uses it to store the data changes that it decodes from
+ * WAL while the table contents is being copied to a new storage.
*/
typedef struct RepackDecodingState
{
/* The relation whose changes we're decoding. */
Oid relid;
- /* Replication slot name. */
- NameData slotname;
-
- /*
- * Decoded changes are stored here. Although we try to avoid excessive
- * batches, it can happen that the changes need to be stored to disk. The
- * tuplestore does this transparently.
- */
- Tuplestorestate *tstore;
-
- /* The current number of changes in tstore. */
- double nchanges;
-
- /*
- * Descriptor to store the ConcurrentChange structure serialized (bytea).
- * We can't store the tuple directly because tuplestore only supports
- * minimum tuple and we may need to transfer OID system column from the
- * output plugin. Also we need to transfer the change kind, so it's better
- * to put everything in the structure than to use 2 tuplestores "in
- * parallel".
- */
- TupleDesc tupdesc_change;
-
- /* Tuple descriptor needed to update indexes. */
+ /* Tuple descriptor of the relation being processed. */
TupleDesc tupdesc;
- /* Slot to retrieve data from tstore. */
- TupleTableSlot *tsslot;
-
- ResourceOwner resowner;
+ /* The current output file. */
+ BufFile *file;
} RepackDecodingState;
+extern PGDLLIMPORT volatile sig_atomic_t RepackMessagePending;
+
+extern bool IsRepackWorker(void);
+extern void HandleRepackMessageInterrupt(void);
+extern void ProcessRepackMessages(void);
+
extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel);
extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid,
@@ -136,6 +103,6 @@ extern void finish_heap_swap(Oid OIDOldHeap, Oid OIDNewHeap,
extern bool am_decoding_for_repack(void);
extern bool change_useless_for_repack(XLogRecordBuffer *buf);
-extern void repack_decode_concurrent_changes(LogicalDecodingContext *ctx,
- XLogRecPtr end_of_wal);
+
+extern void RepackWorkerMain(Datum main_arg);
#endif /* CLUSTER_H */
diff --git a/src/include/storage/procsignal.h b/src/include/storage/procsignal.h
index e52b8eb7697..3ef35ca6b80 100644
--- a/src/include/storage/procsignal.h
+++ b/src/include/storage/procsignal.h
@@ -36,6 +36,7 @@ typedef enum
PROCSIG_BARRIER, /* global barrier interrupt */
PROCSIG_LOG_MEMORY_CONTEXT, /* ask backend to log the memory contexts */
PROCSIG_PARALLEL_APPLY_MESSAGE, /* Message from parallel apply workers */
+ PROCSIG_REPACK_MESSAGE, /* Message from repack worker */
/* Recovery conflict reasons */
PROCSIG_RECOVERY_CONFLICT_FIRST,
diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list
index a0b7b38a5e2..d1a694f9008 100644
--- a/src/tools/pgindent/typedefs.list
+++ b/src/tools/pgindent/typedefs.list
@@ -496,7 +496,6 @@ CompressFileHandle
CompressionLocation
CompressorState
ComputeXidHorizonsResult
-ConcurrentChange
ConcurrentChangeKind
ConditionVariable
ConditionVariableMinimallyPadded
@@ -636,6 +635,9 @@ DeclareCursorStmt
DecodedBkpBlock
DecodedXLogRecord
DecodingOutputState
+DecodingWorker
+DecodingWorkerShared
+DecodingWorkerState
DefElem
DefElemAction
DefaultACLInfo
--
2.47.3
--=-=-=
Content-Type: text/plain
Content-Disposition: attachment;
filename=v29-0006-Use-multiple-snapshots-to-copy-the-data.patch
^ permalink raw reply [nested|flat] 55+ messages in thread
* [PATCH 5/6] Use background worker to do logical decoding.
@ 2026-01-12 16:30 Antonin Houska <[email protected]>
0 siblings, 0 replies; 55+ messages in thread
From: Antonin Houska @ 2026-01-12 16:30 UTC (permalink / raw)
If the backend performing REPACK (CONCURRENTLY) does both data copying and
logical decoding, it has to "travel in time" back and forth and therefore it
has to invalidate system caches quite a few times. (The copying and the
decoding work with different catalog snapshots.) As the decoding worker has
separate caches, the switching is not necessary.
Without the worker, it'd also be difficult to switch between potentiallly long
running tasks like index build and WAL decoding. (No decoding during that time
at all can suspend archiving / recycling of WAL segments for some time, which
in turn may result in full disk.)
Another problem is that, after having acquired AccessExclusiveLock (in order
to swap the files), the backend needs to both decode and apply the data
changes that took place while it was waiting for the lock. With the decoding
worker, the decoding runs all the time, so the backend only needs to apply the
changes. This can reduce the time the exclusive lock is held for.
Note that the code added in order to handle ERRORs in the background worker
almost duplicates the existing code that does the same for other types of
workers (See ProcessParallelMessages() and
ProcessParallelApplyMessages()). Refactoring of the existing code might be
useful, to reduce the duplication.
---
src/backend/access/heap/heapam_handler.c | 44 -
src/backend/commands/cluster.c | 1176 +++++++++++++----
src/backend/libpq/pqmq.c | 5 +
src/backend/postmaster/bgworker.c | 4 +
src/backend/replication/logical/logical.c | 6 +-
.../pgoutput_repack/pgoutput_repack.c | 54 +-
src/backend/storage/ipc/procsignal.c | 4 +
src/backend/tcop/postgres.c | 4 +
.../utils/activity/wait_event_names.txt | 2 +
src/include/access/tableam.h | 7 +-
src/include/commands/cluster.h | 71 +-
src/include/storage/procsignal.h | 1 +
src/tools/pgindent/typedefs.list | 4 +-
13 files changed, 981 insertions(+), 401 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index 3526b6adcb5..475c536ce43 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -33,7 +33,6 @@
#include "catalog/index.h"
#include "catalog/storage.h"
#include "catalog/storage_xlog.h"
-#include "commands/cluster.h"
#include "commands/progress.h"
#include "executor/executor.h"
#include "miscadmin.h"
@@ -688,7 +687,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
Relation OldIndex, bool use_sort,
TransactionId OldestXmin,
Snapshot snapshot,
- LogicalDecodingContext *decoding_ctx,
TransactionId *xid_cutoff,
MultiXactId *multi_cutoff,
double *num_tuples,
@@ -710,7 +708,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
BufferHeapTupleTableSlot *hslot;
BlockNumber prev_cblock = InvalidBlockNumber;
bool concurrent = snapshot != NULL;
- XLogRecPtr end_of_wal_prev = GetFlushRecPtr(NULL);
/* Remember if it's a system catalog */
is_system_catalog = IsSystemRelation(OldHeap);
@@ -957,31 +954,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
ct_val[1] = *num_tuples;
pgstat_progress_update_multi_param(2, ct_index, ct_val);
}
-
- /*
- * Process the WAL produced by the load, as well as by other
- * transactions, so that the replication slot can advance and WAL does
- * not pile up. Use wal_segment_size as a threshold so that we do not
- * introduce the decoding overhead too often.
- *
- * Of course, we must not apply the changes until the initial load has
- * completed.
- *
- * Note that our insertions into the new table should not be decoded
- * as we (intentionally) do not write the logical decoding specific
- * information to WAL.
- */
- if (concurrent)
- {
- XLogRecPtr end_of_wal;
-
- end_of_wal = GetFlushRecPtr(NULL);
- if ((end_of_wal - end_of_wal_prev) > wal_segment_size)
- {
- repack_decode_concurrent_changes(decoding_ctx, end_of_wal);
- end_of_wal_prev = end_of_wal;
- }
- }
}
if (indexScan != NULL)
@@ -1027,22 +999,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
-
- /*
- * Try to keep the amount of not-yet-decoded WAL small, like
- * above.
- */
- if (concurrent)
- {
- XLogRecPtr end_of_wal;
-
- end_of_wal = GetFlushRecPtr(NULL);
- if ((end_of_wal - end_of_wal_prev) > wal_segment_size)
- {
- repack_decode_concurrent_changes(decoding_ctx, end_of_wal);
- end_of_wal_prev = end_of_wal;
- }
- }
}
tuplesort_end(tuplesort);
diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c
index 9f61620a279..49a42f54ae4 100644
--- a/src/backend/commands/cluster.c
+++ b/src/backend/commands/cluster.c
@@ -12,12 +12,13 @@
* In concurrent mode, we lock the table with only ShareUpdateExclusiveLock,
* then do an initial copy as above. However, while the tuples are being
* copied, concurrent transactions could modify the table. To cope with those
- * changes, we rely on logical decoding to obtain them from WAL. The changes
- * are accumulated in a tuplestore. Once the initial copy is complete, we
- * read the changes from the tuplestore and re-apply them on the new heap.
- * Then we upgrade our ShareUpdateExclusiveLock to AccessExclusiveLock and
- * swap the relfilenodes. This way, the time we hold a strong lock on the
- * table is much reduced, and the bloat is eliminated.
+ * changes, we rely on logical decoding to obtain them from WAL. A bgworker
+ * consumes WAL while the initial copy is ongoing (to prevent excessive WAL
+ * from being reserved), and accumulates the changes in a file. Once the
+ * initial copy is complete, we read the changes from the file and re-apply
+ * them on the new heap. Then we upgrade our ShareUpdateExclusiveLock to
+ * AccessExclusiveLock and swap the relfilenodes. This way, the time we hold
+ * a strong lock on the table is much reduced, and the bloat is eliminated.
*
* There is hardly anything left of Paul Brown's original implementation...
*
@@ -45,6 +46,7 @@
#include "access/xlog_internal.h"
#include "access/xloginsert.h"
#include "access/xlogutils.h"
+#include "access/xlogwait.h"
#include "catalog/catalog.h"
#include "catalog/dependency.h"
#include "catalog/heap.h"
@@ -61,6 +63,8 @@
#include "commands/tablecmds.h"
#include "commands/vacuum.h"
#include "executor/executor.h"
+#include "libpq/pqformat.h"
+#include "libpq/pqmq.h"
#include "miscadmin.h"
#include "optimizer/optimizer.h"
#include "pgstat.h"
@@ -71,6 +75,8 @@
#include "storage/ipc.h"
#include "storage/lmgr.h"
#include "storage/predicate.h"
+#include "storage/procsignal.h"
+#include "tcop/tcopprot.h"
#include "utils/acl.h"
#include "utils/fmgroids.h"
#include "utils/guc.h"
@@ -117,6 +123,12 @@ typedef struct IndexInsertState
/* The WAL segment being decoded. */
static XLogSegNo repack_current_segment = 0;
+/*
+ * The first file exported by the decoding worker must contain a snapshot, the
+ * following ones contain the data changes.
+ */
+#define WORKER_FILE_SNAPSHOT 0
+
/*
* Information needed to apply concurrent data changes.
*/
@@ -136,8 +148,113 @@ typedef struct ChangeDest
/* Needed to update indexes of rel_dst. */
IndexInsertState *iistate;
+
+ /*
+ * Sequential number of the file containing the changes.
+ *
+ * TODO This field makes the structure name less descriptive. Should we
+ * rename it, e.g. to ChangeApplyInfo?
+ */
+ int file_seq;
} ChangeDest;
+/*
+ * Layout of shared memory used for communication between backend and the
+ * worker that performs logical decoding of data changes
+ */
+typedef struct DecodingWorkerShared
+{
+ /* Is the decoding initialized? */
+ bool initialized;
+
+ /*
+ * Once the worker has reached this LSN, it should close the current
+ * output file and either create a new one or exit, according to the field
+ * 'done'. If the value is InvalidXLogRecPtr, the worker should decode all
+ * the WAL available and keep checking this field. It is ok if the worker
+ * had already decoded records whose LSN is >= lsn_upto before this field
+ * has been set.
+ */
+ XLogRecPtr lsn_upto;
+
+ /* Exit after closing the current file? */
+ bool done;
+
+ /* The output is stored here. */
+ SharedFileSet sfs;
+
+ /* Number of the last file exported by the worker. */
+ int last_exported;
+
+ /* Synchronize access to the fields above. */
+ slock_t mutex;
+
+ /* Database to connect to. */
+ Oid dbid;
+
+ /* Role to connect as. */
+ Oid roleid;
+
+ /* Decode data changes of this relation. */
+ Oid relid;
+
+ /* The backend uses this to wait for the worker. */
+ ConditionVariable cv;
+
+ /* Info to signal the backend. */
+ PGPROC *backend_proc;
+ pid_t backend_pid;
+ ProcNumber backend_proc_number;
+
+ /* Error queue. */
+ shm_mq *error_mq;
+
+ /*
+ * Memory the queue is located int.
+ *
+ * For considerations on the value see the comments of
+ * PARALLEL_ERROR_QUEUE_SIZE.
+ */
+#define REPACK_ERROR_QUEUE_SIZE 16384
+ char error_queue[FLEXIBLE_ARRAY_MEMBER];
+} DecodingWorkerShared;
+
+/*
+ * Generate worker's output file name. If relations of the same 'relid' happen
+ * to be processed at the same time, they must be from different databases and
+ * therefore different backends must be involved. (PID is already present in
+ * the fileset name.)
+ */
+static inline void
+DecodingWorkerFileName(char *fname, Oid relid, uint32 seq)
+{
+ snprintf(fname, MAXPGPATH, "%u-%u", relid, seq);
+}
+
+/*
+ * Backend-local information to control the decoding worker.
+ */
+typedef struct DecodingWorker
+{
+ /* The worker. */
+ BackgroundWorkerHandle *handle;
+
+ /* DecodingWorkerShared is in this segment. */
+ dsm_segment *seg;
+
+ /* Handle of the error queue. */
+ shm_mq_handle *error_mqh;
+} DecodingWorker;
+
+/* Pointer to currently running decoding worker. */
+static DecodingWorker *decoding_worker = NULL;
+
+/*
+ * Is there a message sent by a repack worker that the backend needs to
+ * receive?
+ */
+volatile sig_atomic_t RepackMessagePending = false;
+
static bool cluster_rel_recheck(RepackCommand cmd, Relation OldHeap,
Oid indexOid, Oid userid, LOCKMODE lmode,
int options);
@@ -145,7 +262,7 @@ static void check_repack_concurrently_requirements(Relation rel);
static void rebuild_relation(Relation OldHeap, Relation index, bool verbose,
bool concurrent);
static void copy_table_data(Relation NewHeap, Relation OldHeap, Relation OldIndex,
- Snapshot snapshot, LogicalDecodingContext *decoding_ctx,
+ Snapshot snapshot,
bool verbose,
bool *pSwapToastByContent,
TransactionId *pFreezeXid,
@@ -158,12 +275,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd,
static bool cluster_is_permitted_for_relation(RepackCommand cmd,
Oid relid, Oid userid);
-static void begin_concurrent_repack(Relation rel);
-static void end_concurrent_repack(void);
static LogicalDecodingContext *setup_logical_decoding(Oid relid);
-static HeapTuple get_changed_tuple(char *change);
-static void apply_concurrent_changes(RepackDecodingState *dstate,
- ChangeDest *dest);
+static bool decode_concurrent_changes(LogicalDecodingContext *ctx,
+ DecodingWorkerShared *shared);
+static void apply_concurrent_changes(BufFile *file, ChangeDest *dest);
static void apply_concurrent_insert(Relation rel, HeapTuple tup,
IndexInsertState *iistate,
TupleTableSlot *index_slot);
@@ -175,9 +290,9 @@ static void apply_concurrent_delete(Relation rel, HeapTuple tup_target);
static HeapTuple find_target_tuple(Relation rel, ChangeDest *dest,
HeapTuple tup_key,
TupleTableSlot *ident_slot);
-static void process_concurrent_changes(LogicalDecodingContext *decoding_ctx,
- XLogRecPtr end_of_wal,
- ChangeDest *dest);
+static void process_concurrent_changes(XLogRecPtr end_of_wal,
+ ChangeDest *dest,
+ bool done);
static IndexInsertState *get_index_insert_state(Relation relation,
Oid ident_index_id,
Relation *ident_index_p);
@@ -186,7 +301,6 @@ static ScanKey build_identity_key(Oid ident_idx_oid, Relation rel_src,
static void free_index_insert_state(IndexInsertState *iistate);
static void cleanup_logical_decoding(LogicalDecodingContext *ctx);
static void rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
- LogicalDecodingContext *decoding_ctx,
TransactionId frozenXid,
MultiXactId cutoffMulti);
static List *build_new_indexes(Relation NewHeap, Relation OldHeap, List *OldIndexes);
@@ -196,6 +310,13 @@ static Relation process_single_relation(RepackStmt *stmt,
ClusterParams *params);
static Oid determine_clustered_index(Relation rel, bool usingindex,
const char *indexname);
+static void start_decoding_worker(Oid relid);
+static void stop_decoding_worker(void);
+static void repack_worker_internal(dsm_segment *seg);
+static void export_initial_snapshot(Snapshot snapshot,
+ DecodingWorkerShared *shared);
+static Snapshot get_initial_snapshot(DecodingWorker *worker);
+static void ProcessRepackMessage(StringInfo msg);
static const char *RepackCommandAsString(RepackCommand cmd);
@@ -619,20 +740,20 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid,
/* rebuild_relation does all the dirty work */
PG_TRY();
{
- /*
- * For concurrent processing, make sure that our logical decoding
- * ignores data changes of other tables than the one we are
- * processing.
- */
- if (concurrent)
- begin_concurrent_repack(OldHeap);
-
rebuild_relation(OldHeap, index, verbose, concurrent);
}
PG_FINALLY();
{
if (concurrent)
- end_concurrent_repack();
+ {
+ /*
+ * Since during normal operation the worker was already asked to
+ * exit, stopping it explicitly is especially important on ERROR.
+ * However it still seems a good practice to make sure that the
+ * worker never survives the REPACK command.
+ */
+ stop_decoding_worker();
+ }
}
PG_END_TRY();
@@ -929,7 +1050,6 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
bool swap_toast_by_content;
TransactionId frozenXid;
MultiXactId cutoffMulti;
- LogicalDecodingContext *decoding_ctx = NULL;
Snapshot snapshot = NULL;
#if USE_ASSERT_CHECKING
LOCKMODE lmode;
@@ -943,19 +1063,36 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
if (concurrent)
{
/*
- * Prepare to capture the concurrent data changes.
+ * The worker needs to be member of the locking group we're the leader
+ * of. We ought to become the leader before the worker starts. The
+ * worker will join the group as soon as it starts.
*
- * Note that this call waits for all transactions with XID already
- * assigned to finish. If some of those transactions is waiting for a
- * lock conflicting with ShareUpdateExclusiveLock on our table (e.g.
- * it runs CREATE INDEX), we can end up in a deadlock. Not sure this
- * risk is worth unlocking/locking the table (and its clustering
- * index) and checking again if its still eligible for REPACK
- * CONCURRENTLY.
+ * This is to make sure that the deadlock described below is
+ * detectable by deadlock.c: if the worker waits for a transaction to
+ * complete and we are waiting for the worker output, then effectively
+ * we (i.e. this backend) are waiting for that transaction.
*/
- decoding_ctx = setup_logical_decoding(tableOid);
+ BecomeLockGroupLeader();
+
+ /*
+ * Start the worker that decodes data changes applied while we're
+ * copying the table contents.
+ *
+ * Note that the worker has to wait for all transactions with XID
+ * already assigned to finish. If some of those transactions is
+ * waiting for a lock conflicting with ShareUpdateExclusiveLock on our
+ * table (e.g. it runs CREATE INDEX), we can end up in a deadlock.
+ * Not sure this risk is worth unlocking/locking the table (and its
+ * clustering index) and checking again if its still eligible for
+ * REPACK CONCURRENTLY.
+ */
+ start_decoding_worker(tableOid);
+
+ /*
+ * Wait until the worker has the initial snapshot and retrieve it.
+ */
+ snapshot = get_initial_snapshot(decoding_worker);
- snapshot = SnapBuildInitialSnapshotForRepack(decoding_ctx->snapshot_builder);
PushActiveSnapshot(snapshot);
}
@@ -980,7 +1117,7 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
NewHeap = table_open(OIDNewHeap, NoLock);
/* Copy the heap data into the new table in the desired order */
- copy_table_data(NewHeap, OldHeap, index, snapshot, decoding_ctx, verbose,
+ copy_table_data(NewHeap, OldHeap, index, snapshot, verbose,
&swap_toast_by_content, &frozenXid, &cutoffMulti);
/* The historic snapshot won't be needed anymore. */
@@ -1001,14 +1138,11 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
if (index)
index_close(index, NoLock);
- rebuild_relation_finish_concurrent(NewHeap, OldHeap, decoding_ctx,
- frozenXid, cutoffMulti);
+ rebuild_relation_finish_concurrent(NewHeap, OldHeap, frozenXid,
+ cutoffMulti);
pgstat_progress_update_param(PROGRESS_REPACK_PHASE,
PROGRESS_REPACK_PHASE_FINAL_CLEANUP);
-
- /* Done with decoding. */
- cleanup_logical_decoding(decoding_ctx);
}
else
{
@@ -1179,8 +1313,7 @@ make_new_heap(Oid OIDOldHeap, Oid NewTableSpace, Oid NewAccessMethod,
*/
static void
copy_table_data(Relation NewHeap, Relation OldHeap, Relation OldIndex,
- Snapshot snapshot, LogicalDecodingContext *decoding_ctx,
- bool verbose, bool *pSwapToastByContent,
+ Snapshot snapshot, bool verbose, bool *pSwapToastByContent,
TransactionId *pFreezeXid, MultiXactId *pCutoffMulti)
{
Relation relRelation;
@@ -1341,7 +1474,6 @@ copy_table_data(Relation NewHeap, Relation OldHeap, Relation OldIndex,
*/
table_relation_copy_for_cluster(OldHeap, NewHeap, OldIndex, use_sort,
cutoffs.OldestXmin, snapshot,
- decoding_ctx,
&cutoffs.FreezeLimit,
&cutoffs.MultiXactCutoff,
&num_tuples, &tups_vacuumed,
@@ -2375,58 +2507,6 @@ RepackCommandAsString(RepackCommand cmd)
}
-/*
- * Call this function before REPACK CONCURRENTLY starts to setup logical
- * decoding. It makes sure that other users of the table put enough
- * information into WAL.
- *
- * The point is that at various places we expect that the table we're
- * processing is treated like a system catalog. For example, we need to be
- * able to scan it using a "historic snapshot" anytime during the processing
- * (as opposed to scanning only at the start point of the decoding, as logical
- * replication does during initial table synchronization), in order to apply
- * concurrent UPDATE / DELETE commands.
- *
- * Note that TOAST table needs no attention here as it's not scanned using
- * historic snapshot.
- */
-static void
-begin_concurrent_repack(Relation rel)
-{
- Oid toastrelid;
-
- /*
- * Avoid logical decoding of other relations by this backend. The lock we
- * have guarantees that the actual locator cannot be changed concurrently:
- * TRUNCATE needs AccessExclusiveLock.
- */
- Assert(CheckRelationLockedByMe(rel, ShareUpdateExclusiveLock, false));
- repacked_rel_locator = rel->rd_locator;
- toastrelid = rel->rd_rel->reltoastrelid;
- if (OidIsValid(toastrelid))
- {
- Relation toastrel;
-
- /* Avoid logical decoding of other TOAST relations. */
- toastrel = table_open(toastrelid, AccessShareLock);
- repacked_rel_toast_locator = toastrel->rd_locator;
- table_close(toastrel, AccessShareLock);
- }
-}
-
-/*
- * Call this when done with REPACK CONCURRENTLY.
- */
-static void
-end_concurrent_repack(void)
-{
- /*
- * Restore normal function of (future) logical decoding for this backend.
- */
- repacked_rel_locator.relNumber = InvalidOid;
- repacked_rel_toast_locator.relNumber = InvalidOid;
-}
-
/*
* Is this backend performing logical decoding on behalf of REPACK
* (CONCURRENTLY) ?
@@ -2491,9 +2571,10 @@ static LogicalDecodingContext *
setup_logical_decoding(Oid relid)
{
Relation rel;
- TupleDesc tupdesc;
+ Oid toastrelid;
LogicalDecodingContext *ctx;
- RepackDecodingState *dstate = palloc0_object(RepackDecodingState);
+ NameData slotname;
+ RepackDecodingState *dstate;
/*
* REPACK CONCURRENTLY is not allowed in a transaction block, so this
@@ -2501,21 +2582,21 @@ setup_logical_decoding(Oid relid)
*/
Assert(!TransactionIdIsValid(GetTopTransactionIdIfAny()));
- /*
- * A single backend should not execute multiple REPACK commands at a time,
- * so use PID to make the slot unique.
- */
- snprintf(NameStr(dstate->slotname), NAMEDATALEN, "repack_%d", MyProcPid);
-
/*
* Check if we can use logical decoding.
*/
CheckSlotPermissions();
CheckLogicalDecodingRequirements();
- /* RS_TEMPORARY so that the slot gets cleaned up on ERROR. */
- ReplicationSlotCreate(NameStr(dstate->slotname), true, RS_TEMPORARY,
- false, false, false);
+ /*
+ * A single backend should not execute multiple REPACK commands at a time,
+ * so use PID to make the slot unique.
+ *
+ * RS_TEMPORARY so that the slot gets cleaned up on ERROR.
+ */
+ snprintf(NameStr(slotname), NAMEDATALEN, "repack_%d", MyProcPid);
+ ReplicationSlotCreate(NameStr(slotname), true, RS_TEMPORARY, false, false,
+ false);
/*
* Neither prepare_write nor do_write callback nor update_progress is
@@ -2537,104 +2618,109 @@ setup_logical_decoding(Oid relid)
DecodingContextFindStartpoint(ctx);
+ /*
+ * decode_concurrent_changes() needs non-blocking callback.
+ */
+ ctx->reader->routine.page_read = read_local_xlog_page_no_wait;
+
+ /*
+ * read_local_xlog_page_no_wait() needs to be able to indicate the end of
+ * WAL.
+ */
+ ctx->reader->private_data = MemoryContextAllocZero(ctx->context,
+ sizeof(ReadLocalXLogPageNoWaitPrivate));
+
+
/* Some WAL records should have been read. */
Assert(ctx->reader->EndRecPtr != InvalidXLogRecPtr);
+ /*
+ * Initialize repack_current_segment so that we can notice WAL segment
+ * boundaries.
+ */
XLByteToSeg(ctx->reader->EndRecPtr, repack_current_segment,
wal_segment_size);
- /*
- * Setup structures to store decoded changes.
- */
+ dstate = palloc0_object(RepackDecodingState);
dstate->relid = relid;
- dstate->tstore = tuplestore_begin_heap(false, false,
- maintenance_work_mem);
- /* Caller should already have the table locked. */
- rel = table_open(relid, NoLock);
- tupdesc = CreateTupleDescCopy(RelationGetDescr(rel));
- dstate->tupdesc = tupdesc;
- table_close(rel, NoLock);
+ /*
+ * Tuple descriptor may be needed to flatten a tuple before we write it to
+ * a file. A copy is needed because the decoding worker invalidates system
+ * caches before it starts to do the actual work.
+ */
+ rel = table_open(relid, AccessShareLock);
+ dstate->tupdesc = CreateTupleDescCopy(RelationGetDescr(rel));
- /* Initialize the descriptor to store the changes ... */
- dstate->tupdesc_change = CreateTemplateTupleDesc(1);
+ /* Avoid logical decoding of other relations. */
+ repacked_rel_locator = rel->rd_locator;
+ toastrelid = rel->rd_rel->reltoastrelid;
+ if (OidIsValid(toastrelid))
+ {
+ Relation toastrel;
- TupleDescInitEntry(dstate->tupdesc_change, 1, NULL, BYTEAOID, -1, 0);
- /* ... as well as the corresponding slot. */
- dstate->tsslot = MakeSingleTupleTableSlot(dstate->tupdesc_change,
- &TTSOpsMinimalTuple);
+ /* Avoid logical decoding of other TOAST relations. */
+ toastrel = table_open(toastrelid, AccessShareLock);
+ repacked_rel_toast_locator = toastrel->rd_locator;
+ table_close(toastrel, AccessShareLock);
+ }
+ table_close(rel, AccessShareLock);
- dstate->resowner = ResourceOwnerCreate(CurrentResourceOwner,
- "logical decoding");
+ /* The file will be set as soon as we have it opened. */
+ dstate->file = NULL;
ctx->output_writer_private = dstate;
+
return ctx;
}
/*
- * Retrieve tuple from ConcurrentChange structure.
+ * Decode logical changes from the WAL sequence and store them to a file.
*
- * The input data starts with the structure but it might not be appropriately
- * aligned.
- */
-static HeapTuple
-get_changed_tuple(char *change)
-{
- HeapTupleData tup_data;
- HeapTuple result;
- char *src;
-
- /*
- * Ensure alignment before accessing the fields. (This is why we can't use
- * heap_copytuple() instead of this function.)
- */
- src = change + offsetof(ConcurrentChange, tup_data);
- memcpy(&tup_data, src, sizeof(HeapTupleData));
-
- result = (HeapTuple) palloc(HEAPTUPLESIZE + tup_data.t_len);
- memcpy(result, &tup_data, sizeof(HeapTupleData));
- result->t_data = (HeapTupleHeader) ((char *) result + HEAPTUPLESIZE);
- src = change + SizeOfConcurrentChange;
- memcpy(result->t_data, src, result->t_len);
-
- return result;
-}
-
-/*
- * Decode logical changes from the WAL sequence up to end_of_wal.
+ * If true is returned, there is no more work for the worker.
*/
-void
-repack_decode_concurrent_changes(LogicalDecodingContext *ctx,
- XLogRecPtr end_of_wal)
+static bool
+decode_concurrent_changes(LogicalDecodingContext *ctx,
+ DecodingWorkerShared *shared)
{
RepackDecodingState *dstate;
- ResourceOwner resowner_old;
+ XLogRecPtr lsn_upto;
+ bool done;
+ char fname[MAXPGPATH];
dstate = (RepackDecodingState *) ctx->output_writer_private;
- resowner_old = CurrentResourceOwner;
- CurrentResourceOwner = dstate->resowner;
- PG_TRY();
+ /* Open the output file. */
+ DecodingWorkerFileName(fname, shared->relid, shared->last_exported + 1);
+ dstate->file = BufFileCreateFileSet(&shared->sfs.fs, fname);
+
+ SpinLockAcquire(&shared->mutex);
+ lsn_upto = shared->lsn_upto;
+ done = shared->done;
+ SpinLockRelease(&shared->mutex);
+
+ while (true)
{
- while (ctx->reader->EndRecPtr < end_of_wal)
- {
- XLogRecord *record;
- XLogSegNo segno_new;
- char *errm = NULL;
- XLogRecPtr end_lsn;
+ XLogRecord *record;
+ XLogSegNo segno_new;
+ char *errm = NULL;
+ XLogRecPtr end_lsn;
- record = XLogReadRecord(ctx->reader, &errm);
- if (errm)
- elog(ERROR, "%s", errm);
+ CHECK_FOR_INTERRUPTS();
- if (record != NULL)
- LogicalDecodingProcessRecord(ctx, ctx->reader);
+ record = XLogReadRecord(ctx->reader, &errm);
+ if (record)
+ {
+ LogicalDecodingProcessRecord(ctx, ctx->reader);
/*
* If WAL segment boundary has been crossed, inform the decoding
- * system that the catalog_xmin can advance. (We can confirm more
- * often, but a filling a single WAL segment should not take much
- * time.)
+ * system that the catalog_xmin can advance.
+ *
+ * TODO Does it make sense to confirm more often? Segment size
+ * seems appropriate for restart_lsn (because less than a segment
+ * cannot be recycled anyway), however more frequent checks might
+ * be beneficial for catalog_xmin.
*/
end_lsn = ctx->reader->EndRecPtr;
XLByteToSeg(end_lsn, segno_new, wal_segment_size);
@@ -2645,80 +2731,137 @@ repack_decode_concurrent_changes(LogicalDecodingContext *ctx,
(uint32) (end_lsn >> 32), (uint32) end_lsn);
repack_current_segment = segno_new;
}
+ }
+ else
+ {
+ ReadLocalXLogPageNoWaitPrivate *priv;
- CHECK_FOR_INTERRUPTS();
+ if (errm)
+ ereport(ERROR, (errmsg("%s", errm)));
+
+ /*
+ * In the decoding loop we do not want to get blocked when there
+ * is no more WAL available, otherwise the loop would become
+ * uninterruptible.
+ */
+ priv = (ReadLocalXLogPageNoWaitPrivate *)
+ ctx->reader->private_data;
+ if (priv->end_of_wal)
+ /* Do not miss the end of WAL condition next time. */
+ priv->end_of_wal = false;
+ else
+ ereport(ERROR, (errmsg("could not read WAL record")));
+ }
+
+ /*
+ * Whether we could read new record or not, keep checking if
+ * 'lsn_upto' was specified.
+ */
+ if (XLogRecPtrIsInvalid(lsn_upto))
+ {
+ SpinLockAcquire(&shared->mutex);
+ lsn_upto = shared->lsn_upto;
+ /* 'done' should be set at the same time as 'lsn_upto' */
+ done = shared->done;
+ SpinLockRelease(&shared->mutex);
+ }
+ if (!XLogRecPtrIsInvalid(lsn_upto) &&
+ ctx->reader->EndRecPtr >= lsn_upto)
+ break;
+
+ if (record == NULL)
+ {
+ int64 timeout = 0;
+ WaitLSNResult res;
+
+ /*
+ * Before we retry reading, wait until new WAL is flushed.
+ *
+ * There is a race condition such that the backend executing
+ * REPACK determines 'lsn_upto', but before it sets the shared
+ * variable, we reach the end of WAL. In that case we'd need to
+ * wait until the next WAL flush (unrelated to REPACK). Although
+ * that should not be a problem in a busy system, it might be
+ * noticeable in other cases, including regression tests (which
+ * are not necessarily executed in parallel). Therefore it makes
+ * sense to use timeout.
+ *
+ * If lsn_upto is valid, WAL records having LSN lower than that
+ * should already have been flushed to disk.
+ */
+ if (XLogRecPtrIsInvalid(lsn_upto))
+ timeout = 100L;
+ res = WaitForLSN(WAIT_LSN_TYPE_PRIMARY_FLUSH,
+ ctx->reader->EndRecPtr + 1,
+ timeout);
+ if (res != WAIT_LSN_RESULT_SUCCESS &&
+ res != WAIT_LSN_RESULT_TIMEOUT)
+ ereport(ERROR, (errmsg("waiting for WAL failed")));
}
- InvalidateSystemCaches();
- CurrentResourceOwner = resowner_old;
- }
- PG_CATCH();
- {
- /* clear all timetravel entries */
- InvalidateSystemCaches();
- CurrentResourceOwner = resowner_old;
- PG_RE_THROW();
}
- PG_END_TRY();
+
+ /*
+ * Close the file so we can make it available to the backend.
+ */
+ BufFileClose(dstate->file);
+ dstate->file = NULL;
+ SpinLockAcquire(&shared->mutex);
+ shared->lsn_upto = InvalidXLogRecPtr;
+ shared->last_exported++;
+ SpinLockRelease(&shared->mutex);
+ ConditionVariableSignal(&shared->cv);
+
+ return done;
}
/*
* Apply changes stored in 'file'.
*/
static void
-apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
+apply_concurrent_changes(BufFile *file, ChangeDest *dest)
{
+ char kind;
+ uint32 t_len;
Relation rel = dest->rel;
TupleTableSlot *index_slot,
*ident_slot;
HeapTuple tup_old = NULL;
- if (dstate->nchanges == 0)
- return;
-
/* TupleTableSlot is needed to pass the tuple to ExecInsertIndexTuples(). */
- index_slot = MakeSingleTupleTableSlot(dstate->tupdesc, &TTSOpsHeapTuple);
+ index_slot = MakeSingleTupleTableSlot(RelationGetDescr(rel),
+ &TTSOpsHeapTuple);
/* A slot to fetch tuples from identity index. */
ident_slot = table_slot_create(rel, NULL);
- while (tuplestore_gettupleslot(dstate->tstore, true, false,
- dstate->tsslot))
+ while (true)
{
- bool shouldFree;
- HeapTuple tup_change,
- tup,
+ size_t nread;
+ HeapTuple tup,
tup_exist;
- char *change_raw,
- *src;
- ConcurrentChange change;
- bool isnull[1];
- Datum values[1];
CHECK_FOR_INTERRUPTS();
- /* Get the change from the single-column tuple. */
- tup_change = ExecFetchSlotHeapTuple(dstate->tsslot, false, &shouldFree);
- heap_deform_tuple(tup_change, dstate->tupdesc_change, values, isnull);
- Assert(!isnull[0]);
-
- /* Make sure we access aligned data. */
- change_raw = (char *) DatumGetByteaP(values[0]);
- src = (char *) VARDATA(change_raw);
- memcpy(&change, src, SizeOfConcurrentChange);
+ nread = BufFileReadMaybeEOF(file, &kind, 1, true);
+ /* Are we done with the file? */
+ if (nread == 0)
+ break;
- /*
- * Extract the tuple from the change. The tuple is copied here because
- * it might be assigned to 'tup_old', in which case it needs to
- * survive into the next iteration.
- */
- tup = get_changed_tuple(src);
+ /* Read the tuple. */
+ BufFileReadExact(file, &t_len, sizeof(t_len));
+ tup = (HeapTuple) palloc(HEAPTUPLESIZE + t_len);
+ tup->t_data = (HeapTupleHeader) ((char *) tup + HEAPTUPLESIZE);
+ BufFileReadExact(file, tup->t_data, t_len);
+ tup->t_len = t_len;
+ ItemPointerSetInvalid(&tup->t_self);
+ tup->t_tableOid = RelationGetRelid(dest->rel);
- if (change.kind == CHANGE_UPDATE_OLD)
+ if (kind == CHANGE_UPDATE_OLD)
{
Assert(tup_old == NULL);
tup_old = tup;
}
- else if (change.kind == CHANGE_INSERT)
+ else if (kind == CHANGE_INSERT)
{
Assert(tup_old == NULL);
@@ -2726,12 +2869,11 @@ apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
pfree(tup);
}
- else if (change.kind == CHANGE_UPDATE_NEW ||
- change.kind == CHANGE_DELETE)
+ else if (kind == CHANGE_UPDATE_NEW || kind == CHANGE_DELETE)
{
HeapTuple tup_key;
- if (change.kind == CHANGE_UPDATE_NEW)
+ if (kind == CHANGE_UPDATE_NEW)
{
tup_key = tup_old != NULL ? tup_old : tup;
}
@@ -2748,7 +2890,7 @@ apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
if (tup_exist == NULL)
elog(ERROR, "failed to find target tuple");
- if (change.kind == CHANGE_UPDATE_NEW)
+ if (kind == CHANGE_UPDATE_NEW)
apply_concurrent_update(rel, tup, tup_exist, dest->iistate,
index_slot);
else
@@ -2763,26 +2905,19 @@ apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
pfree(tup);
}
else
- elog(ERROR, "unrecognized kind of change: %d", change.kind);
+ elog(ERROR, "unrecognized kind of change: %d", kind);
/*
* If a change was applied now, increment CID for next writes and
* update the snapshot so it sees the changes we've applied so far.
*/
- if (change.kind != CHANGE_UPDATE_OLD)
+ if (kind != CHANGE_UPDATE_OLD)
{
CommandCounterIncrement();
UpdateActiveSnapshotCommandId();
}
-
- /* TTSOpsMinimalTuple has .get_heap_tuple==NULL. */
- Assert(shouldFree);
- pfree(tup_change);
}
- tuplestore_clear(dstate->tstore);
- dstate->nchanges = 0;
-
/* Cleanup. */
ExecDropSingleTupleTableSlot(index_slot);
ExecDropSingleTupleTableSlot(ident_slot);
@@ -2957,25 +3092,59 @@ find_target_tuple(Relation rel, ChangeDest *dest, HeapTuple tup_key,
}
/*
- * Decode and apply concurrent changes.
+ * Decode and apply concurrent changes, up to (and including) the record whose
+ * LSN is 'end_of_wal'.
*/
static void
-process_concurrent_changes(LogicalDecodingContext *decoding_ctx,
- XLogRecPtr end_of_wal, ChangeDest *dest)
+process_concurrent_changes(XLogRecPtr end_of_wal, ChangeDest *dest, bool done)
{
- RepackDecodingState *dstate;
+ DecodingWorkerShared *shared;
+ char fname[MAXPGPATH];
+ BufFile *file;
pgstat_progress_update_param(PROGRESS_REPACK_PHASE,
PROGRESS_REPACK_PHASE_CATCH_UP);
- dstate = (RepackDecodingState *) decoding_ctx->output_writer_private;
+ /* Ask the worker for the file. */
+ shared = (DecodingWorkerShared *) dsm_segment_address(decoding_worker->seg);
+ SpinLockAcquire(&shared->mutex);
+ shared->lsn_upto = end_of_wal;
+ shared->done = done;
+ SpinLockRelease(&shared->mutex);
+
+ /*
+ * The worker needs to finish processing of the current WAL record. Even
+ * if it's idle, it'll need to close the output file. Thus we're likely to
+ * wait, so prepare for sleep.
+ */
+ ConditionVariablePrepareToSleep(&shared->cv);
+ for (;;)
+ {
+ int last_exported;
- repack_decode_concurrent_changes(decoding_ctx, end_of_wal);
+ SpinLockAcquire(&shared->mutex);
+ last_exported = shared->last_exported;
+ SpinLockRelease(&shared->mutex);
- if (dstate->nchanges == 0)
- return;
+ /*
+ * Has the worker exported the file we are waiting for?
+ */
+ if (last_exported == dest->file_seq)
+ break;
- apply_concurrent_changes(dstate, dest);
+ ConditionVariableSleep(&shared->cv, WAIT_EVENT_REPACK_WORKER_EXPORT);
+ }
+ ConditionVariableCancelSleep();
+
+ /* Open the file. */
+ DecodingWorkerFileName(fname, shared->relid, dest->file_seq);
+ file = BufFileOpenFileSet(&shared->sfs.fs, fname, O_RDONLY, false);
+ apply_concurrent_changes(file, dest);
+
+ BufFileClose(file);
+
+ /* Get ready for the next file. */
+ dest->file_seq++;
}
/*
@@ -3101,15 +3270,10 @@ cleanup_logical_decoding(LogicalDecodingContext *ctx)
dstate = (RepackDecodingState *) ctx->output_writer_private;
- ExecDropSingleTupleTableSlot(dstate->tsslot);
- FreeTupleDesc(dstate->tupdesc_change);
FreeTupleDesc(dstate->tupdesc);
- tuplestore_end(dstate->tstore);
-
FreeDecodingContext(ctx);
- ReplicationSlotRelease();
- ReplicationSlotDrop(NameStr(dstate->slotname), false);
+ ReplicationSlotDropAcquired();
pfree(dstate);
}
@@ -3123,7 +3287,6 @@ cleanup_logical_decoding(LogicalDecodingContext *ctx)
*/
static void
rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
- LogicalDecodingContext *decoding_ctx,
TransactionId frozenXid,
MultiXactId cutoffMulti)
{
@@ -3204,6 +3367,7 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
&chgdst.ident_index);
chgdst.ident_key = build_identity_key(ident_idx_new, OldHeap,
&chgdst.ident_key_nentries);
+ chgdst.file_seq = WORKER_FILE_SNAPSHOT + 1;
/*
* During testing, wait for another backend to perform concurrent data
@@ -3225,7 +3389,7 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
* hold AccessExclusiveLock. (Quite some amount of WAL could have been
* written during the data copying and index creation.)
*/
- process_concurrent_changes(decoding_ctx, end_of_wal, &chgdst);
+ process_concurrent_changes(end_of_wal, &chgdst, false);
/*
* Acquire AccessExclusiveLock on the table, its TOAST relation (if there
@@ -3306,8 +3470,11 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
XLogFlush(wal_insert_ptr);
end_of_wal = GetFlushRecPtr(NULL);
- /* Apply the concurrent changes again. */
- process_concurrent_changes(decoding_ctx, end_of_wal, &chgdst);
+ /*
+ * Apply the concurrent changes again. Indicate that the decoding worker
+ * won't be needed anymore.
+ */
+ process_concurrent_changes(end_of_wal, &chgdst, true);
/* Remember info about rel before closing OldHeap */
relpersistence = OldHeap->rd_rel->relpersistence;
@@ -3417,3 +3584,510 @@ build_new_indexes(Relation NewHeap, Relation OldHeap, List *OldIndexes)
return result;
}
+
+/*
+ * Try to start a background worker to perform logical decoding of data
+ * changes applied to relation while REPACK CONCURRENTLY is copying its
+ * contents to a new table.
+ */
+static void
+start_decoding_worker(Oid relid)
+{
+ Size size;
+ dsm_segment *seg;
+ DecodingWorkerShared *shared;
+ shm_mq *mq;
+ shm_mq_handle *mqh;
+ BackgroundWorker bgw;
+
+ /* Setup shared memory. */
+ size = BUFFERALIGN(offsetof(DecodingWorkerShared, error_queue)) +
+ BUFFERALIGN(REPACK_ERROR_QUEUE_SIZE);
+ seg = dsm_create(size, 0);
+ shared = (DecodingWorkerShared *) dsm_segment_address(seg);
+ shared->lsn_upto = InvalidXLogRecPtr;
+ shared->done = false;
+ SharedFileSetInit(&shared->sfs, seg);
+ shared->last_exported = -1;
+ SpinLockInit(&shared->mutex);
+ shared->dbid = MyDatabaseId;
+
+ /*
+ * This is the UserId set in cluster_rel(). Security context shouldn't be
+ * needed for decoding worker.
+ */
+ shared->roleid = GetUserId();
+ shared->relid = relid;
+ ConditionVariableInit(&shared->cv);
+ shared->backend_proc = MyProc;
+ shared->backend_pid = MyProcPid;
+ shared->backend_proc_number = MyProcNumber;
+
+ mq = shm_mq_create((char *) BUFFERALIGN(shared->error_queue),
+ REPACK_ERROR_QUEUE_SIZE);
+ shm_mq_set_receiver(mq, MyProc);
+ mqh = shm_mq_attach(mq, seg, NULL);
+
+ memset(&bgw, 0, sizeof(bgw));
+ snprintf(bgw.bgw_name, BGW_MAXLEN,
+ "REPACK decoding worker for relation \"%s\"",
+ get_rel_name(relid));
+ snprintf(bgw.bgw_type, BGW_MAXLEN, "REPACK decoding worker");
+ bgw.bgw_flags = BGWORKER_SHMEM_ACCESS |
+ BGWORKER_BACKEND_DATABASE_CONNECTION;
+ bgw.bgw_start_time = BgWorkerStart_RecoveryFinished;
+ bgw.bgw_restart_time = BGW_NEVER_RESTART;
+ snprintf(bgw.bgw_library_name, MAXPGPATH, "postgres");
+ snprintf(bgw.bgw_function_name, BGW_MAXLEN, "RepackWorkerMain");
+ bgw.bgw_main_arg = UInt32GetDatum(dsm_segment_handle(seg));
+ bgw.bgw_notify_pid = MyProcPid;
+
+ decoding_worker = palloc0_object(DecodingWorker);
+ if (!RegisterDynamicBackgroundWorker(&bgw, &decoding_worker->handle))
+ ereport(ERROR,
+ (errcode(ERRCODE_CONFIGURATION_LIMIT_EXCEEDED),
+ errmsg("out of background worker slots"),
+ errhint("You might need to increase \"%s\".", "max_worker_processes")));
+
+ decoding_worker->seg = seg;
+ decoding_worker->error_mqh = mqh;
+
+ /*
+ * The decoding setup must be done before the caller can have XID assigned
+ * for any reason, otherwise the worker might end up in a deadlock,
+ * waiting for the caller's transaction to end. Therefore wait here until
+ * the worker indicates that it has the logical decoding initialized.
+ */
+ ConditionVariablePrepareToSleep(&shared->cv);
+ for (;;)
+ {
+ int initialized;
+
+ SpinLockAcquire(&shared->mutex);
+ initialized = shared->initialized;
+ SpinLockRelease(&shared->mutex);
+
+ if (initialized)
+ break;
+
+ ConditionVariableSleep(&shared->cv, WAIT_EVENT_REPACK_WORKER_EXPORT);
+ }
+ ConditionVariableCancelSleep();
+}
+
+/*
+ * Stop the decoding worker and cleanup the related resources.
+ *
+ * The worker stops on its own when it knows there is no more work to do, but
+ * we need to stop it explicitly at least on ERROR in the launching backend.
+ */
+static void
+stop_decoding_worker(void)
+{
+ BgwHandleStatus status;
+
+ /* Haven't reached the worker startup? */
+ if (decoding_worker == NULL)
+ return;
+
+ /* Could not register the worker? */
+ if (decoding_worker->handle == NULL)
+ return;
+
+ TerminateBackgroundWorker(decoding_worker->handle);
+ /* The worker should really exit before the REPACK command does. */
+ HOLD_INTERRUPTS();
+ status = WaitForBackgroundWorkerShutdown(decoding_worker->handle);
+ RESUME_INTERRUPTS();
+
+ if (status == BGWH_POSTMASTER_DIED)
+ ereport(FATAL,
+ (errcode(ERRCODE_ADMIN_SHUTDOWN),
+ errmsg("postmaster exited during REPACK command")));
+
+ shm_mq_detach(decoding_worker->error_mqh);
+
+ /*
+ * If we could not cancel the current sleep due to ERROR, do that before
+ * we detach from the shared memory the condition variable is located in.
+ * If we did not, the bgworker ERROR handling code would try and fail
+ * badly.
+ */
+ ConditionVariableCancelSleep();
+
+ dsm_detach(decoding_worker->seg);
+ pfree(decoding_worker);
+ decoding_worker = NULL;
+}
+
+/* Is this process a REPACK worker? */
+static bool is_repack_worker = false;
+
+static pid_t backend_pid;
+static ProcNumber backend_proc_number;
+
+/*
+ * See ParallelWorkerShutdown for details.
+ */
+static void
+RepackWorkerShutdown(int code, Datum arg)
+{
+ SendProcSignal(backend_pid,
+ PROCSIG_REPACK_MESSAGE,
+ backend_proc_number);
+
+ dsm_detach((dsm_segment *) DatumGetPointer(arg));
+}
+
+/* REPACK decoding worker entry point */
+void
+RepackWorkerMain(Datum main_arg)
+{
+ dsm_segment *seg;
+ DecodingWorkerShared *shared;
+ shm_mq *mq;
+ shm_mq_handle *mqh;
+
+ is_repack_worker = true;
+
+ /*
+ * Override the default bgworker_die() with die() so we can use
+ * CHECK_FOR_INTERRUPTS().
+ */
+ pqsignal(SIGTERM, die);
+ BackgroundWorkerUnblockSignals();
+
+ seg = dsm_attach(DatumGetUInt32(main_arg));
+ if (seg == NULL)
+ ereport(ERROR,
+ (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
+ errmsg("could not map dynamic shared memory segment")));
+
+ shared = (DecodingWorkerShared *) dsm_segment_address(seg);
+
+ /* Arrange to signal the leader if we exit. */
+ backend_pid = shared->backend_pid;
+ backend_proc_number = shared->backend_proc_number;
+ before_shmem_exit(RepackWorkerShutdown, PointerGetDatum(seg));
+
+ /*
+ * Join locking group - see the comments around the call of
+ * start_decoding_worker().
+ */
+ if (!BecomeLockGroupMember(shared->backend_proc, backend_pid))
+ /* The leader is not running anymore. */
+ return;
+
+ /*
+ * Setup a queue to send error messages to the backend that launched this
+ * worker.
+ */
+ mq = (shm_mq *) (char *) BUFFERALIGN(shared->error_queue);
+ shm_mq_set_sender(mq, MyProc);
+ mqh = shm_mq_attach(mq, seg, NULL);
+ pq_redirect_to_shm_mq(seg, mqh);
+ pq_set_parallel_leader(shared->backend_pid,
+ shared->backend_proc_number);
+
+ /* Connect to the database. */
+ BackgroundWorkerInitializeConnectionByOid(shared->dbid, shared->roleid, 0);
+
+ repack_worker_internal(seg);
+}
+
+static void
+repack_worker_internal(dsm_segment *seg)
+{
+ DecodingWorkerShared *shared;
+ LogicalDecodingContext *decoding_ctx;
+ SharedFileSet *sfs;
+ Snapshot snapshot;
+
+ /*
+ * Transaction is needed to open relation, and it also provides us with a
+ * resource owner.
+ */
+ StartTransactionCommand();
+
+ shared = (DecodingWorkerShared *) dsm_segment_address(seg);
+
+ /*
+ * Not sure the spinlock is needed here - the backend should not change
+ * anything in the shared memory until we have serialized the snapshot.
+ */
+ SpinLockAcquire(&shared->mutex);
+ Assert(XLogRecPtrIsInvalid(shared->lsn_upto));
+ sfs = &shared->sfs;
+ SpinLockRelease(&shared->mutex);
+
+ SharedFileSetAttach(sfs, seg);
+
+ /*
+ * Prepare to capture the concurrent data changes ourselves.
+ */
+ decoding_ctx = setup_logical_decoding(shared->relid);
+
+ /* Announce that we're ready. */
+ SpinLockAcquire(&shared->mutex);
+ shared->initialized = true;
+ SpinLockRelease(&shared->mutex);
+ ConditionVariableSignal(&shared->cv);
+
+ /* Build the initial snapshot and export it. */
+ snapshot = SnapBuildInitialSnapshotForRepack(decoding_ctx->snapshot_builder);
+ export_initial_snapshot(snapshot, shared);
+
+ /*
+ * Only historic snapshots should be used now. Do not let us restrict the
+ * progress of xmin horizon.
+ */
+ InvalidateCatalogSnapshot();
+
+ while (!decode_concurrent_changes(decoding_ctx, shared))
+ ;
+
+ /* Cleanup. */
+ cleanup_logical_decoding(decoding_ctx);
+ CommitTransactionCommand();
+}
+
+/*
+ * Make snapshot available to the backend that launched the decoding worker.
+ */
+static void
+export_initial_snapshot(Snapshot snapshot, DecodingWorkerShared *shared)
+{
+ char fname[MAXPGPATH];
+ BufFile *file;
+ Size snap_size;
+ char *snap_space;
+
+ snap_size = EstimateSnapshotSpace(snapshot);
+ snap_space = (char *) palloc(snap_size);
+ SerializeSnapshot(snapshot, snap_space);
+ FreeSnapshot(snapshot);
+
+ DecodingWorkerFileName(fname, shared->relid, shared->last_exported + 1);
+ file = BufFileCreateFileSet(&shared->sfs.fs, fname);
+ /* To make restoration easier, write the snapshot size first. */
+ BufFileWrite(file, &snap_size, sizeof(snap_size));
+ BufFileWrite(file, snap_space, snap_size);
+ pfree(snap_space);
+ BufFileClose(file);
+
+ /* Increase the counter to tell the backend that the file is available. */
+ SpinLockAcquire(&shared->mutex);
+ shared->last_exported++;
+ SpinLockRelease(&shared->mutex);
+ ConditionVariableSignal(&shared->cv);
+}
+
+/*
+ * Get the initial snapshot from the decoding worker.
+ */
+static Snapshot
+get_initial_snapshot(DecodingWorker *worker)
+{
+ DecodingWorkerShared *shared;
+ char fname[MAXPGPATH];
+ BufFile *file;
+ Size snap_size;
+ char *snap_space;
+ Snapshot snapshot;
+
+ shared = (DecodingWorkerShared *) dsm_segment_address(worker->seg);
+
+ /*
+ * The worker needs to initialize the logical decoding, which usually
+ * takes some time. Therefore it makes sense to prepare for the sleep
+ * first.
+ */
+ ConditionVariablePrepareToSleep(&shared->cv);
+ for (;;)
+ {
+ int last_exported;
+
+ SpinLockAcquire(&shared->mutex);
+ last_exported = shared->last_exported;
+ SpinLockRelease(&shared->mutex);
+
+ /*
+ * Has the worker exported the file we are waiting for?
+ */
+ if (last_exported == WORKER_FILE_SNAPSHOT)
+ break;
+
+ ConditionVariableSleep(&shared->cv, WAIT_EVENT_REPACK_WORKER_EXPORT);
+ }
+ ConditionVariableCancelSleep();
+
+ /* Read the snapshot from a file. */
+ DecodingWorkerFileName(fname, shared->relid, WORKER_FILE_SNAPSHOT);
+ file = BufFileOpenFileSet(&shared->sfs.fs, fname, O_RDONLY, false);
+ BufFileReadExact(file, &snap_size, sizeof(snap_size));
+ snap_space = (char *) palloc(snap_size);
+ BufFileReadExact(file, snap_space, snap_size);
+ BufFileClose(file);
+
+ /* Restore it. */
+ snapshot = RestoreSnapshot(snap_space);
+ pfree(snap_space);
+
+ return snapshot;
+}
+
+bool
+IsRepackWorker(void)
+{
+ return is_repack_worker;
+}
+
+/*
+ * Handle receipt of an interrupt indicating a repack worker message.
+ *
+ * Note: this is called within a signal handler! All we can do is set
+ * a flag that will cause the next CHECK_FOR_INTERRUPTS() to invoke
+ * ProcessRepackMessages().
+ */
+void
+HandleRepackMessageInterrupt(void)
+{
+ InterruptPending = true;
+ RepackMessagePending = true;
+ SetLatch(MyLatch);
+}
+
+/*
+ * Process any queued protocol messages received from parallel workers.
+ */
+void
+ProcessRepackMessages(void)
+{
+ MemoryContext oldcontext;
+
+ static MemoryContext hpm_context = NULL;
+
+ /*
+ * Nothing to do if we haven't launched the worker yet or have already
+ * terminated it.
+ */
+ if (decoding_worker == NULL)
+ return;
+
+ /*
+ * This is invoked from ProcessInterrupts(), and since some of the
+ * functions it calls contain CHECK_FOR_INTERRUPTS(), there is a potential
+ * for recursive calls if more signals are received while this runs. It's
+ * unclear that recursive entry would be safe, and it doesn't seem useful
+ * even if it is safe, so let's block interrupts until done.
+ */
+ HOLD_INTERRUPTS();
+
+ /*
+ * Moreover, CurrentMemoryContext might be pointing almost anywhere. We
+ * don't want to risk leaking data into long-lived contexts, so let's do
+ * our work here in a private context that we can reset on each use.
+ */
+ if (hpm_context == NULL) /* first time through? */
+ hpm_context = AllocSetContextCreate(TopMemoryContext,
+ "ProcessRepackMessages",
+ ALLOCSET_DEFAULT_SIZES);
+ else
+ MemoryContextReset(hpm_context);
+
+ oldcontext = MemoryContextSwitchTo(hpm_context);
+
+ /* OK to process messages. Reset the flag saying there are more to do. */
+ RepackMessagePending = false;
+
+ /*
+ * Read as many messages as we can from each worker, but stop when no more
+ * messages can be read from the worker without blocking.
+ */
+ while (true)
+ {
+ shm_mq_result res;
+ Size nbytes;
+ void *data;
+
+ res = shm_mq_receive(decoding_worker->error_mqh, &nbytes,
+ &data, true);
+ if (res == SHM_MQ_WOULD_BLOCK)
+ break;
+ else if (res == SHM_MQ_SUCCESS)
+ {
+ StringInfoData msg;
+
+ initStringInfo(&msg);
+ appendBinaryStringInfo(&msg, data, nbytes);
+ ProcessRepackMessage(&msg);
+ pfree(msg.data);
+ }
+ else
+ {
+ /*
+ * The decoding worker is special in that it exits as soon as it
+ * has its work done. Thus the DETACHED result code is fine.
+ */
+ Assert(res == SHM_MQ_DETACHED);
+
+ break;
+ }
+ }
+
+ MemoryContextSwitchTo(oldcontext);
+
+ /* Might as well clear the context on our way out */
+ MemoryContextReset(hpm_context);
+
+ RESUME_INTERRUPTS();
+}
+
+/*
+ * Process a single protocol message received from a single parallel worker.
+ */
+static void
+ProcessRepackMessage(StringInfo msg)
+{
+ char msgtype;
+
+ msgtype = pq_getmsgbyte(msg);
+
+ switch (msgtype)
+ {
+ case PqMsg_ErrorResponse:
+ case PqMsg_NoticeResponse:
+ {
+ ErrorData edata;
+
+ /* Parse ErrorResponse or NoticeResponse. */
+ pq_parse_errornotice(msg, &edata);
+
+ /* Death of a worker isn't enough justification for suicide. */
+ edata.elevel = Min(edata.elevel, ERROR);
+
+ /*
+ * If desired, add a context line to show that this is a
+ * message propagated from a parallel worker. Otherwise, it
+ * can sometimes be confusing to understand what actually
+ * happened.
+ */
+ if (edata.context)
+ edata.context = psprintf("%s\n%s", edata.context,
+ _("decoding worker"));
+ else
+ edata.context = pstrdup(_("decoding worker"));
+
+ /* Rethrow error or print notice. */
+ ThrowErrorData(&edata);
+
+ break;
+ }
+
+ default:
+ {
+ elog(ERROR, "unrecognized message type received from decoding worker: %c (message length %d bytes)",
+ msgtype, msg->len);
+ }
+ }
+}
diff --git a/src/backend/libpq/pqmq.c b/src/backend/libpq/pqmq.c
index 6e4bbfb5aa1..42f6fa472c5 100644
--- a/src/backend/libpq/pqmq.c
+++ b/src/backend/libpq/pqmq.c
@@ -14,6 +14,7 @@
#include "postgres.h"
#include "access/parallel.h"
+#include "commands/cluster.h"
#include "libpq/libpq.h"
#include "libpq/pqformat.h"
#include "libpq/pqmq.h"
@@ -175,6 +176,10 @@ mq_putmessage(char msgtype, const char *s, size_t len)
SendProcSignal(pq_mq_parallel_leader_pid,
PROCSIG_PARALLEL_APPLY_MESSAGE,
pq_mq_parallel_leader_proc_number);
+ else if (IsRepackWorker())
+ SendProcSignal(pq_mq_parallel_leader_pid,
+ PROCSIG_REPACK_MESSAGE,
+ pq_mq_parallel_leader_proc_number);
else
{
Assert(IsParallelWorker());
diff --git a/src/backend/postmaster/bgworker.c b/src/backend/postmaster/bgworker.c
index 65deabe91a7..334bb708c5b 100644
--- a/src/backend/postmaster/bgworker.c
+++ b/src/backend/postmaster/bgworker.c
@@ -13,6 +13,7 @@
#include "postgres.h"
#include "access/parallel.h"
+#include "commands/cluster.h"
#include "libpq/pqsignal.h"
#include "miscadmin.h"
#include "pgstat.h"
@@ -136,6 +137,9 @@ static const struct
},
{
"SequenceSyncWorkerMain", SequenceSyncWorkerMain
+ },
+ {
+ "RepackWorkerMain", RepackWorkerMain
}
};
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index b0ef1a12520..35a46988285 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -194,7 +194,11 @@ StartupDecodingContext(List *output_plugin_options,
ctx->slot = slot;
- ctx->reader = XLogReaderAllocate(wal_segment_size, NULL, xl_routine, ctx);
+ /*
+ * TODO A separate patch for PG core, unless there's really a reason to
+ * pass ctx for private_data (May extensions expect ctx?).
+ */
+ ctx->reader = XLogReaderAllocate(wal_segment_size, NULL, xl_routine, NULL);
if (!ctx->reader)
ereport(ERROR,
(errcode(ERRCODE_OUT_OF_MEMORY),
diff --git a/src/backend/replication/pgoutput_repack/pgoutput_repack.c b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
index a074eed393d..4bc47a72371 100644
--- a/src/backend/replication/pgoutput_repack/pgoutput_repack.c
+++ b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
@@ -169,17 +169,13 @@ store_change(LogicalDecodingContext *ctx, ConcurrentChangeKind kind,
HeapTuple tuple)
{
RepackDecodingState *dstate;
- char *change_raw;
- ConcurrentChange change;
+ char kind_byte = (char) kind;
bool flattened = false;
- Size size;
- Datum values[1];
- bool isnull[1];
- char *dst;
dstate = (RepackDecodingState *) ctx->output_writer_private;
- size = VARHDRSZ + SizeOfConcurrentChange;
+ /* Store the change kind. */
+ BufFileWrite(dstate->file, &kind_byte, 1);
/*
* ReorderBufferCommit() stores the TOAST chunks in its private memory
@@ -196,46 +192,12 @@ store_change(LogicalDecodingContext *ctx, ConcurrentChangeKind kind,
tuple = toast_flatten_tuple(tuple, dstate->tupdesc);
flattened = true;
}
+ /* Store the tuple size ... */
+ BufFileWrite(dstate->file, &tuple->t_len, sizeof(tuple->t_len));
+ /* ... and the tuple itself. */
+ BufFileWrite(dstate->file, tuple->t_data, tuple->t_len);
- size += tuple->t_len;
- if (size >= MaxAllocSize)
- elog(ERROR, "Change is too big.");
-
- /* Construct the change. */
- change_raw = (char *) palloc0(size);
- SET_VARSIZE(change_raw, size);
-
- /*
- * Since the varlena alignment might not be sufficient for the structure,
- * set the fields in a local instance and remember where it should
- * eventually be copied.
- */
- change.kind = kind;
- dst = (char *) VARDATA(change_raw);
-
- /*
- * Copy the tuple.
- *
- * Note: change->tup_data.t_data must be fixed on retrieval!
- */
- memcpy(&change.tup_data, tuple, sizeof(HeapTupleData));
- memcpy(dst, &change, SizeOfConcurrentChange);
- dst += SizeOfConcurrentChange;
- memcpy(dst, tuple->t_data, tuple->t_len);
-
- /* The data has been copied. */
+ /* Free the flat copy if created above. */
if (flattened)
pfree(tuple);
-
- /* Store as tuple of 1 bytea column. */
- values[0] = PointerGetDatum(change_raw);
- isnull[0] = false;
- tuplestore_putvalues(dstate->tstore, dstate->tupdesc_change,
- values, isnull);
-
- /* Accounting. */
- dstate->nchanges++;
-
- /* Cleanup. */
- pfree(change_raw);
}
diff --git a/src/backend/storage/ipc/procsignal.c b/src/backend/storage/ipc/procsignal.c
index 8e56922dcea..6f9e7a7aab7 100644
--- a/src/backend/storage/ipc/procsignal.c
+++ b/src/backend/storage/ipc/procsignal.c
@@ -19,6 +19,7 @@
#include "access/parallel.h"
#include "commands/async.h"
+#include "commands/cluster.h"
#include "miscadmin.h"
#include "pgstat.h"
#include "port/pg_bitutils.h"
@@ -697,6 +698,9 @@ procsignal_sigusr1_handler(SIGNAL_ARGS)
if (CheckProcSignal(PROCSIG_PARALLEL_APPLY_MESSAGE))
HandleParallelApplyMessageInterrupt();
+ if (CheckProcSignal(PROCSIG_REPACK_MESSAGE))
+ HandleRepackMessageInterrupt();
+
if (CheckProcSignal(PROCSIG_RECOVERY_CONFLICT_DATABASE))
HandleRecoveryConflictInterrupt(PROCSIG_RECOVERY_CONFLICT_DATABASE);
diff --git a/src/backend/tcop/postgres.c b/src/backend/tcop/postgres.c
index e54bf1e760f..fc81ad87615 100644
--- a/src/backend/tcop/postgres.c
+++ b/src/backend/tcop/postgres.c
@@ -36,6 +36,7 @@
#include "access/xact.h"
#include "catalog/pg_type.h"
#include "commands/async.h"
+#include "commands/cluster.h"
#include "commands/event_trigger.h"
#include "commands/explain_state.h"
#include "commands/prepare.h"
@@ -3541,6 +3542,9 @@ ProcessInterrupts(void)
if (ParallelApplyMessagePending)
ProcessParallelApplyMessages();
+
+ if (RepackMessagePending)
+ ProcessRepackMessages();
}
/*
diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt
index 3299de23bb3..73a3def69bc 100644
--- a/src/backend/utils/activity/wait_event_names.txt
+++ b/src/backend/utils/activity/wait_event_names.txt
@@ -62,6 +62,7 @@ LOGICAL_APPLY_MAIN "Waiting in main loop of logical replication apply process."
LOGICAL_LAUNCHER_MAIN "Waiting in main loop of logical replication launcher process."
LOGICAL_PARALLEL_APPLY_MAIN "Waiting in main loop of logical replication parallel apply process."
RECOVERY_WAL_STREAM "Waiting in main loop of startup process for WAL to arrive, during streaming recovery."
+REPACK_WORKER_MAIN "Waiting in main loop of REPACK decoding worker process."
REPLICATION_SLOTSYNC_MAIN "Waiting in main loop of slot synchronization."
REPLICATION_SLOTSYNC_SHUTDOWN "Waiting for slot sync worker to shut down."
SYSLOGGER_MAIN "Waiting in main loop of syslogger process."
@@ -154,6 +155,7 @@ RECOVERY_CONFLICT_SNAPSHOT "Waiting for recovery conflict resolution for a vacuu
RECOVERY_CONFLICT_TABLESPACE "Waiting for recovery conflict resolution for dropping a tablespace."
RECOVERY_END_COMMAND "Waiting for <xref linkend="guc-recovery-end-command"/> to complete."
RECOVERY_PAUSE "Waiting for recovery to be resumed."
+REPACK_WORKER_EXPORT "Waiting for decoding worker to export a new output file."
REPLICATION_ORIGIN_DROP "Waiting for a replication origin to become inactive so it can be dropped."
REPLICATION_SLOT_DROP "Waiting for a replication slot to become inactive so it can be dropped."
RESTORE_COMMAND "Waiting for <xref linkend="guc-restore-command"/> to complete."
diff --git a/src/include/access/tableam.h b/src/include/access/tableam.h
index 76aa993009a..15760363a1a 100644
--- a/src/include/access/tableam.h
+++ b/src/include/access/tableam.h
@@ -22,7 +22,6 @@
#include "access/xact.h"
#include "commands/vacuum.h"
#include "executor/tuptable.h"
-#include "replication/logical.h"
#include "storage/read_stream.h"
#include "utils/rel.h"
#include "utils/snapshot.h"
@@ -631,7 +630,6 @@ typedef struct TableAmRoutine
bool use_sort,
TransactionId OldestXmin,
Snapshot snapshot,
- LogicalDecodingContext *decoding_ctx,
TransactionId *xid_cutoff,
MultiXactId *multi_cutoff,
double *num_tuples,
@@ -1651,8 +1649,6 @@ table_relation_copy_data(Relation rel, const RelFileLocator *newrlocator)
* - *multi_cutoff - ditto
* - snapshot - if != NULL, ignore data changes done by transactions that this
* (MVCC) snapshot considers still in-progress or in the future.
- * - decoding_ctx - logical decoding context, to capture concurrent data
- * changes.
*
* Output parameters:
* - *xid_cutoff - rel's new relfrozenxid value, may be invalid
@@ -1666,7 +1662,6 @@ table_relation_copy_for_cluster(Relation OldTable, Relation NewTable,
bool use_sort,
TransactionId OldestXmin,
Snapshot snapshot,
- LogicalDecodingContext *decoding_ctx,
TransactionId *xid_cutoff,
MultiXactId *multi_cutoff,
double *num_tuples,
@@ -1675,7 +1670,7 @@ table_relation_copy_for_cluster(Relation OldTable, Relation NewTable,
{
OldTable->rd_tableam->relation_copy_for_cluster(OldTable, NewTable, OldIndex,
use_sort, OldestXmin,
- snapshot, decoding_ctx,
+ snapshot,
xid_cutoff, multi_cutoff,
num_tuples, tups_vacuumed,
tups_recently_dead);
diff --git a/src/include/commands/cluster.h b/src/include/commands/cluster.h
index 6a5c476294a..1b05d5d418b 100644
--- a/src/include/commands/cluster.h
+++ b/src/include/commands/cluster.h
@@ -17,11 +17,13 @@
#include "nodes/parsenodes.h"
#include "parser/parse_node.h"
#include "replication/decode.h"
+#include "postmaster/bgworker.h"
#include "replication/logical.h"
+#include "storage/buffile.h"
#include "storage/lock.h"
+#include "storage/shm_mq.h"
#include "utils/relcache.h"
#include "utils/resowner.h"
-#include "utils/tuplestore.h"
/* flag bits for ClusterParams->options */
@@ -44,6 +46,9 @@ typedef struct ClusterParams
* The following definitions are used by REPACK CONCURRENTLY.
*/
+/*
+ * Stored as a single byte in the output file.
+ */
typedef enum
{
CHANGE_INSERT,
@@ -52,68 +57,30 @@ typedef enum
CHANGE_DELETE
} ConcurrentChangeKind;
-typedef struct ConcurrentChange
-{
- /* See the enum above. */
- ConcurrentChangeKind kind;
-
- /*
- * The actual tuple.
- *
- * The tuple data follows the ConcurrentChange structure. Before use make
- * sure the tuple is correctly aligned (ConcurrentChange can be stored as
- * bytea) and that tuple->t_data is fixed.
- */
- HeapTupleData tup_data;
-} ConcurrentChange;
-
-#define SizeOfConcurrentChange (offsetof(ConcurrentChange, tup_data) + \
- sizeof(HeapTupleData))
-
/*
* Logical decoding state.
*
- * Here we store the data changes that we decode from WAL while the table
- * contents is being copied to a new storage. Also the necessary metadata
- * needed to apply these changes to the table is stored here.
+ * The output plugin uses it to store the data changes that it decodes from
+ * WAL while the table contents is being copied to a new storage.
*/
typedef struct RepackDecodingState
{
/* The relation whose changes we're decoding. */
Oid relid;
- /* Replication slot name. */
- NameData slotname;
-
- /*
- * Decoded changes are stored here. Although we try to avoid excessive
- * batches, it can happen that the changes need to be stored to disk. The
- * tuplestore does this transparently.
- */
- Tuplestorestate *tstore;
-
- /* The current number of changes in tstore. */
- double nchanges;
-
- /*
- * Descriptor to store the ConcurrentChange structure serialized (bytea).
- * We can't store the tuple directly because tuplestore only supports
- * minimum tuple and we may need to transfer OID system column from the
- * output plugin. Also we need to transfer the change kind, so it's better
- * to put everything in the structure than to use 2 tuplestores "in
- * parallel".
- */
- TupleDesc tupdesc_change;
-
- /* Tuple descriptor needed to update indexes. */
+ /* Tuple descriptor of the relation being processed. */
TupleDesc tupdesc;
- /* Slot to retrieve data from tstore. */
- TupleTableSlot *tsslot;
-
- ResourceOwner resowner;
+ /* The current output file. */
+ BufFile *file;
} RepackDecodingState;
+extern PGDLLIMPORT volatile sig_atomic_t RepackMessagePending;
+
+extern bool IsRepackWorker(void);
+extern void HandleRepackMessageInterrupt(void);
+extern void ProcessRepackMessages(void);
+
extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel);
extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid,
@@ -136,6 +103,6 @@ extern void finish_heap_swap(Oid OIDOldHeap, Oid OIDNewHeap,
extern bool am_decoding_for_repack(void);
extern bool change_useless_for_repack(XLogRecordBuffer *buf);
-extern void repack_decode_concurrent_changes(LogicalDecodingContext *ctx,
- XLogRecPtr end_of_wal);
+
+extern void RepackWorkerMain(Datum main_arg);
#endif /* CLUSTER_H */
diff --git a/src/include/storage/procsignal.h b/src/include/storage/procsignal.h
index e52b8eb7697..3ef35ca6b80 100644
--- a/src/include/storage/procsignal.h
+++ b/src/include/storage/procsignal.h
@@ -36,6 +36,7 @@ typedef enum
PROCSIG_BARRIER, /* global barrier interrupt */
PROCSIG_LOG_MEMORY_CONTEXT, /* ask backend to log the memory contexts */
PROCSIG_PARALLEL_APPLY_MESSAGE, /* Message from parallel apply workers */
+ PROCSIG_REPACK_MESSAGE, /* Message from repack worker */
/* Recovery conflict reasons */
PROCSIG_RECOVERY_CONFLICT_FIRST,
diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list
index a0b7b38a5e2..d1a694f9008 100644
--- a/src/tools/pgindent/typedefs.list
+++ b/src/tools/pgindent/typedefs.list
@@ -496,7 +496,6 @@ CompressFileHandle
CompressionLocation
CompressorState
ComputeXidHorizonsResult
-ConcurrentChange
ConcurrentChangeKind
ConditionVariable
ConditionVariableMinimallyPadded
@@ -636,6 +635,9 @@ DeclareCursorStmt
DecodedBkpBlock
DecodedXLogRecord
DecodingOutputState
+DecodingWorker
+DecodingWorkerShared
+DecodingWorkerState
DefElem
DefElemAction
DefaultACLInfo
--
2.47.3
--=-=-=
Content-Type: text/plain
Content-Disposition: attachment;
filename=v30-0006-Use-multiple-snapshots-to-copy-the-data.patch
^ permalink raw reply [nested|flat] 55+ messages in thread
* [PATCH 5/6] Use background worker to do logical decoding.
@ 2026-01-12 16:30 Antonin Houska <[email protected]>
0 siblings, 0 replies; 55+ messages in thread
From: Antonin Houska @ 2026-01-12 16:30 UTC (permalink / raw)
If the backend performing REPACK (CONCURRENTLY) does both data copying and
logical decoding, it has to "travel in time" back and forth and therefore it
has to invalidate system caches quite a few times. (The copying and the
decoding work with different catalog snapshots.) As the decoding worker has
separate caches, the switching is not necessary.
Without the worker, it'd also be difficult to switch between potentiallly long
running tasks like index build and WAL decoding. (No decoding during that time
at all can suspend archiving / recycling of WAL segments for some time, which
in turn may result in full disk.)
Another problem is that, after having acquired AccessExclusiveLock (in order
to swap the files), the backend needs to both decode and apply the data
changes that took place while it was waiting for the lock. With the decoding
worker, the decoding runs all the time, so the backend only needs to apply the
changes. This can reduce the time the exclusive lock is held for.
Note that the code added in order to handle ERRORs in the background worker
almost duplicates the existing code that does the same for other types of
workers (See ProcessParallelMessages() and
ProcessParallelApplyMessages()). Refactoring of the existing code might be
useful, to reduce the duplication.
---
src/backend/access/heap/heapam_handler.c | 44 -
src/backend/commands/cluster.c | 1176 +++++++++++++----
src/backend/libpq/pqmq.c | 5 +
src/backend/postmaster/bgworker.c | 4 +
src/backend/replication/logical/logical.c | 6 +-
.../pgoutput_repack/pgoutput_repack.c | 54 +-
src/backend/storage/ipc/procsignal.c | 4 +
src/backend/tcop/postgres.c | 4 +
.../utils/activity/wait_event_names.txt | 2 +
src/include/access/tableam.h | 7 +-
src/include/commands/cluster.h | 71 +-
src/include/storage/procsignal.h | 1 +
src/tools/pgindent/typedefs.list | 4 +-
13 files changed, 981 insertions(+), 401 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index 3526b6adcb5..475c536ce43 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -33,7 +33,6 @@
#include "catalog/index.h"
#include "catalog/storage.h"
#include "catalog/storage_xlog.h"
-#include "commands/cluster.h"
#include "commands/progress.h"
#include "executor/executor.h"
#include "miscadmin.h"
@@ -688,7 +687,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
Relation OldIndex, bool use_sort,
TransactionId OldestXmin,
Snapshot snapshot,
- LogicalDecodingContext *decoding_ctx,
TransactionId *xid_cutoff,
MultiXactId *multi_cutoff,
double *num_tuples,
@@ -710,7 +708,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
BufferHeapTupleTableSlot *hslot;
BlockNumber prev_cblock = InvalidBlockNumber;
bool concurrent = snapshot != NULL;
- XLogRecPtr end_of_wal_prev = GetFlushRecPtr(NULL);
/* Remember if it's a system catalog */
is_system_catalog = IsSystemRelation(OldHeap);
@@ -957,31 +954,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
ct_val[1] = *num_tuples;
pgstat_progress_update_multi_param(2, ct_index, ct_val);
}
-
- /*
- * Process the WAL produced by the load, as well as by other
- * transactions, so that the replication slot can advance and WAL does
- * not pile up. Use wal_segment_size as a threshold so that we do not
- * introduce the decoding overhead too often.
- *
- * Of course, we must not apply the changes until the initial load has
- * completed.
- *
- * Note that our insertions into the new table should not be decoded
- * as we (intentionally) do not write the logical decoding specific
- * information to WAL.
- */
- if (concurrent)
- {
- XLogRecPtr end_of_wal;
-
- end_of_wal = GetFlushRecPtr(NULL);
- if ((end_of_wal - end_of_wal_prev) > wal_segment_size)
- {
- repack_decode_concurrent_changes(decoding_ctx, end_of_wal);
- end_of_wal_prev = end_of_wal;
- }
- }
}
if (indexScan != NULL)
@@ -1027,22 +999,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
-
- /*
- * Try to keep the amount of not-yet-decoded WAL small, like
- * above.
- */
- if (concurrent)
- {
- XLogRecPtr end_of_wal;
-
- end_of_wal = GetFlushRecPtr(NULL);
- if ((end_of_wal - end_of_wal_prev) > wal_segment_size)
- {
- repack_decode_concurrent_changes(decoding_ctx, end_of_wal);
- end_of_wal_prev = end_of_wal;
- }
- }
}
tuplesort_end(tuplesort);
diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c
index 9f61620a279..49a42f54ae4 100644
--- a/src/backend/commands/cluster.c
+++ b/src/backend/commands/cluster.c
@@ -12,12 +12,13 @@
* In concurrent mode, we lock the table with only ShareUpdateExclusiveLock,
* then do an initial copy as above. However, while the tuples are being
* copied, concurrent transactions could modify the table. To cope with those
- * changes, we rely on logical decoding to obtain them from WAL. The changes
- * are accumulated in a tuplestore. Once the initial copy is complete, we
- * read the changes from the tuplestore and re-apply them on the new heap.
- * Then we upgrade our ShareUpdateExclusiveLock to AccessExclusiveLock and
- * swap the relfilenodes. This way, the time we hold a strong lock on the
- * table is much reduced, and the bloat is eliminated.
+ * changes, we rely on logical decoding to obtain them from WAL. A bgworker
+ * consumes WAL while the initial copy is ongoing (to prevent excessive WAL
+ * from being reserved), and accumulates the changes in a file. Once the
+ * initial copy is complete, we read the changes from the file and re-apply
+ * them on the new heap. Then we upgrade our ShareUpdateExclusiveLock to
+ * AccessExclusiveLock and swap the relfilenodes. This way, the time we hold
+ * a strong lock on the table is much reduced, and the bloat is eliminated.
*
* There is hardly anything left of Paul Brown's original implementation...
*
@@ -45,6 +46,7 @@
#include "access/xlog_internal.h"
#include "access/xloginsert.h"
#include "access/xlogutils.h"
+#include "access/xlogwait.h"
#include "catalog/catalog.h"
#include "catalog/dependency.h"
#include "catalog/heap.h"
@@ -61,6 +63,8 @@
#include "commands/tablecmds.h"
#include "commands/vacuum.h"
#include "executor/executor.h"
+#include "libpq/pqformat.h"
+#include "libpq/pqmq.h"
#include "miscadmin.h"
#include "optimizer/optimizer.h"
#include "pgstat.h"
@@ -71,6 +75,8 @@
#include "storage/ipc.h"
#include "storage/lmgr.h"
#include "storage/predicate.h"
+#include "storage/procsignal.h"
+#include "tcop/tcopprot.h"
#include "utils/acl.h"
#include "utils/fmgroids.h"
#include "utils/guc.h"
@@ -117,6 +123,12 @@ typedef struct IndexInsertState
/* The WAL segment being decoded. */
static XLogSegNo repack_current_segment = 0;
+/*
+ * The first file exported by the decoding worker must contain a snapshot, the
+ * following ones contain the data changes.
+ */
+#define WORKER_FILE_SNAPSHOT 0
+
/*
* Information needed to apply concurrent data changes.
*/
@@ -136,8 +148,113 @@ typedef struct ChangeDest
/* Needed to update indexes of rel_dst. */
IndexInsertState *iistate;
+
+ /*
+ * Sequential number of the file containing the changes.
+ *
+ * TODO This field makes the structure name less descriptive. Should we
+ * rename it, e.g. to ChangeApplyInfo?
+ */
+ int file_seq;
} ChangeDest;
+/*
+ * Layout of shared memory used for communication between backend and the
+ * worker that performs logical decoding of data changes
+ */
+typedef struct DecodingWorkerShared
+{
+ /* Is the decoding initialized? */
+ bool initialized;
+
+ /*
+ * Once the worker has reached this LSN, it should close the current
+ * output file and either create a new one or exit, according to the field
+ * 'done'. If the value is InvalidXLogRecPtr, the worker should decode all
+ * the WAL available and keep checking this field. It is ok if the worker
+ * had already decoded records whose LSN is >= lsn_upto before this field
+ * has been set.
+ */
+ XLogRecPtr lsn_upto;
+
+ /* Exit after closing the current file? */
+ bool done;
+
+ /* The output is stored here. */
+ SharedFileSet sfs;
+
+ /* Number of the last file exported by the worker. */
+ int last_exported;
+
+ /* Synchronize access to the fields above. */
+ slock_t mutex;
+
+ /* Database to connect to. */
+ Oid dbid;
+
+ /* Role to connect as. */
+ Oid roleid;
+
+ /* Decode data changes of this relation. */
+ Oid relid;
+
+ /* The backend uses this to wait for the worker. */
+ ConditionVariable cv;
+
+ /* Info to signal the backend. */
+ PGPROC *backend_proc;
+ pid_t backend_pid;
+ ProcNumber backend_proc_number;
+
+ /* Error queue. */
+ shm_mq *error_mq;
+
+ /*
+ * Memory the queue is located int.
+ *
+ * For considerations on the value see the comments of
+ * PARALLEL_ERROR_QUEUE_SIZE.
+ */
+#define REPACK_ERROR_QUEUE_SIZE 16384
+ char error_queue[FLEXIBLE_ARRAY_MEMBER];
+} DecodingWorkerShared;
+
+/*
+ * Generate worker's output file name. If relations of the same 'relid' happen
+ * to be processed at the same time, they must be from different databases and
+ * therefore different backends must be involved. (PID is already present in
+ * the fileset name.)
+ */
+static inline void
+DecodingWorkerFileName(char *fname, Oid relid, uint32 seq)
+{
+ snprintf(fname, MAXPGPATH, "%u-%u", relid, seq);
+}
+
+/*
+ * Backend-local information to control the decoding worker.
+ */
+typedef struct DecodingWorker
+{
+ /* The worker. */
+ BackgroundWorkerHandle *handle;
+
+ /* DecodingWorkerShared is in this segment. */
+ dsm_segment *seg;
+
+ /* Handle of the error queue. */
+ shm_mq_handle *error_mqh;
+} DecodingWorker;
+
+/* Pointer to currently running decoding worker. */
+static DecodingWorker *decoding_worker = NULL;
+
+/*
+ * Is there a message sent by a repack worker that the backend needs to
+ * receive?
+ */
+volatile sig_atomic_t RepackMessagePending = false;
+
static bool cluster_rel_recheck(RepackCommand cmd, Relation OldHeap,
Oid indexOid, Oid userid, LOCKMODE lmode,
int options);
@@ -145,7 +262,7 @@ static void check_repack_concurrently_requirements(Relation rel);
static void rebuild_relation(Relation OldHeap, Relation index, bool verbose,
bool concurrent);
static void copy_table_data(Relation NewHeap, Relation OldHeap, Relation OldIndex,
- Snapshot snapshot, LogicalDecodingContext *decoding_ctx,
+ Snapshot snapshot,
bool verbose,
bool *pSwapToastByContent,
TransactionId *pFreezeXid,
@@ -158,12 +275,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd,
static bool cluster_is_permitted_for_relation(RepackCommand cmd,
Oid relid, Oid userid);
-static void begin_concurrent_repack(Relation rel);
-static void end_concurrent_repack(void);
static LogicalDecodingContext *setup_logical_decoding(Oid relid);
-static HeapTuple get_changed_tuple(char *change);
-static void apply_concurrent_changes(RepackDecodingState *dstate,
- ChangeDest *dest);
+static bool decode_concurrent_changes(LogicalDecodingContext *ctx,
+ DecodingWorkerShared *shared);
+static void apply_concurrent_changes(BufFile *file, ChangeDest *dest);
static void apply_concurrent_insert(Relation rel, HeapTuple tup,
IndexInsertState *iistate,
TupleTableSlot *index_slot);
@@ -175,9 +290,9 @@ static void apply_concurrent_delete(Relation rel, HeapTuple tup_target);
static HeapTuple find_target_tuple(Relation rel, ChangeDest *dest,
HeapTuple tup_key,
TupleTableSlot *ident_slot);
-static void process_concurrent_changes(LogicalDecodingContext *decoding_ctx,
- XLogRecPtr end_of_wal,
- ChangeDest *dest);
+static void process_concurrent_changes(XLogRecPtr end_of_wal,
+ ChangeDest *dest,
+ bool done);
static IndexInsertState *get_index_insert_state(Relation relation,
Oid ident_index_id,
Relation *ident_index_p);
@@ -186,7 +301,6 @@ static ScanKey build_identity_key(Oid ident_idx_oid, Relation rel_src,
static void free_index_insert_state(IndexInsertState *iistate);
static void cleanup_logical_decoding(LogicalDecodingContext *ctx);
static void rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
- LogicalDecodingContext *decoding_ctx,
TransactionId frozenXid,
MultiXactId cutoffMulti);
static List *build_new_indexes(Relation NewHeap, Relation OldHeap, List *OldIndexes);
@@ -196,6 +310,13 @@ static Relation process_single_relation(RepackStmt *stmt,
ClusterParams *params);
static Oid determine_clustered_index(Relation rel, bool usingindex,
const char *indexname);
+static void start_decoding_worker(Oid relid);
+static void stop_decoding_worker(void);
+static void repack_worker_internal(dsm_segment *seg);
+static void export_initial_snapshot(Snapshot snapshot,
+ DecodingWorkerShared *shared);
+static Snapshot get_initial_snapshot(DecodingWorker *worker);
+static void ProcessRepackMessage(StringInfo msg);
static const char *RepackCommandAsString(RepackCommand cmd);
@@ -619,20 +740,20 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid,
/* rebuild_relation does all the dirty work */
PG_TRY();
{
- /*
- * For concurrent processing, make sure that our logical decoding
- * ignores data changes of other tables than the one we are
- * processing.
- */
- if (concurrent)
- begin_concurrent_repack(OldHeap);
-
rebuild_relation(OldHeap, index, verbose, concurrent);
}
PG_FINALLY();
{
if (concurrent)
- end_concurrent_repack();
+ {
+ /*
+ * Since during normal operation the worker was already asked to
+ * exit, stopping it explicitly is especially important on ERROR.
+ * However it still seems a good practice to make sure that the
+ * worker never survives the REPACK command.
+ */
+ stop_decoding_worker();
+ }
}
PG_END_TRY();
@@ -929,7 +1050,6 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
bool swap_toast_by_content;
TransactionId frozenXid;
MultiXactId cutoffMulti;
- LogicalDecodingContext *decoding_ctx = NULL;
Snapshot snapshot = NULL;
#if USE_ASSERT_CHECKING
LOCKMODE lmode;
@@ -943,19 +1063,36 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
if (concurrent)
{
/*
- * Prepare to capture the concurrent data changes.
+ * The worker needs to be member of the locking group we're the leader
+ * of. We ought to become the leader before the worker starts. The
+ * worker will join the group as soon as it starts.
*
- * Note that this call waits for all transactions with XID already
- * assigned to finish. If some of those transactions is waiting for a
- * lock conflicting with ShareUpdateExclusiveLock on our table (e.g.
- * it runs CREATE INDEX), we can end up in a deadlock. Not sure this
- * risk is worth unlocking/locking the table (and its clustering
- * index) and checking again if its still eligible for REPACK
- * CONCURRENTLY.
+ * This is to make sure that the deadlock described below is
+ * detectable by deadlock.c: if the worker waits for a transaction to
+ * complete and we are waiting for the worker output, then effectively
+ * we (i.e. this backend) are waiting for that transaction.
*/
- decoding_ctx = setup_logical_decoding(tableOid);
+ BecomeLockGroupLeader();
+
+ /*
+ * Start the worker that decodes data changes applied while we're
+ * copying the table contents.
+ *
+ * Note that the worker has to wait for all transactions with XID
+ * already assigned to finish. If some of those transactions is
+ * waiting for a lock conflicting with ShareUpdateExclusiveLock on our
+ * table (e.g. it runs CREATE INDEX), we can end up in a deadlock.
+ * Not sure this risk is worth unlocking/locking the table (and its
+ * clustering index) and checking again if its still eligible for
+ * REPACK CONCURRENTLY.
+ */
+ start_decoding_worker(tableOid);
+
+ /*
+ * Wait until the worker has the initial snapshot and retrieve it.
+ */
+ snapshot = get_initial_snapshot(decoding_worker);
- snapshot = SnapBuildInitialSnapshotForRepack(decoding_ctx->snapshot_builder);
PushActiveSnapshot(snapshot);
}
@@ -980,7 +1117,7 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
NewHeap = table_open(OIDNewHeap, NoLock);
/* Copy the heap data into the new table in the desired order */
- copy_table_data(NewHeap, OldHeap, index, snapshot, decoding_ctx, verbose,
+ copy_table_data(NewHeap, OldHeap, index, snapshot, verbose,
&swap_toast_by_content, &frozenXid, &cutoffMulti);
/* The historic snapshot won't be needed anymore. */
@@ -1001,14 +1138,11 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
if (index)
index_close(index, NoLock);
- rebuild_relation_finish_concurrent(NewHeap, OldHeap, decoding_ctx,
- frozenXid, cutoffMulti);
+ rebuild_relation_finish_concurrent(NewHeap, OldHeap, frozenXid,
+ cutoffMulti);
pgstat_progress_update_param(PROGRESS_REPACK_PHASE,
PROGRESS_REPACK_PHASE_FINAL_CLEANUP);
-
- /* Done with decoding. */
- cleanup_logical_decoding(decoding_ctx);
}
else
{
@@ -1179,8 +1313,7 @@ make_new_heap(Oid OIDOldHeap, Oid NewTableSpace, Oid NewAccessMethod,
*/
static void
copy_table_data(Relation NewHeap, Relation OldHeap, Relation OldIndex,
- Snapshot snapshot, LogicalDecodingContext *decoding_ctx,
- bool verbose, bool *pSwapToastByContent,
+ Snapshot snapshot, bool verbose, bool *pSwapToastByContent,
TransactionId *pFreezeXid, MultiXactId *pCutoffMulti)
{
Relation relRelation;
@@ -1341,7 +1474,6 @@ copy_table_data(Relation NewHeap, Relation OldHeap, Relation OldIndex,
*/
table_relation_copy_for_cluster(OldHeap, NewHeap, OldIndex, use_sort,
cutoffs.OldestXmin, snapshot,
- decoding_ctx,
&cutoffs.FreezeLimit,
&cutoffs.MultiXactCutoff,
&num_tuples, &tups_vacuumed,
@@ -2375,58 +2507,6 @@ RepackCommandAsString(RepackCommand cmd)
}
-/*
- * Call this function before REPACK CONCURRENTLY starts to setup logical
- * decoding. It makes sure that other users of the table put enough
- * information into WAL.
- *
- * The point is that at various places we expect that the table we're
- * processing is treated like a system catalog. For example, we need to be
- * able to scan it using a "historic snapshot" anytime during the processing
- * (as opposed to scanning only at the start point of the decoding, as logical
- * replication does during initial table synchronization), in order to apply
- * concurrent UPDATE / DELETE commands.
- *
- * Note that TOAST table needs no attention here as it's not scanned using
- * historic snapshot.
- */
-static void
-begin_concurrent_repack(Relation rel)
-{
- Oid toastrelid;
-
- /*
- * Avoid logical decoding of other relations by this backend. The lock we
- * have guarantees that the actual locator cannot be changed concurrently:
- * TRUNCATE needs AccessExclusiveLock.
- */
- Assert(CheckRelationLockedByMe(rel, ShareUpdateExclusiveLock, false));
- repacked_rel_locator = rel->rd_locator;
- toastrelid = rel->rd_rel->reltoastrelid;
- if (OidIsValid(toastrelid))
- {
- Relation toastrel;
-
- /* Avoid logical decoding of other TOAST relations. */
- toastrel = table_open(toastrelid, AccessShareLock);
- repacked_rel_toast_locator = toastrel->rd_locator;
- table_close(toastrel, AccessShareLock);
- }
-}
-
-/*
- * Call this when done with REPACK CONCURRENTLY.
- */
-static void
-end_concurrent_repack(void)
-{
- /*
- * Restore normal function of (future) logical decoding for this backend.
- */
- repacked_rel_locator.relNumber = InvalidOid;
- repacked_rel_toast_locator.relNumber = InvalidOid;
-}
-
/*
* Is this backend performing logical decoding on behalf of REPACK
* (CONCURRENTLY) ?
@@ -2491,9 +2571,10 @@ static LogicalDecodingContext *
setup_logical_decoding(Oid relid)
{
Relation rel;
- TupleDesc tupdesc;
+ Oid toastrelid;
LogicalDecodingContext *ctx;
- RepackDecodingState *dstate = palloc0_object(RepackDecodingState);
+ NameData slotname;
+ RepackDecodingState *dstate;
/*
* REPACK CONCURRENTLY is not allowed in a transaction block, so this
@@ -2501,21 +2582,21 @@ setup_logical_decoding(Oid relid)
*/
Assert(!TransactionIdIsValid(GetTopTransactionIdIfAny()));
- /*
- * A single backend should not execute multiple REPACK commands at a time,
- * so use PID to make the slot unique.
- */
- snprintf(NameStr(dstate->slotname), NAMEDATALEN, "repack_%d", MyProcPid);
-
/*
* Check if we can use logical decoding.
*/
CheckSlotPermissions();
CheckLogicalDecodingRequirements();
- /* RS_TEMPORARY so that the slot gets cleaned up on ERROR. */
- ReplicationSlotCreate(NameStr(dstate->slotname), true, RS_TEMPORARY,
- false, false, false);
+ /*
+ * A single backend should not execute multiple REPACK commands at a time,
+ * so use PID to make the slot unique.
+ *
+ * RS_TEMPORARY so that the slot gets cleaned up on ERROR.
+ */
+ snprintf(NameStr(slotname), NAMEDATALEN, "repack_%d", MyProcPid);
+ ReplicationSlotCreate(NameStr(slotname), true, RS_TEMPORARY, false, false,
+ false);
/*
* Neither prepare_write nor do_write callback nor update_progress is
@@ -2537,104 +2618,109 @@ setup_logical_decoding(Oid relid)
DecodingContextFindStartpoint(ctx);
+ /*
+ * decode_concurrent_changes() needs non-blocking callback.
+ */
+ ctx->reader->routine.page_read = read_local_xlog_page_no_wait;
+
+ /*
+ * read_local_xlog_page_no_wait() needs to be able to indicate the end of
+ * WAL.
+ */
+ ctx->reader->private_data = MemoryContextAllocZero(ctx->context,
+ sizeof(ReadLocalXLogPageNoWaitPrivate));
+
+
/* Some WAL records should have been read. */
Assert(ctx->reader->EndRecPtr != InvalidXLogRecPtr);
+ /*
+ * Initialize repack_current_segment so that we can notice WAL segment
+ * boundaries.
+ */
XLByteToSeg(ctx->reader->EndRecPtr, repack_current_segment,
wal_segment_size);
- /*
- * Setup structures to store decoded changes.
- */
+ dstate = palloc0_object(RepackDecodingState);
dstate->relid = relid;
- dstate->tstore = tuplestore_begin_heap(false, false,
- maintenance_work_mem);
- /* Caller should already have the table locked. */
- rel = table_open(relid, NoLock);
- tupdesc = CreateTupleDescCopy(RelationGetDescr(rel));
- dstate->tupdesc = tupdesc;
- table_close(rel, NoLock);
+ /*
+ * Tuple descriptor may be needed to flatten a tuple before we write it to
+ * a file. A copy is needed because the decoding worker invalidates system
+ * caches before it starts to do the actual work.
+ */
+ rel = table_open(relid, AccessShareLock);
+ dstate->tupdesc = CreateTupleDescCopy(RelationGetDescr(rel));
- /* Initialize the descriptor to store the changes ... */
- dstate->tupdesc_change = CreateTemplateTupleDesc(1);
+ /* Avoid logical decoding of other relations. */
+ repacked_rel_locator = rel->rd_locator;
+ toastrelid = rel->rd_rel->reltoastrelid;
+ if (OidIsValid(toastrelid))
+ {
+ Relation toastrel;
- TupleDescInitEntry(dstate->tupdesc_change, 1, NULL, BYTEAOID, -1, 0);
- /* ... as well as the corresponding slot. */
- dstate->tsslot = MakeSingleTupleTableSlot(dstate->tupdesc_change,
- &TTSOpsMinimalTuple);
+ /* Avoid logical decoding of other TOAST relations. */
+ toastrel = table_open(toastrelid, AccessShareLock);
+ repacked_rel_toast_locator = toastrel->rd_locator;
+ table_close(toastrel, AccessShareLock);
+ }
+ table_close(rel, AccessShareLock);
- dstate->resowner = ResourceOwnerCreate(CurrentResourceOwner,
- "logical decoding");
+ /* The file will be set as soon as we have it opened. */
+ dstate->file = NULL;
ctx->output_writer_private = dstate;
+
return ctx;
}
/*
- * Retrieve tuple from ConcurrentChange structure.
+ * Decode logical changes from the WAL sequence and store them to a file.
*
- * The input data starts with the structure but it might not be appropriately
- * aligned.
- */
-static HeapTuple
-get_changed_tuple(char *change)
-{
- HeapTupleData tup_data;
- HeapTuple result;
- char *src;
-
- /*
- * Ensure alignment before accessing the fields. (This is why we can't use
- * heap_copytuple() instead of this function.)
- */
- src = change + offsetof(ConcurrentChange, tup_data);
- memcpy(&tup_data, src, sizeof(HeapTupleData));
-
- result = (HeapTuple) palloc(HEAPTUPLESIZE + tup_data.t_len);
- memcpy(result, &tup_data, sizeof(HeapTupleData));
- result->t_data = (HeapTupleHeader) ((char *) result + HEAPTUPLESIZE);
- src = change + SizeOfConcurrentChange;
- memcpy(result->t_data, src, result->t_len);
-
- return result;
-}
-
-/*
- * Decode logical changes from the WAL sequence up to end_of_wal.
+ * If true is returned, there is no more work for the worker.
*/
-void
-repack_decode_concurrent_changes(LogicalDecodingContext *ctx,
- XLogRecPtr end_of_wal)
+static bool
+decode_concurrent_changes(LogicalDecodingContext *ctx,
+ DecodingWorkerShared *shared)
{
RepackDecodingState *dstate;
- ResourceOwner resowner_old;
+ XLogRecPtr lsn_upto;
+ bool done;
+ char fname[MAXPGPATH];
dstate = (RepackDecodingState *) ctx->output_writer_private;
- resowner_old = CurrentResourceOwner;
- CurrentResourceOwner = dstate->resowner;
- PG_TRY();
+ /* Open the output file. */
+ DecodingWorkerFileName(fname, shared->relid, shared->last_exported + 1);
+ dstate->file = BufFileCreateFileSet(&shared->sfs.fs, fname);
+
+ SpinLockAcquire(&shared->mutex);
+ lsn_upto = shared->lsn_upto;
+ done = shared->done;
+ SpinLockRelease(&shared->mutex);
+
+ while (true)
{
- while (ctx->reader->EndRecPtr < end_of_wal)
- {
- XLogRecord *record;
- XLogSegNo segno_new;
- char *errm = NULL;
- XLogRecPtr end_lsn;
+ XLogRecord *record;
+ XLogSegNo segno_new;
+ char *errm = NULL;
+ XLogRecPtr end_lsn;
- record = XLogReadRecord(ctx->reader, &errm);
- if (errm)
- elog(ERROR, "%s", errm);
+ CHECK_FOR_INTERRUPTS();
- if (record != NULL)
- LogicalDecodingProcessRecord(ctx, ctx->reader);
+ record = XLogReadRecord(ctx->reader, &errm);
+ if (record)
+ {
+ LogicalDecodingProcessRecord(ctx, ctx->reader);
/*
* If WAL segment boundary has been crossed, inform the decoding
- * system that the catalog_xmin can advance. (We can confirm more
- * often, but a filling a single WAL segment should not take much
- * time.)
+ * system that the catalog_xmin can advance.
+ *
+ * TODO Does it make sense to confirm more often? Segment size
+ * seems appropriate for restart_lsn (because less than a segment
+ * cannot be recycled anyway), however more frequent checks might
+ * be beneficial for catalog_xmin.
*/
end_lsn = ctx->reader->EndRecPtr;
XLByteToSeg(end_lsn, segno_new, wal_segment_size);
@@ -2645,80 +2731,137 @@ repack_decode_concurrent_changes(LogicalDecodingContext *ctx,
(uint32) (end_lsn >> 32), (uint32) end_lsn);
repack_current_segment = segno_new;
}
+ }
+ else
+ {
+ ReadLocalXLogPageNoWaitPrivate *priv;
- CHECK_FOR_INTERRUPTS();
+ if (errm)
+ ereport(ERROR, (errmsg("%s", errm)));
+
+ /*
+ * In the decoding loop we do not want to get blocked when there
+ * is no more WAL available, otherwise the loop would become
+ * uninterruptible.
+ */
+ priv = (ReadLocalXLogPageNoWaitPrivate *)
+ ctx->reader->private_data;
+ if (priv->end_of_wal)
+ /* Do not miss the end of WAL condition next time. */
+ priv->end_of_wal = false;
+ else
+ ereport(ERROR, (errmsg("could not read WAL record")));
+ }
+
+ /*
+ * Whether we could read new record or not, keep checking if
+ * 'lsn_upto' was specified.
+ */
+ if (XLogRecPtrIsInvalid(lsn_upto))
+ {
+ SpinLockAcquire(&shared->mutex);
+ lsn_upto = shared->lsn_upto;
+ /* 'done' should be set at the same time as 'lsn_upto' */
+ done = shared->done;
+ SpinLockRelease(&shared->mutex);
+ }
+ if (!XLogRecPtrIsInvalid(lsn_upto) &&
+ ctx->reader->EndRecPtr >= lsn_upto)
+ break;
+
+ if (record == NULL)
+ {
+ int64 timeout = 0;
+ WaitLSNResult res;
+
+ /*
+ * Before we retry reading, wait until new WAL is flushed.
+ *
+ * There is a race condition such that the backend executing
+ * REPACK determines 'lsn_upto', but before it sets the shared
+ * variable, we reach the end of WAL. In that case we'd need to
+ * wait until the next WAL flush (unrelated to REPACK). Although
+ * that should not be a problem in a busy system, it might be
+ * noticeable in other cases, including regression tests (which
+ * are not necessarily executed in parallel). Therefore it makes
+ * sense to use timeout.
+ *
+ * If lsn_upto is valid, WAL records having LSN lower than that
+ * should already have been flushed to disk.
+ */
+ if (XLogRecPtrIsInvalid(lsn_upto))
+ timeout = 100L;
+ res = WaitForLSN(WAIT_LSN_TYPE_PRIMARY_FLUSH,
+ ctx->reader->EndRecPtr + 1,
+ timeout);
+ if (res != WAIT_LSN_RESULT_SUCCESS &&
+ res != WAIT_LSN_RESULT_TIMEOUT)
+ ereport(ERROR, (errmsg("waiting for WAL failed")));
}
- InvalidateSystemCaches();
- CurrentResourceOwner = resowner_old;
- }
- PG_CATCH();
- {
- /* clear all timetravel entries */
- InvalidateSystemCaches();
- CurrentResourceOwner = resowner_old;
- PG_RE_THROW();
}
- PG_END_TRY();
+
+ /*
+ * Close the file so we can make it available to the backend.
+ */
+ BufFileClose(dstate->file);
+ dstate->file = NULL;
+ SpinLockAcquire(&shared->mutex);
+ shared->lsn_upto = InvalidXLogRecPtr;
+ shared->last_exported++;
+ SpinLockRelease(&shared->mutex);
+ ConditionVariableSignal(&shared->cv);
+
+ return done;
}
/*
* Apply changes stored in 'file'.
*/
static void
-apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
+apply_concurrent_changes(BufFile *file, ChangeDest *dest)
{
+ char kind;
+ uint32 t_len;
Relation rel = dest->rel;
TupleTableSlot *index_slot,
*ident_slot;
HeapTuple tup_old = NULL;
- if (dstate->nchanges == 0)
- return;
-
/* TupleTableSlot is needed to pass the tuple to ExecInsertIndexTuples(). */
- index_slot = MakeSingleTupleTableSlot(dstate->tupdesc, &TTSOpsHeapTuple);
+ index_slot = MakeSingleTupleTableSlot(RelationGetDescr(rel),
+ &TTSOpsHeapTuple);
/* A slot to fetch tuples from identity index. */
ident_slot = table_slot_create(rel, NULL);
- while (tuplestore_gettupleslot(dstate->tstore, true, false,
- dstate->tsslot))
+ while (true)
{
- bool shouldFree;
- HeapTuple tup_change,
- tup,
+ size_t nread;
+ HeapTuple tup,
tup_exist;
- char *change_raw,
- *src;
- ConcurrentChange change;
- bool isnull[1];
- Datum values[1];
CHECK_FOR_INTERRUPTS();
- /* Get the change from the single-column tuple. */
- tup_change = ExecFetchSlotHeapTuple(dstate->tsslot, false, &shouldFree);
- heap_deform_tuple(tup_change, dstate->tupdesc_change, values, isnull);
- Assert(!isnull[0]);
-
- /* Make sure we access aligned data. */
- change_raw = (char *) DatumGetByteaP(values[0]);
- src = (char *) VARDATA(change_raw);
- memcpy(&change, src, SizeOfConcurrentChange);
+ nread = BufFileReadMaybeEOF(file, &kind, 1, true);
+ /* Are we done with the file? */
+ if (nread == 0)
+ break;
- /*
- * Extract the tuple from the change. The tuple is copied here because
- * it might be assigned to 'tup_old', in which case it needs to
- * survive into the next iteration.
- */
- tup = get_changed_tuple(src);
+ /* Read the tuple. */
+ BufFileReadExact(file, &t_len, sizeof(t_len));
+ tup = (HeapTuple) palloc(HEAPTUPLESIZE + t_len);
+ tup->t_data = (HeapTupleHeader) ((char *) tup + HEAPTUPLESIZE);
+ BufFileReadExact(file, tup->t_data, t_len);
+ tup->t_len = t_len;
+ ItemPointerSetInvalid(&tup->t_self);
+ tup->t_tableOid = RelationGetRelid(dest->rel);
- if (change.kind == CHANGE_UPDATE_OLD)
+ if (kind == CHANGE_UPDATE_OLD)
{
Assert(tup_old == NULL);
tup_old = tup;
}
- else if (change.kind == CHANGE_INSERT)
+ else if (kind == CHANGE_INSERT)
{
Assert(tup_old == NULL);
@@ -2726,12 +2869,11 @@ apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
pfree(tup);
}
- else if (change.kind == CHANGE_UPDATE_NEW ||
- change.kind == CHANGE_DELETE)
+ else if (kind == CHANGE_UPDATE_NEW || kind == CHANGE_DELETE)
{
HeapTuple tup_key;
- if (change.kind == CHANGE_UPDATE_NEW)
+ if (kind == CHANGE_UPDATE_NEW)
{
tup_key = tup_old != NULL ? tup_old : tup;
}
@@ -2748,7 +2890,7 @@ apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
if (tup_exist == NULL)
elog(ERROR, "failed to find target tuple");
- if (change.kind == CHANGE_UPDATE_NEW)
+ if (kind == CHANGE_UPDATE_NEW)
apply_concurrent_update(rel, tup, tup_exist, dest->iistate,
index_slot);
else
@@ -2763,26 +2905,19 @@ apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
pfree(tup);
}
else
- elog(ERROR, "unrecognized kind of change: %d", change.kind);
+ elog(ERROR, "unrecognized kind of change: %d", kind);
/*
* If a change was applied now, increment CID for next writes and
* update the snapshot so it sees the changes we've applied so far.
*/
- if (change.kind != CHANGE_UPDATE_OLD)
+ if (kind != CHANGE_UPDATE_OLD)
{
CommandCounterIncrement();
UpdateActiveSnapshotCommandId();
}
-
- /* TTSOpsMinimalTuple has .get_heap_tuple==NULL. */
- Assert(shouldFree);
- pfree(tup_change);
}
- tuplestore_clear(dstate->tstore);
- dstate->nchanges = 0;
-
/* Cleanup. */
ExecDropSingleTupleTableSlot(index_slot);
ExecDropSingleTupleTableSlot(ident_slot);
@@ -2957,25 +3092,59 @@ find_target_tuple(Relation rel, ChangeDest *dest, HeapTuple tup_key,
}
/*
- * Decode and apply concurrent changes.
+ * Decode and apply concurrent changes, up to (and including) the record whose
+ * LSN is 'end_of_wal'.
*/
static void
-process_concurrent_changes(LogicalDecodingContext *decoding_ctx,
- XLogRecPtr end_of_wal, ChangeDest *dest)
+process_concurrent_changes(XLogRecPtr end_of_wal, ChangeDest *dest, bool done)
{
- RepackDecodingState *dstate;
+ DecodingWorkerShared *shared;
+ char fname[MAXPGPATH];
+ BufFile *file;
pgstat_progress_update_param(PROGRESS_REPACK_PHASE,
PROGRESS_REPACK_PHASE_CATCH_UP);
- dstate = (RepackDecodingState *) decoding_ctx->output_writer_private;
+ /* Ask the worker for the file. */
+ shared = (DecodingWorkerShared *) dsm_segment_address(decoding_worker->seg);
+ SpinLockAcquire(&shared->mutex);
+ shared->lsn_upto = end_of_wal;
+ shared->done = done;
+ SpinLockRelease(&shared->mutex);
+
+ /*
+ * The worker needs to finish processing of the current WAL record. Even
+ * if it's idle, it'll need to close the output file. Thus we're likely to
+ * wait, so prepare for sleep.
+ */
+ ConditionVariablePrepareToSleep(&shared->cv);
+ for (;;)
+ {
+ int last_exported;
- repack_decode_concurrent_changes(decoding_ctx, end_of_wal);
+ SpinLockAcquire(&shared->mutex);
+ last_exported = shared->last_exported;
+ SpinLockRelease(&shared->mutex);
- if (dstate->nchanges == 0)
- return;
+ /*
+ * Has the worker exported the file we are waiting for?
+ */
+ if (last_exported == dest->file_seq)
+ break;
- apply_concurrent_changes(dstate, dest);
+ ConditionVariableSleep(&shared->cv, WAIT_EVENT_REPACK_WORKER_EXPORT);
+ }
+ ConditionVariableCancelSleep();
+
+ /* Open the file. */
+ DecodingWorkerFileName(fname, shared->relid, dest->file_seq);
+ file = BufFileOpenFileSet(&shared->sfs.fs, fname, O_RDONLY, false);
+ apply_concurrent_changes(file, dest);
+
+ BufFileClose(file);
+
+ /* Get ready for the next file. */
+ dest->file_seq++;
}
/*
@@ -3101,15 +3270,10 @@ cleanup_logical_decoding(LogicalDecodingContext *ctx)
dstate = (RepackDecodingState *) ctx->output_writer_private;
- ExecDropSingleTupleTableSlot(dstate->tsslot);
- FreeTupleDesc(dstate->tupdesc_change);
FreeTupleDesc(dstate->tupdesc);
- tuplestore_end(dstate->tstore);
-
FreeDecodingContext(ctx);
- ReplicationSlotRelease();
- ReplicationSlotDrop(NameStr(dstate->slotname), false);
+ ReplicationSlotDropAcquired();
pfree(dstate);
}
@@ -3123,7 +3287,6 @@ cleanup_logical_decoding(LogicalDecodingContext *ctx)
*/
static void
rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
- LogicalDecodingContext *decoding_ctx,
TransactionId frozenXid,
MultiXactId cutoffMulti)
{
@@ -3204,6 +3367,7 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
&chgdst.ident_index);
chgdst.ident_key = build_identity_key(ident_idx_new, OldHeap,
&chgdst.ident_key_nentries);
+ chgdst.file_seq = WORKER_FILE_SNAPSHOT + 1;
/*
* During testing, wait for another backend to perform concurrent data
@@ -3225,7 +3389,7 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
* hold AccessExclusiveLock. (Quite some amount of WAL could have been
* written during the data copying and index creation.)
*/
- process_concurrent_changes(decoding_ctx, end_of_wal, &chgdst);
+ process_concurrent_changes(end_of_wal, &chgdst, false);
/*
* Acquire AccessExclusiveLock on the table, its TOAST relation (if there
@@ -3306,8 +3470,11 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
XLogFlush(wal_insert_ptr);
end_of_wal = GetFlushRecPtr(NULL);
- /* Apply the concurrent changes again. */
- process_concurrent_changes(decoding_ctx, end_of_wal, &chgdst);
+ /*
+ * Apply the concurrent changes again. Indicate that the decoding worker
+ * won't be needed anymore.
+ */
+ process_concurrent_changes(end_of_wal, &chgdst, true);
/* Remember info about rel before closing OldHeap */
relpersistence = OldHeap->rd_rel->relpersistence;
@@ -3417,3 +3584,510 @@ build_new_indexes(Relation NewHeap, Relation OldHeap, List *OldIndexes)
return result;
}
+
+/*
+ * Try to start a background worker to perform logical decoding of data
+ * changes applied to relation while REPACK CONCURRENTLY is copying its
+ * contents to a new table.
+ */
+static void
+start_decoding_worker(Oid relid)
+{
+ Size size;
+ dsm_segment *seg;
+ DecodingWorkerShared *shared;
+ shm_mq *mq;
+ shm_mq_handle *mqh;
+ BackgroundWorker bgw;
+
+ /* Setup shared memory. */
+ size = BUFFERALIGN(offsetof(DecodingWorkerShared, error_queue)) +
+ BUFFERALIGN(REPACK_ERROR_QUEUE_SIZE);
+ seg = dsm_create(size, 0);
+ shared = (DecodingWorkerShared *) dsm_segment_address(seg);
+ shared->lsn_upto = InvalidXLogRecPtr;
+ shared->done = false;
+ SharedFileSetInit(&shared->sfs, seg);
+ shared->last_exported = -1;
+ SpinLockInit(&shared->mutex);
+ shared->dbid = MyDatabaseId;
+
+ /*
+ * This is the UserId set in cluster_rel(). Security context shouldn't be
+ * needed for decoding worker.
+ */
+ shared->roleid = GetUserId();
+ shared->relid = relid;
+ ConditionVariableInit(&shared->cv);
+ shared->backend_proc = MyProc;
+ shared->backend_pid = MyProcPid;
+ shared->backend_proc_number = MyProcNumber;
+
+ mq = shm_mq_create((char *) BUFFERALIGN(shared->error_queue),
+ REPACK_ERROR_QUEUE_SIZE);
+ shm_mq_set_receiver(mq, MyProc);
+ mqh = shm_mq_attach(mq, seg, NULL);
+
+ memset(&bgw, 0, sizeof(bgw));
+ snprintf(bgw.bgw_name, BGW_MAXLEN,
+ "REPACK decoding worker for relation \"%s\"",
+ get_rel_name(relid));
+ snprintf(bgw.bgw_type, BGW_MAXLEN, "REPACK decoding worker");
+ bgw.bgw_flags = BGWORKER_SHMEM_ACCESS |
+ BGWORKER_BACKEND_DATABASE_CONNECTION;
+ bgw.bgw_start_time = BgWorkerStart_RecoveryFinished;
+ bgw.bgw_restart_time = BGW_NEVER_RESTART;
+ snprintf(bgw.bgw_library_name, MAXPGPATH, "postgres");
+ snprintf(bgw.bgw_function_name, BGW_MAXLEN, "RepackWorkerMain");
+ bgw.bgw_main_arg = UInt32GetDatum(dsm_segment_handle(seg));
+ bgw.bgw_notify_pid = MyProcPid;
+
+ decoding_worker = palloc0_object(DecodingWorker);
+ if (!RegisterDynamicBackgroundWorker(&bgw, &decoding_worker->handle))
+ ereport(ERROR,
+ (errcode(ERRCODE_CONFIGURATION_LIMIT_EXCEEDED),
+ errmsg("out of background worker slots"),
+ errhint("You might need to increase \"%s\".", "max_worker_processes")));
+
+ decoding_worker->seg = seg;
+ decoding_worker->error_mqh = mqh;
+
+ /*
+ * The decoding setup must be done before the caller can have XID assigned
+ * for any reason, otherwise the worker might end up in a deadlock,
+ * waiting for the caller's transaction to end. Therefore wait here until
+ * the worker indicates that it has the logical decoding initialized.
+ */
+ ConditionVariablePrepareToSleep(&shared->cv);
+ for (;;)
+ {
+ int initialized;
+
+ SpinLockAcquire(&shared->mutex);
+ initialized = shared->initialized;
+ SpinLockRelease(&shared->mutex);
+
+ if (initialized)
+ break;
+
+ ConditionVariableSleep(&shared->cv, WAIT_EVENT_REPACK_WORKER_EXPORT);
+ }
+ ConditionVariableCancelSleep();
+}
+
+/*
+ * Stop the decoding worker and cleanup the related resources.
+ *
+ * The worker stops on its own when it knows there is no more work to do, but
+ * we need to stop it explicitly at least on ERROR in the launching backend.
+ */
+static void
+stop_decoding_worker(void)
+{
+ BgwHandleStatus status;
+
+ /* Haven't reached the worker startup? */
+ if (decoding_worker == NULL)
+ return;
+
+ /* Could not register the worker? */
+ if (decoding_worker->handle == NULL)
+ return;
+
+ TerminateBackgroundWorker(decoding_worker->handle);
+ /* The worker should really exit before the REPACK command does. */
+ HOLD_INTERRUPTS();
+ status = WaitForBackgroundWorkerShutdown(decoding_worker->handle);
+ RESUME_INTERRUPTS();
+
+ if (status == BGWH_POSTMASTER_DIED)
+ ereport(FATAL,
+ (errcode(ERRCODE_ADMIN_SHUTDOWN),
+ errmsg("postmaster exited during REPACK command")));
+
+ shm_mq_detach(decoding_worker->error_mqh);
+
+ /*
+ * If we could not cancel the current sleep due to ERROR, do that before
+ * we detach from the shared memory the condition variable is located in.
+ * If we did not, the bgworker ERROR handling code would try and fail
+ * badly.
+ */
+ ConditionVariableCancelSleep();
+
+ dsm_detach(decoding_worker->seg);
+ pfree(decoding_worker);
+ decoding_worker = NULL;
+}
+
+/* Is this process a REPACK worker? */
+static bool is_repack_worker = false;
+
+static pid_t backend_pid;
+static ProcNumber backend_proc_number;
+
+/*
+ * See ParallelWorkerShutdown for details.
+ */
+static void
+RepackWorkerShutdown(int code, Datum arg)
+{
+ SendProcSignal(backend_pid,
+ PROCSIG_REPACK_MESSAGE,
+ backend_proc_number);
+
+ dsm_detach((dsm_segment *) DatumGetPointer(arg));
+}
+
+/* REPACK decoding worker entry point */
+void
+RepackWorkerMain(Datum main_arg)
+{
+ dsm_segment *seg;
+ DecodingWorkerShared *shared;
+ shm_mq *mq;
+ shm_mq_handle *mqh;
+
+ is_repack_worker = true;
+
+ /*
+ * Override the default bgworker_die() with die() so we can use
+ * CHECK_FOR_INTERRUPTS().
+ */
+ pqsignal(SIGTERM, die);
+ BackgroundWorkerUnblockSignals();
+
+ seg = dsm_attach(DatumGetUInt32(main_arg));
+ if (seg == NULL)
+ ereport(ERROR,
+ (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
+ errmsg("could not map dynamic shared memory segment")));
+
+ shared = (DecodingWorkerShared *) dsm_segment_address(seg);
+
+ /* Arrange to signal the leader if we exit. */
+ backend_pid = shared->backend_pid;
+ backend_proc_number = shared->backend_proc_number;
+ before_shmem_exit(RepackWorkerShutdown, PointerGetDatum(seg));
+
+ /*
+ * Join locking group - see the comments around the call of
+ * start_decoding_worker().
+ */
+ if (!BecomeLockGroupMember(shared->backend_proc, backend_pid))
+ /* The leader is not running anymore. */
+ return;
+
+ /*
+ * Setup a queue to send error messages to the backend that launched this
+ * worker.
+ */
+ mq = (shm_mq *) (char *) BUFFERALIGN(shared->error_queue);
+ shm_mq_set_sender(mq, MyProc);
+ mqh = shm_mq_attach(mq, seg, NULL);
+ pq_redirect_to_shm_mq(seg, mqh);
+ pq_set_parallel_leader(shared->backend_pid,
+ shared->backend_proc_number);
+
+ /* Connect to the database. */
+ BackgroundWorkerInitializeConnectionByOid(shared->dbid, shared->roleid, 0);
+
+ repack_worker_internal(seg);
+}
+
+static void
+repack_worker_internal(dsm_segment *seg)
+{
+ DecodingWorkerShared *shared;
+ LogicalDecodingContext *decoding_ctx;
+ SharedFileSet *sfs;
+ Snapshot snapshot;
+
+ /*
+ * Transaction is needed to open relation, and it also provides us with a
+ * resource owner.
+ */
+ StartTransactionCommand();
+
+ shared = (DecodingWorkerShared *) dsm_segment_address(seg);
+
+ /*
+ * Not sure the spinlock is needed here - the backend should not change
+ * anything in the shared memory until we have serialized the snapshot.
+ */
+ SpinLockAcquire(&shared->mutex);
+ Assert(XLogRecPtrIsInvalid(shared->lsn_upto));
+ sfs = &shared->sfs;
+ SpinLockRelease(&shared->mutex);
+
+ SharedFileSetAttach(sfs, seg);
+
+ /*
+ * Prepare to capture the concurrent data changes ourselves.
+ */
+ decoding_ctx = setup_logical_decoding(shared->relid);
+
+ /* Announce that we're ready. */
+ SpinLockAcquire(&shared->mutex);
+ shared->initialized = true;
+ SpinLockRelease(&shared->mutex);
+ ConditionVariableSignal(&shared->cv);
+
+ /* Build the initial snapshot and export it. */
+ snapshot = SnapBuildInitialSnapshotForRepack(decoding_ctx->snapshot_builder);
+ export_initial_snapshot(snapshot, shared);
+
+ /*
+ * Only historic snapshots should be used now. Do not let us restrict the
+ * progress of xmin horizon.
+ */
+ InvalidateCatalogSnapshot();
+
+ while (!decode_concurrent_changes(decoding_ctx, shared))
+ ;
+
+ /* Cleanup. */
+ cleanup_logical_decoding(decoding_ctx);
+ CommitTransactionCommand();
+}
+
+/*
+ * Make snapshot available to the backend that launched the decoding worker.
+ */
+static void
+export_initial_snapshot(Snapshot snapshot, DecodingWorkerShared *shared)
+{
+ char fname[MAXPGPATH];
+ BufFile *file;
+ Size snap_size;
+ char *snap_space;
+
+ snap_size = EstimateSnapshotSpace(snapshot);
+ snap_space = (char *) palloc(snap_size);
+ SerializeSnapshot(snapshot, snap_space);
+ FreeSnapshot(snapshot);
+
+ DecodingWorkerFileName(fname, shared->relid, shared->last_exported + 1);
+ file = BufFileCreateFileSet(&shared->sfs.fs, fname);
+ /* To make restoration easier, write the snapshot size first. */
+ BufFileWrite(file, &snap_size, sizeof(snap_size));
+ BufFileWrite(file, snap_space, snap_size);
+ pfree(snap_space);
+ BufFileClose(file);
+
+ /* Increase the counter to tell the backend that the file is available. */
+ SpinLockAcquire(&shared->mutex);
+ shared->last_exported++;
+ SpinLockRelease(&shared->mutex);
+ ConditionVariableSignal(&shared->cv);
+}
+
+/*
+ * Get the initial snapshot from the decoding worker.
+ */
+static Snapshot
+get_initial_snapshot(DecodingWorker *worker)
+{
+ DecodingWorkerShared *shared;
+ char fname[MAXPGPATH];
+ BufFile *file;
+ Size snap_size;
+ char *snap_space;
+ Snapshot snapshot;
+
+ shared = (DecodingWorkerShared *) dsm_segment_address(worker->seg);
+
+ /*
+ * The worker needs to initialize the logical decoding, which usually
+ * takes some time. Therefore it makes sense to prepare for the sleep
+ * first.
+ */
+ ConditionVariablePrepareToSleep(&shared->cv);
+ for (;;)
+ {
+ int last_exported;
+
+ SpinLockAcquire(&shared->mutex);
+ last_exported = shared->last_exported;
+ SpinLockRelease(&shared->mutex);
+
+ /*
+ * Has the worker exported the file we are waiting for?
+ */
+ if (last_exported == WORKER_FILE_SNAPSHOT)
+ break;
+
+ ConditionVariableSleep(&shared->cv, WAIT_EVENT_REPACK_WORKER_EXPORT);
+ }
+ ConditionVariableCancelSleep();
+
+ /* Read the snapshot from a file. */
+ DecodingWorkerFileName(fname, shared->relid, WORKER_FILE_SNAPSHOT);
+ file = BufFileOpenFileSet(&shared->sfs.fs, fname, O_RDONLY, false);
+ BufFileReadExact(file, &snap_size, sizeof(snap_size));
+ snap_space = (char *) palloc(snap_size);
+ BufFileReadExact(file, snap_space, snap_size);
+ BufFileClose(file);
+
+ /* Restore it. */
+ snapshot = RestoreSnapshot(snap_space);
+ pfree(snap_space);
+
+ return snapshot;
+}
+
+bool
+IsRepackWorker(void)
+{
+ return is_repack_worker;
+}
+
+/*
+ * Handle receipt of an interrupt indicating a repack worker message.
+ *
+ * Note: this is called within a signal handler! All we can do is set
+ * a flag that will cause the next CHECK_FOR_INTERRUPTS() to invoke
+ * ProcessRepackMessages().
+ */
+void
+HandleRepackMessageInterrupt(void)
+{
+ InterruptPending = true;
+ RepackMessagePending = true;
+ SetLatch(MyLatch);
+}
+
+/*
+ * Process any queued protocol messages received from parallel workers.
+ */
+void
+ProcessRepackMessages(void)
+{
+ MemoryContext oldcontext;
+
+ static MemoryContext hpm_context = NULL;
+
+ /*
+ * Nothing to do if we haven't launched the worker yet or have already
+ * terminated it.
+ */
+ if (decoding_worker == NULL)
+ return;
+
+ /*
+ * This is invoked from ProcessInterrupts(), and since some of the
+ * functions it calls contain CHECK_FOR_INTERRUPTS(), there is a potential
+ * for recursive calls if more signals are received while this runs. It's
+ * unclear that recursive entry would be safe, and it doesn't seem useful
+ * even if it is safe, so let's block interrupts until done.
+ */
+ HOLD_INTERRUPTS();
+
+ /*
+ * Moreover, CurrentMemoryContext might be pointing almost anywhere. We
+ * don't want to risk leaking data into long-lived contexts, so let's do
+ * our work here in a private context that we can reset on each use.
+ */
+ if (hpm_context == NULL) /* first time through? */
+ hpm_context = AllocSetContextCreate(TopMemoryContext,
+ "ProcessRepackMessages",
+ ALLOCSET_DEFAULT_SIZES);
+ else
+ MemoryContextReset(hpm_context);
+
+ oldcontext = MemoryContextSwitchTo(hpm_context);
+
+ /* OK to process messages. Reset the flag saying there are more to do. */
+ RepackMessagePending = false;
+
+ /*
+ * Read as many messages as we can from each worker, but stop when no more
+ * messages can be read from the worker without blocking.
+ */
+ while (true)
+ {
+ shm_mq_result res;
+ Size nbytes;
+ void *data;
+
+ res = shm_mq_receive(decoding_worker->error_mqh, &nbytes,
+ &data, true);
+ if (res == SHM_MQ_WOULD_BLOCK)
+ break;
+ else if (res == SHM_MQ_SUCCESS)
+ {
+ StringInfoData msg;
+
+ initStringInfo(&msg);
+ appendBinaryStringInfo(&msg, data, nbytes);
+ ProcessRepackMessage(&msg);
+ pfree(msg.data);
+ }
+ else
+ {
+ /*
+ * The decoding worker is special in that it exits as soon as it
+ * has its work done. Thus the DETACHED result code is fine.
+ */
+ Assert(res == SHM_MQ_DETACHED);
+
+ break;
+ }
+ }
+
+ MemoryContextSwitchTo(oldcontext);
+
+ /* Might as well clear the context on our way out */
+ MemoryContextReset(hpm_context);
+
+ RESUME_INTERRUPTS();
+}
+
+/*
+ * Process a single protocol message received from a single parallel worker.
+ */
+static void
+ProcessRepackMessage(StringInfo msg)
+{
+ char msgtype;
+
+ msgtype = pq_getmsgbyte(msg);
+
+ switch (msgtype)
+ {
+ case PqMsg_ErrorResponse:
+ case PqMsg_NoticeResponse:
+ {
+ ErrorData edata;
+
+ /* Parse ErrorResponse or NoticeResponse. */
+ pq_parse_errornotice(msg, &edata);
+
+ /* Death of a worker isn't enough justification for suicide. */
+ edata.elevel = Min(edata.elevel, ERROR);
+
+ /*
+ * If desired, add a context line to show that this is a
+ * message propagated from a parallel worker. Otherwise, it
+ * can sometimes be confusing to understand what actually
+ * happened.
+ */
+ if (edata.context)
+ edata.context = psprintf("%s\n%s", edata.context,
+ _("decoding worker"));
+ else
+ edata.context = pstrdup(_("decoding worker"));
+
+ /* Rethrow error or print notice. */
+ ThrowErrorData(&edata);
+
+ break;
+ }
+
+ default:
+ {
+ elog(ERROR, "unrecognized message type received from decoding worker: %c (message length %d bytes)",
+ msgtype, msg->len);
+ }
+ }
+}
diff --git a/src/backend/libpq/pqmq.c b/src/backend/libpq/pqmq.c
index 6e4bbfb5aa1..42f6fa472c5 100644
--- a/src/backend/libpq/pqmq.c
+++ b/src/backend/libpq/pqmq.c
@@ -14,6 +14,7 @@
#include "postgres.h"
#include "access/parallel.h"
+#include "commands/cluster.h"
#include "libpq/libpq.h"
#include "libpq/pqformat.h"
#include "libpq/pqmq.h"
@@ -175,6 +176,10 @@ mq_putmessage(char msgtype, const char *s, size_t len)
SendProcSignal(pq_mq_parallel_leader_pid,
PROCSIG_PARALLEL_APPLY_MESSAGE,
pq_mq_parallel_leader_proc_number);
+ else if (IsRepackWorker())
+ SendProcSignal(pq_mq_parallel_leader_pid,
+ PROCSIG_REPACK_MESSAGE,
+ pq_mq_parallel_leader_proc_number);
else
{
Assert(IsParallelWorker());
diff --git a/src/backend/postmaster/bgworker.c b/src/backend/postmaster/bgworker.c
index 65deabe91a7..334bb708c5b 100644
--- a/src/backend/postmaster/bgworker.c
+++ b/src/backend/postmaster/bgworker.c
@@ -13,6 +13,7 @@
#include "postgres.h"
#include "access/parallel.h"
+#include "commands/cluster.h"
#include "libpq/pqsignal.h"
#include "miscadmin.h"
#include "pgstat.h"
@@ -136,6 +137,9 @@ static const struct
},
{
"SequenceSyncWorkerMain", SequenceSyncWorkerMain
+ },
+ {
+ "RepackWorkerMain", RepackWorkerMain
}
};
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index b0ef1a12520..35a46988285 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -194,7 +194,11 @@ StartupDecodingContext(List *output_plugin_options,
ctx->slot = slot;
- ctx->reader = XLogReaderAllocate(wal_segment_size, NULL, xl_routine, ctx);
+ /*
+ * TODO A separate patch for PG core, unless there's really a reason to
+ * pass ctx for private_data (May extensions expect ctx?).
+ */
+ ctx->reader = XLogReaderAllocate(wal_segment_size, NULL, xl_routine, NULL);
if (!ctx->reader)
ereport(ERROR,
(errcode(ERRCODE_OUT_OF_MEMORY),
diff --git a/src/backend/replication/pgoutput_repack/pgoutput_repack.c b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
index a074eed393d..4bc47a72371 100644
--- a/src/backend/replication/pgoutput_repack/pgoutput_repack.c
+++ b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
@@ -169,17 +169,13 @@ store_change(LogicalDecodingContext *ctx, ConcurrentChangeKind kind,
HeapTuple tuple)
{
RepackDecodingState *dstate;
- char *change_raw;
- ConcurrentChange change;
+ char kind_byte = (char) kind;
bool flattened = false;
- Size size;
- Datum values[1];
- bool isnull[1];
- char *dst;
dstate = (RepackDecodingState *) ctx->output_writer_private;
- size = VARHDRSZ + SizeOfConcurrentChange;
+ /* Store the change kind. */
+ BufFileWrite(dstate->file, &kind_byte, 1);
/*
* ReorderBufferCommit() stores the TOAST chunks in its private memory
@@ -196,46 +192,12 @@ store_change(LogicalDecodingContext *ctx, ConcurrentChangeKind kind,
tuple = toast_flatten_tuple(tuple, dstate->tupdesc);
flattened = true;
}
+ /* Store the tuple size ... */
+ BufFileWrite(dstate->file, &tuple->t_len, sizeof(tuple->t_len));
+ /* ... and the tuple itself. */
+ BufFileWrite(dstate->file, tuple->t_data, tuple->t_len);
- size += tuple->t_len;
- if (size >= MaxAllocSize)
- elog(ERROR, "Change is too big.");
-
- /* Construct the change. */
- change_raw = (char *) palloc0(size);
- SET_VARSIZE(change_raw, size);
-
- /*
- * Since the varlena alignment might not be sufficient for the structure,
- * set the fields in a local instance and remember where it should
- * eventually be copied.
- */
- change.kind = kind;
- dst = (char *) VARDATA(change_raw);
-
- /*
- * Copy the tuple.
- *
- * Note: change->tup_data.t_data must be fixed on retrieval!
- */
- memcpy(&change.tup_data, tuple, sizeof(HeapTupleData));
- memcpy(dst, &change, SizeOfConcurrentChange);
- dst += SizeOfConcurrentChange;
- memcpy(dst, tuple->t_data, tuple->t_len);
-
- /* The data has been copied. */
+ /* Free the flat copy if created above. */
if (flattened)
pfree(tuple);
-
- /* Store as tuple of 1 bytea column. */
- values[0] = PointerGetDatum(change_raw);
- isnull[0] = false;
- tuplestore_putvalues(dstate->tstore, dstate->tupdesc_change,
- values, isnull);
-
- /* Accounting. */
- dstate->nchanges++;
-
- /* Cleanup. */
- pfree(change_raw);
}
diff --git a/src/backend/storage/ipc/procsignal.c b/src/backend/storage/ipc/procsignal.c
index 8e56922dcea..6f9e7a7aab7 100644
--- a/src/backend/storage/ipc/procsignal.c
+++ b/src/backend/storage/ipc/procsignal.c
@@ -19,6 +19,7 @@
#include "access/parallel.h"
#include "commands/async.h"
+#include "commands/cluster.h"
#include "miscadmin.h"
#include "pgstat.h"
#include "port/pg_bitutils.h"
@@ -697,6 +698,9 @@ procsignal_sigusr1_handler(SIGNAL_ARGS)
if (CheckProcSignal(PROCSIG_PARALLEL_APPLY_MESSAGE))
HandleParallelApplyMessageInterrupt();
+ if (CheckProcSignal(PROCSIG_REPACK_MESSAGE))
+ HandleRepackMessageInterrupt();
+
if (CheckProcSignal(PROCSIG_RECOVERY_CONFLICT_DATABASE))
HandleRecoveryConflictInterrupt(PROCSIG_RECOVERY_CONFLICT_DATABASE);
diff --git a/src/backend/tcop/postgres.c b/src/backend/tcop/postgres.c
index e54bf1e760f..fc81ad87615 100644
--- a/src/backend/tcop/postgres.c
+++ b/src/backend/tcop/postgres.c
@@ -36,6 +36,7 @@
#include "access/xact.h"
#include "catalog/pg_type.h"
#include "commands/async.h"
+#include "commands/cluster.h"
#include "commands/event_trigger.h"
#include "commands/explain_state.h"
#include "commands/prepare.h"
@@ -3541,6 +3542,9 @@ ProcessInterrupts(void)
if (ParallelApplyMessagePending)
ProcessParallelApplyMessages();
+
+ if (RepackMessagePending)
+ ProcessRepackMessages();
}
/*
diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt
index 3299de23bb3..73a3def69bc 100644
--- a/src/backend/utils/activity/wait_event_names.txt
+++ b/src/backend/utils/activity/wait_event_names.txt
@@ -62,6 +62,7 @@ LOGICAL_APPLY_MAIN "Waiting in main loop of logical replication apply process."
LOGICAL_LAUNCHER_MAIN "Waiting in main loop of logical replication launcher process."
LOGICAL_PARALLEL_APPLY_MAIN "Waiting in main loop of logical replication parallel apply process."
RECOVERY_WAL_STREAM "Waiting in main loop of startup process for WAL to arrive, during streaming recovery."
+REPACK_WORKER_MAIN "Waiting in main loop of REPACK decoding worker process."
REPLICATION_SLOTSYNC_MAIN "Waiting in main loop of slot synchronization."
REPLICATION_SLOTSYNC_SHUTDOWN "Waiting for slot sync worker to shut down."
SYSLOGGER_MAIN "Waiting in main loop of syslogger process."
@@ -154,6 +155,7 @@ RECOVERY_CONFLICT_SNAPSHOT "Waiting for recovery conflict resolution for a vacuu
RECOVERY_CONFLICT_TABLESPACE "Waiting for recovery conflict resolution for dropping a tablespace."
RECOVERY_END_COMMAND "Waiting for <xref linkend="guc-recovery-end-command"/> to complete."
RECOVERY_PAUSE "Waiting for recovery to be resumed."
+REPACK_WORKER_EXPORT "Waiting for decoding worker to export a new output file."
REPLICATION_ORIGIN_DROP "Waiting for a replication origin to become inactive so it can be dropped."
REPLICATION_SLOT_DROP "Waiting for a replication slot to become inactive so it can be dropped."
RESTORE_COMMAND "Waiting for <xref linkend="guc-restore-command"/> to complete."
diff --git a/src/include/access/tableam.h b/src/include/access/tableam.h
index 76aa993009a..15760363a1a 100644
--- a/src/include/access/tableam.h
+++ b/src/include/access/tableam.h
@@ -22,7 +22,6 @@
#include "access/xact.h"
#include "commands/vacuum.h"
#include "executor/tuptable.h"
-#include "replication/logical.h"
#include "storage/read_stream.h"
#include "utils/rel.h"
#include "utils/snapshot.h"
@@ -631,7 +630,6 @@ typedef struct TableAmRoutine
bool use_sort,
TransactionId OldestXmin,
Snapshot snapshot,
- LogicalDecodingContext *decoding_ctx,
TransactionId *xid_cutoff,
MultiXactId *multi_cutoff,
double *num_tuples,
@@ -1651,8 +1649,6 @@ table_relation_copy_data(Relation rel, const RelFileLocator *newrlocator)
* - *multi_cutoff - ditto
* - snapshot - if != NULL, ignore data changes done by transactions that this
* (MVCC) snapshot considers still in-progress or in the future.
- * - decoding_ctx - logical decoding context, to capture concurrent data
- * changes.
*
* Output parameters:
* - *xid_cutoff - rel's new relfrozenxid value, may be invalid
@@ -1666,7 +1662,6 @@ table_relation_copy_for_cluster(Relation OldTable, Relation NewTable,
bool use_sort,
TransactionId OldestXmin,
Snapshot snapshot,
- LogicalDecodingContext *decoding_ctx,
TransactionId *xid_cutoff,
MultiXactId *multi_cutoff,
double *num_tuples,
@@ -1675,7 +1670,7 @@ table_relation_copy_for_cluster(Relation OldTable, Relation NewTable,
{
OldTable->rd_tableam->relation_copy_for_cluster(OldTable, NewTable, OldIndex,
use_sort, OldestXmin,
- snapshot, decoding_ctx,
+ snapshot,
xid_cutoff, multi_cutoff,
num_tuples, tups_vacuumed,
tups_recently_dead);
diff --git a/src/include/commands/cluster.h b/src/include/commands/cluster.h
index 6a5c476294a..1b05d5d418b 100644
--- a/src/include/commands/cluster.h
+++ b/src/include/commands/cluster.h
@@ -17,11 +17,13 @@
#include "nodes/parsenodes.h"
#include "parser/parse_node.h"
#include "replication/decode.h"
+#include "postmaster/bgworker.h"
#include "replication/logical.h"
+#include "storage/buffile.h"
#include "storage/lock.h"
+#include "storage/shm_mq.h"
#include "utils/relcache.h"
#include "utils/resowner.h"
-#include "utils/tuplestore.h"
/* flag bits for ClusterParams->options */
@@ -44,6 +46,9 @@ typedef struct ClusterParams
* The following definitions are used by REPACK CONCURRENTLY.
*/
+/*
+ * Stored as a single byte in the output file.
+ */
typedef enum
{
CHANGE_INSERT,
@@ -52,68 +57,30 @@ typedef enum
CHANGE_DELETE
} ConcurrentChangeKind;
-typedef struct ConcurrentChange
-{
- /* See the enum above. */
- ConcurrentChangeKind kind;
-
- /*
- * The actual tuple.
- *
- * The tuple data follows the ConcurrentChange structure. Before use make
- * sure the tuple is correctly aligned (ConcurrentChange can be stored as
- * bytea) and that tuple->t_data is fixed.
- */
- HeapTupleData tup_data;
-} ConcurrentChange;
-
-#define SizeOfConcurrentChange (offsetof(ConcurrentChange, tup_data) + \
- sizeof(HeapTupleData))
-
/*
* Logical decoding state.
*
- * Here we store the data changes that we decode from WAL while the table
- * contents is being copied to a new storage. Also the necessary metadata
- * needed to apply these changes to the table is stored here.
+ * The output plugin uses it to store the data changes that it decodes from
+ * WAL while the table contents is being copied to a new storage.
*/
typedef struct RepackDecodingState
{
/* The relation whose changes we're decoding. */
Oid relid;
- /* Replication slot name. */
- NameData slotname;
-
- /*
- * Decoded changes are stored here. Although we try to avoid excessive
- * batches, it can happen that the changes need to be stored to disk. The
- * tuplestore does this transparently.
- */
- Tuplestorestate *tstore;
-
- /* The current number of changes in tstore. */
- double nchanges;
-
- /*
- * Descriptor to store the ConcurrentChange structure serialized (bytea).
- * We can't store the tuple directly because tuplestore only supports
- * minimum tuple and we may need to transfer OID system column from the
- * output plugin. Also we need to transfer the change kind, so it's better
- * to put everything in the structure than to use 2 tuplestores "in
- * parallel".
- */
- TupleDesc tupdesc_change;
-
- /* Tuple descriptor needed to update indexes. */
+ /* Tuple descriptor of the relation being processed. */
TupleDesc tupdesc;
- /* Slot to retrieve data from tstore. */
- TupleTableSlot *tsslot;
-
- ResourceOwner resowner;
+ /* The current output file. */
+ BufFile *file;
} RepackDecodingState;
+extern PGDLLIMPORT volatile sig_atomic_t RepackMessagePending;
+
+extern bool IsRepackWorker(void);
+extern void HandleRepackMessageInterrupt(void);
+extern void ProcessRepackMessages(void);
+
extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel);
extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid,
@@ -136,6 +103,6 @@ extern void finish_heap_swap(Oid OIDOldHeap, Oid OIDNewHeap,
extern bool am_decoding_for_repack(void);
extern bool change_useless_for_repack(XLogRecordBuffer *buf);
-extern void repack_decode_concurrent_changes(LogicalDecodingContext *ctx,
- XLogRecPtr end_of_wal);
+
+extern void RepackWorkerMain(Datum main_arg);
#endif /* CLUSTER_H */
diff --git a/src/include/storage/procsignal.h b/src/include/storage/procsignal.h
index e52b8eb7697..3ef35ca6b80 100644
--- a/src/include/storage/procsignal.h
+++ b/src/include/storage/procsignal.h
@@ -36,6 +36,7 @@ typedef enum
PROCSIG_BARRIER, /* global barrier interrupt */
PROCSIG_LOG_MEMORY_CONTEXT, /* ask backend to log the memory contexts */
PROCSIG_PARALLEL_APPLY_MESSAGE, /* Message from parallel apply workers */
+ PROCSIG_REPACK_MESSAGE, /* Message from repack worker */
/* Recovery conflict reasons */
PROCSIG_RECOVERY_CONFLICT_FIRST,
diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list
index a0b7b38a5e2..d1a694f9008 100644
--- a/src/tools/pgindent/typedefs.list
+++ b/src/tools/pgindent/typedefs.list
@@ -496,7 +496,6 @@ CompressFileHandle
CompressionLocation
CompressorState
ComputeXidHorizonsResult
-ConcurrentChange
ConcurrentChangeKind
ConditionVariable
ConditionVariableMinimallyPadded
@@ -636,6 +635,9 @@ DeclareCursorStmt
DecodedBkpBlock
DecodedXLogRecord
DecodingOutputState
+DecodingWorker
+DecodingWorkerShared
+DecodingWorkerState
DefElem
DefElemAction
DefaultACLInfo
--
2.47.3
--=-=-=
Content-Type: text/plain
Content-Disposition: attachment;
filename=v30-0006-Use-multiple-snapshots-to-copy-the-data.patch
^ permalink raw reply [nested|flat] 55+ messages in thread
* [PATCH 5/6] Use background worker to do logical decoding.
@ 2026-01-22 08:23 Antonin Houska <[email protected]>
0 siblings, 0 replies; 55+ messages in thread
From: Antonin Houska @ 2026-01-22 08:23 UTC (permalink / raw)
If the backend performing REPACK (CONCURRENTLY) does both data copying and
logical decoding, it has to "travel in time" back and forth and therefore it
has to invalidate system caches quite a few times. (The copying and the
decoding work with different catalog snapshots.) As the decoding worker has
separate caches, the switching is not necessary.
Without the worker, it'd also be difficult to switch between potentially long
running tasks like index build and WAL decoding. (No decoding during that time
at all can suspend archiving / recycling of WAL segments for some time, which
in turn may result in full disk.)
Another problem is that, after having acquired AccessExclusiveLock (in order
to swap the files), the backend needs to both decode and apply the data
changes that took place while it was waiting for the lock. With the decoding
worker, the decoding runs all the time, so the backend only needs to apply the
changes. This can reduce the time the exclusive lock is held for.
Note that the code added in order to handle ERRORs in the background worker
almost duplicates the existing code that does the same for other types of
workers (See ProcessParallelMessages() and
ProcessParallelApplyMessages()). Refactoring of the existing code might be
useful, to reduce the duplication.
---
src/backend/access/heap/heapam_handler.c | 44 -
src/backend/commands/cluster.c | 1179 +++++++++++++----
src/backend/libpq/pqmq.c | 5 +
src/backend/postmaster/bgworker.c | 4 +
src/backend/replication/logical/logical.c | 6 +-
.../pgoutput_repack/pgoutput_repack.c | 54 +-
src/backend/storage/ipc/procsignal.c | 4 +
src/backend/tcop/postgres.c | 4 +
.../utils/activity/wait_event_names.txt | 1 +
src/include/access/tableam.h | 7 +-
src/include/commands/cluster.h | 71 +-
src/include/storage/procsignal.h | 1 +
src/tools/pgindent/typedefs.list | 3 +-
13 files changed, 981 insertions(+), 402 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index 908f1ef66c6..8589b3c940e 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -33,7 +33,6 @@
#include "catalog/index.h"
#include "catalog/storage.h"
#include "catalog/storage_xlog.h"
-#include "commands/cluster.h"
#include "commands/progress.h"
#include "executor/executor.h"
#include "miscadmin.h"
@@ -688,7 +687,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
Relation OldIndex, bool use_sort,
TransactionId OldestXmin,
Snapshot snapshot,
- LogicalDecodingContext *decoding_ctx,
TransactionId *xid_cutoff,
MultiXactId *multi_cutoff,
double *num_tuples,
@@ -710,7 +708,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
BufferHeapTupleTableSlot *hslot;
BlockNumber prev_cblock = InvalidBlockNumber;
bool concurrent = snapshot != NULL;
- XLogRecPtr end_of_wal_prev = GetFlushRecPtr(NULL);
/* Remember if it's a system catalog */
is_system_catalog = IsSystemRelation(OldHeap);
@@ -971,31 +968,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
ct_val[1] = *num_tuples;
pgstat_progress_update_multi_param(2, ct_index, ct_val);
}
-
- /*
- * Process the WAL produced by the load, as well as by other
- * transactions, so that the replication slot can advance and WAL does
- * not pile up. Use wal_segment_size as a threshold so that we do not
- * introduce the decoding overhead too often.
- *
- * Of course, we must not apply the changes until the initial load has
- * completed.
- *
- * Note that our insertions into the new table should not be decoded
- * as we (intentionally) do not write the logical decoding specific
- * information to WAL.
- */
- if (concurrent)
- {
- XLogRecPtr end_of_wal;
-
- end_of_wal = GetFlushRecPtr(NULL);
- if ((end_of_wal - end_of_wal_prev) > wal_segment_size)
- {
- repack_decode_concurrent_changes(decoding_ctx, end_of_wal);
- end_of_wal_prev = end_of_wal;
- }
- }
}
if (indexScan != NULL)
@@ -1041,22 +1013,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
-
- /*
- * Try to keep the amount of not-yet-decoded WAL small, like
- * above.
- */
- if (concurrent)
- {
- XLogRecPtr end_of_wal;
-
- end_of_wal = GetFlushRecPtr(NULL);
- if ((end_of_wal - end_of_wal_prev) > wal_segment_size)
- {
- repack_decode_concurrent_changes(decoding_ctx, end_of_wal);
- end_of_wal_prev = end_of_wal;
- }
- }
}
tuplesort_end(tuplesort);
diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c
index 03ccf10b782..e988a7a7296 100644
--- a/src/backend/commands/cluster.c
+++ b/src/backend/commands/cluster.c
@@ -12,12 +12,13 @@
* In concurrent mode, we lock the table with only ShareUpdateExclusiveLock,
* then do an initial copy as above. However, while the tuples are being
* copied, concurrent transactions could modify the table. To cope with those
- * changes, we rely on logical decoding to obtain them from WAL. The changes
- * are accumulated in a tuplestore. Once the initial copy is complete, we
- * read the changes from the tuplestore and re-apply them on the new heap.
- * Then we upgrade our ShareUpdateExclusiveLock to AccessExclusiveLock and
- * swap the relfilenodes. This way, the time we hold a strong lock on the
- * table is much reduced, and the bloat is eliminated.
+ * changes, we rely on logical decoding to obtain them from WAL. A bgworker
+ * consumes WAL while the initial copy is ongoing (to prevent excessive WAL
+ * from being reserved), and accumulates the changes in a file. Once the
+ * initial copy is complete, we read the changes from the file and re-apply
+ * them on the new heap. Then we upgrade our ShareUpdateExclusiveLock to
+ * AccessExclusiveLock and swap the relfilenodes. This way, the time we hold
+ * a strong lock on the table is much reduced, and the bloat is eliminated.
*
* There is hardly anything left of Paul Brown's original implementation...
*
@@ -45,6 +46,7 @@
#include "access/xlog_internal.h"
#include "access/xloginsert.h"
#include "access/xlogutils.h"
+#include "access/xlogwait.h"
#include "catalog/catalog.h"
#include "catalog/dependency.h"
#include "catalog/heap.h"
@@ -61,6 +63,8 @@
#include "commands/tablecmds.h"
#include "commands/vacuum.h"
#include "executor/executor.h"
+#include "libpq/pqformat.h"
+#include "libpq/pqmq.h"
#include "miscadmin.h"
#include "optimizer/optimizer.h"
#include "pgstat.h"
@@ -71,6 +75,8 @@
#include "storage/ipc.h"
#include "storage/lmgr.h"
#include "storage/predicate.h"
+#include "storage/procsignal.h"
+#include "tcop/tcopprot.h"
#include "utils/acl.h"
#include "utils/fmgroids.h"
#include "utils/guc.h"
@@ -117,6 +123,12 @@ typedef struct IndexInsertState
/* The WAL segment being decoded. */
static XLogSegNo repack_current_segment = 0;
+/*
+ * The first file exported by the decoding worker must contain a snapshot, the
+ * following ones contain the data changes.
+ */
+#define WORKER_FILE_SNAPSHOT 0
+
/*
* Information needed to apply concurrent data changes.
*/
@@ -136,8 +148,113 @@ typedef struct ChangeDest
/* Needed to update indexes of rel_dst. */
IndexInsertState *iistate;
+
+ /*
+ * Sequential number of the file containing the changes.
+ *
+ * TODO This field makes the structure name less descriptive. Should we
+ * rename it, e.g. to ChangeApplyInfo?
+ */
+ int file_seq;
} ChangeDest;
+/*
+ * Layout of shared memory used for communication between backend and the
+ * worker that performs logical decoding of data changes
+ */
+typedef struct DecodingWorkerShared
+{
+ /* Is the decoding initialized? */
+ bool initialized;
+
+ /*
+ * Once the worker has reached this LSN, it should close the current
+ * output file and either create a new one or exit, according to the field
+ * 'done'. If the value is InvalidXLogRecPtr, the worker should decode all
+ * the WAL available and keep checking this field. It is ok if the worker
+ * had already decoded records whose LSN is >= lsn_upto before this field
+ * has been set.
+ */
+ XLogRecPtr lsn_upto;
+
+ /* Exit after closing the current file? */
+ bool done;
+
+ /* The output is stored here. */
+ SharedFileSet sfs;
+
+ /* Number of the last file exported by the worker. */
+ int last_exported;
+
+ /* Synchronize access to the fields above. */
+ slock_t mutex;
+
+ /* Database to connect to. */
+ Oid dbid;
+
+ /* Role to connect as. */
+ Oid roleid;
+
+ /* Decode data changes of this relation. */
+ Oid relid;
+
+ /* The backend uses this to wait for the worker. */
+ ConditionVariable cv;
+
+ /* Info to signal the backend. */
+ PGPROC *backend_proc;
+ pid_t backend_pid;
+ ProcNumber backend_proc_number;
+
+ /* Error queue. */
+ shm_mq *error_mq;
+
+ /*
+ * Memory the queue is located in.
+ *
+ * For considerations on the value see the comments of
+ * PARALLEL_ERROR_QUEUE_SIZE.
+ */
+#define REPACK_ERROR_QUEUE_SIZE 16384
+ char error_queue[FLEXIBLE_ARRAY_MEMBER];
+} DecodingWorkerShared;
+
+/*
+ * Generate worker's output file name. If relations of the same 'relid' happen
+ * to be processed at the same time, they must be from different databases and
+ * therefore different backends must be involved. (PID is already present in
+ * the fileset name.)
+ */
+static inline void
+DecodingWorkerFileName(char *fname, Oid relid, uint32 seq)
+{
+ snprintf(fname, MAXPGPATH, "%u-%u", relid, seq);
+}
+
+/*
+ * Backend-local information to control the decoding worker.
+ */
+typedef struct DecodingWorker
+{
+ /* The worker. */
+ BackgroundWorkerHandle *handle;
+
+ /* DecodingWorkerShared is in this segment. */
+ dsm_segment *seg;
+
+ /* Handle of the error queue. */
+ shm_mq_handle *error_mqh;
+} DecodingWorker;
+
+/* Pointer to currently running decoding worker. */
+static DecodingWorker *decoding_worker = NULL;
+
+/*
+ * Is there a message sent by a repack worker that the backend needs to
+ * receive?
+ */
+volatile sig_atomic_t RepackMessagePending = false;
+
static bool cluster_rel_recheck(RepackCommand cmd, Relation OldHeap,
Oid indexOid, Oid userid, LOCKMODE lmode,
int options);
@@ -145,7 +262,7 @@ static void check_repack_concurrently_requirements(Relation rel);
static void rebuild_relation(Relation OldHeap, Relation index, bool verbose,
bool concurrent);
static void copy_table_data(Relation NewHeap, Relation OldHeap, Relation OldIndex,
- Snapshot snapshot, LogicalDecodingContext *decoding_ctx,
+ Snapshot snapshot,
bool verbose,
bool *pSwapToastByContent,
TransactionId *pFreezeXid,
@@ -158,12 +275,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd,
static bool cluster_is_permitted_for_relation(RepackCommand cmd,
Oid relid, Oid userid);
-static void begin_concurrent_repack(Relation rel);
-static void end_concurrent_repack(void);
static LogicalDecodingContext *setup_logical_decoding(Oid relid);
-static HeapTuple get_changed_tuple(char *change);
-static void apply_concurrent_changes(RepackDecodingState *dstate,
- ChangeDest *dest);
+static bool decode_concurrent_changes(LogicalDecodingContext *ctx,
+ DecodingWorkerShared *shared);
+static void apply_concurrent_changes(BufFile *file, ChangeDest *dest);
static void apply_concurrent_insert(Relation rel, HeapTuple tup,
IndexInsertState *iistate,
TupleTableSlot *index_slot);
@@ -175,9 +290,9 @@ static void apply_concurrent_delete(Relation rel, HeapTuple tup_target);
static HeapTuple find_target_tuple(Relation rel, ChangeDest *dest,
HeapTuple tup_key,
TupleTableSlot *ident_slot);
-static void process_concurrent_changes(LogicalDecodingContext *decoding_ctx,
- XLogRecPtr end_of_wal,
- ChangeDest *dest);
+static void process_concurrent_changes(XLogRecPtr end_of_wal,
+ ChangeDest *dest,
+ bool done);
static IndexInsertState *get_index_insert_state(Relation relation,
Oid ident_index_id,
Relation *ident_index_p);
@@ -186,7 +301,6 @@ static ScanKey build_identity_key(Oid ident_idx_oid, Relation rel_src,
static void free_index_insert_state(IndexInsertState *iistate);
static void cleanup_logical_decoding(LogicalDecodingContext *ctx);
static void rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
- LogicalDecodingContext *decoding_ctx,
TransactionId frozenXid,
MultiXactId cutoffMulti);
static List *build_new_indexes(Relation NewHeap, Relation OldHeap, List *OldIndexes);
@@ -196,6 +310,13 @@ static Relation process_single_relation(RepackStmt *stmt,
ClusterParams *params);
static Oid determine_clustered_index(Relation rel, bool usingindex,
const char *indexname);
+static void start_decoding_worker(Oid relid);
+static void stop_decoding_worker(void);
+static void repack_worker_internal(dsm_segment *seg);
+static void export_initial_snapshot(Snapshot snapshot,
+ DecodingWorkerShared *shared);
+static Snapshot get_initial_snapshot(DecodingWorker *worker);
+static void ProcessRepackMessage(StringInfo msg);
static const char *RepackCommandAsString(RepackCommand cmd);
@@ -619,20 +740,20 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid,
/* rebuild_relation does all the dirty work */
PG_TRY();
{
- /*
- * For concurrent processing, make sure that our logical decoding
- * ignores data changes of other tables than the one we are
- * processing.
- */
- if (concurrent)
- begin_concurrent_repack(OldHeap);
-
rebuild_relation(OldHeap, index, verbose, concurrent);
}
PG_FINALLY();
{
if (concurrent)
- end_concurrent_repack();
+ {
+ /*
+ * Since during normal operation the worker was already asked to
+ * exit, stopping it explicitly is especially important on ERROR.
+ * However it still seems a good practice to make sure that the
+ * worker never survives the REPACK command.
+ */
+ stop_decoding_worker();
+ }
}
PG_END_TRY();
@@ -929,7 +1050,6 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
bool swap_toast_by_content;
TransactionId frozenXid;
MultiXactId cutoffMulti;
- LogicalDecodingContext *decoding_ctx = NULL;
Snapshot snapshot = NULL;
#if USE_ASSERT_CHECKING
LOCKMODE lmode;
@@ -943,19 +1063,36 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
if (concurrent)
{
/*
- * Prepare to capture the concurrent data changes.
+ * The worker needs to be member of the locking group we're the leader
+ * of. We ought to become the leader before the worker starts. The
+ * worker will join the group as soon as it starts.
*
- * Note that this call waits for all transactions with XID already
- * assigned to finish. If some of those transactions is waiting for a
- * lock conflicting with ShareUpdateExclusiveLock on our table (e.g.
- * it runs CREATE INDEX), we can end up in a deadlock. Not sure this
- * risk is worth unlocking/locking the table (and its clustering
- * index) and checking again if its still eligible for REPACK
- * CONCURRENTLY.
+ * This is to make sure that the deadlock described below is
+ * detectable by deadlock.c: if the worker waits for a transaction to
+ * complete and we are waiting for the worker output, then effectively
+ * we (i.e. this backend) are waiting for that transaction.
*/
- decoding_ctx = setup_logical_decoding(tableOid);
+ BecomeLockGroupLeader();
+
+ /*
+ * Start the worker that decodes data changes applied while we're
+ * copying the table contents.
+ *
+ * Note that the worker has to wait for all transactions with XID
+ * already assigned to finish. If some of those transactions is
+ * waiting for a lock conflicting with ShareUpdateExclusiveLock on our
+ * table (e.g. it runs CREATE INDEX), we can end up in a deadlock.
+ * Not sure this risk is worth unlocking/locking the table (and its
+ * clustering index) and checking again if it's still eligible for
+ * REPACK CONCURRENTLY.
+ */
+ start_decoding_worker(tableOid);
+
+ /*
+ * Wait until the worker has the initial snapshot and retrieve it.
+ */
+ snapshot = get_initial_snapshot(decoding_worker);
- snapshot = SnapBuildInitialSnapshotForRepack(decoding_ctx->snapshot_builder);
PushActiveSnapshot(snapshot);
}
@@ -980,7 +1117,7 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
NewHeap = table_open(OIDNewHeap, NoLock);
/* Copy the heap data into the new table in the desired order */
- copy_table_data(NewHeap, OldHeap, index, snapshot, decoding_ctx, verbose,
+ copy_table_data(NewHeap, OldHeap, index, snapshot, verbose,
&swap_toast_by_content, &frozenXid, &cutoffMulti);
/* The historic snapshot won't be needed anymore. */
@@ -1001,14 +1138,11 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
if (index)
index_close(index, NoLock);
- rebuild_relation_finish_concurrent(NewHeap, OldHeap, decoding_ctx,
- frozenXid, cutoffMulti);
+ rebuild_relation_finish_concurrent(NewHeap, OldHeap, frozenXid,
+ cutoffMulti);
pgstat_progress_update_param(PROGRESS_REPACK_PHASE,
PROGRESS_REPACK_PHASE_FINAL_CLEANUP);
-
- /* Done with decoding. */
- cleanup_logical_decoding(decoding_ctx);
}
else
{
@@ -1179,8 +1313,7 @@ make_new_heap(Oid OIDOldHeap, Oid NewTableSpace, Oid NewAccessMethod,
*/
static void
copy_table_data(Relation NewHeap, Relation OldHeap, Relation OldIndex,
- Snapshot snapshot, LogicalDecodingContext *decoding_ctx,
- bool verbose, bool *pSwapToastByContent,
+ Snapshot snapshot, bool verbose, bool *pSwapToastByContent,
TransactionId *pFreezeXid, MultiXactId *pCutoffMulti)
{
Relation relRelation;
@@ -1341,7 +1474,6 @@ copy_table_data(Relation NewHeap, Relation OldHeap, Relation OldIndex,
*/
table_relation_copy_for_cluster(OldHeap, NewHeap, OldIndex, use_sort,
cutoffs.OldestXmin, snapshot,
- decoding_ctx,
&cutoffs.FreezeLimit,
&cutoffs.MultiXactCutoff,
&num_tuples, &tups_vacuumed,
@@ -2371,62 +2503,10 @@ RepackCommandAsString(RepackCommand cmd)
case REPACK_COMMAND_CLUSTER:
return "CLUSTER";
}
- return "???"; /* keep compiler quiet */
+ return "???"; /* keep compiler quiet */
}
-/*
- * Call this function before REPACK CONCURRENTLY starts to setup logical
- * decoding. It makes sure that other users of the table put enough
- * information into WAL.
- *
- * The point is that at various places we expect that the table we're
- * processing is treated like a system catalog. For example, we need to be
- * able to scan it using a "historic snapshot" anytime during the processing
- * (as opposed to scanning only at the start point of the decoding, as logical
- * replication does during initial table synchronization), in order to apply
- * concurrent UPDATE / DELETE commands.
- *
- * Note that TOAST table needs no attention here as it's not scanned using
- * historic snapshot.
- */
-static void
-begin_concurrent_repack(Relation rel)
-{
- Oid toastrelid;
-
- /*
- * Avoid logical decoding of other relations by this backend. The lock we
- * have guarantees that the actual locator cannot be changed concurrently:
- * TRUNCATE needs AccessExclusiveLock.
- */
- Assert(CheckRelationLockedByMe(rel, ShareUpdateExclusiveLock, false));
- repacked_rel_locator = rel->rd_locator;
- toastrelid = rel->rd_rel->reltoastrelid;
- if (OidIsValid(toastrelid))
- {
- Relation toastrel;
-
- /* Avoid logical decoding of other TOAST relations. */
- toastrel = table_open(toastrelid, AccessShareLock);
- repacked_rel_toast_locator = toastrel->rd_locator;
- table_close(toastrel, AccessShareLock);
- }
-}
-
-/*
- * Call this when done with REPACK CONCURRENTLY.
- */
-static void
-end_concurrent_repack(void)
-{
- /*
- * Restore normal function of (future) logical decoding for this backend.
- */
- repacked_rel_locator.relNumber = InvalidOid;
- repacked_rel_toast_locator.relNumber = InvalidOid;
-}
-
/*
* Is this backend performing logical decoding on behalf of REPACK
* (CONCURRENTLY) ?
@@ -2491,9 +2571,10 @@ static LogicalDecodingContext *
setup_logical_decoding(Oid relid)
{
Relation rel;
- TupleDesc tupdesc;
+ Oid toastrelid;
LogicalDecodingContext *ctx;
- RepackDecodingState *dstate = palloc0_object(RepackDecodingState);
+ NameData slotname;
+ RepackDecodingState *dstate;
/*
* REPACK CONCURRENTLY is not allowed in a transaction block, so this
@@ -2501,20 +2582,21 @@ setup_logical_decoding(Oid relid)
*/
Assert(!TransactionIdIsValid(GetTopTransactionIdIfAny()));
- /*
- * A single backend should not execute multiple REPACK commands at a time,
- * so use PID to make the slot unique.
- */
- snprintf(NameStr(dstate->slotname), NAMEDATALEN, "repack_%d", MyProcPid);
-
/*
* Make sure we can use logical decoding.
*/
CheckSlotPermissions();
CheckLogicalDecodingRequirements();
- /* RS_TEMPORARY so that the slot gets cleaned up on ERROR. */
- ReplicationSlotCreate(NameStr(dstate->slotname), true, RS_TEMPORARY,
- false, false, false);
+ /*
+ * A single backend should not execute multiple REPACK commands at a time,
+ * so use PID to make the slot unique.
+ *
+ * RS_TEMPORARY so that the slot gets cleaned up on ERROR.
+ */
+ snprintf(NameStr(slotname), NAMEDATALEN, "repack_%d", MyProcPid);
+ ReplicationSlotCreate(NameStr(slotname), true, RS_TEMPORARY, false, false,
+ false);
+
EnsureLogicalDecodingEnabled();
/*
@@ -2537,104 +2619,109 @@ setup_logical_decoding(Oid relid)
DecodingContextFindStartpoint(ctx);
+ /*
+ * decode_concurrent_changes() needs non-blocking callback.
+ */
+ ctx->reader->routine.page_read = read_local_xlog_page_no_wait;
+
+ /*
+ * read_local_xlog_page_no_wait() needs to be able to indicate the end of
+ * WAL.
+ */
+ ctx->reader->private_data = MemoryContextAllocZero(ctx->context,
+ sizeof(ReadLocalXLogPageNoWaitPrivate));
+
+
/* Some WAL records should have been read. */
Assert(ctx->reader->EndRecPtr != InvalidXLogRecPtr);
+ /*
+ * Initialize repack_current_segment so that we can notice WAL segment
+ * boundaries.
+ */
XLByteToSeg(ctx->reader->EndRecPtr, repack_current_segment,
wal_segment_size);
- /*
- * Setup structures to store decoded changes.
- */
+ dstate = palloc0_object(RepackDecodingState);
dstate->relid = relid;
- dstate->tstore = tuplestore_begin_heap(false, false,
- maintenance_work_mem);
- /* Caller should already have the table locked. */
- rel = table_open(relid, NoLock);
- tupdesc = CreateTupleDescCopy(RelationGetDescr(rel));
- dstate->tupdesc = tupdesc;
- table_close(rel, NoLock);
+ /*
+ * Tuple descriptor may be needed to flatten a tuple before we write it to
+ * a file. A copy is needed because the decoding worker invalidates system
+ * caches before it starts to do the actual work.
+ */
+ rel = table_open(relid, AccessShareLock);
+ dstate->tupdesc = CreateTupleDescCopy(RelationGetDescr(rel));
- /* Initialize the descriptor to store the changes ... */
- dstate->tupdesc_change = CreateTemplateTupleDesc(1);
+ /* Avoid logical decoding of other relations. */
+ repacked_rel_locator = rel->rd_locator;
+ toastrelid = rel->rd_rel->reltoastrelid;
+ if (OidIsValid(toastrelid))
+ {
+ Relation toastrel;
- TupleDescInitEntry(dstate->tupdesc_change, 1, NULL, BYTEAOID, -1, 0);
- /* ... as well as the corresponding slot. */
- dstate->tsslot = MakeSingleTupleTableSlot(dstate->tupdesc_change,
- &TTSOpsMinimalTuple);
+ /* Avoid logical decoding of other TOAST relations. */
+ toastrel = table_open(toastrelid, AccessShareLock);
+ repacked_rel_toast_locator = toastrel->rd_locator;
+ table_close(toastrel, AccessShareLock);
+ }
+ table_close(rel, AccessShareLock);
- dstate->resowner = ResourceOwnerCreate(CurrentResourceOwner,
- "logical decoding");
+ /* The file will be set as soon as we have it opened. */
+ dstate->file = NULL;
ctx->output_writer_private = dstate;
+
return ctx;
}
/*
- * Retrieve tuple from ConcurrentChange structure.
+ * Decode logical changes from the WAL sequence and store them to a file.
*
- * The input data starts with the structure but it might not be appropriately
- * aligned.
- */
-static HeapTuple
-get_changed_tuple(char *change)
-{
- HeapTupleData tup_data;
- HeapTuple result;
- char *src;
-
- /*
- * Ensure alignment before accessing the fields. (This is why we can't use
- * heap_copytuple() instead of this function.)
- */
- src = change + offsetof(ConcurrentChange, tup_data);
- memcpy(&tup_data, src, sizeof(HeapTupleData));
-
- result = (HeapTuple) palloc(HEAPTUPLESIZE + tup_data.t_len);
- memcpy(result, &tup_data, sizeof(HeapTupleData));
- result->t_data = (HeapTupleHeader) ((char *) result + HEAPTUPLESIZE);
- src = change + SizeOfConcurrentChange;
- memcpy(result->t_data, src, result->t_len);
-
- return result;
-}
-
-/*
- * Decode logical changes from the WAL sequence up to end_of_wal.
+ * If true is returned, there is no more work for the worker.
*/
-void
-repack_decode_concurrent_changes(LogicalDecodingContext *ctx,
- XLogRecPtr end_of_wal)
+static bool
+decode_concurrent_changes(LogicalDecodingContext *ctx,
+ DecodingWorkerShared *shared)
{
RepackDecodingState *dstate;
- ResourceOwner resowner_old;
+ XLogRecPtr lsn_upto;
+ bool done;
+ char fname[MAXPGPATH];
dstate = (RepackDecodingState *) ctx->output_writer_private;
- resowner_old = CurrentResourceOwner;
- CurrentResourceOwner = dstate->resowner;
- PG_TRY();
+ /* Open the output file. */
+ DecodingWorkerFileName(fname, shared->relid, shared->last_exported + 1);
+ dstate->file = BufFileCreateFileSet(&shared->sfs.fs, fname);
+
+ SpinLockAcquire(&shared->mutex);
+ lsn_upto = shared->lsn_upto;
+ done = shared->done;
+ SpinLockRelease(&shared->mutex);
+
+ while (true)
{
- while (ctx->reader->EndRecPtr < end_of_wal)
- {
- XLogRecord *record;
- XLogSegNo segno_new;
- char *errm = NULL;
- XLogRecPtr end_lsn;
+ XLogRecord *record;
+ XLogSegNo segno_new;
+ char *errm = NULL;
+ XLogRecPtr end_lsn;
- record = XLogReadRecord(ctx->reader, &errm);
- if (errm)
- elog(ERROR, "%s", errm);
+ CHECK_FOR_INTERRUPTS();
- if (record != NULL)
- LogicalDecodingProcessRecord(ctx, ctx->reader);
+ record = XLogReadRecord(ctx->reader, &errm);
+ if (record)
+ {
+ LogicalDecodingProcessRecord(ctx, ctx->reader);
/*
* If WAL segment boundary has been crossed, inform the decoding
- * system that the catalog_xmin can advance. (We can confirm more
- * often, but a filling a single WAL segment should not take much
- * time.)
+ * system that the catalog_xmin can advance.
+ *
+ * TODO Does it make sense to confirm more often? Segment size
+ * seems appropriate for restart_lsn (because less than a segment
+ * cannot be recycled anyway), however more frequent checks might
+ * be beneficial for catalog_xmin.
*/
end_lsn = ctx->reader->EndRecPtr;
XLByteToSeg(end_lsn, segno_new, wal_segment_size);
@@ -2645,80 +2732,137 @@ repack_decode_concurrent_changes(LogicalDecodingContext *ctx,
(uint32) (end_lsn >> 32), (uint32) end_lsn);
repack_current_segment = segno_new;
}
+ }
+ else
+ {
+ ReadLocalXLogPageNoWaitPrivate *priv;
+
+ if (errm)
+ ereport(ERROR, (errmsg("%s", errm)));
- CHECK_FOR_INTERRUPTS();
+ /*
+ * In the decoding loop we do not want to get blocked when there
+ * is no more WAL available, otherwise the loop would become
+ * uninterruptible.
+ */
+ priv = (ReadLocalXLogPageNoWaitPrivate *)
+ ctx->reader->private_data;
+ if (priv->end_of_wal)
+ /* Do not miss the end of WAL condition next time. */
+ priv->end_of_wal = false;
+ else
+ ereport(ERROR, (errmsg("could not read WAL record")));
+ }
+
+ /*
+ * Whether we could read new record or not, keep checking if
+ * 'lsn_upto' was specified.
+ */
+ if (XLogRecPtrIsInvalid(lsn_upto))
+ {
+ SpinLockAcquire(&shared->mutex);
+ lsn_upto = shared->lsn_upto;
+ /* 'done' should be set at the same time as 'lsn_upto' */
+ done = shared->done;
+ SpinLockRelease(&shared->mutex);
+ }
+ if (!XLogRecPtrIsInvalid(lsn_upto) &&
+ ctx->reader->EndRecPtr >= lsn_upto)
+ break;
+
+ if (record == NULL)
+ {
+ int64 timeout = 0;
+ WaitLSNResult res;
+
+ /*
+ * Before we retry reading, wait until new WAL is flushed.
+ *
+ * There is a race condition such that the backend executing
+ * REPACK determines 'lsn_upto', but before it sets the shared
+ * variable, we reach the end of WAL. In that case we'd need to
+ * wait until the next WAL flush (unrelated to REPACK). Although
+ * that should not be a problem in a busy system, it might be
+ * noticeable in other cases, including regression tests (which
+ * are not necessarily executed in parallel). Therefore it makes
+ * sense to use timeout.
+ *
+ * If lsn_upto is valid, WAL records having LSN lower than that
+ * should already have been flushed to disk.
+ */
+ if (XLogRecPtrIsInvalid(lsn_upto))
+ timeout = 100L;
+ res = WaitForLSN(WAIT_LSN_TYPE_PRIMARY_FLUSH,
+ ctx->reader->EndRecPtr + 1,
+ timeout);
+ if (res != WAIT_LSN_RESULT_SUCCESS &&
+ res != WAIT_LSN_RESULT_TIMEOUT)
+ ereport(ERROR, (errmsg("waiting for WAL failed")));
}
- InvalidateSystemCaches();
- CurrentResourceOwner = resowner_old;
- }
- PG_CATCH();
- {
- /* clear all timetravel entries */
- InvalidateSystemCaches();
- CurrentResourceOwner = resowner_old;
- PG_RE_THROW();
}
- PG_END_TRY();
+
+ /*
+ * Close the file so we can make it available to the backend.
+ */
+ BufFileClose(dstate->file);
+ dstate->file = NULL;
+ SpinLockAcquire(&shared->mutex);
+ shared->lsn_upto = InvalidXLogRecPtr;
+ shared->last_exported++;
+ SpinLockRelease(&shared->mutex);
+ ConditionVariableSignal(&shared->cv);
+
+ return done;
}
/*
* Apply changes stored in 'file'.
*/
static void
-apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
+apply_concurrent_changes(BufFile *file, ChangeDest *dest)
{
+ char kind;
+ uint32 t_len;
Relation rel = dest->rel;
TupleTableSlot *index_slot,
*ident_slot;
HeapTuple tup_old = NULL;
- if (dstate->nchanges == 0)
- return;
-
/* TupleTableSlot is needed to pass the tuple to ExecInsertIndexTuples(). */
- index_slot = MakeSingleTupleTableSlot(dstate->tupdesc, &TTSOpsHeapTuple);
+ index_slot = MakeSingleTupleTableSlot(RelationGetDescr(rel),
+ &TTSOpsHeapTuple);
/* A slot to fetch tuples from identity index. */
ident_slot = table_slot_create(rel, NULL);
- while (tuplestore_gettupleslot(dstate->tstore, true, false,
- dstate->tsslot))
+ while (true)
{
- bool shouldFree;
- HeapTuple tup_change,
- tup,
+ size_t nread;
+ HeapTuple tup,
tup_exist;
- char *change_raw,
- *src;
- ConcurrentChange change;
- bool isnull[1];
- Datum values[1];
CHECK_FOR_INTERRUPTS();
- /* Get the change from the single-column tuple. */
- tup_change = ExecFetchSlotHeapTuple(dstate->tsslot, false, &shouldFree);
- heap_deform_tuple(tup_change, dstate->tupdesc_change, values, isnull);
- Assert(!isnull[0]);
-
- /* Make sure we access aligned data. */
- change_raw = (char *) DatumGetByteaP(values[0]);
- src = (char *) VARDATA(change_raw);
- memcpy(&change, src, SizeOfConcurrentChange);
+ nread = BufFileReadMaybeEOF(file, &kind, 1, true);
+ /* Are we done with the file? */
+ if (nread == 0)
+ break;
- /*
- * Extract the tuple from the change. The tuple is copied here because
- * it might be assigned to 'tup_old', in which case it needs to
- * survive into the next iteration.
- */
- tup = get_changed_tuple(src);
+ /* Read the tuple. */
+ BufFileReadExact(file, &t_len, sizeof(t_len));
+ tup = (HeapTuple) palloc(HEAPTUPLESIZE + t_len);
+ tup->t_data = (HeapTupleHeader) ((char *) tup + HEAPTUPLESIZE);
+ BufFileReadExact(file, tup->t_data, t_len);
+ tup->t_len = t_len;
+ ItemPointerSetInvalid(&tup->t_self);
+ tup->t_tableOid = RelationGetRelid(dest->rel);
- if (change.kind == CHANGE_UPDATE_OLD)
+ if (kind == CHANGE_UPDATE_OLD)
{
Assert(tup_old == NULL);
tup_old = tup;
}
- else if (change.kind == CHANGE_INSERT)
+ else if (kind == CHANGE_INSERT)
{
Assert(tup_old == NULL);
@@ -2726,12 +2870,11 @@ apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
pfree(tup);
}
- else if (change.kind == CHANGE_UPDATE_NEW ||
- change.kind == CHANGE_DELETE)
+ else if (kind == CHANGE_UPDATE_NEW || kind == CHANGE_DELETE)
{
HeapTuple tup_key;
- if (change.kind == CHANGE_UPDATE_NEW)
+ if (kind == CHANGE_UPDATE_NEW)
{
tup_key = tup_old != NULL ? tup_old : tup;
}
@@ -2748,7 +2891,7 @@ apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
if (tup_exist == NULL)
elog(ERROR, "failed to find target tuple");
- if (change.kind == CHANGE_UPDATE_NEW)
+ if (kind == CHANGE_UPDATE_NEW)
apply_concurrent_update(rel, tup, tup_exist, dest->iistate,
index_slot);
else
@@ -2763,26 +2906,19 @@ apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
pfree(tup);
}
else
- elog(ERROR, "unrecognized kind of change: %d", change.kind);
+ elog(ERROR, "unrecognized kind of change: %d", kind);
/*
* If a change was applied now, increment CID for next writes and
* update the snapshot so it sees the changes we've applied so far.
*/
- if (change.kind != CHANGE_UPDATE_OLD)
+ if (kind != CHANGE_UPDATE_OLD)
{
CommandCounterIncrement();
UpdateActiveSnapshotCommandId();
}
-
- /* TTSOpsMinimalTuple has .get_heap_tuple==NULL. */
- Assert(shouldFree);
- pfree(tup_change);
}
- tuplestore_clear(dstate->tstore);
- dstate->nchanges = 0;
-
/* Cleanup. */
ExecDropSingleTupleTableSlot(index_slot);
ExecDropSingleTupleTableSlot(ident_slot);
@@ -2957,25 +3093,59 @@ find_target_tuple(Relation rel, ChangeDest *dest, HeapTuple tup_key,
}
/*
- * Decode and apply concurrent changes.
+ * Decode and apply concurrent changes, up to (and including) the record whose
+ * LSN is 'end_of_wal'.
*/
static void
-process_concurrent_changes(LogicalDecodingContext *decoding_ctx,
- XLogRecPtr end_of_wal, ChangeDest *dest)
+process_concurrent_changes(XLogRecPtr end_of_wal, ChangeDest *dest, bool done)
{
- RepackDecodingState *dstate;
+ DecodingWorkerShared *shared;
+ char fname[MAXPGPATH];
+ BufFile *file;
pgstat_progress_update_param(PROGRESS_REPACK_PHASE,
PROGRESS_REPACK_PHASE_CATCH_UP);
- dstate = (RepackDecodingState *) decoding_ctx->output_writer_private;
+ /* Ask the worker for the file. */
+ shared = (DecodingWorkerShared *) dsm_segment_address(decoding_worker->seg);
+ SpinLockAcquire(&shared->mutex);
+ shared->lsn_upto = end_of_wal;
+ shared->done = done;
+ SpinLockRelease(&shared->mutex);
+
+ /*
+ * The worker needs to finish processing of the current WAL record. Even
+ * if it's idle, it'll need to close the output file. Thus we're likely to
+ * wait, so prepare for sleep.
+ */
+ ConditionVariablePrepareToSleep(&shared->cv);
+ for (;;)
+ {
+ int last_exported;
- repack_decode_concurrent_changes(decoding_ctx, end_of_wal);
+ SpinLockAcquire(&shared->mutex);
+ last_exported = shared->last_exported;
+ SpinLockRelease(&shared->mutex);
- if (dstate->nchanges == 0)
- return;
+ /*
+ * Has the worker exported the file we are waiting for?
+ */
+ if (last_exported == dest->file_seq)
+ break;
- apply_concurrent_changes(dstate, dest);
+ ConditionVariableSleep(&shared->cv, WAIT_EVENT_REPACK_WORKER_EXPORT);
+ }
+ ConditionVariableCancelSleep();
+
+ /* Open the file. */
+ DecodingWorkerFileName(fname, shared->relid, dest->file_seq);
+ file = BufFileOpenFileSet(&shared->sfs.fs, fname, O_RDONLY, false);
+ apply_concurrent_changes(file, dest);
+
+ BufFileClose(file);
+
+ /* Get ready for the next file. */
+ dest->file_seq++;
}
/*
@@ -3101,15 +3271,10 @@ cleanup_logical_decoding(LogicalDecodingContext *ctx)
dstate = (RepackDecodingState *) ctx->output_writer_private;
- ExecDropSingleTupleTableSlot(dstate->tsslot);
- FreeTupleDesc(dstate->tupdesc_change);
FreeTupleDesc(dstate->tupdesc);
- tuplestore_end(dstate->tstore);
-
FreeDecodingContext(ctx);
- ReplicationSlotRelease();
- ReplicationSlotDrop(NameStr(dstate->slotname), false);
+ ReplicationSlotDropAcquired();
pfree(dstate);
}
@@ -3123,7 +3288,6 @@ cleanup_logical_decoding(LogicalDecodingContext *ctx)
*/
static void
rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
- LogicalDecodingContext *decoding_ctx,
TransactionId frozenXid,
MultiXactId cutoffMulti)
{
@@ -3204,6 +3368,7 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
&chgdst.ident_index);
chgdst.ident_key = build_identity_key(ident_idx_new, OldHeap,
&chgdst.ident_key_nentries);
+ chgdst.file_seq = WORKER_FILE_SNAPSHOT + 1;
/*
* During testing, wait for another backend to perform concurrent data
@@ -3225,7 +3390,7 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
* hold AccessExclusiveLock. (Quite some amount of WAL could have been
* written during the data copying and index creation.)
*/
- process_concurrent_changes(decoding_ctx, end_of_wal, &chgdst);
+ process_concurrent_changes(end_of_wal, &chgdst, false);
/*
* Acquire AccessExclusiveLock on the table, its TOAST relation (if there
@@ -3306,8 +3471,11 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
XLogFlush(wal_insert_ptr);
end_of_wal = GetFlushRecPtr(NULL);
- /* Apply the concurrent changes again. */
- process_concurrent_changes(decoding_ctx, end_of_wal, &chgdst);
+ /*
+ * Apply the concurrent changes again. Indicate that the decoding worker
+ * won't be needed anymore.
+ */
+ process_concurrent_changes(end_of_wal, &chgdst, true);
/* Remember info about rel before closing OldHeap */
relpersistence = OldHeap->rd_rel->relpersistence;
@@ -3417,3 +3585,510 @@ build_new_indexes(Relation NewHeap, Relation OldHeap, List *OldIndexes)
return result;
}
+
+/*
+ * Try to start a background worker to perform logical decoding of data
+ * changes applied to relation while REPACK CONCURRENTLY is copying its
+ * contents to a new table.
+ */
+static void
+start_decoding_worker(Oid relid)
+{
+ Size size;
+ dsm_segment *seg;
+ DecodingWorkerShared *shared;
+ shm_mq *mq;
+ shm_mq_handle *mqh;
+ BackgroundWorker bgw;
+
+ /* Setup shared memory. */
+ size = BUFFERALIGN(offsetof(DecodingWorkerShared, error_queue)) +
+ BUFFERALIGN(REPACK_ERROR_QUEUE_SIZE);
+ seg = dsm_create(size, 0);
+ shared = (DecodingWorkerShared *) dsm_segment_address(seg);
+ shared->lsn_upto = InvalidXLogRecPtr;
+ shared->done = false;
+ SharedFileSetInit(&shared->sfs, seg);
+ shared->last_exported = -1;
+ SpinLockInit(&shared->mutex);
+ shared->dbid = MyDatabaseId;
+
+ /*
+ * This is the UserId set in cluster_rel(). Security context shouldn't be
+ * needed for decoding worker.
+ */
+ shared->roleid = GetUserId();
+ shared->relid = relid;
+ ConditionVariableInit(&shared->cv);
+ shared->backend_proc = MyProc;
+ shared->backend_pid = MyProcPid;
+ shared->backend_proc_number = MyProcNumber;
+
+ mq = shm_mq_create((char *) BUFFERALIGN(shared->error_queue),
+ REPACK_ERROR_QUEUE_SIZE);
+ shm_mq_set_receiver(mq, MyProc);
+ mqh = shm_mq_attach(mq, seg, NULL);
+
+ memset(&bgw, 0, sizeof(bgw));
+ snprintf(bgw.bgw_name, BGW_MAXLEN,
+ "REPACK decoding worker for relation \"%s\"",
+ get_rel_name(relid));
+ snprintf(bgw.bgw_type, BGW_MAXLEN, "REPACK decoding worker");
+ bgw.bgw_flags = BGWORKER_SHMEM_ACCESS |
+ BGWORKER_BACKEND_DATABASE_CONNECTION;
+ bgw.bgw_start_time = BgWorkerStart_RecoveryFinished;
+ bgw.bgw_restart_time = BGW_NEVER_RESTART;
+ snprintf(bgw.bgw_library_name, MAXPGPATH, "postgres");
+ snprintf(bgw.bgw_function_name, BGW_MAXLEN, "RepackWorkerMain");
+ bgw.bgw_main_arg = UInt32GetDatum(dsm_segment_handle(seg));
+ bgw.bgw_notify_pid = MyProcPid;
+
+ decoding_worker = palloc0_object(DecodingWorker);
+ if (!RegisterDynamicBackgroundWorker(&bgw, &decoding_worker->handle))
+ ereport(ERROR,
+ (errcode(ERRCODE_CONFIGURATION_LIMIT_EXCEEDED),
+ errmsg("out of background worker slots"),
+ errhint("You might need to increase \"%s\".", "max_worker_processes")));
+
+ decoding_worker->seg = seg;
+ decoding_worker->error_mqh = mqh;
+
+ /*
+ * The decoding setup must be done before the caller can have XID assigned
+ * for any reason, otherwise the worker might end up in a deadlock,
+ * waiting for the caller's transaction to end. Therefore wait here until
+ * the worker indicates that it has the logical decoding initialized.
+ */
+ ConditionVariablePrepareToSleep(&shared->cv);
+ for (;;)
+ {
+ bool initialized;
+
+ SpinLockAcquire(&shared->mutex);
+ initialized = shared->initialized;
+ SpinLockRelease(&shared->mutex);
+
+ if (initialized)
+ break;
+
+ ConditionVariableSleep(&shared->cv, WAIT_EVENT_REPACK_WORKER_EXPORT);
+ }
+ ConditionVariableCancelSleep();
+}
+
+/*
+ * Stop the decoding worker and cleanup the related resources.
+ *
+ * The worker stops on its own when it knows there is no more work to do, but
+ * we need to stop it explicitly at least on ERROR in the launching backend.
+ */
+static void
+stop_decoding_worker(void)
+{
+ BgwHandleStatus status;
+
+ /* Haven't reached the worker startup? */
+ if (decoding_worker == NULL)
+ return;
+
+ /* Could not register the worker? */
+ if (decoding_worker->handle == NULL)
+ return;
+
+ TerminateBackgroundWorker(decoding_worker->handle);
+ /* The worker should really exit before the REPACK command does. */
+ HOLD_INTERRUPTS();
+ status = WaitForBackgroundWorkerShutdown(decoding_worker->handle);
+ RESUME_INTERRUPTS();
+
+ if (status == BGWH_POSTMASTER_DIED)
+ ereport(FATAL,
+ (errcode(ERRCODE_ADMIN_SHUTDOWN),
+ errmsg("postmaster exited during REPACK command")));
+
+ shm_mq_detach(decoding_worker->error_mqh);
+
+ /*
+ * If we could not cancel the current sleep due to ERROR, do that before
+ * we detach from the shared memory the condition variable is located in.
+ * If we did not, the bgworker ERROR handling code would try and fail
+ * badly.
+ */
+ ConditionVariableCancelSleep();
+
+ dsm_detach(decoding_worker->seg);
+ pfree(decoding_worker);
+ decoding_worker = NULL;
+}
+
+/* Is this process a REPACK worker? */
+static bool is_repack_worker = false;
+
+static pid_t backend_pid;
+static ProcNumber backend_proc_number;
+
+/*
+ * See ParallelWorkerShutdown for details.
+ */
+static void
+RepackWorkerShutdown(int code, Datum arg)
+{
+ SendProcSignal(backend_pid,
+ PROCSIG_REPACK_MESSAGE,
+ backend_proc_number);
+
+ dsm_detach((dsm_segment *) DatumGetPointer(arg));
+}
+
+/* REPACK decoding worker entry point */
+void
+RepackWorkerMain(Datum main_arg)
+{
+ dsm_segment *seg;
+ DecodingWorkerShared *shared;
+ shm_mq *mq;
+ shm_mq_handle *mqh;
+
+ is_repack_worker = true;
+
+ /*
+ * Override the default bgworker_die() with die() so we can use
+ * CHECK_FOR_INTERRUPTS().
+ */
+ pqsignal(SIGTERM, die);
+ BackgroundWorkerUnblockSignals();
+
+ seg = dsm_attach(DatumGetUInt32(main_arg));
+ if (seg == NULL)
+ ereport(ERROR,
+ (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
+ errmsg("could not map dynamic shared memory segment")));
+
+ shared = (DecodingWorkerShared *) dsm_segment_address(seg);
+
+ /* Arrange to signal the leader if we exit. */
+ backend_pid = shared->backend_pid;
+ backend_proc_number = shared->backend_proc_number;
+ before_shmem_exit(RepackWorkerShutdown, PointerGetDatum(seg));
+
+ /*
+ * Join locking group - see the comments around the call of
+ * start_decoding_worker().
+ */
+ if (!BecomeLockGroupMember(shared->backend_proc, backend_pid))
+ /* The leader is not running anymore. */
+ return;
+
+ /*
+ * Setup a queue to send error messages to the backend that launched this
+ * worker.
+ */
+ mq = (shm_mq *) (char *) BUFFERALIGN(shared->error_queue);
+ shm_mq_set_sender(mq, MyProc);
+ mqh = shm_mq_attach(mq, seg, NULL);
+ pq_redirect_to_shm_mq(seg, mqh);
+ pq_set_parallel_leader(shared->backend_pid,
+ shared->backend_proc_number);
+
+ /* Connect to the database. */
+ BackgroundWorkerInitializeConnectionByOid(shared->dbid, shared->roleid, 0);
+
+ repack_worker_internal(seg);
+}
+
+static void
+repack_worker_internal(dsm_segment *seg)
+{
+ DecodingWorkerShared *shared;
+ LogicalDecodingContext *decoding_ctx;
+ SharedFileSet *sfs;
+ Snapshot snapshot;
+
+ /*
+ * Transaction is needed to open relation, and it also provides us with a
+ * resource owner.
+ */
+ StartTransactionCommand();
+
+ shared = (DecodingWorkerShared *) dsm_segment_address(seg);
+
+ /*
+ * Not sure the spinlock is needed here - the backend should not change
+ * anything in the shared memory until we have serialized the snapshot.
+ */
+ SpinLockAcquire(&shared->mutex);
+ Assert(XLogRecPtrIsInvalid(shared->lsn_upto));
+ sfs = &shared->sfs;
+ SpinLockRelease(&shared->mutex);
+
+ SharedFileSetAttach(sfs, seg);
+
+ /*
+ * Prepare to capture the concurrent data changes ourselves.
+ */
+ decoding_ctx = setup_logical_decoding(shared->relid);
+
+ /* Announce that we're ready. */
+ SpinLockAcquire(&shared->mutex);
+ shared->initialized = true;
+ SpinLockRelease(&shared->mutex);
+ ConditionVariableSignal(&shared->cv);
+
+ /* Build the initial snapshot and export it. */
+ snapshot = SnapBuildInitialSnapshotForRepack(decoding_ctx->snapshot_builder);
+ export_initial_snapshot(snapshot, shared);
+
+ /*
+ * Only historic snapshots should be used now. Do not let us restrict the
+ * progress of xmin horizon.
+ */
+ InvalidateCatalogSnapshot();
+
+ while (!decode_concurrent_changes(decoding_ctx, shared))
+ ;
+
+ /* Cleanup. */
+ cleanup_logical_decoding(decoding_ctx);
+ CommitTransactionCommand();
+}
+
+/*
+ * Make snapshot available to the backend that launched the decoding worker.
+ */
+static void
+export_initial_snapshot(Snapshot snapshot, DecodingWorkerShared *shared)
+{
+ char fname[MAXPGPATH];
+ BufFile *file;
+ Size snap_size;
+ char *snap_space;
+
+ snap_size = EstimateSnapshotSpace(snapshot);
+ snap_space = (char *) palloc(snap_size);
+ SerializeSnapshot(snapshot, snap_space);
+ FreeSnapshot(snapshot);
+
+ DecodingWorkerFileName(fname, shared->relid, shared->last_exported + 1);
+ file = BufFileCreateFileSet(&shared->sfs.fs, fname);
+ /* To make restoration easier, write the snapshot size first. */
+ BufFileWrite(file, &snap_size, sizeof(snap_size));
+ BufFileWrite(file, snap_space, snap_size);
+ pfree(snap_space);
+ BufFileClose(file);
+
+ /* Increase the counter to tell the backend that the file is available. */
+ SpinLockAcquire(&shared->mutex);
+ shared->last_exported++;
+ SpinLockRelease(&shared->mutex);
+ ConditionVariableSignal(&shared->cv);
+}
+
+/*
+ * Get the initial snapshot from the decoding worker.
+ */
+static Snapshot
+get_initial_snapshot(DecodingWorker *worker)
+{
+ DecodingWorkerShared *shared;
+ char fname[MAXPGPATH];
+ BufFile *file;
+ Size snap_size;
+ char *snap_space;
+ Snapshot snapshot;
+
+ shared = (DecodingWorkerShared *) dsm_segment_address(worker->seg);
+
+ /*
+ * The worker needs to initialize the logical decoding, which usually
+ * takes some time. Therefore it makes sense to prepare for the sleep
+ * first.
+ */
+ ConditionVariablePrepareToSleep(&shared->cv);
+ for (;;)
+ {
+ int last_exported;
+
+ SpinLockAcquire(&shared->mutex);
+ last_exported = shared->last_exported;
+ SpinLockRelease(&shared->mutex);
+
+ /*
+ * Has the worker exported the file we are waiting for?
+ */
+ if (last_exported == WORKER_FILE_SNAPSHOT)
+ break;
+
+ ConditionVariableSleep(&shared->cv, WAIT_EVENT_REPACK_WORKER_EXPORT);
+ }
+ ConditionVariableCancelSleep();
+
+ /* Read the snapshot from a file. */
+ DecodingWorkerFileName(fname, shared->relid, WORKER_FILE_SNAPSHOT);
+ file = BufFileOpenFileSet(&shared->sfs.fs, fname, O_RDONLY, false);
+ BufFileReadExact(file, &snap_size, sizeof(snap_size));
+ snap_space = (char *) palloc(snap_size);
+ BufFileReadExact(file, snap_space, snap_size);
+ BufFileClose(file);
+
+ /* Restore it. */
+ snapshot = RestoreSnapshot(snap_space);
+ pfree(snap_space);
+
+ return snapshot;
+}
+
+bool
+IsRepackWorker(void)
+{
+ return is_repack_worker;
+}
+
+/*
+ * Handle receipt of an interrupt indicating a repack worker message.
+ *
+ * Note: this is called within a signal handler! All we can do is set
+ * a flag that will cause the next CHECK_FOR_INTERRUPTS() to invoke
+ * ProcessRepackMessages().
+ */
+void
+HandleRepackMessageInterrupt(void)
+{
+ InterruptPending = true;
+ RepackMessagePending = true;
+ SetLatch(MyLatch);
+}
+
+/*
+ * Process any queued protocol messages received from parallel workers.
+ */
+void
+ProcessRepackMessages(void)
+{
+ MemoryContext oldcontext;
+
+ static MemoryContext hpm_context = NULL;
+
+ /*
+ * Nothing to do if we haven't launched the worker yet or have already
+ * terminated it.
+ */
+ if (decoding_worker == NULL)
+ return;
+
+ /*
+ * This is invoked from ProcessInterrupts(), and since some of the
+ * functions it calls contain CHECK_FOR_INTERRUPTS(), there is a potential
+ * for recursive calls if more signals are received while this runs. It's
+ * unclear that recursive entry would be safe, and it doesn't seem useful
+ * even if it is safe, so let's block interrupts until done.
+ */
+ HOLD_INTERRUPTS();
+
+ /*
+ * Moreover, CurrentMemoryContext might be pointing almost anywhere. We
+ * don't want to risk leaking data into long-lived contexts, so let's do
+ * our work here in a private context that we can reset on each use.
+ */
+ if (hpm_context == NULL) /* first time through? */
+ hpm_context = AllocSetContextCreate(TopMemoryContext,
+ "ProcessRepackMessages",
+ ALLOCSET_DEFAULT_SIZES);
+ else
+ MemoryContextReset(hpm_context);
+
+ oldcontext = MemoryContextSwitchTo(hpm_context);
+
+ /* OK to process messages. Reset the flag saying there are more to do. */
+ RepackMessagePending = false;
+
+ /*
+ * Read as many messages as we can from each worker, but stop when no more
+ * messages can be read from the worker without blocking.
+ */
+ while (true)
+ {
+ shm_mq_result res;
+ Size nbytes;
+ void *data;
+
+ res = shm_mq_receive(decoding_worker->error_mqh, &nbytes,
+ &data, true);
+ if (res == SHM_MQ_WOULD_BLOCK)
+ break;
+ else if (res == SHM_MQ_SUCCESS)
+ {
+ StringInfoData msg;
+
+ initStringInfo(&msg);
+ appendBinaryStringInfo(&msg, data, nbytes);
+ ProcessRepackMessage(&msg);
+ pfree(msg.data);
+ }
+ else
+ {
+ /*
+ * The decoding worker is special in that it exits as soon as it
+ * has its work done. Thus the DETACHED result code is fine.
+ */
+ Assert(res == SHM_MQ_DETACHED);
+
+ break;
+ }
+ }
+
+ MemoryContextSwitchTo(oldcontext);
+
+ /* Might as well clear the context on our way out */
+ MemoryContextReset(hpm_context);
+
+ RESUME_INTERRUPTS();
+}
+
+/*
+ * Process a single protocol message received from a single parallel worker.
+ */
+static void
+ProcessRepackMessage(StringInfo msg)
+{
+ char msgtype;
+
+ msgtype = pq_getmsgbyte(msg);
+
+ switch (msgtype)
+ {
+ case PqMsg_ErrorResponse:
+ case PqMsg_NoticeResponse:
+ {
+ ErrorData edata;
+
+ /* Parse ErrorResponse or NoticeResponse. */
+ pq_parse_errornotice(msg, &edata);
+
+ /* Death of a worker isn't enough justification for suicide. */
+ edata.elevel = Min(edata.elevel, ERROR);
+
+ /*
+ * If desired, add a context line to show that this is a
+ * message propagated from a parallel worker. Otherwise, it
+ * can sometimes be confusing to understand what actually
+ * happened.
+ */
+ if (edata.context)
+ edata.context = psprintf("%s\n%s", edata.context,
+ _("decoding worker"));
+ else
+ edata.context = pstrdup(_("decoding worker"));
+
+ /* Rethrow error or print notice. */
+ ThrowErrorData(&edata);
+
+ break;
+ }
+
+ default:
+ {
+ elog(ERROR, "unrecognized message type received from decoding worker: %c (message length %d bytes)",
+ msgtype, msg->len);
+ }
+ }
+}
diff --git a/src/backend/libpq/pqmq.c b/src/backend/libpq/pqmq.c
index 6e4bbfb5aa1..42f6fa472c5 100644
--- a/src/backend/libpq/pqmq.c
+++ b/src/backend/libpq/pqmq.c
@@ -14,6 +14,7 @@
#include "postgres.h"
#include "access/parallel.h"
+#include "commands/cluster.h"
#include "libpq/libpq.h"
#include "libpq/pqformat.h"
#include "libpq/pqmq.h"
@@ -175,6 +176,10 @@ mq_putmessage(char msgtype, const char *s, size_t len)
SendProcSignal(pq_mq_parallel_leader_pid,
PROCSIG_PARALLEL_APPLY_MESSAGE,
pq_mq_parallel_leader_proc_number);
+ else if (IsRepackWorker())
+ SendProcSignal(pq_mq_parallel_leader_pid,
+ PROCSIG_REPACK_MESSAGE,
+ pq_mq_parallel_leader_proc_number);
else
{
Assert(IsParallelWorker());
diff --git a/src/backend/postmaster/bgworker.c b/src/backend/postmaster/bgworker.c
index 65deabe91a7..334bb708c5b 100644
--- a/src/backend/postmaster/bgworker.c
+++ b/src/backend/postmaster/bgworker.c
@@ -13,6 +13,7 @@
#include "postgres.h"
#include "access/parallel.h"
+#include "commands/cluster.h"
#include "libpq/pqsignal.h"
#include "miscadmin.h"
#include "pgstat.h"
@@ -136,6 +137,9 @@ static const struct
},
{
"SequenceSyncWorkerMain", SequenceSyncWorkerMain
+ },
+ {
+ "RepackWorkerMain", RepackWorkerMain
}
};
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index b0ef1a12520..35a46988285 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -194,7 +194,11 @@ StartupDecodingContext(List *output_plugin_options,
ctx->slot = slot;
- ctx->reader = XLogReaderAllocate(wal_segment_size, NULL, xl_routine, ctx);
+ /*
+ * TODO A separate patch for PG core, unless there's really a reason to
+ * pass ctx for private_data (May extensions expect ctx?).
+ */
+ ctx->reader = XLogReaderAllocate(wal_segment_size, NULL, xl_routine, NULL);
if (!ctx->reader)
ereport(ERROR,
(errcode(ERRCODE_OUT_OF_MEMORY),
diff --git a/src/backend/replication/pgoutput_repack/pgoutput_repack.c b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
index a074eed393d..4bc47a72371 100644
--- a/src/backend/replication/pgoutput_repack/pgoutput_repack.c
+++ b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
@@ -169,17 +169,13 @@ store_change(LogicalDecodingContext *ctx, ConcurrentChangeKind kind,
HeapTuple tuple)
{
RepackDecodingState *dstate;
- char *change_raw;
- ConcurrentChange change;
+ char kind_byte = (char) kind;
bool flattened = false;
- Size size;
- Datum values[1];
- bool isnull[1];
- char *dst;
dstate = (RepackDecodingState *) ctx->output_writer_private;
- size = VARHDRSZ + SizeOfConcurrentChange;
+ /* Store the change kind. */
+ BufFileWrite(dstate->file, &kind_byte, 1);
/*
* ReorderBufferCommit() stores the TOAST chunks in its private memory
@@ -196,46 +192,12 @@ store_change(LogicalDecodingContext *ctx, ConcurrentChangeKind kind,
tuple = toast_flatten_tuple(tuple, dstate->tupdesc);
flattened = true;
}
+ /* Store the tuple size ... */
+ BufFileWrite(dstate->file, &tuple->t_len, sizeof(tuple->t_len));
+ /* ... and the tuple itself. */
+ BufFileWrite(dstate->file, tuple->t_data, tuple->t_len);
- size += tuple->t_len;
- if (size >= MaxAllocSize)
- elog(ERROR, "Change is too big.");
-
- /* Construct the change. */
- change_raw = (char *) palloc0(size);
- SET_VARSIZE(change_raw, size);
-
- /*
- * Since the varlena alignment might not be sufficient for the structure,
- * set the fields in a local instance and remember where it should
- * eventually be copied.
- */
- change.kind = kind;
- dst = (char *) VARDATA(change_raw);
-
- /*
- * Copy the tuple.
- *
- * Note: change->tup_data.t_data must be fixed on retrieval!
- */
- memcpy(&change.tup_data, tuple, sizeof(HeapTupleData));
- memcpy(dst, &change, SizeOfConcurrentChange);
- dst += SizeOfConcurrentChange;
- memcpy(dst, tuple->t_data, tuple->t_len);
-
- /* The data has been copied. */
+ /* Free the flat copy if created above. */
if (flattened)
pfree(tuple);
-
- /* Store as tuple of 1 bytea column. */
- values[0] = PointerGetDatum(change_raw);
- isnull[0] = false;
- tuplestore_putvalues(dstate->tstore, dstate->tupdesc_change,
- values, isnull);
-
- /* Accounting. */
- dstate->nchanges++;
-
- /* Cleanup. */
- pfree(change_raw);
}
diff --git a/src/backend/storage/ipc/procsignal.c b/src/backend/storage/ipc/procsignal.c
index 8e56922dcea..6f9e7a7aab7 100644
--- a/src/backend/storage/ipc/procsignal.c
+++ b/src/backend/storage/ipc/procsignal.c
@@ -19,6 +19,7 @@
#include "access/parallel.h"
#include "commands/async.h"
+#include "commands/cluster.h"
#include "miscadmin.h"
#include "pgstat.h"
#include "port/pg_bitutils.h"
@@ -697,6 +698,9 @@ procsignal_sigusr1_handler(SIGNAL_ARGS)
if (CheckProcSignal(PROCSIG_PARALLEL_APPLY_MESSAGE))
HandleParallelApplyMessageInterrupt();
+ if (CheckProcSignal(PROCSIG_REPACK_MESSAGE))
+ HandleRepackMessageInterrupt();
+
if (CheckProcSignal(PROCSIG_RECOVERY_CONFLICT_DATABASE))
HandleRecoveryConflictInterrupt(PROCSIG_RECOVERY_CONFLICT_DATABASE);
diff --git a/src/backend/tcop/postgres.c b/src/backend/tcop/postgres.c
index e54bf1e760f..fc81ad87615 100644
--- a/src/backend/tcop/postgres.c
+++ b/src/backend/tcop/postgres.c
@@ -36,6 +36,7 @@
#include "access/xact.h"
#include "catalog/pg_type.h"
#include "commands/async.h"
+#include "commands/cluster.h"
#include "commands/event_trigger.h"
#include "commands/explain_state.h"
#include "commands/prepare.h"
@@ -3541,6 +3542,9 @@ ProcessInterrupts(void)
if (ParallelApplyMessagePending)
ProcessParallelApplyMessages();
+
+ if (RepackMessagePending)
+ ProcessRepackMessages();
}
/*
diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt
index 5537a2d2530..56d01ccd5af 100644
--- a/src/backend/utils/activity/wait_event_names.txt
+++ b/src/backend/utils/activity/wait_event_names.txt
@@ -154,6 +154,7 @@ RECOVERY_CONFLICT_SNAPSHOT "Waiting for recovery conflict resolution for a vacuu
RECOVERY_CONFLICT_TABLESPACE "Waiting for recovery conflict resolution for dropping a tablespace."
RECOVERY_END_COMMAND "Waiting for <xref linkend="guc-recovery-end-command"/> to complete."
RECOVERY_PAUSE "Waiting for recovery to be resumed."
+REPACK_WORKER_EXPORT "Waiting for decoding worker to export a new output file."
REPLICATION_ORIGIN_DROP "Waiting for a replication origin to become inactive so it can be dropped."
REPLICATION_SLOT_DROP "Waiting for a replication slot to become inactive so it can be dropped."
RESTORE_COMMAND "Waiting for <xref linkend="guc-restore-command"/> to complete."
diff --git a/src/include/access/tableam.h b/src/include/access/tableam.h
index 76aa993009a..15760363a1a 100644
--- a/src/include/access/tableam.h
+++ b/src/include/access/tableam.h
@@ -22,7 +22,6 @@
#include "access/xact.h"
#include "commands/vacuum.h"
#include "executor/tuptable.h"
-#include "replication/logical.h"
#include "storage/read_stream.h"
#include "utils/rel.h"
#include "utils/snapshot.h"
@@ -631,7 +630,6 @@ typedef struct TableAmRoutine
bool use_sort,
TransactionId OldestXmin,
Snapshot snapshot,
- LogicalDecodingContext *decoding_ctx,
TransactionId *xid_cutoff,
MultiXactId *multi_cutoff,
double *num_tuples,
@@ -1651,8 +1649,6 @@ table_relation_copy_data(Relation rel, const RelFileLocator *newrlocator)
* - *multi_cutoff - ditto
* - snapshot - if != NULL, ignore data changes done by transactions that this
* (MVCC) snapshot considers still in-progress or in the future.
- * - decoding_ctx - logical decoding context, to capture concurrent data
- * changes.
*
* Output parameters:
* - *xid_cutoff - rel's new relfrozenxid value, may be invalid
@@ -1666,7 +1662,6 @@ table_relation_copy_for_cluster(Relation OldTable, Relation NewTable,
bool use_sort,
TransactionId OldestXmin,
Snapshot snapshot,
- LogicalDecodingContext *decoding_ctx,
TransactionId *xid_cutoff,
MultiXactId *multi_cutoff,
double *num_tuples,
@@ -1675,7 +1670,7 @@ table_relation_copy_for_cluster(Relation OldTable, Relation NewTable,
{
OldTable->rd_tableam->relation_copy_for_cluster(OldTable, NewTable, OldIndex,
use_sort, OldestXmin,
- snapshot, decoding_ctx,
+ snapshot,
xid_cutoff, multi_cutoff,
num_tuples, tups_vacuumed,
tups_recently_dead);
diff --git a/src/include/commands/cluster.h b/src/include/commands/cluster.h
index 6a5c476294a..1b05d5d418b 100644
--- a/src/include/commands/cluster.h
+++ b/src/include/commands/cluster.h
@@ -17,11 +17,13 @@
#include "nodes/parsenodes.h"
#include "parser/parse_node.h"
#include "replication/decode.h"
+#include "postmaster/bgworker.h"
#include "replication/logical.h"
+#include "storage/buffile.h"
#include "storage/lock.h"
+#include "storage/shm_mq.h"
#include "utils/relcache.h"
#include "utils/resowner.h"
-#include "utils/tuplestore.h"
/* flag bits for ClusterParams->options */
@@ -44,6 +46,9 @@ typedef struct ClusterParams
* The following definitions are used by REPACK CONCURRENTLY.
*/
+/*
+ * Stored as a single byte in the output file.
+ */
typedef enum
{
CHANGE_INSERT,
@@ -52,68 +57,30 @@ typedef enum
CHANGE_DELETE
} ConcurrentChangeKind;
-typedef struct ConcurrentChange
-{
- /* See the enum above. */
- ConcurrentChangeKind kind;
-
- /*
- * The actual tuple.
- *
- * The tuple data follows the ConcurrentChange structure. Before use make
- * sure the tuple is correctly aligned (ConcurrentChange can be stored as
- * bytea) and that tuple->t_data is fixed.
- */
- HeapTupleData tup_data;
-} ConcurrentChange;
-
-#define SizeOfConcurrentChange (offsetof(ConcurrentChange, tup_data) + \
- sizeof(HeapTupleData))
-
/*
* Logical decoding state.
*
- * Here we store the data changes that we decode from WAL while the table
- * contents is being copied to a new storage. Also the necessary metadata
- * needed to apply these changes to the table is stored here.
+ * The output plugin uses it to store the data changes that it decodes from
+ * WAL while the table contents is being copied to a new storage.
*/
typedef struct RepackDecodingState
{
/* The relation whose changes we're decoding. */
Oid relid;
- /* Replication slot name. */
- NameData slotname;
-
- /*
- * Decoded changes are stored here. Although we try to avoid excessive
- * batches, it can happen that the changes need to be stored to disk. The
- * tuplestore does this transparently.
- */
- Tuplestorestate *tstore;
-
- /* The current number of changes in tstore. */
- double nchanges;
-
- /*
- * Descriptor to store the ConcurrentChange structure serialized (bytea).
- * We can't store the tuple directly because tuplestore only supports
- * minimum tuple and we may need to transfer OID system column from the
- * output plugin. Also we need to transfer the change kind, so it's better
- * to put everything in the structure than to use 2 tuplestores "in
- * parallel".
- */
- TupleDesc tupdesc_change;
-
- /* Tuple descriptor needed to update indexes. */
+ /* Tuple descriptor of the relation being processed. */
TupleDesc tupdesc;
- /* Slot to retrieve data from tstore. */
- TupleTableSlot *tsslot;
-
- ResourceOwner resowner;
+ /* The current output file. */
+ BufFile *file;
} RepackDecodingState;
+extern PGDLLIMPORT volatile sig_atomic_t RepackMessagePending;
+
+extern bool IsRepackWorker(void);
+extern void HandleRepackMessageInterrupt(void);
+extern void ProcessRepackMessages(void);
+
extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel);
extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid,
@@ -136,6 +103,6 @@ extern void finish_heap_swap(Oid OIDOldHeap, Oid OIDNewHeap,
extern bool am_decoding_for_repack(void);
extern bool change_useless_for_repack(XLogRecordBuffer *buf);
-extern void repack_decode_concurrent_changes(LogicalDecodingContext *ctx,
- XLogRecPtr end_of_wal);
+
+extern void RepackWorkerMain(Datum main_arg);
#endif /* CLUSTER_H */
diff --git a/src/include/storage/procsignal.h b/src/include/storage/procsignal.h
index e52b8eb7697..3ef35ca6b80 100644
--- a/src/include/storage/procsignal.h
+++ b/src/include/storage/procsignal.h
@@ -36,6 +36,7 @@ typedef enum
PROCSIG_BARRIER, /* global barrier interrupt */
PROCSIG_LOG_MEMORY_CONTEXT, /* ask backend to log the memory contexts */
PROCSIG_PARALLEL_APPLY_MESSAGE, /* Message from parallel apply workers */
+ PROCSIG_REPACK_MESSAGE, /* Message from repack worker */
/* Recovery conflict reasons */
PROCSIG_RECOVERY_CONFLICT_FIRST,
diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list
index 556128e9c57..3cae074c7e7 100644
--- a/src/tools/pgindent/typedefs.list
+++ b/src/tools/pgindent/typedefs.list
@@ -498,7 +498,6 @@ CompressFileHandle
CompressionLocation
CompressorState
ComputeXidHorizonsResult
-ConcurrentChange
ConcurrentChangeKind
ConditionVariable
ConditionVariableMinimallyPadded
@@ -638,6 +637,8 @@ DeclareCursorStmt
DecodedBkpBlock
DecodedXLogRecord
DecodingOutputState
+DecodingWorker
+DecodingWorkerShared
DefElem
DefElemAction
DefaultACLInfo
--
2.47.3
--=-=-=
Content-Type: text/plain
Content-Disposition: attachment;
filename=v31-0006-Use-multiple-snapshots-to-copy-the-data.patch
^ permalink raw reply [nested|flat] 55+ messages in thread
* [PATCH 5/6] Use background worker to do logical decoding.
@ 2026-01-22 08:23 Antonin Houska <[email protected]>
0 siblings, 0 replies; 55+ messages in thread
From: Antonin Houska @ 2026-01-22 08:23 UTC (permalink / raw)
If the backend performing REPACK (CONCURRENTLY) does both data copying and
logical decoding, it has to "travel in time" back and forth and therefore it
has to invalidate system caches quite a few times. (The copying and the
decoding work with different catalog snapshots.) As the decoding worker has
separate caches, the switching is not necessary.
Without the worker, it'd also be difficult to switch between potentially long
running tasks like index build and WAL decoding. (No decoding during that time
at all can suspend archiving / recycling of WAL segments for some time, which
in turn may result in full disk.)
Another problem is that, after having acquired AccessExclusiveLock (in order
to swap the files), the backend needs to both decode and apply the data
changes that took place while it was waiting for the lock. With the decoding
worker, the decoding runs all the time, so the backend only needs to apply the
changes. This can reduce the time the exclusive lock is held for.
Note that the code added in order to handle ERRORs in the background worker
almost duplicates the existing code that does the same for other types of
workers (See ProcessParallelMessages() and
ProcessParallelApplyMessages()). Refactoring of the existing code might be
useful, to reduce the duplication.
---
src/backend/access/heap/heapam_handler.c | 44 -
src/backend/commands/cluster.c | 1179 +++++++++++++----
src/backend/libpq/pqmq.c | 5 +
src/backend/postmaster/bgworker.c | 4 +
src/backend/replication/logical/logical.c | 6 +-
.../pgoutput_repack/pgoutput_repack.c | 54 +-
src/backend/storage/ipc/procsignal.c | 4 +
src/backend/tcop/postgres.c | 4 +
.../utils/activity/wait_event_names.txt | 1 +
src/include/access/tableam.h | 7 +-
src/include/commands/cluster.h | 71 +-
src/include/storage/procsignal.h | 1 +
src/tools/pgindent/typedefs.list | 3 +-
13 files changed, 981 insertions(+), 402 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index 908f1ef66c6..8589b3c940e 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -33,7 +33,6 @@
#include "catalog/index.h"
#include "catalog/storage.h"
#include "catalog/storage_xlog.h"
-#include "commands/cluster.h"
#include "commands/progress.h"
#include "executor/executor.h"
#include "miscadmin.h"
@@ -688,7 +687,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
Relation OldIndex, bool use_sort,
TransactionId OldestXmin,
Snapshot snapshot,
- LogicalDecodingContext *decoding_ctx,
TransactionId *xid_cutoff,
MultiXactId *multi_cutoff,
double *num_tuples,
@@ -710,7 +708,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
BufferHeapTupleTableSlot *hslot;
BlockNumber prev_cblock = InvalidBlockNumber;
bool concurrent = snapshot != NULL;
- XLogRecPtr end_of_wal_prev = GetFlushRecPtr(NULL);
/* Remember if it's a system catalog */
is_system_catalog = IsSystemRelation(OldHeap);
@@ -971,31 +968,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
ct_val[1] = *num_tuples;
pgstat_progress_update_multi_param(2, ct_index, ct_val);
}
-
- /*
- * Process the WAL produced by the load, as well as by other
- * transactions, so that the replication slot can advance and WAL does
- * not pile up. Use wal_segment_size as a threshold so that we do not
- * introduce the decoding overhead too often.
- *
- * Of course, we must not apply the changes until the initial load has
- * completed.
- *
- * Note that our insertions into the new table should not be decoded
- * as we (intentionally) do not write the logical decoding specific
- * information to WAL.
- */
- if (concurrent)
- {
- XLogRecPtr end_of_wal;
-
- end_of_wal = GetFlushRecPtr(NULL);
- if ((end_of_wal - end_of_wal_prev) > wal_segment_size)
- {
- repack_decode_concurrent_changes(decoding_ctx, end_of_wal);
- end_of_wal_prev = end_of_wal;
- }
- }
}
if (indexScan != NULL)
@@ -1041,22 +1013,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
-
- /*
- * Try to keep the amount of not-yet-decoded WAL small, like
- * above.
- */
- if (concurrent)
- {
- XLogRecPtr end_of_wal;
-
- end_of_wal = GetFlushRecPtr(NULL);
- if ((end_of_wal - end_of_wal_prev) > wal_segment_size)
- {
- repack_decode_concurrent_changes(decoding_ctx, end_of_wal);
- end_of_wal_prev = end_of_wal;
- }
- }
}
tuplesort_end(tuplesort);
diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c
index 03ccf10b782..e988a7a7296 100644
--- a/src/backend/commands/cluster.c
+++ b/src/backend/commands/cluster.c
@@ -12,12 +12,13 @@
* In concurrent mode, we lock the table with only ShareUpdateExclusiveLock,
* then do an initial copy as above. However, while the tuples are being
* copied, concurrent transactions could modify the table. To cope with those
- * changes, we rely on logical decoding to obtain them from WAL. The changes
- * are accumulated in a tuplestore. Once the initial copy is complete, we
- * read the changes from the tuplestore and re-apply them on the new heap.
- * Then we upgrade our ShareUpdateExclusiveLock to AccessExclusiveLock and
- * swap the relfilenodes. This way, the time we hold a strong lock on the
- * table is much reduced, and the bloat is eliminated.
+ * changes, we rely on logical decoding to obtain them from WAL. A bgworker
+ * consumes WAL while the initial copy is ongoing (to prevent excessive WAL
+ * from being reserved), and accumulates the changes in a file. Once the
+ * initial copy is complete, we read the changes from the file and re-apply
+ * them on the new heap. Then we upgrade our ShareUpdateExclusiveLock to
+ * AccessExclusiveLock and swap the relfilenodes. This way, the time we hold
+ * a strong lock on the table is much reduced, and the bloat is eliminated.
*
* There is hardly anything left of Paul Brown's original implementation...
*
@@ -45,6 +46,7 @@
#include "access/xlog_internal.h"
#include "access/xloginsert.h"
#include "access/xlogutils.h"
+#include "access/xlogwait.h"
#include "catalog/catalog.h"
#include "catalog/dependency.h"
#include "catalog/heap.h"
@@ -61,6 +63,8 @@
#include "commands/tablecmds.h"
#include "commands/vacuum.h"
#include "executor/executor.h"
+#include "libpq/pqformat.h"
+#include "libpq/pqmq.h"
#include "miscadmin.h"
#include "optimizer/optimizer.h"
#include "pgstat.h"
@@ -71,6 +75,8 @@
#include "storage/ipc.h"
#include "storage/lmgr.h"
#include "storage/predicate.h"
+#include "storage/procsignal.h"
+#include "tcop/tcopprot.h"
#include "utils/acl.h"
#include "utils/fmgroids.h"
#include "utils/guc.h"
@@ -117,6 +123,12 @@ typedef struct IndexInsertState
/* The WAL segment being decoded. */
static XLogSegNo repack_current_segment = 0;
+/*
+ * The first file exported by the decoding worker must contain a snapshot, the
+ * following ones contain the data changes.
+ */
+#define WORKER_FILE_SNAPSHOT 0
+
/*
* Information needed to apply concurrent data changes.
*/
@@ -136,8 +148,113 @@ typedef struct ChangeDest
/* Needed to update indexes of rel_dst. */
IndexInsertState *iistate;
+
+ /*
+ * Sequential number of the file containing the changes.
+ *
+ * TODO This field makes the structure name less descriptive. Should we
+ * rename it, e.g. to ChangeApplyInfo?
+ */
+ int file_seq;
} ChangeDest;
+/*
+ * Layout of shared memory used for communication between backend and the
+ * worker that performs logical decoding of data changes
+ */
+typedef struct DecodingWorkerShared
+{
+ /* Is the decoding initialized? */
+ bool initialized;
+
+ /*
+ * Once the worker has reached this LSN, it should close the current
+ * output file and either create a new one or exit, according to the field
+ * 'done'. If the value is InvalidXLogRecPtr, the worker should decode all
+ * the WAL available and keep checking this field. It is ok if the worker
+ * had already decoded records whose LSN is >= lsn_upto before this field
+ * has been set.
+ */
+ XLogRecPtr lsn_upto;
+
+ /* Exit after closing the current file? */
+ bool done;
+
+ /* The output is stored here. */
+ SharedFileSet sfs;
+
+ /* Number of the last file exported by the worker. */
+ int last_exported;
+
+ /* Synchronize access to the fields above. */
+ slock_t mutex;
+
+ /* Database to connect to. */
+ Oid dbid;
+
+ /* Role to connect as. */
+ Oid roleid;
+
+ /* Decode data changes of this relation. */
+ Oid relid;
+
+ /* The backend uses this to wait for the worker. */
+ ConditionVariable cv;
+
+ /* Info to signal the backend. */
+ PGPROC *backend_proc;
+ pid_t backend_pid;
+ ProcNumber backend_proc_number;
+
+ /* Error queue. */
+ shm_mq *error_mq;
+
+ /*
+ * Memory the queue is located in.
+ *
+ * For considerations on the value see the comments of
+ * PARALLEL_ERROR_QUEUE_SIZE.
+ */
+#define REPACK_ERROR_QUEUE_SIZE 16384
+ char error_queue[FLEXIBLE_ARRAY_MEMBER];
+} DecodingWorkerShared;
+
+/*
+ * Generate worker's output file name. If relations of the same 'relid' happen
+ * to be processed at the same time, they must be from different databases and
+ * therefore different backends must be involved. (PID is already present in
+ * the fileset name.)
+ */
+static inline void
+DecodingWorkerFileName(char *fname, Oid relid, uint32 seq)
+{
+ snprintf(fname, MAXPGPATH, "%u-%u", relid, seq);
+}
+
+/*
+ * Backend-local information to control the decoding worker.
+ */
+typedef struct DecodingWorker
+{
+ /* The worker. */
+ BackgroundWorkerHandle *handle;
+
+ /* DecodingWorkerShared is in this segment. */
+ dsm_segment *seg;
+
+ /* Handle of the error queue. */
+ shm_mq_handle *error_mqh;
+} DecodingWorker;
+
+/* Pointer to currently running decoding worker. */
+static DecodingWorker *decoding_worker = NULL;
+
+/*
+ * Is there a message sent by a repack worker that the backend needs to
+ * receive?
+ */
+volatile sig_atomic_t RepackMessagePending = false;
+
static bool cluster_rel_recheck(RepackCommand cmd, Relation OldHeap,
Oid indexOid, Oid userid, LOCKMODE lmode,
int options);
@@ -145,7 +262,7 @@ static void check_repack_concurrently_requirements(Relation rel);
static void rebuild_relation(Relation OldHeap, Relation index, bool verbose,
bool concurrent);
static void copy_table_data(Relation NewHeap, Relation OldHeap, Relation OldIndex,
- Snapshot snapshot, LogicalDecodingContext *decoding_ctx,
+ Snapshot snapshot,
bool verbose,
bool *pSwapToastByContent,
TransactionId *pFreezeXid,
@@ -158,12 +275,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd,
static bool cluster_is_permitted_for_relation(RepackCommand cmd,
Oid relid, Oid userid);
-static void begin_concurrent_repack(Relation rel);
-static void end_concurrent_repack(void);
static LogicalDecodingContext *setup_logical_decoding(Oid relid);
-static HeapTuple get_changed_tuple(char *change);
-static void apply_concurrent_changes(RepackDecodingState *dstate,
- ChangeDest *dest);
+static bool decode_concurrent_changes(LogicalDecodingContext *ctx,
+ DecodingWorkerShared *shared);
+static void apply_concurrent_changes(BufFile *file, ChangeDest *dest);
static void apply_concurrent_insert(Relation rel, HeapTuple tup,
IndexInsertState *iistate,
TupleTableSlot *index_slot);
@@ -175,9 +290,9 @@ static void apply_concurrent_delete(Relation rel, HeapTuple tup_target);
static HeapTuple find_target_tuple(Relation rel, ChangeDest *dest,
HeapTuple tup_key,
TupleTableSlot *ident_slot);
-static void process_concurrent_changes(LogicalDecodingContext *decoding_ctx,
- XLogRecPtr end_of_wal,
- ChangeDest *dest);
+static void process_concurrent_changes(XLogRecPtr end_of_wal,
+ ChangeDest *dest,
+ bool done);
static IndexInsertState *get_index_insert_state(Relation relation,
Oid ident_index_id,
Relation *ident_index_p);
@@ -186,7 +301,6 @@ static ScanKey build_identity_key(Oid ident_idx_oid, Relation rel_src,
static void free_index_insert_state(IndexInsertState *iistate);
static void cleanup_logical_decoding(LogicalDecodingContext *ctx);
static void rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
- LogicalDecodingContext *decoding_ctx,
TransactionId frozenXid,
MultiXactId cutoffMulti);
static List *build_new_indexes(Relation NewHeap, Relation OldHeap, List *OldIndexes);
@@ -196,6 +310,13 @@ static Relation process_single_relation(RepackStmt *stmt,
ClusterParams *params);
static Oid determine_clustered_index(Relation rel, bool usingindex,
const char *indexname);
+static void start_decoding_worker(Oid relid);
+static void stop_decoding_worker(void);
+static void repack_worker_internal(dsm_segment *seg);
+static void export_initial_snapshot(Snapshot snapshot,
+ DecodingWorkerShared *shared);
+static Snapshot get_initial_snapshot(DecodingWorker *worker);
+static void ProcessRepackMessage(StringInfo msg);
static const char *RepackCommandAsString(RepackCommand cmd);
@@ -619,20 +740,20 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid,
/* rebuild_relation does all the dirty work */
PG_TRY();
{
- /*
- * For concurrent processing, make sure that our logical decoding
- * ignores data changes of other tables than the one we are
- * processing.
- */
- if (concurrent)
- begin_concurrent_repack(OldHeap);
-
rebuild_relation(OldHeap, index, verbose, concurrent);
}
PG_FINALLY();
{
if (concurrent)
- end_concurrent_repack();
+ {
+ /*
+ * Since during normal operation the worker was already asked to
+ * exit, stopping it explicitly is especially important on ERROR.
+ * However it still seems a good practice to make sure that the
+ * worker never survives the REPACK command.
+ */
+ stop_decoding_worker();
+ }
}
PG_END_TRY();
@@ -929,7 +1050,6 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
bool swap_toast_by_content;
TransactionId frozenXid;
MultiXactId cutoffMulti;
- LogicalDecodingContext *decoding_ctx = NULL;
Snapshot snapshot = NULL;
#if USE_ASSERT_CHECKING
LOCKMODE lmode;
@@ -943,19 +1063,36 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
if (concurrent)
{
/*
- * Prepare to capture the concurrent data changes.
+ * The worker needs to be member of the locking group we're the leader
+ * of. We ought to become the leader before the worker starts. The
+ * worker will join the group as soon as it starts.
*
- * Note that this call waits for all transactions with XID already
- * assigned to finish. If some of those transactions is waiting for a
- * lock conflicting with ShareUpdateExclusiveLock on our table (e.g.
- * it runs CREATE INDEX), we can end up in a deadlock. Not sure this
- * risk is worth unlocking/locking the table (and its clustering
- * index) and checking again if its still eligible for REPACK
- * CONCURRENTLY.
+ * This is to make sure that the deadlock described below is
+ * detectable by deadlock.c: if the worker waits for a transaction to
+ * complete and we are waiting for the worker output, then effectively
+ * we (i.e. this backend) are waiting for that transaction.
*/
- decoding_ctx = setup_logical_decoding(tableOid);
+ BecomeLockGroupLeader();
+
+ /*
+ * Start the worker that decodes data changes applied while we're
+ * copying the table contents.
+ *
+ * Note that the worker has to wait for all transactions with XID
+ * already assigned to finish. If some of those transactions is
+ * waiting for a lock conflicting with ShareUpdateExclusiveLock on our
+ * table (e.g. it runs CREATE INDEX), we can end up in a deadlock.
+ * Not sure this risk is worth unlocking/locking the table (and its
+ * clustering index) and checking again if it's still eligible for
+ * REPACK CONCURRENTLY.
+ */
+ start_decoding_worker(tableOid);
+
+ /*
+ * Wait until the worker has the initial snapshot and retrieve it.
+ */
+ snapshot = get_initial_snapshot(decoding_worker);
- snapshot = SnapBuildInitialSnapshotForRepack(decoding_ctx->snapshot_builder);
PushActiveSnapshot(snapshot);
}
@@ -980,7 +1117,7 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
NewHeap = table_open(OIDNewHeap, NoLock);
/* Copy the heap data into the new table in the desired order */
- copy_table_data(NewHeap, OldHeap, index, snapshot, decoding_ctx, verbose,
+ copy_table_data(NewHeap, OldHeap, index, snapshot, verbose,
&swap_toast_by_content, &frozenXid, &cutoffMulti);
/* The historic snapshot won't be needed anymore. */
@@ -1001,14 +1138,11 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
if (index)
index_close(index, NoLock);
- rebuild_relation_finish_concurrent(NewHeap, OldHeap, decoding_ctx,
- frozenXid, cutoffMulti);
+ rebuild_relation_finish_concurrent(NewHeap, OldHeap, frozenXid,
+ cutoffMulti);
pgstat_progress_update_param(PROGRESS_REPACK_PHASE,
PROGRESS_REPACK_PHASE_FINAL_CLEANUP);
-
- /* Done with decoding. */
- cleanup_logical_decoding(decoding_ctx);
}
else
{
@@ -1179,8 +1313,7 @@ make_new_heap(Oid OIDOldHeap, Oid NewTableSpace, Oid NewAccessMethod,
*/
static void
copy_table_data(Relation NewHeap, Relation OldHeap, Relation OldIndex,
- Snapshot snapshot, LogicalDecodingContext *decoding_ctx,
- bool verbose, bool *pSwapToastByContent,
+ Snapshot snapshot, bool verbose, bool *pSwapToastByContent,
TransactionId *pFreezeXid, MultiXactId *pCutoffMulti)
{
Relation relRelation;
@@ -1341,7 +1474,6 @@ copy_table_data(Relation NewHeap, Relation OldHeap, Relation OldIndex,
*/
table_relation_copy_for_cluster(OldHeap, NewHeap, OldIndex, use_sort,
cutoffs.OldestXmin, snapshot,
- decoding_ctx,
&cutoffs.FreezeLimit,
&cutoffs.MultiXactCutoff,
&num_tuples, &tups_vacuumed,
@@ -2371,62 +2503,10 @@ RepackCommandAsString(RepackCommand cmd)
case REPACK_COMMAND_CLUSTER:
return "CLUSTER";
}
- return "???"; /* keep compiler quiet */
+ return "???"; /* keep compiler quiet */
}
-/*
- * Call this function before REPACK CONCURRENTLY starts to setup logical
- * decoding. It makes sure that other users of the table put enough
- * information into WAL.
- *
- * The point is that at various places we expect that the table we're
- * processing is treated like a system catalog. For example, we need to be
- * able to scan it using a "historic snapshot" anytime during the processing
- * (as opposed to scanning only at the start point of the decoding, as logical
- * replication does during initial table synchronization), in order to apply
- * concurrent UPDATE / DELETE commands.
- *
- * Note that TOAST table needs no attention here as it's not scanned using
- * historic snapshot.
- */
-static void
-begin_concurrent_repack(Relation rel)
-{
- Oid toastrelid;
-
- /*
- * Avoid logical decoding of other relations by this backend. The lock we
- * have guarantees that the actual locator cannot be changed concurrently:
- * TRUNCATE needs AccessExclusiveLock.
- */
- Assert(CheckRelationLockedByMe(rel, ShareUpdateExclusiveLock, false));
- repacked_rel_locator = rel->rd_locator;
- toastrelid = rel->rd_rel->reltoastrelid;
- if (OidIsValid(toastrelid))
- {
- Relation toastrel;
-
- /* Avoid logical decoding of other TOAST relations. */
- toastrel = table_open(toastrelid, AccessShareLock);
- repacked_rel_toast_locator = toastrel->rd_locator;
- table_close(toastrel, AccessShareLock);
- }
-}
-
-/*
- * Call this when done with REPACK CONCURRENTLY.
- */
-static void
-end_concurrent_repack(void)
-{
- /*
- * Restore normal function of (future) logical decoding for this backend.
- */
- repacked_rel_locator.relNumber = InvalidOid;
- repacked_rel_toast_locator.relNumber = InvalidOid;
-}
-
/*
* Is this backend performing logical decoding on behalf of REPACK
* (CONCURRENTLY) ?
@@ -2491,9 +2571,10 @@ static LogicalDecodingContext *
setup_logical_decoding(Oid relid)
{
Relation rel;
- TupleDesc tupdesc;
+ Oid toastrelid;
LogicalDecodingContext *ctx;
- RepackDecodingState *dstate = palloc0_object(RepackDecodingState);
+ NameData slotname;
+ RepackDecodingState *dstate;
/*
* REPACK CONCURRENTLY is not allowed in a transaction block, so this
@@ -2501,20 +2582,21 @@ setup_logical_decoding(Oid relid)
*/
Assert(!TransactionIdIsValid(GetTopTransactionIdIfAny()));
- /*
- * A single backend should not execute multiple REPACK commands at a time,
- * so use PID to make the slot unique.
- */
- snprintf(NameStr(dstate->slotname), NAMEDATALEN, "repack_%d", MyProcPid);
-
/*
* Make sure we can use logical decoding.
*/
CheckSlotPermissions();
CheckLogicalDecodingRequirements();
- /* RS_TEMPORARY so that the slot gets cleaned up on ERROR. */
- ReplicationSlotCreate(NameStr(dstate->slotname), true, RS_TEMPORARY,
- false, false, false);
+ /*
+ * A single backend should not execute multiple REPACK commands at a time,
+ * so use PID to make the slot unique.
+ *
+ * RS_TEMPORARY so that the slot gets cleaned up on ERROR.
+ */
+ snprintf(NameStr(slotname), NAMEDATALEN, "repack_%d", MyProcPid);
+ ReplicationSlotCreate(NameStr(slotname), true, RS_TEMPORARY, false, false,
+ false);
+
EnsureLogicalDecodingEnabled();
/*
@@ -2537,104 +2619,109 @@ setup_logical_decoding(Oid relid)
DecodingContextFindStartpoint(ctx);
+ /*
+ * decode_concurrent_changes() needs non-blocking callback.
+ */
+ ctx->reader->routine.page_read = read_local_xlog_page_no_wait;
+
+ /*
+ * read_local_xlog_page_no_wait() needs to be able to indicate the end of
+ * WAL.
+ */
+ ctx->reader->private_data = MemoryContextAllocZero(ctx->context,
+ sizeof(ReadLocalXLogPageNoWaitPrivate));
+
+
/* Some WAL records should have been read. */
Assert(ctx->reader->EndRecPtr != InvalidXLogRecPtr);
+ /*
+ * Initialize repack_current_segment so that we can notice WAL segment
+ * boundaries.
+ */
XLByteToSeg(ctx->reader->EndRecPtr, repack_current_segment,
wal_segment_size);
- /*
- * Setup structures to store decoded changes.
- */
+ dstate = palloc0_object(RepackDecodingState);
dstate->relid = relid;
- dstate->tstore = tuplestore_begin_heap(false, false,
- maintenance_work_mem);
- /* Caller should already have the table locked. */
- rel = table_open(relid, NoLock);
- tupdesc = CreateTupleDescCopy(RelationGetDescr(rel));
- dstate->tupdesc = tupdesc;
- table_close(rel, NoLock);
+ /*
+ * Tuple descriptor may be needed to flatten a tuple before we write it to
+ * a file. A copy is needed because the decoding worker invalidates system
+ * caches before it starts to do the actual work.
+ */
+ rel = table_open(relid, AccessShareLock);
+ dstate->tupdesc = CreateTupleDescCopy(RelationGetDescr(rel));
- /* Initialize the descriptor to store the changes ... */
- dstate->tupdesc_change = CreateTemplateTupleDesc(1);
+ /* Avoid logical decoding of other relations. */
+ repacked_rel_locator = rel->rd_locator;
+ toastrelid = rel->rd_rel->reltoastrelid;
+ if (OidIsValid(toastrelid))
+ {
+ Relation toastrel;
- TupleDescInitEntry(dstate->tupdesc_change, 1, NULL, BYTEAOID, -1, 0);
- /* ... as well as the corresponding slot. */
- dstate->tsslot = MakeSingleTupleTableSlot(dstate->tupdesc_change,
- &TTSOpsMinimalTuple);
+ /* Avoid logical decoding of other TOAST relations. */
+ toastrel = table_open(toastrelid, AccessShareLock);
+ repacked_rel_toast_locator = toastrel->rd_locator;
+ table_close(toastrel, AccessShareLock);
+ }
+ table_close(rel, AccessShareLock);
- dstate->resowner = ResourceOwnerCreate(CurrentResourceOwner,
- "logical decoding");
+ /* The file will be set as soon as we have it opened. */
+ dstate->file = NULL;
ctx->output_writer_private = dstate;
+
return ctx;
}
/*
- * Retrieve tuple from ConcurrentChange structure.
+ * Decode logical changes from the WAL sequence and store them to a file.
*
- * The input data starts with the structure but it might not be appropriately
- * aligned.
- */
-static HeapTuple
-get_changed_tuple(char *change)
-{
- HeapTupleData tup_data;
- HeapTuple result;
- char *src;
-
- /*
- * Ensure alignment before accessing the fields. (This is why we can't use
- * heap_copytuple() instead of this function.)
- */
- src = change + offsetof(ConcurrentChange, tup_data);
- memcpy(&tup_data, src, sizeof(HeapTupleData));
-
- result = (HeapTuple) palloc(HEAPTUPLESIZE + tup_data.t_len);
- memcpy(result, &tup_data, sizeof(HeapTupleData));
- result->t_data = (HeapTupleHeader) ((char *) result + HEAPTUPLESIZE);
- src = change + SizeOfConcurrentChange;
- memcpy(result->t_data, src, result->t_len);
-
- return result;
-}
-
-/*
- * Decode logical changes from the WAL sequence up to end_of_wal.
+ * If true is returned, there is no more work for the worker.
*/
-void
-repack_decode_concurrent_changes(LogicalDecodingContext *ctx,
- XLogRecPtr end_of_wal)
+static bool
+decode_concurrent_changes(LogicalDecodingContext *ctx,
+ DecodingWorkerShared *shared)
{
RepackDecodingState *dstate;
- ResourceOwner resowner_old;
+ XLogRecPtr lsn_upto;
+ bool done;
+ char fname[MAXPGPATH];
dstate = (RepackDecodingState *) ctx->output_writer_private;
- resowner_old = CurrentResourceOwner;
- CurrentResourceOwner = dstate->resowner;
- PG_TRY();
+ /* Open the output file. */
+ DecodingWorkerFileName(fname, shared->relid, shared->last_exported + 1);
+ dstate->file = BufFileCreateFileSet(&shared->sfs.fs, fname);
+
+ SpinLockAcquire(&shared->mutex);
+ lsn_upto = shared->lsn_upto;
+ done = shared->done;
+ SpinLockRelease(&shared->mutex);
+
+ while (true)
{
- while (ctx->reader->EndRecPtr < end_of_wal)
- {
- XLogRecord *record;
- XLogSegNo segno_new;
- char *errm = NULL;
- XLogRecPtr end_lsn;
+ XLogRecord *record;
+ XLogSegNo segno_new;
+ char *errm = NULL;
+ XLogRecPtr end_lsn;
- record = XLogReadRecord(ctx->reader, &errm);
- if (errm)
- elog(ERROR, "%s", errm);
+ CHECK_FOR_INTERRUPTS();
- if (record != NULL)
- LogicalDecodingProcessRecord(ctx, ctx->reader);
+ record = XLogReadRecord(ctx->reader, &errm);
+ if (record)
+ {
+ LogicalDecodingProcessRecord(ctx, ctx->reader);
/*
* If WAL segment boundary has been crossed, inform the decoding
- * system that the catalog_xmin can advance. (We can confirm more
- * often, but a filling a single WAL segment should not take much
- * time.)
+ * system that the catalog_xmin can advance.
+ *
+ * TODO Does it make sense to confirm more often? Segment size
+ * seems appropriate for restart_lsn (because less than a segment
+ * cannot be recycled anyway), however more frequent checks might
+ * be beneficial for catalog_xmin.
*/
end_lsn = ctx->reader->EndRecPtr;
XLByteToSeg(end_lsn, segno_new, wal_segment_size);
@@ -2645,80 +2732,137 @@ repack_decode_concurrent_changes(LogicalDecodingContext *ctx,
(uint32) (end_lsn >> 32), (uint32) end_lsn);
repack_current_segment = segno_new;
}
+ }
+ else
+ {
+ ReadLocalXLogPageNoWaitPrivate *priv;
+
+ if (errm)
+ ereport(ERROR, (errmsg("%s", errm)));
- CHECK_FOR_INTERRUPTS();
+ /*
+ * In the decoding loop we do not want to get blocked when there
+ * is no more WAL available, otherwise the loop would become
+ * uninterruptible.
+ */
+ priv = (ReadLocalXLogPageNoWaitPrivate *)
+ ctx->reader->private_data;
+ if (priv->end_of_wal)
+ /* Do not miss the end of WAL condition next time. */
+ priv->end_of_wal = false;
+ else
+ ereport(ERROR, (errmsg("could not read WAL record")));
+ }
+
+ /*
+ * Whether we could read new record or not, keep checking if
+ * 'lsn_upto' was specified.
+ */
+ if (XLogRecPtrIsInvalid(lsn_upto))
+ {
+ SpinLockAcquire(&shared->mutex);
+ lsn_upto = shared->lsn_upto;
+ /* 'done' should be set at the same time as 'lsn_upto' */
+ done = shared->done;
+ SpinLockRelease(&shared->mutex);
+ }
+ if (!XLogRecPtrIsInvalid(lsn_upto) &&
+ ctx->reader->EndRecPtr >= lsn_upto)
+ break;
+
+ if (record == NULL)
+ {
+ int64 timeout = 0;
+ WaitLSNResult res;
+
+ /*
+ * Before we retry reading, wait until new WAL is flushed.
+ *
+ * There is a race condition such that the backend executing
+ * REPACK determines 'lsn_upto', but before it sets the shared
+ * variable, we reach the end of WAL. In that case we'd need to
+ * wait until the next WAL flush (unrelated to REPACK). Although
+ * that should not be a problem in a busy system, it might be
+ * noticeable in other cases, including regression tests (which
+ * are not necessarily executed in parallel). Therefore it makes
+ * sense to use timeout.
+ *
+ * If lsn_upto is valid, WAL records having LSN lower than that
+ * should already have been flushed to disk.
+ */
+ if (XLogRecPtrIsInvalid(lsn_upto))
+ timeout = 100L;
+ res = WaitForLSN(WAIT_LSN_TYPE_PRIMARY_FLUSH,
+ ctx->reader->EndRecPtr + 1,
+ timeout);
+ if (res != WAIT_LSN_RESULT_SUCCESS &&
+ res != WAIT_LSN_RESULT_TIMEOUT)
+ ereport(ERROR, (errmsg("waiting for WAL failed")));
}
- InvalidateSystemCaches();
- CurrentResourceOwner = resowner_old;
- }
- PG_CATCH();
- {
- /* clear all timetravel entries */
- InvalidateSystemCaches();
- CurrentResourceOwner = resowner_old;
- PG_RE_THROW();
}
- PG_END_TRY();
+
+ /*
+ * Close the file so we can make it available to the backend.
+ */
+ BufFileClose(dstate->file);
+ dstate->file = NULL;
+ SpinLockAcquire(&shared->mutex);
+ shared->lsn_upto = InvalidXLogRecPtr;
+ shared->last_exported++;
+ SpinLockRelease(&shared->mutex);
+ ConditionVariableSignal(&shared->cv);
+
+ return done;
}
/*
* Apply changes stored in 'file'.
*/
static void
-apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
+apply_concurrent_changes(BufFile *file, ChangeDest *dest)
{
+ char kind;
+ uint32 t_len;
Relation rel = dest->rel;
TupleTableSlot *index_slot,
*ident_slot;
HeapTuple tup_old = NULL;
- if (dstate->nchanges == 0)
- return;
-
/* TupleTableSlot is needed to pass the tuple to ExecInsertIndexTuples(). */
- index_slot = MakeSingleTupleTableSlot(dstate->tupdesc, &TTSOpsHeapTuple);
+ index_slot = MakeSingleTupleTableSlot(RelationGetDescr(rel),
+ &TTSOpsHeapTuple);
/* A slot to fetch tuples from identity index. */
ident_slot = table_slot_create(rel, NULL);
- while (tuplestore_gettupleslot(dstate->tstore, true, false,
- dstate->tsslot))
+ while (true)
{
- bool shouldFree;
- HeapTuple tup_change,
- tup,
+ size_t nread;
+ HeapTuple tup,
tup_exist;
- char *change_raw,
- *src;
- ConcurrentChange change;
- bool isnull[1];
- Datum values[1];
CHECK_FOR_INTERRUPTS();
- /* Get the change from the single-column tuple. */
- tup_change = ExecFetchSlotHeapTuple(dstate->tsslot, false, &shouldFree);
- heap_deform_tuple(tup_change, dstate->tupdesc_change, values, isnull);
- Assert(!isnull[0]);
-
- /* Make sure we access aligned data. */
- change_raw = (char *) DatumGetByteaP(values[0]);
- src = (char *) VARDATA(change_raw);
- memcpy(&change, src, SizeOfConcurrentChange);
+ nread = BufFileReadMaybeEOF(file, &kind, 1, true);
+ /* Are we done with the file? */
+ if (nread == 0)
+ break;
- /*
- * Extract the tuple from the change. The tuple is copied here because
- * it might be assigned to 'tup_old', in which case it needs to
- * survive into the next iteration.
- */
- tup = get_changed_tuple(src);
+ /* Read the tuple. */
+ BufFileReadExact(file, &t_len, sizeof(t_len));
+ tup = (HeapTuple) palloc(HEAPTUPLESIZE + t_len);
+ tup->t_data = (HeapTupleHeader) ((char *) tup + HEAPTUPLESIZE);
+ BufFileReadExact(file, tup->t_data, t_len);
+ tup->t_len = t_len;
+ ItemPointerSetInvalid(&tup->t_self);
+ tup->t_tableOid = RelationGetRelid(dest->rel);
- if (change.kind == CHANGE_UPDATE_OLD)
+ if (kind == CHANGE_UPDATE_OLD)
{
Assert(tup_old == NULL);
tup_old = tup;
}
- else if (change.kind == CHANGE_INSERT)
+ else if (kind == CHANGE_INSERT)
{
Assert(tup_old == NULL);
@@ -2726,12 +2870,11 @@ apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
pfree(tup);
}
- else if (change.kind == CHANGE_UPDATE_NEW ||
- change.kind == CHANGE_DELETE)
+ else if (kind == CHANGE_UPDATE_NEW || kind == CHANGE_DELETE)
{
HeapTuple tup_key;
- if (change.kind == CHANGE_UPDATE_NEW)
+ if (kind == CHANGE_UPDATE_NEW)
{
tup_key = tup_old != NULL ? tup_old : tup;
}
@@ -2748,7 +2891,7 @@ apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
if (tup_exist == NULL)
elog(ERROR, "failed to find target tuple");
- if (change.kind == CHANGE_UPDATE_NEW)
+ if (kind == CHANGE_UPDATE_NEW)
apply_concurrent_update(rel, tup, tup_exist, dest->iistate,
index_slot);
else
@@ -2763,26 +2906,19 @@ apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
pfree(tup);
}
else
- elog(ERROR, "unrecognized kind of change: %d", change.kind);
+ elog(ERROR, "unrecognized kind of change: %d", kind);
/*
* If a change was applied now, increment CID for next writes and
* update the snapshot so it sees the changes we've applied so far.
*/
- if (change.kind != CHANGE_UPDATE_OLD)
+ if (kind != CHANGE_UPDATE_OLD)
{
CommandCounterIncrement();
UpdateActiveSnapshotCommandId();
}
-
- /* TTSOpsMinimalTuple has .get_heap_tuple==NULL. */
- Assert(shouldFree);
- pfree(tup_change);
}
- tuplestore_clear(dstate->tstore);
- dstate->nchanges = 0;
-
/* Cleanup. */
ExecDropSingleTupleTableSlot(index_slot);
ExecDropSingleTupleTableSlot(ident_slot);
@@ -2957,25 +3093,59 @@ find_target_tuple(Relation rel, ChangeDest *dest, HeapTuple tup_key,
}
/*
- * Decode and apply concurrent changes.
+ * Decode and apply concurrent changes, up to (and including) the record whose
+ * LSN is 'end_of_wal'.
*/
static void
-process_concurrent_changes(LogicalDecodingContext *decoding_ctx,
- XLogRecPtr end_of_wal, ChangeDest *dest)
+process_concurrent_changes(XLogRecPtr end_of_wal, ChangeDest *dest, bool done)
{
- RepackDecodingState *dstate;
+ DecodingWorkerShared *shared;
+ char fname[MAXPGPATH];
+ BufFile *file;
pgstat_progress_update_param(PROGRESS_REPACK_PHASE,
PROGRESS_REPACK_PHASE_CATCH_UP);
- dstate = (RepackDecodingState *) decoding_ctx->output_writer_private;
+ /* Ask the worker for the file. */
+ shared = (DecodingWorkerShared *) dsm_segment_address(decoding_worker->seg);
+ SpinLockAcquire(&shared->mutex);
+ shared->lsn_upto = end_of_wal;
+ shared->done = done;
+ SpinLockRelease(&shared->mutex);
+
+ /*
+ * The worker needs to finish processing of the current WAL record. Even
+ * if it's idle, it'll need to close the output file. Thus we're likely to
+ * wait, so prepare for sleep.
+ */
+ ConditionVariablePrepareToSleep(&shared->cv);
+ for (;;)
+ {
+ int last_exported;
- repack_decode_concurrent_changes(decoding_ctx, end_of_wal);
+ SpinLockAcquire(&shared->mutex);
+ last_exported = shared->last_exported;
+ SpinLockRelease(&shared->mutex);
- if (dstate->nchanges == 0)
- return;
+ /*
+ * Has the worker exported the file we are waiting for?
+ */
+ if (last_exported == dest->file_seq)
+ break;
- apply_concurrent_changes(dstate, dest);
+ ConditionVariableSleep(&shared->cv, WAIT_EVENT_REPACK_WORKER_EXPORT);
+ }
+ ConditionVariableCancelSleep();
+
+ /* Open the file. */
+ DecodingWorkerFileName(fname, shared->relid, dest->file_seq);
+ file = BufFileOpenFileSet(&shared->sfs.fs, fname, O_RDONLY, false);
+ apply_concurrent_changes(file, dest);
+
+ BufFileClose(file);
+
+ /* Get ready for the next file. */
+ dest->file_seq++;
}
/*
@@ -3101,15 +3271,10 @@ cleanup_logical_decoding(LogicalDecodingContext *ctx)
dstate = (RepackDecodingState *) ctx->output_writer_private;
- ExecDropSingleTupleTableSlot(dstate->tsslot);
- FreeTupleDesc(dstate->tupdesc_change);
FreeTupleDesc(dstate->tupdesc);
- tuplestore_end(dstate->tstore);
-
FreeDecodingContext(ctx);
- ReplicationSlotRelease();
- ReplicationSlotDrop(NameStr(dstate->slotname), false);
+ ReplicationSlotDropAcquired();
pfree(dstate);
}
@@ -3123,7 +3288,6 @@ cleanup_logical_decoding(LogicalDecodingContext *ctx)
*/
static void
rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
- LogicalDecodingContext *decoding_ctx,
TransactionId frozenXid,
MultiXactId cutoffMulti)
{
@@ -3204,6 +3368,7 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
&chgdst.ident_index);
chgdst.ident_key = build_identity_key(ident_idx_new, OldHeap,
&chgdst.ident_key_nentries);
+ chgdst.file_seq = WORKER_FILE_SNAPSHOT + 1;
/*
* During testing, wait for another backend to perform concurrent data
@@ -3225,7 +3390,7 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
* hold AccessExclusiveLock. (Quite some amount of WAL could have been
* written during the data copying and index creation.)
*/
- process_concurrent_changes(decoding_ctx, end_of_wal, &chgdst);
+ process_concurrent_changes(end_of_wal, &chgdst, false);
/*
* Acquire AccessExclusiveLock on the table, its TOAST relation (if there
@@ -3306,8 +3471,11 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
XLogFlush(wal_insert_ptr);
end_of_wal = GetFlushRecPtr(NULL);
- /* Apply the concurrent changes again. */
- process_concurrent_changes(decoding_ctx, end_of_wal, &chgdst);
+ /*
+ * Apply the concurrent changes again. Indicate that the decoding worker
+ * won't be needed anymore.
+ */
+ process_concurrent_changes(end_of_wal, &chgdst, true);
/* Remember info about rel before closing OldHeap */
relpersistence = OldHeap->rd_rel->relpersistence;
@@ -3417,3 +3585,510 @@ build_new_indexes(Relation NewHeap, Relation OldHeap, List *OldIndexes)
return result;
}
+
+/*
+ * Try to start a background worker to perform logical decoding of data
+ * changes applied to relation while REPACK CONCURRENTLY is copying its
+ * contents to a new table.
+ */
+static void
+start_decoding_worker(Oid relid)
+{
+ Size size;
+ dsm_segment *seg;
+ DecodingWorkerShared *shared;
+ shm_mq *mq;
+ shm_mq_handle *mqh;
+ BackgroundWorker bgw;
+
+ /* Setup shared memory. */
+ size = BUFFERALIGN(offsetof(DecodingWorkerShared, error_queue)) +
+ BUFFERALIGN(REPACK_ERROR_QUEUE_SIZE);
+ seg = dsm_create(size, 0);
+ shared = (DecodingWorkerShared *) dsm_segment_address(seg);
+ shared->lsn_upto = InvalidXLogRecPtr;
+ shared->done = false;
+ SharedFileSetInit(&shared->sfs, seg);
+ shared->last_exported = -1;
+ SpinLockInit(&shared->mutex);
+ shared->dbid = MyDatabaseId;
+
+ /*
+ * This is the UserId set in cluster_rel(). Security context shouldn't be
+ * needed for decoding worker.
+ */
+ shared->roleid = GetUserId();
+ shared->relid = relid;
+ ConditionVariableInit(&shared->cv);
+ shared->backend_proc = MyProc;
+ shared->backend_pid = MyProcPid;
+ shared->backend_proc_number = MyProcNumber;
+
+ mq = shm_mq_create((char *) BUFFERALIGN(shared->error_queue),
+ REPACK_ERROR_QUEUE_SIZE);
+ shm_mq_set_receiver(mq, MyProc);
+ mqh = shm_mq_attach(mq, seg, NULL);
+
+ memset(&bgw, 0, sizeof(bgw));
+ snprintf(bgw.bgw_name, BGW_MAXLEN,
+ "REPACK decoding worker for relation \"%s\"",
+ get_rel_name(relid));
+ snprintf(bgw.bgw_type, BGW_MAXLEN, "REPACK decoding worker");
+ bgw.bgw_flags = BGWORKER_SHMEM_ACCESS |
+ BGWORKER_BACKEND_DATABASE_CONNECTION;
+ bgw.bgw_start_time = BgWorkerStart_RecoveryFinished;
+ bgw.bgw_restart_time = BGW_NEVER_RESTART;
+ snprintf(bgw.bgw_library_name, MAXPGPATH, "postgres");
+ snprintf(bgw.bgw_function_name, BGW_MAXLEN, "RepackWorkerMain");
+ bgw.bgw_main_arg = UInt32GetDatum(dsm_segment_handle(seg));
+ bgw.bgw_notify_pid = MyProcPid;
+
+ decoding_worker = palloc0_object(DecodingWorker);
+ if (!RegisterDynamicBackgroundWorker(&bgw, &decoding_worker->handle))
+ ereport(ERROR,
+ (errcode(ERRCODE_CONFIGURATION_LIMIT_EXCEEDED),
+ errmsg("out of background worker slots"),
+ errhint("You might need to increase \"%s\".", "max_worker_processes")));
+
+ decoding_worker->seg = seg;
+ decoding_worker->error_mqh = mqh;
+
+ /*
+ * The decoding setup must be done before the caller can have XID assigned
+ * for any reason, otherwise the worker might end up in a deadlock,
+ * waiting for the caller's transaction to end. Therefore wait here until
+ * the worker indicates that it has the logical decoding initialized.
+ */
+ ConditionVariablePrepareToSleep(&shared->cv);
+ for (;;)
+ {
+ bool initialized;
+
+ SpinLockAcquire(&shared->mutex);
+ initialized = shared->initialized;
+ SpinLockRelease(&shared->mutex);
+
+ if (initialized)
+ break;
+
+ ConditionVariableSleep(&shared->cv, WAIT_EVENT_REPACK_WORKER_EXPORT);
+ }
+ ConditionVariableCancelSleep();
+}
+
+/*
+ * Stop the decoding worker and cleanup the related resources.
+ *
+ * The worker stops on its own when it knows there is no more work to do, but
+ * we need to stop it explicitly at least on ERROR in the launching backend.
+ */
+static void
+stop_decoding_worker(void)
+{
+ BgwHandleStatus status;
+
+ /* Haven't reached the worker startup? */
+ if (decoding_worker == NULL)
+ return;
+
+ /* Could not register the worker? */
+ if (decoding_worker->handle == NULL)
+ return;
+
+ TerminateBackgroundWorker(decoding_worker->handle);
+ /* The worker should really exit before the REPACK command does. */
+ HOLD_INTERRUPTS();
+ status = WaitForBackgroundWorkerShutdown(decoding_worker->handle);
+ RESUME_INTERRUPTS();
+
+ if (status == BGWH_POSTMASTER_DIED)
+ ereport(FATAL,
+ (errcode(ERRCODE_ADMIN_SHUTDOWN),
+ errmsg("postmaster exited during REPACK command")));
+
+ shm_mq_detach(decoding_worker->error_mqh);
+
+ /*
+ * If we could not cancel the current sleep due to ERROR, do that before
+ * we detach from the shared memory the condition variable is located in.
+ * If we did not, the bgworker ERROR handling code would try and fail
+ * badly.
+ */
+ ConditionVariableCancelSleep();
+
+ dsm_detach(decoding_worker->seg);
+ pfree(decoding_worker);
+ decoding_worker = NULL;
+}
+
+/* Is this process a REPACK worker? */
+static bool is_repack_worker = false;
+
+static pid_t backend_pid;
+static ProcNumber backend_proc_number;
+
+/*
+ * See ParallelWorkerShutdown for details.
+ */
+static void
+RepackWorkerShutdown(int code, Datum arg)
+{
+ SendProcSignal(backend_pid,
+ PROCSIG_REPACK_MESSAGE,
+ backend_proc_number);
+
+ dsm_detach((dsm_segment *) DatumGetPointer(arg));
+}
+
+/* REPACK decoding worker entry point */
+void
+RepackWorkerMain(Datum main_arg)
+{
+ dsm_segment *seg;
+ DecodingWorkerShared *shared;
+ shm_mq *mq;
+ shm_mq_handle *mqh;
+
+ is_repack_worker = true;
+
+ /*
+ * Override the default bgworker_die() with die() so we can use
+ * CHECK_FOR_INTERRUPTS().
+ */
+ pqsignal(SIGTERM, die);
+ BackgroundWorkerUnblockSignals();
+
+ seg = dsm_attach(DatumGetUInt32(main_arg));
+ if (seg == NULL)
+ ereport(ERROR,
+ (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
+ errmsg("could not map dynamic shared memory segment")));
+
+ shared = (DecodingWorkerShared *) dsm_segment_address(seg);
+
+ /* Arrange to signal the leader if we exit. */
+ backend_pid = shared->backend_pid;
+ backend_proc_number = shared->backend_proc_number;
+ before_shmem_exit(RepackWorkerShutdown, PointerGetDatum(seg));
+
+ /*
+ * Join locking group - see the comments around the call of
+ * start_decoding_worker().
+ */
+ if (!BecomeLockGroupMember(shared->backend_proc, backend_pid))
+ /* The leader is not running anymore. */
+ return;
+
+ /*
+ * Setup a queue to send error messages to the backend that launched this
+ * worker.
+ */
+ mq = (shm_mq *) (char *) BUFFERALIGN(shared->error_queue);
+ shm_mq_set_sender(mq, MyProc);
+ mqh = shm_mq_attach(mq, seg, NULL);
+ pq_redirect_to_shm_mq(seg, mqh);
+ pq_set_parallel_leader(shared->backend_pid,
+ shared->backend_proc_number);
+
+ /* Connect to the database. */
+ BackgroundWorkerInitializeConnectionByOid(shared->dbid, shared->roleid, 0);
+
+ repack_worker_internal(seg);
+}
+
+static void
+repack_worker_internal(dsm_segment *seg)
+{
+ DecodingWorkerShared *shared;
+ LogicalDecodingContext *decoding_ctx;
+ SharedFileSet *sfs;
+ Snapshot snapshot;
+
+ /*
+ * Transaction is needed to open relation, and it also provides us with a
+ * resource owner.
+ */
+ StartTransactionCommand();
+
+ shared = (DecodingWorkerShared *) dsm_segment_address(seg);
+
+ /*
+ * Not sure the spinlock is needed here - the backend should not change
+ * anything in the shared memory until we have serialized the snapshot.
+ */
+ SpinLockAcquire(&shared->mutex);
+ Assert(XLogRecPtrIsInvalid(shared->lsn_upto));
+ sfs = &shared->sfs;
+ SpinLockRelease(&shared->mutex);
+
+ SharedFileSetAttach(sfs, seg);
+
+ /*
+ * Prepare to capture the concurrent data changes ourselves.
+ */
+ decoding_ctx = setup_logical_decoding(shared->relid);
+
+ /* Announce that we're ready. */
+ SpinLockAcquire(&shared->mutex);
+ shared->initialized = true;
+ SpinLockRelease(&shared->mutex);
+ ConditionVariableSignal(&shared->cv);
+
+ /* Build the initial snapshot and export it. */
+ snapshot = SnapBuildInitialSnapshotForRepack(decoding_ctx->snapshot_builder);
+ export_initial_snapshot(snapshot, shared);
+
+ /*
+ * Only historic snapshots should be used now. Do not let us restrict the
+ * progress of xmin horizon.
+ */
+ InvalidateCatalogSnapshot();
+
+ while (!decode_concurrent_changes(decoding_ctx, shared))
+ ;
+
+ /* Cleanup. */
+ cleanup_logical_decoding(decoding_ctx);
+ CommitTransactionCommand();
+}
+
+/*
+ * Make snapshot available to the backend that launched the decoding worker.
+ */
+static void
+export_initial_snapshot(Snapshot snapshot, DecodingWorkerShared *shared)
+{
+ char fname[MAXPGPATH];
+ BufFile *file;
+ Size snap_size;
+ char *snap_space;
+
+ snap_size = EstimateSnapshotSpace(snapshot);
+ snap_space = (char *) palloc(snap_size);
+ SerializeSnapshot(snapshot, snap_space);
+ FreeSnapshot(snapshot);
+
+ DecodingWorkerFileName(fname, shared->relid, shared->last_exported + 1);
+ file = BufFileCreateFileSet(&shared->sfs.fs, fname);
+ /* To make restoration easier, write the snapshot size first. */
+ BufFileWrite(file, &snap_size, sizeof(snap_size));
+ BufFileWrite(file, snap_space, snap_size);
+ pfree(snap_space);
+ BufFileClose(file);
+
+ /* Increase the counter to tell the backend that the file is available. */
+ SpinLockAcquire(&shared->mutex);
+ shared->last_exported++;
+ SpinLockRelease(&shared->mutex);
+ ConditionVariableSignal(&shared->cv);
+}
+
+/*
+ * Get the initial snapshot from the decoding worker.
+ */
+static Snapshot
+get_initial_snapshot(DecodingWorker *worker)
+{
+ DecodingWorkerShared *shared;
+ char fname[MAXPGPATH];
+ BufFile *file;
+ Size snap_size;
+ char *snap_space;
+ Snapshot snapshot;
+
+ shared = (DecodingWorkerShared *) dsm_segment_address(worker->seg);
+
+ /*
+ * The worker needs to initialize the logical decoding, which usually
+ * takes some time. Therefore it makes sense to prepare for the sleep
+ * first.
+ */
+ ConditionVariablePrepareToSleep(&shared->cv);
+ for (;;)
+ {
+ int last_exported;
+
+ SpinLockAcquire(&shared->mutex);
+ last_exported = shared->last_exported;
+ SpinLockRelease(&shared->mutex);
+
+ /*
+ * Has the worker exported the file we are waiting for?
+ */
+ if (last_exported == WORKER_FILE_SNAPSHOT)
+ break;
+
+ ConditionVariableSleep(&shared->cv, WAIT_EVENT_REPACK_WORKER_EXPORT);
+ }
+ ConditionVariableCancelSleep();
+
+ /* Read the snapshot from a file. */
+ DecodingWorkerFileName(fname, shared->relid, WORKER_FILE_SNAPSHOT);
+ file = BufFileOpenFileSet(&shared->sfs.fs, fname, O_RDONLY, false);
+ BufFileReadExact(file, &snap_size, sizeof(snap_size));
+ snap_space = (char *) palloc(snap_size);
+ BufFileReadExact(file, snap_space, snap_size);
+ BufFileClose(file);
+
+ /* Restore it. */
+ snapshot = RestoreSnapshot(snap_space);
+ pfree(snap_space);
+
+ return snapshot;
+}
+
+bool
+IsRepackWorker(void)
+{
+ return is_repack_worker;
+}
+
+/*
+ * Handle receipt of an interrupt indicating a repack worker message.
+ *
+ * Note: this is called within a signal handler! All we can do is set
+ * a flag that will cause the next CHECK_FOR_INTERRUPTS() to invoke
+ * ProcessRepackMessages().
+ */
+void
+HandleRepackMessageInterrupt(void)
+{
+ InterruptPending = true;
+ RepackMessagePending = true;
+ SetLatch(MyLatch);
+}
+
+/*
+ * Process any queued protocol messages received from parallel workers.
+ */
+void
+ProcessRepackMessages(void)
+{
+ MemoryContext oldcontext;
+
+ static MemoryContext hpm_context = NULL;
+
+ /*
+ * Nothing to do if we haven't launched the worker yet or have already
+ * terminated it.
+ */
+ if (decoding_worker == NULL)
+ return;
+
+ /*
+ * This is invoked from ProcessInterrupts(), and since some of the
+ * functions it calls contain CHECK_FOR_INTERRUPTS(), there is a potential
+ * for recursive calls if more signals are received while this runs. It's
+ * unclear that recursive entry would be safe, and it doesn't seem useful
+ * even if it is safe, so let's block interrupts until done.
+ */
+ HOLD_INTERRUPTS();
+
+ /*
+ * Moreover, CurrentMemoryContext might be pointing almost anywhere. We
+ * don't want to risk leaking data into long-lived contexts, so let's do
+ * our work here in a private context that we can reset on each use.
+ */
+ if (hpm_context == NULL) /* first time through? */
+ hpm_context = AllocSetContextCreate(TopMemoryContext,
+ "ProcessRepackMessages",
+ ALLOCSET_DEFAULT_SIZES);
+ else
+ MemoryContextReset(hpm_context);
+
+ oldcontext = MemoryContextSwitchTo(hpm_context);
+
+ /* OK to process messages. Reset the flag saying there are more to do. */
+ RepackMessagePending = false;
+
+ /*
+ * Read as many messages as we can from each worker, but stop when no more
+ * messages can be read from the worker without blocking.
+ */
+ while (true)
+ {
+ shm_mq_result res;
+ Size nbytes;
+ void *data;
+
+ res = shm_mq_receive(decoding_worker->error_mqh, &nbytes,
+ &data, true);
+ if (res == SHM_MQ_WOULD_BLOCK)
+ break;
+ else if (res == SHM_MQ_SUCCESS)
+ {
+ StringInfoData msg;
+
+ initStringInfo(&msg);
+ appendBinaryStringInfo(&msg, data, nbytes);
+ ProcessRepackMessage(&msg);
+ pfree(msg.data);
+ }
+ else
+ {
+ /*
+ * The decoding worker is special in that it exits as soon as it
+ * has its work done. Thus the DETACHED result code is fine.
+ */
+ Assert(res == SHM_MQ_DETACHED);
+
+ break;
+ }
+ }
+
+ MemoryContextSwitchTo(oldcontext);
+
+ /* Might as well clear the context on our way out */
+ MemoryContextReset(hpm_context);
+
+ RESUME_INTERRUPTS();
+}
+
+/*
+ * Process a single protocol message received from a single parallel worker.
+ */
+static void
+ProcessRepackMessage(StringInfo msg)
+{
+ char msgtype;
+
+ msgtype = pq_getmsgbyte(msg);
+
+ switch (msgtype)
+ {
+ case PqMsg_ErrorResponse:
+ case PqMsg_NoticeResponse:
+ {
+ ErrorData edata;
+
+ /* Parse ErrorResponse or NoticeResponse. */
+ pq_parse_errornotice(msg, &edata);
+
+ /* Death of a worker isn't enough justification for suicide. */
+ edata.elevel = Min(edata.elevel, ERROR);
+
+ /*
+ * If desired, add a context line to show that this is a
+ * message propagated from a parallel worker. Otherwise, it
+ * can sometimes be confusing to understand what actually
+ * happened.
+ */
+ if (edata.context)
+ edata.context = psprintf("%s\n%s", edata.context,
+ _("decoding worker"));
+ else
+ edata.context = pstrdup(_("decoding worker"));
+
+ /* Rethrow error or print notice. */
+ ThrowErrorData(&edata);
+
+ break;
+ }
+
+ default:
+ {
+ elog(ERROR, "unrecognized message type received from decoding worker: %c (message length %d bytes)",
+ msgtype, msg->len);
+ }
+ }
+}
diff --git a/src/backend/libpq/pqmq.c b/src/backend/libpq/pqmq.c
index 6e4bbfb5aa1..42f6fa472c5 100644
--- a/src/backend/libpq/pqmq.c
+++ b/src/backend/libpq/pqmq.c
@@ -14,6 +14,7 @@
#include "postgres.h"
#include "access/parallel.h"
+#include "commands/cluster.h"
#include "libpq/libpq.h"
#include "libpq/pqformat.h"
#include "libpq/pqmq.h"
@@ -175,6 +176,10 @@ mq_putmessage(char msgtype, const char *s, size_t len)
SendProcSignal(pq_mq_parallel_leader_pid,
PROCSIG_PARALLEL_APPLY_MESSAGE,
pq_mq_parallel_leader_proc_number);
+ else if (IsRepackWorker())
+ SendProcSignal(pq_mq_parallel_leader_pid,
+ PROCSIG_REPACK_MESSAGE,
+ pq_mq_parallel_leader_proc_number);
else
{
Assert(IsParallelWorker());
diff --git a/src/backend/postmaster/bgworker.c b/src/backend/postmaster/bgworker.c
index 65deabe91a7..334bb708c5b 100644
--- a/src/backend/postmaster/bgworker.c
+++ b/src/backend/postmaster/bgworker.c
@@ -13,6 +13,7 @@
#include "postgres.h"
#include "access/parallel.h"
+#include "commands/cluster.h"
#include "libpq/pqsignal.h"
#include "miscadmin.h"
#include "pgstat.h"
@@ -136,6 +137,9 @@ static const struct
},
{
"SequenceSyncWorkerMain", SequenceSyncWorkerMain
+ },
+ {
+ "RepackWorkerMain", RepackWorkerMain
}
};
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index b0ef1a12520..35a46988285 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -194,7 +194,11 @@ StartupDecodingContext(List *output_plugin_options,
ctx->slot = slot;
- ctx->reader = XLogReaderAllocate(wal_segment_size, NULL, xl_routine, ctx);
+ /*
+ * TODO A separate patch for PG core, unless there's really a reason to
+ * pass ctx for private_data (May extensions expect ctx?).
+ */
+ ctx->reader = XLogReaderAllocate(wal_segment_size, NULL, xl_routine, NULL);
if (!ctx->reader)
ereport(ERROR,
(errcode(ERRCODE_OUT_OF_MEMORY),
diff --git a/src/backend/replication/pgoutput_repack/pgoutput_repack.c b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
index a074eed393d..4bc47a72371 100644
--- a/src/backend/replication/pgoutput_repack/pgoutput_repack.c
+++ b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
@@ -169,17 +169,13 @@ store_change(LogicalDecodingContext *ctx, ConcurrentChangeKind kind,
HeapTuple tuple)
{
RepackDecodingState *dstate;
- char *change_raw;
- ConcurrentChange change;
+ char kind_byte = (char) kind;
bool flattened = false;
- Size size;
- Datum values[1];
- bool isnull[1];
- char *dst;
dstate = (RepackDecodingState *) ctx->output_writer_private;
- size = VARHDRSZ + SizeOfConcurrentChange;
+ /* Store the change kind. */
+ BufFileWrite(dstate->file, &kind_byte, 1);
/*
* ReorderBufferCommit() stores the TOAST chunks in its private memory
@@ -196,46 +192,12 @@ store_change(LogicalDecodingContext *ctx, ConcurrentChangeKind kind,
tuple = toast_flatten_tuple(tuple, dstate->tupdesc);
flattened = true;
}
+ /* Store the tuple size ... */
+ BufFileWrite(dstate->file, &tuple->t_len, sizeof(tuple->t_len));
+ /* ... and the tuple itself. */
+ BufFileWrite(dstate->file, tuple->t_data, tuple->t_len);
- size += tuple->t_len;
- if (size >= MaxAllocSize)
- elog(ERROR, "Change is too big.");
-
- /* Construct the change. */
- change_raw = (char *) palloc0(size);
- SET_VARSIZE(change_raw, size);
-
- /*
- * Since the varlena alignment might not be sufficient for the structure,
- * set the fields in a local instance and remember where it should
- * eventually be copied.
- */
- change.kind = kind;
- dst = (char *) VARDATA(change_raw);
-
- /*
- * Copy the tuple.
- *
- * Note: change->tup_data.t_data must be fixed on retrieval!
- */
- memcpy(&change.tup_data, tuple, sizeof(HeapTupleData));
- memcpy(dst, &change, SizeOfConcurrentChange);
- dst += SizeOfConcurrentChange;
- memcpy(dst, tuple->t_data, tuple->t_len);
-
- /* The data has been copied. */
+ /* Free the flat copy if created above. */
if (flattened)
pfree(tuple);
-
- /* Store as tuple of 1 bytea column. */
- values[0] = PointerGetDatum(change_raw);
- isnull[0] = false;
- tuplestore_putvalues(dstate->tstore, dstate->tupdesc_change,
- values, isnull);
-
- /* Accounting. */
- dstate->nchanges++;
-
- /* Cleanup. */
- pfree(change_raw);
}
diff --git a/src/backend/storage/ipc/procsignal.c b/src/backend/storage/ipc/procsignal.c
index 8e56922dcea..6f9e7a7aab7 100644
--- a/src/backend/storage/ipc/procsignal.c
+++ b/src/backend/storage/ipc/procsignal.c
@@ -19,6 +19,7 @@
#include "access/parallel.h"
#include "commands/async.h"
+#include "commands/cluster.h"
#include "miscadmin.h"
#include "pgstat.h"
#include "port/pg_bitutils.h"
@@ -697,6 +698,9 @@ procsignal_sigusr1_handler(SIGNAL_ARGS)
if (CheckProcSignal(PROCSIG_PARALLEL_APPLY_MESSAGE))
HandleParallelApplyMessageInterrupt();
+ if (CheckProcSignal(PROCSIG_REPACK_MESSAGE))
+ HandleRepackMessageInterrupt();
+
if (CheckProcSignal(PROCSIG_RECOVERY_CONFLICT_DATABASE))
HandleRecoveryConflictInterrupt(PROCSIG_RECOVERY_CONFLICT_DATABASE);
diff --git a/src/backend/tcop/postgres.c b/src/backend/tcop/postgres.c
index e54bf1e760f..fc81ad87615 100644
--- a/src/backend/tcop/postgres.c
+++ b/src/backend/tcop/postgres.c
@@ -36,6 +36,7 @@
#include "access/xact.h"
#include "catalog/pg_type.h"
#include "commands/async.h"
+#include "commands/cluster.h"
#include "commands/event_trigger.h"
#include "commands/explain_state.h"
#include "commands/prepare.h"
@@ -3541,6 +3542,9 @@ ProcessInterrupts(void)
if (ParallelApplyMessagePending)
ProcessParallelApplyMessages();
+
+ if (RepackMessagePending)
+ ProcessRepackMessages();
}
/*
diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt
index 5537a2d2530..56d01ccd5af 100644
--- a/src/backend/utils/activity/wait_event_names.txt
+++ b/src/backend/utils/activity/wait_event_names.txt
@@ -154,6 +154,7 @@ RECOVERY_CONFLICT_SNAPSHOT "Waiting for recovery conflict resolution for a vacuu
RECOVERY_CONFLICT_TABLESPACE "Waiting for recovery conflict resolution for dropping a tablespace."
RECOVERY_END_COMMAND "Waiting for <xref linkend="guc-recovery-end-command"/> to complete."
RECOVERY_PAUSE "Waiting for recovery to be resumed."
+REPACK_WORKER_EXPORT "Waiting for decoding worker to export a new output file."
REPLICATION_ORIGIN_DROP "Waiting for a replication origin to become inactive so it can be dropped."
REPLICATION_SLOT_DROP "Waiting for a replication slot to become inactive so it can be dropped."
RESTORE_COMMAND "Waiting for <xref linkend="guc-restore-command"/> to complete."
diff --git a/src/include/access/tableam.h b/src/include/access/tableam.h
index 76aa993009a..15760363a1a 100644
--- a/src/include/access/tableam.h
+++ b/src/include/access/tableam.h
@@ -22,7 +22,6 @@
#include "access/xact.h"
#include "commands/vacuum.h"
#include "executor/tuptable.h"
-#include "replication/logical.h"
#include "storage/read_stream.h"
#include "utils/rel.h"
#include "utils/snapshot.h"
@@ -631,7 +630,6 @@ typedef struct TableAmRoutine
bool use_sort,
TransactionId OldestXmin,
Snapshot snapshot,
- LogicalDecodingContext *decoding_ctx,
TransactionId *xid_cutoff,
MultiXactId *multi_cutoff,
double *num_tuples,
@@ -1651,8 +1649,6 @@ table_relation_copy_data(Relation rel, const RelFileLocator *newrlocator)
* - *multi_cutoff - ditto
* - snapshot - if != NULL, ignore data changes done by transactions that this
* (MVCC) snapshot considers still in-progress or in the future.
- * - decoding_ctx - logical decoding context, to capture concurrent data
- * changes.
*
* Output parameters:
* - *xid_cutoff - rel's new relfrozenxid value, may be invalid
@@ -1666,7 +1662,6 @@ table_relation_copy_for_cluster(Relation OldTable, Relation NewTable,
bool use_sort,
TransactionId OldestXmin,
Snapshot snapshot,
- LogicalDecodingContext *decoding_ctx,
TransactionId *xid_cutoff,
MultiXactId *multi_cutoff,
double *num_tuples,
@@ -1675,7 +1670,7 @@ table_relation_copy_for_cluster(Relation OldTable, Relation NewTable,
{
OldTable->rd_tableam->relation_copy_for_cluster(OldTable, NewTable, OldIndex,
use_sort, OldestXmin,
- snapshot, decoding_ctx,
+ snapshot,
xid_cutoff, multi_cutoff,
num_tuples, tups_vacuumed,
tups_recently_dead);
diff --git a/src/include/commands/cluster.h b/src/include/commands/cluster.h
index 6a5c476294a..1b05d5d418b 100644
--- a/src/include/commands/cluster.h
+++ b/src/include/commands/cluster.h
@@ -17,11 +17,13 @@
#include "nodes/parsenodes.h"
#include "parser/parse_node.h"
#include "replication/decode.h"
+#include "postmaster/bgworker.h"
#include "replication/logical.h"
+#include "storage/buffile.h"
#include "storage/lock.h"
+#include "storage/shm_mq.h"
#include "utils/relcache.h"
#include "utils/resowner.h"
-#include "utils/tuplestore.h"
/* flag bits for ClusterParams->options */
@@ -44,6 +46,9 @@ typedef struct ClusterParams
* The following definitions are used by REPACK CONCURRENTLY.
*/
+/*
+ * Stored as a single byte in the output file.
+ */
typedef enum
{
CHANGE_INSERT,
@@ -52,68 +57,30 @@ typedef enum
CHANGE_DELETE
} ConcurrentChangeKind;
-typedef struct ConcurrentChange
-{
- /* See the enum above. */
- ConcurrentChangeKind kind;
-
- /*
- * The actual tuple.
- *
- * The tuple data follows the ConcurrentChange structure. Before use make
- * sure the tuple is correctly aligned (ConcurrentChange can be stored as
- * bytea) and that tuple->t_data is fixed.
- */
- HeapTupleData tup_data;
-} ConcurrentChange;
-
-#define SizeOfConcurrentChange (offsetof(ConcurrentChange, tup_data) + \
- sizeof(HeapTupleData))
-
/*
* Logical decoding state.
*
- * Here we store the data changes that we decode from WAL while the table
- * contents is being copied to a new storage. Also the necessary metadata
- * needed to apply these changes to the table is stored here.
+ * The output plugin uses it to store the data changes that it decodes from
+ * WAL while the table contents is being copied to a new storage.
*/
typedef struct RepackDecodingState
{
/* The relation whose changes we're decoding. */
Oid relid;
- /* Replication slot name. */
- NameData slotname;
-
- /*
- * Decoded changes are stored here. Although we try to avoid excessive
- * batches, it can happen that the changes need to be stored to disk. The
- * tuplestore does this transparently.
- */
- Tuplestorestate *tstore;
-
- /* The current number of changes in tstore. */
- double nchanges;
-
- /*
- * Descriptor to store the ConcurrentChange structure serialized (bytea).
- * We can't store the tuple directly because tuplestore only supports
- * minimum tuple and we may need to transfer OID system column from the
- * output plugin. Also we need to transfer the change kind, so it's better
- * to put everything in the structure than to use 2 tuplestores "in
- * parallel".
- */
- TupleDesc tupdesc_change;
-
- /* Tuple descriptor needed to update indexes. */
+ /* Tuple descriptor of the relation being processed. */
TupleDesc tupdesc;
- /* Slot to retrieve data from tstore. */
- TupleTableSlot *tsslot;
-
- ResourceOwner resowner;
+ /* The current output file. */
+ BufFile *file;
} RepackDecodingState;
+extern PGDLLIMPORT volatile sig_atomic_t RepackMessagePending;
+
+extern bool IsRepackWorker(void);
+extern void HandleRepackMessageInterrupt(void);
+extern void ProcessRepackMessages(void);
+
extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel);
extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid,
@@ -136,6 +103,6 @@ extern void finish_heap_swap(Oid OIDOldHeap, Oid OIDNewHeap,
extern bool am_decoding_for_repack(void);
extern bool change_useless_for_repack(XLogRecordBuffer *buf);
-extern void repack_decode_concurrent_changes(LogicalDecodingContext *ctx,
- XLogRecPtr end_of_wal);
+
+extern void RepackWorkerMain(Datum main_arg);
#endif /* CLUSTER_H */
diff --git a/src/include/storage/procsignal.h b/src/include/storage/procsignal.h
index e52b8eb7697..3ef35ca6b80 100644
--- a/src/include/storage/procsignal.h
+++ b/src/include/storage/procsignal.h
@@ -36,6 +36,7 @@ typedef enum
PROCSIG_BARRIER, /* global barrier interrupt */
PROCSIG_LOG_MEMORY_CONTEXT, /* ask backend to log the memory contexts */
PROCSIG_PARALLEL_APPLY_MESSAGE, /* Message from parallel apply workers */
+ PROCSIG_REPACK_MESSAGE, /* Message from repack worker */
/* Recovery conflict reasons */
PROCSIG_RECOVERY_CONFLICT_FIRST,
diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list
index 556128e9c57..3cae074c7e7 100644
--- a/src/tools/pgindent/typedefs.list
+++ b/src/tools/pgindent/typedefs.list
@@ -498,7 +498,6 @@ CompressFileHandle
CompressionLocation
CompressorState
ComputeXidHorizonsResult
-ConcurrentChange
ConcurrentChangeKind
ConditionVariable
ConditionVariableMinimallyPadded
@@ -638,6 +637,8 @@ DeclareCursorStmt
DecodedBkpBlock
DecodedXLogRecord
DecodingOutputState
+DecodingWorker
+DecodingWorkerShared
DefElem
DefElemAction
DefaultACLInfo
--
2.47.3
--=-=-=
Content-Type: text/plain
Content-Disposition: attachment;
filename=v31-0006-Use-multiple-snapshots-to-copy-the-data.patch
^ permalink raw reply [nested|flat] 55+ messages in thread
* [PATCH 5/6] Use background worker to do logical decoding.
@ 2026-01-27 10:48 Antonin Houska <[email protected]>
0 siblings, 0 replies; 55+ messages in thread
From: Antonin Houska @ 2026-01-27 10:48 UTC (permalink / raw)
If the backend performing REPACK (CONCURRENTLY) does both data copying and
logical decoding, it has to "travel in time" back and forth and therefore it
has to invalidate system caches quite a few times. (The copying and the
decoding work with different catalog snapshots.) As the decoding worker has
separate caches, the switching is not necessary.
Without the worker, it'd also be difficult to switch between potentially long
running tasks like index build and WAL decoding. (No decoding during that time
at all can suspend archiving / recycling of WAL segments for some time, which
in turn may result in full disk.)
Another problem is that, after having acquired AccessExclusiveLock (in order
to swap the files), the backend needs to both decode and apply the data
changes that took place while it was waiting for the lock. With the decoding
worker, the decoding runs all the time, so the backend only needs to apply the
changes. This can reduce the time the exclusive lock is held for.
Note that the code added in order to handle ERRORs in the background worker
almost duplicates the existing code that does the same for other types of
workers (See ProcessParallelMessages() and
ProcessParallelApplyMessages()). Refactoring of the existing code might be
useful, to reduce the duplication.
---
src/backend/access/heap/heapam_handler.c | 44 -
src/backend/commands/cluster.c | 1179 +++++++++++++----
src/backend/libpq/pqmq.c | 5 +
src/backend/postmaster/bgworker.c | 4 +
src/backend/replication/logical/logical.c | 6 +-
.../pgoutput_repack/pgoutput_repack.c | 54 +-
src/backend/storage/ipc/procsignal.c | 4 +
src/backend/tcop/postgres.c | 4 +
.../utils/activity/wait_event_names.txt | 1 +
src/include/access/tableam.h | 7 +-
src/include/commands/cluster.h | 71 +-
src/include/storage/procsignal.h | 1 +
src/tools/pgindent/typedefs.list | 3 +-
13 files changed, 981 insertions(+), 402 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index 908f1ef66c6..8589b3c940e 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -33,7 +33,6 @@
#include "catalog/index.h"
#include "catalog/storage.h"
#include "catalog/storage_xlog.h"
-#include "commands/cluster.h"
#include "commands/progress.h"
#include "executor/executor.h"
#include "miscadmin.h"
@@ -688,7 +687,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
Relation OldIndex, bool use_sort,
TransactionId OldestXmin,
Snapshot snapshot,
- LogicalDecodingContext *decoding_ctx,
TransactionId *xid_cutoff,
MultiXactId *multi_cutoff,
double *num_tuples,
@@ -710,7 +708,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
BufferHeapTupleTableSlot *hslot;
BlockNumber prev_cblock = InvalidBlockNumber;
bool concurrent = snapshot != NULL;
- XLogRecPtr end_of_wal_prev = GetFlushRecPtr(NULL);
/* Remember if it's a system catalog */
is_system_catalog = IsSystemRelation(OldHeap);
@@ -971,31 +968,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
ct_val[1] = *num_tuples;
pgstat_progress_update_multi_param(2, ct_index, ct_val);
}
-
- /*
- * Process the WAL produced by the load, as well as by other
- * transactions, so that the replication slot can advance and WAL does
- * not pile up. Use wal_segment_size as a threshold so that we do not
- * introduce the decoding overhead too often.
- *
- * Of course, we must not apply the changes until the initial load has
- * completed.
- *
- * Note that our insertions into the new table should not be decoded
- * as we (intentionally) do not write the logical decoding specific
- * information to WAL.
- */
- if (concurrent)
- {
- XLogRecPtr end_of_wal;
-
- end_of_wal = GetFlushRecPtr(NULL);
- if ((end_of_wal - end_of_wal_prev) > wal_segment_size)
- {
- repack_decode_concurrent_changes(decoding_ctx, end_of_wal);
- end_of_wal_prev = end_of_wal;
- }
- }
}
if (indexScan != NULL)
@@ -1041,22 +1013,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
-
- /*
- * Try to keep the amount of not-yet-decoded WAL small, like
- * above.
- */
- if (concurrent)
- {
- XLogRecPtr end_of_wal;
-
- end_of_wal = GetFlushRecPtr(NULL);
- if ((end_of_wal - end_of_wal_prev) > wal_segment_size)
- {
- repack_decode_concurrent_changes(decoding_ctx, end_of_wal);
- end_of_wal_prev = end_of_wal;
- }
- }
}
tuplesort_end(tuplesort);
diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c
index 03ccf10b782..e988a7a7296 100644
--- a/src/backend/commands/cluster.c
+++ b/src/backend/commands/cluster.c
@@ -12,12 +12,13 @@
* In concurrent mode, we lock the table with only ShareUpdateExclusiveLock,
* then do an initial copy as above. However, while the tuples are being
* copied, concurrent transactions could modify the table. To cope with those
- * changes, we rely on logical decoding to obtain them from WAL. The changes
- * are accumulated in a tuplestore. Once the initial copy is complete, we
- * read the changes from the tuplestore and re-apply them on the new heap.
- * Then we upgrade our ShareUpdateExclusiveLock to AccessExclusiveLock and
- * swap the relfilenodes. This way, the time we hold a strong lock on the
- * table is much reduced, and the bloat is eliminated.
+ * changes, we rely on logical decoding to obtain them from WAL. A bgworker
+ * consumes WAL while the initial copy is ongoing (to prevent excessive WAL
+ * from being reserved), and accumulates the changes in a file. Once the
+ * initial copy is complete, we read the changes from the file and re-apply
+ * them on the new heap. Then we upgrade our ShareUpdateExclusiveLock to
+ * AccessExclusiveLock and swap the relfilenodes. This way, the time we hold
+ * a strong lock on the table is much reduced, and the bloat is eliminated.
*
* There is hardly anything left of Paul Brown's original implementation...
*
@@ -45,6 +46,7 @@
#include "access/xlog_internal.h"
#include "access/xloginsert.h"
#include "access/xlogutils.h"
+#include "access/xlogwait.h"
#include "catalog/catalog.h"
#include "catalog/dependency.h"
#include "catalog/heap.h"
@@ -61,6 +63,8 @@
#include "commands/tablecmds.h"
#include "commands/vacuum.h"
#include "executor/executor.h"
+#include "libpq/pqformat.h"
+#include "libpq/pqmq.h"
#include "miscadmin.h"
#include "optimizer/optimizer.h"
#include "pgstat.h"
@@ -71,6 +75,8 @@
#include "storage/ipc.h"
#include "storage/lmgr.h"
#include "storage/predicate.h"
+#include "storage/procsignal.h"
+#include "tcop/tcopprot.h"
#include "utils/acl.h"
#include "utils/fmgroids.h"
#include "utils/guc.h"
@@ -117,6 +123,12 @@ typedef struct IndexInsertState
/* The WAL segment being decoded. */
static XLogSegNo repack_current_segment = 0;
+/*
+ * The first file exported by the decoding worker must contain a snapshot, the
+ * following ones contain the data changes.
+ */
+#define WORKER_FILE_SNAPSHOT 0
+
/*
* Information needed to apply concurrent data changes.
*/
@@ -136,8 +148,113 @@ typedef struct ChangeDest
/* Needed to update indexes of rel_dst. */
IndexInsertState *iistate;
+
+ /*
+ * Sequential number of the file containing the changes.
+ *
+ * TODO This field makes the structure name less descriptive. Should we
+ * rename it, e.g. to ChangeApplyInfo?
+ */
+ int file_seq;
} ChangeDest;
+/*
+ * Layout of shared memory used for communication between backend and the
+ * worker that performs logical decoding of data changes
+ */
+typedef struct DecodingWorkerShared
+{
+ /* Is the decoding initialized? */
+ bool initialized;
+
+ /*
+ * Once the worker has reached this LSN, it should close the current
+ * output file and either create a new one or exit, according to the field
+ * 'done'. If the value is InvalidXLogRecPtr, the worker should decode all
+ * the WAL available and keep checking this field. It is ok if the worker
+ * had already decoded records whose LSN is >= lsn_upto before this field
+ * has been set.
+ */
+ XLogRecPtr lsn_upto;
+
+ /* Exit after closing the current file? */
+ bool done;
+
+ /* The output is stored here. */
+ SharedFileSet sfs;
+
+ /* Number of the last file exported by the worker. */
+ int last_exported;
+
+ /* Synchronize access to the fields above. */
+ slock_t mutex;
+
+ /* Database to connect to. */
+ Oid dbid;
+
+ /* Role to connect as. */
+ Oid roleid;
+
+ /* Decode data changes of this relation. */
+ Oid relid;
+
+ /* The backend uses this to wait for the worker. */
+ ConditionVariable cv;
+
+ /* Info to signal the backend. */
+ PGPROC *backend_proc;
+ pid_t backend_pid;
+ ProcNumber backend_proc_number;
+
+ /* Error queue. */
+ shm_mq *error_mq;
+
+ /*
+ * Memory the queue is located in.
+ *
+ * For considerations on the value see the comments of
+ * PARALLEL_ERROR_QUEUE_SIZE.
+ */
+#define REPACK_ERROR_QUEUE_SIZE 16384
+ char error_queue[FLEXIBLE_ARRAY_MEMBER];
+} DecodingWorkerShared;
+
+/*
+ * Generate worker's output file name. If relations of the same 'relid' happen
+ * to be processed at the same time, they must be from different databases and
+ * therefore different backends must be involved. (PID is already present in
+ * the fileset name.)
+ */
+static inline void
+DecodingWorkerFileName(char *fname, Oid relid, uint32 seq)
+{
+ snprintf(fname, MAXPGPATH, "%u-%u", relid, seq);
+}
+
+/*
+ * Backend-local information to control the decoding worker.
+ */
+typedef struct DecodingWorker
+{
+ /* The worker. */
+ BackgroundWorkerHandle *handle;
+
+ /* DecodingWorkerShared is in this segment. */
+ dsm_segment *seg;
+
+ /* Handle of the error queue. */
+ shm_mq_handle *error_mqh;
+} DecodingWorker;
+
+/* Pointer to currently running decoding worker. */
+static DecodingWorker *decoding_worker = NULL;
+
+/*
+ * Is there a message sent by a repack worker that the backend needs to
+ * receive?
+ */
+volatile sig_atomic_t RepackMessagePending = false;
+
static bool cluster_rel_recheck(RepackCommand cmd, Relation OldHeap,
Oid indexOid, Oid userid, LOCKMODE lmode,
int options);
@@ -145,7 +262,7 @@ static void check_repack_concurrently_requirements(Relation rel);
static void rebuild_relation(Relation OldHeap, Relation index, bool verbose,
bool concurrent);
static void copy_table_data(Relation NewHeap, Relation OldHeap, Relation OldIndex,
- Snapshot snapshot, LogicalDecodingContext *decoding_ctx,
+ Snapshot snapshot,
bool verbose,
bool *pSwapToastByContent,
TransactionId *pFreezeXid,
@@ -158,12 +275,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd,
static bool cluster_is_permitted_for_relation(RepackCommand cmd,
Oid relid, Oid userid);
-static void begin_concurrent_repack(Relation rel);
-static void end_concurrent_repack(void);
static LogicalDecodingContext *setup_logical_decoding(Oid relid);
-static HeapTuple get_changed_tuple(char *change);
-static void apply_concurrent_changes(RepackDecodingState *dstate,
- ChangeDest *dest);
+static bool decode_concurrent_changes(LogicalDecodingContext *ctx,
+ DecodingWorkerShared *shared);
+static void apply_concurrent_changes(BufFile *file, ChangeDest *dest);
static void apply_concurrent_insert(Relation rel, HeapTuple tup,
IndexInsertState *iistate,
TupleTableSlot *index_slot);
@@ -175,9 +290,9 @@ static void apply_concurrent_delete(Relation rel, HeapTuple tup_target);
static HeapTuple find_target_tuple(Relation rel, ChangeDest *dest,
HeapTuple tup_key,
TupleTableSlot *ident_slot);
-static void process_concurrent_changes(LogicalDecodingContext *decoding_ctx,
- XLogRecPtr end_of_wal,
- ChangeDest *dest);
+static void process_concurrent_changes(XLogRecPtr end_of_wal,
+ ChangeDest *dest,
+ bool done);
static IndexInsertState *get_index_insert_state(Relation relation,
Oid ident_index_id,
Relation *ident_index_p);
@@ -186,7 +301,6 @@ static ScanKey build_identity_key(Oid ident_idx_oid, Relation rel_src,
static void free_index_insert_state(IndexInsertState *iistate);
static void cleanup_logical_decoding(LogicalDecodingContext *ctx);
static void rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
- LogicalDecodingContext *decoding_ctx,
TransactionId frozenXid,
MultiXactId cutoffMulti);
static List *build_new_indexes(Relation NewHeap, Relation OldHeap, List *OldIndexes);
@@ -196,6 +310,13 @@ static Relation process_single_relation(RepackStmt *stmt,
ClusterParams *params);
static Oid determine_clustered_index(Relation rel, bool usingindex,
const char *indexname);
+static void start_decoding_worker(Oid relid);
+static void stop_decoding_worker(void);
+static void repack_worker_internal(dsm_segment *seg);
+static void export_initial_snapshot(Snapshot snapshot,
+ DecodingWorkerShared *shared);
+static Snapshot get_initial_snapshot(DecodingWorker *worker);
+static void ProcessRepackMessage(StringInfo msg);
static const char *RepackCommandAsString(RepackCommand cmd);
@@ -619,20 +740,20 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid,
/* rebuild_relation does all the dirty work */
PG_TRY();
{
- /*
- * For concurrent processing, make sure that our logical decoding
- * ignores data changes of other tables than the one we are
- * processing.
- */
- if (concurrent)
- begin_concurrent_repack(OldHeap);
-
rebuild_relation(OldHeap, index, verbose, concurrent);
}
PG_FINALLY();
{
if (concurrent)
- end_concurrent_repack();
+ {
+ /*
+ * Since during normal operation the worker was already asked to
+ * exit, stopping it explicitly is especially important on ERROR.
+ * However it still seems a good practice to make sure that the
+ * worker never survives the REPACK command.
+ */
+ stop_decoding_worker();
+ }
}
PG_END_TRY();
@@ -929,7 +1050,6 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
bool swap_toast_by_content;
TransactionId frozenXid;
MultiXactId cutoffMulti;
- LogicalDecodingContext *decoding_ctx = NULL;
Snapshot snapshot = NULL;
#if USE_ASSERT_CHECKING
LOCKMODE lmode;
@@ -943,19 +1063,36 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
if (concurrent)
{
/*
- * Prepare to capture the concurrent data changes.
+ * The worker needs to be member of the locking group we're the leader
+ * of. We ought to become the leader before the worker starts. The
+ * worker will join the group as soon as it starts.
*
- * Note that this call waits for all transactions with XID already
- * assigned to finish. If some of those transactions is waiting for a
- * lock conflicting with ShareUpdateExclusiveLock on our table (e.g.
- * it runs CREATE INDEX), we can end up in a deadlock. Not sure this
- * risk is worth unlocking/locking the table (and its clustering
- * index) and checking again if its still eligible for REPACK
- * CONCURRENTLY.
+ * This is to make sure that the deadlock described below is
+ * detectable by deadlock.c: if the worker waits for a transaction to
+ * complete and we are waiting for the worker output, then effectively
+ * we (i.e. this backend) are waiting for that transaction.
*/
- decoding_ctx = setup_logical_decoding(tableOid);
+ BecomeLockGroupLeader();
+
+ /*
+ * Start the worker that decodes data changes applied while we're
+ * copying the table contents.
+ *
+ * Note that the worker has to wait for all transactions with XID
+ * already assigned to finish. If some of those transactions is
+ * waiting for a lock conflicting with ShareUpdateExclusiveLock on our
+ * table (e.g. it runs CREATE INDEX), we can end up in a deadlock.
+ * Not sure this risk is worth unlocking/locking the table (and its
+ * clustering index) and checking again if it's still eligible for
+ * REPACK CONCURRENTLY.
+ */
+ start_decoding_worker(tableOid);
+
+ /*
+ * Wait until the worker has the initial snapshot and retrieve it.
+ */
+ snapshot = get_initial_snapshot(decoding_worker);
- snapshot = SnapBuildInitialSnapshotForRepack(decoding_ctx->snapshot_builder);
PushActiveSnapshot(snapshot);
}
@@ -980,7 +1117,7 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
NewHeap = table_open(OIDNewHeap, NoLock);
/* Copy the heap data into the new table in the desired order */
- copy_table_data(NewHeap, OldHeap, index, snapshot, decoding_ctx, verbose,
+ copy_table_data(NewHeap, OldHeap, index, snapshot, verbose,
&swap_toast_by_content, &frozenXid, &cutoffMulti);
/* The historic snapshot won't be needed anymore. */
@@ -1001,14 +1138,11 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
if (index)
index_close(index, NoLock);
- rebuild_relation_finish_concurrent(NewHeap, OldHeap, decoding_ctx,
- frozenXid, cutoffMulti);
+ rebuild_relation_finish_concurrent(NewHeap, OldHeap, frozenXid,
+ cutoffMulti);
pgstat_progress_update_param(PROGRESS_REPACK_PHASE,
PROGRESS_REPACK_PHASE_FINAL_CLEANUP);
-
- /* Done with decoding. */
- cleanup_logical_decoding(decoding_ctx);
}
else
{
@@ -1179,8 +1313,7 @@ make_new_heap(Oid OIDOldHeap, Oid NewTableSpace, Oid NewAccessMethod,
*/
static void
copy_table_data(Relation NewHeap, Relation OldHeap, Relation OldIndex,
- Snapshot snapshot, LogicalDecodingContext *decoding_ctx,
- bool verbose, bool *pSwapToastByContent,
+ Snapshot snapshot, bool verbose, bool *pSwapToastByContent,
TransactionId *pFreezeXid, MultiXactId *pCutoffMulti)
{
Relation relRelation;
@@ -1341,7 +1474,6 @@ copy_table_data(Relation NewHeap, Relation OldHeap, Relation OldIndex,
*/
table_relation_copy_for_cluster(OldHeap, NewHeap, OldIndex, use_sort,
cutoffs.OldestXmin, snapshot,
- decoding_ctx,
&cutoffs.FreezeLimit,
&cutoffs.MultiXactCutoff,
&num_tuples, &tups_vacuumed,
@@ -2371,62 +2503,10 @@ RepackCommandAsString(RepackCommand cmd)
case REPACK_COMMAND_CLUSTER:
return "CLUSTER";
}
- return "???"; /* keep compiler quiet */
+ return "???"; /* keep compiler quiet */
}
-/*
- * Call this function before REPACK CONCURRENTLY starts to setup logical
- * decoding. It makes sure that other users of the table put enough
- * information into WAL.
- *
- * The point is that at various places we expect that the table we're
- * processing is treated like a system catalog. For example, we need to be
- * able to scan it using a "historic snapshot" anytime during the processing
- * (as opposed to scanning only at the start point of the decoding, as logical
- * replication does during initial table synchronization), in order to apply
- * concurrent UPDATE / DELETE commands.
- *
- * Note that TOAST table needs no attention here as it's not scanned using
- * historic snapshot.
- */
-static void
-begin_concurrent_repack(Relation rel)
-{
- Oid toastrelid;
-
- /*
- * Avoid logical decoding of other relations by this backend. The lock we
- * have guarantees that the actual locator cannot be changed concurrently:
- * TRUNCATE needs AccessExclusiveLock.
- */
- Assert(CheckRelationLockedByMe(rel, ShareUpdateExclusiveLock, false));
- repacked_rel_locator = rel->rd_locator;
- toastrelid = rel->rd_rel->reltoastrelid;
- if (OidIsValid(toastrelid))
- {
- Relation toastrel;
-
- /* Avoid logical decoding of other TOAST relations. */
- toastrel = table_open(toastrelid, AccessShareLock);
- repacked_rel_toast_locator = toastrel->rd_locator;
- table_close(toastrel, AccessShareLock);
- }
-}
-
-/*
- * Call this when done with REPACK CONCURRENTLY.
- */
-static void
-end_concurrent_repack(void)
-{
- /*
- * Restore normal function of (future) logical decoding for this backend.
- */
- repacked_rel_locator.relNumber = InvalidOid;
- repacked_rel_toast_locator.relNumber = InvalidOid;
-}
-
/*
* Is this backend performing logical decoding on behalf of REPACK
* (CONCURRENTLY) ?
@@ -2491,9 +2571,10 @@ static LogicalDecodingContext *
setup_logical_decoding(Oid relid)
{
Relation rel;
- TupleDesc tupdesc;
+ Oid toastrelid;
LogicalDecodingContext *ctx;
- RepackDecodingState *dstate = palloc0_object(RepackDecodingState);
+ NameData slotname;
+ RepackDecodingState *dstate;
/*
* REPACK CONCURRENTLY is not allowed in a transaction block, so this
@@ -2501,20 +2582,21 @@ setup_logical_decoding(Oid relid)
*/
Assert(!TransactionIdIsValid(GetTopTransactionIdIfAny()));
- /*
- * A single backend should not execute multiple REPACK commands at a time,
- * so use PID to make the slot unique.
- */
- snprintf(NameStr(dstate->slotname), NAMEDATALEN, "repack_%d", MyProcPid);
-
/*
* Make sure we can use logical decoding.
*/
CheckSlotPermissions();
CheckLogicalDecodingRequirements();
- /* RS_TEMPORARY so that the slot gets cleaned up on ERROR. */
- ReplicationSlotCreate(NameStr(dstate->slotname), true, RS_TEMPORARY,
- false, false, false);
+ /*
+ * A single backend should not execute multiple REPACK commands at a time,
+ * so use PID to make the slot unique.
+ *
+ * RS_TEMPORARY so that the slot gets cleaned up on ERROR.
+ */
+ snprintf(NameStr(slotname), NAMEDATALEN, "repack_%d", MyProcPid);
+ ReplicationSlotCreate(NameStr(slotname), true, RS_TEMPORARY, false, false,
+ false);
+
EnsureLogicalDecodingEnabled();
/*
@@ -2537,104 +2619,109 @@ setup_logical_decoding(Oid relid)
DecodingContextFindStartpoint(ctx);
+ /*
+ * decode_concurrent_changes() needs non-blocking callback.
+ */
+ ctx->reader->routine.page_read = read_local_xlog_page_no_wait;
+
+ /*
+ * read_local_xlog_page_no_wait() needs to be able to indicate the end of
+ * WAL.
+ */
+ ctx->reader->private_data = MemoryContextAllocZero(ctx->context,
+ sizeof(ReadLocalXLogPageNoWaitPrivate));
+
+
/* Some WAL records should have been read. */
Assert(ctx->reader->EndRecPtr != InvalidXLogRecPtr);
+ /*
+ * Initialize repack_current_segment so that we can notice WAL segment
+ * boundaries.
+ */
XLByteToSeg(ctx->reader->EndRecPtr, repack_current_segment,
wal_segment_size);
- /*
- * Setup structures to store decoded changes.
- */
+ dstate = palloc0_object(RepackDecodingState);
dstate->relid = relid;
- dstate->tstore = tuplestore_begin_heap(false, false,
- maintenance_work_mem);
- /* Caller should already have the table locked. */
- rel = table_open(relid, NoLock);
- tupdesc = CreateTupleDescCopy(RelationGetDescr(rel));
- dstate->tupdesc = tupdesc;
- table_close(rel, NoLock);
+ /*
+ * Tuple descriptor may be needed to flatten a tuple before we write it to
+ * a file. A copy is needed because the decoding worker invalidates system
+ * caches before it starts to do the actual work.
+ */
+ rel = table_open(relid, AccessShareLock);
+ dstate->tupdesc = CreateTupleDescCopy(RelationGetDescr(rel));
- /* Initialize the descriptor to store the changes ... */
- dstate->tupdesc_change = CreateTemplateTupleDesc(1);
+ /* Avoid logical decoding of other relations. */
+ repacked_rel_locator = rel->rd_locator;
+ toastrelid = rel->rd_rel->reltoastrelid;
+ if (OidIsValid(toastrelid))
+ {
+ Relation toastrel;
- TupleDescInitEntry(dstate->tupdesc_change, 1, NULL, BYTEAOID, -1, 0);
- /* ... as well as the corresponding slot. */
- dstate->tsslot = MakeSingleTupleTableSlot(dstate->tupdesc_change,
- &TTSOpsMinimalTuple);
+ /* Avoid logical decoding of other TOAST relations. */
+ toastrel = table_open(toastrelid, AccessShareLock);
+ repacked_rel_toast_locator = toastrel->rd_locator;
+ table_close(toastrel, AccessShareLock);
+ }
+ table_close(rel, AccessShareLock);
- dstate->resowner = ResourceOwnerCreate(CurrentResourceOwner,
- "logical decoding");
+ /* The file will be set as soon as we have it opened. */
+ dstate->file = NULL;
ctx->output_writer_private = dstate;
+
return ctx;
}
/*
- * Retrieve tuple from ConcurrentChange structure.
+ * Decode logical changes from the WAL sequence and store them to a file.
*
- * The input data starts with the structure but it might not be appropriately
- * aligned.
- */
-static HeapTuple
-get_changed_tuple(char *change)
-{
- HeapTupleData tup_data;
- HeapTuple result;
- char *src;
-
- /*
- * Ensure alignment before accessing the fields. (This is why we can't use
- * heap_copytuple() instead of this function.)
- */
- src = change + offsetof(ConcurrentChange, tup_data);
- memcpy(&tup_data, src, sizeof(HeapTupleData));
-
- result = (HeapTuple) palloc(HEAPTUPLESIZE + tup_data.t_len);
- memcpy(result, &tup_data, sizeof(HeapTupleData));
- result->t_data = (HeapTupleHeader) ((char *) result + HEAPTUPLESIZE);
- src = change + SizeOfConcurrentChange;
- memcpy(result->t_data, src, result->t_len);
-
- return result;
-}
-
-/*
- * Decode logical changes from the WAL sequence up to end_of_wal.
+ * If true is returned, there is no more work for the worker.
*/
-void
-repack_decode_concurrent_changes(LogicalDecodingContext *ctx,
- XLogRecPtr end_of_wal)
+static bool
+decode_concurrent_changes(LogicalDecodingContext *ctx,
+ DecodingWorkerShared *shared)
{
RepackDecodingState *dstate;
- ResourceOwner resowner_old;
+ XLogRecPtr lsn_upto;
+ bool done;
+ char fname[MAXPGPATH];
dstate = (RepackDecodingState *) ctx->output_writer_private;
- resowner_old = CurrentResourceOwner;
- CurrentResourceOwner = dstate->resowner;
- PG_TRY();
+ /* Open the output file. */
+ DecodingWorkerFileName(fname, shared->relid, shared->last_exported + 1);
+ dstate->file = BufFileCreateFileSet(&shared->sfs.fs, fname);
+
+ SpinLockAcquire(&shared->mutex);
+ lsn_upto = shared->lsn_upto;
+ done = shared->done;
+ SpinLockRelease(&shared->mutex);
+
+ while (true)
{
- while (ctx->reader->EndRecPtr < end_of_wal)
- {
- XLogRecord *record;
- XLogSegNo segno_new;
- char *errm = NULL;
- XLogRecPtr end_lsn;
+ XLogRecord *record;
+ XLogSegNo segno_new;
+ char *errm = NULL;
+ XLogRecPtr end_lsn;
- record = XLogReadRecord(ctx->reader, &errm);
- if (errm)
- elog(ERROR, "%s", errm);
+ CHECK_FOR_INTERRUPTS();
- if (record != NULL)
- LogicalDecodingProcessRecord(ctx, ctx->reader);
+ record = XLogReadRecord(ctx->reader, &errm);
+ if (record)
+ {
+ LogicalDecodingProcessRecord(ctx, ctx->reader);
/*
* If WAL segment boundary has been crossed, inform the decoding
- * system that the catalog_xmin can advance. (We can confirm more
- * often, but a filling a single WAL segment should not take much
- * time.)
+ * system that the catalog_xmin can advance.
+ *
+ * TODO Does it make sense to confirm more often? Segment size
+ * seems appropriate for restart_lsn (because less than a segment
+ * cannot be recycled anyway), however more frequent checks might
+ * be beneficial for catalog_xmin.
*/
end_lsn = ctx->reader->EndRecPtr;
XLByteToSeg(end_lsn, segno_new, wal_segment_size);
@@ -2645,80 +2732,137 @@ repack_decode_concurrent_changes(LogicalDecodingContext *ctx,
(uint32) (end_lsn >> 32), (uint32) end_lsn);
repack_current_segment = segno_new;
}
+ }
+ else
+ {
+ ReadLocalXLogPageNoWaitPrivate *priv;
+
+ if (errm)
+ ereport(ERROR, (errmsg("%s", errm)));
- CHECK_FOR_INTERRUPTS();
+ /*
+ * In the decoding loop we do not want to get blocked when there
+ * is no more WAL available, otherwise the loop would become
+ * uninterruptible.
+ */
+ priv = (ReadLocalXLogPageNoWaitPrivate *)
+ ctx->reader->private_data;
+ if (priv->end_of_wal)
+ /* Do not miss the end of WAL condition next time. */
+ priv->end_of_wal = false;
+ else
+ ereport(ERROR, (errmsg("could not read WAL record")));
+ }
+
+ /*
+ * Whether we could read new record or not, keep checking if
+ * 'lsn_upto' was specified.
+ */
+ if (XLogRecPtrIsInvalid(lsn_upto))
+ {
+ SpinLockAcquire(&shared->mutex);
+ lsn_upto = shared->lsn_upto;
+ /* 'done' should be set at the same time as 'lsn_upto' */
+ done = shared->done;
+ SpinLockRelease(&shared->mutex);
+ }
+ if (!XLogRecPtrIsInvalid(lsn_upto) &&
+ ctx->reader->EndRecPtr >= lsn_upto)
+ break;
+
+ if (record == NULL)
+ {
+ int64 timeout = 0;
+ WaitLSNResult res;
+
+ /*
+ * Before we retry reading, wait until new WAL is flushed.
+ *
+ * There is a race condition such that the backend executing
+ * REPACK determines 'lsn_upto', but before it sets the shared
+ * variable, we reach the end of WAL. In that case we'd need to
+ * wait until the next WAL flush (unrelated to REPACK). Although
+ * that should not be a problem in a busy system, it might be
+ * noticeable in other cases, including regression tests (which
+ * are not necessarily executed in parallel). Therefore it makes
+ * sense to use timeout.
+ *
+ * If lsn_upto is valid, WAL records having LSN lower than that
+ * should already have been flushed to disk.
+ */
+ if (XLogRecPtrIsInvalid(lsn_upto))
+ timeout = 100L;
+ res = WaitForLSN(WAIT_LSN_TYPE_PRIMARY_FLUSH,
+ ctx->reader->EndRecPtr + 1,
+ timeout);
+ if (res != WAIT_LSN_RESULT_SUCCESS &&
+ res != WAIT_LSN_RESULT_TIMEOUT)
+ ereport(ERROR, (errmsg("waiting for WAL failed")));
}
- InvalidateSystemCaches();
- CurrentResourceOwner = resowner_old;
- }
- PG_CATCH();
- {
- /* clear all timetravel entries */
- InvalidateSystemCaches();
- CurrentResourceOwner = resowner_old;
- PG_RE_THROW();
}
- PG_END_TRY();
+
+ /*
+ * Close the file so we can make it available to the backend.
+ */
+ BufFileClose(dstate->file);
+ dstate->file = NULL;
+ SpinLockAcquire(&shared->mutex);
+ shared->lsn_upto = InvalidXLogRecPtr;
+ shared->last_exported++;
+ SpinLockRelease(&shared->mutex);
+ ConditionVariableSignal(&shared->cv);
+
+ return done;
}
/*
* Apply changes stored in 'file'.
*/
static void
-apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
+apply_concurrent_changes(BufFile *file, ChangeDest *dest)
{
+ char kind;
+ uint32 t_len;
Relation rel = dest->rel;
TupleTableSlot *index_slot,
*ident_slot;
HeapTuple tup_old = NULL;
- if (dstate->nchanges == 0)
- return;
-
/* TupleTableSlot is needed to pass the tuple to ExecInsertIndexTuples(). */
- index_slot = MakeSingleTupleTableSlot(dstate->tupdesc, &TTSOpsHeapTuple);
+ index_slot = MakeSingleTupleTableSlot(RelationGetDescr(rel),
+ &TTSOpsHeapTuple);
/* A slot to fetch tuples from identity index. */
ident_slot = table_slot_create(rel, NULL);
- while (tuplestore_gettupleslot(dstate->tstore, true, false,
- dstate->tsslot))
+ while (true)
{
- bool shouldFree;
- HeapTuple tup_change,
- tup,
+ size_t nread;
+ HeapTuple tup,
tup_exist;
- char *change_raw,
- *src;
- ConcurrentChange change;
- bool isnull[1];
- Datum values[1];
CHECK_FOR_INTERRUPTS();
- /* Get the change from the single-column tuple. */
- tup_change = ExecFetchSlotHeapTuple(dstate->tsslot, false, &shouldFree);
- heap_deform_tuple(tup_change, dstate->tupdesc_change, values, isnull);
- Assert(!isnull[0]);
-
- /* Make sure we access aligned data. */
- change_raw = (char *) DatumGetByteaP(values[0]);
- src = (char *) VARDATA(change_raw);
- memcpy(&change, src, SizeOfConcurrentChange);
+ nread = BufFileReadMaybeEOF(file, &kind, 1, true);
+ /* Are we done with the file? */
+ if (nread == 0)
+ break;
- /*
- * Extract the tuple from the change. The tuple is copied here because
- * it might be assigned to 'tup_old', in which case it needs to
- * survive into the next iteration.
- */
- tup = get_changed_tuple(src);
+ /* Read the tuple. */
+ BufFileReadExact(file, &t_len, sizeof(t_len));
+ tup = (HeapTuple) palloc(HEAPTUPLESIZE + t_len);
+ tup->t_data = (HeapTupleHeader) ((char *) tup + HEAPTUPLESIZE);
+ BufFileReadExact(file, tup->t_data, t_len);
+ tup->t_len = t_len;
+ ItemPointerSetInvalid(&tup->t_self);
+ tup->t_tableOid = RelationGetRelid(dest->rel);
- if (change.kind == CHANGE_UPDATE_OLD)
+ if (kind == CHANGE_UPDATE_OLD)
{
Assert(tup_old == NULL);
tup_old = tup;
}
- else if (change.kind == CHANGE_INSERT)
+ else if (kind == CHANGE_INSERT)
{
Assert(tup_old == NULL);
@@ -2726,12 +2870,11 @@ apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
pfree(tup);
}
- else if (change.kind == CHANGE_UPDATE_NEW ||
- change.kind == CHANGE_DELETE)
+ else if (kind == CHANGE_UPDATE_NEW || kind == CHANGE_DELETE)
{
HeapTuple tup_key;
- if (change.kind == CHANGE_UPDATE_NEW)
+ if (kind == CHANGE_UPDATE_NEW)
{
tup_key = tup_old != NULL ? tup_old : tup;
}
@@ -2748,7 +2891,7 @@ apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
if (tup_exist == NULL)
elog(ERROR, "failed to find target tuple");
- if (change.kind == CHANGE_UPDATE_NEW)
+ if (kind == CHANGE_UPDATE_NEW)
apply_concurrent_update(rel, tup, tup_exist, dest->iistate,
index_slot);
else
@@ -2763,26 +2906,19 @@ apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
pfree(tup);
}
else
- elog(ERROR, "unrecognized kind of change: %d", change.kind);
+ elog(ERROR, "unrecognized kind of change: %d", kind);
/*
* If a change was applied now, increment CID for next writes and
* update the snapshot so it sees the changes we've applied so far.
*/
- if (change.kind != CHANGE_UPDATE_OLD)
+ if (kind != CHANGE_UPDATE_OLD)
{
CommandCounterIncrement();
UpdateActiveSnapshotCommandId();
}
-
- /* TTSOpsMinimalTuple has .get_heap_tuple==NULL. */
- Assert(shouldFree);
- pfree(tup_change);
}
- tuplestore_clear(dstate->tstore);
- dstate->nchanges = 0;
-
/* Cleanup. */
ExecDropSingleTupleTableSlot(index_slot);
ExecDropSingleTupleTableSlot(ident_slot);
@@ -2957,25 +3093,59 @@ find_target_tuple(Relation rel, ChangeDest *dest, HeapTuple tup_key,
}
/*
- * Decode and apply concurrent changes.
+ * Decode and apply concurrent changes, up to (and including) the record whose
+ * LSN is 'end_of_wal'.
*/
static void
-process_concurrent_changes(LogicalDecodingContext *decoding_ctx,
- XLogRecPtr end_of_wal, ChangeDest *dest)
+process_concurrent_changes(XLogRecPtr end_of_wal, ChangeDest *dest, bool done)
{
- RepackDecodingState *dstate;
+ DecodingWorkerShared *shared;
+ char fname[MAXPGPATH];
+ BufFile *file;
pgstat_progress_update_param(PROGRESS_REPACK_PHASE,
PROGRESS_REPACK_PHASE_CATCH_UP);
- dstate = (RepackDecodingState *) decoding_ctx->output_writer_private;
+ /* Ask the worker for the file. */
+ shared = (DecodingWorkerShared *) dsm_segment_address(decoding_worker->seg);
+ SpinLockAcquire(&shared->mutex);
+ shared->lsn_upto = end_of_wal;
+ shared->done = done;
+ SpinLockRelease(&shared->mutex);
+
+ /*
+ * The worker needs to finish processing of the current WAL record. Even
+ * if it's idle, it'll need to close the output file. Thus we're likely to
+ * wait, so prepare for sleep.
+ */
+ ConditionVariablePrepareToSleep(&shared->cv);
+ for (;;)
+ {
+ int last_exported;
- repack_decode_concurrent_changes(decoding_ctx, end_of_wal);
+ SpinLockAcquire(&shared->mutex);
+ last_exported = shared->last_exported;
+ SpinLockRelease(&shared->mutex);
- if (dstate->nchanges == 0)
- return;
+ /*
+ * Has the worker exported the file we are waiting for?
+ */
+ if (last_exported == dest->file_seq)
+ break;
- apply_concurrent_changes(dstate, dest);
+ ConditionVariableSleep(&shared->cv, WAIT_EVENT_REPACK_WORKER_EXPORT);
+ }
+ ConditionVariableCancelSleep();
+
+ /* Open the file. */
+ DecodingWorkerFileName(fname, shared->relid, dest->file_seq);
+ file = BufFileOpenFileSet(&shared->sfs.fs, fname, O_RDONLY, false);
+ apply_concurrent_changes(file, dest);
+
+ BufFileClose(file);
+
+ /* Get ready for the next file. */
+ dest->file_seq++;
}
/*
@@ -3101,15 +3271,10 @@ cleanup_logical_decoding(LogicalDecodingContext *ctx)
dstate = (RepackDecodingState *) ctx->output_writer_private;
- ExecDropSingleTupleTableSlot(dstate->tsslot);
- FreeTupleDesc(dstate->tupdesc_change);
FreeTupleDesc(dstate->tupdesc);
- tuplestore_end(dstate->tstore);
-
FreeDecodingContext(ctx);
- ReplicationSlotRelease();
- ReplicationSlotDrop(NameStr(dstate->slotname), false);
+ ReplicationSlotDropAcquired();
pfree(dstate);
}
@@ -3123,7 +3288,6 @@ cleanup_logical_decoding(LogicalDecodingContext *ctx)
*/
static void
rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
- LogicalDecodingContext *decoding_ctx,
TransactionId frozenXid,
MultiXactId cutoffMulti)
{
@@ -3204,6 +3368,7 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
&chgdst.ident_index);
chgdst.ident_key = build_identity_key(ident_idx_new, OldHeap,
&chgdst.ident_key_nentries);
+ chgdst.file_seq = WORKER_FILE_SNAPSHOT + 1;
/*
* During testing, wait for another backend to perform concurrent data
@@ -3225,7 +3390,7 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
* hold AccessExclusiveLock. (Quite some amount of WAL could have been
* written during the data copying and index creation.)
*/
- process_concurrent_changes(decoding_ctx, end_of_wal, &chgdst);
+ process_concurrent_changes(end_of_wal, &chgdst, false);
/*
* Acquire AccessExclusiveLock on the table, its TOAST relation (if there
@@ -3306,8 +3471,11 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
XLogFlush(wal_insert_ptr);
end_of_wal = GetFlushRecPtr(NULL);
- /* Apply the concurrent changes again. */
- process_concurrent_changes(decoding_ctx, end_of_wal, &chgdst);
+ /*
+ * Apply the concurrent changes again. Indicate that the decoding worker
+ * won't be needed anymore.
+ */
+ process_concurrent_changes(end_of_wal, &chgdst, true);
/* Remember info about rel before closing OldHeap */
relpersistence = OldHeap->rd_rel->relpersistence;
@@ -3417,3 +3585,510 @@ build_new_indexes(Relation NewHeap, Relation OldHeap, List *OldIndexes)
return result;
}
+
+/*
+ * Try to start a background worker to perform logical decoding of data
+ * changes applied to relation while REPACK CONCURRENTLY is copying its
+ * contents to a new table.
+ */
+static void
+start_decoding_worker(Oid relid)
+{
+ Size size;
+ dsm_segment *seg;
+ DecodingWorkerShared *shared;
+ shm_mq *mq;
+ shm_mq_handle *mqh;
+ BackgroundWorker bgw;
+
+ /* Setup shared memory. */
+ size = BUFFERALIGN(offsetof(DecodingWorkerShared, error_queue)) +
+ BUFFERALIGN(REPACK_ERROR_QUEUE_SIZE);
+ seg = dsm_create(size, 0);
+ shared = (DecodingWorkerShared *) dsm_segment_address(seg);
+ shared->lsn_upto = InvalidXLogRecPtr;
+ shared->done = false;
+ SharedFileSetInit(&shared->sfs, seg);
+ shared->last_exported = -1;
+ SpinLockInit(&shared->mutex);
+ shared->dbid = MyDatabaseId;
+
+ /*
+ * This is the UserId set in cluster_rel(). Security context shouldn't be
+ * needed for decoding worker.
+ */
+ shared->roleid = GetUserId();
+ shared->relid = relid;
+ ConditionVariableInit(&shared->cv);
+ shared->backend_proc = MyProc;
+ shared->backend_pid = MyProcPid;
+ shared->backend_proc_number = MyProcNumber;
+
+ mq = shm_mq_create((char *) BUFFERALIGN(shared->error_queue),
+ REPACK_ERROR_QUEUE_SIZE);
+ shm_mq_set_receiver(mq, MyProc);
+ mqh = shm_mq_attach(mq, seg, NULL);
+
+ memset(&bgw, 0, sizeof(bgw));
+ snprintf(bgw.bgw_name, BGW_MAXLEN,
+ "REPACK decoding worker for relation \"%s\"",
+ get_rel_name(relid));
+ snprintf(bgw.bgw_type, BGW_MAXLEN, "REPACK decoding worker");
+ bgw.bgw_flags = BGWORKER_SHMEM_ACCESS |
+ BGWORKER_BACKEND_DATABASE_CONNECTION;
+ bgw.bgw_start_time = BgWorkerStart_RecoveryFinished;
+ bgw.bgw_restart_time = BGW_NEVER_RESTART;
+ snprintf(bgw.bgw_library_name, MAXPGPATH, "postgres");
+ snprintf(bgw.bgw_function_name, BGW_MAXLEN, "RepackWorkerMain");
+ bgw.bgw_main_arg = UInt32GetDatum(dsm_segment_handle(seg));
+ bgw.bgw_notify_pid = MyProcPid;
+
+ decoding_worker = palloc0_object(DecodingWorker);
+ if (!RegisterDynamicBackgroundWorker(&bgw, &decoding_worker->handle))
+ ereport(ERROR,
+ (errcode(ERRCODE_CONFIGURATION_LIMIT_EXCEEDED),
+ errmsg("out of background worker slots"),
+ errhint("You might need to increase \"%s\".", "max_worker_processes")));
+
+ decoding_worker->seg = seg;
+ decoding_worker->error_mqh = mqh;
+
+ /*
+ * The decoding setup must be done before the caller can have XID assigned
+ * for any reason, otherwise the worker might end up in a deadlock,
+ * waiting for the caller's transaction to end. Therefore wait here until
+ * the worker indicates that it has the logical decoding initialized.
+ */
+ ConditionVariablePrepareToSleep(&shared->cv);
+ for (;;)
+ {
+ bool initialized;
+
+ SpinLockAcquire(&shared->mutex);
+ initialized = shared->initialized;
+ SpinLockRelease(&shared->mutex);
+
+ if (initialized)
+ break;
+
+ ConditionVariableSleep(&shared->cv, WAIT_EVENT_REPACK_WORKER_EXPORT);
+ }
+ ConditionVariableCancelSleep();
+}
+
+/*
+ * Stop the decoding worker and cleanup the related resources.
+ *
+ * The worker stops on its own when it knows there is no more work to do, but
+ * we need to stop it explicitly at least on ERROR in the launching backend.
+ */
+static void
+stop_decoding_worker(void)
+{
+ BgwHandleStatus status;
+
+ /* Haven't reached the worker startup? */
+ if (decoding_worker == NULL)
+ return;
+
+ /* Could not register the worker? */
+ if (decoding_worker->handle == NULL)
+ return;
+
+ TerminateBackgroundWorker(decoding_worker->handle);
+ /* The worker should really exit before the REPACK command does. */
+ HOLD_INTERRUPTS();
+ status = WaitForBackgroundWorkerShutdown(decoding_worker->handle);
+ RESUME_INTERRUPTS();
+
+ if (status == BGWH_POSTMASTER_DIED)
+ ereport(FATAL,
+ (errcode(ERRCODE_ADMIN_SHUTDOWN),
+ errmsg("postmaster exited during REPACK command")));
+
+ shm_mq_detach(decoding_worker->error_mqh);
+
+ /*
+ * If we could not cancel the current sleep due to ERROR, do that before
+ * we detach from the shared memory the condition variable is located in.
+ * If we did not, the bgworker ERROR handling code would try and fail
+ * badly.
+ */
+ ConditionVariableCancelSleep();
+
+ dsm_detach(decoding_worker->seg);
+ pfree(decoding_worker);
+ decoding_worker = NULL;
+}
+
+/* Is this process a REPACK worker? */
+static bool is_repack_worker = false;
+
+static pid_t backend_pid;
+static ProcNumber backend_proc_number;
+
+/*
+ * See ParallelWorkerShutdown for details.
+ */
+static void
+RepackWorkerShutdown(int code, Datum arg)
+{
+ SendProcSignal(backend_pid,
+ PROCSIG_REPACK_MESSAGE,
+ backend_proc_number);
+
+ dsm_detach((dsm_segment *) DatumGetPointer(arg));
+}
+
+/* REPACK decoding worker entry point */
+void
+RepackWorkerMain(Datum main_arg)
+{
+ dsm_segment *seg;
+ DecodingWorkerShared *shared;
+ shm_mq *mq;
+ shm_mq_handle *mqh;
+
+ is_repack_worker = true;
+
+ /*
+ * Override the default bgworker_die() with die() so we can use
+ * CHECK_FOR_INTERRUPTS().
+ */
+ pqsignal(SIGTERM, die);
+ BackgroundWorkerUnblockSignals();
+
+ seg = dsm_attach(DatumGetUInt32(main_arg));
+ if (seg == NULL)
+ ereport(ERROR,
+ (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
+ errmsg("could not map dynamic shared memory segment")));
+
+ shared = (DecodingWorkerShared *) dsm_segment_address(seg);
+
+ /* Arrange to signal the leader if we exit. */
+ backend_pid = shared->backend_pid;
+ backend_proc_number = shared->backend_proc_number;
+ before_shmem_exit(RepackWorkerShutdown, PointerGetDatum(seg));
+
+ /*
+ * Join locking group - see the comments around the call of
+ * start_decoding_worker().
+ */
+ if (!BecomeLockGroupMember(shared->backend_proc, backend_pid))
+ /* The leader is not running anymore. */
+ return;
+
+ /*
+ * Setup a queue to send error messages to the backend that launched this
+ * worker.
+ */
+ mq = (shm_mq *) (char *) BUFFERALIGN(shared->error_queue);
+ shm_mq_set_sender(mq, MyProc);
+ mqh = shm_mq_attach(mq, seg, NULL);
+ pq_redirect_to_shm_mq(seg, mqh);
+ pq_set_parallel_leader(shared->backend_pid,
+ shared->backend_proc_number);
+
+ /* Connect to the database. */
+ BackgroundWorkerInitializeConnectionByOid(shared->dbid, shared->roleid, 0);
+
+ repack_worker_internal(seg);
+}
+
+static void
+repack_worker_internal(dsm_segment *seg)
+{
+ DecodingWorkerShared *shared;
+ LogicalDecodingContext *decoding_ctx;
+ SharedFileSet *sfs;
+ Snapshot snapshot;
+
+ /*
+ * Transaction is needed to open relation, and it also provides us with a
+ * resource owner.
+ */
+ StartTransactionCommand();
+
+ shared = (DecodingWorkerShared *) dsm_segment_address(seg);
+
+ /*
+ * Not sure the spinlock is needed here - the backend should not change
+ * anything in the shared memory until we have serialized the snapshot.
+ */
+ SpinLockAcquire(&shared->mutex);
+ Assert(XLogRecPtrIsInvalid(shared->lsn_upto));
+ sfs = &shared->sfs;
+ SpinLockRelease(&shared->mutex);
+
+ SharedFileSetAttach(sfs, seg);
+
+ /*
+ * Prepare to capture the concurrent data changes ourselves.
+ */
+ decoding_ctx = setup_logical_decoding(shared->relid);
+
+ /* Announce that we're ready. */
+ SpinLockAcquire(&shared->mutex);
+ shared->initialized = true;
+ SpinLockRelease(&shared->mutex);
+ ConditionVariableSignal(&shared->cv);
+
+ /* Build the initial snapshot and export it. */
+ snapshot = SnapBuildInitialSnapshotForRepack(decoding_ctx->snapshot_builder);
+ export_initial_snapshot(snapshot, shared);
+
+ /*
+ * Only historic snapshots should be used now. Do not let us restrict the
+ * progress of xmin horizon.
+ */
+ InvalidateCatalogSnapshot();
+
+ while (!decode_concurrent_changes(decoding_ctx, shared))
+ ;
+
+ /* Cleanup. */
+ cleanup_logical_decoding(decoding_ctx);
+ CommitTransactionCommand();
+}
+
+/*
+ * Make snapshot available to the backend that launched the decoding worker.
+ */
+static void
+export_initial_snapshot(Snapshot snapshot, DecodingWorkerShared *shared)
+{
+ char fname[MAXPGPATH];
+ BufFile *file;
+ Size snap_size;
+ char *snap_space;
+
+ snap_size = EstimateSnapshotSpace(snapshot);
+ snap_space = (char *) palloc(snap_size);
+ SerializeSnapshot(snapshot, snap_space);
+ FreeSnapshot(snapshot);
+
+ DecodingWorkerFileName(fname, shared->relid, shared->last_exported + 1);
+ file = BufFileCreateFileSet(&shared->sfs.fs, fname);
+ /* To make restoration easier, write the snapshot size first. */
+ BufFileWrite(file, &snap_size, sizeof(snap_size));
+ BufFileWrite(file, snap_space, snap_size);
+ pfree(snap_space);
+ BufFileClose(file);
+
+ /* Increase the counter to tell the backend that the file is available. */
+ SpinLockAcquire(&shared->mutex);
+ shared->last_exported++;
+ SpinLockRelease(&shared->mutex);
+ ConditionVariableSignal(&shared->cv);
+}
+
+/*
+ * Get the initial snapshot from the decoding worker.
+ */
+static Snapshot
+get_initial_snapshot(DecodingWorker *worker)
+{
+ DecodingWorkerShared *shared;
+ char fname[MAXPGPATH];
+ BufFile *file;
+ Size snap_size;
+ char *snap_space;
+ Snapshot snapshot;
+
+ shared = (DecodingWorkerShared *) dsm_segment_address(worker->seg);
+
+ /*
+ * The worker needs to initialize the logical decoding, which usually
+ * takes some time. Therefore it makes sense to prepare for the sleep
+ * first.
+ */
+ ConditionVariablePrepareToSleep(&shared->cv);
+ for (;;)
+ {
+ int last_exported;
+
+ SpinLockAcquire(&shared->mutex);
+ last_exported = shared->last_exported;
+ SpinLockRelease(&shared->mutex);
+
+ /*
+ * Has the worker exported the file we are waiting for?
+ */
+ if (last_exported == WORKER_FILE_SNAPSHOT)
+ break;
+
+ ConditionVariableSleep(&shared->cv, WAIT_EVENT_REPACK_WORKER_EXPORT);
+ }
+ ConditionVariableCancelSleep();
+
+ /* Read the snapshot from a file. */
+ DecodingWorkerFileName(fname, shared->relid, WORKER_FILE_SNAPSHOT);
+ file = BufFileOpenFileSet(&shared->sfs.fs, fname, O_RDONLY, false);
+ BufFileReadExact(file, &snap_size, sizeof(snap_size));
+ snap_space = (char *) palloc(snap_size);
+ BufFileReadExact(file, snap_space, snap_size);
+ BufFileClose(file);
+
+ /* Restore it. */
+ snapshot = RestoreSnapshot(snap_space);
+ pfree(snap_space);
+
+ return snapshot;
+}
+
+bool
+IsRepackWorker(void)
+{
+ return is_repack_worker;
+}
+
+/*
+ * Handle receipt of an interrupt indicating a repack worker message.
+ *
+ * Note: this is called within a signal handler! All we can do is set
+ * a flag that will cause the next CHECK_FOR_INTERRUPTS() to invoke
+ * ProcessRepackMessages().
+ */
+void
+HandleRepackMessageInterrupt(void)
+{
+ InterruptPending = true;
+ RepackMessagePending = true;
+ SetLatch(MyLatch);
+}
+
+/*
+ * Process any queued protocol messages received from parallel workers.
+ */
+void
+ProcessRepackMessages(void)
+{
+ MemoryContext oldcontext;
+
+ static MemoryContext hpm_context = NULL;
+
+ /*
+ * Nothing to do if we haven't launched the worker yet or have already
+ * terminated it.
+ */
+ if (decoding_worker == NULL)
+ return;
+
+ /*
+ * This is invoked from ProcessInterrupts(), and since some of the
+ * functions it calls contain CHECK_FOR_INTERRUPTS(), there is a potential
+ * for recursive calls if more signals are received while this runs. It's
+ * unclear that recursive entry would be safe, and it doesn't seem useful
+ * even if it is safe, so let's block interrupts until done.
+ */
+ HOLD_INTERRUPTS();
+
+ /*
+ * Moreover, CurrentMemoryContext might be pointing almost anywhere. We
+ * don't want to risk leaking data into long-lived contexts, so let's do
+ * our work here in a private context that we can reset on each use.
+ */
+ if (hpm_context == NULL) /* first time through? */
+ hpm_context = AllocSetContextCreate(TopMemoryContext,
+ "ProcessRepackMessages",
+ ALLOCSET_DEFAULT_SIZES);
+ else
+ MemoryContextReset(hpm_context);
+
+ oldcontext = MemoryContextSwitchTo(hpm_context);
+
+ /* OK to process messages. Reset the flag saying there are more to do. */
+ RepackMessagePending = false;
+
+ /*
+ * Read as many messages as we can from each worker, but stop when no more
+ * messages can be read from the worker without blocking.
+ */
+ while (true)
+ {
+ shm_mq_result res;
+ Size nbytes;
+ void *data;
+
+ res = shm_mq_receive(decoding_worker->error_mqh, &nbytes,
+ &data, true);
+ if (res == SHM_MQ_WOULD_BLOCK)
+ break;
+ else if (res == SHM_MQ_SUCCESS)
+ {
+ StringInfoData msg;
+
+ initStringInfo(&msg);
+ appendBinaryStringInfo(&msg, data, nbytes);
+ ProcessRepackMessage(&msg);
+ pfree(msg.data);
+ }
+ else
+ {
+ /*
+ * The decoding worker is special in that it exits as soon as it
+ * has its work done. Thus the DETACHED result code is fine.
+ */
+ Assert(res == SHM_MQ_DETACHED);
+
+ break;
+ }
+ }
+
+ MemoryContextSwitchTo(oldcontext);
+
+ /* Might as well clear the context on our way out */
+ MemoryContextReset(hpm_context);
+
+ RESUME_INTERRUPTS();
+}
+
+/*
+ * Process a single protocol message received from a single parallel worker.
+ */
+static void
+ProcessRepackMessage(StringInfo msg)
+{
+ char msgtype;
+
+ msgtype = pq_getmsgbyte(msg);
+
+ switch (msgtype)
+ {
+ case PqMsg_ErrorResponse:
+ case PqMsg_NoticeResponse:
+ {
+ ErrorData edata;
+
+ /* Parse ErrorResponse or NoticeResponse. */
+ pq_parse_errornotice(msg, &edata);
+
+ /* Death of a worker isn't enough justification for suicide. */
+ edata.elevel = Min(edata.elevel, ERROR);
+
+ /*
+ * If desired, add a context line to show that this is a
+ * message propagated from a parallel worker. Otherwise, it
+ * can sometimes be confusing to understand what actually
+ * happened.
+ */
+ if (edata.context)
+ edata.context = psprintf("%s\n%s", edata.context,
+ _("decoding worker"));
+ else
+ edata.context = pstrdup(_("decoding worker"));
+
+ /* Rethrow error or print notice. */
+ ThrowErrorData(&edata);
+
+ break;
+ }
+
+ default:
+ {
+ elog(ERROR, "unrecognized message type received from decoding worker: %c (message length %d bytes)",
+ msgtype, msg->len);
+ }
+ }
+}
diff --git a/src/backend/libpq/pqmq.c b/src/backend/libpq/pqmq.c
index 6e4bbfb5aa1..42f6fa472c5 100644
--- a/src/backend/libpq/pqmq.c
+++ b/src/backend/libpq/pqmq.c
@@ -14,6 +14,7 @@
#include "postgres.h"
#include "access/parallel.h"
+#include "commands/cluster.h"
#include "libpq/libpq.h"
#include "libpq/pqformat.h"
#include "libpq/pqmq.h"
@@ -175,6 +176,10 @@ mq_putmessage(char msgtype, const char *s, size_t len)
SendProcSignal(pq_mq_parallel_leader_pid,
PROCSIG_PARALLEL_APPLY_MESSAGE,
pq_mq_parallel_leader_proc_number);
+ else if (IsRepackWorker())
+ SendProcSignal(pq_mq_parallel_leader_pid,
+ PROCSIG_REPACK_MESSAGE,
+ pq_mq_parallel_leader_proc_number);
else
{
Assert(IsParallelWorker());
diff --git a/src/backend/postmaster/bgworker.c b/src/backend/postmaster/bgworker.c
index 65deabe91a7..334bb708c5b 100644
--- a/src/backend/postmaster/bgworker.c
+++ b/src/backend/postmaster/bgworker.c
@@ -13,6 +13,7 @@
#include "postgres.h"
#include "access/parallel.h"
+#include "commands/cluster.h"
#include "libpq/pqsignal.h"
#include "miscadmin.h"
#include "pgstat.h"
@@ -136,6 +137,9 @@ static const struct
},
{
"SequenceSyncWorkerMain", SequenceSyncWorkerMain
+ },
+ {
+ "RepackWorkerMain", RepackWorkerMain
}
};
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index b0ef1a12520..35a46988285 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -194,7 +194,11 @@ StartupDecodingContext(List *output_plugin_options,
ctx->slot = slot;
- ctx->reader = XLogReaderAllocate(wal_segment_size, NULL, xl_routine, ctx);
+ /*
+ * TODO A separate patch for PG core, unless there's really a reason to
+ * pass ctx for private_data (May extensions expect ctx?).
+ */
+ ctx->reader = XLogReaderAllocate(wal_segment_size, NULL, xl_routine, NULL);
if (!ctx->reader)
ereport(ERROR,
(errcode(ERRCODE_OUT_OF_MEMORY),
diff --git a/src/backend/replication/pgoutput_repack/pgoutput_repack.c b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
index a074eed393d..4bc47a72371 100644
--- a/src/backend/replication/pgoutput_repack/pgoutput_repack.c
+++ b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
@@ -169,17 +169,13 @@ store_change(LogicalDecodingContext *ctx, ConcurrentChangeKind kind,
HeapTuple tuple)
{
RepackDecodingState *dstate;
- char *change_raw;
- ConcurrentChange change;
+ char kind_byte = (char) kind;
bool flattened = false;
- Size size;
- Datum values[1];
- bool isnull[1];
- char *dst;
dstate = (RepackDecodingState *) ctx->output_writer_private;
- size = VARHDRSZ + SizeOfConcurrentChange;
+ /* Store the change kind. */
+ BufFileWrite(dstate->file, &kind_byte, 1);
/*
* ReorderBufferCommit() stores the TOAST chunks in its private memory
@@ -196,46 +192,12 @@ store_change(LogicalDecodingContext *ctx, ConcurrentChangeKind kind,
tuple = toast_flatten_tuple(tuple, dstate->tupdesc);
flattened = true;
}
+ /* Store the tuple size ... */
+ BufFileWrite(dstate->file, &tuple->t_len, sizeof(tuple->t_len));
+ /* ... and the tuple itself. */
+ BufFileWrite(dstate->file, tuple->t_data, tuple->t_len);
- size += tuple->t_len;
- if (size >= MaxAllocSize)
- elog(ERROR, "Change is too big.");
-
- /* Construct the change. */
- change_raw = (char *) palloc0(size);
- SET_VARSIZE(change_raw, size);
-
- /*
- * Since the varlena alignment might not be sufficient for the structure,
- * set the fields in a local instance and remember where it should
- * eventually be copied.
- */
- change.kind = kind;
- dst = (char *) VARDATA(change_raw);
-
- /*
- * Copy the tuple.
- *
- * Note: change->tup_data.t_data must be fixed on retrieval!
- */
- memcpy(&change.tup_data, tuple, sizeof(HeapTupleData));
- memcpy(dst, &change, SizeOfConcurrentChange);
- dst += SizeOfConcurrentChange;
- memcpy(dst, tuple->t_data, tuple->t_len);
-
- /* The data has been copied. */
+ /* Free the flat copy if created above. */
if (flattened)
pfree(tuple);
-
- /* Store as tuple of 1 bytea column. */
- values[0] = PointerGetDatum(change_raw);
- isnull[0] = false;
- tuplestore_putvalues(dstate->tstore, dstate->tupdesc_change,
- values, isnull);
-
- /* Accounting. */
- dstate->nchanges++;
-
- /* Cleanup. */
- pfree(change_raw);
}
diff --git a/src/backend/storage/ipc/procsignal.c b/src/backend/storage/ipc/procsignal.c
index 8e56922dcea..6f9e7a7aab7 100644
--- a/src/backend/storage/ipc/procsignal.c
+++ b/src/backend/storage/ipc/procsignal.c
@@ -19,6 +19,7 @@
#include "access/parallel.h"
#include "commands/async.h"
+#include "commands/cluster.h"
#include "miscadmin.h"
#include "pgstat.h"
#include "port/pg_bitutils.h"
@@ -697,6 +698,9 @@ procsignal_sigusr1_handler(SIGNAL_ARGS)
if (CheckProcSignal(PROCSIG_PARALLEL_APPLY_MESSAGE))
HandleParallelApplyMessageInterrupt();
+ if (CheckProcSignal(PROCSIG_REPACK_MESSAGE))
+ HandleRepackMessageInterrupt();
+
if (CheckProcSignal(PROCSIG_RECOVERY_CONFLICT_DATABASE))
HandleRecoveryConflictInterrupt(PROCSIG_RECOVERY_CONFLICT_DATABASE);
diff --git a/src/backend/tcop/postgres.c b/src/backend/tcop/postgres.c
index e54bf1e760f..fc81ad87615 100644
--- a/src/backend/tcop/postgres.c
+++ b/src/backend/tcop/postgres.c
@@ -36,6 +36,7 @@
#include "access/xact.h"
#include "catalog/pg_type.h"
#include "commands/async.h"
+#include "commands/cluster.h"
#include "commands/event_trigger.h"
#include "commands/explain_state.h"
#include "commands/prepare.h"
@@ -3541,6 +3542,9 @@ ProcessInterrupts(void)
if (ParallelApplyMessagePending)
ProcessParallelApplyMessages();
+
+ if (RepackMessagePending)
+ ProcessRepackMessages();
}
/*
diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt
index 5537a2d2530..56d01ccd5af 100644
--- a/src/backend/utils/activity/wait_event_names.txt
+++ b/src/backend/utils/activity/wait_event_names.txt
@@ -154,6 +154,7 @@ RECOVERY_CONFLICT_SNAPSHOT "Waiting for recovery conflict resolution for a vacuu
RECOVERY_CONFLICT_TABLESPACE "Waiting for recovery conflict resolution for dropping a tablespace."
RECOVERY_END_COMMAND "Waiting for <xref linkend="guc-recovery-end-command"/> to complete."
RECOVERY_PAUSE "Waiting for recovery to be resumed."
+REPACK_WORKER_EXPORT "Waiting for decoding worker to export a new output file."
REPLICATION_ORIGIN_DROP "Waiting for a replication origin to become inactive so it can be dropped."
REPLICATION_SLOT_DROP "Waiting for a replication slot to become inactive so it can be dropped."
RESTORE_COMMAND "Waiting for <xref linkend="guc-restore-command"/> to complete."
diff --git a/src/include/access/tableam.h b/src/include/access/tableam.h
index 76aa993009a..15760363a1a 100644
--- a/src/include/access/tableam.h
+++ b/src/include/access/tableam.h
@@ -22,7 +22,6 @@
#include "access/xact.h"
#include "commands/vacuum.h"
#include "executor/tuptable.h"
-#include "replication/logical.h"
#include "storage/read_stream.h"
#include "utils/rel.h"
#include "utils/snapshot.h"
@@ -631,7 +630,6 @@ typedef struct TableAmRoutine
bool use_sort,
TransactionId OldestXmin,
Snapshot snapshot,
- LogicalDecodingContext *decoding_ctx,
TransactionId *xid_cutoff,
MultiXactId *multi_cutoff,
double *num_tuples,
@@ -1651,8 +1649,6 @@ table_relation_copy_data(Relation rel, const RelFileLocator *newrlocator)
* - *multi_cutoff - ditto
* - snapshot - if != NULL, ignore data changes done by transactions that this
* (MVCC) snapshot considers still in-progress or in the future.
- * - decoding_ctx - logical decoding context, to capture concurrent data
- * changes.
*
* Output parameters:
* - *xid_cutoff - rel's new relfrozenxid value, may be invalid
@@ -1666,7 +1662,6 @@ table_relation_copy_for_cluster(Relation OldTable, Relation NewTable,
bool use_sort,
TransactionId OldestXmin,
Snapshot snapshot,
- LogicalDecodingContext *decoding_ctx,
TransactionId *xid_cutoff,
MultiXactId *multi_cutoff,
double *num_tuples,
@@ -1675,7 +1670,7 @@ table_relation_copy_for_cluster(Relation OldTable, Relation NewTable,
{
OldTable->rd_tableam->relation_copy_for_cluster(OldTable, NewTable, OldIndex,
use_sort, OldestXmin,
- snapshot, decoding_ctx,
+ snapshot,
xid_cutoff, multi_cutoff,
num_tuples, tups_vacuumed,
tups_recently_dead);
diff --git a/src/include/commands/cluster.h b/src/include/commands/cluster.h
index 6a5c476294a..1b05d5d418b 100644
--- a/src/include/commands/cluster.h
+++ b/src/include/commands/cluster.h
@@ -17,11 +17,13 @@
#include "nodes/parsenodes.h"
#include "parser/parse_node.h"
#include "replication/decode.h"
+#include "postmaster/bgworker.h"
#include "replication/logical.h"
+#include "storage/buffile.h"
#include "storage/lock.h"
+#include "storage/shm_mq.h"
#include "utils/relcache.h"
#include "utils/resowner.h"
-#include "utils/tuplestore.h"
/* flag bits for ClusterParams->options */
@@ -44,6 +46,9 @@ typedef struct ClusterParams
* The following definitions are used by REPACK CONCURRENTLY.
*/
+/*
+ * Stored as a single byte in the output file.
+ */
typedef enum
{
CHANGE_INSERT,
@@ -52,68 +57,30 @@ typedef enum
CHANGE_DELETE
} ConcurrentChangeKind;
-typedef struct ConcurrentChange
-{
- /* See the enum above. */
- ConcurrentChangeKind kind;
-
- /*
- * The actual tuple.
- *
- * The tuple data follows the ConcurrentChange structure. Before use make
- * sure the tuple is correctly aligned (ConcurrentChange can be stored as
- * bytea) and that tuple->t_data is fixed.
- */
- HeapTupleData tup_data;
-} ConcurrentChange;
-
-#define SizeOfConcurrentChange (offsetof(ConcurrentChange, tup_data) + \
- sizeof(HeapTupleData))
-
/*
* Logical decoding state.
*
- * Here we store the data changes that we decode from WAL while the table
- * contents is being copied to a new storage. Also the necessary metadata
- * needed to apply these changes to the table is stored here.
+ * The output plugin uses it to store the data changes that it decodes from
+ * WAL while the table contents is being copied to a new storage.
*/
typedef struct RepackDecodingState
{
/* The relation whose changes we're decoding. */
Oid relid;
- /* Replication slot name. */
- NameData slotname;
-
- /*
- * Decoded changes are stored here. Although we try to avoid excessive
- * batches, it can happen that the changes need to be stored to disk. The
- * tuplestore does this transparently.
- */
- Tuplestorestate *tstore;
-
- /* The current number of changes in tstore. */
- double nchanges;
-
- /*
- * Descriptor to store the ConcurrentChange structure serialized (bytea).
- * We can't store the tuple directly because tuplestore only supports
- * minimum tuple and we may need to transfer OID system column from the
- * output plugin. Also we need to transfer the change kind, so it's better
- * to put everything in the structure than to use 2 tuplestores "in
- * parallel".
- */
- TupleDesc tupdesc_change;
-
- /* Tuple descriptor needed to update indexes. */
+ /* Tuple descriptor of the relation being processed. */
TupleDesc tupdesc;
- /* Slot to retrieve data from tstore. */
- TupleTableSlot *tsslot;
-
- ResourceOwner resowner;
+ /* The current output file. */
+ BufFile *file;
} RepackDecodingState;
+extern PGDLLIMPORT volatile sig_atomic_t RepackMessagePending;
+
+extern bool IsRepackWorker(void);
+extern void HandleRepackMessageInterrupt(void);
+extern void ProcessRepackMessages(void);
+
extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel);
extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid,
@@ -136,6 +103,6 @@ extern void finish_heap_swap(Oid OIDOldHeap, Oid OIDNewHeap,
extern bool am_decoding_for_repack(void);
extern bool change_useless_for_repack(XLogRecordBuffer *buf);
-extern void repack_decode_concurrent_changes(LogicalDecodingContext *ctx,
- XLogRecPtr end_of_wal);
+
+extern void RepackWorkerMain(Datum main_arg);
#endif /* CLUSTER_H */
diff --git a/src/include/storage/procsignal.h b/src/include/storage/procsignal.h
index e52b8eb7697..3ef35ca6b80 100644
--- a/src/include/storage/procsignal.h
+++ b/src/include/storage/procsignal.h
@@ -36,6 +36,7 @@ typedef enum
PROCSIG_BARRIER, /* global barrier interrupt */
PROCSIG_LOG_MEMORY_CONTEXT, /* ask backend to log the memory contexts */
PROCSIG_PARALLEL_APPLY_MESSAGE, /* Message from parallel apply workers */
+ PROCSIG_REPACK_MESSAGE, /* Message from repack worker */
/* Recovery conflict reasons */
PROCSIG_RECOVERY_CONFLICT_FIRST,
diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list
index 1aa4fb442f1..1e2e4ba80e9 100644
--- a/src/tools/pgindent/typedefs.list
+++ b/src/tools/pgindent/typedefs.list
@@ -498,7 +498,6 @@ CompressFileHandle
CompressionLocation
CompressorState
ComputeXidHorizonsResult
-ConcurrentChange
ConcurrentChangeKind
ConditionVariable
ConditionVariableMinimallyPadded
@@ -638,6 +637,8 @@ DeclareCursorStmt
DecodedBkpBlock
DecodedXLogRecord
DecodingOutputState
+DecodingWorker
+DecodingWorkerShared
DefElem
DefElemAction
DefaultACLInfo
--
2.47.3
--=-=-=
Content-Type: text/plain
Content-Disposition: attachment;
filename=v32-0006-Use-multiple-snapshots-to-copy-the-data.patch
^ permalink raw reply [nested|flat] 55+ messages in thread
* [PATCH v33 5/5] Use background worker to do logical decoding.
@ 2026-01-27 10:48 Antonin Houska <[email protected]>
0 siblings, 0 replies; 55+ messages in thread
From: Antonin Houska @ 2026-01-27 10:48 UTC (permalink / raw)
If the backend performing REPACK (CONCURRENTLY) does both data copying and
logical decoding, it has to "travel in time" back and forth and therefore it
has to invalidate system caches quite a few times. (The copying and the
decoding work with different catalog snapshots.) As the decoding worker has
separate caches, the switching is not necessary.
Without the worker, it'd also be difficult to switch between potentially long
running tasks like index build and WAL decoding. (No decoding during that time
at all can suspend archiving / recycling of WAL segments for some time, which
in turn may result in full disk.)
Another problem is that, after having acquired AccessExclusiveLock (in order
to swap the files), the backend needs to both decode and apply the data
changes that took place while it was waiting for the lock. With the decoding
worker, the decoding runs all the time, so the backend only needs to apply the
changes. This can reduce the time the exclusive lock is held for.
Note that the code added in order to handle ERRORs in the background worker
almost duplicates the existing code that does the same for other types of
workers (See ProcessParallelMessages() and
ProcessParallelApplyMessages()). Refactoring of the existing code might be
useful, to reduce the duplication.
---
src/backend/access/heap/heapam_handler.c | 44 -
src/backend/commands/cluster.c | 1183 +++++++++++++----
src/backend/libpq/pqmq.c | 5 +
src/backend/postmaster/bgworker.c | 5 +
src/backend/replication/logical/logical.c | 6 +-
.../pgoutput_repack/pgoutput_repack.c | 54 +-
src/backend/storage/ipc/procsignal.c | 4 +
src/backend/tcop/postgres.c | 4 +
.../utils/activity/wait_event_names.txt | 1 +
src/include/access/tableam.h | 7 +-
src/include/commands/cluster.h | 71 +-
src/include/storage/procsignal.h | 1 +
src/tools/pgindent/typedefs.list | 3 +-
13 files changed, 984 insertions(+), 404 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index 908f1ef66c6..8589b3c940e 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -33,7 +33,6 @@
#include "catalog/index.h"
#include "catalog/storage.h"
#include "catalog/storage_xlog.h"
-#include "commands/cluster.h"
#include "commands/progress.h"
#include "executor/executor.h"
#include "miscadmin.h"
@@ -688,7 +687,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
Relation OldIndex, bool use_sort,
TransactionId OldestXmin,
Snapshot snapshot,
- LogicalDecodingContext *decoding_ctx,
TransactionId *xid_cutoff,
MultiXactId *multi_cutoff,
double *num_tuples,
@@ -710,7 +708,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
BufferHeapTupleTableSlot *hslot;
BlockNumber prev_cblock = InvalidBlockNumber;
bool concurrent = snapshot != NULL;
- XLogRecPtr end_of_wal_prev = GetFlushRecPtr(NULL);
/* Remember if it's a system catalog */
is_system_catalog = IsSystemRelation(OldHeap);
@@ -971,31 +968,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
ct_val[1] = *num_tuples;
pgstat_progress_update_multi_param(2, ct_index, ct_val);
}
-
- /*
- * Process the WAL produced by the load, as well as by other
- * transactions, so that the replication slot can advance and WAL does
- * not pile up. Use wal_segment_size as a threshold so that we do not
- * introduce the decoding overhead too often.
- *
- * Of course, we must not apply the changes until the initial load has
- * completed.
- *
- * Note that our insertions into the new table should not be decoded
- * as we (intentionally) do not write the logical decoding specific
- * information to WAL.
- */
- if (concurrent)
- {
- XLogRecPtr end_of_wal;
-
- end_of_wal = GetFlushRecPtr(NULL);
- if ((end_of_wal - end_of_wal_prev) > wal_segment_size)
- {
- repack_decode_concurrent_changes(decoding_ctx, end_of_wal);
- end_of_wal_prev = end_of_wal;
- }
- }
}
if (indexScan != NULL)
@@ -1041,22 +1013,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
-
- /*
- * Try to keep the amount of not-yet-decoded WAL small, like
- * above.
- */
- if (concurrent)
- {
- XLogRecPtr end_of_wal;
-
- end_of_wal = GetFlushRecPtr(NULL);
- if ((end_of_wal - end_of_wal_prev) > wal_segment_size)
- {
- repack_decode_concurrent_changes(decoding_ctx, end_of_wal);
- end_of_wal_prev = end_of_wal;
- }
- }
}
tuplesort_end(tuplesort);
diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c
index 03ccf10b782..e988a7a7296 100644
--- a/src/backend/commands/cluster.c
+++ b/src/backend/commands/cluster.c
@@ -12,12 +12,13 @@
* In concurrent mode, we lock the table with only ShareUpdateExclusiveLock,
* then do an initial copy as above. However, while the tuples are being
* copied, concurrent transactions could modify the table. To cope with those
- * changes, we rely on logical decoding to obtain them from WAL. The changes
- * are accumulated in a tuplestore. Once the initial copy is complete, we
- * read the changes from the tuplestore and re-apply them on the new heap.
- * Then we upgrade our ShareUpdateExclusiveLock to AccessExclusiveLock and
- * swap the relfilenodes. This way, the time we hold a strong lock on the
- * table is much reduced, and the bloat is eliminated.
+ * changes, we rely on logical decoding to obtain them from WAL. A bgworker
+ * consumes WAL while the initial copy is ongoing (to prevent excessive WAL
+ * from being reserved), and accumulates the changes in a file. Once the
+ * initial copy is complete, we read the changes from the file and re-apply
+ * them on the new heap. Then we upgrade our ShareUpdateExclusiveLock to
+ * AccessExclusiveLock and swap the relfilenodes. This way, the time we hold
+ * a strong lock on the table is much reduced, and the bloat is eliminated.
*
* There is hardly anything left of Paul Brown's original implementation...
*
@@ -45,6 +46,7 @@
#include "access/xlog_internal.h"
#include "access/xloginsert.h"
#include "access/xlogutils.h"
+#include "access/xlogwait.h"
#include "catalog/catalog.h"
#include "catalog/dependency.h"
#include "catalog/heap.h"
@@ -61,6 +63,8 @@
#include "commands/tablecmds.h"
#include "commands/vacuum.h"
#include "executor/executor.h"
+#include "libpq/pqformat.h"
+#include "libpq/pqmq.h"
#include "miscadmin.h"
#include "optimizer/optimizer.h"
#include "pgstat.h"
@@ -71,6 +75,8 @@
#include "storage/ipc.h"
#include "storage/lmgr.h"
#include "storage/predicate.h"
+#include "storage/procsignal.h"
+#include "tcop/tcopprot.h"
#include "utils/acl.h"
#include "utils/fmgroids.h"
#include "utils/guc.h"
@@ -117,6 +123,12 @@ typedef struct IndexInsertState
/* The WAL segment being decoded. */
static XLogSegNo repack_current_segment = 0;
+/*
+ * The first file exported by the decoding worker must contain a snapshot, the
+ * following ones contain the data changes.
+ */
+#define WORKER_FILE_SNAPSHOT 0
+
/*
* Information needed to apply concurrent data changes.
*/
@@ -136,8 +148,113 @@ typedef struct ChangeDest
/* Needed to update indexes of rel_dst. */
IndexInsertState *iistate;
+
+ /*
+ * Sequential number of the file containing the changes.
+ *
+ * TODO This field makes the structure name less descriptive. Should we
+ * rename it, e.g. to ChangeApplyInfo?
+ */
+ int file_seq;
} ChangeDest;
+/*
+ * Layout of shared memory used for communication between backend and the
+ * worker that performs logical decoding of data changes
+ */
+typedef struct DecodingWorkerShared
+{
+ /* Is the decoding initialized? */
+ bool initialized;
+
+ /*
+ * Once the worker has reached this LSN, it should close the current
+ * output file and either create a new one or exit, according to the field
+ * 'done'. If the value is InvalidXLogRecPtr, the worker should decode all
+ * the WAL available and keep checking this field. It is ok if the worker
+ * had already decoded records whose LSN is >= lsn_upto before this field
+ * has been set.
+ */
+ XLogRecPtr lsn_upto;
+
+ /* Exit after closing the current file? */
+ bool done;
+
+ /* The output is stored here. */
+ SharedFileSet sfs;
+
+ /* Number of the last file exported by the worker. */
+ int last_exported;
+
+ /* Synchronize access to the fields above. */
+ slock_t mutex;
+
+ /* Database to connect to. */
+ Oid dbid;
+
+ /* Role to connect as. */
+ Oid roleid;
+
+ /* Decode data changes of this relation. */
+ Oid relid;
+
+ /* The backend uses this to wait for the worker. */
+ ConditionVariable cv;
+
+ /* Info to signal the backend. */
+ PGPROC *backend_proc;
+ pid_t backend_pid;
+ ProcNumber backend_proc_number;
+
+ /* Error queue. */
+ shm_mq *error_mq;
+
+ /*
+ * Memory the queue is located in.
+ *
+ * For considerations on the value see the comments of
+ * PARALLEL_ERROR_QUEUE_SIZE.
+ */
+#define REPACK_ERROR_QUEUE_SIZE 16384
+ char error_queue[FLEXIBLE_ARRAY_MEMBER];
+} DecodingWorkerShared;
+
+/*
+ * Generate worker's output file name. If relations of the same 'relid' happen
+ * to be processed at the same time, they must be from different databases and
+ * therefore different backends must be involved. (PID is already present in
+ * the fileset name.)
+ */
+static inline void
+DecodingWorkerFileName(char *fname, Oid relid, uint32 seq)
+{
+ snprintf(fname, MAXPGPATH, "%u-%u", relid, seq);
+}
+
+/*
+ * Backend-local information to control the decoding worker.
+ */
+typedef struct DecodingWorker
+{
+ /* The worker. */
+ BackgroundWorkerHandle *handle;
+
+ /* DecodingWorkerShared is in this segment. */
+ dsm_segment *seg;
+
+ /* Handle of the error queue. */
+ shm_mq_handle *error_mqh;
+} DecodingWorker;
+
+/* Pointer to currently running decoding worker. */
+static DecodingWorker *decoding_worker = NULL;
+
+/*
+ * Is there a message sent by a repack worker that the backend needs to
+ * receive?
+ */
+volatile sig_atomic_t RepackMessagePending = false;
+
static bool cluster_rel_recheck(RepackCommand cmd, Relation OldHeap,
Oid indexOid, Oid userid, LOCKMODE lmode,
int options);
@@ -145,7 +262,7 @@ static void check_repack_concurrently_requirements(Relation rel);
static void rebuild_relation(Relation OldHeap, Relation index, bool verbose,
bool concurrent);
static void copy_table_data(Relation NewHeap, Relation OldHeap, Relation OldIndex,
- Snapshot snapshot, LogicalDecodingContext *decoding_ctx,
+ Snapshot snapshot,
bool verbose,
bool *pSwapToastByContent,
TransactionId *pFreezeXid,
@@ -158,12 +275,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd,
static bool cluster_is_permitted_for_relation(RepackCommand cmd,
Oid relid, Oid userid);
-static void begin_concurrent_repack(Relation rel);
-static void end_concurrent_repack(void);
static LogicalDecodingContext *setup_logical_decoding(Oid relid);
-static HeapTuple get_changed_tuple(char *change);
-static void apply_concurrent_changes(RepackDecodingState *dstate,
- ChangeDest *dest);
+static bool decode_concurrent_changes(LogicalDecodingContext *ctx,
+ DecodingWorkerShared *shared);
+static void apply_concurrent_changes(BufFile *file, ChangeDest *dest);
static void apply_concurrent_insert(Relation rel, HeapTuple tup,
IndexInsertState *iistate,
TupleTableSlot *index_slot);
@@ -175,9 +290,9 @@ static void apply_concurrent_delete(Relation rel, HeapTuple tup_target);
static HeapTuple find_target_tuple(Relation rel, ChangeDest *dest,
HeapTuple tup_key,
TupleTableSlot *ident_slot);
-static void process_concurrent_changes(LogicalDecodingContext *decoding_ctx,
- XLogRecPtr end_of_wal,
- ChangeDest *dest);
+static void process_concurrent_changes(XLogRecPtr end_of_wal,
+ ChangeDest *dest,
+ bool done);
static IndexInsertState *get_index_insert_state(Relation relation,
Oid ident_index_id,
Relation *ident_index_p);
@@ -186,7 +301,6 @@ static ScanKey build_identity_key(Oid ident_idx_oid, Relation rel_src,
static void free_index_insert_state(IndexInsertState *iistate);
static void cleanup_logical_decoding(LogicalDecodingContext *ctx);
static void rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
- LogicalDecodingContext *decoding_ctx,
TransactionId frozenXid,
MultiXactId cutoffMulti);
static List *build_new_indexes(Relation NewHeap, Relation OldHeap, List *OldIndexes);
@@ -196,6 +310,13 @@ static Relation process_single_relation(RepackStmt *stmt,
ClusterParams *params);
static Oid determine_clustered_index(Relation rel, bool usingindex,
const char *indexname);
+static void start_decoding_worker(Oid relid);
+static void stop_decoding_worker(void);
+static void repack_worker_internal(dsm_segment *seg);
+static void export_initial_snapshot(Snapshot snapshot,
+ DecodingWorkerShared *shared);
+static Snapshot get_initial_snapshot(DecodingWorker *worker);
+static void ProcessRepackMessage(StringInfo msg);
static const char *RepackCommandAsString(RepackCommand cmd);
@@ -619,20 +740,20 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid,
/* rebuild_relation does all the dirty work */
PG_TRY();
{
- /*
- * For concurrent processing, make sure that our logical decoding
- * ignores data changes of other tables than the one we are
- * processing.
- */
- if (concurrent)
- begin_concurrent_repack(OldHeap);
-
rebuild_relation(OldHeap, index, verbose, concurrent);
}
PG_FINALLY();
{
if (concurrent)
- end_concurrent_repack();
+ {
+ /*
+ * Since during normal operation the worker was already asked to
+ * exit, stopping it explicitly is especially important on ERROR.
+ * However it still seems a good practice to make sure that the
+ * worker never survives the REPACK command.
+ */
+ stop_decoding_worker();
+ }
}
PG_END_TRY();
@@ -929,7 +1050,6 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
bool swap_toast_by_content;
TransactionId frozenXid;
MultiXactId cutoffMulti;
- LogicalDecodingContext *decoding_ctx = NULL;
Snapshot snapshot = NULL;
#if USE_ASSERT_CHECKING
LOCKMODE lmode;
@@ -943,19 +1063,36 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
if (concurrent)
{
/*
- * Prepare to capture the concurrent data changes.
+ * The worker needs to be member of the locking group we're the leader
+ * of. We ought to become the leader before the worker starts. The
+ * worker will join the group as soon as it starts.
*
- * Note that this call waits for all transactions with XID already
- * assigned to finish. If some of those transactions is waiting for a
- * lock conflicting with ShareUpdateExclusiveLock on our table (e.g.
- * it runs CREATE INDEX), we can end up in a deadlock. Not sure this
- * risk is worth unlocking/locking the table (and its clustering
- * index) and checking again if its still eligible for REPACK
- * CONCURRENTLY.
+ * This is to make sure that the deadlock described below is
+ * detectable by deadlock.c: if the worker waits for a transaction to
+ * complete and we are waiting for the worker output, then effectively
+ * we (i.e. this backend) are waiting for that transaction.
*/
- decoding_ctx = setup_logical_decoding(tableOid);
+ BecomeLockGroupLeader();
+
+ /*
+ * Start the worker that decodes data changes applied while we're
+ * copying the table contents.
+ *
+ * Note that the worker has to wait for all transactions with XID
+ * already assigned to finish. If some of those transactions is
+ * waiting for a lock conflicting with ShareUpdateExclusiveLock on our
+ * table (e.g. it runs CREATE INDEX), we can end up in a deadlock.
+ * Not sure this risk is worth unlocking/locking the table (and its
+ * clustering index) and checking again if it's still eligible for
+ * REPACK CONCURRENTLY.
+ */
+ start_decoding_worker(tableOid);
+
+ /*
+ * Wait until the worker has the initial snapshot and retrieve it.
+ */
+ snapshot = get_initial_snapshot(decoding_worker);
- snapshot = SnapBuildInitialSnapshotForRepack(decoding_ctx->snapshot_builder);
PushActiveSnapshot(snapshot);
}
@@ -980,7 +1117,7 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
NewHeap = table_open(OIDNewHeap, NoLock);
/* Copy the heap data into the new table in the desired order */
- copy_table_data(NewHeap, OldHeap, index, snapshot, decoding_ctx, verbose,
+ copy_table_data(NewHeap, OldHeap, index, snapshot, verbose,
&swap_toast_by_content, &frozenXid, &cutoffMulti);
/* The historic snapshot won't be needed anymore. */
@@ -1001,14 +1138,11 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
if (index)
index_close(index, NoLock);
- rebuild_relation_finish_concurrent(NewHeap, OldHeap, decoding_ctx,
- frozenXid, cutoffMulti);
+ rebuild_relation_finish_concurrent(NewHeap, OldHeap, frozenXid,
+ cutoffMulti);
pgstat_progress_update_param(PROGRESS_REPACK_PHASE,
PROGRESS_REPACK_PHASE_FINAL_CLEANUP);
-
- /* Done with decoding. */
- cleanup_logical_decoding(decoding_ctx);
}
else
{
@@ -1179,8 +1313,7 @@ make_new_heap(Oid OIDOldHeap, Oid NewTableSpace, Oid NewAccessMethod,
*/
static void
copy_table_data(Relation NewHeap, Relation OldHeap, Relation OldIndex,
- Snapshot snapshot, LogicalDecodingContext *decoding_ctx,
- bool verbose, bool *pSwapToastByContent,
+ Snapshot snapshot, bool verbose, bool *pSwapToastByContent,
TransactionId *pFreezeXid, MultiXactId *pCutoffMulti)
{
Relation relRelation;
@@ -1341,7 +1474,6 @@ copy_table_data(Relation NewHeap, Relation OldHeap, Relation OldIndex,
*/
table_relation_copy_for_cluster(OldHeap, NewHeap, OldIndex, use_sort,
cutoffs.OldestXmin, snapshot,
- decoding_ctx,
&cutoffs.FreezeLimit,
&cutoffs.MultiXactCutoff,
&num_tuples, &tups_vacuumed,
@@ -2371,62 +2503,10 @@ RepackCommandAsString(RepackCommand cmd)
case REPACK_COMMAND_CLUSTER:
return "CLUSTER";
}
- return "???"; /* keep compiler quiet */
+ return "???"; /* keep compiler quiet */
}
-/*
- * Call this function before REPACK CONCURRENTLY starts to setup logical
- * decoding. It makes sure that other users of the table put enough
- * information into WAL.
- *
- * The point is that at various places we expect that the table we're
- * processing is treated like a system catalog. For example, we need to be
- * able to scan it using a "historic snapshot" anytime during the processing
- * (as opposed to scanning only at the start point of the decoding, as logical
- * replication does during initial table synchronization), in order to apply
- * concurrent UPDATE / DELETE commands.
- *
- * Note that TOAST table needs no attention here as it's not scanned using
- * historic snapshot.
- */
-static void
-begin_concurrent_repack(Relation rel)
-{
- Oid toastrelid;
-
- /*
- * Avoid logical decoding of other relations by this backend. The lock we
- * have guarantees that the actual locator cannot be changed concurrently:
- * TRUNCATE needs AccessExclusiveLock.
- */
- Assert(CheckRelationLockedByMe(rel, ShareUpdateExclusiveLock, false));
- repacked_rel_locator = rel->rd_locator;
- toastrelid = rel->rd_rel->reltoastrelid;
- if (OidIsValid(toastrelid))
- {
- Relation toastrel;
-
- /* Avoid logical decoding of other TOAST relations. */
- toastrel = table_open(toastrelid, AccessShareLock);
- repacked_rel_toast_locator = toastrel->rd_locator;
- table_close(toastrel, AccessShareLock);
- }
-}
-
-/*
- * Call this when done with REPACK CONCURRENTLY.
- */
-static void
-end_concurrent_repack(void)
-{
- /*
- * Restore normal function of (future) logical decoding for this backend.
- */
- repacked_rel_locator.relNumber = InvalidOid;
- repacked_rel_toast_locator.relNumber = InvalidOid;
-}
-
/*
* Is this backend performing logical decoding on behalf of REPACK
* (CONCURRENTLY) ?
@@ -2491,9 +2571,10 @@ static LogicalDecodingContext *
setup_logical_decoding(Oid relid)
{
Relation rel;
- TupleDesc tupdesc;
+ Oid toastrelid;
LogicalDecodingContext *ctx;
- RepackDecodingState *dstate = palloc0_object(RepackDecodingState);
+ NameData slotname;
+ RepackDecodingState *dstate;
/*
* REPACK CONCURRENTLY is not allowed in a transaction block, so this
@@ -2501,20 +2582,21 @@ setup_logical_decoding(Oid relid)
*/
Assert(!TransactionIdIsValid(GetTopTransactionIdIfAny()));
- /*
- * A single backend should not execute multiple REPACK commands at a time,
- * so use PID to make the slot unique.
- */
- snprintf(NameStr(dstate->slotname), NAMEDATALEN, "repack_%d", MyProcPid);
-
/*
* Make sure we can use logical decoding.
*/
CheckSlotPermissions();
CheckLogicalDecodingRequirements();
- /* RS_TEMPORARY so that the slot gets cleaned up on ERROR. */
- ReplicationSlotCreate(NameStr(dstate->slotname), true, RS_TEMPORARY,
- false, false, false);
+ /*
+ * A single backend should not execute multiple REPACK commands at a time,
+ * so use PID to make the slot unique.
+ *
+ * RS_TEMPORARY so that the slot gets cleaned up on ERROR.
+ */
+ snprintf(NameStr(slotname), NAMEDATALEN, "repack_%d", MyProcPid);
+ ReplicationSlotCreate(NameStr(slotname), true, RS_TEMPORARY, false, false,
+ false);
+
EnsureLogicalDecodingEnabled();
/*
@@ -2537,104 +2619,109 @@ setup_logical_decoding(Oid relid)
DecodingContextFindStartpoint(ctx);
+ /*
+ * decode_concurrent_changes() needs non-blocking callback.
+ */
+ ctx->reader->routine.page_read = read_local_xlog_page_no_wait;
+
+ /*
+ * read_local_xlog_page_no_wait() needs to be able to indicate the end of
+ * WAL.
+ */
+ ctx->reader->private_data = MemoryContextAllocZero(ctx->context,
+ sizeof(ReadLocalXLogPageNoWaitPrivate));
+
+
/* Some WAL records should have been read. */
Assert(ctx->reader->EndRecPtr != InvalidXLogRecPtr);
+ /*
+ * Initialize repack_current_segment so that we can notice WAL segment
+ * boundaries.
+ */
XLByteToSeg(ctx->reader->EndRecPtr, repack_current_segment,
wal_segment_size);
- /*
- * Setup structures to store decoded changes.
- */
+ dstate = palloc0_object(RepackDecodingState);
dstate->relid = relid;
- dstate->tstore = tuplestore_begin_heap(false, false,
- maintenance_work_mem);
- /* Caller should already have the table locked. */
- rel = table_open(relid, NoLock);
- tupdesc = CreateTupleDescCopy(RelationGetDescr(rel));
- dstate->tupdesc = tupdesc;
- table_close(rel, NoLock);
+ /*
+ * Tuple descriptor may be needed to flatten a tuple before we write it to
+ * a file. A copy is needed because the decoding worker invalidates system
+ * caches before it starts to do the actual work.
+ */
+ rel = table_open(relid, AccessShareLock);
+ dstate->tupdesc = CreateTupleDescCopy(RelationGetDescr(rel));
- /* Initialize the descriptor to store the changes ... */
- dstate->tupdesc_change = CreateTemplateTupleDesc(1);
+ /* Avoid logical decoding of other relations. */
+ repacked_rel_locator = rel->rd_locator;
+ toastrelid = rel->rd_rel->reltoastrelid;
+ if (OidIsValid(toastrelid))
+ {
+ Relation toastrel;
- TupleDescInitEntry(dstate->tupdesc_change, 1, NULL, BYTEAOID, -1, 0);
- /* ... as well as the corresponding slot. */
- dstate->tsslot = MakeSingleTupleTableSlot(dstate->tupdesc_change,
- &TTSOpsMinimalTuple);
+ /* Avoid logical decoding of other TOAST relations. */
+ toastrel = table_open(toastrelid, AccessShareLock);
+ repacked_rel_toast_locator = toastrel->rd_locator;
+ table_close(toastrel, AccessShareLock);
+ }
+ table_close(rel, AccessShareLock);
- dstate->resowner = ResourceOwnerCreate(CurrentResourceOwner,
- "logical decoding");
+ /* The file will be set as soon as we have it opened. */
+ dstate->file = NULL;
ctx->output_writer_private = dstate;
+
return ctx;
}
/*
- * Retrieve tuple from ConcurrentChange structure.
+ * Decode logical changes from the WAL sequence and store them to a file.
*
- * The input data starts with the structure but it might not be appropriately
- * aligned.
+ * If true is returned, there is no more work for the worker.
*/
-static HeapTuple
-get_changed_tuple(char *change)
-{
- HeapTupleData tup_data;
- HeapTuple result;
- char *src;
-
- /*
- * Ensure alignment before accessing the fields. (This is why we can't use
- * heap_copytuple() instead of this function.)
- */
- src = change + offsetof(ConcurrentChange, tup_data);
- memcpy(&tup_data, src, sizeof(HeapTupleData));
-
- result = (HeapTuple) palloc(HEAPTUPLESIZE + tup_data.t_len);
- memcpy(result, &tup_data, sizeof(HeapTupleData));
- result->t_data = (HeapTupleHeader) ((char *) result + HEAPTUPLESIZE);
- src = change + SizeOfConcurrentChange;
- memcpy(result->t_data, src, result->t_len);
-
- return result;
-}
-
-/*
- * Decode logical changes from the WAL sequence up to end_of_wal.
- */
-void
-repack_decode_concurrent_changes(LogicalDecodingContext *ctx,
- XLogRecPtr end_of_wal)
+static bool
+decode_concurrent_changes(LogicalDecodingContext *ctx,
+ DecodingWorkerShared *shared)
{
RepackDecodingState *dstate;
- ResourceOwner resowner_old;
+ XLogRecPtr lsn_upto;
+ bool done;
+ char fname[MAXPGPATH];
dstate = (RepackDecodingState *) ctx->output_writer_private;
- resowner_old = CurrentResourceOwner;
- CurrentResourceOwner = dstate->resowner;
- PG_TRY();
+ /* Open the output file. */
+ DecodingWorkerFileName(fname, shared->relid, shared->last_exported + 1);
+ dstate->file = BufFileCreateFileSet(&shared->sfs.fs, fname);
+
+ SpinLockAcquire(&shared->mutex);
+ lsn_upto = shared->lsn_upto;
+ done = shared->done;
+ SpinLockRelease(&shared->mutex);
+
+ while (true)
{
- while (ctx->reader->EndRecPtr < end_of_wal)
+ XLogRecord *record;
+ XLogSegNo segno_new;
+ char *errm = NULL;
+ XLogRecPtr end_lsn;
+
+ CHECK_FOR_INTERRUPTS();
+
+ record = XLogReadRecord(ctx->reader, &errm);
+ if (record)
{
- XLogRecord *record;
- XLogSegNo segno_new;
- char *errm = NULL;
- XLogRecPtr end_lsn;
-
- record = XLogReadRecord(ctx->reader, &errm);
- if (errm)
- elog(ERROR, "%s", errm);
-
- if (record != NULL)
- LogicalDecodingProcessRecord(ctx, ctx->reader);
+ LogicalDecodingProcessRecord(ctx, ctx->reader);
/*
* If WAL segment boundary has been crossed, inform the decoding
- * system that the catalog_xmin can advance. (We can confirm more
- * often, but a filling a single WAL segment should not take much
- * time.)
+ * system that the catalog_xmin can advance.
+ *
+ * TODO Does it make sense to confirm more often? Segment size
+ * seems appropriate for restart_lsn (because less than a segment
+ * cannot be recycled anyway), however more frequent checks might
+ * be beneficial for catalog_xmin.
*/
end_lsn = ctx->reader->EndRecPtr;
XLByteToSeg(end_lsn, segno_new, wal_segment_size);
@@ -2645,80 +2732,137 @@ repack_decode_concurrent_changes(LogicalDecodingContext *ctx,
(uint32) (end_lsn >> 32), (uint32) end_lsn);
repack_current_segment = segno_new;
}
-
- CHECK_FOR_INTERRUPTS();
}
- InvalidateSystemCaches();
- CurrentResourceOwner = resowner_old;
+ else
+ {
+ ReadLocalXLogPageNoWaitPrivate *priv;
+
+ if (errm)
+ ereport(ERROR, (errmsg("%s", errm)));
+
+ /*
+ * In the decoding loop we do not want to get blocked when there
+ * is no more WAL available, otherwise the loop would become
+ * uninterruptible.
+ */
+ priv = (ReadLocalXLogPageNoWaitPrivate *)
+ ctx->reader->private_data;
+ if (priv->end_of_wal)
+ /* Do not miss the end of WAL condition next time. */
+ priv->end_of_wal = false;
+ else
+ ereport(ERROR, (errmsg("could not read WAL record")));
+ }
+
+ /*
+ * Whether we could read new record or not, keep checking if
+ * 'lsn_upto' was specified.
+ */
+ if (XLogRecPtrIsInvalid(lsn_upto))
+ {
+ SpinLockAcquire(&shared->mutex);
+ lsn_upto = shared->lsn_upto;
+ /* 'done' should be set at the same time as 'lsn_upto' */
+ done = shared->done;
+ SpinLockRelease(&shared->mutex);
+ }
+ if (!XLogRecPtrIsInvalid(lsn_upto) &&
+ ctx->reader->EndRecPtr >= lsn_upto)
+ break;
+
+ if (record == NULL)
+ {
+ int64 timeout = 0;
+ WaitLSNResult res;
+
+ /*
+ * Before we retry reading, wait until new WAL is flushed.
+ *
+ * There is a race condition such that the backend executing
+ * REPACK determines 'lsn_upto', but before it sets the shared
+ * variable, we reach the end of WAL. In that case we'd need to
+ * wait until the next WAL flush (unrelated to REPACK). Although
+ * that should not be a problem in a busy system, it might be
+ * noticeable in other cases, including regression tests (which
+ * are not necessarily executed in parallel). Therefore it makes
+ * sense to use timeout.
+ *
+ * If lsn_upto is valid, WAL records having LSN lower than that
+ * should already have been flushed to disk.
+ */
+ if (XLogRecPtrIsInvalid(lsn_upto))
+ timeout = 100L;
+ res = WaitForLSN(WAIT_LSN_TYPE_PRIMARY_FLUSH,
+ ctx->reader->EndRecPtr + 1,
+ timeout);
+ if (res != WAIT_LSN_RESULT_SUCCESS &&
+ res != WAIT_LSN_RESULT_TIMEOUT)
+ ereport(ERROR, (errmsg("waiting for WAL failed")));
+ }
}
- PG_CATCH();
- {
- /* clear all timetravel entries */
- InvalidateSystemCaches();
- CurrentResourceOwner = resowner_old;
- PG_RE_THROW();
- }
- PG_END_TRY();
+
+ /*
+ * Close the file so we can make it available to the backend.
+ */
+ BufFileClose(dstate->file);
+ dstate->file = NULL;
+ SpinLockAcquire(&shared->mutex);
+ shared->lsn_upto = InvalidXLogRecPtr;
+ shared->last_exported++;
+ SpinLockRelease(&shared->mutex);
+ ConditionVariableSignal(&shared->cv);
+
+ return done;
}
/*
* Apply changes stored in 'file'.
*/
static void
-apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
+apply_concurrent_changes(BufFile *file, ChangeDest *dest)
{
+ char kind;
+ uint32 t_len;
Relation rel = dest->rel;
TupleTableSlot *index_slot,
*ident_slot;
HeapTuple tup_old = NULL;
- if (dstate->nchanges == 0)
- return;
-
/* TupleTableSlot is needed to pass the tuple to ExecInsertIndexTuples(). */
- index_slot = MakeSingleTupleTableSlot(dstate->tupdesc, &TTSOpsHeapTuple);
+ index_slot = MakeSingleTupleTableSlot(RelationGetDescr(rel),
+ &TTSOpsHeapTuple);
/* A slot to fetch tuples from identity index. */
ident_slot = table_slot_create(rel, NULL);
- while (tuplestore_gettupleslot(dstate->tstore, true, false,
- dstate->tsslot))
+ while (true)
{
- bool shouldFree;
- HeapTuple tup_change,
- tup,
+ size_t nread;
+ HeapTuple tup,
tup_exist;
- char *change_raw,
- *src;
- ConcurrentChange change;
- bool isnull[1];
- Datum values[1];
CHECK_FOR_INTERRUPTS();
- /* Get the change from the single-column tuple. */
- tup_change = ExecFetchSlotHeapTuple(dstate->tsslot, false, &shouldFree);
- heap_deform_tuple(tup_change, dstate->tupdesc_change, values, isnull);
- Assert(!isnull[0]);
+ nread = BufFileReadMaybeEOF(file, &kind, 1, true);
+ /* Are we done with the file? */
+ if (nread == 0)
+ break;
- /* Make sure we access aligned data. */
- change_raw = (char *) DatumGetByteaP(values[0]);
- src = (char *) VARDATA(change_raw);
- memcpy(&change, src, SizeOfConcurrentChange);
+ /* Read the tuple. */
+ BufFileReadExact(file, &t_len, sizeof(t_len));
+ tup = (HeapTuple) palloc(HEAPTUPLESIZE + t_len);
+ tup->t_data = (HeapTupleHeader) ((char *) tup + HEAPTUPLESIZE);
+ BufFileReadExact(file, tup->t_data, t_len);
+ tup->t_len = t_len;
+ ItemPointerSetInvalid(&tup->t_self);
+ tup->t_tableOid = RelationGetRelid(dest->rel);
- /*
- * Extract the tuple from the change. The tuple is copied here because
- * it might be assigned to 'tup_old', in which case it needs to
- * survive into the next iteration.
- */
- tup = get_changed_tuple(src);
-
- if (change.kind == CHANGE_UPDATE_OLD)
+ if (kind == CHANGE_UPDATE_OLD)
{
Assert(tup_old == NULL);
tup_old = tup;
}
- else if (change.kind == CHANGE_INSERT)
+ else if (kind == CHANGE_INSERT)
{
Assert(tup_old == NULL);
@@ -2726,12 +2870,11 @@ apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
pfree(tup);
}
- else if (change.kind == CHANGE_UPDATE_NEW ||
- change.kind == CHANGE_DELETE)
+ else if (kind == CHANGE_UPDATE_NEW || kind == CHANGE_DELETE)
{
HeapTuple tup_key;
- if (change.kind == CHANGE_UPDATE_NEW)
+ if (kind == CHANGE_UPDATE_NEW)
{
tup_key = tup_old != NULL ? tup_old : tup;
}
@@ -2748,7 +2891,7 @@ apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
if (tup_exist == NULL)
elog(ERROR, "failed to find target tuple");
- if (change.kind == CHANGE_UPDATE_NEW)
+ if (kind == CHANGE_UPDATE_NEW)
apply_concurrent_update(rel, tup, tup_exist, dest->iistate,
index_slot);
else
@@ -2763,26 +2906,19 @@ apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
pfree(tup);
}
else
- elog(ERROR, "unrecognized kind of change: %d", change.kind);
+ elog(ERROR, "unrecognized kind of change: %d", kind);
/*
* If a change was applied now, increment CID for next writes and
* update the snapshot so it sees the changes we've applied so far.
*/
- if (change.kind != CHANGE_UPDATE_OLD)
+ if (kind != CHANGE_UPDATE_OLD)
{
CommandCounterIncrement();
UpdateActiveSnapshotCommandId();
}
-
- /* TTSOpsMinimalTuple has .get_heap_tuple==NULL. */
- Assert(shouldFree);
- pfree(tup_change);
}
- tuplestore_clear(dstate->tstore);
- dstate->nchanges = 0;
-
/* Cleanup. */
ExecDropSingleTupleTableSlot(index_slot);
ExecDropSingleTupleTableSlot(ident_slot);
@@ -2957,25 +3093,59 @@ find_target_tuple(Relation rel, ChangeDest *dest, HeapTuple tup_key,
}
/*
- * Decode and apply concurrent changes.
+ * Decode and apply concurrent changes, up to (and including) the record whose
+ * LSN is 'end_of_wal'.
*/
static void
-process_concurrent_changes(LogicalDecodingContext *decoding_ctx,
- XLogRecPtr end_of_wal, ChangeDest *dest)
+process_concurrent_changes(XLogRecPtr end_of_wal, ChangeDest *dest, bool done)
{
- RepackDecodingState *dstate;
+ DecodingWorkerShared *shared;
+ char fname[MAXPGPATH];
+ BufFile *file;
pgstat_progress_update_param(PROGRESS_REPACK_PHASE,
PROGRESS_REPACK_PHASE_CATCH_UP);
- dstate = (RepackDecodingState *) decoding_ctx->output_writer_private;
+ /* Ask the worker for the file. */
+ shared = (DecodingWorkerShared *) dsm_segment_address(decoding_worker->seg);
+ SpinLockAcquire(&shared->mutex);
+ shared->lsn_upto = end_of_wal;
+ shared->done = done;
+ SpinLockRelease(&shared->mutex);
- repack_decode_concurrent_changes(decoding_ctx, end_of_wal);
+ /*
+ * The worker needs to finish processing of the current WAL record. Even
+ * if it's idle, it'll need to close the output file. Thus we're likely to
+ * wait, so prepare for sleep.
+ */
+ ConditionVariablePrepareToSleep(&shared->cv);
+ for (;;)
+ {
+ int last_exported;
- if (dstate->nchanges == 0)
- return;
+ SpinLockAcquire(&shared->mutex);
+ last_exported = shared->last_exported;
+ SpinLockRelease(&shared->mutex);
- apply_concurrent_changes(dstate, dest);
+ /*
+ * Has the worker exported the file we are waiting for?
+ */
+ if (last_exported == dest->file_seq)
+ break;
+
+ ConditionVariableSleep(&shared->cv, WAIT_EVENT_REPACK_WORKER_EXPORT);
+ }
+ ConditionVariableCancelSleep();
+
+ /* Open the file. */
+ DecodingWorkerFileName(fname, shared->relid, dest->file_seq);
+ file = BufFileOpenFileSet(&shared->sfs.fs, fname, O_RDONLY, false);
+ apply_concurrent_changes(file, dest);
+
+ BufFileClose(file);
+
+ /* Get ready for the next file. */
+ dest->file_seq++;
}
/*
@@ -3101,15 +3271,10 @@ cleanup_logical_decoding(LogicalDecodingContext *ctx)
dstate = (RepackDecodingState *) ctx->output_writer_private;
- ExecDropSingleTupleTableSlot(dstate->tsslot);
- FreeTupleDesc(dstate->tupdesc_change);
FreeTupleDesc(dstate->tupdesc);
- tuplestore_end(dstate->tstore);
-
FreeDecodingContext(ctx);
- ReplicationSlotRelease();
- ReplicationSlotDrop(NameStr(dstate->slotname), false);
+ ReplicationSlotDropAcquired();
pfree(dstate);
}
@@ -3123,7 +3288,6 @@ cleanup_logical_decoding(LogicalDecodingContext *ctx)
*/
static void
rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
- LogicalDecodingContext *decoding_ctx,
TransactionId frozenXid,
MultiXactId cutoffMulti)
{
@@ -3204,6 +3368,7 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
&chgdst.ident_index);
chgdst.ident_key = build_identity_key(ident_idx_new, OldHeap,
&chgdst.ident_key_nentries);
+ chgdst.file_seq = WORKER_FILE_SNAPSHOT + 1;
/*
* During testing, wait for another backend to perform concurrent data
@@ -3225,7 +3390,7 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
* hold AccessExclusiveLock. (Quite some amount of WAL could have been
* written during the data copying and index creation.)
*/
- process_concurrent_changes(decoding_ctx, end_of_wal, &chgdst);
+ process_concurrent_changes(end_of_wal, &chgdst, false);
/*
* Acquire AccessExclusiveLock on the table, its TOAST relation (if there
@@ -3306,8 +3471,11 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
XLogFlush(wal_insert_ptr);
end_of_wal = GetFlushRecPtr(NULL);
- /* Apply the concurrent changes again. */
- process_concurrent_changes(decoding_ctx, end_of_wal, &chgdst);
+ /*
+ * Apply the concurrent changes again. Indicate that the decoding worker
+ * won't be needed anymore.
+ */
+ process_concurrent_changes(end_of_wal, &chgdst, true);
/* Remember info about rel before closing OldHeap */
relpersistence = OldHeap->rd_rel->relpersistence;
@@ -3417,3 +3585,510 @@ build_new_indexes(Relation NewHeap, Relation OldHeap, List *OldIndexes)
return result;
}
+
+/*
+ * Try to start a background worker to perform logical decoding of data
+ * changes applied to relation while REPACK CONCURRENTLY is copying its
+ * contents to a new table.
+ */
+static void
+start_decoding_worker(Oid relid)
+{
+ Size size;
+ dsm_segment *seg;
+ DecodingWorkerShared *shared;
+ shm_mq *mq;
+ shm_mq_handle *mqh;
+ BackgroundWorker bgw;
+
+ /* Setup shared memory. */
+ size = BUFFERALIGN(offsetof(DecodingWorkerShared, error_queue)) +
+ BUFFERALIGN(REPACK_ERROR_QUEUE_SIZE);
+ seg = dsm_create(size, 0);
+ shared = (DecodingWorkerShared *) dsm_segment_address(seg);
+ shared->lsn_upto = InvalidXLogRecPtr;
+ shared->done = false;
+ SharedFileSetInit(&shared->sfs, seg);
+ shared->last_exported = -1;
+ SpinLockInit(&shared->mutex);
+ shared->dbid = MyDatabaseId;
+
+ /*
+ * This is the UserId set in cluster_rel(). Security context shouldn't be
+ * needed for decoding worker.
+ */
+ shared->roleid = GetUserId();
+ shared->relid = relid;
+ ConditionVariableInit(&shared->cv);
+ shared->backend_proc = MyProc;
+ shared->backend_pid = MyProcPid;
+ shared->backend_proc_number = MyProcNumber;
+
+ mq = shm_mq_create((char *) BUFFERALIGN(shared->error_queue),
+ REPACK_ERROR_QUEUE_SIZE);
+ shm_mq_set_receiver(mq, MyProc);
+ mqh = shm_mq_attach(mq, seg, NULL);
+
+ memset(&bgw, 0, sizeof(bgw));
+ snprintf(bgw.bgw_name, BGW_MAXLEN,
+ "REPACK decoding worker for relation \"%s\"",
+ get_rel_name(relid));
+ snprintf(bgw.bgw_type, BGW_MAXLEN, "REPACK decoding worker");
+ bgw.bgw_flags = BGWORKER_SHMEM_ACCESS |
+ BGWORKER_BACKEND_DATABASE_CONNECTION;
+ bgw.bgw_start_time = BgWorkerStart_RecoveryFinished;
+ bgw.bgw_restart_time = BGW_NEVER_RESTART;
+ snprintf(bgw.bgw_library_name, MAXPGPATH, "postgres");
+ snprintf(bgw.bgw_function_name, BGW_MAXLEN, "RepackWorkerMain");
+ bgw.bgw_main_arg = UInt32GetDatum(dsm_segment_handle(seg));
+ bgw.bgw_notify_pid = MyProcPid;
+
+ decoding_worker = palloc0_object(DecodingWorker);
+ if (!RegisterDynamicBackgroundWorker(&bgw, &decoding_worker->handle))
+ ereport(ERROR,
+ (errcode(ERRCODE_CONFIGURATION_LIMIT_EXCEEDED),
+ errmsg("out of background worker slots"),
+ errhint("You might need to increase \"%s\".", "max_worker_processes")));
+
+ decoding_worker->seg = seg;
+ decoding_worker->error_mqh = mqh;
+
+ /*
+ * The decoding setup must be done before the caller can have XID assigned
+ * for any reason, otherwise the worker might end up in a deadlock,
+ * waiting for the caller's transaction to end. Therefore wait here until
+ * the worker indicates that it has the logical decoding initialized.
+ */
+ ConditionVariablePrepareToSleep(&shared->cv);
+ for (;;)
+ {
+ bool initialized;
+
+ SpinLockAcquire(&shared->mutex);
+ initialized = shared->initialized;
+ SpinLockRelease(&shared->mutex);
+
+ if (initialized)
+ break;
+
+ ConditionVariableSleep(&shared->cv, WAIT_EVENT_REPACK_WORKER_EXPORT);
+ }
+ ConditionVariableCancelSleep();
+}
+
+/*
+ * Stop the decoding worker and cleanup the related resources.
+ *
+ * The worker stops on its own when it knows there is no more work to do, but
+ * we need to stop it explicitly at least on ERROR in the launching backend.
+ */
+static void
+stop_decoding_worker(void)
+{
+ BgwHandleStatus status;
+
+ /* Haven't reached the worker startup? */
+ if (decoding_worker == NULL)
+ return;
+
+ /* Could not register the worker? */
+ if (decoding_worker->handle == NULL)
+ return;
+
+ TerminateBackgroundWorker(decoding_worker->handle);
+ /* The worker should really exit before the REPACK command does. */
+ HOLD_INTERRUPTS();
+ status = WaitForBackgroundWorkerShutdown(decoding_worker->handle);
+ RESUME_INTERRUPTS();
+
+ if (status == BGWH_POSTMASTER_DIED)
+ ereport(FATAL,
+ (errcode(ERRCODE_ADMIN_SHUTDOWN),
+ errmsg("postmaster exited during REPACK command")));
+
+ shm_mq_detach(decoding_worker->error_mqh);
+
+ /*
+ * If we could not cancel the current sleep due to ERROR, do that before
+ * we detach from the shared memory the condition variable is located in.
+ * If we did not, the bgworker ERROR handling code would try and fail
+ * badly.
+ */
+ ConditionVariableCancelSleep();
+
+ dsm_detach(decoding_worker->seg);
+ pfree(decoding_worker);
+ decoding_worker = NULL;
+}
+
+/* Is this process a REPACK worker? */
+static bool is_repack_worker = false;
+
+static pid_t backend_pid;
+static ProcNumber backend_proc_number;
+
+/*
+ * See ParallelWorkerShutdown for details.
+ */
+static void
+RepackWorkerShutdown(int code, Datum arg)
+{
+ SendProcSignal(backend_pid,
+ PROCSIG_REPACK_MESSAGE,
+ backend_proc_number);
+
+ dsm_detach((dsm_segment *) DatumGetPointer(arg));
+}
+
+/* REPACK decoding worker entry point */
+void
+RepackWorkerMain(Datum main_arg)
+{
+ dsm_segment *seg;
+ DecodingWorkerShared *shared;
+ shm_mq *mq;
+ shm_mq_handle *mqh;
+
+ is_repack_worker = true;
+
+ /*
+ * Override the default bgworker_die() with die() so we can use
+ * CHECK_FOR_INTERRUPTS().
+ */
+ pqsignal(SIGTERM, die);
+ BackgroundWorkerUnblockSignals();
+
+ seg = dsm_attach(DatumGetUInt32(main_arg));
+ if (seg == NULL)
+ ereport(ERROR,
+ (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
+ errmsg("could not map dynamic shared memory segment")));
+
+ shared = (DecodingWorkerShared *) dsm_segment_address(seg);
+
+ /* Arrange to signal the leader if we exit. */
+ backend_pid = shared->backend_pid;
+ backend_proc_number = shared->backend_proc_number;
+ before_shmem_exit(RepackWorkerShutdown, PointerGetDatum(seg));
+
+ /*
+ * Join locking group - see the comments around the call of
+ * start_decoding_worker().
+ */
+ if (!BecomeLockGroupMember(shared->backend_proc, backend_pid))
+ /* The leader is not running anymore. */
+ return;
+
+ /*
+ * Setup a queue to send error messages to the backend that launched this
+ * worker.
+ */
+ mq = (shm_mq *) (char *) BUFFERALIGN(shared->error_queue);
+ shm_mq_set_sender(mq, MyProc);
+ mqh = shm_mq_attach(mq, seg, NULL);
+ pq_redirect_to_shm_mq(seg, mqh);
+ pq_set_parallel_leader(shared->backend_pid,
+ shared->backend_proc_number);
+
+ /* Connect to the database. */
+ BackgroundWorkerInitializeConnectionByOid(shared->dbid, shared->roleid, 0);
+
+ repack_worker_internal(seg);
+}
+
+static void
+repack_worker_internal(dsm_segment *seg)
+{
+ DecodingWorkerShared *shared;
+ LogicalDecodingContext *decoding_ctx;
+ SharedFileSet *sfs;
+ Snapshot snapshot;
+
+ /*
+ * Transaction is needed to open relation, and it also provides us with a
+ * resource owner.
+ */
+ StartTransactionCommand();
+
+ shared = (DecodingWorkerShared *) dsm_segment_address(seg);
+
+ /*
+ * Not sure the spinlock is needed here - the backend should not change
+ * anything in the shared memory until we have serialized the snapshot.
+ */
+ SpinLockAcquire(&shared->mutex);
+ Assert(XLogRecPtrIsInvalid(shared->lsn_upto));
+ sfs = &shared->sfs;
+ SpinLockRelease(&shared->mutex);
+
+ SharedFileSetAttach(sfs, seg);
+
+ /*
+ * Prepare to capture the concurrent data changes ourselves.
+ */
+ decoding_ctx = setup_logical_decoding(shared->relid);
+
+ /* Announce that we're ready. */
+ SpinLockAcquire(&shared->mutex);
+ shared->initialized = true;
+ SpinLockRelease(&shared->mutex);
+ ConditionVariableSignal(&shared->cv);
+
+ /* Build the initial snapshot and export it. */
+ snapshot = SnapBuildInitialSnapshotForRepack(decoding_ctx->snapshot_builder);
+ export_initial_snapshot(snapshot, shared);
+
+ /*
+ * Only historic snapshots should be used now. Do not let us restrict the
+ * progress of xmin horizon.
+ */
+ InvalidateCatalogSnapshot();
+
+ while (!decode_concurrent_changes(decoding_ctx, shared))
+ ;
+
+ /* Cleanup. */
+ cleanup_logical_decoding(decoding_ctx);
+ CommitTransactionCommand();
+}
+
+/*
+ * Make snapshot available to the backend that launched the decoding worker.
+ */
+static void
+export_initial_snapshot(Snapshot snapshot, DecodingWorkerShared *shared)
+{
+ char fname[MAXPGPATH];
+ BufFile *file;
+ Size snap_size;
+ char *snap_space;
+
+ snap_size = EstimateSnapshotSpace(snapshot);
+ snap_space = (char *) palloc(snap_size);
+ SerializeSnapshot(snapshot, snap_space);
+ FreeSnapshot(snapshot);
+
+ DecodingWorkerFileName(fname, shared->relid, shared->last_exported + 1);
+ file = BufFileCreateFileSet(&shared->sfs.fs, fname);
+ /* To make restoration easier, write the snapshot size first. */
+ BufFileWrite(file, &snap_size, sizeof(snap_size));
+ BufFileWrite(file, snap_space, snap_size);
+ pfree(snap_space);
+ BufFileClose(file);
+
+ /* Increase the counter to tell the backend that the file is available. */
+ SpinLockAcquire(&shared->mutex);
+ shared->last_exported++;
+ SpinLockRelease(&shared->mutex);
+ ConditionVariableSignal(&shared->cv);
+}
+
+/*
+ * Get the initial snapshot from the decoding worker.
+ */
+static Snapshot
+get_initial_snapshot(DecodingWorker *worker)
+{
+ DecodingWorkerShared *shared;
+ char fname[MAXPGPATH];
+ BufFile *file;
+ Size snap_size;
+ char *snap_space;
+ Snapshot snapshot;
+
+ shared = (DecodingWorkerShared *) dsm_segment_address(worker->seg);
+
+ /*
+ * The worker needs to initialize the logical decoding, which usually
+ * takes some time. Therefore it makes sense to prepare for the sleep
+ * first.
+ */
+ ConditionVariablePrepareToSleep(&shared->cv);
+ for (;;)
+ {
+ int last_exported;
+
+ SpinLockAcquire(&shared->mutex);
+ last_exported = shared->last_exported;
+ SpinLockRelease(&shared->mutex);
+
+ /*
+ * Has the worker exported the file we are waiting for?
+ */
+ if (last_exported == WORKER_FILE_SNAPSHOT)
+ break;
+
+ ConditionVariableSleep(&shared->cv, WAIT_EVENT_REPACK_WORKER_EXPORT);
+ }
+ ConditionVariableCancelSleep();
+
+ /* Read the snapshot from a file. */
+ DecodingWorkerFileName(fname, shared->relid, WORKER_FILE_SNAPSHOT);
+ file = BufFileOpenFileSet(&shared->sfs.fs, fname, O_RDONLY, false);
+ BufFileReadExact(file, &snap_size, sizeof(snap_size));
+ snap_space = (char *) palloc(snap_size);
+ BufFileReadExact(file, snap_space, snap_size);
+ BufFileClose(file);
+
+ /* Restore it. */
+ snapshot = RestoreSnapshot(snap_space);
+ pfree(snap_space);
+
+ return snapshot;
+}
+
+bool
+IsRepackWorker(void)
+{
+ return is_repack_worker;
+}
+
+/*
+ * Handle receipt of an interrupt indicating a repack worker message.
+ *
+ * Note: this is called within a signal handler! All we can do is set
+ * a flag that will cause the next CHECK_FOR_INTERRUPTS() to invoke
+ * ProcessRepackMessages().
+ */
+void
+HandleRepackMessageInterrupt(void)
+{
+ InterruptPending = true;
+ RepackMessagePending = true;
+ SetLatch(MyLatch);
+}
+
+/*
+ * Process any queued protocol messages received from parallel workers.
+ */
+void
+ProcessRepackMessages(void)
+{
+ MemoryContext oldcontext;
+
+ static MemoryContext hpm_context = NULL;
+
+ /*
+ * Nothing to do if we haven't launched the worker yet or have already
+ * terminated it.
+ */
+ if (decoding_worker == NULL)
+ return;
+
+ /*
+ * This is invoked from ProcessInterrupts(), and since some of the
+ * functions it calls contain CHECK_FOR_INTERRUPTS(), there is a potential
+ * for recursive calls if more signals are received while this runs. It's
+ * unclear that recursive entry would be safe, and it doesn't seem useful
+ * even if it is safe, so let's block interrupts until done.
+ */
+ HOLD_INTERRUPTS();
+
+ /*
+ * Moreover, CurrentMemoryContext might be pointing almost anywhere. We
+ * don't want to risk leaking data into long-lived contexts, so let's do
+ * our work here in a private context that we can reset on each use.
+ */
+ if (hpm_context == NULL) /* first time through? */
+ hpm_context = AllocSetContextCreate(TopMemoryContext,
+ "ProcessRepackMessages",
+ ALLOCSET_DEFAULT_SIZES);
+ else
+ MemoryContextReset(hpm_context);
+
+ oldcontext = MemoryContextSwitchTo(hpm_context);
+
+ /* OK to process messages. Reset the flag saying there are more to do. */
+ RepackMessagePending = false;
+
+ /*
+ * Read as many messages as we can from each worker, but stop when no more
+ * messages can be read from the worker without blocking.
+ */
+ while (true)
+ {
+ shm_mq_result res;
+ Size nbytes;
+ void *data;
+
+ res = shm_mq_receive(decoding_worker->error_mqh, &nbytes,
+ &data, true);
+ if (res == SHM_MQ_WOULD_BLOCK)
+ break;
+ else if (res == SHM_MQ_SUCCESS)
+ {
+ StringInfoData msg;
+
+ initStringInfo(&msg);
+ appendBinaryStringInfo(&msg, data, nbytes);
+ ProcessRepackMessage(&msg);
+ pfree(msg.data);
+ }
+ else
+ {
+ /*
+ * The decoding worker is special in that it exits as soon as it
+ * has its work done. Thus the DETACHED result code is fine.
+ */
+ Assert(res == SHM_MQ_DETACHED);
+
+ break;
+ }
+ }
+
+ MemoryContextSwitchTo(oldcontext);
+
+ /* Might as well clear the context on our way out */
+ MemoryContextReset(hpm_context);
+
+ RESUME_INTERRUPTS();
+}
+
+/*
+ * Process a single protocol message received from a single parallel worker.
+ */
+static void
+ProcessRepackMessage(StringInfo msg)
+{
+ char msgtype;
+
+ msgtype = pq_getmsgbyte(msg);
+
+ switch (msgtype)
+ {
+ case PqMsg_ErrorResponse:
+ case PqMsg_NoticeResponse:
+ {
+ ErrorData edata;
+
+ /* Parse ErrorResponse or NoticeResponse. */
+ pq_parse_errornotice(msg, &edata);
+
+ /* Death of a worker isn't enough justification for suicide. */
+ edata.elevel = Min(edata.elevel, ERROR);
+
+ /*
+ * If desired, add a context line to show that this is a
+ * message propagated from a parallel worker. Otherwise, it
+ * can sometimes be confusing to understand what actually
+ * happened.
+ */
+ if (edata.context)
+ edata.context = psprintf("%s\n%s", edata.context,
+ _("decoding worker"));
+ else
+ edata.context = pstrdup(_("decoding worker"));
+
+ /* Rethrow error or print notice. */
+ ThrowErrorData(&edata);
+
+ break;
+ }
+
+ default:
+ {
+ elog(ERROR, "unrecognized message type received from decoding worker: %c (message length %d bytes)",
+ msgtype, msg->len);
+ }
+ }
+}
diff --git a/src/backend/libpq/pqmq.c b/src/backend/libpq/pqmq.c
index 6e4bbfb5aa1..42f6fa472c5 100644
--- a/src/backend/libpq/pqmq.c
+++ b/src/backend/libpq/pqmq.c
@@ -14,6 +14,7 @@
#include "postgres.h"
#include "access/parallel.h"
+#include "commands/cluster.h"
#include "libpq/libpq.h"
#include "libpq/pqformat.h"
#include "libpq/pqmq.h"
@@ -175,6 +176,10 @@ mq_putmessage(char msgtype, const char *s, size_t len)
SendProcSignal(pq_mq_parallel_leader_pid,
PROCSIG_PARALLEL_APPLY_MESSAGE,
pq_mq_parallel_leader_proc_number);
+ else if (IsRepackWorker())
+ SendProcSignal(pq_mq_parallel_leader_pid,
+ PROCSIG_REPACK_MESSAGE,
+ pq_mq_parallel_leader_proc_number);
else
{
Assert(IsParallelWorker());
diff --git a/src/backend/postmaster/bgworker.c b/src/backend/postmaster/bgworker.c
index 261ccd3f59c..09c58371e8e 100644
--- a/src/backend/postmaster/bgworker.c
+++ b/src/backend/postmaster/bgworker.c
@@ -13,6 +13,7 @@
#include "postgres.h"
#include "access/parallel.h"
+#include "commands/cluster.h"
#include "libpq/pqsignal.h"
#include "miscadmin.h"
#include "pgstat.h"
@@ -142,6 +143,10 @@ static const struct
{
.fn_name = "SequenceSyncWorkerMain",
.fn_addr = SequenceSyncWorkerMain
+ },
+ {
+ .fn_name = "RepackWorkerMain",
+ .fn_addr = RepackWorkerMain
}
};
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index 603a2b94d05..7651b187418 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -194,7 +194,11 @@ StartupDecodingContext(List *output_plugin_options,
ctx->slot = slot;
- ctx->reader = XLogReaderAllocate(wal_segment_size, NULL, xl_routine, ctx);
+ /*
+ * TODO A separate patch for PG core, unless there's really a reason to
+ * pass ctx for private_data (May extensions expect ctx?).
+ */
+ ctx->reader = XLogReaderAllocate(wal_segment_size, NULL, xl_routine, NULL);
if (!ctx->reader)
ereport(ERROR,
(errcode(ERRCODE_OUT_OF_MEMORY),
diff --git a/src/backend/replication/pgoutput_repack/pgoutput_repack.c b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
index 6b54ea040ac..3a109dbfaff 100644
--- a/src/backend/replication/pgoutput_repack/pgoutput_repack.c
+++ b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
@@ -167,17 +167,13 @@ store_change(LogicalDecodingContext *ctx, ConcurrentChangeKind kind,
HeapTuple tuple)
{
RepackDecodingState *dstate;
- char *change_raw;
- ConcurrentChange change;
+ char kind_byte = (char) kind;
bool flattened = false;
- Size size;
- Datum values[1];
- bool isnull[1];
- char *dst;
dstate = (RepackDecodingState *) ctx->output_writer_private;
- size = VARHDRSZ + SizeOfConcurrentChange;
+ /* Store the change kind. */
+ BufFileWrite(dstate->file, &kind_byte, 1);
/*
* ReorderBufferCommit() stores the TOAST chunks in its private memory
@@ -194,46 +190,12 @@ store_change(LogicalDecodingContext *ctx, ConcurrentChangeKind kind,
tuple = toast_flatten_tuple(tuple, dstate->tupdesc);
flattened = true;
}
+ /* Store the tuple size ... */
+ BufFileWrite(dstate->file, &tuple->t_len, sizeof(tuple->t_len));
+ /* ... and the tuple itself. */
+ BufFileWrite(dstate->file, tuple->t_data, tuple->t_len);
- size += tuple->t_len;
- if (size >= MaxAllocSize)
- elog(ERROR, "Change is too big.");
-
- /* Construct the change. */
- change_raw = (char *) palloc0(size);
- SET_VARSIZE(change_raw, size);
-
- /*
- * Since the varlena alignment might not be sufficient for the structure,
- * set the fields in a local instance and remember where it should
- * eventually be copied.
- */
- change.kind = kind;
- dst = (char *) VARDATA(change_raw);
-
- /*
- * Copy the tuple.
- *
- * Note: change->tup_data.t_data must be fixed on retrieval!
- */
- memcpy(&change.tup_data, tuple, sizeof(HeapTupleData));
- memcpy(dst, &change, SizeOfConcurrentChange);
- dst += SizeOfConcurrentChange;
- memcpy(dst, tuple->t_data, tuple->t_len);
-
- /* The data has been copied. */
+ /* Free the flat copy if created above. */
if (flattened)
pfree(tuple);
-
- /* Store as tuple of 1 bytea column. */
- values[0] = PointerGetDatum(change_raw);
- isnull[0] = false;
- tuplestore_putvalues(dstate->tstore, dstate->tupdesc_change,
- values, isnull);
-
- /* Accounting. */
- dstate->nchanges++;
-
- /* Cleanup. */
- pfree(change_raw);
}
diff --git a/src/backend/storage/ipc/procsignal.c b/src/backend/storage/ipc/procsignal.c
index 8e56922dcea..6f9e7a7aab7 100644
--- a/src/backend/storage/ipc/procsignal.c
+++ b/src/backend/storage/ipc/procsignal.c
@@ -19,6 +19,7 @@
#include "access/parallel.h"
#include "commands/async.h"
+#include "commands/cluster.h"
#include "miscadmin.h"
#include "pgstat.h"
#include "port/pg_bitutils.h"
@@ -697,6 +698,9 @@ procsignal_sigusr1_handler(SIGNAL_ARGS)
if (CheckProcSignal(PROCSIG_PARALLEL_APPLY_MESSAGE))
HandleParallelApplyMessageInterrupt();
+ if (CheckProcSignal(PROCSIG_REPACK_MESSAGE))
+ HandleRepackMessageInterrupt();
+
if (CheckProcSignal(PROCSIG_RECOVERY_CONFLICT_DATABASE))
HandleRecoveryConflictInterrupt(PROCSIG_RECOVERY_CONFLICT_DATABASE);
diff --git a/src/backend/tcop/postgres.c b/src/backend/tcop/postgres.c
index 02e9aaa6bca..e08bf56f11e 100644
--- a/src/backend/tcop/postgres.c
+++ b/src/backend/tcop/postgres.c
@@ -36,6 +36,7 @@
#include "access/xact.h"
#include "catalog/pg_type.h"
#include "commands/async.h"
+#include "commands/cluster.h"
#include "commands/event_trigger.h"
#include "commands/explain_state.h"
#include "commands/prepare.h"
@@ -3520,6 +3521,9 @@ ProcessInterrupts(void)
if (ParallelApplyMessagePending)
ProcessParallelApplyMessages();
+
+ if (RepackMessagePending)
+ ProcessRepackMessages();
}
/*
diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt
index 4aa864fe3c3..b00bd794759 100644
--- a/src/backend/utils/activity/wait_event_names.txt
+++ b/src/backend/utils/activity/wait_event_names.txt
@@ -154,6 +154,7 @@ RECOVERY_CONFLICT_SNAPSHOT "Waiting for recovery conflict resolution for a vacuu
RECOVERY_CONFLICT_TABLESPACE "Waiting for recovery conflict resolution for dropping a tablespace."
RECOVERY_END_COMMAND "Waiting for <xref linkend="guc-recovery-end-command"/> to complete."
RECOVERY_PAUSE "Waiting for recovery to be resumed."
+REPACK_WORKER_EXPORT "Waiting for decoding worker to export a new output file."
REPLICATION_ORIGIN_DROP "Waiting for a replication origin to become inactive so it can be dropped."
REPLICATION_SLOT_DROP "Waiting for a replication slot to become inactive so it can be dropped."
RESTORE_COMMAND "Waiting for <xref linkend="guc-restore-command"/> to complete."
diff --git a/src/include/access/tableam.h b/src/include/access/tableam.h
index 14928cd04a1..f1005afde9d 100644
--- a/src/include/access/tableam.h
+++ b/src/include/access/tableam.h
@@ -22,7 +22,6 @@
#include "access/xact.h"
#include "commands/vacuum.h"
#include "executor/tuptable.h"
-#include "replication/logical.h"
#include "storage/read_stream.h"
#include "utils/rel.h"
#include "utils/snapshot.h"
@@ -631,7 +630,6 @@ typedef struct TableAmRoutine
bool use_sort,
TransactionId OldestXmin,
Snapshot snapshot,
- LogicalDecodingContext *decoding_ctx,
TransactionId *xid_cutoff,
MultiXactId *multi_cutoff,
double *num_tuples,
@@ -1663,8 +1661,6 @@ table_relation_copy_data(Relation rel, const RelFileLocator *newrlocator)
* - *multi_cutoff - ditto
* - snapshot - if != NULL, ignore data changes done by transactions that this
* (MVCC) snapshot considers still in-progress or in the future.
- * - decoding_ctx - logical decoding context, to capture concurrent data
- * changes.
*
* Output parameters:
* - *xid_cutoff - rel's new relfrozenxid value, may be invalid
@@ -1678,7 +1674,6 @@ table_relation_copy_for_cluster(Relation OldTable, Relation NewTable,
bool use_sort,
TransactionId OldestXmin,
Snapshot snapshot,
- LogicalDecodingContext *decoding_ctx,
TransactionId *xid_cutoff,
MultiXactId *multi_cutoff,
double *num_tuples,
@@ -1687,7 +1682,7 @@ table_relation_copy_for_cluster(Relation OldTable, Relation NewTable,
{
OldTable->rd_tableam->relation_copy_for_cluster(OldTable, NewTable, OldIndex,
use_sort, OldestXmin,
- snapshot, decoding_ctx,
+ snapshot,
xid_cutoff, multi_cutoff,
num_tuples, tups_vacuumed,
tups_recently_dead);
diff --git a/src/include/commands/cluster.h b/src/include/commands/cluster.h
index 6a5c476294a..1b05d5d418b 100644
--- a/src/include/commands/cluster.h
+++ b/src/include/commands/cluster.h
@@ -17,11 +17,13 @@
#include "nodes/parsenodes.h"
#include "parser/parse_node.h"
#include "replication/decode.h"
+#include "postmaster/bgworker.h"
#include "replication/logical.h"
+#include "storage/buffile.h"
#include "storage/lock.h"
+#include "storage/shm_mq.h"
#include "utils/relcache.h"
#include "utils/resowner.h"
-#include "utils/tuplestore.h"
/* flag bits for ClusterParams->options */
@@ -44,6 +46,9 @@ typedef struct ClusterParams
* The following definitions are used by REPACK CONCURRENTLY.
*/
+/*
+ * Stored as a single byte in the output file.
+ */
typedef enum
{
CHANGE_INSERT,
@@ -52,68 +57,30 @@ typedef enum
CHANGE_DELETE
} ConcurrentChangeKind;
-typedef struct ConcurrentChange
-{
- /* See the enum above. */
- ConcurrentChangeKind kind;
-
- /*
- * The actual tuple.
- *
- * The tuple data follows the ConcurrentChange structure. Before use make
- * sure the tuple is correctly aligned (ConcurrentChange can be stored as
- * bytea) and that tuple->t_data is fixed.
- */
- HeapTupleData tup_data;
-} ConcurrentChange;
-
-#define SizeOfConcurrentChange (offsetof(ConcurrentChange, tup_data) + \
- sizeof(HeapTupleData))
-
/*
* Logical decoding state.
*
- * Here we store the data changes that we decode from WAL while the table
- * contents is being copied to a new storage. Also the necessary metadata
- * needed to apply these changes to the table is stored here.
+ * The output plugin uses it to store the data changes that it decodes from
+ * WAL while the table contents is being copied to a new storage.
*/
typedef struct RepackDecodingState
{
/* The relation whose changes we're decoding. */
Oid relid;
- /* Replication slot name. */
- NameData slotname;
-
- /*
- * Decoded changes are stored here. Although we try to avoid excessive
- * batches, it can happen that the changes need to be stored to disk. The
- * tuplestore does this transparently.
- */
- Tuplestorestate *tstore;
-
- /* The current number of changes in tstore. */
- double nchanges;
-
- /*
- * Descriptor to store the ConcurrentChange structure serialized (bytea).
- * We can't store the tuple directly because tuplestore only supports
- * minimum tuple and we may need to transfer OID system column from the
- * output plugin. Also we need to transfer the change kind, so it's better
- * to put everything in the structure than to use 2 tuplestores "in
- * parallel".
- */
- TupleDesc tupdesc_change;
-
- /* Tuple descriptor needed to update indexes. */
+ /* Tuple descriptor of the relation being processed. */
TupleDesc tupdesc;
- /* Slot to retrieve data from tstore. */
- TupleTableSlot *tsslot;
-
- ResourceOwner resowner;
+ /* The current output file. */
+ BufFile *file;
} RepackDecodingState;
+extern PGDLLIMPORT volatile sig_atomic_t RepackMessagePending;
+
+extern bool IsRepackWorker(void);
+extern void HandleRepackMessageInterrupt(void);
+extern void ProcessRepackMessages(void);
+
extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel);
extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid,
@@ -136,6 +103,6 @@ extern void finish_heap_swap(Oid OIDOldHeap, Oid OIDNewHeap,
extern bool am_decoding_for_repack(void);
extern bool change_useless_for_repack(XLogRecordBuffer *buf);
-extern void repack_decode_concurrent_changes(LogicalDecodingContext *ctx,
- XLogRecPtr end_of_wal);
+
+extern void RepackWorkerMain(Datum main_arg);
#endif /* CLUSTER_H */
diff --git a/src/include/storage/procsignal.h b/src/include/storage/procsignal.h
index e52b8eb7697..3ef35ca6b80 100644
--- a/src/include/storage/procsignal.h
+++ b/src/include/storage/procsignal.h
@@ -36,6 +36,7 @@ typedef enum
PROCSIG_BARRIER, /* global barrier interrupt */
PROCSIG_LOG_MEMORY_CONTEXT, /* ask backend to log the memory contexts */
PROCSIG_PARALLEL_APPLY_MESSAGE, /* Message from parallel apply workers */
+ PROCSIG_REPACK_MESSAGE, /* Message from repack worker */
/* Recovery conflict reasons */
PROCSIG_RECOVERY_CONFLICT_FIRST,
diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list
index 2676823d2a7..84c7b680dc5 100644
--- a/src/tools/pgindent/typedefs.list
+++ b/src/tools/pgindent/typedefs.list
@@ -498,7 +498,6 @@ CompressFileHandle
CompressionLocation
CompressorState
ComputeXidHorizonsResult
-ConcurrentChange
ConcurrentChangeKind
ConditionVariable
ConditionVariableMinimallyPadded
@@ -638,6 +637,8 @@ DeclareCursorStmt
DecodedBkpBlock
DecodedXLogRecord
DecodingOutputState
+DecodingWorker
+DecodingWorkerShared
DefElem
DefElemAction
DefaultACLInfo
--
2.47.3
--r7amdyq3rmktb267--
^ permalink raw reply [nested|flat] 55+ messages in thread
* [PATCH v33 5/5] Use background worker to do logical decoding.
@ 2026-01-27 10:48 Antonin Houska <[email protected]>
0 siblings, 0 replies; 55+ messages in thread
From: Antonin Houska @ 2026-01-27 10:48 UTC (permalink / raw)
If the backend performing REPACK (CONCURRENTLY) does both data copying and
logical decoding, it has to "travel in time" back and forth and therefore it
has to invalidate system caches quite a few times. (The copying and the
decoding work with different catalog snapshots.) As the decoding worker has
separate caches, the switching is not necessary.
Without the worker, it'd also be difficult to switch between potentially long
running tasks like index build and WAL decoding. (No decoding during that time
at all can suspend archiving / recycling of WAL segments for some time, which
in turn may result in full disk.)
Another problem is that, after having acquired AccessExclusiveLock (in order
to swap the files), the backend needs to both decode and apply the data
changes that took place while it was waiting for the lock. With the decoding
worker, the decoding runs all the time, so the backend only needs to apply the
changes. This can reduce the time the exclusive lock is held for.
Note that the code added in order to handle ERRORs in the background worker
almost duplicates the existing code that does the same for other types of
workers (See ProcessParallelMessages() and
ProcessParallelApplyMessages()). Refactoring of the existing code might be
useful, to reduce the duplication.
---
src/backend/access/heap/heapam_handler.c | 44 -
src/backend/commands/cluster.c | 1183 +++++++++++++----
src/backend/libpq/pqmq.c | 5 +
src/backend/postmaster/bgworker.c | 5 +
src/backend/replication/logical/logical.c | 6 +-
.../pgoutput_repack/pgoutput_repack.c | 54 +-
src/backend/storage/ipc/procsignal.c | 4 +
src/backend/tcop/postgres.c | 4 +
.../utils/activity/wait_event_names.txt | 1 +
src/include/access/tableam.h | 7 +-
src/include/commands/cluster.h | 71 +-
src/include/storage/procsignal.h | 1 +
src/tools/pgindent/typedefs.list | 3 +-
13 files changed, 984 insertions(+), 404 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index 908f1ef66c6..8589b3c940e 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -33,7 +33,6 @@
#include "catalog/index.h"
#include "catalog/storage.h"
#include "catalog/storage_xlog.h"
-#include "commands/cluster.h"
#include "commands/progress.h"
#include "executor/executor.h"
#include "miscadmin.h"
@@ -688,7 +687,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
Relation OldIndex, bool use_sort,
TransactionId OldestXmin,
Snapshot snapshot,
- LogicalDecodingContext *decoding_ctx,
TransactionId *xid_cutoff,
MultiXactId *multi_cutoff,
double *num_tuples,
@@ -710,7 +708,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
BufferHeapTupleTableSlot *hslot;
BlockNumber prev_cblock = InvalidBlockNumber;
bool concurrent = snapshot != NULL;
- XLogRecPtr end_of_wal_prev = GetFlushRecPtr(NULL);
/* Remember if it's a system catalog */
is_system_catalog = IsSystemRelation(OldHeap);
@@ -971,31 +968,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
ct_val[1] = *num_tuples;
pgstat_progress_update_multi_param(2, ct_index, ct_val);
}
-
- /*
- * Process the WAL produced by the load, as well as by other
- * transactions, so that the replication slot can advance and WAL does
- * not pile up. Use wal_segment_size as a threshold so that we do not
- * introduce the decoding overhead too often.
- *
- * Of course, we must not apply the changes until the initial load has
- * completed.
- *
- * Note that our insertions into the new table should not be decoded
- * as we (intentionally) do not write the logical decoding specific
- * information to WAL.
- */
- if (concurrent)
- {
- XLogRecPtr end_of_wal;
-
- end_of_wal = GetFlushRecPtr(NULL);
- if ((end_of_wal - end_of_wal_prev) > wal_segment_size)
- {
- repack_decode_concurrent_changes(decoding_ctx, end_of_wal);
- end_of_wal_prev = end_of_wal;
- }
- }
}
if (indexScan != NULL)
@@ -1041,22 +1013,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
-
- /*
- * Try to keep the amount of not-yet-decoded WAL small, like
- * above.
- */
- if (concurrent)
- {
- XLogRecPtr end_of_wal;
-
- end_of_wal = GetFlushRecPtr(NULL);
- if ((end_of_wal - end_of_wal_prev) > wal_segment_size)
- {
- repack_decode_concurrent_changes(decoding_ctx, end_of_wal);
- end_of_wal_prev = end_of_wal;
- }
- }
}
tuplesort_end(tuplesort);
diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c
index 03ccf10b782..e988a7a7296 100644
--- a/src/backend/commands/cluster.c
+++ b/src/backend/commands/cluster.c
@@ -12,12 +12,13 @@
* In concurrent mode, we lock the table with only ShareUpdateExclusiveLock,
* then do an initial copy as above. However, while the tuples are being
* copied, concurrent transactions could modify the table. To cope with those
- * changes, we rely on logical decoding to obtain them from WAL. The changes
- * are accumulated in a tuplestore. Once the initial copy is complete, we
- * read the changes from the tuplestore and re-apply them on the new heap.
- * Then we upgrade our ShareUpdateExclusiveLock to AccessExclusiveLock and
- * swap the relfilenodes. This way, the time we hold a strong lock on the
- * table is much reduced, and the bloat is eliminated.
+ * changes, we rely on logical decoding to obtain them from WAL. A bgworker
+ * consumes WAL while the initial copy is ongoing (to prevent excessive WAL
+ * from being reserved), and accumulates the changes in a file. Once the
+ * initial copy is complete, we read the changes from the file and re-apply
+ * them on the new heap. Then we upgrade our ShareUpdateExclusiveLock to
+ * AccessExclusiveLock and swap the relfilenodes. This way, the time we hold
+ * a strong lock on the table is much reduced, and the bloat is eliminated.
*
* There is hardly anything left of Paul Brown's original implementation...
*
@@ -45,6 +46,7 @@
#include "access/xlog_internal.h"
#include "access/xloginsert.h"
#include "access/xlogutils.h"
+#include "access/xlogwait.h"
#include "catalog/catalog.h"
#include "catalog/dependency.h"
#include "catalog/heap.h"
@@ -61,6 +63,8 @@
#include "commands/tablecmds.h"
#include "commands/vacuum.h"
#include "executor/executor.h"
+#include "libpq/pqformat.h"
+#include "libpq/pqmq.h"
#include "miscadmin.h"
#include "optimizer/optimizer.h"
#include "pgstat.h"
@@ -71,6 +75,8 @@
#include "storage/ipc.h"
#include "storage/lmgr.h"
#include "storage/predicate.h"
+#include "storage/procsignal.h"
+#include "tcop/tcopprot.h"
#include "utils/acl.h"
#include "utils/fmgroids.h"
#include "utils/guc.h"
@@ -117,6 +123,12 @@ typedef struct IndexInsertState
/* The WAL segment being decoded. */
static XLogSegNo repack_current_segment = 0;
+/*
+ * The first file exported by the decoding worker must contain a snapshot, the
+ * following ones contain the data changes.
+ */
+#define WORKER_FILE_SNAPSHOT 0
+
/*
* Information needed to apply concurrent data changes.
*/
@@ -136,8 +148,113 @@ typedef struct ChangeDest
/* Needed to update indexes of rel_dst. */
IndexInsertState *iistate;
+
+ /*
+ * Sequential number of the file containing the changes.
+ *
+ * TODO This field makes the structure name less descriptive. Should we
+ * rename it, e.g. to ChangeApplyInfo?
+ */
+ int file_seq;
} ChangeDest;
+/*
+ * Layout of shared memory used for communication between backend and the
+ * worker that performs logical decoding of data changes
+ */
+typedef struct DecodingWorkerShared
+{
+ /* Is the decoding initialized? */
+ bool initialized;
+
+ /*
+ * Once the worker has reached this LSN, it should close the current
+ * output file and either create a new one or exit, according to the field
+ * 'done'. If the value is InvalidXLogRecPtr, the worker should decode all
+ * the WAL available and keep checking this field. It is ok if the worker
+ * had already decoded records whose LSN is >= lsn_upto before this field
+ * has been set.
+ */
+ XLogRecPtr lsn_upto;
+
+ /* Exit after closing the current file? */
+ bool done;
+
+ /* The output is stored here. */
+ SharedFileSet sfs;
+
+ /* Number of the last file exported by the worker. */
+ int last_exported;
+
+ /* Synchronize access to the fields above. */
+ slock_t mutex;
+
+ /* Database to connect to. */
+ Oid dbid;
+
+ /* Role to connect as. */
+ Oid roleid;
+
+ /* Decode data changes of this relation. */
+ Oid relid;
+
+ /* The backend uses this to wait for the worker. */
+ ConditionVariable cv;
+
+ /* Info to signal the backend. */
+ PGPROC *backend_proc;
+ pid_t backend_pid;
+ ProcNumber backend_proc_number;
+
+ /* Error queue. */
+ shm_mq *error_mq;
+
+ /*
+ * Memory the queue is located in.
+ *
+ * For considerations on the value see the comments of
+ * PARALLEL_ERROR_QUEUE_SIZE.
+ */
+#define REPACK_ERROR_QUEUE_SIZE 16384
+ char error_queue[FLEXIBLE_ARRAY_MEMBER];
+} DecodingWorkerShared;
+
+/*
+ * Generate worker's output file name. If relations of the same 'relid' happen
+ * to be processed at the same time, they must be from different databases and
+ * therefore different backends must be involved. (PID is already present in
+ * the fileset name.)
+ */
+static inline void
+DecodingWorkerFileName(char *fname, Oid relid, uint32 seq)
+{
+ snprintf(fname, MAXPGPATH, "%u-%u", relid, seq);
+}
+
+/*
+ * Backend-local information to control the decoding worker.
+ */
+typedef struct DecodingWorker
+{
+ /* The worker. */
+ BackgroundWorkerHandle *handle;
+
+ /* DecodingWorkerShared is in this segment. */
+ dsm_segment *seg;
+
+ /* Handle of the error queue. */
+ shm_mq_handle *error_mqh;
+} DecodingWorker;
+
+/* Pointer to currently running decoding worker. */
+static DecodingWorker *decoding_worker = NULL;
+
+/*
+ * Is there a message sent by a repack worker that the backend needs to
+ * receive?
+ */
+volatile sig_atomic_t RepackMessagePending = false;
+
static bool cluster_rel_recheck(RepackCommand cmd, Relation OldHeap,
Oid indexOid, Oid userid, LOCKMODE lmode,
int options);
@@ -145,7 +262,7 @@ static void check_repack_concurrently_requirements(Relation rel);
static void rebuild_relation(Relation OldHeap, Relation index, bool verbose,
bool concurrent);
static void copy_table_data(Relation NewHeap, Relation OldHeap, Relation OldIndex,
- Snapshot snapshot, LogicalDecodingContext *decoding_ctx,
+ Snapshot snapshot,
bool verbose,
bool *pSwapToastByContent,
TransactionId *pFreezeXid,
@@ -158,12 +275,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd,
static bool cluster_is_permitted_for_relation(RepackCommand cmd,
Oid relid, Oid userid);
-static void begin_concurrent_repack(Relation rel);
-static void end_concurrent_repack(void);
static LogicalDecodingContext *setup_logical_decoding(Oid relid);
-static HeapTuple get_changed_tuple(char *change);
-static void apply_concurrent_changes(RepackDecodingState *dstate,
- ChangeDest *dest);
+static bool decode_concurrent_changes(LogicalDecodingContext *ctx,
+ DecodingWorkerShared *shared);
+static void apply_concurrent_changes(BufFile *file, ChangeDest *dest);
static void apply_concurrent_insert(Relation rel, HeapTuple tup,
IndexInsertState *iistate,
TupleTableSlot *index_slot);
@@ -175,9 +290,9 @@ static void apply_concurrent_delete(Relation rel, HeapTuple tup_target);
static HeapTuple find_target_tuple(Relation rel, ChangeDest *dest,
HeapTuple tup_key,
TupleTableSlot *ident_slot);
-static void process_concurrent_changes(LogicalDecodingContext *decoding_ctx,
- XLogRecPtr end_of_wal,
- ChangeDest *dest);
+static void process_concurrent_changes(XLogRecPtr end_of_wal,
+ ChangeDest *dest,
+ bool done);
static IndexInsertState *get_index_insert_state(Relation relation,
Oid ident_index_id,
Relation *ident_index_p);
@@ -186,7 +301,6 @@ static ScanKey build_identity_key(Oid ident_idx_oid, Relation rel_src,
static void free_index_insert_state(IndexInsertState *iistate);
static void cleanup_logical_decoding(LogicalDecodingContext *ctx);
static void rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
- LogicalDecodingContext *decoding_ctx,
TransactionId frozenXid,
MultiXactId cutoffMulti);
static List *build_new_indexes(Relation NewHeap, Relation OldHeap, List *OldIndexes);
@@ -196,6 +310,13 @@ static Relation process_single_relation(RepackStmt *stmt,
ClusterParams *params);
static Oid determine_clustered_index(Relation rel, bool usingindex,
const char *indexname);
+static void start_decoding_worker(Oid relid);
+static void stop_decoding_worker(void);
+static void repack_worker_internal(dsm_segment *seg);
+static void export_initial_snapshot(Snapshot snapshot,
+ DecodingWorkerShared *shared);
+static Snapshot get_initial_snapshot(DecodingWorker *worker);
+static void ProcessRepackMessage(StringInfo msg);
static const char *RepackCommandAsString(RepackCommand cmd);
@@ -619,20 +740,20 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid,
/* rebuild_relation does all the dirty work */
PG_TRY();
{
- /*
- * For concurrent processing, make sure that our logical decoding
- * ignores data changes of other tables than the one we are
- * processing.
- */
- if (concurrent)
- begin_concurrent_repack(OldHeap);
-
rebuild_relation(OldHeap, index, verbose, concurrent);
}
PG_FINALLY();
{
if (concurrent)
- end_concurrent_repack();
+ {
+ /*
+ * Since during normal operation the worker was already asked to
+ * exit, stopping it explicitly is especially important on ERROR.
+ * However it still seems a good practice to make sure that the
+ * worker never survives the REPACK command.
+ */
+ stop_decoding_worker();
+ }
}
PG_END_TRY();
@@ -929,7 +1050,6 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
bool swap_toast_by_content;
TransactionId frozenXid;
MultiXactId cutoffMulti;
- LogicalDecodingContext *decoding_ctx = NULL;
Snapshot snapshot = NULL;
#if USE_ASSERT_CHECKING
LOCKMODE lmode;
@@ -943,19 +1063,36 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
if (concurrent)
{
/*
- * Prepare to capture the concurrent data changes.
+ * The worker needs to be member of the locking group we're the leader
+ * of. We ought to become the leader before the worker starts. The
+ * worker will join the group as soon as it starts.
*
- * Note that this call waits for all transactions with XID already
- * assigned to finish. If some of those transactions is waiting for a
- * lock conflicting with ShareUpdateExclusiveLock on our table (e.g.
- * it runs CREATE INDEX), we can end up in a deadlock. Not sure this
- * risk is worth unlocking/locking the table (and its clustering
- * index) and checking again if its still eligible for REPACK
- * CONCURRENTLY.
+ * This is to make sure that the deadlock described below is
+ * detectable by deadlock.c: if the worker waits for a transaction to
+ * complete and we are waiting for the worker output, then effectively
+ * we (i.e. this backend) are waiting for that transaction.
*/
- decoding_ctx = setup_logical_decoding(tableOid);
+ BecomeLockGroupLeader();
+
+ /*
+ * Start the worker that decodes data changes applied while we're
+ * copying the table contents.
+ *
+ * Note that the worker has to wait for all transactions with XID
+ * already assigned to finish. If some of those transactions is
+ * waiting for a lock conflicting with ShareUpdateExclusiveLock on our
+ * table (e.g. it runs CREATE INDEX), we can end up in a deadlock.
+ * Not sure this risk is worth unlocking/locking the table (and its
+ * clustering index) and checking again if it's still eligible for
+ * REPACK CONCURRENTLY.
+ */
+ start_decoding_worker(tableOid);
+
+ /*
+ * Wait until the worker has the initial snapshot and retrieve it.
+ */
+ snapshot = get_initial_snapshot(decoding_worker);
- snapshot = SnapBuildInitialSnapshotForRepack(decoding_ctx->snapshot_builder);
PushActiveSnapshot(snapshot);
}
@@ -980,7 +1117,7 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
NewHeap = table_open(OIDNewHeap, NoLock);
/* Copy the heap data into the new table in the desired order */
- copy_table_data(NewHeap, OldHeap, index, snapshot, decoding_ctx, verbose,
+ copy_table_data(NewHeap, OldHeap, index, snapshot, verbose,
&swap_toast_by_content, &frozenXid, &cutoffMulti);
/* The historic snapshot won't be needed anymore. */
@@ -1001,14 +1138,11 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
if (index)
index_close(index, NoLock);
- rebuild_relation_finish_concurrent(NewHeap, OldHeap, decoding_ctx,
- frozenXid, cutoffMulti);
+ rebuild_relation_finish_concurrent(NewHeap, OldHeap, frozenXid,
+ cutoffMulti);
pgstat_progress_update_param(PROGRESS_REPACK_PHASE,
PROGRESS_REPACK_PHASE_FINAL_CLEANUP);
-
- /* Done with decoding. */
- cleanup_logical_decoding(decoding_ctx);
}
else
{
@@ -1179,8 +1313,7 @@ make_new_heap(Oid OIDOldHeap, Oid NewTableSpace, Oid NewAccessMethod,
*/
static void
copy_table_data(Relation NewHeap, Relation OldHeap, Relation OldIndex,
- Snapshot snapshot, LogicalDecodingContext *decoding_ctx,
- bool verbose, bool *pSwapToastByContent,
+ Snapshot snapshot, bool verbose, bool *pSwapToastByContent,
TransactionId *pFreezeXid, MultiXactId *pCutoffMulti)
{
Relation relRelation;
@@ -1341,7 +1474,6 @@ copy_table_data(Relation NewHeap, Relation OldHeap, Relation OldIndex,
*/
table_relation_copy_for_cluster(OldHeap, NewHeap, OldIndex, use_sort,
cutoffs.OldestXmin, snapshot,
- decoding_ctx,
&cutoffs.FreezeLimit,
&cutoffs.MultiXactCutoff,
&num_tuples, &tups_vacuumed,
@@ -2371,62 +2503,10 @@ RepackCommandAsString(RepackCommand cmd)
case REPACK_COMMAND_CLUSTER:
return "CLUSTER";
}
- return "???"; /* keep compiler quiet */
+ return "???"; /* keep compiler quiet */
}
-/*
- * Call this function before REPACK CONCURRENTLY starts to setup logical
- * decoding. It makes sure that other users of the table put enough
- * information into WAL.
- *
- * The point is that at various places we expect that the table we're
- * processing is treated like a system catalog. For example, we need to be
- * able to scan it using a "historic snapshot" anytime during the processing
- * (as opposed to scanning only at the start point of the decoding, as logical
- * replication does during initial table synchronization), in order to apply
- * concurrent UPDATE / DELETE commands.
- *
- * Note that TOAST table needs no attention here as it's not scanned using
- * historic snapshot.
- */
-static void
-begin_concurrent_repack(Relation rel)
-{
- Oid toastrelid;
-
- /*
- * Avoid logical decoding of other relations by this backend. The lock we
- * have guarantees that the actual locator cannot be changed concurrently:
- * TRUNCATE needs AccessExclusiveLock.
- */
- Assert(CheckRelationLockedByMe(rel, ShareUpdateExclusiveLock, false));
- repacked_rel_locator = rel->rd_locator;
- toastrelid = rel->rd_rel->reltoastrelid;
- if (OidIsValid(toastrelid))
- {
- Relation toastrel;
-
- /* Avoid logical decoding of other TOAST relations. */
- toastrel = table_open(toastrelid, AccessShareLock);
- repacked_rel_toast_locator = toastrel->rd_locator;
- table_close(toastrel, AccessShareLock);
- }
-}
-
-/*
- * Call this when done with REPACK CONCURRENTLY.
- */
-static void
-end_concurrent_repack(void)
-{
- /*
- * Restore normal function of (future) logical decoding for this backend.
- */
- repacked_rel_locator.relNumber = InvalidOid;
- repacked_rel_toast_locator.relNumber = InvalidOid;
-}
-
/*
* Is this backend performing logical decoding on behalf of REPACK
* (CONCURRENTLY) ?
@@ -2491,9 +2571,10 @@ static LogicalDecodingContext *
setup_logical_decoding(Oid relid)
{
Relation rel;
- TupleDesc tupdesc;
+ Oid toastrelid;
LogicalDecodingContext *ctx;
- RepackDecodingState *dstate = palloc0_object(RepackDecodingState);
+ NameData slotname;
+ RepackDecodingState *dstate;
/*
* REPACK CONCURRENTLY is not allowed in a transaction block, so this
@@ -2501,20 +2582,21 @@ setup_logical_decoding(Oid relid)
*/
Assert(!TransactionIdIsValid(GetTopTransactionIdIfAny()));
- /*
- * A single backend should not execute multiple REPACK commands at a time,
- * so use PID to make the slot unique.
- */
- snprintf(NameStr(dstate->slotname), NAMEDATALEN, "repack_%d", MyProcPid);
-
/*
* Make sure we can use logical decoding.
*/
CheckSlotPermissions();
CheckLogicalDecodingRequirements();
- /* RS_TEMPORARY so that the slot gets cleaned up on ERROR. */
- ReplicationSlotCreate(NameStr(dstate->slotname), true, RS_TEMPORARY,
- false, false, false);
+ /*
+ * A single backend should not execute multiple REPACK commands at a time,
+ * so use PID to make the slot unique.
+ *
+ * RS_TEMPORARY so that the slot gets cleaned up on ERROR.
+ */
+ snprintf(NameStr(slotname), NAMEDATALEN, "repack_%d", MyProcPid);
+ ReplicationSlotCreate(NameStr(slotname), true, RS_TEMPORARY, false, false,
+ false);
+
EnsureLogicalDecodingEnabled();
/*
@@ -2537,104 +2619,109 @@ setup_logical_decoding(Oid relid)
DecodingContextFindStartpoint(ctx);
+ /*
+ * decode_concurrent_changes() needs non-blocking callback.
+ */
+ ctx->reader->routine.page_read = read_local_xlog_page_no_wait;
+
+ /*
+ * read_local_xlog_page_no_wait() needs to be able to indicate the end of
+ * WAL.
+ */
+ ctx->reader->private_data = MemoryContextAllocZero(ctx->context,
+ sizeof(ReadLocalXLogPageNoWaitPrivate));
+
+
/* Some WAL records should have been read. */
Assert(ctx->reader->EndRecPtr != InvalidXLogRecPtr);
+ /*
+ * Initialize repack_current_segment so that we can notice WAL segment
+ * boundaries.
+ */
XLByteToSeg(ctx->reader->EndRecPtr, repack_current_segment,
wal_segment_size);
- /*
- * Setup structures to store decoded changes.
- */
+ dstate = palloc0_object(RepackDecodingState);
dstate->relid = relid;
- dstate->tstore = tuplestore_begin_heap(false, false,
- maintenance_work_mem);
- /* Caller should already have the table locked. */
- rel = table_open(relid, NoLock);
- tupdesc = CreateTupleDescCopy(RelationGetDescr(rel));
- dstate->tupdesc = tupdesc;
- table_close(rel, NoLock);
+ /*
+ * Tuple descriptor may be needed to flatten a tuple before we write it to
+ * a file. A copy is needed because the decoding worker invalidates system
+ * caches before it starts to do the actual work.
+ */
+ rel = table_open(relid, AccessShareLock);
+ dstate->tupdesc = CreateTupleDescCopy(RelationGetDescr(rel));
- /* Initialize the descriptor to store the changes ... */
- dstate->tupdesc_change = CreateTemplateTupleDesc(1);
+ /* Avoid logical decoding of other relations. */
+ repacked_rel_locator = rel->rd_locator;
+ toastrelid = rel->rd_rel->reltoastrelid;
+ if (OidIsValid(toastrelid))
+ {
+ Relation toastrel;
- TupleDescInitEntry(dstate->tupdesc_change, 1, NULL, BYTEAOID, -1, 0);
- /* ... as well as the corresponding slot. */
- dstate->tsslot = MakeSingleTupleTableSlot(dstate->tupdesc_change,
- &TTSOpsMinimalTuple);
+ /* Avoid logical decoding of other TOAST relations. */
+ toastrel = table_open(toastrelid, AccessShareLock);
+ repacked_rel_toast_locator = toastrel->rd_locator;
+ table_close(toastrel, AccessShareLock);
+ }
+ table_close(rel, AccessShareLock);
- dstate->resowner = ResourceOwnerCreate(CurrentResourceOwner,
- "logical decoding");
+ /* The file will be set as soon as we have it opened. */
+ dstate->file = NULL;
ctx->output_writer_private = dstate;
+
return ctx;
}
/*
- * Retrieve tuple from ConcurrentChange structure.
+ * Decode logical changes from the WAL sequence and store them to a file.
*
- * The input data starts with the structure but it might not be appropriately
- * aligned.
+ * If true is returned, there is no more work for the worker.
*/
-static HeapTuple
-get_changed_tuple(char *change)
-{
- HeapTupleData tup_data;
- HeapTuple result;
- char *src;
-
- /*
- * Ensure alignment before accessing the fields. (This is why we can't use
- * heap_copytuple() instead of this function.)
- */
- src = change + offsetof(ConcurrentChange, tup_data);
- memcpy(&tup_data, src, sizeof(HeapTupleData));
-
- result = (HeapTuple) palloc(HEAPTUPLESIZE + tup_data.t_len);
- memcpy(result, &tup_data, sizeof(HeapTupleData));
- result->t_data = (HeapTupleHeader) ((char *) result + HEAPTUPLESIZE);
- src = change + SizeOfConcurrentChange;
- memcpy(result->t_data, src, result->t_len);
-
- return result;
-}
-
-/*
- * Decode logical changes from the WAL sequence up to end_of_wal.
- */
-void
-repack_decode_concurrent_changes(LogicalDecodingContext *ctx,
- XLogRecPtr end_of_wal)
+static bool
+decode_concurrent_changes(LogicalDecodingContext *ctx,
+ DecodingWorkerShared *shared)
{
RepackDecodingState *dstate;
- ResourceOwner resowner_old;
+ XLogRecPtr lsn_upto;
+ bool done;
+ char fname[MAXPGPATH];
dstate = (RepackDecodingState *) ctx->output_writer_private;
- resowner_old = CurrentResourceOwner;
- CurrentResourceOwner = dstate->resowner;
- PG_TRY();
+ /* Open the output file. */
+ DecodingWorkerFileName(fname, shared->relid, shared->last_exported + 1);
+ dstate->file = BufFileCreateFileSet(&shared->sfs.fs, fname);
+
+ SpinLockAcquire(&shared->mutex);
+ lsn_upto = shared->lsn_upto;
+ done = shared->done;
+ SpinLockRelease(&shared->mutex);
+
+ while (true)
{
- while (ctx->reader->EndRecPtr < end_of_wal)
+ XLogRecord *record;
+ XLogSegNo segno_new;
+ char *errm = NULL;
+ XLogRecPtr end_lsn;
+
+ CHECK_FOR_INTERRUPTS();
+
+ record = XLogReadRecord(ctx->reader, &errm);
+ if (record)
{
- XLogRecord *record;
- XLogSegNo segno_new;
- char *errm = NULL;
- XLogRecPtr end_lsn;
-
- record = XLogReadRecord(ctx->reader, &errm);
- if (errm)
- elog(ERROR, "%s", errm);
-
- if (record != NULL)
- LogicalDecodingProcessRecord(ctx, ctx->reader);
+ LogicalDecodingProcessRecord(ctx, ctx->reader);
/*
* If WAL segment boundary has been crossed, inform the decoding
- * system that the catalog_xmin can advance. (We can confirm more
- * often, but a filling a single WAL segment should not take much
- * time.)
+ * system that the catalog_xmin can advance.
+ *
+ * TODO Does it make sense to confirm more often? Segment size
+ * seems appropriate for restart_lsn (because less than a segment
+ * cannot be recycled anyway), however more frequent checks might
+ * be beneficial for catalog_xmin.
*/
end_lsn = ctx->reader->EndRecPtr;
XLByteToSeg(end_lsn, segno_new, wal_segment_size);
@@ -2645,80 +2732,137 @@ repack_decode_concurrent_changes(LogicalDecodingContext *ctx,
(uint32) (end_lsn >> 32), (uint32) end_lsn);
repack_current_segment = segno_new;
}
-
- CHECK_FOR_INTERRUPTS();
}
- InvalidateSystemCaches();
- CurrentResourceOwner = resowner_old;
+ else
+ {
+ ReadLocalXLogPageNoWaitPrivate *priv;
+
+ if (errm)
+ ereport(ERROR, (errmsg("%s", errm)));
+
+ /*
+ * In the decoding loop we do not want to get blocked when there
+ * is no more WAL available, otherwise the loop would become
+ * uninterruptible.
+ */
+ priv = (ReadLocalXLogPageNoWaitPrivate *)
+ ctx->reader->private_data;
+ if (priv->end_of_wal)
+ /* Do not miss the end of WAL condition next time. */
+ priv->end_of_wal = false;
+ else
+ ereport(ERROR, (errmsg("could not read WAL record")));
+ }
+
+ /*
+ * Whether we could read new record or not, keep checking if
+ * 'lsn_upto' was specified.
+ */
+ if (XLogRecPtrIsInvalid(lsn_upto))
+ {
+ SpinLockAcquire(&shared->mutex);
+ lsn_upto = shared->lsn_upto;
+ /* 'done' should be set at the same time as 'lsn_upto' */
+ done = shared->done;
+ SpinLockRelease(&shared->mutex);
+ }
+ if (!XLogRecPtrIsInvalid(lsn_upto) &&
+ ctx->reader->EndRecPtr >= lsn_upto)
+ break;
+
+ if (record == NULL)
+ {
+ int64 timeout = 0;
+ WaitLSNResult res;
+
+ /*
+ * Before we retry reading, wait until new WAL is flushed.
+ *
+ * There is a race condition such that the backend executing
+ * REPACK determines 'lsn_upto', but before it sets the shared
+ * variable, we reach the end of WAL. In that case we'd need to
+ * wait until the next WAL flush (unrelated to REPACK). Although
+ * that should not be a problem in a busy system, it might be
+ * noticeable in other cases, including regression tests (which
+ * are not necessarily executed in parallel). Therefore it makes
+ * sense to use timeout.
+ *
+ * If lsn_upto is valid, WAL records having LSN lower than that
+ * should already have been flushed to disk.
+ */
+ if (XLogRecPtrIsInvalid(lsn_upto))
+ timeout = 100L;
+ res = WaitForLSN(WAIT_LSN_TYPE_PRIMARY_FLUSH,
+ ctx->reader->EndRecPtr + 1,
+ timeout);
+ if (res != WAIT_LSN_RESULT_SUCCESS &&
+ res != WAIT_LSN_RESULT_TIMEOUT)
+ ereport(ERROR, (errmsg("waiting for WAL failed")));
+ }
}
- PG_CATCH();
- {
- /* clear all timetravel entries */
- InvalidateSystemCaches();
- CurrentResourceOwner = resowner_old;
- PG_RE_THROW();
- }
- PG_END_TRY();
+
+ /*
+ * Close the file so we can make it available to the backend.
+ */
+ BufFileClose(dstate->file);
+ dstate->file = NULL;
+ SpinLockAcquire(&shared->mutex);
+ shared->lsn_upto = InvalidXLogRecPtr;
+ shared->last_exported++;
+ SpinLockRelease(&shared->mutex);
+ ConditionVariableSignal(&shared->cv);
+
+ return done;
}
/*
* Apply changes stored in 'file'.
*/
static void
-apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
+apply_concurrent_changes(BufFile *file, ChangeDest *dest)
{
+ char kind;
+ uint32 t_len;
Relation rel = dest->rel;
TupleTableSlot *index_slot,
*ident_slot;
HeapTuple tup_old = NULL;
- if (dstate->nchanges == 0)
- return;
-
/* TupleTableSlot is needed to pass the tuple to ExecInsertIndexTuples(). */
- index_slot = MakeSingleTupleTableSlot(dstate->tupdesc, &TTSOpsHeapTuple);
+ index_slot = MakeSingleTupleTableSlot(RelationGetDescr(rel),
+ &TTSOpsHeapTuple);
/* A slot to fetch tuples from identity index. */
ident_slot = table_slot_create(rel, NULL);
- while (tuplestore_gettupleslot(dstate->tstore, true, false,
- dstate->tsslot))
+ while (true)
{
- bool shouldFree;
- HeapTuple tup_change,
- tup,
+ size_t nread;
+ HeapTuple tup,
tup_exist;
- char *change_raw,
- *src;
- ConcurrentChange change;
- bool isnull[1];
- Datum values[1];
CHECK_FOR_INTERRUPTS();
- /* Get the change from the single-column tuple. */
- tup_change = ExecFetchSlotHeapTuple(dstate->tsslot, false, &shouldFree);
- heap_deform_tuple(tup_change, dstate->tupdesc_change, values, isnull);
- Assert(!isnull[0]);
+ nread = BufFileReadMaybeEOF(file, &kind, 1, true);
+ /* Are we done with the file? */
+ if (nread == 0)
+ break;
- /* Make sure we access aligned data. */
- change_raw = (char *) DatumGetByteaP(values[0]);
- src = (char *) VARDATA(change_raw);
- memcpy(&change, src, SizeOfConcurrentChange);
+ /* Read the tuple. */
+ BufFileReadExact(file, &t_len, sizeof(t_len));
+ tup = (HeapTuple) palloc(HEAPTUPLESIZE + t_len);
+ tup->t_data = (HeapTupleHeader) ((char *) tup + HEAPTUPLESIZE);
+ BufFileReadExact(file, tup->t_data, t_len);
+ tup->t_len = t_len;
+ ItemPointerSetInvalid(&tup->t_self);
+ tup->t_tableOid = RelationGetRelid(dest->rel);
- /*
- * Extract the tuple from the change. The tuple is copied here because
- * it might be assigned to 'tup_old', in which case it needs to
- * survive into the next iteration.
- */
- tup = get_changed_tuple(src);
-
- if (change.kind == CHANGE_UPDATE_OLD)
+ if (kind == CHANGE_UPDATE_OLD)
{
Assert(tup_old == NULL);
tup_old = tup;
}
- else if (change.kind == CHANGE_INSERT)
+ else if (kind == CHANGE_INSERT)
{
Assert(tup_old == NULL);
@@ -2726,12 +2870,11 @@ apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
pfree(tup);
}
- else if (change.kind == CHANGE_UPDATE_NEW ||
- change.kind == CHANGE_DELETE)
+ else if (kind == CHANGE_UPDATE_NEW || kind == CHANGE_DELETE)
{
HeapTuple tup_key;
- if (change.kind == CHANGE_UPDATE_NEW)
+ if (kind == CHANGE_UPDATE_NEW)
{
tup_key = tup_old != NULL ? tup_old : tup;
}
@@ -2748,7 +2891,7 @@ apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
if (tup_exist == NULL)
elog(ERROR, "failed to find target tuple");
- if (change.kind == CHANGE_UPDATE_NEW)
+ if (kind == CHANGE_UPDATE_NEW)
apply_concurrent_update(rel, tup, tup_exist, dest->iistate,
index_slot);
else
@@ -2763,26 +2906,19 @@ apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
pfree(tup);
}
else
- elog(ERROR, "unrecognized kind of change: %d", change.kind);
+ elog(ERROR, "unrecognized kind of change: %d", kind);
/*
* If a change was applied now, increment CID for next writes and
* update the snapshot so it sees the changes we've applied so far.
*/
- if (change.kind != CHANGE_UPDATE_OLD)
+ if (kind != CHANGE_UPDATE_OLD)
{
CommandCounterIncrement();
UpdateActiveSnapshotCommandId();
}
-
- /* TTSOpsMinimalTuple has .get_heap_tuple==NULL. */
- Assert(shouldFree);
- pfree(tup_change);
}
- tuplestore_clear(dstate->tstore);
- dstate->nchanges = 0;
-
/* Cleanup. */
ExecDropSingleTupleTableSlot(index_slot);
ExecDropSingleTupleTableSlot(ident_slot);
@@ -2957,25 +3093,59 @@ find_target_tuple(Relation rel, ChangeDest *dest, HeapTuple tup_key,
}
/*
- * Decode and apply concurrent changes.
+ * Decode and apply concurrent changes, up to (and including) the record whose
+ * LSN is 'end_of_wal'.
*/
static void
-process_concurrent_changes(LogicalDecodingContext *decoding_ctx,
- XLogRecPtr end_of_wal, ChangeDest *dest)
+process_concurrent_changes(XLogRecPtr end_of_wal, ChangeDest *dest, bool done)
{
- RepackDecodingState *dstate;
+ DecodingWorkerShared *shared;
+ char fname[MAXPGPATH];
+ BufFile *file;
pgstat_progress_update_param(PROGRESS_REPACK_PHASE,
PROGRESS_REPACK_PHASE_CATCH_UP);
- dstate = (RepackDecodingState *) decoding_ctx->output_writer_private;
+ /* Ask the worker for the file. */
+ shared = (DecodingWorkerShared *) dsm_segment_address(decoding_worker->seg);
+ SpinLockAcquire(&shared->mutex);
+ shared->lsn_upto = end_of_wal;
+ shared->done = done;
+ SpinLockRelease(&shared->mutex);
- repack_decode_concurrent_changes(decoding_ctx, end_of_wal);
+ /*
+ * The worker needs to finish processing of the current WAL record. Even
+ * if it's idle, it'll need to close the output file. Thus we're likely to
+ * wait, so prepare for sleep.
+ */
+ ConditionVariablePrepareToSleep(&shared->cv);
+ for (;;)
+ {
+ int last_exported;
- if (dstate->nchanges == 0)
- return;
+ SpinLockAcquire(&shared->mutex);
+ last_exported = shared->last_exported;
+ SpinLockRelease(&shared->mutex);
- apply_concurrent_changes(dstate, dest);
+ /*
+ * Has the worker exported the file we are waiting for?
+ */
+ if (last_exported == dest->file_seq)
+ break;
+
+ ConditionVariableSleep(&shared->cv, WAIT_EVENT_REPACK_WORKER_EXPORT);
+ }
+ ConditionVariableCancelSleep();
+
+ /* Open the file. */
+ DecodingWorkerFileName(fname, shared->relid, dest->file_seq);
+ file = BufFileOpenFileSet(&shared->sfs.fs, fname, O_RDONLY, false);
+ apply_concurrent_changes(file, dest);
+
+ BufFileClose(file);
+
+ /* Get ready for the next file. */
+ dest->file_seq++;
}
/*
@@ -3101,15 +3271,10 @@ cleanup_logical_decoding(LogicalDecodingContext *ctx)
dstate = (RepackDecodingState *) ctx->output_writer_private;
- ExecDropSingleTupleTableSlot(dstate->tsslot);
- FreeTupleDesc(dstate->tupdesc_change);
FreeTupleDesc(dstate->tupdesc);
- tuplestore_end(dstate->tstore);
-
FreeDecodingContext(ctx);
- ReplicationSlotRelease();
- ReplicationSlotDrop(NameStr(dstate->slotname), false);
+ ReplicationSlotDropAcquired();
pfree(dstate);
}
@@ -3123,7 +3288,6 @@ cleanup_logical_decoding(LogicalDecodingContext *ctx)
*/
static void
rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
- LogicalDecodingContext *decoding_ctx,
TransactionId frozenXid,
MultiXactId cutoffMulti)
{
@@ -3204,6 +3368,7 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
&chgdst.ident_index);
chgdst.ident_key = build_identity_key(ident_idx_new, OldHeap,
&chgdst.ident_key_nentries);
+ chgdst.file_seq = WORKER_FILE_SNAPSHOT + 1;
/*
* During testing, wait for another backend to perform concurrent data
@@ -3225,7 +3390,7 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
* hold AccessExclusiveLock. (Quite some amount of WAL could have been
* written during the data copying and index creation.)
*/
- process_concurrent_changes(decoding_ctx, end_of_wal, &chgdst);
+ process_concurrent_changes(end_of_wal, &chgdst, false);
/*
* Acquire AccessExclusiveLock on the table, its TOAST relation (if there
@@ -3306,8 +3471,11 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
XLogFlush(wal_insert_ptr);
end_of_wal = GetFlushRecPtr(NULL);
- /* Apply the concurrent changes again. */
- process_concurrent_changes(decoding_ctx, end_of_wal, &chgdst);
+ /*
+ * Apply the concurrent changes again. Indicate that the decoding worker
+ * won't be needed anymore.
+ */
+ process_concurrent_changes(end_of_wal, &chgdst, true);
/* Remember info about rel before closing OldHeap */
relpersistence = OldHeap->rd_rel->relpersistence;
@@ -3417,3 +3585,510 @@ build_new_indexes(Relation NewHeap, Relation OldHeap, List *OldIndexes)
return result;
}
+
+/*
+ * Try to start a background worker to perform logical decoding of data
+ * changes applied to relation while REPACK CONCURRENTLY is copying its
+ * contents to a new table.
+ */
+static void
+start_decoding_worker(Oid relid)
+{
+ Size size;
+ dsm_segment *seg;
+ DecodingWorkerShared *shared;
+ shm_mq *mq;
+ shm_mq_handle *mqh;
+ BackgroundWorker bgw;
+
+ /* Setup shared memory. */
+ size = BUFFERALIGN(offsetof(DecodingWorkerShared, error_queue)) +
+ BUFFERALIGN(REPACK_ERROR_QUEUE_SIZE);
+ seg = dsm_create(size, 0);
+ shared = (DecodingWorkerShared *) dsm_segment_address(seg);
+ shared->lsn_upto = InvalidXLogRecPtr;
+ shared->done = false;
+ SharedFileSetInit(&shared->sfs, seg);
+ shared->last_exported = -1;
+ SpinLockInit(&shared->mutex);
+ shared->dbid = MyDatabaseId;
+
+ /*
+ * This is the UserId set in cluster_rel(). Security context shouldn't be
+ * needed for decoding worker.
+ */
+ shared->roleid = GetUserId();
+ shared->relid = relid;
+ ConditionVariableInit(&shared->cv);
+ shared->backend_proc = MyProc;
+ shared->backend_pid = MyProcPid;
+ shared->backend_proc_number = MyProcNumber;
+
+ mq = shm_mq_create((char *) BUFFERALIGN(shared->error_queue),
+ REPACK_ERROR_QUEUE_SIZE);
+ shm_mq_set_receiver(mq, MyProc);
+ mqh = shm_mq_attach(mq, seg, NULL);
+
+ memset(&bgw, 0, sizeof(bgw));
+ snprintf(bgw.bgw_name, BGW_MAXLEN,
+ "REPACK decoding worker for relation \"%s\"",
+ get_rel_name(relid));
+ snprintf(bgw.bgw_type, BGW_MAXLEN, "REPACK decoding worker");
+ bgw.bgw_flags = BGWORKER_SHMEM_ACCESS |
+ BGWORKER_BACKEND_DATABASE_CONNECTION;
+ bgw.bgw_start_time = BgWorkerStart_RecoveryFinished;
+ bgw.bgw_restart_time = BGW_NEVER_RESTART;
+ snprintf(bgw.bgw_library_name, MAXPGPATH, "postgres");
+ snprintf(bgw.bgw_function_name, BGW_MAXLEN, "RepackWorkerMain");
+ bgw.bgw_main_arg = UInt32GetDatum(dsm_segment_handle(seg));
+ bgw.bgw_notify_pid = MyProcPid;
+
+ decoding_worker = palloc0_object(DecodingWorker);
+ if (!RegisterDynamicBackgroundWorker(&bgw, &decoding_worker->handle))
+ ereport(ERROR,
+ (errcode(ERRCODE_CONFIGURATION_LIMIT_EXCEEDED),
+ errmsg("out of background worker slots"),
+ errhint("You might need to increase \"%s\".", "max_worker_processes")));
+
+ decoding_worker->seg = seg;
+ decoding_worker->error_mqh = mqh;
+
+ /*
+ * The decoding setup must be done before the caller can have XID assigned
+ * for any reason, otherwise the worker might end up in a deadlock,
+ * waiting for the caller's transaction to end. Therefore wait here until
+ * the worker indicates that it has the logical decoding initialized.
+ */
+ ConditionVariablePrepareToSleep(&shared->cv);
+ for (;;)
+ {
+ bool initialized;
+
+ SpinLockAcquire(&shared->mutex);
+ initialized = shared->initialized;
+ SpinLockRelease(&shared->mutex);
+
+ if (initialized)
+ break;
+
+ ConditionVariableSleep(&shared->cv, WAIT_EVENT_REPACK_WORKER_EXPORT);
+ }
+ ConditionVariableCancelSleep();
+}
+
+/*
+ * Stop the decoding worker and cleanup the related resources.
+ *
+ * The worker stops on its own when it knows there is no more work to do, but
+ * we need to stop it explicitly at least on ERROR in the launching backend.
+ */
+static void
+stop_decoding_worker(void)
+{
+ BgwHandleStatus status;
+
+ /* Haven't reached the worker startup? */
+ if (decoding_worker == NULL)
+ return;
+
+ /* Could not register the worker? */
+ if (decoding_worker->handle == NULL)
+ return;
+
+ TerminateBackgroundWorker(decoding_worker->handle);
+ /* The worker should really exit before the REPACK command does. */
+ HOLD_INTERRUPTS();
+ status = WaitForBackgroundWorkerShutdown(decoding_worker->handle);
+ RESUME_INTERRUPTS();
+
+ if (status == BGWH_POSTMASTER_DIED)
+ ereport(FATAL,
+ (errcode(ERRCODE_ADMIN_SHUTDOWN),
+ errmsg("postmaster exited during REPACK command")));
+
+ shm_mq_detach(decoding_worker->error_mqh);
+
+ /*
+ * If we could not cancel the current sleep due to ERROR, do that before
+ * we detach from the shared memory the condition variable is located in.
+ * If we did not, the bgworker ERROR handling code would try and fail
+ * badly.
+ */
+ ConditionVariableCancelSleep();
+
+ dsm_detach(decoding_worker->seg);
+ pfree(decoding_worker);
+ decoding_worker = NULL;
+}
+
+/* Is this process a REPACK worker? */
+static bool is_repack_worker = false;
+
+static pid_t backend_pid;
+static ProcNumber backend_proc_number;
+
+/*
+ * See ParallelWorkerShutdown for details.
+ */
+static void
+RepackWorkerShutdown(int code, Datum arg)
+{
+ SendProcSignal(backend_pid,
+ PROCSIG_REPACK_MESSAGE,
+ backend_proc_number);
+
+ dsm_detach((dsm_segment *) DatumGetPointer(arg));
+}
+
+/* REPACK decoding worker entry point */
+void
+RepackWorkerMain(Datum main_arg)
+{
+ dsm_segment *seg;
+ DecodingWorkerShared *shared;
+ shm_mq *mq;
+ shm_mq_handle *mqh;
+
+ is_repack_worker = true;
+
+ /*
+ * Override the default bgworker_die() with die() so we can use
+ * CHECK_FOR_INTERRUPTS().
+ */
+ pqsignal(SIGTERM, die);
+ BackgroundWorkerUnblockSignals();
+
+ seg = dsm_attach(DatumGetUInt32(main_arg));
+ if (seg == NULL)
+ ereport(ERROR,
+ (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
+ errmsg("could not map dynamic shared memory segment")));
+
+ shared = (DecodingWorkerShared *) dsm_segment_address(seg);
+
+ /* Arrange to signal the leader if we exit. */
+ backend_pid = shared->backend_pid;
+ backend_proc_number = shared->backend_proc_number;
+ before_shmem_exit(RepackWorkerShutdown, PointerGetDatum(seg));
+
+ /*
+ * Join locking group - see the comments around the call of
+ * start_decoding_worker().
+ */
+ if (!BecomeLockGroupMember(shared->backend_proc, backend_pid))
+ /* The leader is not running anymore. */
+ return;
+
+ /*
+ * Setup a queue to send error messages to the backend that launched this
+ * worker.
+ */
+ mq = (shm_mq *) (char *) BUFFERALIGN(shared->error_queue);
+ shm_mq_set_sender(mq, MyProc);
+ mqh = shm_mq_attach(mq, seg, NULL);
+ pq_redirect_to_shm_mq(seg, mqh);
+ pq_set_parallel_leader(shared->backend_pid,
+ shared->backend_proc_number);
+
+ /* Connect to the database. */
+ BackgroundWorkerInitializeConnectionByOid(shared->dbid, shared->roleid, 0);
+
+ repack_worker_internal(seg);
+}
+
+static void
+repack_worker_internal(dsm_segment *seg)
+{
+ DecodingWorkerShared *shared;
+ LogicalDecodingContext *decoding_ctx;
+ SharedFileSet *sfs;
+ Snapshot snapshot;
+
+ /*
+ * Transaction is needed to open relation, and it also provides us with a
+ * resource owner.
+ */
+ StartTransactionCommand();
+
+ shared = (DecodingWorkerShared *) dsm_segment_address(seg);
+
+ /*
+ * Not sure the spinlock is needed here - the backend should not change
+ * anything in the shared memory until we have serialized the snapshot.
+ */
+ SpinLockAcquire(&shared->mutex);
+ Assert(XLogRecPtrIsInvalid(shared->lsn_upto));
+ sfs = &shared->sfs;
+ SpinLockRelease(&shared->mutex);
+
+ SharedFileSetAttach(sfs, seg);
+
+ /*
+ * Prepare to capture the concurrent data changes ourselves.
+ */
+ decoding_ctx = setup_logical_decoding(shared->relid);
+
+ /* Announce that we're ready. */
+ SpinLockAcquire(&shared->mutex);
+ shared->initialized = true;
+ SpinLockRelease(&shared->mutex);
+ ConditionVariableSignal(&shared->cv);
+
+ /* Build the initial snapshot and export it. */
+ snapshot = SnapBuildInitialSnapshotForRepack(decoding_ctx->snapshot_builder);
+ export_initial_snapshot(snapshot, shared);
+
+ /*
+ * Only historic snapshots should be used now. Do not let us restrict the
+ * progress of xmin horizon.
+ */
+ InvalidateCatalogSnapshot();
+
+ while (!decode_concurrent_changes(decoding_ctx, shared))
+ ;
+
+ /* Cleanup. */
+ cleanup_logical_decoding(decoding_ctx);
+ CommitTransactionCommand();
+}
+
+/*
+ * Make snapshot available to the backend that launched the decoding worker.
+ */
+static void
+export_initial_snapshot(Snapshot snapshot, DecodingWorkerShared *shared)
+{
+ char fname[MAXPGPATH];
+ BufFile *file;
+ Size snap_size;
+ char *snap_space;
+
+ snap_size = EstimateSnapshotSpace(snapshot);
+ snap_space = (char *) palloc(snap_size);
+ SerializeSnapshot(snapshot, snap_space);
+ FreeSnapshot(snapshot);
+
+ DecodingWorkerFileName(fname, shared->relid, shared->last_exported + 1);
+ file = BufFileCreateFileSet(&shared->sfs.fs, fname);
+ /* To make restoration easier, write the snapshot size first. */
+ BufFileWrite(file, &snap_size, sizeof(snap_size));
+ BufFileWrite(file, snap_space, snap_size);
+ pfree(snap_space);
+ BufFileClose(file);
+
+ /* Increase the counter to tell the backend that the file is available. */
+ SpinLockAcquire(&shared->mutex);
+ shared->last_exported++;
+ SpinLockRelease(&shared->mutex);
+ ConditionVariableSignal(&shared->cv);
+}
+
+/*
+ * Get the initial snapshot from the decoding worker.
+ */
+static Snapshot
+get_initial_snapshot(DecodingWorker *worker)
+{
+ DecodingWorkerShared *shared;
+ char fname[MAXPGPATH];
+ BufFile *file;
+ Size snap_size;
+ char *snap_space;
+ Snapshot snapshot;
+
+ shared = (DecodingWorkerShared *) dsm_segment_address(worker->seg);
+
+ /*
+ * The worker needs to initialize the logical decoding, which usually
+ * takes some time. Therefore it makes sense to prepare for the sleep
+ * first.
+ */
+ ConditionVariablePrepareToSleep(&shared->cv);
+ for (;;)
+ {
+ int last_exported;
+
+ SpinLockAcquire(&shared->mutex);
+ last_exported = shared->last_exported;
+ SpinLockRelease(&shared->mutex);
+
+ /*
+ * Has the worker exported the file we are waiting for?
+ */
+ if (last_exported == WORKER_FILE_SNAPSHOT)
+ break;
+
+ ConditionVariableSleep(&shared->cv, WAIT_EVENT_REPACK_WORKER_EXPORT);
+ }
+ ConditionVariableCancelSleep();
+
+ /* Read the snapshot from a file. */
+ DecodingWorkerFileName(fname, shared->relid, WORKER_FILE_SNAPSHOT);
+ file = BufFileOpenFileSet(&shared->sfs.fs, fname, O_RDONLY, false);
+ BufFileReadExact(file, &snap_size, sizeof(snap_size));
+ snap_space = (char *) palloc(snap_size);
+ BufFileReadExact(file, snap_space, snap_size);
+ BufFileClose(file);
+
+ /* Restore it. */
+ snapshot = RestoreSnapshot(snap_space);
+ pfree(snap_space);
+
+ return snapshot;
+}
+
+bool
+IsRepackWorker(void)
+{
+ return is_repack_worker;
+}
+
+/*
+ * Handle receipt of an interrupt indicating a repack worker message.
+ *
+ * Note: this is called within a signal handler! All we can do is set
+ * a flag that will cause the next CHECK_FOR_INTERRUPTS() to invoke
+ * ProcessRepackMessages().
+ */
+void
+HandleRepackMessageInterrupt(void)
+{
+ InterruptPending = true;
+ RepackMessagePending = true;
+ SetLatch(MyLatch);
+}
+
+/*
+ * Process any queued protocol messages received from parallel workers.
+ */
+void
+ProcessRepackMessages(void)
+{
+ MemoryContext oldcontext;
+
+ static MemoryContext hpm_context = NULL;
+
+ /*
+ * Nothing to do if we haven't launched the worker yet or have already
+ * terminated it.
+ */
+ if (decoding_worker == NULL)
+ return;
+
+ /*
+ * This is invoked from ProcessInterrupts(), and since some of the
+ * functions it calls contain CHECK_FOR_INTERRUPTS(), there is a potential
+ * for recursive calls if more signals are received while this runs. It's
+ * unclear that recursive entry would be safe, and it doesn't seem useful
+ * even if it is safe, so let's block interrupts until done.
+ */
+ HOLD_INTERRUPTS();
+
+ /*
+ * Moreover, CurrentMemoryContext might be pointing almost anywhere. We
+ * don't want to risk leaking data into long-lived contexts, so let's do
+ * our work here in a private context that we can reset on each use.
+ */
+ if (hpm_context == NULL) /* first time through? */
+ hpm_context = AllocSetContextCreate(TopMemoryContext,
+ "ProcessRepackMessages",
+ ALLOCSET_DEFAULT_SIZES);
+ else
+ MemoryContextReset(hpm_context);
+
+ oldcontext = MemoryContextSwitchTo(hpm_context);
+
+ /* OK to process messages. Reset the flag saying there are more to do. */
+ RepackMessagePending = false;
+
+ /*
+ * Read as many messages as we can from each worker, but stop when no more
+ * messages can be read from the worker without blocking.
+ */
+ while (true)
+ {
+ shm_mq_result res;
+ Size nbytes;
+ void *data;
+
+ res = shm_mq_receive(decoding_worker->error_mqh, &nbytes,
+ &data, true);
+ if (res == SHM_MQ_WOULD_BLOCK)
+ break;
+ else if (res == SHM_MQ_SUCCESS)
+ {
+ StringInfoData msg;
+
+ initStringInfo(&msg);
+ appendBinaryStringInfo(&msg, data, nbytes);
+ ProcessRepackMessage(&msg);
+ pfree(msg.data);
+ }
+ else
+ {
+ /*
+ * The decoding worker is special in that it exits as soon as it
+ * has its work done. Thus the DETACHED result code is fine.
+ */
+ Assert(res == SHM_MQ_DETACHED);
+
+ break;
+ }
+ }
+
+ MemoryContextSwitchTo(oldcontext);
+
+ /* Might as well clear the context on our way out */
+ MemoryContextReset(hpm_context);
+
+ RESUME_INTERRUPTS();
+}
+
+/*
+ * Process a single protocol message received from a single parallel worker.
+ */
+static void
+ProcessRepackMessage(StringInfo msg)
+{
+ char msgtype;
+
+ msgtype = pq_getmsgbyte(msg);
+
+ switch (msgtype)
+ {
+ case PqMsg_ErrorResponse:
+ case PqMsg_NoticeResponse:
+ {
+ ErrorData edata;
+
+ /* Parse ErrorResponse or NoticeResponse. */
+ pq_parse_errornotice(msg, &edata);
+
+ /* Death of a worker isn't enough justification for suicide. */
+ edata.elevel = Min(edata.elevel, ERROR);
+
+ /*
+ * If desired, add a context line to show that this is a
+ * message propagated from a parallel worker. Otherwise, it
+ * can sometimes be confusing to understand what actually
+ * happened.
+ */
+ if (edata.context)
+ edata.context = psprintf("%s\n%s", edata.context,
+ _("decoding worker"));
+ else
+ edata.context = pstrdup(_("decoding worker"));
+
+ /* Rethrow error or print notice. */
+ ThrowErrorData(&edata);
+
+ break;
+ }
+
+ default:
+ {
+ elog(ERROR, "unrecognized message type received from decoding worker: %c (message length %d bytes)",
+ msgtype, msg->len);
+ }
+ }
+}
diff --git a/src/backend/libpq/pqmq.c b/src/backend/libpq/pqmq.c
index 6e4bbfb5aa1..42f6fa472c5 100644
--- a/src/backend/libpq/pqmq.c
+++ b/src/backend/libpq/pqmq.c
@@ -14,6 +14,7 @@
#include "postgres.h"
#include "access/parallel.h"
+#include "commands/cluster.h"
#include "libpq/libpq.h"
#include "libpq/pqformat.h"
#include "libpq/pqmq.h"
@@ -175,6 +176,10 @@ mq_putmessage(char msgtype, const char *s, size_t len)
SendProcSignal(pq_mq_parallel_leader_pid,
PROCSIG_PARALLEL_APPLY_MESSAGE,
pq_mq_parallel_leader_proc_number);
+ else if (IsRepackWorker())
+ SendProcSignal(pq_mq_parallel_leader_pid,
+ PROCSIG_REPACK_MESSAGE,
+ pq_mq_parallel_leader_proc_number);
else
{
Assert(IsParallelWorker());
diff --git a/src/backend/postmaster/bgworker.c b/src/backend/postmaster/bgworker.c
index 261ccd3f59c..09c58371e8e 100644
--- a/src/backend/postmaster/bgworker.c
+++ b/src/backend/postmaster/bgworker.c
@@ -13,6 +13,7 @@
#include "postgres.h"
#include "access/parallel.h"
+#include "commands/cluster.h"
#include "libpq/pqsignal.h"
#include "miscadmin.h"
#include "pgstat.h"
@@ -142,6 +143,10 @@ static const struct
{
.fn_name = "SequenceSyncWorkerMain",
.fn_addr = SequenceSyncWorkerMain
+ },
+ {
+ .fn_name = "RepackWorkerMain",
+ .fn_addr = RepackWorkerMain
}
};
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index 603a2b94d05..7651b187418 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -194,7 +194,11 @@ StartupDecodingContext(List *output_plugin_options,
ctx->slot = slot;
- ctx->reader = XLogReaderAllocate(wal_segment_size, NULL, xl_routine, ctx);
+ /*
+ * TODO A separate patch for PG core, unless there's really a reason to
+ * pass ctx for private_data (May extensions expect ctx?).
+ */
+ ctx->reader = XLogReaderAllocate(wal_segment_size, NULL, xl_routine, NULL);
if (!ctx->reader)
ereport(ERROR,
(errcode(ERRCODE_OUT_OF_MEMORY),
diff --git a/src/backend/replication/pgoutput_repack/pgoutput_repack.c b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
index 6b54ea040ac..3a109dbfaff 100644
--- a/src/backend/replication/pgoutput_repack/pgoutput_repack.c
+++ b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
@@ -167,17 +167,13 @@ store_change(LogicalDecodingContext *ctx, ConcurrentChangeKind kind,
HeapTuple tuple)
{
RepackDecodingState *dstate;
- char *change_raw;
- ConcurrentChange change;
+ char kind_byte = (char) kind;
bool flattened = false;
- Size size;
- Datum values[1];
- bool isnull[1];
- char *dst;
dstate = (RepackDecodingState *) ctx->output_writer_private;
- size = VARHDRSZ + SizeOfConcurrentChange;
+ /* Store the change kind. */
+ BufFileWrite(dstate->file, &kind_byte, 1);
/*
* ReorderBufferCommit() stores the TOAST chunks in its private memory
@@ -194,46 +190,12 @@ store_change(LogicalDecodingContext *ctx, ConcurrentChangeKind kind,
tuple = toast_flatten_tuple(tuple, dstate->tupdesc);
flattened = true;
}
+ /* Store the tuple size ... */
+ BufFileWrite(dstate->file, &tuple->t_len, sizeof(tuple->t_len));
+ /* ... and the tuple itself. */
+ BufFileWrite(dstate->file, tuple->t_data, tuple->t_len);
- size += tuple->t_len;
- if (size >= MaxAllocSize)
- elog(ERROR, "Change is too big.");
-
- /* Construct the change. */
- change_raw = (char *) palloc0(size);
- SET_VARSIZE(change_raw, size);
-
- /*
- * Since the varlena alignment might not be sufficient for the structure,
- * set the fields in a local instance and remember where it should
- * eventually be copied.
- */
- change.kind = kind;
- dst = (char *) VARDATA(change_raw);
-
- /*
- * Copy the tuple.
- *
- * Note: change->tup_data.t_data must be fixed on retrieval!
- */
- memcpy(&change.tup_data, tuple, sizeof(HeapTupleData));
- memcpy(dst, &change, SizeOfConcurrentChange);
- dst += SizeOfConcurrentChange;
- memcpy(dst, tuple->t_data, tuple->t_len);
-
- /* The data has been copied. */
+ /* Free the flat copy if created above. */
if (flattened)
pfree(tuple);
-
- /* Store as tuple of 1 bytea column. */
- values[0] = PointerGetDatum(change_raw);
- isnull[0] = false;
- tuplestore_putvalues(dstate->tstore, dstate->tupdesc_change,
- values, isnull);
-
- /* Accounting. */
- dstate->nchanges++;
-
- /* Cleanup. */
- pfree(change_raw);
}
diff --git a/src/backend/storage/ipc/procsignal.c b/src/backend/storage/ipc/procsignal.c
index 8e56922dcea..6f9e7a7aab7 100644
--- a/src/backend/storage/ipc/procsignal.c
+++ b/src/backend/storage/ipc/procsignal.c
@@ -19,6 +19,7 @@
#include "access/parallel.h"
#include "commands/async.h"
+#include "commands/cluster.h"
#include "miscadmin.h"
#include "pgstat.h"
#include "port/pg_bitutils.h"
@@ -697,6 +698,9 @@ procsignal_sigusr1_handler(SIGNAL_ARGS)
if (CheckProcSignal(PROCSIG_PARALLEL_APPLY_MESSAGE))
HandleParallelApplyMessageInterrupt();
+ if (CheckProcSignal(PROCSIG_REPACK_MESSAGE))
+ HandleRepackMessageInterrupt();
+
if (CheckProcSignal(PROCSIG_RECOVERY_CONFLICT_DATABASE))
HandleRecoveryConflictInterrupt(PROCSIG_RECOVERY_CONFLICT_DATABASE);
diff --git a/src/backend/tcop/postgres.c b/src/backend/tcop/postgres.c
index 02e9aaa6bca..e08bf56f11e 100644
--- a/src/backend/tcop/postgres.c
+++ b/src/backend/tcop/postgres.c
@@ -36,6 +36,7 @@
#include "access/xact.h"
#include "catalog/pg_type.h"
#include "commands/async.h"
+#include "commands/cluster.h"
#include "commands/event_trigger.h"
#include "commands/explain_state.h"
#include "commands/prepare.h"
@@ -3520,6 +3521,9 @@ ProcessInterrupts(void)
if (ParallelApplyMessagePending)
ProcessParallelApplyMessages();
+
+ if (RepackMessagePending)
+ ProcessRepackMessages();
}
/*
diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt
index 4aa864fe3c3..b00bd794759 100644
--- a/src/backend/utils/activity/wait_event_names.txt
+++ b/src/backend/utils/activity/wait_event_names.txt
@@ -154,6 +154,7 @@ RECOVERY_CONFLICT_SNAPSHOT "Waiting for recovery conflict resolution for a vacuu
RECOVERY_CONFLICT_TABLESPACE "Waiting for recovery conflict resolution for dropping a tablespace."
RECOVERY_END_COMMAND "Waiting for <xref linkend="guc-recovery-end-command"/> to complete."
RECOVERY_PAUSE "Waiting for recovery to be resumed."
+REPACK_WORKER_EXPORT "Waiting for decoding worker to export a new output file."
REPLICATION_ORIGIN_DROP "Waiting for a replication origin to become inactive so it can be dropped."
REPLICATION_SLOT_DROP "Waiting for a replication slot to become inactive so it can be dropped."
RESTORE_COMMAND "Waiting for <xref linkend="guc-restore-command"/> to complete."
diff --git a/src/include/access/tableam.h b/src/include/access/tableam.h
index 14928cd04a1..f1005afde9d 100644
--- a/src/include/access/tableam.h
+++ b/src/include/access/tableam.h
@@ -22,7 +22,6 @@
#include "access/xact.h"
#include "commands/vacuum.h"
#include "executor/tuptable.h"
-#include "replication/logical.h"
#include "storage/read_stream.h"
#include "utils/rel.h"
#include "utils/snapshot.h"
@@ -631,7 +630,6 @@ typedef struct TableAmRoutine
bool use_sort,
TransactionId OldestXmin,
Snapshot snapshot,
- LogicalDecodingContext *decoding_ctx,
TransactionId *xid_cutoff,
MultiXactId *multi_cutoff,
double *num_tuples,
@@ -1663,8 +1661,6 @@ table_relation_copy_data(Relation rel, const RelFileLocator *newrlocator)
* - *multi_cutoff - ditto
* - snapshot - if != NULL, ignore data changes done by transactions that this
* (MVCC) snapshot considers still in-progress or in the future.
- * - decoding_ctx - logical decoding context, to capture concurrent data
- * changes.
*
* Output parameters:
* - *xid_cutoff - rel's new relfrozenxid value, may be invalid
@@ -1678,7 +1674,6 @@ table_relation_copy_for_cluster(Relation OldTable, Relation NewTable,
bool use_sort,
TransactionId OldestXmin,
Snapshot snapshot,
- LogicalDecodingContext *decoding_ctx,
TransactionId *xid_cutoff,
MultiXactId *multi_cutoff,
double *num_tuples,
@@ -1687,7 +1682,7 @@ table_relation_copy_for_cluster(Relation OldTable, Relation NewTable,
{
OldTable->rd_tableam->relation_copy_for_cluster(OldTable, NewTable, OldIndex,
use_sort, OldestXmin,
- snapshot, decoding_ctx,
+ snapshot,
xid_cutoff, multi_cutoff,
num_tuples, tups_vacuumed,
tups_recently_dead);
diff --git a/src/include/commands/cluster.h b/src/include/commands/cluster.h
index 6a5c476294a..1b05d5d418b 100644
--- a/src/include/commands/cluster.h
+++ b/src/include/commands/cluster.h
@@ -17,11 +17,13 @@
#include "nodes/parsenodes.h"
#include "parser/parse_node.h"
#include "replication/decode.h"
+#include "postmaster/bgworker.h"
#include "replication/logical.h"
+#include "storage/buffile.h"
#include "storage/lock.h"
+#include "storage/shm_mq.h"
#include "utils/relcache.h"
#include "utils/resowner.h"
-#include "utils/tuplestore.h"
/* flag bits for ClusterParams->options */
@@ -44,6 +46,9 @@ typedef struct ClusterParams
* The following definitions are used by REPACK CONCURRENTLY.
*/
+/*
+ * Stored as a single byte in the output file.
+ */
typedef enum
{
CHANGE_INSERT,
@@ -52,68 +57,30 @@ typedef enum
CHANGE_DELETE
} ConcurrentChangeKind;
-typedef struct ConcurrentChange
-{
- /* See the enum above. */
- ConcurrentChangeKind kind;
-
- /*
- * The actual tuple.
- *
- * The tuple data follows the ConcurrentChange structure. Before use make
- * sure the tuple is correctly aligned (ConcurrentChange can be stored as
- * bytea) and that tuple->t_data is fixed.
- */
- HeapTupleData tup_data;
-} ConcurrentChange;
-
-#define SizeOfConcurrentChange (offsetof(ConcurrentChange, tup_data) + \
- sizeof(HeapTupleData))
-
/*
* Logical decoding state.
*
- * Here we store the data changes that we decode from WAL while the table
- * contents is being copied to a new storage. Also the necessary metadata
- * needed to apply these changes to the table is stored here.
+ * The output plugin uses it to store the data changes that it decodes from
+ * WAL while the table contents is being copied to a new storage.
*/
typedef struct RepackDecodingState
{
/* The relation whose changes we're decoding. */
Oid relid;
- /* Replication slot name. */
- NameData slotname;
-
- /*
- * Decoded changes are stored here. Although we try to avoid excessive
- * batches, it can happen that the changes need to be stored to disk. The
- * tuplestore does this transparently.
- */
- Tuplestorestate *tstore;
-
- /* The current number of changes in tstore. */
- double nchanges;
-
- /*
- * Descriptor to store the ConcurrentChange structure serialized (bytea).
- * We can't store the tuple directly because tuplestore only supports
- * minimum tuple and we may need to transfer OID system column from the
- * output plugin. Also we need to transfer the change kind, so it's better
- * to put everything in the structure than to use 2 tuplestores "in
- * parallel".
- */
- TupleDesc tupdesc_change;
-
- /* Tuple descriptor needed to update indexes. */
+ /* Tuple descriptor of the relation being processed. */
TupleDesc tupdesc;
- /* Slot to retrieve data from tstore. */
- TupleTableSlot *tsslot;
-
- ResourceOwner resowner;
+ /* The current output file. */
+ BufFile *file;
} RepackDecodingState;
+extern PGDLLIMPORT volatile sig_atomic_t RepackMessagePending;
+
+extern bool IsRepackWorker(void);
+extern void HandleRepackMessageInterrupt(void);
+extern void ProcessRepackMessages(void);
+
extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel);
extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid,
@@ -136,6 +103,6 @@ extern void finish_heap_swap(Oid OIDOldHeap, Oid OIDNewHeap,
extern bool am_decoding_for_repack(void);
extern bool change_useless_for_repack(XLogRecordBuffer *buf);
-extern void repack_decode_concurrent_changes(LogicalDecodingContext *ctx,
- XLogRecPtr end_of_wal);
+
+extern void RepackWorkerMain(Datum main_arg);
#endif /* CLUSTER_H */
diff --git a/src/include/storage/procsignal.h b/src/include/storage/procsignal.h
index e52b8eb7697..3ef35ca6b80 100644
--- a/src/include/storage/procsignal.h
+++ b/src/include/storage/procsignal.h
@@ -36,6 +36,7 @@ typedef enum
PROCSIG_BARRIER, /* global barrier interrupt */
PROCSIG_LOG_MEMORY_CONTEXT, /* ask backend to log the memory contexts */
PROCSIG_PARALLEL_APPLY_MESSAGE, /* Message from parallel apply workers */
+ PROCSIG_REPACK_MESSAGE, /* Message from repack worker */
/* Recovery conflict reasons */
PROCSIG_RECOVERY_CONFLICT_FIRST,
diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list
index 2676823d2a7..84c7b680dc5 100644
--- a/src/tools/pgindent/typedefs.list
+++ b/src/tools/pgindent/typedefs.list
@@ -498,7 +498,6 @@ CompressFileHandle
CompressionLocation
CompressorState
ComputeXidHorizonsResult
-ConcurrentChange
ConcurrentChangeKind
ConditionVariable
ConditionVariableMinimallyPadded
@@ -638,6 +637,8 @@ DeclareCursorStmt
DecodedBkpBlock
DecodedXLogRecord
DecodingOutputState
+DecodingWorker
+DecodingWorkerShared
DefElem
DefElemAction
DefaultACLInfo
--
2.47.3
--r7amdyq3rmktb267--
^ permalink raw reply [nested|flat] 55+ messages in thread
* [PATCH 5/6] Use background worker to do logical decoding.
@ 2026-01-27 10:48 Antonin Houska <[email protected]>
0 siblings, 0 replies; 55+ messages in thread
From: Antonin Houska @ 2026-01-27 10:48 UTC (permalink / raw)
If the backend performing REPACK (CONCURRENTLY) does both data copying and
logical decoding, it has to "travel in time" back and forth and therefore it
has to invalidate system caches quite a few times. (The copying and the
decoding work with different catalog snapshots.) As the decoding worker has
separate caches, the switching is not necessary.
Without the worker, it'd also be difficult to switch between potentially long
running tasks like index build and WAL decoding. (No decoding during that time
at all can suspend archiving / recycling of WAL segments for some time, which
in turn may result in full disk.)
Another problem is that, after having acquired AccessExclusiveLock (in order
to swap the files), the backend needs to both decode and apply the data
changes that took place while it was waiting for the lock. With the decoding
worker, the decoding runs all the time, so the backend only needs to apply the
changes. This can reduce the time the exclusive lock is held for.
Note that the code added in order to handle ERRORs in the background worker
almost duplicates the existing code that does the same for other types of
workers (See ProcessParallelMessages() and
ProcessParallelApplyMessages()). Refactoring of the existing code might be
useful, to reduce the duplication.
---
src/backend/access/heap/heapam_handler.c | 44 -
src/backend/commands/cluster.c | 1179 +++++++++++++----
src/backend/libpq/pqmq.c | 5 +
src/backend/postmaster/bgworker.c | 4 +
src/backend/replication/logical/logical.c | 6 +-
.../pgoutput_repack/pgoutput_repack.c | 54 +-
src/backend/storage/ipc/procsignal.c | 4 +
src/backend/tcop/postgres.c | 4 +
.../utils/activity/wait_event_names.txt | 1 +
src/include/access/tableam.h | 7 +-
src/include/commands/cluster.h | 71 +-
src/include/storage/procsignal.h | 1 +
src/tools/pgindent/typedefs.list | 3 +-
13 files changed, 981 insertions(+), 402 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index 908f1ef66c6..8589b3c940e 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -33,7 +33,6 @@
#include "catalog/index.h"
#include "catalog/storage.h"
#include "catalog/storage_xlog.h"
-#include "commands/cluster.h"
#include "commands/progress.h"
#include "executor/executor.h"
#include "miscadmin.h"
@@ -688,7 +687,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
Relation OldIndex, bool use_sort,
TransactionId OldestXmin,
Snapshot snapshot,
- LogicalDecodingContext *decoding_ctx,
TransactionId *xid_cutoff,
MultiXactId *multi_cutoff,
double *num_tuples,
@@ -710,7 +708,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
BufferHeapTupleTableSlot *hslot;
BlockNumber prev_cblock = InvalidBlockNumber;
bool concurrent = snapshot != NULL;
- XLogRecPtr end_of_wal_prev = GetFlushRecPtr(NULL);
/* Remember if it's a system catalog */
is_system_catalog = IsSystemRelation(OldHeap);
@@ -971,31 +968,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
ct_val[1] = *num_tuples;
pgstat_progress_update_multi_param(2, ct_index, ct_val);
}
-
- /*
- * Process the WAL produced by the load, as well as by other
- * transactions, so that the replication slot can advance and WAL does
- * not pile up. Use wal_segment_size as a threshold so that we do not
- * introduce the decoding overhead too often.
- *
- * Of course, we must not apply the changes until the initial load has
- * completed.
- *
- * Note that our insertions into the new table should not be decoded
- * as we (intentionally) do not write the logical decoding specific
- * information to WAL.
- */
- if (concurrent)
- {
- XLogRecPtr end_of_wal;
-
- end_of_wal = GetFlushRecPtr(NULL);
- if ((end_of_wal - end_of_wal_prev) > wal_segment_size)
- {
- repack_decode_concurrent_changes(decoding_ctx, end_of_wal);
- end_of_wal_prev = end_of_wal;
- }
- }
}
if (indexScan != NULL)
@@ -1041,22 +1013,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
-
- /*
- * Try to keep the amount of not-yet-decoded WAL small, like
- * above.
- */
- if (concurrent)
- {
- XLogRecPtr end_of_wal;
-
- end_of_wal = GetFlushRecPtr(NULL);
- if ((end_of_wal - end_of_wal_prev) > wal_segment_size)
- {
- repack_decode_concurrent_changes(decoding_ctx, end_of_wal);
- end_of_wal_prev = end_of_wal;
- }
- }
}
tuplesort_end(tuplesort);
diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c
index 03ccf10b782..e988a7a7296 100644
--- a/src/backend/commands/cluster.c
+++ b/src/backend/commands/cluster.c
@@ -12,12 +12,13 @@
* In concurrent mode, we lock the table with only ShareUpdateExclusiveLock,
* then do an initial copy as above. However, while the tuples are being
* copied, concurrent transactions could modify the table. To cope with those
- * changes, we rely on logical decoding to obtain them from WAL. The changes
- * are accumulated in a tuplestore. Once the initial copy is complete, we
- * read the changes from the tuplestore and re-apply them on the new heap.
- * Then we upgrade our ShareUpdateExclusiveLock to AccessExclusiveLock and
- * swap the relfilenodes. This way, the time we hold a strong lock on the
- * table is much reduced, and the bloat is eliminated.
+ * changes, we rely on logical decoding to obtain them from WAL. A bgworker
+ * consumes WAL while the initial copy is ongoing (to prevent excessive WAL
+ * from being reserved), and accumulates the changes in a file. Once the
+ * initial copy is complete, we read the changes from the file and re-apply
+ * them on the new heap. Then we upgrade our ShareUpdateExclusiveLock to
+ * AccessExclusiveLock and swap the relfilenodes. This way, the time we hold
+ * a strong lock on the table is much reduced, and the bloat is eliminated.
*
* There is hardly anything left of Paul Brown's original implementation...
*
@@ -45,6 +46,7 @@
#include "access/xlog_internal.h"
#include "access/xloginsert.h"
#include "access/xlogutils.h"
+#include "access/xlogwait.h"
#include "catalog/catalog.h"
#include "catalog/dependency.h"
#include "catalog/heap.h"
@@ -61,6 +63,8 @@
#include "commands/tablecmds.h"
#include "commands/vacuum.h"
#include "executor/executor.h"
+#include "libpq/pqformat.h"
+#include "libpq/pqmq.h"
#include "miscadmin.h"
#include "optimizer/optimizer.h"
#include "pgstat.h"
@@ -71,6 +75,8 @@
#include "storage/ipc.h"
#include "storage/lmgr.h"
#include "storage/predicate.h"
+#include "storage/procsignal.h"
+#include "tcop/tcopprot.h"
#include "utils/acl.h"
#include "utils/fmgroids.h"
#include "utils/guc.h"
@@ -117,6 +123,12 @@ typedef struct IndexInsertState
/* The WAL segment being decoded. */
static XLogSegNo repack_current_segment = 0;
+/*
+ * The first file exported by the decoding worker must contain a snapshot, the
+ * following ones contain the data changes.
+ */
+#define WORKER_FILE_SNAPSHOT 0
+
/*
* Information needed to apply concurrent data changes.
*/
@@ -136,8 +148,113 @@ typedef struct ChangeDest
/* Needed to update indexes of rel_dst. */
IndexInsertState *iistate;
+
+ /*
+ * Sequential number of the file containing the changes.
+ *
+ * TODO This field makes the structure name less descriptive. Should we
+ * rename it, e.g. to ChangeApplyInfo?
+ */
+ int file_seq;
} ChangeDest;
+/*
+ * Layout of shared memory used for communication between backend and the
+ * worker that performs logical decoding of data changes
+ */
+typedef struct DecodingWorkerShared
+{
+ /* Is the decoding initialized? */
+ bool initialized;
+
+ /*
+ * Once the worker has reached this LSN, it should close the current
+ * output file and either create a new one or exit, according to the field
+ * 'done'. If the value is InvalidXLogRecPtr, the worker should decode all
+ * the WAL available and keep checking this field. It is ok if the worker
+ * had already decoded records whose LSN is >= lsn_upto before this field
+ * has been set.
+ */
+ XLogRecPtr lsn_upto;
+
+ /* Exit after closing the current file? */
+ bool done;
+
+ /* The output is stored here. */
+ SharedFileSet sfs;
+
+ /* Number of the last file exported by the worker. */
+ int last_exported;
+
+ /* Synchronize access to the fields above. */
+ slock_t mutex;
+
+ /* Database to connect to. */
+ Oid dbid;
+
+ /* Role to connect as. */
+ Oid roleid;
+
+ /* Decode data changes of this relation. */
+ Oid relid;
+
+ /* The backend uses this to wait for the worker. */
+ ConditionVariable cv;
+
+ /* Info to signal the backend. */
+ PGPROC *backend_proc;
+ pid_t backend_pid;
+ ProcNumber backend_proc_number;
+
+ /* Error queue. */
+ shm_mq *error_mq;
+
+ /*
+ * Memory the queue is located in.
+ *
+ * For considerations on the value see the comments of
+ * PARALLEL_ERROR_QUEUE_SIZE.
+ */
+#define REPACK_ERROR_QUEUE_SIZE 16384
+ char error_queue[FLEXIBLE_ARRAY_MEMBER];
+} DecodingWorkerShared;
+
+/*
+ * Generate worker's output file name. If relations of the same 'relid' happen
+ * to be processed at the same time, they must be from different databases and
+ * therefore different backends must be involved. (PID is already present in
+ * the fileset name.)
+ */
+static inline void
+DecodingWorkerFileName(char *fname, Oid relid, uint32 seq)
+{
+ snprintf(fname, MAXPGPATH, "%u-%u", relid, seq);
+}
+
+/*
+ * Backend-local information to control the decoding worker.
+ */
+typedef struct DecodingWorker
+{
+ /* The worker. */
+ BackgroundWorkerHandle *handle;
+
+ /* DecodingWorkerShared is in this segment. */
+ dsm_segment *seg;
+
+ /* Handle of the error queue. */
+ shm_mq_handle *error_mqh;
+} DecodingWorker;
+
+/* Pointer to currently running decoding worker. */
+static DecodingWorker *decoding_worker = NULL;
+
+/*
+ * Is there a message sent by a repack worker that the backend needs to
+ * receive?
+ */
+volatile sig_atomic_t RepackMessagePending = false;
+
static bool cluster_rel_recheck(RepackCommand cmd, Relation OldHeap,
Oid indexOid, Oid userid, LOCKMODE lmode,
int options);
@@ -145,7 +262,7 @@ static void check_repack_concurrently_requirements(Relation rel);
static void rebuild_relation(Relation OldHeap, Relation index, bool verbose,
bool concurrent);
static void copy_table_data(Relation NewHeap, Relation OldHeap, Relation OldIndex,
- Snapshot snapshot, LogicalDecodingContext *decoding_ctx,
+ Snapshot snapshot,
bool verbose,
bool *pSwapToastByContent,
TransactionId *pFreezeXid,
@@ -158,12 +275,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd,
static bool cluster_is_permitted_for_relation(RepackCommand cmd,
Oid relid, Oid userid);
-static void begin_concurrent_repack(Relation rel);
-static void end_concurrent_repack(void);
static LogicalDecodingContext *setup_logical_decoding(Oid relid);
-static HeapTuple get_changed_tuple(char *change);
-static void apply_concurrent_changes(RepackDecodingState *dstate,
- ChangeDest *dest);
+static bool decode_concurrent_changes(LogicalDecodingContext *ctx,
+ DecodingWorkerShared *shared);
+static void apply_concurrent_changes(BufFile *file, ChangeDest *dest);
static void apply_concurrent_insert(Relation rel, HeapTuple tup,
IndexInsertState *iistate,
TupleTableSlot *index_slot);
@@ -175,9 +290,9 @@ static void apply_concurrent_delete(Relation rel, HeapTuple tup_target);
static HeapTuple find_target_tuple(Relation rel, ChangeDest *dest,
HeapTuple tup_key,
TupleTableSlot *ident_slot);
-static void process_concurrent_changes(LogicalDecodingContext *decoding_ctx,
- XLogRecPtr end_of_wal,
- ChangeDest *dest);
+static void process_concurrent_changes(XLogRecPtr end_of_wal,
+ ChangeDest *dest,
+ bool done);
static IndexInsertState *get_index_insert_state(Relation relation,
Oid ident_index_id,
Relation *ident_index_p);
@@ -186,7 +301,6 @@ static ScanKey build_identity_key(Oid ident_idx_oid, Relation rel_src,
static void free_index_insert_state(IndexInsertState *iistate);
static void cleanup_logical_decoding(LogicalDecodingContext *ctx);
static void rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
- LogicalDecodingContext *decoding_ctx,
TransactionId frozenXid,
MultiXactId cutoffMulti);
static List *build_new_indexes(Relation NewHeap, Relation OldHeap, List *OldIndexes);
@@ -196,6 +310,13 @@ static Relation process_single_relation(RepackStmt *stmt,
ClusterParams *params);
static Oid determine_clustered_index(Relation rel, bool usingindex,
const char *indexname);
+static void start_decoding_worker(Oid relid);
+static void stop_decoding_worker(void);
+static void repack_worker_internal(dsm_segment *seg);
+static void export_initial_snapshot(Snapshot snapshot,
+ DecodingWorkerShared *shared);
+static Snapshot get_initial_snapshot(DecodingWorker *worker);
+static void ProcessRepackMessage(StringInfo msg);
static const char *RepackCommandAsString(RepackCommand cmd);
@@ -619,20 +740,20 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid,
/* rebuild_relation does all the dirty work */
PG_TRY();
{
- /*
- * For concurrent processing, make sure that our logical decoding
- * ignores data changes of other tables than the one we are
- * processing.
- */
- if (concurrent)
- begin_concurrent_repack(OldHeap);
-
rebuild_relation(OldHeap, index, verbose, concurrent);
}
PG_FINALLY();
{
if (concurrent)
- end_concurrent_repack();
+ {
+ /*
+ * Since during normal operation the worker was already asked to
+ * exit, stopping it explicitly is especially important on ERROR.
+ * However it still seems a good practice to make sure that the
+ * worker never survives the REPACK command.
+ */
+ stop_decoding_worker();
+ }
}
PG_END_TRY();
@@ -929,7 +1050,6 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
bool swap_toast_by_content;
TransactionId frozenXid;
MultiXactId cutoffMulti;
- LogicalDecodingContext *decoding_ctx = NULL;
Snapshot snapshot = NULL;
#if USE_ASSERT_CHECKING
LOCKMODE lmode;
@@ -943,19 +1063,36 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
if (concurrent)
{
/*
- * Prepare to capture the concurrent data changes.
+ * The worker needs to be member of the locking group we're the leader
+ * of. We ought to become the leader before the worker starts. The
+ * worker will join the group as soon as it starts.
*
- * Note that this call waits for all transactions with XID already
- * assigned to finish. If some of those transactions is waiting for a
- * lock conflicting with ShareUpdateExclusiveLock on our table (e.g.
- * it runs CREATE INDEX), we can end up in a deadlock. Not sure this
- * risk is worth unlocking/locking the table (and its clustering
- * index) and checking again if its still eligible for REPACK
- * CONCURRENTLY.
+ * This is to make sure that the deadlock described below is
+ * detectable by deadlock.c: if the worker waits for a transaction to
+ * complete and we are waiting for the worker output, then effectively
+ * we (i.e. this backend) are waiting for that transaction.
*/
- decoding_ctx = setup_logical_decoding(tableOid);
+ BecomeLockGroupLeader();
+
+ /*
+ * Start the worker that decodes data changes applied while we're
+ * copying the table contents.
+ *
+ * Note that the worker has to wait for all transactions with XID
+ * already assigned to finish. If some of those transactions is
+ * waiting for a lock conflicting with ShareUpdateExclusiveLock on our
+ * table (e.g. it runs CREATE INDEX), we can end up in a deadlock.
+ * Not sure this risk is worth unlocking/locking the table (and its
+ * clustering index) and checking again if it's still eligible for
+ * REPACK CONCURRENTLY.
+ */
+ start_decoding_worker(tableOid);
+
+ /*
+ * Wait until the worker has the initial snapshot and retrieve it.
+ */
+ snapshot = get_initial_snapshot(decoding_worker);
- snapshot = SnapBuildInitialSnapshotForRepack(decoding_ctx->snapshot_builder);
PushActiveSnapshot(snapshot);
}
@@ -980,7 +1117,7 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
NewHeap = table_open(OIDNewHeap, NoLock);
/* Copy the heap data into the new table in the desired order */
- copy_table_data(NewHeap, OldHeap, index, snapshot, decoding_ctx, verbose,
+ copy_table_data(NewHeap, OldHeap, index, snapshot, verbose,
&swap_toast_by_content, &frozenXid, &cutoffMulti);
/* The historic snapshot won't be needed anymore. */
@@ -1001,14 +1138,11 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
if (index)
index_close(index, NoLock);
- rebuild_relation_finish_concurrent(NewHeap, OldHeap, decoding_ctx,
- frozenXid, cutoffMulti);
+ rebuild_relation_finish_concurrent(NewHeap, OldHeap, frozenXid,
+ cutoffMulti);
pgstat_progress_update_param(PROGRESS_REPACK_PHASE,
PROGRESS_REPACK_PHASE_FINAL_CLEANUP);
-
- /* Done with decoding. */
- cleanup_logical_decoding(decoding_ctx);
}
else
{
@@ -1179,8 +1313,7 @@ make_new_heap(Oid OIDOldHeap, Oid NewTableSpace, Oid NewAccessMethod,
*/
static void
copy_table_data(Relation NewHeap, Relation OldHeap, Relation OldIndex,
- Snapshot snapshot, LogicalDecodingContext *decoding_ctx,
- bool verbose, bool *pSwapToastByContent,
+ Snapshot snapshot, bool verbose, bool *pSwapToastByContent,
TransactionId *pFreezeXid, MultiXactId *pCutoffMulti)
{
Relation relRelation;
@@ -1341,7 +1474,6 @@ copy_table_data(Relation NewHeap, Relation OldHeap, Relation OldIndex,
*/
table_relation_copy_for_cluster(OldHeap, NewHeap, OldIndex, use_sort,
cutoffs.OldestXmin, snapshot,
- decoding_ctx,
&cutoffs.FreezeLimit,
&cutoffs.MultiXactCutoff,
&num_tuples, &tups_vacuumed,
@@ -2371,62 +2503,10 @@ RepackCommandAsString(RepackCommand cmd)
case REPACK_COMMAND_CLUSTER:
return "CLUSTER";
}
- return "???"; /* keep compiler quiet */
+ return "???"; /* keep compiler quiet */
}
-/*
- * Call this function before REPACK CONCURRENTLY starts to setup logical
- * decoding. It makes sure that other users of the table put enough
- * information into WAL.
- *
- * The point is that at various places we expect that the table we're
- * processing is treated like a system catalog. For example, we need to be
- * able to scan it using a "historic snapshot" anytime during the processing
- * (as opposed to scanning only at the start point of the decoding, as logical
- * replication does during initial table synchronization), in order to apply
- * concurrent UPDATE / DELETE commands.
- *
- * Note that TOAST table needs no attention here as it's not scanned using
- * historic snapshot.
- */
-static void
-begin_concurrent_repack(Relation rel)
-{
- Oid toastrelid;
-
- /*
- * Avoid logical decoding of other relations by this backend. The lock we
- * have guarantees that the actual locator cannot be changed concurrently:
- * TRUNCATE needs AccessExclusiveLock.
- */
- Assert(CheckRelationLockedByMe(rel, ShareUpdateExclusiveLock, false));
- repacked_rel_locator = rel->rd_locator;
- toastrelid = rel->rd_rel->reltoastrelid;
- if (OidIsValid(toastrelid))
- {
- Relation toastrel;
-
- /* Avoid logical decoding of other TOAST relations. */
- toastrel = table_open(toastrelid, AccessShareLock);
- repacked_rel_toast_locator = toastrel->rd_locator;
- table_close(toastrel, AccessShareLock);
- }
-}
-
-/*
- * Call this when done with REPACK CONCURRENTLY.
- */
-static void
-end_concurrent_repack(void)
-{
- /*
- * Restore normal function of (future) logical decoding for this backend.
- */
- repacked_rel_locator.relNumber = InvalidOid;
- repacked_rel_toast_locator.relNumber = InvalidOid;
-}
-
/*
* Is this backend performing logical decoding on behalf of REPACK
* (CONCURRENTLY) ?
@@ -2491,9 +2571,10 @@ static LogicalDecodingContext *
setup_logical_decoding(Oid relid)
{
Relation rel;
- TupleDesc tupdesc;
+ Oid toastrelid;
LogicalDecodingContext *ctx;
- RepackDecodingState *dstate = palloc0_object(RepackDecodingState);
+ NameData slotname;
+ RepackDecodingState *dstate;
/*
* REPACK CONCURRENTLY is not allowed in a transaction block, so this
@@ -2501,20 +2582,21 @@ setup_logical_decoding(Oid relid)
*/
Assert(!TransactionIdIsValid(GetTopTransactionIdIfAny()));
- /*
- * A single backend should not execute multiple REPACK commands at a time,
- * so use PID to make the slot unique.
- */
- snprintf(NameStr(dstate->slotname), NAMEDATALEN, "repack_%d", MyProcPid);
-
/*
* Make sure we can use logical decoding.
*/
CheckSlotPermissions();
CheckLogicalDecodingRequirements();
- /* RS_TEMPORARY so that the slot gets cleaned up on ERROR. */
- ReplicationSlotCreate(NameStr(dstate->slotname), true, RS_TEMPORARY,
- false, false, false);
+ /*
+ * A single backend should not execute multiple REPACK commands at a time,
+ * so use PID to make the slot unique.
+ *
+ * RS_TEMPORARY so that the slot gets cleaned up on ERROR.
+ */
+ snprintf(NameStr(slotname), NAMEDATALEN, "repack_%d", MyProcPid);
+ ReplicationSlotCreate(NameStr(slotname), true, RS_TEMPORARY, false, false,
+ false);
+
EnsureLogicalDecodingEnabled();
/*
@@ -2537,104 +2619,109 @@ setup_logical_decoding(Oid relid)
DecodingContextFindStartpoint(ctx);
+ /*
+ * decode_concurrent_changes() needs non-blocking callback.
+ */
+ ctx->reader->routine.page_read = read_local_xlog_page_no_wait;
+
+ /*
+ * read_local_xlog_page_no_wait() needs to be able to indicate the end of
+ * WAL.
+ */
+ ctx->reader->private_data = MemoryContextAllocZero(ctx->context,
+ sizeof(ReadLocalXLogPageNoWaitPrivate));
+
+
/* Some WAL records should have been read. */
Assert(ctx->reader->EndRecPtr != InvalidXLogRecPtr);
+ /*
+ * Initialize repack_current_segment so that we can notice WAL segment
+ * boundaries.
+ */
XLByteToSeg(ctx->reader->EndRecPtr, repack_current_segment,
wal_segment_size);
- /*
- * Setup structures to store decoded changes.
- */
+ dstate = palloc0_object(RepackDecodingState);
dstate->relid = relid;
- dstate->tstore = tuplestore_begin_heap(false, false,
- maintenance_work_mem);
- /* Caller should already have the table locked. */
- rel = table_open(relid, NoLock);
- tupdesc = CreateTupleDescCopy(RelationGetDescr(rel));
- dstate->tupdesc = tupdesc;
- table_close(rel, NoLock);
+ /*
+ * Tuple descriptor may be needed to flatten a tuple before we write it to
+ * a file. A copy is needed because the decoding worker invalidates system
+ * caches before it starts to do the actual work.
+ */
+ rel = table_open(relid, AccessShareLock);
+ dstate->tupdesc = CreateTupleDescCopy(RelationGetDescr(rel));
- /* Initialize the descriptor to store the changes ... */
- dstate->tupdesc_change = CreateTemplateTupleDesc(1);
+ /* Avoid logical decoding of other relations. */
+ repacked_rel_locator = rel->rd_locator;
+ toastrelid = rel->rd_rel->reltoastrelid;
+ if (OidIsValid(toastrelid))
+ {
+ Relation toastrel;
- TupleDescInitEntry(dstate->tupdesc_change, 1, NULL, BYTEAOID, -1, 0);
- /* ... as well as the corresponding slot. */
- dstate->tsslot = MakeSingleTupleTableSlot(dstate->tupdesc_change,
- &TTSOpsMinimalTuple);
+ /* Avoid logical decoding of other TOAST relations. */
+ toastrel = table_open(toastrelid, AccessShareLock);
+ repacked_rel_toast_locator = toastrel->rd_locator;
+ table_close(toastrel, AccessShareLock);
+ }
+ table_close(rel, AccessShareLock);
- dstate->resowner = ResourceOwnerCreate(CurrentResourceOwner,
- "logical decoding");
+ /* The file will be set as soon as we have it opened. */
+ dstate->file = NULL;
ctx->output_writer_private = dstate;
+
return ctx;
}
/*
- * Retrieve tuple from ConcurrentChange structure.
+ * Decode logical changes from the WAL sequence and store them to a file.
*
- * The input data starts with the structure but it might not be appropriately
- * aligned.
- */
-static HeapTuple
-get_changed_tuple(char *change)
-{
- HeapTupleData tup_data;
- HeapTuple result;
- char *src;
-
- /*
- * Ensure alignment before accessing the fields. (This is why we can't use
- * heap_copytuple() instead of this function.)
- */
- src = change + offsetof(ConcurrentChange, tup_data);
- memcpy(&tup_data, src, sizeof(HeapTupleData));
-
- result = (HeapTuple) palloc(HEAPTUPLESIZE + tup_data.t_len);
- memcpy(result, &tup_data, sizeof(HeapTupleData));
- result->t_data = (HeapTupleHeader) ((char *) result + HEAPTUPLESIZE);
- src = change + SizeOfConcurrentChange;
- memcpy(result->t_data, src, result->t_len);
-
- return result;
-}
-
-/*
- * Decode logical changes from the WAL sequence up to end_of_wal.
+ * If true is returned, there is no more work for the worker.
*/
-void
-repack_decode_concurrent_changes(LogicalDecodingContext *ctx,
- XLogRecPtr end_of_wal)
+static bool
+decode_concurrent_changes(LogicalDecodingContext *ctx,
+ DecodingWorkerShared *shared)
{
RepackDecodingState *dstate;
- ResourceOwner resowner_old;
+ XLogRecPtr lsn_upto;
+ bool done;
+ char fname[MAXPGPATH];
dstate = (RepackDecodingState *) ctx->output_writer_private;
- resowner_old = CurrentResourceOwner;
- CurrentResourceOwner = dstate->resowner;
- PG_TRY();
+ /* Open the output file. */
+ DecodingWorkerFileName(fname, shared->relid, shared->last_exported + 1);
+ dstate->file = BufFileCreateFileSet(&shared->sfs.fs, fname);
+
+ SpinLockAcquire(&shared->mutex);
+ lsn_upto = shared->lsn_upto;
+ done = shared->done;
+ SpinLockRelease(&shared->mutex);
+
+ while (true)
{
- while (ctx->reader->EndRecPtr < end_of_wal)
- {
- XLogRecord *record;
- XLogSegNo segno_new;
- char *errm = NULL;
- XLogRecPtr end_lsn;
+ XLogRecord *record;
+ XLogSegNo segno_new;
+ char *errm = NULL;
+ XLogRecPtr end_lsn;
- record = XLogReadRecord(ctx->reader, &errm);
- if (errm)
- elog(ERROR, "%s", errm);
+ CHECK_FOR_INTERRUPTS();
- if (record != NULL)
- LogicalDecodingProcessRecord(ctx, ctx->reader);
+ record = XLogReadRecord(ctx->reader, &errm);
+ if (record)
+ {
+ LogicalDecodingProcessRecord(ctx, ctx->reader);
/*
* If WAL segment boundary has been crossed, inform the decoding
- * system that the catalog_xmin can advance. (We can confirm more
- * often, but a filling a single WAL segment should not take much
- * time.)
+ * system that the catalog_xmin can advance.
+ *
+ * TODO Does it make sense to confirm more often? Segment size
+ * seems appropriate for restart_lsn (because less than a segment
+ * cannot be recycled anyway), however more frequent checks might
+ * be beneficial for catalog_xmin.
*/
end_lsn = ctx->reader->EndRecPtr;
XLByteToSeg(end_lsn, segno_new, wal_segment_size);
@@ -2645,80 +2732,137 @@ repack_decode_concurrent_changes(LogicalDecodingContext *ctx,
(uint32) (end_lsn >> 32), (uint32) end_lsn);
repack_current_segment = segno_new;
}
+ }
+ else
+ {
+ ReadLocalXLogPageNoWaitPrivate *priv;
+
+ if (errm)
+ ereport(ERROR, (errmsg("%s", errm)));
- CHECK_FOR_INTERRUPTS();
+ /*
+ * In the decoding loop we do not want to get blocked when there
+ * is no more WAL available, otherwise the loop would become
+ * uninterruptible.
+ */
+ priv = (ReadLocalXLogPageNoWaitPrivate *)
+ ctx->reader->private_data;
+ if (priv->end_of_wal)
+ /* Do not miss the end of WAL condition next time. */
+ priv->end_of_wal = false;
+ else
+ ereport(ERROR, (errmsg("could not read WAL record")));
+ }
+
+ /*
+ * Whether we could read new record or not, keep checking if
+ * 'lsn_upto' was specified.
+ */
+ if (XLogRecPtrIsInvalid(lsn_upto))
+ {
+ SpinLockAcquire(&shared->mutex);
+ lsn_upto = shared->lsn_upto;
+ /* 'done' should be set at the same time as 'lsn_upto' */
+ done = shared->done;
+ SpinLockRelease(&shared->mutex);
+ }
+ if (!XLogRecPtrIsInvalid(lsn_upto) &&
+ ctx->reader->EndRecPtr >= lsn_upto)
+ break;
+
+ if (record == NULL)
+ {
+ int64 timeout = 0;
+ WaitLSNResult res;
+
+ /*
+ * Before we retry reading, wait until new WAL is flushed.
+ *
+ * There is a race condition such that the backend executing
+ * REPACK determines 'lsn_upto', but before it sets the shared
+ * variable, we reach the end of WAL. In that case we'd need to
+ * wait until the next WAL flush (unrelated to REPACK). Although
+ * that should not be a problem in a busy system, it might be
+ * noticeable in other cases, including regression tests (which
+ * are not necessarily executed in parallel). Therefore it makes
+ * sense to use timeout.
+ *
+ * If lsn_upto is valid, WAL records having LSN lower than that
+ * should already have been flushed to disk.
+ */
+ if (XLogRecPtrIsInvalid(lsn_upto))
+ timeout = 100L;
+ res = WaitForLSN(WAIT_LSN_TYPE_PRIMARY_FLUSH,
+ ctx->reader->EndRecPtr + 1,
+ timeout);
+ if (res != WAIT_LSN_RESULT_SUCCESS &&
+ res != WAIT_LSN_RESULT_TIMEOUT)
+ ereport(ERROR, (errmsg("waiting for WAL failed")));
}
- InvalidateSystemCaches();
- CurrentResourceOwner = resowner_old;
- }
- PG_CATCH();
- {
- /* clear all timetravel entries */
- InvalidateSystemCaches();
- CurrentResourceOwner = resowner_old;
- PG_RE_THROW();
}
- PG_END_TRY();
+
+ /*
+ * Close the file so we can make it available to the backend.
+ */
+ BufFileClose(dstate->file);
+ dstate->file = NULL;
+ SpinLockAcquire(&shared->mutex);
+ shared->lsn_upto = InvalidXLogRecPtr;
+ shared->last_exported++;
+ SpinLockRelease(&shared->mutex);
+ ConditionVariableSignal(&shared->cv);
+
+ return done;
}
/*
* Apply changes stored in 'file'.
*/
static void
-apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
+apply_concurrent_changes(BufFile *file, ChangeDest *dest)
{
+ char kind;
+ uint32 t_len;
Relation rel = dest->rel;
TupleTableSlot *index_slot,
*ident_slot;
HeapTuple tup_old = NULL;
- if (dstate->nchanges == 0)
- return;
-
/* TupleTableSlot is needed to pass the tuple to ExecInsertIndexTuples(). */
- index_slot = MakeSingleTupleTableSlot(dstate->tupdesc, &TTSOpsHeapTuple);
+ index_slot = MakeSingleTupleTableSlot(RelationGetDescr(rel),
+ &TTSOpsHeapTuple);
/* A slot to fetch tuples from identity index. */
ident_slot = table_slot_create(rel, NULL);
- while (tuplestore_gettupleslot(dstate->tstore, true, false,
- dstate->tsslot))
+ while (true)
{
- bool shouldFree;
- HeapTuple tup_change,
- tup,
+ size_t nread;
+ HeapTuple tup,
tup_exist;
- char *change_raw,
- *src;
- ConcurrentChange change;
- bool isnull[1];
- Datum values[1];
CHECK_FOR_INTERRUPTS();
- /* Get the change from the single-column tuple. */
- tup_change = ExecFetchSlotHeapTuple(dstate->tsslot, false, &shouldFree);
- heap_deform_tuple(tup_change, dstate->tupdesc_change, values, isnull);
- Assert(!isnull[0]);
-
- /* Make sure we access aligned data. */
- change_raw = (char *) DatumGetByteaP(values[0]);
- src = (char *) VARDATA(change_raw);
- memcpy(&change, src, SizeOfConcurrentChange);
+ nread = BufFileReadMaybeEOF(file, &kind, 1, true);
+ /* Are we done with the file? */
+ if (nread == 0)
+ break;
- /*
- * Extract the tuple from the change. The tuple is copied here because
- * it might be assigned to 'tup_old', in which case it needs to
- * survive into the next iteration.
- */
- tup = get_changed_tuple(src);
+ /* Read the tuple. */
+ BufFileReadExact(file, &t_len, sizeof(t_len));
+ tup = (HeapTuple) palloc(HEAPTUPLESIZE + t_len);
+ tup->t_data = (HeapTupleHeader) ((char *) tup + HEAPTUPLESIZE);
+ BufFileReadExact(file, tup->t_data, t_len);
+ tup->t_len = t_len;
+ ItemPointerSetInvalid(&tup->t_self);
+ tup->t_tableOid = RelationGetRelid(dest->rel);
- if (change.kind == CHANGE_UPDATE_OLD)
+ if (kind == CHANGE_UPDATE_OLD)
{
Assert(tup_old == NULL);
tup_old = tup;
}
- else if (change.kind == CHANGE_INSERT)
+ else if (kind == CHANGE_INSERT)
{
Assert(tup_old == NULL);
@@ -2726,12 +2870,11 @@ apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
pfree(tup);
}
- else if (change.kind == CHANGE_UPDATE_NEW ||
- change.kind == CHANGE_DELETE)
+ else if (kind == CHANGE_UPDATE_NEW || kind == CHANGE_DELETE)
{
HeapTuple tup_key;
- if (change.kind == CHANGE_UPDATE_NEW)
+ if (kind == CHANGE_UPDATE_NEW)
{
tup_key = tup_old != NULL ? tup_old : tup;
}
@@ -2748,7 +2891,7 @@ apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
if (tup_exist == NULL)
elog(ERROR, "failed to find target tuple");
- if (change.kind == CHANGE_UPDATE_NEW)
+ if (kind == CHANGE_UPDATE_NEW)
apply_concurrent_update(rel, tup, tup_exist, dest->iistate,
index_slot);
else
@@ -2763,26 +2906,19 @@ apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
pfree(tup);
}
else
- elog(ERROR, "unrecognized kind of change: %d", change.kind);
+ elog(ERROR, "unrecognized kind of change: %d", kind);
/*
* If a change was applied now, increment CID for next writes and
* update the snapshot so it sees the changes we've applied so far.
*/
- if (change.kind != CHANGE_UPDATE_OLD)
+ if (kind != CHANGE_UPDATE_OLD)
{
CommandCounterIncrement();
UpdateActiveSnapshotCommandId();
}
-
- /* TTSOpsMinimalTuple has .get_heap_tuple==NULL. */
- Assert(shouldFree);
- pfree(tup_change);
}
- tuplestore_clear(dstate->tstore);
- dstate->nchanges = 0;
-
/* Cleanup. */
ExecDropSingleTupleTableSlot(index_slot);
ExecDropSingleTupleTableSlot(ident_slot);
@@ -2957,25 +3093,59 @@ find_target_tuple(Relation rel, ChangeDest *dest, HeapTuple tup_key,
}
/*
- * Decode and apply concurrent changes.
+ * Decode and apply concurrent changes, up to (and including) the record whose
+ * LSN is 'end_of_wal'.
*/
static void
-process_concurrent_changes(LogicalDecodingContext *decoding_ctx,
- XLogRecPtr end_of_wal, ChangeDest *dest)
+process_concurrent_changes(XLogRecPtr end_of_wal, ChangeDest *dest, bool done)
{
- RepackDecodingState *dstate;
+ DecodingWorkerShared *shared;
+ char fname[MAXPGPATH];
+ BufFile *file;
pgstat_progress_update_param(PROGRESS_REPACK_PHASE,
PROGRESS_REPACK_PHASE_CATCH_UP);
- dstate = (RepackDecodingState *) decoding_ctx->output_writer_private;
+ /* Ask the worker for the file. */
+ shared = (DecodingWorkerShared *) dsm_segment_address(decoding_worker->seg);
+ SpinLockAcquire(&shared->mutex);
+ shared->lsn_upto = end_of_wal;
+ shared->done = done;
+ SpinLockRelease(&shared->mutex);
+
+ /*
+ * The worker needs to finish processing of the current WAL record. Even
+ * if it's idle, it'll need to close the output file. Thus we're likely to
+ * wait, so prepare for sleep.
+ */
+ ConditionVariablePrepareToSleep(&shared->cv);
+ for (;;)
+ {
+ int last_exported;
- repack_decode_concurrent_changes(decoding_ctx, end_of_wal);
+ SpinLockAcquire(&shared->mutex);
+ last_exported = shared->last_exported;
+ SpinLockRelease(&shared->mutex);
- if (dstate->nchanges == 0)
- return;
+ /*
+ * Has the worker exported the file we are waiting for?
+ */
+ if (last_exported == dest->file_seq)
+ break;
- apply_concurrent_changes(dstate, dest);
+ ConditionVariableSleep(&shared->cv, WAIT_EVENT_REPACK_WORKER_EXPORT);
+ }
+ ConditionVariableCancelSleep();
+
+ /* Open the file. */
+ DecodingWorkerFileName(fname, shared->relid, dest->file_seq);
+ file = BufFileOpenFileSet(&shared->sfs.fs, fname, O_RDONLY, false);
+ apply_concurrent_changes(file, dest);
+
+ BufFileClose(file);
+
+ /* Get ready for the next file. */
+ dest->file_seq++;
}
/*
@@ -3101,15 +3271,10 @@ cleanup_logical_decoding(LogicalDecodingContext *ctx)
dstate = (RepackDecodingState *) ctx->output_writer_private;
- ExecDropSingleTupleTableSlot(dstate->tsslot);
- FreeTupleDesc(dstate->tupdesc_change);
FreeTupleDesc(dstate->tupdesc);
- tuplestore_end(dstate->tstore);
-
FreeDecodingContext(ctx);
- ReplicationSlotRelease();
- ReplicationSlotDrop(NameStr(dstate->slotname), false);
+ ReplicationSlotDropAcquired();
pfree(dstate);
}
@@ -3123,7 +3288,6 @@ cleanup_logical_decoding(LogicalDecodingContext *ctx)
*/
static void
rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
- LogicalDecodingContext *decoding_ctx,
TransactionId frozenXid,
MultiXactId cutoffMulti)
{
@@ -3204,6 +3368,7 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
&chgdst.ident_index);
chgdst.ident_key = build_identity_key(ident_idx_new, OldHeap,
&chgdst.ident_key_nentries);
+ chgdst.file_seq = WORKER_FILE_SNAPSHOT + 1;
/*
* During testing, wait for another backend to perform concurrent data
@@ -3225,7 +3390,7 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
* hold AccessExclusiveLock. (Quite some amount of WAL could have been
* written during the data copying and index creation.)
*/
- process_concurrent_changes(decoding_ctx, end_of_wal, &chgdst);
+ process_concurrent_changes(end_of_wal, &chgdst, false);
/*
* Acquire AccessExclusiveLock on the table, its TOAST relation (if there
@@ -3306,8 +3471,11 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
XLogFlush(wal_insert_ptr);
end_of_wal = GetFlushRecPtr(NULL);
- /* Apply the concurrent changes again. */
- process_concurrent_changes(decoding_ctx, end_of_wal, &chgdst);
+ /*
+ * Apply the concurrent changes again. Indicate that the decoding worker
+ * won't be needed anymore.
+ */
+ process_concurrent_changes(end_of_wal, &chgdst, true);
/* Remember info about rel before closing OldHeap */
relpersistence = OldHeap->rd_rel->relpersistence;
@@ -3417,3 +3585,510 @@ build_new_indexes(Relation NewHeap, Relation OldHeap, List *OldIndexes)
return result;
}
+
+/*
+ * Try to start a background worker to perform logical decoding of data
+ * changes applied to relation while REPACK CONCURRENTLY is copying its
+ * contents to a new table.
+ */
+static void
+start_decoding_worker(Oid relid)
+{
+ Size size;
+ dsm_segment *seg;
+ DecodingWorkerShared *shared;
+ shm_mq *mq;
+ shm_mq_handle *mqh;
+ BackgroundWorker bgw;
+
+ /* Setup shared memory. */
+ size = BUFFERALIGN(offsetof(DecodingWorkerShared, error_queue)) +
+ BUFFERALIGN(REPACK_ERROR_QUEUE_SIZE);
+ seg = dsm_create(size, 0);
+ shared = (DecodingWorkerShared *) dsm_segment_address(seg);
+ shared->lsn_upto = InvalidXLogRecPtr;
+ shared->done = false;
+ SharedFileSetInit(&shared->sfs, seg);
+ shared->last_exported = -1;
+ SpinLockInit(&shared->mutex);
+ shared->dbid = MyDatabaseId;
+
+ /*
+ * This is the UserId set in cluster_rel(). Security context shouldn't be
+ * needed for decoding worker.
+ */
+ shared->roleid = GetUserId();
+ shared->relid = relid;
+ ConditionVariableInit(&shared->cv);
+ shared->backend_proc = MyProc;
+ shared->backend_pid = MyProcPid;
+ shared->backend_proc_number = MyProcNumber;
+
+ mq = shm_mq_create((char *) BUFFERALIGN(shared->error_queue),
+ REPACK_ERROR_QUEUE_SIZE);
+ shm_mq_set_receiver(mq, MyProc);
+ mqh = shm_mq_attach(mq, seg, NULL);
+
+ memset(&bgw, 0, sizeof(bgw));
+ snprintf(bgw.bgw_name, BGW_MAXLEN,
+ "REPACK decoding worker for relation \"%s\"",
+ get_rel_name(relid));
+ snprintf(bgw.bgw_type, BGW_MAXLEN, "REPACK decoding worker");
+ bgw.bgw_flags = BGWORKER_SHMEM_ACCESS |
+ BGWORKER_BACKEND_DATABASE_CONNECTION;
+ bgw.bgw_start_time = BgWorkerStart_RecoveryFinished;
+ bgw.bgw_restart_time = BGW_NEVER_RESTART;
+ snprintf(bgw.bgw_library_name, MAXPGPATH, "postgres");
+ snprintf(bgw.bgw_function_name, BGW_MAXLEN, "RepackWorkerMain");
+ bgw.bgw_main_arg = UInt32GetDatum(dsm_segment_handle(seg));
+ bgw.bgw_notify_pid = MyProcPid;
+
+ decoding_worker = palloc0_object(DecodingWorker);
+ if (!RegisterDynamicBackgroundWorker(&bgw, &decoding_worker->handle))
+ ereport(ERROR,
+ (errcode(ERRCODE_CONFIGURATION_LIMIT_EXCEEDED),
+ errmsg("out of background worker slots"),
+ errhint("You might need to increase \"%s\".", "max_worker_processes")));
+
+ decoding_worker->seg = seg;
+ decoding_worker->error_mqh = mqh;
+
+ /*
+ * The decoding setup must be done before the caller can have XID assigned
+ * for any reason, otherwise the worker might end up in a deadlock,
+ * waiting for the caller's transaction to end. Therefore wait here until
+ * the worker indicates that it has the logical decoding initialized.
+ */
+ ConditionVariablePrepareToSleep(&shared->cv);
+ for (;;)
+ {
+ bool initialized;
+
+ SpinLockAcquire(&shared->mutex);
+ initialized = shared->initialized;
+ SpinLockRelease(&shared->mutex);
+
+ if (initialized)
+ break;
+
+ ConditionVariableSleep(&shared->cv, WAIT_EVENT_REPACK_WORKER_EXPORT);
+ }
+ ConditionVariableCancelSleep();
+}
+
+/*
+ * Stop the decoding worker and cleanup the related resources.
+ *
+ * The worker stops on its own when it knows there is no more work to do, but
+ * we need to stop it explicitly at least on ERROR in the launching backend.
+ */
+static void
+stop_decoding_worker(void)
+{
+ BgwHandleStatus status;
+
+ /* Haven't reached the worker startup? */
+ if (decoding_worker == NULL)
+ return;
+
+ /* Could not register the worker? */
+ if (decoding_worker->handle == NULL)
+ return;
+
+ TerminateBackgroundWorker(decoding_worker->handle);
+ /* The worker should really exit before the REPACK command does. */
+ HOLD_INTERRUPTS();
+ status = WaitForBackgroundWorkerShutdown(decoding_worker->handle);
+ RESUME_INTERRUPTS();
+
+ if (status == BGWH_POSTMASTER_DIED)
+ ereport(FATAL,
+ (errcode(ERRCODE_ADMIN_SHUTDOWN),
+ errmsg("postmaster exited during REPACK command")));
+
+ shm_mq_detach(decoding_worker->error_mqh);
+
+ /*
+ * If we could not cancel the current sleep due to ERROR, do that before
+ * we detach from the shared memory the condition variable is located in.
+ * If we did not, the bgworker ERROR handling code would try and fail
+ * badly.
+ */
+ ConditionVariableCancelSleep();
+
+ dsm_detach(decoding_worker->seg);
+ pfree(decoding_worker);
+ decoding_worker = NULL;
+}
+
+/* Is this process a REPACK worker? */
+static bool is_repack_worker = false;
+
+static pid_t backend_pid;
+static ProcNumber backend_proc_number;
+
+/*
+ * See ParallelWorkerShutdown for details.
+ */
+static void
+RepackWorkerShutdown(int code, Datum arg)
+{
+ SendProcSignal(backend_pid,
+ PROCSIG_REPACK_MESSAGE,
+ backend_proc_number);
+
+ dsm_detach((dsm_segment *) DatumGetPointer(arg));
+}
+
+/* REPACK decoding worker entry point */
+void
+RepackWorkerMain(Datum main_arg)
+{
+ dsm_segment *seg;
+ DecodingWorkerShared *shared;
+ shm_mq *mq;
+ shm_mq_handle *mqh;
+
+ is_repack_worker = true;
+
+ /*
+ * Override the default bgworker_die() with die() so we can use
+ * CHECK_FOR_INTERRUPTS().
+ */
+ pqsignal(SIGTERM, die);
+ BackgroundWorkerUnblockSignals();
+
+ seg = dsm_attach(DatumGetUInt32(main_arg));
+ if (seg == NULL)
+ ereport(ERROR,
+ (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
+ errmsg("could not map dynamic shared memory segment")));
+
+ shared = (DecodingWorkerShared *) dsm_segment_address(seg);
+
+ /* Arrange to signal the leader if we exit. */
+ backend_pid = shared->backend_pid;
+ backend_proc_number = shared->backend_proc_number;
+ before_shmem_exit(RepackWorkerShutdown, PointerGetDatum(seg));
+
+ /*
+ * Join locking group - see the comments around the call of
+ * start_decoding_worker().
+ */
+ if (!BecomeLockGroupMember(shared->backend_proc, backend_pid))
+ /* The leader is not running anymore. */
+ return;
+
+ /*
+ * Setup a queue to send error messages to the backend that launched this
+ * worker.
+ */
+ mq = (shm_mq *) (char *) BUFFERALIGN(shared->error_queue);
+ shm_mq_set_sender(mq, MyProc);
+ mqh = shm_mq_attach(mq, seg, NULL);
+ pq_redirect_to_shm_mq(seg, mqh);
+ pq_set_parallel_leader(shared->backend_pid,
+ shared->backend_proc_number);
+
+ /* Connect to the database. */
+ BackgroundWorkerInitializeConnectionByOid(shared->dbid, shared->roleid, 0);
+
+ repack_worker_internal(seg);
+}
+
+static void
+repack_worker_internal(dsm_segment *seg)
+{
+ DecodingWorkerShared *shared;
+ LogicalDecodingContext *decoding_ctx;
+ SharedFileSet *sfs;
+ Snapshot snapshot;
+
+ /*
+ * Transaction is needed to open relation, and it also provides us with a
+ * resource owner.
+ */
+ StartTransactionCommand();
+
+ shared = (DecodingWorkerShared *) dsm_segment_address(seg);
+
+ /*
+ * Not sure the spinlock is needed here - the backend should not change
+ * anything in the shared memory until we have serialized the snapshot.
+ */
+ SpinLockAcquire(&shared->mutex);
+ Assert(XLogRecPtrIsInvalid(shared->lsn_upto));
+ sfs = &shared->sfs;
+ SpinLockRelease(&shared->mutex);
+
+ SharedFileSetAttach(sfs, seg);
+
+ /*
+ * Prepare to capture the concurrent data changes ourselves.
+ */
+ decoding_ctx = setup_logical_decoding(shared->relid);
+
+ /* Announce that we're ready. */
+ SpinLockAcquire(&shared->mutex);
+ shared->initialized = true;
+ SpinLockRelease(&shared->mutex);
+ ConditionVariableSignal(&shared->cv);
+
+ /* Build the initial snapshot and export it. */
+ snapshot = SnapBuildInitialSnapshotForRepack(decoding_ctx->snapshot_builder);
+ export_initial_snapshot(snapshot, shared);
+
+ /*
+ * Only historic snapshots should be used now. Do not let us restrict the
+ * progress of xmin horizon.
+ */
+ InvalidateCatalogSnapshot();
+
+ while (!decode_concurrent_changes(decoding_ctx, shared))
+ ;
+
+ /* Cleanup. */
+ cleanup_logical_decoding(decoding_ctx);
+ CommitTransactionCommand();
+}
+
+/*
+ * Make snapshot available to the backend that launched the decoding worker.
+ */
+static void
+export_initial_snapshot(Snapshot snapshot, DecodingWorkerShared *shared)
+{
+ char fname[MAXPGPATH];
+ BufFile *file;
+ Size snap_size;
+ char *snap_space;
+
+ snap_size = EstimateSnapshotSpace(snapshot);
+ snap_space = (char *) palloc(snap_size);
+ SerializeSnapshot(snapshot, snap_space);
+ FreeSnapshot(snapshot);
+
+ DecodingWorkerFileName(fname, shared->relid, shared->last_exported + 1);
+ file = BufFileCreateFileSet(&shared->sfs.fs, fname);
+ /* To make restoration easier, write the snapshot size first. */
+ BufFileWrite(file, &snap_size, sizeof(snap_size));
+ BufFileWrite(file, snap_space, snap_size);
+ pfree(snap_space);
+ BufFileClose(file);
+
+ /* Increase the counter to tell the backend that the file is available. */
+ SpinLockAcquire(&shared->mutex);
+ shared->last_exported++;
+ SpinLockRelease(&shared->mutex);
+ ConditionVariableSignal(&shared->cv);
+}
+
+/*
+ * Get the initial snapshot from the decoding worker.
+ */
+static Snapshot
+get_initial_snapshot(DecodingWorker *worker)
+{
+ DecodingWorkerShared *shared;
+ char fname[MAXPGPATH];
+ BufFile *file;
+ Size snap_size;
+ char *snap_space;
+ Snapshot snapshot;
+
+ shared = (DecodingWorkerShared *) dsm_segment_address(worker->seg);
+
+ /*
+ * The worker needs to initialize the logical decoding, which usually
+ * takes some time. Therefore it makes sense to prepare for the sleep
+ * first.
+ */
+ ConditionVariablePrepareToSleep(&shared->cv);
+ for (;;)
+ {
+ int last_exported;
+
+ SpinLockAcquire(&shared->mutex);
+ last_exported = shared->last_exported;
+ SpinLockRelease(&shared->mutex);
+
+ /*
+ * Has the worker exported the file we are waiting for?
+ */
+ if (last_exported == WORKER_FILE_SNAPSHOT)
+ break;
+
+ ConditionVariableSleep(&shared->cv, WAIT_EVENT_REPACK_WORKER_EXPORT);
+ }
+ ConditionVariableCancelSleep();
+
+ /* Read the snapshot from a file. */
+ DecodingWorkerFileName(fname, shared->relid, WORKER_FILE_SNAPSHOT);
+ file = BufFileOpenFileSet(&shared->sfs.fs, fname, O_RDONLY, false);
+ BufFileReadExact(file, &snap_size, sizeof(snap_size));
+ snap_space = (char *) palloc(snap_size);
+ BufFileReadExact(file, snap_space, snap_size);
+ BufFileClose(file);
+
+ /* Restore it. */
+ snapshot = RestoreSnapshot(snap_space);
+ pfree(snap_space);
+
+ return snapshot;
+}
+
+bool
+IsRepackWorker(void)
+{
+ return is_repack_worker;
+}
+
+/*
+ * Handle receipt of an interrupt indicating a repack worker message.
+ *
+ * Note: this is called within a signal handler! All we can do is set
+ * a flag that will cause the next CHECK_FOR_INTERRUPTS() to invoke
+ * ProcessRepackMessages().
+ */
+void
+HandleRepackMessageInterrupt(void)
+{
+ InterruptPending = true;
+ RepackMessagePending = true;
+ SetLatch(MyLatch);
+}
+
+/*
+ * Process any queued protocol messages received from parallel workers.
+ */
+void
+ProcessRepackMessages(void)
+{
+ MemoryContext oldcontext;
+
+ static MemoryContext hpm_context = NULL;
+
+ /*
+ * Nothing to do if we haven't launched the worker yet or have already
+ * terminated it.
+ */
+ if (decoding_worker == NULL)
+ return;
+
+ /*
+ * This is invoked from ProcessInterrupts(), and since some of the
+ * functions it calls contain CHECK_FOR_INTERRUPTS(), there is a potential
+ * for recursive calls if more signals are received while this runs. It's
+ * unclear that recursive entry would be safe, and it doesn't seem useful
+ * even if it is safe, so let's block interrupts until done.
+ */
+ HOLD_INTERRUPTS();
+
+ /*
+ * Moreover, CurrentMemoryContext might be pointing almost anywhere. We
+ * don't want to risk leaking data into long-lived contexts, so let's do
+ * our work here in a private context that we can reset on each use.
+ */
+ if (hpm_context == NULL) /* first time through? */
+ hpm_context = AllocSetContextCreate(TopMemoryContext,
+ "ProcessRepackMessages",
+ ALLOCSET_DEFAULT_SIZES);
+ else
+ MemoryContextReset(hpm_context);
+
+ oldcontext = MemoryContextSwitchTo(hpm_context);
+
+ /* OK to process messages. Reset the flag saying there are more to do. */
+ RepackMessagePending = false;
+
+ /*
+ * Read as many messages as we can from each worker, but stop when no more
+ * messages can be read from the worker without blocking.
+ */
+ while (true)
+ {
+ shm_mq_result res;
+ Size nbytes;
+ void *data;
+
+ res = shm_mq_receive(decoding_worker->error_mqh, &nbytes,
+ &data, true);
+ if (res == SHM_MQ_WOULD_BLOCK)
+ break;
+ else if (res == SHM_MQ_SUCCESS)
+ {
+ StringInfoData msg;
+
+ initStringInfo(&msg);
+ appendBinaryStringInfo(&msg, data, nbytes);
+ ProcessRepackMessage(&msg);
+ pfree(msg.data);
+ }
+ else
+ {
+ /*
+ * The decoding worker is special in that it exits as soon as it
+ * has its work done. Thus the DETACHED result code is fine.
+ */
+ Assert(res == SHM_MQ_DETACHED);
+
+ break;
+ }
+ }
+
+ MemoryContextSwitchTo(oldcontext);
+
+ /* Might as well clear the context on our way out */
+ MemoryContextReset(hpm_context);
+
+ RESUME_INTERRUPTS();
+}
+
+/*
+ * Process a single protocol message received from a single parallel worker.
+ */
+static void
+ProcessRepackMessage(StringInfo msg)
+{
+ char msgtype;
+
+ msgtype = pq_getmsgbyte(msg);
+
+ switch (msgtype)
+ {
+ case PqMsg_ErrorResponse:
+ case PqMsg_NoticeResponse:
+ {
+ ErrorData edata;
+
+ /* Parse ErrorResponse or NoticeResponse. */
+ pq_parse_errornotice(msg, &edata);
+
+ /* Death of a worker isn't enough justification for suicide. */
+ edata.elevel = Min(edata.elevel, ERROR);
+
+ /*
+ * If desired, add a context line to show that this is a
+ * message propagated from a parallel worker. Otherwise, it
+ * can sometimes be confusing to understand what actually
+ * happened.
+ */
+ if (edata.context)
+ edata.context = psprintf("%s\n%s", edata.context,
+ _("decoding worker"));
+ else
+ edata.context = pstrdup(_("decoding worker"));
+
+ /* Rethrow error or print notice. */
+ ThrowErrorData(&edata);
+
+ break;
+ }
+
+ default:
+ {
+ elog(ERROR, "unrecognized message type received from decoding worker: %c (message length %d bytes)",
+ msgtype, msg->len);
+ }
+ }
+}
diff --git a/src/backend/libpq/pqmq.c b/src/backend/libpq/pqmq.c
index 6e4bbfb5aa1..42f6fa472c5 100644
--- a/src/backend/libpq/pqmq.c
+++ b/src/backend/libpq/pqmq.c
@@ -14,6 +14,7 @@
#include "postgres.h"
#include "access/parallel.h"
+#include "commands/cluster.h"
#include "libpq/libpq.h"
#include "libpq/pqformat.h"
#include "libpq/pqmq.h"
@@ -175,6 +176,10 @@ mq_putmessage(char msgtype, const char *s, size_t len)
SendProcSignal(pq_mq_parallel_leader_pid,
PROCSIG_PARALLEL_APPLY_MESSAGE,
pq_mq_parallel_leader_proc_number);
+ else if (IsRepackWorker())
+ SendProcSignal(pq_mq_parallel_leader_pid,
+ PROCSIG_REPACK_MESSAGE,
+ pq_mq_parallel_leader_proc_number);
else
{
Assert(IsParallelWorker());
diff --git a/src/backend/postmaster/bgworker.c b/src/backend/postmaster/bgworker.c
index 65deabe91a7..334bb708c5b 100644
--- a/src/backend/postmaster/bgworker.c
+++ b/src/backend/postmaster/bgworker.c
@@ -13,6 +13,7 @@
#include "postgres.h"
#include "access/parallel.h"
+#include "commands/cluster.h"
#include "libpq/pqsignal.h"
#include "miscadmin.h"
#include "pgstat.h"
@@ -136,6 +137,9 @@ static const struct
},
{
"SequenceSyncWorkerMain", SequenceSyncWorkerMain
+ },
+ {
+ "RepackWorkerMain", RepackWorkerMain
}
};
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index b0ef1a12520..35a46988285 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -194,7 +194,11 @@ StartupDecodingContext(List *output_plugin_options,
ctx->slot = slot;
- ctx->reader = XLogReaderAllocate(wal_segment_size, NULL, xl_routine, ctx);
+ /*
+ * TODO A separate patch for PG core, unless there's really a reason to
+ * pass ctx for private_data (May extensions expect ctx?).
+ */
+ ctx->reader = XLogReaderAllocate(wal_segment_size, NULL, xl_routine, NULL);
if (!ctx->reader)
ereport(ERROR,
(errcode(ERRCODE_OUT_OF_MEMORY),
diff --git a/src/backend/replication/pgoutput_repack/pgoutput_repack.c b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
index a074eed393d..4bc47a72371 100644
--- a/src/backend/replication/pgoutput_repack/pgoutput_repack.c
+++ b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
@@ -169,17 +169,13 @@ store_change(LogicalDecodingContext *ctx, ConcurrentChangeKind kind,
HeapTuple tuple)
{
RepackDecodingState *dstate;
- char *change_raw;
- ConcurrentChange change;
+ char kind_byte = (char) kind;
bool flattened = false;
- Size size;
- Datum values[1];
- bool isnull[1];
- char *dst;
dstate = (RepackDecodingState *) ctx->output_writer_private;
- size = VARHDRSZ + SizeOfConcurrentChange;
+ /* Store the change kind. */
+ BufFileWrite(dstate->file, &kind_byte, 1);
/*
* ReorderBufferCommit() stores the TOAST chunks in its private memory
@@ -196,46 +192,12 @@ store_change(LogicalDecodingContext *ctx, ConcurrentChangeKind kind,
tuple = toast_flatten_tuple(tuple, dstate->tupdesc);
flattened = true;
}
+ /* Store the tuple size ... */
+ BufFileWrite(dstate->file, &tuple->t_len, sizeof(tuple->t_len));
+ /* ... and the tuple itself. */
+ BufFileWrite(dstate->file, tuple->t_data, tuple->t_len);
- size += tuple->t_len;
- if (size >= MaxAllocSize)
- elog(ERROR, "Change is too big.");
-
- /* Construct the change. */
- change_raw = (char *) palloc0(size);
- SET_VARSIZE(change_raw, size);
-
- /*
- * Since the varlena alignment might not be sufficient for the structure,
- * set the fields in a local instance and remember where it should
- * eventually be copied.
- */
- change.kind = kind;
- dst = (char *) VARDATA(change_raw);
-
- /*
- * Copy the tuple.
- *
- * Note: change->tup_data.t_data must be fixed on retrieval!
- */
- memcpy(&change.tup_data, tuple, sizeof(HeapTupleData));
- memcpy(dst, &change, SizeOfConcurrentChange);
- dst += SizeOfConcurrentChange;
- memcpy(dst, tuple->t_data, tuple->t_len);
-
- /* The data has been copied. */
+ /* Free the flat copy if created above. */
if (flattened)
pfree(tuple);
-
- /* Store as tuple of 1 bytea column. */
- values[0] = PointerGetDatum(change_raw);
- isnull[0] = false;
- tuplestore_putvalues(dstate->tstore, dstate->tupdesc_change,
- values, isnull);
-
- /* Accounting. */
- dstate->nchanges++;
-
- /* Cleanup. */
- pfree(change_raw);
}
diff --git a/src/backend/storage/ipc/procsignal.c b/src/backend/storage/ipc/procsignal.c
index 8e56922dcea..6f9e7a7aab7 100644
--- a/src/backend/storage/ipc/procsignal.c
+++ b/src/backend/storage/ipc/procsignal.c
@@ -19,6 +19,7 @@
#include "access/parallel.h"
#include "commands/async.h"
+#include "commands/cluster.h"
#include "miscadmin.h"
#include "pgstat.h"
#include "port/pg_bitutils.h"
@@ -697,6 +698,9 @@ procsignal_sigusr1_handler(SIGNAL_ARGS)
if (CheckProcSignal(PROCSIG_PARALLEL_APPLY_MESSAGE))
HandleParallelApplyMessageInterrupt();
+ if (CheckProcSignal(PROCSIG_REPACK_MESSAGE))
+ HandleRepackMessageInterrupt();
+
if (CheckProcSignal(PROCSIG_RECOVERY_CONFLICT_DATABASE))
HandleRecoveryConflictInterrupt(PROCSIG_RECOVERY_CONFLICT_DATABASE);
diff --git a/src/backend/tcop/postgres.c b/src/backend/tcop/postgres.c
index e54bf1e760f..fc81ad87615 100644
--- a/src/backend/tcop/postgres.c
+++ b/src/backend/tcop/postgres.c
@@ -36,6 +36,7 @@
#include "access/xact.h"
#include "catalog/pg_type.h"
#include "commands/async.h"
+#include "commands/cluster.h"
#include "commands/event_trigger.h"
#include "commands/explain_state.h"
#include "commands/prepare.h"
@@ -3541,6 +3542,9 @@ ProcessInterrupts(void)
if (ParallelApplyMessagePending)
ProcessParallelApplyMessages();
+
+ if (RepackMessagePending)
+ ProcessRepackMessages();
}
/*
diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt
index 5537a2d2530..56d01ccd5af 100644
--- a/src/backend/utils/activity/wait_event_names.txt
+++ b/src/backend/utils/activity/wait_event_names.txt
@@ -154,6 +154,7 @@ RECOVERY_CONFLICT_SNAPSHOT "Waiting for recovery conflict resolution for a vacuu
RECOVERY_CONFLICT_TABLESPACE "Waiting for recovery conflict resolution for dropping a tablespace."
RECOVERY_END_COMMAND "Waiting for <xref linkend="guc-recovery-end-command"/> to complete."
RECOVERY_PAUSE "Waiting for recovery to be resumed."
+REPACK_WORKER_EXPORT "Waiting for decoding worker to export a new output file."
REPLICATION_ORIGIN_DROP "Waiting for a replication origin to become inactive so it can be dropped."
REPLICATION_SLOT_DROP "Waiting for a replication slot to become inactive so it can be dropped."
RESTORE_COMMAND "Waiting for <xref linkend="guc-restore-command"/> to complete."
diff --git a/src/include/access/tableam.h b/src/include/access/tableam.h
index 76aa993009a..15760363a1a 100644
--- a/src/include/access/tableam.h
+++ b/src/include/access/tableam.h
@@ -22,7 +22,6 @@
#include "access/xact.h"
#include "commands/vacuum.h"
#include "executor/tuptable.h"
-#include "replication/logical.h"
#include "storage/read_stream.h"
#include "utils/rel.h"
#include "utils/snapshot.h"
@@ -631,7 +630,6 @@ typedef struct TableAmRoutine
bool use_sort,
TransactionId OldestXmin,
Snapshot snapshot,
- LogicalDecodingContext *decoding_ctx,
TransactionId *xid_cutoff,
MultiXactId *multi_cutoff,
double *num_tuples,
@@ -1651,8 +1649,6 @@ table_relation_copy_data(Relation rel, const RelFileLocator *newrlocator)
* - *multi_cutoff - ditto
* - snapshot - if != NULL, ignore data changes done by transactions that this
* (MVCC) snapshot considers still in-progress or in the future.
- * - decoding_ctx - logical decoding context, to capture concurrent data
- * changes.
*
* Output parameters:
* - *xid_cutoff - rel's new relfrozenxid value, may be invalid
@@ -1666,7 +1662,6 @@ table_relation_copy_for_cluster(Relation OldTable, Relation NewTable,
bool use_sort,
TransactionId OldestXmin,
Snapshot snapshot,
- LogicalDecodingContext *decoding_ctx,
TransactionId *xid_cutoff,
MultiXactId *multi_cutoff,
double *num_tuples,
@@ -1675,7 +1670,7 @@ table_relation_copy_for_cluster(Relation OldTable, Relation NewTable,
{
OldTable->rd_tableam->relation_copy_for_cluster(OldTable, NewTable, OldIndex,
use_sort, OldestXmin,
- snapshot, decoding_ctx,
+ snapshot,
xid_cutoff, multi_cutoff,
num_tuples, tups_vacuumed,
tups_recently_dead);
diff --git a/src/include/commands/cluster.h b/src/include/commands/cluster.h
index 6a5c476294a..1b05d5d418b 100644
--- a/src/include/commands/cluster.h
+++ b/src/include/commands/cluster.h
@@ -17,11 +17,13 @@
#include "nodes/parsenodes.h"
#include "parser/parse_node.h"
#include "replication/decode.h"
+#include "postmaster/bgworker.h"
#include "replication/logical.h"
+#include "storage/buffile.h"
#include "storage/lock.h"
+#include "storage/shm_mq.h"
#include "utils/relcache.h"
#include "utils/resowner.h"
-#include "utils/tuplestore.h"
/* flag bits for ClusterParams->options */
@@ -44,6 +46,9 @@ typedef struct ClusterParams
* The following definitions are used by REPACK CONCURRENTLY.
*/
+/*
+ * Stored as a single byte in the output file.
+ */
typedef enum
{
CHANGE_INSERT,
@@ -52,68 +57,30 @@ typedef enum
CHANGE_DELETE
} ConcurrentChangeKind;
-typedef struct ConcurrentChange
-{
- /* See the enum above. */
- ConcurrentChangeKind kind;
-
- /*
- * The actual tuple.
- *
- * The tuple data follows the ConcurrentChange structure. Before use make
- * sure the tuple is correctly aligned (ConcurrentChange can be stored as
- * bytea) and that tuple->t_data is fixed.
- */
- HeapTupleData tup_data;
-} ConcurrentChange;
-
-#define SizeOfConcurrentChange (offsetof(ConcurrentChange, tup_data) + \
- sizeof(HeapTupleData))
-
/*
* Logical decoding state.
*
- * Here we store the data changes that we decode from WAL while the table
- * contents is being copied to a new storage. Also the necessary metadata
- * needed to apply these changes to the table is stored here.
+ * The output plugin uses it to store the data changes that it decodes from
+ * WAL while the table contents is being copied to a new storage.
*/
typedef struct RepackDecodingState
{
/* The relation whose changes we're decoding. */
Oid relid;
- /* Replication slot name. */
- NameData slotname;
-
- /*
- * Decoded changes are stored here. Although we try to avoid excessive
- * batches, it can happen that the changes need to be stored to disk. The
- * tuplestore does this transparently.
- */
- Tuplestorestate *tstore;
-
- /* The current number of changes in tstore. */
- double nchanges;
-
- /*
- * Descriptor to store the ConcurrentChange structure serialized (bytea).
- * We can't store the tuple directly because tuplestore only supports
- * minimum tuple and we may need to transfer OID system column from the
- * output plugin. Also we need to transfer the change kind, so it's better
- * to put everything in the structure than to use 2 tuplestores "in
- * parallel".
- */
- TupleDesc tupdesc_change;
-
- /* Tuple descriptor needed to update indexes. */
+ /* Tuple descriptor of the relation being processed. */
TupleDesc tupdesc;
- /* Slot to retrieve data from tstore. */
- TupleTableSlot *tsslot;
-
- ResourceOwner resowner;
+ /* The current output file. */
+ BufFile *file;
} RepackDecodingState;
+extern PGDLLIMPORT volatile sig_atomic_t RepackMessagePending;
+
+extern bool IsRepackWorker(void);
+extern void HandleRepackMessageInterrupt(void);
+extern void ProcessRepackMessages(void);
+
extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel);
extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid,
@@ -136,6 +103,6 @@ extern void finish_heap_swap(Oid OIDOldHeap, Oid OIDNewHeap,
extern bool am_decoding_for_repack(void);
extern bool change_useless_for_repack(XLogRecordBuffer *buf);
-extern void repack_decode_concurrent_changes(LogicalDecodingContext *ctx,
- XLogRecPtr end_of_wal);
+
+extern void RepackWorkerMain(Datum main_arg);
#endif /* CLUSTER_H */
diff --git a/src/include/storage/procsignal.h b/src/include/storage/procsignal.h
index e52b8eb7697..3ef35ca6b80 100644
--- a/src/include/storage/procsignal.h
+++ b/src/include/storage/procsignal.h
@@ -36,6 +36,7 @@ typedef enum
PROCSIG_BARRIER, /* global barrier interrupt */
PROCSIG_LOG_MEMORY_CONTEXT, /* ask backend to log the memory contexts */
PROCSIG_PARALLEL_APPLY_MESSAGE, /* Message from parallel apply workers */
+ PROCSIG_REPACK_MESSAGE, /* Message from repack worker */
/* Recovery conflict reasons */
PROCSIG_RECOVERY_CONFLICT_FIRST,
diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list
index 1aa4fb442f1..1e2e4ba80e9 100644
--- a/src/tools/pgindent/typedefs.list
+++ b/src/tools/pgindent/typedefs.list
@@ -498,7 +498,6 @@ CompressFileHandle
CompressionLocation
CompressorState
ComputeXidHorizonsResult
-ConcurrentChange
ConcurrentChangeKind
ConditionVariable
ConditionVariableMinimallyPadded
@@ -638,6 +637,8 @@ DeclareCursorStmt
DecodedBkpBlock
DecodedXLogRecord
DecodingOutputState
+DecodingWorker
+DecodingWorkerShared
DefElem
DefElemAction
DefaultACLInfo
--
2.47.3
--=-=-=
Content-Type: text/plain
Content-Disposition: attachment;
filename=v32-0006-Use-multiple-snapshots-to-copy-the-data.patch
^ permalink raw reply [nested|flat] 55+ messages in thread
* [PATCH v35 5/5] Use background worker to do logical decoding.
@ 2026-02-16 19:49 Antonin Houska <[email protected]>
0 siblings, 0 replies; 55+ messages in thread
From: Antonin Houska @ 2026-02-16 19:49 UTC (permalink / raw)
If the backend performing REPACK (CONCURRENTLY) does both data copying and
logical decoding, it has to "travel in time" back and forth and therefore it
has to invalidate system caches quite a few times. (The copying and the
decoding work with different catalog snapshots.) As the decoding worker has
separate caches, the switching is not necessary.
Without the worker, it'd also be difficult to switch between potentially long
running tasks like index build and WAL decoding. (No decoding during that time
at all can suspend archiving / recycling of WAL segments for some time, which
in turn may result in full disk.)
Another problem is that, after having acquired AccessExclusiveLock (in order
to swap the files), the backend needs to both decode and apply the data
changes that took place while it was waiting for the lock. With the decoding
worker, the decoding runs all the time, so the backend only needs to apply the
changes. This can reduce the time the exclusive lock is held for.
Note that the code added in order to handle ERRORs in the background worker
almost duplicates the existing code that does the same for other types of
workers (See ProcessParallelMessages() and
ProcessParallelApplyMessages()). Refactoring of the existing code might be
useful, to reduce the duplication.
---
src/backend/access/heap/heapam_handler.c | 44 -
src/backend/commands/cluster.c | 1179 +++++++++++++----
src/backend/libpq/pqmq.c | 5 +
src/backend/postmaster/bgworker.c | 5 +
src/backend/replication/logical/logical.c | 6 +-
.../pgoutput_repack/pgoutput_repack.c | 54 +-
src/backend/storage/ipc/procsignal.c | 4 +
src/backend/tcop/postgres.c | 4 +
.../utils/activity/wait_event_names.txt | 1 +
src/include/access/tableam.h | 7 +-
src/include/commands/cluster.h | 71 +-
src/include/storage/procsignal.h | 1 +
src/tools/pgindent/typedefs.list | 3 +-
13 files changed, 981 insertions(+), 403 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index 41c52046e14..4c503a63832 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -33,7 +33,6 @@
#include "catalog/index.h"
#include "catalog/storage.h"
#include "catalog/storage_xlog.h"
-#include "commands/cluster.h"
#include "commands/progress.h"
#include "executor/executor.h"
#include "miscadmin.h"
@@ -688,7 +687,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
Relation OldIndex, bool use_sort,
TransactionId OldestXmin,
Snapshot snapshot,
- LogicalDecodingContext *decoding_ctx,
TransactionId *xid_cutoff,
MultiXactId *multi_cutoff,
double *num_tuples,
@@ -710,7 +708,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
BufferHeapTupleTableSlot *hslot;
BlockNumber prev_cblock = InvalidBlockNumber;
bool concurrent = snapshot != NULL;
- XLogRecPtr end_of_wal_prev = GetFlushRecPtr(NULL);
/* Remember if it's a system catalog */
is_system_catalog = IsSystemRelation(OldHeap);
@@ -972,31 +969,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
ct_val[1] = *num_tuples;
pgstat_progress_update_multi_param(2, ct_index, ct_val);
}
-
- /*
- * Process the WAL produced by the load, as well as by other
- * transactions, so that the replication slot can advance and WAL does
- * not pile up. Use wal_segment_size as a threshold so that we do not
- * introduce the decoding overhead too often.
- *
- * Of course, we must not apply the changes until the initial load has
- * completed.
- *
- * Note that our insertions into the new table should not be decoded
- * as we (intentionally) do not write the logical decoding specific
- * information to WAL.
- */
- if (concurrent)
- {
- XLogRecPtr end_of_wal;
-
- end_of_wal = GetFlushRecPtr(NULL);
- if ((end_of_wal - end_of_wal_prev) > wal_segment_size)
- {
- repack_decode_concurrent_changes(decoding_ctx, end_of_wal);
- end_of_wal_prev = end_of_wal;
- }
- }
}
if (indexScan != NULL)
@@ -1042,22 +1014,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
-
- /*
- * Try to keep the amount of not-yet-decoded WAL small, like
- * above.
- */
- if (concurrent)
- {
- XLogRecPtr end_of_wal;
-
- end_of_wal = GetFlushRecPtr(NULL);
- if ((end_of_wal - end_of_wal_prev) > wal_segment_size)
- {
- repack_decode_concurrent_changes(decoding_ctx, end_of_wal);
- end_of_wal_prev = end_of_wal;
- }
- }
}
tuplesort_end(tuplesort);
diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c
index 6111f7985ab..86c0d44f2d1 100644
--- a/src/backend/commands/cluster.c
+++ b/src/backend/commands/cluster.c
@@ -12,12 +12,13 @@
* In concurrent mode, we lock the table with only ShareUpdateExclusiveLock,
* then do an initial copy as above. However, while the tuples are being
* copied, concurrent transactions could modify the table. To cope with those
- * changes, we rely on logical decoding to obtain them from WAL. The changes
- * are accumulated in a tuplestore. Once the initial copy is complete, we
- * read the changes from the tuplestore and re-apply them on the new heap.
- * Then we upgrade our ShareUpdateExclusiveLock to AccessExclusiveLock and
- * swap the relfilenodes. This way, the time we hold a strong lock on the
- * table is much reduced, and the bloat is eliminated.
+ * changes, we rely on logical decoding to obtain them from WAL. A bgworker
+ * consumes WAL while the initial copy is ongoing (to prevent excessive WAL
+ * from being reserved), and accumulates the changes in a file. Once the
+ * initial copy is complete, we read the changes from the file and re-apply
+ * them on the new heap. Then we upgrade our ShareUpdateExclusiveLock to
+ * AccessExclusiveLock and swap the relfilenodes. This way, the time we hold
+ * a strong lock on the table is much reduced, and the bloat is eliminated.
*
* There is hardly anything left of Paul Brown's original implementation...
*
@@ -45,6 +46,7 @@
#include "access/xlog_internal.h"
#include "access/xloginsert.h"
#include "access/xlogutils.h"
+#include "access/xlogwait.h"
#include "catalog/catalog.h"
#include "catalog/dependency.h"
#include "catalog/heap.h"
@@ -61,6 +63,8 @@
#include "commands/tablecmds.h"
#include "commands/vacuum.h"
#include "executor/executor.h"
+#include "libpq/pqformat.h"
+#include "libpq/pqmq.h"
#include "miscadmin.h"
#include "optimizer/optimizer.h"
#include "pgstat.h"
@@ -71,6 +75,8 @@
#include "storage/ipc.h"
#include "storage/lmgr.h"
#include "storage/predicate.h"
+#include "storage/procsignal.h"
+#include "tcop/tcopprot.h"
#include "utils/acl.h"
#include "utils/fmgroids.h"
#include "utils/guc.h"
@@ -117,6 +123,12 @@ typedef struct IndexInsertState
/* The WAL segment being decoded. */
static XLogSegNo repack_current_segment = 0;
+/*
+ * The first file exported by the decoding worker must contain a snapshot, the
+ * following ones contain the data changes.
+ */
+#define WORKER_FILE_SNAPSHOT 0
+
/*
* Information needed to apply concurrent data changes.
*/
@@ -136,8 +148,110 @@ typedef struct ChangeDest
/* Needed to update indexes of rel_dst. */
IndexInsertState *iistate;
+
+ /*
+ * Sequential number of the file containing the changes.
+ *
+ * TODO This field makes the structure name less descriptive. Should we
+ * rename it, e.g. to ChangeApplyInfo?
+ */
+ int file_seq;
} ChangeDest;
+/*
+ * Layout of shared memory used for communication between backend and the
+ * worker that performs logical decoding of data changes
+ */
+typedef struct DecodingWorkerShared
+{
+ /* Is the decoding initialized? */
+ bool initialized;
+
+ /*
+ * Once the worker has reached this LSN, it should close the current
+ * output file and either create a new one or exit, according to the field
+ * 'done'. If the value is InvalidXLogRecPtr, the worker should decode all
+ * the WAL available and keep checking this field. It is ok if the worker
+ * had already decoded records whose LSN is >= lsn_upto before this field
+ * has been set.
+ */
+ XLogRecPtr lsn_upto;
+
+ /* Exit after closing the current file? */
+ bool done;
+
+ /* The output is stored here. */
+ SharedFileSet sfs;
+
+ /* Number of the last file exported by the worker. */
+ int last_exported;
+
+ /* Synchronize access to the fields above. */
+ slock_t mutex;
+
+ /* Database to connect to. */
+ Oid dbid;
+
+ /* Role to connect as. */
+ Oid roleid;
+
+ /* Decode data changes of this relation. */
+ Oid relid;
+
+ /* The backend uses this to wait for the worker. */
+ ConditionVariable cv;
+
+ /* Info to signal the backend. */
+ PGPROC *backend_proc;
+ pid_t backend_pid;
+ ProcNumber backend_proc_number;
+
+ /*
+ * Memory the queue is located in.
+ *
+ * For considerations on the value see the comments of
+ * PARALLEL_ERROR_QUEUE_SIZE.
+ */
+#define REPACK_ERROR_QUEUE_SIZE 16384
+ char error_queue[FLEXIBLE_ARRAY_MEMBER];
+} DecodingWorkerShared;
+
+/*
+ * Generate worker's output file name. If relations of the same 'relid' happen
+ * to be processed at the same time, they must be from different databases and
+ * therefore different backends must be involved. (PID is already present in
+ * the fileset name.)
+ */
+static inline void
+DecodingWorkerFileName(char *fname, Oid relid, uint32 seq)
+{
+ snprintf(fname, MAXPGPATH, "%u-%u", relid, seq);
+}
+
+/*
+ * Backend-local information to control the decoding worker.
+ */
+typedef struct DecodingWorker
+{
+ /* The worker. */
+ BackgroundWorkerHandle *handle;
+
+ /* DecodingWorkerShared is in this segment. */
+ dsm_segment *seg;
+
+ /* Handle of the error queue. */
+ shm_mq_handle *error_mqh;
+} DecodingWorker;
+
+/* Pointer to currently running decoding worker. */
+static DecodingWorker *decoding_worker = NULL;
+
+/*
+ * Is there a message sent by a repack worker that the backend needs to
+ * receive?
+ */
+volatile sig_atomic_t RepackMessagePending = false;
+
static bool cluster_rel_recheck(RepackCommand cmd, Relation OldHeap,
Oid indexOid, Oid userid, LOCKMODE lmode,
int options);
@@ -145,7 +259,7 @@ static void check_repack_concurrently_requirements(Relation rel);
static void rebuild_relation(Relation OldHeap, Relation index, bool verbose,
bool concurrent);
static void copy_table_data(Relation NewHeap, Relation OldHeap, Relation OldIndex,
- Snapshot snapshot, LogicalDecodingContext *decoding_ctx,
+ Snapshot snapshot,
bool verbose,
bool *pSwapToastByContent,
TransactionId *pFreezeXid,
@@ -158,12 +272,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd,
static bool cluster_is_permitted_for_relation(RepackCommand cmd,
Oid relid, Oid userid);
-static void begin_concurrent_repack(Relation rel);
-static void end_concurrent_repack(void);
static LogicalDecodingContext *setup_logical_decoding(Oid relid);
-static HeapTuple get_changed_tuple(char *change);
-static void apply_concurrent_changes(RepackDecodingState *dstate,
- ChangeDest *dest);
+static bool decode_concurrent_changes(LogicalDecodingContext *ctx,
+ DecodingWorkerShared *shared);
+static void apply_concurrent_changes(BufFile *file, ChangeDest *dest);
static void apply_concurrent_insert(Relation rel, HeapTuple tup,
IndexInsertState *iistate,
TupleTableSlot *index_slot);
@@ -175,9 +287,9 @@ static void apply_concurrent_delete(Relation rel, HeapTuple tup_target);
static HeapTuple find_target_tuple(Relation rel, ChangeDest *dest,
HeapTuple tup_key,
TupleTableSlot *ident_slot);
-static void process_concurrent_changes(LogicalDecodingContext *decoding_ctx,
- XLogRecPtr end_of_wal,
- ChangeDest *dest);
+static void process_concurrent_changes(XLogRecPtr end_of_wal,
+ ChangeDest *dest,
+ bool done);
static IndexInsertState *get_index_insert_state(Relation relation,
Oid ident_index_id,
Relation *ident_index_p);
@@ -186,7 +298,6 @@ static ScanKey build_identity_key(Oid ident_idx_oid, Relation rel_src,
static void free_index_insert_state(IndexInsertState *iistate);
static void cleanup_logical_decoding(LogicalDecodingContext *ctx);
static void rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
- LogicalDecodingContext *decoding_ctx,
TransactionId frozenXid,
MultiXactId cutoffMulti);
static List *build_new_indexes(Relation NewHeap, Relation OldHeap, List *OldIndexes);
@@ -196,6 +307,13 @@ static Relation process_single_relation(RepackStmt *stmt,
ClusterParams *params);
static Oid determine_clustered_index(Relation rel, bool usingindex,
const char *indexname);
+static void start_decoding_worker(Oid relid);
+static void stop_decoding_worker(void);
+static void repack_worker_internal(dsm_segment *seg);
+static void export_initial_snapshot(Snapshot snapshot,
+ DecodingWorkerShared *shared);
+static Snapshot get_initial_snapshot(DecodingWorker *worker);
+static void ProcessRepackMessage(StringInfo msg);
static const char *RepackCommandAsString(RepackCommand cmd);
@@ -619,20 +737,20 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid,
/* rebuild_relation does all the dirty work */
PG_TRY();
{
- /*
- * For concurrent processing, make sure that our logical decoding
- * ignores data changes of other tables than the one we are
- * processing.
- */
- if (concurrent)
- begin_concurrent_repack(OldHeap);
-
rebuild_relation(OldHeap, index, verbose, concurrent);
}
PG_FINALLY();
{
if (concurrent)
- end_concurrent_repack();
+ {
+ /*
+ * Since during normal operation the worker was already asked to
+ * exit, stopping it explicitly is especially important on ERROR.
+ * However it still seems a good practice to make sure that the
+ * worker never survives the REPACK command.
+ */
+ stop_decoding_worker();
+ }
}
PG_END_TRY();
@@ -929,7 +1047,6 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
bool swap_toast_by_content;
TransactionId frozenXid;
MultiXactId cutoffMulti;
- LogicalDecodingContext *decoding_ctx = NULL;
Snapshot snapshot = NULL;
#if USE_ASSERT_CHECKING
LOCKMODE lmode;
@@ -943,19 +1060,36 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
if (concurrent)
{
/*
- * Prepare to capture the concurrent data changes.
+ * The worker needs to be member of the locking group we're the leader
+ * of. We ought to become the leader before the worker starts. The
+ * worker will join the group as soon as it starts.
*
- * Note that this call waits for all transactions with XID already
- * assigned to finish. If some of those transactions is waiting for a
- * lock conflicting with ShareUpdateExclusiveLock on our table (e.g.
- * it runs CREATE INDEX), we can end up in a deadlock. Not sure this
- * risk is worth unlocking/locking the table (and its clustering
- * index) and checking again if its still eligible for REPACK
- * CONCURRENTLY.
+ * This is to make sure that the deadlock described below is
+ * detectable by deadlock.c: if the worker waits for a transaction to
+ * complete and we are waiting for the worker output, then effectively
+ * we (i.e. this backend) are waiting for that transaction.
*/
- decoding_ctx = setup_logical_decoding(tableOid);
+ BecomeLockGroupLeader();
+
+ /*
+ * Start the worker that decodes data changes applied while we're
+ * copying the table contents.
+ *
+ * Note that the worker has to wait for all transactions with XID
+ * already assigned to finish. If some of those transactions is
+ * waiting for a lock conflicting with ShareUpdateExclusiveLock on our
+ * table (e.g. it runs CREATE INDEX), we can end up in a deadlock.
+ * Not sure this risk is worth unlocking/locking the table (and its
+ * clustering index) and checking again if it's still eligible for
+ * REPACK CONCURRENTLY.
+ */
+ start_decoding_worker(tableOid);
+
+ /*
+ * Wait until the worker has the initial snapshot and retrieve it.
+ */
+ snapshot = get_initial_snapshot(decoding_worker);
- snapshot = SnapBuildInitialSnapshotForRepack(decoding_ctx->snapshot_builder);
PushActiveSnapshot(snapshot);
}
@@ -980,7 +1114,7 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
NewHeap = table_open(OIDNewHeap, NoLock);
/* Copy the heap data into the new table in the desired order */
- copy_table_data(NewHeap, OldHeap, index, snapshot, decoding_ctx, verbose,
+ copy_table_data(NewHeap, OldHeap, index, snapshot, verbose,
&swap_toast_by_content, &frozenXid, &cutoffMulti);
/* The historic snapshot won't be needed anymore. */
@@ -1001,14 +1135,11 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
if (index)
index_close(index, NoLock);
- rebuild_relation_finish_concurrent(NewHeap, OldHeap, decoding_ctx,
- frozenXid, cutoffMulti);
+ rebuild_relation_finish_concurrent(NewHeap, OldHeap, frozenXid,
+ cutoffMulti);
pgstat_progress_update_param(PROGRESS_REPACK_PHASE,
PROGRESS_REPACK_PHASE_FINAL_CLEANUP);
-
- /* Done with decoding. */
- cleanup_logical_decoding(decoding_ctx);
}
else
{
@@ -1180,8 +1311,7 @@ make_new_heap(Oid OIDOldHeap, Oid NewTableSpace, Oid NewAccessMethod,
*/
static void
copy_table_data(Relation NewHeap, Relation OldHeap, Relation OldIndex,
- Snapshot snapshot, LogicalDecodingContext *decoding_ctx,
- bool verbose, bool *pSwapToastByContent,
+ Snapshot snapshot, bool verbose, bool *pSwapToastByContent,
TransactionId *pFreezeXid, MultiXactId *pCutoffMulti)
{
Relation relRelation;
@@ -1342,7 +1472,6 @@ copy_table_data(Relation NewHeap, Relation OldHeap, Relation OldIndex,
*/
table_relation_copy_for_cluster(OldHeap, NewHeap, OldIndex, use_sort,
cutoffs.OldestXmin, snapshot,
- decoding_ctx,
&cutoffs.FreezeLimit,
&cutoffs.MultiXactCutoff,
&num_tuples, &tups_vacuumed,
@@ -2376,58 +2505,6 @@ RepackCommandAsString(RepackCommand cmd)
}
-/*
- * Call this function before REPACK CONCURRENTLY starts to setup logical
- * decoding. It makes sure that other users of the table put enough
- * information into WAL.
- *
- * The point is that at various places we expect that the table we're
- * processing is treated like a system catalog. For example, we need to be
- * able to scan it using a "historic snapshot" anytime during the processing
- * (as opposed to scanning only at the start point of the decoding, as logical
- * replication does during initial table synchronization), in order to apply
- * concurrent UPDATE / DELETE commands.
- *
- * Note that TOAST table needs no attention here as it's not scanned using
- * historic snapshot.
- */
-static void
-begin_concurrent_repack(Relation rel)
-{
- Oid toastrelid;
-
- /*
- * Avoid logical decoding of other relations by this backend. The lock we
- * have guarantees that the actual locator cannot be changed concurrently:
- * TRUNCATE needs AccessExclusiveLock.
- */
- Assert(CheckRelationLockedByMe(rel, ShareUpdateExclusiveLock, false));
- repacked_rel_locator = rel->rd_locator;
- toastrelid = rel->rd_rel->reltoastrelid;
- if (OidIsValid(toastrelid))
- {
- Relation toastrel;
-
- /* Avoid logical decoding of other TOAST relations. */
- toastrel = table_open(toastrelid, AccessShareLock);
- repacked_rel_toast_locator = toastrel->rd_locator;
- table_close(toastrel, AccessShareLock);
- }
-}
-
-/*
- * Call this when done with REPACK CONCURRENTLY.
- */
-static void
-end_concurrent_repack(void)
-{
- /*
- * Restore normal function of (future) logical decoding for this backend.
- */
- repacked_rel_locator.relNumber = InvalidOid;
- repacked_rel_toast_locator.relNumber = InvalidOid;
-}
-
/*
* Is this backend performing logical decoding on behalf of REPACK
* (CONCURRENTLY) ?
@@ -2492,9 +2569,10 @@ static LogicalDecodingContext *
setup_logical_decoding(Oid relid)
{
Relation rel;
- TupleDesc tupdesc;
+ Oid toastrelid;
LogicalDecodingContext *ctx;
- RepackDecodingState *dstate = palloc0_object(RepackDecodingState);
+ NameData slotname;
+ RepackDecodingState *dstate;
/*
* REPACK CONCURRENTLY is not allowed in a transaction block, so this
@@ -2502,20 +2580,22 @@ setup_logical_decoding(Oid relid)
*/
Assert(!TransactionIdIsValid(GetTopTransactionIdIfAny()));
- /*
- * A single backend should not execute multiple REPACK commands at a time,
- * so use PID to make the slot unique.
- */
- snprintf(NameStr(dstate->slotname), NAMEDATALEN, "repack_%d", MyProcPid);
-
/*
* Make sure we can use logical decoding.
*/
CheckSlotPermissions();
CheckLogicalDecodingRequirements();
- /* RS_TEMPORARY so that the slot gets cleaned up on ERROR. */
- ReplicationSlotCreate(NameStr(dstate->slotname), true, RS_TEMPORARY,
- false, false, false);
+
+ /*
+ * A single backend should not execute multiple REPACK commands at a time,
+ * so use PID to make the slot unique.
+ *
+ * RS_TEMPORARY so that the slot gets cleaned up on ERROR.
+ */
+ snprintf(NameStr(slotname), NAMEDATALEN, "repack_%d", MyProcPid);
+ ReplicationSlotCreate(NameStr(slotname), true, RS_TEMPORARY, false, false,
+ false);
+
EnsureLogicalDecodingEnabled();
/*
@@ -2538,104 +2618,109 @@ setup_logical_decoding(Oid relid)
DecodingContextFindStartpoint(ctx);
+ /*
+ * decode_concurrent_changes() needs non-blocking callback.
+ */
+ ctx->reader->routine.page_read = read_local_xlog_page_no_wait;
+
+ /*
+ * read_local_xlog_page_no_wait() needs to be able to indicate the end of
+ * WAL.
+ */
+ ctx->reader->private_data = MemoryContextAllocZero(ctx->context,
+ sizeof(ReadLocalXLogPageNoWaitPrivate));
+
+
/* Some WAL records should have been read. */
Assert(ctx->reader->EndRecPtr != InvalidXLogRecPtr);
+ /*
+ * Initialize repack_current_segment so that we can notice WAL segment
+ * boundaries.
+ */
XLByteToSeg(ctx->reader->EndRecPtr, repack_current_segment,
wal_segment_size);
- /*
- * Setup structures to store decoded changes.
- */
+ dstate = palloc0_object(RepackDecodingState);
dstate->relid = relid;
- dstate->tstore = tuplestore_begin_heap(false, false,
- maintenance_work_mem);
- /* Caller should already have the table locked. */
- rel = table_open(relid, NoLock);
- tupdesc = CreateTupleDescCopy(RelationGetDescr(rel));
- dstate->tupdesc = tupdesc;
- table_close(rel, NoLock);
+ /*
+ * Tuple descriptor may be needed to flatten a tuple before we write it to
+ * a file. A copy is needed because the decoding worker invalidates system
+ * caches before it starts to do the actual work.
+ */
+ rel = table_open(relid, AccessShareLock);
+ dstate->tupdesc = CreateTupleDescCopy(RelationGetDescr(rel));
- /* Initialize the descriptor to store the changes ... */
- dstate->tupdesc_change = CreateTemplateTupleDesc(1);
+ /* Avoid logical decoding of other relations. */
+ repacked_rel_locator = rel->rd_locator;
+ toastrelid = rel->rd_rel->reltoastrelid;
+ if (OidIsValid(toastrelid))
+ {
+ Relation toastrel;
- TupleDescInitEntry(dstate->tupdesc_change, 1, NULL, BYTEAOID, -1, 0);
- /* ... as well as the corresponding slot. */
- dstate->tsslot = MakeSingleTupleTableSlot(dstate->tupdesc_change,
- &TTSOpsMinimalTuple);
+ /* Avoid logical decoding of other TOAST relations. */
+ toastrel = table_open(toastrelid, AccessShareLock);
+ repacked_rel_toast_locator = toastrel->rd_locator;
+ table_close(toastrel, AccessShareLock);
+ }
+ table_close(rel, AccessShareLock);
- dstate->resowner = ResourceOwnerCreate(CurrentResourceOwner,
- "logical decoding");
+ /* The file will be set as soon as we have it opened. */
+ dstate->file = NULL;
ctx->output_writer_private = dstate;
+
return ctx;
}
/*
- * Retrieve tuple from ConcurrentChange structure.
+ * Decode logical changes from the WAL sequence and store them to a file.
*
- * The input data starts with the structure but it might not be appropriately
- * aligned.
+ * If true is returned, there is no more work for the worker.
*/
-static HeapTuple
-get_changed_tuple(char *change)
-{
- HeapTupleData tup_data;
- HeapTuple result;
- char *src;
-
- /*
- * Ensure alignment before accessing the fields. (This is why we can't use
- * heap_copytuple() instead of this function.)
- */
- src = change + offsetof(ConcurrentChange, tup_data);
- memcpy(&tup_data, src, sizeof(HeapTupleData));
-
- result = (HeapTuple) palloc(HEAPTUPLESIZE + tup_data.t_len);
- memcpy(result, &tup_data, sizeof(HeapTupleData));
- result->t_data = (HeapTupleHeader) ((char *) result + HEAPTUPLESIZE);
- src = change + SizeOfConcurrentChange;
- memcpy(result->t_data, src, result->t_len);
-
- return result;
-}
-
-/*
- * Decode logical changes from the WAL sequence up to end_of_wal.
- */
-void
-repack_decode_concurrent_changes(LogicalDecodingContext *ctx,
- XLogRecPtr end_of_wal)
+static bool
+decode_concurrent_changes(LogicalDecodingContext *ctx,
+ DecodingWorkerShared *shared)
{
RepackDecodingState *dstate;
- ResourceOwner resowner_old;
+ XLogRecPtr lsn_upto;
+ bool done;
+ char fname[MAXPGPATH];
dstate = (RepackDecodingState *) ctx->output_writer_private;
- resowner_old = CurrentResourceOwner;
- CurrentResourceOwner = dstate->resowner;
- PG_TRY();
+ /* Open the output file. */
+ DecodingWorkerFileName(fname, shared->relid, shared->last_exported + 1);
+ dstate->file = BufFileCreateFileSet(&shared->sfs.fs, fname);
+
+ SpinLockAcquire(&shared->mutex);
+ lsn_upto = shared->lsn_upto;
+ done = shared->done;
+ SpinLockRelease(&shared->mutex);
+
+ while (true)
{
- while (ctx->reader->EndRecPtr < end_of_wal)
+ XLogRecord *record;
+ XLogSegNo segno_new;
+ char *errm = NULL;
+ XLogRecPtr end_lsn;
+
+ CHECK_FOR_INTERRUPTS();
+
+ record = XLogReadRecord(ctx->reader, &errm);
+ if (record)
{
- XLogRecord *record;
- XLogSegNo segno_new;
- char *errm = NULL;
- XLogRecPtr end_lsn;
-
- record = XLogReadRecord(ctx->reader, &errm);
- if (errm)
- elog(ERROR, "%s", errm);
-
- if (record != NULL)
- LogicalDecodingProcessRecord(ctx, ctx->reader);
+ LogicalDecodingProcessRecord(ctx, ctx->reader);
/*
* If WAL segment boundary has been crossed, inform the decoding
- * system that the catalog_xmin can advance. (We can confirm more
- * often, but a filling a single WAL segment should not take much
- * time.)
+ * system that the catalog_xmin can advance.
+ *
+ * TODO Does it make sense to confirm more often? Segment size
+ * seems appropriate for restart_lsn (because less than a segment
+ * cannot be recycled anyway), however more frequent checks might
+ * be beneficial for catalog_xmin.
*/
end_lsn = ctx->reader->EndRecPtr;
XLByteToSeg(end_lsn, segno_new, wal_segment_size);
@@ -2646,80 +2731,137 @@ repack_decode_concurrent_changes(LogicalDecodingContext *ctx,
(uint32) (end_lsn >> 32), (uint32) end_lsn);
repack_current_segment = segno_new;
}
-
- CHECK_FOR_INTERRUPTS();
}
- InvalidateSystemCaches();
- CurrentResourceOwner = resowner_old;
+ else
+ {
+ ReadLocalXLogPageNoWaitPrivate *priv;
+
+ if (errm)
+ ereport(ERROR, (errmsg("%s", errm)));
+
+ /*
+ * In the decoding loop we do not want to get blocked when there
+ * is no more WAL available, otherwise the loop would become
+ * uninterruptible.
+ */
+ priv = (ReadLocalXLogPageNoWaitPrivate *)
+ ctx->reader->private_data;
+ if (priv->end_of_wal)
+ /* Do not miss the end of WAL condition next time. */
+ priv->end_of_wal = false;
+ else
+ ereport(ERROR, (errmsg("could not read WAL record")));
+ }
+
+ /*
+ * Whether we could read new record or not, keep checking if
+ * 'lsn_upto' was specified.
+ */
+ if (XLogRecPtrIsInvalid(lsn_upto))
+ {
+ SpinLockAcquire(&shared->mutex);
+ lsn_upto = shared->lsn_upto;
+ /* 'done' should be set at the same time as 'lsn_upto' */
+ done = shared->done;
+ SpinLockRelease(&shared->mutex);
+ }
+ if (!XLogRecPtrIsInvalid(lsn_upto) &&
+ ctx->reader->EndRecPtr >= lsn_upto)
+ break;
+
+ if (record == NULL)
+ {
+ int64 timeout = 0;
+ WaitLSNResult res;
+
+ /*
+ * Before we retry reading, wait until new WAL is flushed.
+ *
+ * There is a race condition such that the backend executing
+ * REPACK determines 'lsn_upto', but before it sets the shared
+ * variable, we reach the end of WAL. In that case we'd need to
+ * wait until the next WAL flush (unrelated to REPACK). Although
+ * that should not be a problem in a busy system, it might be
+ * noticeable in other cases, including regression tests (which
+ * are not necessarily executed in parallel). Therefore it makes
+ * sense to use timeout.
+ *
+ * If lsn_upto is valid, WAL records having LSN lower than that
+ * should already have been flushed to disk.
+ */
+ if (XLogRecPtrIsInvalid(lsn_upto))
+ timeout = 100L;
+ res = WaitForLSN(WAIT_LSN_TYPE_PRIMARY_FLUSH,
+ ctx->reader->EndRecPtr + 1,
+ timeout);
+ if (res != WAIT_LSN_RESULT_SUCCESS &&
+ res != WAIT_LSN_RESULT_TIMEOUT)
+ ereport(ERROR, (errmsg("waiting for WAL failed")));
+ }
}
- PG_CATCH();
- {
- /* clear all timetravel entries */
- InvalidateSystemCaches();
- CurrentResourceOwner = resowner_old;
- PG_RE_THROW();
- }
- PG_END_TRY();
+
+ /*
+ * Close the file so we can make it available to the backend.
+ */
+ BufFileClose(dstate->file);
+ dstate->file = NULL;
+ SpinLockAcquire(&shared->mutex);
+ shared->lsn_upto = InvalidXLogRecPtr;
+ shared->last_exported++;
+ SpinLockRelease(&shared->mutex);
+ ConditionVariableSignal(&shared->cv);
+
+ return done;
}
/*
* Apply changes stored in 'file'.
*/
static void
-apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
+apply_concurrent_changes(BufFile *file, ChangeDest *dest)
{
+ char kind;
+ uint32 t_len;
Relation rel = dest->rel;
TupleTableSlot *index_slot,
*ident_slot;
HeapTuple tup_old = NULL;
- if (dstate->nchanges == 0)
- return;
-
/* TupleTableSlot is needed to pass the tuple to ExecInsertIndexTuples(). */
- index_slot = MakeSingleTupleTableSlot(dstate->tupdesc, &TTSOpsHeapTuple);
+ index_slot = MakeSingleTupleTableSlot(RelationGetDescr(rel),
+ &TTSOpsHeapTuple);
/* A slot to fetch tuples from identity index. */
ident_slot = table_slot_create(rel, NULL);
- while (tuplestore_gettupleslot(dstate->tstore, true, false,
- dstate->tsslot))
+ while (true)
{
- bool shouldFree;
- HeapTuple tup_change,
- tup,
+ size_t nread;
+ HeapTuple tup,
tup_exist;
- char *change_raw,
- *src;
- ConcurrentChange change;
- bool isnull[1];
- Datum values[1];
CHECK_FOR_INTERRUPTS();
- /* Get the change from the single-column tuple. */
- tup_change = ExecFetchSlotHeapTuple(dstate->tsslot, false, &shouldFree);
- heap_deform_tuple(tup_change, dstate->tupdesc_change, values, isnull);
- Assert(!isnull[0]);
+ nread = BufFileReadMaybeEOF(file, &kind, 1, true);
+ /* Are we done with the file? */
+ if (nread == 0)
+ break;
- /* Make sure we access aligned data. */
- change_raw = (char *) DatumGetByteaP(values[0]);
- src = (char *) VARDATA(change_raw);
- memcpy(&change, src, SizeOfConcurrentChange);
+ /* Read the tuple. */
+ BufFileReadExact(file, &t_len, sizeof(t_len));
+ tup = (HeapTuple) palloc(HEAPTUPLESIZE + t_len);
+ tup->t_data = (HeapTupleHeader) ((char *) tup + HEAPTUPLESIZE);
+ BufFileReadExact(file, tup->t_data, t_len);
+ tup->t_len = t_len;
+ ItemPointerSetInvalid(&tup->t_self);
+ tup->t_tableOid = RelationGetRelid(dest->rel);
- /*
- * Extract the tuple from the change. The tuple is copied here because
- * it might be assigned to 'tup_old', in which case it needs to
- * survive into the next iteration.
- */
- tup = get_changed_tuple(src);
-
- if (change.kind == CHANGE_UPDATE_OLD)
+ if (kind == CHANGE_UPDATE_OLD)
{
Assert(tup_old == NULL);
tup_old = tup;
}
- else if (change.kind == CHANGE_INSERT)
+ else if (kind == CHANGE_INSERT)
{
Assert(tup_old == NULL);
@@ -2727,12 +2869,11 @@ apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
pfree(tup);
}
- else if (change.kind == CHANGE_UPDATE_NEW ||
- change.kind == CHANGE_DELETE)
+ else if (kind == CHANGE_UPDATE_NEW || kind == CHANGE_DELETE)
{
HeapTuple tup_key;
- if (change.kind == CHANGE_UPDATE_NEW)
+ if (kind == CHANGE_UPDATE_NEW)
{
tup_key = tup_old != NULL ? tup_old : tup;
}
@@ -2749,7 +2890,7 @@ apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
if (tup_exist == NULL)
elog(ERROR, "failed to find target tuple");
- if (change.kind == CHANGE_UPDATE_NEW)
+ if (kind == CHANGE_UPDATE_NEW)
apply_concurrent_update(rel, tup, tup_exist, dest->iistate,
index_slot);
else
@@ -2764,26 +2905,19 @@ apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
pfree(tup);
}
else
- elog(ERROR, "unrecognized kind of change: %d", change.kind);
+ elog(ERROR, "unrecognized kind of change: %d", kind);
/*
* If a change was applied now, increment CID for next writes and
* update the snapshot so it sees the changes we've applied so far.
*/
- if (change.kind != CHANGE_UPDATE_OLD)
+ if (kind != CHANGE_UPDATE_OLD)
{
CommandCounterIncrement();
UpdateActiveSnapshotCommandId();
}
-
- /* TTSOpsMinimalTuple has .get_heap_tuple==NULL. */
- Assert(shouldFree);
- pfree(tup_change);
}
- tuplestore_clear(dstate->tstore);
- dstate->nchanges = 0;
-
/* Cleanup. */
ExecDropSingleTupleTableSlot(index_slot);
ExecDropSingleTupleTableSlot(ident_slot);
@@ -2954,25 +3088,59 @@ find_target_tuple(Relation rel, ChangeDest *dest, HeapTuple tup_key,
}
/*
- * Decode and apply concurrent changes.
+ * Decode and apply concurrent changes, up to (and including) the record whose
+ * LSN is 'end_of_wal'.
*/
static void
-process_concurrent_changes(LogicalDecodingContext *decoding_ctx,
- XLogRecPtr end_of_wal, ChangeDest *dest)
+process_concurrent_changes(XLogRecPtr end_of_wal, ChangeDest *dest, bool done)
{
- RepackDecodingState *dstate;
+ DecodingWorkerShared *shared;
+ char fname[MAXPGPATH];
+ BufFile *file;
pgstat_progress_update_param(PROGRESS_REPACK_PHASE,
PROGRESS_REPACK_PHASE_CATCH_UP);
- dstate = (RepackDecodingState *) decoding_ctx->output_writer_private;
+ /* Ask the worker for the file. */
+ shared = (DecodingWorkerShared *) dsm_segment_address(decoding_worker->seg);
+ SpinLockAcquire(&shared->mutex);
+ shared->lsn_upto = end_of_wal;
+ shared->done = done;
+ SpinLockRelease(&shared->mutex);
- repack_decode_concurrent_changes(decoding_ctx, end_of_wal);
+ /*
+ * The worker needs to finish processing of the current WAL record. Even
+ * if it's idle, it'll need to close the output file. Thus we're likely to
+ * wait, so prepare for sleep.
+ */
+ ConditionVariablePrepareToSleep(&shared->cv);
+ for (;;)
+ {
+ int last_exported;
- if (dstate->nchanges == 0)
- return;
+ SpinLockAcquire(&shared->mutex);
+ last_exported = shared->last_exported;
+ SpinLockRelease(&shared->mutex);
- apply_concurrent_changes(dstate, dest);
+ /*
+ * Has the worker exported the file we are waiting for?
+ */
+ if (last_exported == dest->file_seq)
+ break;
+
+ ConditionVariableSleep(&shared->cv, WAIT_EVENT_REPACK_WORKER_EXPORT);
+ }
+ ConditionVariableCancelSleep();
+
+ /* Open the file. */
+ DecodingWorkerFileName(fname, shared->relid, dest->file_seq);
+ file = BufFileOpenFileSet(&shared->sfs.fs, fname, O_RDONLY, false);
+ apply_concurrent_changes(file, dest);
+
+ BufFileClose(file);
+
+ /* Get ready for the next file. */
+ dest->file_seq++;
}
/*
@@ -3098,15 +3266,10 @@ cleanup_logical_decoding(LogicalDecodingContext *ctx)
dstate = (RepackDecodingState *) ctx->output_writer_private;
- ExecDropSingleTupleTableSlot(dstate->tsslot);
- FreeTupleDesc(dstate->tupdesc_change);
FreeTupleDesc(dstate->tupdesc);
- tuplestore_end(dstate->tstore);
-
FreeDecodingContext(ctx);
- ReplicationSlotRelease();
- ReplicationSlotDrop(NameStr(dstate->slotname), false);
+ ReplicationSlotDropAcquired();
pfree(dstate);
}
@@ -3120,7 +3283,6 @@ cleanup_logical_decoding(LogicalDecodingContext *ctx)
*/
static void
rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
- LogicalDecodingContext *decoding_ctx,
TransactionId frozenXid,
MultiXactId cutoffMulti)
{
@@ -3201,6 +3363,7 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
&chgdst.ident_index);
chgdst.ident_key = build_identity_key(ident_idx_new, OldHeap,
&chgdst.ident_key_nentries);
+ chgdst.file_seq = WORKER_FILE_SNAPSHOT + 1;
/*
* During testing, wait for another backend to perform concurrent data
@@ -3222,7 +3385,7 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
* hold AccessExclusiveLock. (Quite some amount of WAL could have been
* written during the data copying and index creation.)
*/
- process_concurrent_changes(decoding_ctx, end_of_wal, &chgdst);
+ process_concurrent_changes(end_of_wal, &chgdst, false);
/*
* Acquire AccessExclusiveLock on the table, its TOAST relation (if there
@@ -3303,8 +3466,11 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
XLogFlush(wal_insert_ptr);
end_of_wal = GetFlushRecPtr(NULL);
- /* Apply the concurrent changes again. */
- process_concurrent_changes(decoding_ctx, end_of_wal, &chgdst);
+ /*
+ * Apply the concurrent changes again. Indicate that the decoding worker
+ * won't be needed anymore.
+ */
+ process_concurrent_changes(end_of_wal, &chgdst, true);
/* Remember info about rel before closing OldHeap */
relpersistence = OldHeap->rd_rel->relpersistence;
@@ -3416,3 +3582,510 @@ build_new_indexes(Relation NewHeap, Relation OldHeap, List *OldIndexes)
return result;
}
+
+/*
+ * Try to start a background worker to perform logical decoding of data
+ * changes applied to relation while REPACK CONCURRENTLY is copying its
+ * contents to a new table.
+ */
+static void
+start_decoding_worker(Oid relid)
+{
+ Size size;
+ dsm_segment *seg;
+ DecodingWorkerShared *shared;
+ shm_mq *mq;
+ shm_mq_handle *mqh;
+ BackgroundWorker bgw;
+
+ /* Setup shared memory. */
+ size = BUFFERALIGN(offsetof(DecodingWorkerShared, error_queue)) +
+ BUFFERALIGN(REPACK_ERROR_QUEUE_SIZE);
+ seg = dsm_create(size, 0);
+ shared = (DecodingWorkerShared *) dsm_segment_address(seg);
+ shared->lsn_upto = InvalidXLogRecPtr;
+ shared->done = false;
+ SharedFileSetInit(&shared->sfs, seg);
+ shared->last_exported = -1;
+ SpinLockInit(&shared->mutex);
+ shared->dbid = MyDatabaseId;
+
+ /*
+ * This is the UserId set in cluster_rel(). Security context shouldn't be
+ * needed for decoding worker.
+ */
+ shared->roleid = GetUserId();
+ shared->relid = relid;
+ ConditionVariableInit(&shared->cv);
+ shared->backend_proc = MyProc;
+ shared->backend_pid = MyProcPid;
+ shared->backend_proc_number = MyProcNumber;
+
+ mq = shm_mq_create((char *) BUFFERALIGN(shared->error_queue),
+ REPACK_ERROR_QUEUE_SIZE);
+ shm_mq_set_receiver(mq, MyProc);
+ mqh = shm_mq_attach(mq, seg, NULL);
+
+ memset(&bgw, 0, sizeof(bgw));
+ snprintf(bgw.bgw_name, BGW_MAXLEN,
+ "REPACK decoding worker for relation \"%s\"",
+ get_rel_name(relid));
+ snprintf(bgw.bgw_type, BGW_MAXLEN, "REPACK decoding worker");
+ bgw.bgw_flags = BGWORKER_SHMEM_ACCESS |
+ BGWORKER_BACKEND_DATABASE_CONNECTION;
+ bgw.bgw_start_time = BgWorkerStart_RecoveryFinished;
+ bgw.bgw_restart_time = BGW_NEVER_RESTART;
+ snprintf(bgw.bgw_library_name, MAXPGPATH, "postgres");
+ snprintf(bgw.bgw_function_name, BGW_MAXLEN, "RepackWorkerMain");
+ bgw.bgw_main_arg = UInt32GetDatum(dsm_segment_handle(seg));
+ bgw.bgw_notify_pid = MyProcPid;
+
+ decoding_worker = palloc0_object(DecodingWorker);
+ if (!RegisterDynamicBackgroundWorker(&bgw, &decoding_worker->handle))
+ ereport(ERROR,
+ (errcode(ERRCODE_CONFIGURATION_LIMIT_EXCEEDED),
+ errmsg("out of background worker slots"),
+ errhint("You might need to increase \"%s\".", "max_worker_processes")));
+
+ decoding_worker->seg = seg;
+ decoding_worker->error_mqh = mqh;
+
+ /*
+ * The decoding setup must be done before the caller can have XID assigned
+ * for any reason, otherwise the worker might end up in a deadlock,
+ * waiting for the caller's transaction to end. Therefore wait here until
+ * the worker indicates that it has the logical decoding initialized.
+ */
+ ConditionVariablePrepareToSleep(&shared->cv);
+ for (;;)
+ {
+ bool initialized;
+
+ SpinLockAcquire(&shared->mutex);
+ initialized = shared->initialized;
+ SpinLockRelease(&shared->mutex);
+
+ if (initialized)
+ break;
+
+ ConditionVariableSleep(&shared->cv, WAIT_EVENT_REPACK_WORKER_EXPORT);
+ }
+ ConditionVariableCancelSleep();
+}
+
+/*
+ * Stop the decoding worker and cleanup the related resources.
+ *
+ * The worker stops on its own when it knows there is no more work to do, but
+ * we need to stop it explicitly at least on ERROR in the launching backend.
+ */
+static void
+stop_decoding_worker(void)
+{
+ BgwHandleStatus status;
+
+ /* Haven't reached the worker startup? */
+ if (decoding_worker == NULL)
+ return;
+
+ /* Could not register the worker? */
+ if (decoding_worker->handle == NULL)
+ return;
+
+ TerminateBackgroundWorker(decoding_worker->handle);
+ /* The worker should really exit before the REPACK command does. */
+ HOLD_INTERRUPTS();
+ status = WaitForBackgroundWorkerShutdown(decoding_worker->handle);
+ RESUME_INTERRUPTS();
+
+ if (status == BGWH_POSTMASTER_DIED)
+ ereport(FATAL,
+ (errcode(ERRCODE_ADMIN_SHUTDOWN),
+ errmsg("postmaster exited during REPACK command")));
+
+ shm_mq_detach(decoding_worker->error_mqh);
+
+ /*
+ * If we could not cancel the current sleep due to ERROR, do that before
+ * we detach from the shared memory the condition variable is located in.
+ * If we did not, the bgworker ERROR handling code would try and fail
+ * badly.
+ */
+ ConditionVariableCancelSleep();
+
+ dsm_detach(decoding_worker->seg);
+ pfree(decoding_worker);
+ decoding_worker = NULL;
+}
+
+/* Is this process a REPACK worker? */
+static bool is_repack_worker = false;
+
+static pid_t backend_pid;
+static ProcNumber backend_proc_number;
+
+/*
+ * See ParallelWorkerShutdown for details.
+ */
+static void
+RepackWorkerShutdown(int code, Datum arg)
+{
+ SendProcSignal(backend_pid,
+ PROCSIG_REPACK_MESSAGE,
+ backend_proc_number);
+
+ dsm_detach((dsm_segment *) DatumGetPointer(arg));
+}
+
+/* REPACK decoding worker entry point */
+void
+RepackWorkerMain(Datum main_arg)
+{
+ dsm_segment *seg;
+ DecodingWorkerShared *shared;
+ shm_mq *mq;
+ shm_mq_handle *mqh;
+
+ is_repack_worker = true;
+
+ /*
+ * Override the default bgworker_die() with die() so we can use
+ * CHECK_FOR_INTERRUPTS().
+ */
+ pqsignal(SIGTERM, die);
+ BackgroundWorkerUnblockSignals();
+
+ seg = dsm_attach(DatumGetUInt32(main_arg));
+ if (seg == NULL)
+ ereport(ERROR,
+ (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
+ errmsg("could not map dynamic shared memory segment")));
+
+ shared = (DecodingWorkerShared *) dsm_segment_address(seg);
+
+ /* Arrange to signal the leader if we exit. */
+ backend_pid = shared->backend_pid;
+ backend_proc_number = shared->backend_proc_number;
+ before_shmem_exit(RepackWorkerShutdown, PointerGetDatum(seg));
+
+ /*
+ * Join locking group - see the comments around the call of
+ * start_decoding_worker().
+ */
+ if (!BecomeLockGroupMember(shared->backend_proc, backend_pid))
+ /* The leader is not running anymore. */
+ return;
+
+ /*
+ * Setup a queue to send error messages to the backend that launched this
+ * worker.
+ */
+ mq = (shm_mq *) (char *) BUFFERALIGN(shared->error_queue);
+ shm_mq_set_sender(mq, MyProc);
+ mqh = shm_mq_attach(mq, seg, NULL);
+ pq_redirect_to_shm_mq(seg, mqh);
+ pq_set_parallel_leader(shared->backend_pid,
+ shared->backend_proc_number);
+
+ /* Connect to the database. */
+ BackgroundWorkerInitializeConnectionByOid(shared->dbid, shared->roleid, 0);
+
+ repack_worker_internal(seg);
+}
+
+static void
+repack_worker_internal(dsm_segment *seg)
+{
+ DecodingWorkerShared *shared;
+ LogicalDecodingContext *decoding_ctx;
+ SharedFileSet *sfs;
+ Snapshot snapshot;
+
+ /*
+ * Transaction is needed to open relation, and it also provides us with a
+ * resource owner.
+ */
+ StartTransactionCommand();
+
+ shared = (DecodingWorkerShared *) dsm_segment_address(seg);
+
+ /*
+ * Not sure the spinlock is needed here - the backend should not change
+ * anything in the shared memory until we have serialized the snapshot.
+ */
+ SpinLockAcquire(&shared->mutex);
+ Assert(XLogRecPtrIsInvalid(shared->lsn_upto));
+ sfs = &shared->sfs;
+ SpinLockRelease(&shared->mutex);
+
+ SharedFileSetAttach(sfs, seg);
+
+ /*
+ * Prepare to capture the concurrent data changes ourselves.
+ */
+ decoding_ctx = setup_logical_decoding(shared->relid);
+
+ /* Announce that we're ready. */
+ SpinLockAcquire(&shared->mutex);
+ shared->initialized = true;
+ SpinLockRelease(&shared->mutex);
+ ConditionVariableSignal(&shared->cv);
+
+ /* Build the initial snapshot and export it. */
+ snapshot = SnapBuildInitialSnapshotForRepack(decoding_ctx->snapshot_builder);
+ export_initial_snapshot(snapshot, shared);
+
+ /*
+ * Only historic snapshots should be used now. Do not let us restrict the
+ * progress of xmin horizon.
+ */
+ InvalidateCatalogSnapshot();
+
+ while (!decode_concurrent_changes(decoding_ctx, shared))
+ ;
+
+ /* Cleanup. */
+ cleanup_logical_decoding(decoding_ctx);
+ CommitTransactionCommand();
+}
+
+/*
+ * Make snapshot available to the backend that launched the decoding worker.
+ */
+static void
+export_initial_snapshot(Snapshot snapshot, DecodingWorkerShared *shared)
+{
+ char fname[MAXPGPATH];
+ BufFile *file;
+ Size snap_size;
+ char *snap_space;
+
+ snap_size = EstimateSnapshotSpace(snapshot);
+ snap_space = (char *) palloc(snap_size);
+ SerializeSnapshot(snapshot, snap_space);
+ FreeSnapshot(snapshot);
+
+ DecodingWorkerFileName(fname, shared->relid, shared->last_exported + 1);
+ file = BufFileCreateFileSet(&shared->sfs.fs, fname);
+ /* To make restoration easier, write the snapshot size first. */
+ BufFileWrite(file, &snap_size, sizeof(snap_size));
+ BufFileWrite(file, snap_space, snap_size);
+ pfree(snap_space);
+ BufFileClose(file);
+
+ /* Increase the counter to tell the backend that the file is available. */
+ SpinLockAcquire(&shared->mutex);
+ shared->last_exported++;
+ SpinLockRelease(&shared->mutex);
+ ConditionVariableSignal(&shared->cv);
+}
+
+/*
+ * Get the initial snapshot from the decoding worker.
+ */
+static Snapshot
+get_initial_snapshot(DecodingWorker *worker)
+{
+ DecodingWorkerShared *shared;
+ char fname[MAXPGPATH];
+ BufFile *file;
+ Size snap_size;
+ char *snap_space;
+ Snapshot snapshot;
+
+ shared = (DecodingWorkerShared *) dsm_segment_address(worker->seg);
+
+ /*
+ * The worker needs to initialize the logical decoding, which usually
+ * takes some time. Therefore it makes sense to prepare for the sleep
+ * first.
+ */
+ ConditionVariablePrepareToSleep(&shared->cv);
+ for (;;)
+ {
+ int last_exported;
+
+ SpinLockAcquire(&shared->mutex);
+ last_exported = shared->last_exported;
+ SpinLockRelease(&shared->mutex);
+
+ /*
+ * Has the worker exported the file we are waiting for?
+ */
+ if (last_exported == WORKER_FILE_SNAPSHOT)
+ break;
+
+ ConditionVariableSleep(&shared->cv, WAIT_EVENT_REPACK_WORKER_EXPORT);
+ }
+ ConditionVariableCancelSleep();
+
+ /* Read the snapshot from a file. */
+ DecodingWorkerFileName(fname, shared->relid, WORKER_FILE_SNAPSHOT);
+ file = BufFileOpenFileSet(&shared->sfs.fs, fname, O_RDONLY, false);
+ BufFileReadExact(file, &snap_size, sizeof(snap_size));
+ snap_space = (char *) palloc(snap_size);
+ BufFileReadExact(file, snap_space, snap_size);
+ BufFileClose(file);
+
+ /* Restore it. */
+ snapshot = RestoreSnapshot(snap_space);
+ pfree(snap_space);
+
+ return snapshot;
+}
+
+bool
+IsRepackWorker(void)
+{
+ return is_repack_worker;
+}
+
+/*
+ * Handle receipt of an interrupt indicating a repack worker message.
+ *
+ * Note: this is called within a signal handler! All we can do is set
+ * a flag that will cause the next CHECK_FOR_INTERRUPTS() to invoke
+ * ProcessRepackMessages().
+ */
+void
+HandleRepackMessageInterrupt(void)
+{
+ InterruptPending = true;
+ RepackMessagePending = true;
+ SetLatch(MyLatch);
+}
+
+/*
+ * Process any queued protocol messages received from parallel workers.
+ */
+void
+ProcessRepackMessages(void)
+{
+ MemoryContext oldcontext;
+
+ static MemoryContext hpm_context = NULL;
+
+ /*
+ * Nothing to do if we haven't launched the worker yet or have already
+ * terminated it.
+ */
+ if (decoding_worker == NULL)
+ return;
+
+ /*
+ * This is invoked from ProcessInterrupts(), and since some of the
+ * functions it calls contain CHECK_FOR_INTERRUPTS(), there is a potential
+ * for recursive calls if more signals are received while this runs. It's
+ * unclear that recursive entry would be safe, and it doesn't seem useful
+ * even if it is safe, so let's block interrupts until done.
+ */
+ HOLD_INTERRUPTS();
+
+ /*
+ * Moreover, CurrentMemoryContext might be pointing almost anywhere. We
+ * don't want to risk leaking data into long-lived contexts, so let's do
+ * our work here in a private context that we can reset on each use.
+ */
+ if (hpm_context == NULL) /* first time through? */
+ hpm_context = AllocSetContextCreate(TopMemoryContext,
+ "ProcessRepackMessages",
+ ALLOCSET_DEFAULT_SIZES);
+ else
+ MemoryContextReset(hpm_context);
+
+ oldcontext = MemoryContextSwitchTo(hpm_context);
+
+ /* OK to process messages. Reset the flag saying there are more to do. */
+ RepackMessagePending = false;
+
+ /*
+ * Read as many messages as we can from each worker, but stop when no more
+ * messages can be read from the worker without blocking.
+ */
+ while (true)
+ {
+ shm_mq_result res;
+ Size nbytes;
+ void *data;
+
+ res = shm_mq_receive(decoding_worker->error_mqh, &nbytes,
+ &data, true);
+ if (res == SHM_MQ_WOULD_BLOCK)
+ break;
+ else if (res == SHM_MQ_SUCCESS)
+ {
+ StringInfoData msg;
+
+ initStringInfo(&msg);
+ appendBinaryStringInfo(&msg, data, nbytes);
+ ProcessRepackMessage(&msg);
+ pfree(msg.data);
+ }
+ else
+ {
+ /*
+ * The decoding worker is special in that it exits as soon as it
+ * has its work done. Thus the DETACHED result code is fine.
+ */
+ Assert(res == SHM_MQ_DETACHED);
+
+ break;
+ }
+ }
+
+ MemoryContextSwitchTo(oldcontext);
+
+ /* Might as well clear the context on our way out */
+ MemoryContextReset(hpm_context);
+
+ RESUME_INTERRUPTS();
+}
+
+/*
+ * Process a single protocol message received from a single parallel worker.
+ */
+static void
+ProcessRepackMessage(StringInfo msg)
+{
+ char msgtype;
+
+ msgtype = pq_getmsgbyte(msg);
+
+ switch (msgtype)
+ {
+ case PqMsg_ErrorResponse:
+ case PqMsg_NoticeResponse:
+ {
+ ErrorData edata;
+
+ /* Parse ErrorResponse or NoticeResponse. */
+ pq_parse_errornotice(msg, &edata);
+
+ /* Death of a worker isn't enough justification for suicide. */
+ edata.elevel = Min(edata.elevel, ERROR);
+
+ /*
+ * If desired, add a context line to show that this is a
+ * message propagated from a parallel worker. Otherwise, it
+ * can sometimes be confusing to understand what actually
+ * happened.
+ */
+ if (edata.context)
+ edata.context = psprintf("%s\n%s", edata.context,
+ _("decoding worker"));
+ else
+ edata.context = pstrdup(_("decoding worker"));
+
+ /* Rethrow error or print notice. */
+ ThrowErrorData(&edata);
+
+ break;
+ }
+
+ default:
+ {
+ elog(ERROR, "unrecognized message type received from decoding worker: %c (message length %d bytes)",
+ msgtype, msg->len);
+ }
+ }
+}
diff --git a/src/backend/libpq/pqmq.c b/src/backend/libpq/pqmq.c
index 6e4bbfb5aa1..42f6fa472c5 100644
--- a/src/backend/libpq/pqmq.c
+++ b/src/backend/libpq/pqmq.c
@@ -14,6 +14,7 @@
#include "postgres.h"
#include "access/parallel.h"
+#include "commands/cluster.h"
#include "libpq/libpq.h"
#include "libpq/pqformat.h"
#include "libpq/pqmq.h"
@@ -175,6 +176,10 @@ mq_putmessage(char msgtype, const char *s, size_t len)
SendProcSignal(pq_mq_parallel_leader_pid,
PROCSIG_PARALLEL_APPLY_MESSAGE,
pq_mq_parallel_leader_proc_number);
+ else if (IsRepackWorker())
+ SendProcSignal(pq_mq_parallel_leader_pid,
+ PROCSIG_REPACK_MESSAGE,
+ pq_mq_parallel_leader_proc_number);
else
{
Assert(IsParallelWorker());
diff --git a/src/backend/postmaster/bgworker.c b/src/backend/postmaster/bgworker.c
index 8678ea4e139..1edb77ba431 100644
--- a/src/backend/postmaster/bgworker.c
+++ b/src/backend/postmaster/bgworker.c
@@ -13,6 +13,7 @@
#include "postgres.h"
#include "access/parallel.h"
+#include "commands/cluster.h"
#include "libpq/pqsignal.h"
#include "miscadmin.h"
#include "pgstat.h"
@@ -142,6 +143,10 @@ static const struct
{
.fn_name = "SequenceSyncWorkerMain",
.fn_addr = SequenceSyncWorkerMain
+ },
+ {
+ .fn_name = "RepackWorkerMain",
+ .fn_addr = RepackWorkerMain
}
};
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index 603a2b94d05..7651b187418 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -194,7 +194,11 @@ StartupDecodingContext(List *output_plugin_options,
ctx->slot = slot;
- ctx->reader = XLogReaderAllocate(wal_segment_size, NULL, xl_routine, ctx);
+ /*
+ * TODO A separate patch for PG core, unless there's really a reason to
+ * pass ctx for private_data (May extensions expect ctx?).
+ */
+ ctx->reader = XLogReaderAllocate(wal_segment_size, NULL, xl_routine, NULL);
if (!ctx->reader)
ereport(ERROR,
(errcode(ERRCODE_OUT_OF_MEMORY),
diff --git a/src/backend/replication/pgoutput_repack/pgoutput_repack.c b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
index 601016716b3..d0c464a98d5 100644
--- a/src/backend/replication/pgoutput_repack/pgoutput_repack.c
+++ b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
@@ -168,17 +168,13 @@ store_change(LogicalDecodingContext *ctx, ConcurrentChangeKind kind,
HeapTuple tuple)
{
RepackDecodingState *dstate;
- char *change_raw;
- ConcurrentChange change;
+ char kind_byte = (char) kind;
bool flattened = false;
- Size size;
- Datum values[1];
- bool isnull[1];
- char *dst;
dstate = (RepackDecodingState *) ctx->output_writer_private;
- size = VARHDRSZ + SizeOfConcurrentChange;
+ /* Store the change kind. */
+ BufFileWrite(dstate->file, &kind_byte, 1);
/*
* ReorderBufferCommit() stores the TOAST chunks in its private memory
@@ -195,46 +191,12 @@ store_change(LogicalDecodingContext *ctx, ConcurrentChangeKind kind,
tuple = toast_flatten_tuple(tuple, dstate->tupdesc);
flattened = true;
}
+ /* Store the tuple size ... */
+ BufFileWrite(dstate->file, &tuple->t_len, sizeof(tuple->t_len));
+ /* ... and the tuple itself. */
+ BufFileWrite(dstate->file, tuple->t_data, tuple->t_len);
- size += tuple->t_len;
- if (size >= MaxAllocSize)
- elog(ERROR, "Change is too big.");
-
- /* Construct the change. */
- change_raw = (char *) palloc0(size);
- SET_VARSIZE(change_raw, size);
-
- /*
- * Since the varlena alignment might not be sufficient for the structure,
- * set the fields in a local instance and remember where it should
- * eventually be copied.
- */
- change.kind = kind;
- dst = (char *) VARDATA(change_raw);
-
- /*
- * Copy the tuple.
- *
- * Note: change->tup_data.t_data must be fixed on retrieval!
- */
- memcpy(&change.tup_data, tuple, sizeof(HeapTupleData));
- memcpy(dst, &change, SizeOfConcurrentChange);
- dst += SizeOfConcurrentChange;
- memcpy(dst, tuple->t_data, tuple->t_len);
-
- /* The data has been copied. */
+ /* Free the flat copy if created above. */
if (flattened)
pfree(tuple);
-
- /* Store as tuple of 1 bytea column. */
- values[0] = PointerGetDatum(change_raw);
- isnull[0] = false;
- tuplestore_putvalues(dstate->tstore, dstate->tupdesc_change,
- values, isnull);
-
- /* Accounting. */
- dstate->nchanges++;
-
- /* Cleanup. */
- pfree(change_raw);
}
diff --git a/src/backend/storage/ipc/procsignal.c b/src/backend/storage/ipc/procsignal.c
index 7505c9d3a37..7c2d637bd33 100644
--- a/src/backend/storage/ipc/procsignal.c
+++ b/src/backend/storage/ipc/procsignal.c
@@ -19,6 +19,7 @@
#include "access/parallel.h"
#include "commands/async.h"
+#include "commands/cluster.h"
#include "miscadmin.h"
#include "pgstat.h"
#include "port/pg_bitutils.h"
@@ -698,6 +699,9 @@ procsignal_sigusr1_handler(SIGNAL_ARGS)
if (CheckProcSignal(PROCSIG_PARALLEL_APPLY_MESSAGE))
HandleParallelApplyMessageInterrupt();
+ if (CheckProcSignal(PROCSIG_REPACK_MESSAGE))
+ HandleRepackMessageInterrupt();
+
if (CheckProcSignal(PROCSIG_RECOVERY_CONFLICT))
HandleRecoveryConflictInterrupt();
diff --git a/src/backend/tcop/postgres.c b/src/backend/tcop/postgres.c
index d01a09dd0c4..9711abaf2eb 100644
--- a/src/backend/tcop/postgres.c
+++ b/src/backend/tcop/postgres.c
@@ -36,6 +36,7 @@
#include "access/xact.h"
#include "catalog/pg_type.h"
#include "commands/async.h"
+#include "commands/cluster.h"
#include "commands/event_trigger.h"
#include "commands/explain_state.h"
#include "commands/prepare.h"
@@ -3575,6 +3576,9 @@ ProcessInterrupts(void)
if (ParallelApplyMessagePending)
ProcessParallelApplyMessages();
+
+ if (RepackMessagePending)
+ ProcessRepackMessages();
}
/*
diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt
index 4aa864fe3c3..b00bd794759 100644
--- a/src/backend/utils/activity/wait_event_names.txt
+++ b/src/backend/utils/activity/wait_event_names.txt
@@ -154,6 +154,7 @@ RECOVERY_CONFLICT_SNAPSHOT "Waiting for recovery conflict resolution for a vacuu
RECOVERY_CONFLICT_TABLESPACE "Waiting for recovery conflict resolution for dropping a tablespace."
RECOVERY_END_COMMAND "Waiting for <xref linkend="guc-recovery-end-command"/> to complete."
RECOVERY_PAUSE "Waiting for recovery to be resumed."
+REPACK_WORKER_EXPORT "Waiting for decoding worker to export a new output file."
REPLICATION_ORIGIN_DROP "Waiting for a replication origin to become inactive so it can be dropped."
REPLICATION_SLOT_DROP "Waiting for a replication slot to become inactive so it can be dropped."
RESTORE_COMMAND "Waiting for <xref linkend="guc-restore-command"/> to complete."
diff --git a/src/include/access/tableam.h b/src/include/access/tableam.h
index 82aa258f72d..ce72e7e017d 100644
--- a/src/include/access/tableam.h
+++ b/src/include/access/tableam.h
@@ -22,7 +22,6 @@
#include "access/xact.h"
#include "commands/vacuum.h"
#include "executor/tuptable.h"
-#include "replication/logical.h"
#include "storage/read_stream.h"
#include "utils/rel.h"
#include "utils/snapshot.h"
@@ -631,7 +630,6 @@ typedef struct TableAmRoutine
bool use_sort,
TransactionId OldestXmin,
Snapshot snapshot,
- LogicalDecodingContext *decoding_ctx,
TransactionId *xid_cutoff,
MultiXactId *multi_cutoff,
double *num_tuples,
@@ -1663,8 +1661,6 @@ table_relation_copy_data(Relation rel, const RelFileLocator *newrlocator)
* - *multi_cutoff - ditto
* - snapshot - if != NULL, ignore data changes done by transactions that this
* (MVCC) snapshot considers still in-progress or in the future.
- * - decoding_ctx - logical decoding context, to capture concurrent data
- * changes.
*
* Output parameters:
* - *xid_cutoff - rel's new relfrozenxid value, may be invalid
@@ -1678,7 +1674,6 @@ table_relation_copy_for_cluster(Relation OldTable, Relation NewTable,
bool use_sort,
TransactionId OldestXmin,
Snapshot snapshot,
- LogicalDecodingContext *decoding_ctx,
TransactionId *xid_cutoff,
MultiXactId *multi_cutoff,
double *num_tuples,
@@ -1687,7 +1682,7 @@ table_relation_copy_for_cluster(Relation OldTable, Relation NewTable,
{
OldTable->rd_tableam->relation_copy_for_cluster(OldTable, NewTable, OldIndex,
use_sort, OldestXmin,
- snapshot, decoding_ctx,
+ snapshot,
xid_cutoff, multi_cutoff,
num_tuples, tups_vacuumed,
tups_recently_dead);
diff --git a/src/include/commands/cluster.h b/src/include/commands/cluster.h
index 6a5c476294a..1b05d5d418b 100644
--- a/src/include/commands/cluster.h
+++ b/src/include/commands/cluster.h
@@ -17,11 +17,13 @@
#include "nodes/parsenodes.h"
#include "parser/parse_node.h"
#include "replication/decode.h"
+#include "postmaster/bgworker.h"
#include "replication/logical.h"
+#include "storage/buffile.h"
#include "storage/lock.h"
+#include "storage/shm_mq.h"
#include "utils/relcache.h"
#include "utils/resowner.h"
-#include "utils/tuplestore.h"
/* flag bits for ClusterParams->options */
@@ -44,6 +46,9 @@ typedef struct ClusterParams
* The following definitions are used by REPACK CONCURRENTLY.
*/
+/*
+ * Stored as a single byte in the output file.
+ */
typedef enum
{
CHANGE_INSERT,
@@ -52,68 +57,30 @@ typedef enum
CHANGE_DELETE
} ConcurrentChangeKind;
-typedef struct ConcurrentChange
-{
- /* See the enum above. */
- ConcurrentChangeKind kind;
-
- /*
- * The actual tuple.
- *
- * The tuple data follows the ConcurrentChange structure. Before use make
- * sure the tuple is correctly aligned (ConcurrentChange can be stored as
- * bytea) and that tuple->t_data is fixed.
- */
- HeapTupleData tup_data;
-} ConcurrentChange;
-
-#define SizeOfConcurrentChange (offsetof(ConcurrentChange, tup_data) + \
- sizeof(HeapTupleData))
-
/*
* Logical decoding state.
*
- * Here we store the data changes that we decode from WAL while the table
- * contents is being copied to a new storage. Also the necessary metadata
- * needed to apply these changes to the table is stored here.
+ * The output plugin uses it to store the data changes that it decodes from
+ * WAL while the table contents is being copied to a new storage.
*/
typedef struct RepackDecodingState
{
/* The relation whose changes we're decoding. */
Oid relid;
- /* Replication slot name. */
- NameData slotname;
-
- /*
- * Decoded changes are stored here. Although we try to avoid excessive
- * batches, it can happen that the changes need to be stored to disk. The
- * tuplestore does this transparently.
- */
- Tuplestorestate *tstore;
-
- /* The current number of changes in tstore. */
- double nchanges;
-
- /*
- * Descriptor to store the ConcurrentChange structure serialized (bytea).
- * We can't store the tuple directly because tuplestore only supports
- * minimum tuple and we may need to transfer OID system column from the
- * output plugin. Also we need to transfer the change kind, so it's better
- * to put everything in the structure than to use 2 tuplestores "in
- * parallel".
- */
- TupleDesc tupdesc_change;
-
- /* Tuple descriptor needed to update indexes. */
+ /* Tuple descriptor of the relation being processed. */
TupleDesc tupdesc;
- /* Slot to retrieve data from tstore. */
- TupleTableSlot *tsslot;
-
- ResourceOwner resowner;
+ /* The current output file. */
+ BufFile *file;
} RepackDecodingState;
+extern PGDLLIMPORT volatile sig_atomic_t RepackMessagePending;
+
+extern bool IsRepackWorker(void);
+extern void HandleRepackMessageInterrupt(void);
+extern void ProcessRepackMessages(void);
+
extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel);
extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid,
@@ -136,6 +103,6 @@ extern void finish_heap_swap(Oid OIDOldHeap, Oid OIDNewHeap,
extern bool am_decoding_for_repack(void);
extern bool change_useless_for_repack(XLogRecordBuffer *buf);
-extern void repack_decode_concurrent_changes(LogicalDecodingContext *ctx,
- XLogRecPtr end_of_wal);
+
+extern void RepackWorkerMain(Datum main_arg);
#endif /* CLUSTER_H */
diff --git a/src/include/storage/procsignal.h b/src/include/storage/procsignal.h
index 348fba53a93..a944ee0d211 100644
--- a/src/include/storage/procsignal.h
+++ b/src/include/storage/procsignal.h
@@ -36,6 +36,7 @@ typedef enum
PROCSIG_BARRIER, /* global barrier interrupt */
PROCSIG_LOG_MEMORY_CONTEXT, /* ask backend to log the memory contexts */
PROCSIG_PARALLEL_APPLY_MESSAGE, /* Message from parallel apply workers */
+ PROCSIG_REPACK_MESSAGE, /* Message from repack worker */
PROCSIG_RECOVERY_CONFLICT, /* backend is blocking recovery, check
* PGPROC->pendingRecoveryConflicts for the
* reason */
diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list
index 71a0802187e..b35d4e318fc 100644
--- a/src/tools/pgindent/typedefs.list
+++ b/src/tools/pgindent/typedefs.list
@@ -499,7 +499,6 @@ CompressFileHandle
CompressionLocation
CompressorState
ComputeXidHorizonsResult
-ConcurrentChange
ConcurrentChangeKind
ConditionVariable
ConditionVariableMinimallyPadded
@@ -639,6 +638,8 @@ DeclareCursorStmt
DecodedBkpBlock
DecodedXLogRecord
DecodingOutputState
+DecodingWorker
+DecodingWorkerShared
DefElem
DefElemAction
DefaultACLInfo
--
2.47.3
--kmtyk4ln3767qrma--
^ permalink raw reply [nested|flat] 55+ messages in thread
* [PATCH 5/5] Use background worker to do logical decoding.
@ 2026-02-16 19:49 Antonin Houska <[email protected]>
0 siblings, 0 replies; 55+ messages in thread
From: Antonin Houska @ 2026-02-16 19:49 UTC (permalink / raw)
If the backend performing REPACK (CONCURRENTLY) does both data copying and
logical decoding, it has to "travel in time" back and forth and therefore it
has to invalidate system caches quite a few times. (The copying and the
decoding work with different catalog snapshots.) As the decoding worker has
separate caches, the switching is not necessary.
Without the worker, it'd also be difficult to switch between potentially long
running tasks like index build and WAL decoding. (No decoding during that time
at all can suspend archiving / recycling of WAL segments for some time, which
in turn may result in full disk.)
Another problem is that, after having acquired AccessExclusiveLock (in order
to swap the files), the backend needs to both decode and apply the data
changes that took place while it was waiting for the lock. With the decoding
worker, the decoding runs all the time, so the backend only needs to apply the
changes. This can reduce the time the exclusive lock is held for.
Note that the code added in order to handle ERRORs in the background worker
almost duplicates the existing code that does the same for other types of
workers (See ProcessParallelMessages() and
ProcessParallelApplyMessages()). Refactoring of the existing code might be
useful, to reduce the duplication.
---
src/backend/access/heap/heapam_handler.c | 44 -
src/backend/commands/cluster.c | 1175 +++++++++++++----
src/backend/libpq/pqmq.c | 5 +
src/backend/postmaster/bgworker.c | 5 +
src/backend/replication/logical/logical.c | 6 +-
.../pgoutput_repack/pgoutput_repack.c | 54 +-
src/backend/storage/ipc/procsignal.c | 4 +
src/backend/tcop/postgres.c | 4 +
.../utils/activity/wait_event_names.txt | 1 +
src/include/access/tableam.h | 7 +-
src/include/commands/cluster.h | 71 +-
src/include/storage/procsignal.h | 1 +
src/tools/pgindent/typedefs.list | 3 +-
13 files changed, 979 insertions(+), 401 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index 48b66823523..e55a4bf727e 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -33,7 +33,6 @@
#include "catalog/index.h"
#include "catalog/storage.h"
#include "catalog/storage_xlog.h"
-#include "commands/cluster.h"
#include "commands/progress.h"
#include "executor/executor.h"
#include "miscadmin.h"
@@ -688,7 +687,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
Relation OldIndex, bool use_sort,
TransactionId OldestXmin,
Snapshot snapshot,
- LogicalDecodingContext *decoding_ctx,
TransactionId *xid_cutoff,
MultiXactId *multi_cutoff,
double *num_tuples,
@@ -710,7 +708,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
BufferHeapTupleTableSlot *hslot;
BlockNumber prev_cblock = InvalidBlockNumber;
bool concurrent = snapshot != NULL;
- XLogRecPtr end_of_wal_prev = GetFlushRecPtr(NULL);
/* Remember if it's a system catalog */
is_system_catalog = IsSystemRelation(OldHeap);
@@ -972,31 +969,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
ct_val[1] = *num_tuples;
pgstat_progress_update_multi_param(2, ct_index, ct_val);
}
-
- /*
- * Process the WAL produced by the load, as well as by other
- * transactions, so that the replication slot can advance and WAL does
- * not pile up. Use wal_segment_size as a threshold so that we do not
- * introduce the decoding overhead too often.
- *
- * Of course, we must not apply the changes until the initial load has
- * completed.
- *
- * Note that our insertions into the new table should not be decoded
- * as we (intentionally) do not write the logical decoding specific
- * information to WAL.
- */
- if (concurrent)
- {
- XLogRecPtr end_of_wal;
-
- end_of_wal = GetFlushRecPtr(NULL);
- if ((end_of_wal - end_of_wal_prev) > wal_segment_size)
- {
- repack_decode_concurrent_changes(decoding_ctx, end_of_wal);
- end_of_wal_prev = end_of_wal;
- }
- }
}
if (indexScan != NULL)
@@ -1042,22 +1014,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
-
- /*
- * Try to keep the amount of not-yet-decoded WAL small, like
- * above.
- */
- if (concurrent)
- {
- XLogRecPtr end_of_wal;
-
- end_of_wal = GetFlushRecPtr(NULL);
- if ((end_of_wal - end_of_wal_prev) > wal_segment_size)
- {
- repack_decode_concurrent_changes(decoding_ctx, end_of_wal);
- end_of_wal_prev = end_of_wal;
- }
- }
}
tuplesort_end(tuplesort);
diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c
index b60cda202d2..7c9945846e2 100644
--- a/src/backend/commands/cluster.c
+++ b/src/backend/commands/cluster.c
@@ -12,12 +12,13 @@
* In concurrent mode, we lock the table with only ShareUpdateExclusiveLock,
* then do an initial copy as above. However, while the tuples are being
* copied, concurrent transactions could modify the table. To cope with those
- * changes, we rely on logical decoding to obtain them from WAL. The changes
- * are accumulated in a tuplestore. Once the initial copy is complete, we
- * read the changes from the tuplestore and re-apply them on the new heap.
- * Then we upgrade our ShareUpdateExclusiveLock to AccessExclusiveLock and
- * swap the relfilenodes. This way, the time we hold a strong lock on the
- * table is much reduced, and the bloat is eliminated.
+ * changes, we rely on logical decoding to obtain them from WAL. A bgworker
+ * consumes WAL while the initial copy is ongoing (to prevent excessive WAL
+ * from being reserved), and accumulates the changes in a file. Once the
+ * initial copy is complete, we read the changes from the file and re-apply
+ * them on the new heap. Then we upgrade our ShareUpdateExclusiveLock to
+ * AccessExclusiveLock and swap the relfilenodes. This way, the time we hold
+ * a strong lock on the table is much reduced, and the bloat is eliminated.
*
* There is hardly anything left of Paul Brown's original implementation...
*
@@ -45,6 +46,7 @@
#include "access/xlog_internal.h"
#include "access/xloginsert.h"
#include "access/xlogutils.h"
+#include "access/xlogwait.h"
#include "catalog/catalog.h"
#include "catalog/dependency.h"
#include "catalog/heap.h"
@@ -61,6 +63,8 @@
#include "commands/tablecmds.h"
#include "commands/vacuum.h"
#include "executor/executor.h"
+#include "libpq/pqformat.h"
+#include "libpq/pqmq.h"
#include "miscadmin.h"
#include "optimizer/optimizer.h"
#include "pgstat.h"
@@ -71,6 +75,8 @@
#include "storage/ipc.h"
#include "storage/lmgr.h"
#include "storage/predicate.h"
+#include "storage/procsignal.h"
+#include "tcop/tcopprot.h"
#include "utils/acl.h"
#include "utils/fmgroids.h"
#include "utils/guc.h"
@@ -117,6 +123,12 @@ typedef struct IndexInsertState
/* The WAL segment being decoded. */
static XLogSegNo repack_current_segment = 0;
+/*
+ * The first file exported by the decoding worker must contain a snapshot, the
+ * following ones contain the data changes.
+ */
+#define WORKER_FILE_SNAPSHOT 0
+
/*
* Information needed to apply concurrent data changes.
*/
@@ -136,8 +148,110 @@ typedef struct ChangeDest
/* Needed to update indexes of rel_dst. */
IndexInsertState *iistate;
+
+ /*
+ * Sequential number of the file containing the changes.
+ *
+ * TODO This field makes the structure name less descriptive. Should we
+ * rename it, e.g. to ChangeApplyInfo?
+ */
+ int file_seq;
} ChangeDest;
+/*
+ * Layout of shared memory used for communication between backend and the
+ * worker that performs logical decoding of data changes
+ */
+typedef struct DecodingWorkerShared
+{
+ /* Is the decoding initialized? */
+ bool initialized;
+
+ /*
+ * Once the worker has reached this LSN, it should close the current
+ * output file and either create a new one or exit, according to the field
+ * 'done'. If the value is InvalidXLogRecPtr, the worker should decode all
+ * the WAL available and keep checking this field. It is ok if the worker
+ * had already decoded records whose LSN is >= lsn_upto before this field
+ * has been set.
+ */
+ XLogRecPtr lsn_upto;
+
+ /* Exit after closing the current file? */
+ bool done;
+
+ /* The output is stored here. */
+ SharedFileSet sfs;
+
+ /* Number of the last file exported by the worker. */
+ int last_exported;
+
+ /* Synchronize access to the fields above. */
+ slock_t mutex;
+
+ /* Database to connect to. */
+ Oid dbid;
+
+ /* Role to connect as. */
+ Oid roleid;
+
+ /* Decode data changes of this relation. */
+ Oid relid;
+
+ /* The backend uses this to wait for the worker. */
+ ConditionVariable cv;
+
+ /* Info to signal the backend. */
+ PGPROC *backend_proc;
+ pid_t backend_pid;
+ ProcNumber backend_proc_number;
+
+ /*
+ * Memory the queue is located in.
+ *
+ * For considerations on the value see the comments of
+ * PARALLEL_ERROR_QUEUE_SIZE.
+ */
+#define REPACK_ERROR_QUEUE_SIZE 16384
+ char error_queue[FLEXIBLE_ARRAY_MEMBER];
+} DecodingWorkerShared;
+
+/*
+ * Generate worker's output file name. If relations of the same 'relid' happen
+ * to be processed at the same time, they must be from different databases and
+ * therefore different backends must be involved. (PID is already present in
+ * the fileset name.)
+ */
+static inline void
+DecodingWorkerFileName(char *fname, Oid relid, uint32 seq)
+{
+ snprintf(fname, MAXPGPATH, "%u-%u", relid, seq);
+}
+
+/*
+ * Backend-local information to control the decoding worker.
+ */
+typedef struct DecodingWorker
+{
+ /* The worker. */
+ BackgroundWorkerHandle *handle;
+
+ /* DecodingWorkerShared is in this segment. */
+ dsm_segment *seg;
+
+ /* Handle of the error queue. */
+ shm_mq_handle *error_mqh;
+} DecodingWorker;
+
+/* Pointer to currently running decoding worker. */
+static DecodingWorker *decoding_worker = NULL;
+
+/*
+ * Is there a message sent by a repack worker that the backend needs to
+ * receive?
+ */
+volatile sig_atomic_t RepackMessagePending = false;
+
static bool cluster_rel_recheck(RepackCommand cmd, Relation OldHeap,
Oid indexOid, Oid userid, LOCKMODE lmode,
int options);
@@ -145,7 +259,7 @@ static void check_repack_concurrently_requirements(Relation rel);
static void rebuild_relation(Relation OldHeap, Relation index, bool verbose,
bool concurrent);
static void copy_table_data(Relation NewHeap, Relation OldHeap, Relation OldIndex,
- Snapshot snapshot, LogicalDecodingContext *decoding_ctx,
+ Snapshot snapshot,
bool verbose,
bool *pSwapToastByContent,
TransactionId *pFreezeXid,
@@ -158,12 +272,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd,
static bool cluster_is_permitted_for_relation(RepackCommand cmd,
Oid relid, Oid userid);
-static void begin_concurrent_repack(Relation rel);
-static void end_concurrent_repack(void);
static LogicalDecodingContext *setup_logical_decoding(Oid relid);
-static HeapTuple get_changed_tuple(char *change);
-static void apply_concurrent_changes(RepackDecodingState *dstate,
- ChangeDest *dest);
+static bool decode_concurrent_changes(LogicalDecodingContext *ctx,
+ DecodingWorkerShared *shared);
+static void apply_concurrent_changes(BufFile *file, ChangeDest *dest);
static void apply_concurrent_insert(Relation rel, HeapTuple tup,
IndexInsertState *iistate,
TupleTableSlot *index_slot);
@@ -175,9 +287,9 @@ static void apply_concurrent_delete(Relation rel, HeapTuple tup_target);
static HeapTuple find_target_tuple(Relation rel, ChangeDest *dest,
HeapTuple tup_key,
TupleTableSlot *ident_slot);
-static void process_concurrent_changes(LogicalDecodingContext *decoding_ctx,
- XLogRecPtr end_of_wal,
- ChangeDest *dest);
+static void process_concurrent_changes(XLogRecPtr end_of_wal,
+ ChangeDest *dest,
+ bool done);
static IndexInsertState *get_index_insert_state(Relation relation,
Oid ident_index_id,
Relation *ident_index_p);
@@ -186,7 +298,6 @@ static ScanKey build_identity_key(Oid ident_idx_oid, Relation rel_src,
static void free_index_insert_state(IndexInsertState *iistate);
static void cleanup_logical_decoding(LogicalDecodingContext *ctx);
static void rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
- LogicalDecodingContext *decoding_ctx,
TransactionId frozenXid,
MultiXactId cutoffMulti);
static List *build_new_indexes(Relation NewHeap, Relation OldHeap, List *OldIndexes);
@@ -196,6 +307,13 @@ static Relation process_single_relation(RepackStmt *stmt,
ClusterParams *params);
static Oid determine_clustered_index(Relation rel, bool usingindex,
const char *indexname);
+static void start_decoding_worker(Oid relid);
+static void stop_decoding_worker(void);
+static void repack_worker_internal(dsm_segment *seg);
+static void export_initial_snapshot(Snapshot snapshot,
+ DecodingWorkerShared *shared);
+static Snapshot get_initial_snapshot(DecodingWorker *worker);
+static void ProcessRepackMessage(StringInfo msg);
static const char *RepackCommandAsString(RepackCommand cmd);
@@ -619,20 +737,20 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid,
/* rebuild_relation does all the dirty work */
PG_TRY();
{
- /*
- * For concurrent processing, make sure that our logical decoding
- * ignores data changes of other tables than the one we are
- * processing.
- */
- if (concurrent)
- begin_concurrent_repack(OldHeap);
-
rebuild_relation(OldHeap, index, verbose, concurrent);
}
PG_FINALLY();
{
if (concurrent)
- end_concurrent_repack();
+ {
+ /*
+ * Since during normal operation the worker was already asked to
+ * exit, stopping it explicitly is especially important on ERROR.
+ * However it still seems a good practice to make sure that the
+ * worker never survives the REPACK command.
+ */
+ stop_decoding_worker();
+ }
}
PG_END_TRY();
@@ -929,7 +1047,6 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
bool swap_toast_by_content;
TransactionId frozenXid;
MultiXactId cutoffMulti;
- LogicalDecodingContext *decoding_ctx = NULL;
Snapshot snapshot = NULL;
#if USE_ASSERT_CHECKING
LOCKMODE lmode;
@@ -943,19 +1060,36 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
if (concurrent)
{
/*
- * Prepare to capture the concurrent data changes.
+ * The worker needs to be member of the locking group we're the leader
+ * of. We ought to become the leader before the worker starts. The
+ * worker will join the group as soon as it starts.
*
- * Note that this call waits for all transactions with XID already
- * assigned to finish. If some of those transactions is waiting for a
- * lock conflicting with ShareUpdateExclusiveLock on our table (e.g.
- * it runs CREATE INDEX), we can end up in a deadlock. Not sure this
- * risk is worth unlocking/locking the table (and its clustering
- * index) and checking again if its still eligible for REPACK
- * CONCURRENTLY.
+ * This is to make sure that the deadlock described below is
+ * detectable by deadlock.c: if the worker waits for a transaction to
+ * complete and we are waiting for the worker output, then effectively
+ * we (i.e. this backend) are waiting for that transaction.
*/
- decoding_ctx = setup_logical_decoding(tableOid);
+ BecomeLockGroupLeader();
+
+ /*
+ * Start the worker that decodes data changes applied while we're
+ * copying the table contents.
+ *
+ * Note that the worker has to wait for all transactions with XID
+ * already assigned to finish. If some of those transactions is
+ * waiting for a lock conflicting with ShareUpdateExclusiveLock on our
+ * table (e.g. it runs CREATE INDEX), we can end up in a deadlock.
+ * Not sure this risk is worth unlocking/locking the table (and its
+ * clustering index) and checking again if it's still eligible for
+ * REPACK CONCURRENTLY.
+ */
+ start_decoding_worker(tableOid);
+
+ /*
+ * Wait until the worker has the initial snapshot and retrieve it.
+ */
+ snapshot = get_initial_snapshot(decoding_worker);
- snapshot = SnapBuildInitialSnapshotForRepack(decoding_ctx->snapshot_builder);
PushActiveSnapshot(snapshot);
}
@@ -980,7 +1114,7 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
NewHeap = table_open(OIDNewHeap, NoLock);
/* Copy the heap data into the new table in the desired order */
- copy_table_data(NewHeap, OldHeap, index, snapshot, decoding_ctx, verbose,
+ copy_table_data(NewHeap, OldHeap, index, snapshot, verbose,
&swap_toast_by_content, &frozenXid, &cutoffMulti);
/* The historic snapshot won't be needed anymore. */
@@ -1001,14 +1135,11 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
if (index)
index_close(index, NoLock);
- rebuild_relation_finish_concurrent(NewHeap, OldHeap, decoding_ctx,
- frozenXid, cutoffMulti);
+ rebuild_relation_finish_concurrent(NewHeap, OldHeap, frozenXid,
+ cutoffMulti);
pgstat_progress_update_param(PROGRESS_REPACK_PHASE,
PROGRESS_REPACK_PHASE_FINAL_CLEANUP);
-
- /* Done with decoding. */
- cleanup_logical_decoding(decoding_ctx);
}
else
{
@@ -1180,8 +1311,7 @@ make_new_heap(Oid OIDOldHeap, Oid NewTableSpace, Oid NewAccessMethod,
*/
static void
copy_table_data(Relation NewHeap, Relation OldHeap, Relation OldIndex,
- Snapshot snapshot, LogicalDecodingContext *decoding_ctx,
- bool verbose, bool *pSwapToastByContent,
+ Snapshot snapshot, bool verbose, bool *pSwapToastByContent,
TransactionId *pFreezeXid, MultiXactId *pCutoffMulti)
{
Relation relRelation;
@@ -1342,7 +1472,6 @@ copy_table_data(Relation NewHeap, Relation OldHeap, Relation OldIndex,
*/
table_relation_copy_for_cluster(OldHeap, NewHeap, OldIndex, use_sort,
cutoffs.OldestXmin, snapshot,
- decoding_ctx,
&cutoffs.FreezeLimit,
&cutoffs.MultiXactCutoff,
&num_tuples, &tups_vacuumed,
@@ -2376,58 +2505,6 @@ RepackCommandAsString(RepackCommand cmd)
}
-/*
- * Call this function before REPACK CONCURRENTLY starts to setup logical
- * decoding. It makes sure that other users of the table put enough
- * information into WAL.
- *
- * The point is that at various places we expect that the table we're
- * processing is treated like a system catalog. For example, we need to be
- * able to scan it using a "historic snapshot" anytime during the processing
- * (as opposed to scanning only at the start point of the decoding, as logical
- * replication does during initial table synchronization), in order to apply
- * concurrent UPDATE / DELETE commands.
- *
- * Note that TOAST table needs no attention here as it's not scanned using
- * historic snapshot.
- */
-static void
-begin_concurrent_repack(Relation rel)
-{
- Oid toastrelid;
-
- /*
- * Avoid logical decoding of other relations by this backend. The lock we
- * have guarantees that the actual locator cannot be changed concurrently:
- * TRUNCATE needs AccessExclusiveLock.
- */
- Assert(CheckRelationLockedByMe(rel, ShareUpdateExclusiveLock, false));
- repacked_rel_locator = rel->rd_locator;
- toastrelid = rel->rd_rel->reltoastrelid;
- if (OidIsValid(toastrelid))
- {
- Relation toastrel;
-
- /* Avoid logical decoding of other TOAST relations. */
- toastrel = table_open(toastrelid, AccessShareLock);
- repacked_rel_toast_locator = toastrel->rd_locator;
- table_close(toastrel, AccessShareLock);
- }
-}
-
-/*
- * Call this when done with REPACK CONCURRENTLY.
- */
-static void
-end_concurrent_repack(void)
-{
- /*
- * Restore normal function of (future) logical decoding for this backend.
- */
- repacked_rel_locator.relNumber = InvalidOid;
- repacked_rel_toast_locator.relNumber = InvalidOid;
-}
-
/*
* Is this backend performing logical decoding on behalf of REPACK
* (CONCURRENTLY) ?
@@ -2492,9 +2569,10 @@ static LogicalDecodingContext *
setup_logical_decoding(Oid relid)
{
Relation rel;
- TupleDesc tupdesc;
+ Oid toastrelid;
LogicalDecodingContext *ctx;
- RepackDecodingState *dstate = palloc0_object(RepackDecodingState);
+ NameData slotname;
+ RepackDecodingState *dstate;
/*
* REPACK CONCURRENTLY is not allowed in a transaction block, so this
@@ -2502,20 +2580,22 @@ setup_logical_decoding(Oid relid)
*/
Assert(!TransactionIdIsValid(GetTopTransactionIdIfAny()));
- /*
- * A single backend should not execute multiple REPACK commands at a time,
- * so use PID to make the slot unique.
- */
- snprintf(NameStr(dstate->slotname), NAMEDATALEN, "repack_%d", MyProcPid);
-
/*
* Make sure we can use logical decoding.
*/
CheckSlotPermissions();
CheckLogicalDecodingRequirements();
- /* RS_TEMPORARY so that the slot gets cleaned up on ERROR. */
- ReplicationSlotCreate(NameStr(dstate->slotname), true, RS_TEMPORARY,
- false, false, false);
+
+ /*
+ * A single backend should not execute multiple REPACK commands at a time,
+ * so use PID to make the slot unique.
+ *
+ * RS_TEMPORARY so that the slot gets cleaned up on ERROR.
+ */
+ snprintf(NameStr(slotname), NAMEDATALEN, "repack_%d", MyProcPid);
+ ReplicationSlotCreate(NameStr(slotname), true, RS_TEMPORARY, false, false,
+ false);
+
EnsureLogicalDecodingEnabled();
/*
@@ -2538,104 +2618,109 @@ setup_logical_decoding(Oid relid)
DecodingContextFindStartpoint(ctx);
+ /*
+ * decode_concurrent_changes() needs non-blocking callback.
+ */
+ ctx->reader->routine.page_read = read_local_xlog_page_no_wait;
+
+ /*
+ * read_local_xlog_page_no_wait() needs to be able to indicate the end of
+ * WAL.
+ */
+ ctx->reader->private_data = MemoryContextAllocZero(ctx->context,
+ sizeof(ReadLocalXLogPageNoWaitPrivate));
+
+
/* Some WAL records should have been read. */
Assert(ctx->reader->EndRecPtr != InvalidXLogRecPtr);
+ /*
+ * Initialize repack_current_segment so that we can notice WAL segment
+ * boundaries.
+ */
XLByteToSeg(ctx->reader->EndRecPtr, repack_current_segment,
wal_segment_size);
- /*
- * Setup structures to store decoded changes.
- */
+ dstate = palloc0_object(RepackDecodingState);
dstate->relid = relid;
- dstate->tstore = tuplestore_begin_heap(false, false,
- maintenance_work_mem);
- /* Caller should already have the table locked. */
- rel = table_open(relid, NoLock);
- tupdesc = CreateTupleDescCopy(RelationGetDescr(rel));
- dstate->tupdesc = tupdesc;
- table_close(rel, NoLock);
+ /*
+ * Tuple descriptor may be needed to flatten a tuple before we write it to
+ * a file. A copy is needed because the decoding worker invalidates system
+ * caches before it starts to do the actual work.
+ */
+ rel = table_open(relid, AccessShareLock);
+ dstate->tupdesc = CreateTupleDescCopy(RelationGetDescr(rel));
- /* Initialize the descriptor to store the changes ... */
- dstate->tupdesc_change = CreateTemplateTupleDesc(1);
+ /* Avoid logical decoding of other relations. */
+ repacked_rel_locator = rel->rd_locator;
+ toastrelid = rel->rd_rel->reltoastrelid;
+ if (OidIsValid(toastrelid))
+ {
+ Relation toastrel;
- TupleDescInitEntry(dstate->tupdesc_change, 1, NULL, BYTEAOID, -1, 0);
- /* ... as well as the corresponding slot. */
- dstate->tsslot = MakeSingleTupleTableSlot(dstate->tupdesc_change,
- &TTSOpsMinimalTuple);
+ /* Avoid logical decoding of other TOAST relations. */
+ toastrel = table_open(toastrelid, AccessShareLock);
+ repacked_rel_toast_locator = toastrel->rd_locator;
+ table_close(toastrel, AccessShareLock);
+ }
+ table_close(rel, AccessShareLock);
- dstate->resowner = ResourceOwnerCreate(CurrentResourceOwner,
- "logical decoding");
+ /* The file will be set as soon as we have it opened. */
+ dstate->file = NULL;
ctx->output_writer_private = dstate;
+
return ctx;
}
/*
- * Retrieve tuple from ConcurrentChange structure.
+ * Decode logical changes from the WAL sequence and store them to a file.
*
- * The input data starts with the structure but it might not be appropriately
- * aligned.
- */
-static HeapTuple
-get_changed_tuple(char *change)
-{
- HeapTupleData tup_data;
- HeapTuple result;
- char *src;
-
- /*
- * Ensure alignment before accessing the fields. (This is why we can't use
- * heap_copytuple() instead of this function.)
- */
- src = change + offsetof(ConcurrentChange, tup_data);
- memcpy(&tup_data, src, sizeof(HeapTupleData));
-
- result = (HeapTuple) palloc(HEAPTUPLESIZE + tup_data.t_len);
- memcpy(result, &tup_data, sizeof(HeapTupleData));
- result->t_data = (HeapTupleHeader) ((char *) result + HEAPTUPLESIZE);
- src = change + SizeOfConcurrentChange;
- memcpy(result->t_data, src, result->t_len);
-
- return result;
-}
-
-/*
- * Decode logical changes from the WAL sequence up to end_of_wal.
+ * If true is returned, there is no more work for the worker.
*/
-void
-repack_decode_concurrent_changes(LogicalDecodingContext *ctx,
- XLogRecPtr end_of_wal)
+static bool
+decode_concurrent_changes(LogicalDecodingContext *ctx,
+ DecodingWorkerShared *shared)
{
RepackDecodingState *dstate;
- ResourceOwner resowner_old;
+ XLogRecPtr lsn_upto;
+ bool done;
+ char fname[MAXPGPATH];
dstate = (RepackDecodingState *) ctx->output_writer_private;
- resowner_old = CurrentResourceOwner;
- CurrentResourceOwner = dstate->resowner;
- PG_TRY();
+ /* Open the output file. */
+ DecodingWorkerFileName(fname, shared->relid, shared->last_exported + 1);
+ dstate->file = BufFileCreateFileSet(&shared->sfs.fs, fname);
+
+ SpinLockAcquire(&shared->mutex);
+ lsn_upto = shared->lsn_upto;
+ done = shared->done;
+ SpinLockRelease(&shared->mutex);
+
+ while (true)
{
- while (ctx->reader->EndRecPtr < end_of_wal)
- {
- XLogRecord *record;
- XLogSegNo segno_new;
- char *errm = NULL;
- XLogRecPtr end_lsn;
+ XLogRecord *record;
+ XLogSegNo segno_new;
+ char *errm = NULL;
+ XLogRecPtr end_lsn;
- record = XLogReadRecord(ctx->reader, &errm);
- if (errm)
- elog(ERROR, "%s", errm);
+ CHECK_FOR_INTERRUPTS();
- if (record != NULL)
- LogicalDecodingProcessRecord(ctx, ctx->reader);
+ record = XLogReadRecord(ctx->reader, &errm);
+ if (record)
+ {
+ LogicalDecodingProcessRecord(ctx, ctx->reader);
/*
* If WAL segment boundary has been crossed, inform the decoding
- * system that the catalog_xmin can advance. (We can confirm more
- * often, but a filling a single WAL segment should not take much
- * time.)
+ * system that the catalog_xmin can advance.
+ *
+ * TODO Does it make sense to confirm more often? Segment size
+ * seems appropriate for restart_lsn (because less than a segment
+ * cannot be recycled anyway), however more frequent checks might
+ * be beneficial for catalog_xmin.
*/
end_lsn = ctx->reader->EndRecPtr;
XLByteToSeg(end_lsn, segno_new, wal_segment_size);
@@ -2646,80 +2731,137 @@ repack_decode_concurrent_changes(LogicalDecodingContext *ctx,
(uint32) (end_lsn >> 32), (uint32) end_lsn);
repack_current_segment = segno_new;
}
+ }
+ else
+ {
+ ReadLocalXLogPageNoWaitPrivate *priv;
+
+ if (errm)
+ ereport(ERROR, (errmsg("%s", errm)));
- CHECK_FOR_INTERRUPTS();
+ /*
+ * In the decoding loop we do not want to get blocked when there
+ * is no more WAL available, otherwise the loop would become
+ * uninterruptible.
+ */
+ priv = (ReadLocalXLogPageNoWaitPrivate *)
+ ctx->reader->private_data;
+ if (priv->end_of_wal)
+ /* Do not miss the end of WAL condition next time. */
+ priv->end_of_wal = false;
+ else
+ ereport(ERROR, (errmsg("could not read WAL record")));
+ }
+
+ /*
+ * Whether we could read new record or not, keep checking if
+ * 'lsn_upto' was specified.
+ */
+ if (XLogRecPtrIsInvalid(lsn_upto))
+ {
+ SpinLockAcquire(&shared->mutex);
+ lsn_upto = shared->lsn_upto;
+ /* 'done' should be set at the same time as 'lsn_upto' */
+ done = shared->done;
+ SpinLockRelease(&shared->mutex);
+ }
+ if (!XLogRecPtrIsInvalid(lsn_upto) &&
+ ctx->reader->EndRecPtr >= lsn_upto)
+ break;
+
+ if (record == NULL)
+ {
+ int64 timeout = 0;
+ WaitLSNResult res;
+
+ /*
+ * Before we retry reading, wait until new WAL is flushed.
+ *
+ * There is a race condition such that the backend executing
+ * REPACK determines 'lsn_upto', but before it sets the shared
+ * variable, we reach the end of WAL. In that case we'd need to
+ * wait until the next WAL flush (unrelated to REPACK). Although
+ * that should not be a problem in a busy system, it might be
+ * noticeable in other cases, including regression tests (which
+ * are not necessarily executed in parallel). Therefore it makes
+ * sense to use timeout.
+ *
+ * If lsn_upto is valid, WAL records having LSN lower than that
+ * should already have been flushed to disk.
+ */
+ if (XLogRecPtrIsInvalid(lsn_upto))
+ timeout = 100L;
+ res = WaitForLSN(WAIT_LSN_TYPE_PRIMARY_FLUSH,
+ ctx->reader->EndRecPtr + 1,
+ timeout);
+ if (res != WAIT_LSN_RESULT_SUCCESS &&
+ res != WAIT_LSN_RESULT_TIMEOUT)
+ ereport(ERROR, (errmsg("waiting for WAL failed")));
}
- InvalidateSystemCaches();
- CurrentResourceOwner = resowner_old;
- }
- PG_CATCH();
- {
- /* clear all timetravel entries */
- InvalidateSystemCaches();
- CurrentResourceOwner = resowner_old;
- PG_RE_THROW();
}
- PG_END_TRY();
+
+ /*
+ * Close the file so we can make it available to the backend.
+ */
+ BufFileClose(dstate->file);
+ dstate->file = NULL;
+ SpinLockAcquire(&shared->mutex);
+ shared->lsn_upto = InvalidXLogRecPtr;
+ shared->last_exported++;
+ SpinLockRelease(&shared->mutex);
+ ConditionVariableSignal(&shared->cv);
+
+ return done;
}
/*
* Apply changes stored in 'file'.
*/
static void
-apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
+apply_concurrent_changes(BufFile *file, ChangeDest *dest)
{
+ char kind;
+ uint32 t_len;
Relation rel = dest->rel;
TupleTableSlot *index_slot,
*ident_slot;
HeapTuple tup_old = NULL;
- if (dstate->nchanges == 0)
- return;
-
/* TupleTableSlot is needed to pass the tuple to ExecInsertIndexTuples(). */
- index_slot = MakeSingleTupleTableSlot(dstate->tupdesc, &TTSOpsHeapTuple);
+ index_slot = MakeSingleTupleTableSlot(RelationGetDescr(rel),
+ &TTSOpsHeapTuple);
/* A slot to fetch tuples from identity index. */
ident_slot = table_slot_create(rel, NULL);
- while (tuplestore_gettupleslot(dstate->tstore, true, false,
- dstate->tsslot))
+ while (true)
{
- bool shouldFree;
- HeapTuple tup_change,
- tup,
+ size_t nread;
+ HeapTuple tup,
tup_exist;
- char *change_raw,
- *src;
- ConcurrentChange change;
- bool isnull[1];
- Datum values[1];
CHECK_FOR_INTERRUPTS();
- /* Get the change from the single-column tuple. */
- tup_change = ExecFetchSlotHeapTuple(dstate->tsslot, false, &shouldFree);
- heap_deform_tuple(tup_change, dstate->tupdesc_change, values, isnull);
- Assert(!isnull[0]);
-
- /* Make sure we access aligned data. */
- change_raw = (char *) DatumGetByteaP(values[0]);
- src = (char *) VARDATA(change_raw);
- memcpy(&change, src, SizeOfConcurrentChange);
+ nread = BufFileReadMaybeEOF(file, &kind, 1, true);
+ /* Are we done with the file? */
+ if (nread == 0)
+ break;
- /*
- * Extract the tuple from the change. The tuple is copied here because
- * it might be assigned to 'tup_old', in which case it needs to
- * survive into the next iteration.
- */
- tup = get_changed_tuple(src);
+ /* Read the tuple. */
+ BufFileReadExact(file, &t_len, sizeof(t_len));
+ tup = (HeapTuple) palloc(HEAPTUPLESIZE + t_len);
+ tup->t_data = (HeapTupleHeader) ((char *) tup + HEAPTUPLESIZE);
+ BufFileReadExact(file, tup->t_data, t_len);
+ tup->t_len = t_len;
+ ItemPointerSetInvalid(&tup->t_self);
+ tup->t_tableOid = RelationGetRelid(dest->rel);
- if (change.kind == CHANGE_UPDATE_OLD)
+ if (kind == CHANGE_UPDATE_OLD)
{
Assert(tup_old == NULL);
tup_old = tup;
}
- else if (change.kind == CHANGE_INSERT)
+ else if (kind == CHANGE_INSERT)
{
Assert(tup_old == NULL);
@@ -2727,12 +2869,11 @@ apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
pfree(tup);
}
- else if (change.kind == CHANGE_UPDATE_NEW ||
- change.kind == CHANGE_DELETE)
+ else if (kind == CHANGE_UPDATE_NEW || kind == CHANGE_DELETE)
{
HeapTuple tup_key;
- if (change.kind == CHANGE_UPDATE_NEW)
+ if (kind == CHANGE_UPDATE_NEW)
{
tup_key = tup_old != NULL ? tup_old : tup;
}
@@ -2749,7 +2890,7 @@ apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
if (tup_exist == NULL)
elog(ERROR, "failed to find target tuple");
- if (change.kind == CHANGE_UPDATE_NEW)
+ if (kind == CHANGE_UPDATE_NEW)
apply_concurrent_update(rel, tup, tup_exist, dest->iistate,
index_slot);
else
@@ -2764,26 +2905,19 @@ apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
pfree(tup);
}
else
- elog(ERROR, "unrecognized kind of change: %d", change.kind);
+ elog(ERROR, "unrecognized kind of change: %d", kind);
/*
* If a change was applied now, increment CID for next writes and
* update the snapshot so it sees the changes we've applied so far.
*/
- if (change.kind != CHANGE_UPDATE_OLD)
+ if (kind != CHANGE_UPDATE_OLD)
{
CommandCounterIncrement();
UpdateActiveSnapshotCommandId();
}
-
- /* TTSOpsMinimalTuple has .get_heap_tuple==NULL. */
- Assert(shouldFree);
- pfree(tup_change);
}
- tuplestore_clear(dstate->tstore);
- dstate->nchanges = 0;
-
/* Cleanup. */
ExecDropSingleTupleTableSlot(index_slot);
ExecDropSingleTupleTableSlot(ident_slot);
@@ -2958,25 +3092,59 @@ find_target_tuple(Relation rel, ChangeDest *dest, HeapTuple tup_key,
}
/*
- * Decode and apply concurrent changes.
+ * Decode and apply concurrent changes, up to (and including) the record whose
+ * LSN is 'end_of_wal'.
*/
static void
-process_concurrent_changes(LogicalDecodingContext *decoding_ctx,
- XLogRecPtr end_of_wal, ChangeDest *dest)
+process_concurrent_changes(XLogRecPtr end_of_wal, ChangeDest *dest, bool done)
{
- RepackDecodingState *dstate;
+ DecodingWorkerShared *shared;
+ char fname[MAXPGPATH];
+ BufFile *file;
pgstat_progress_update_param(PROGRESS_REPACK_PHASE,
PROGRESS_REPACK_PHASE_CATCH_UP);
- dstate = (RepackDecodingState *) decoding_ctx->output_writer_private;
+ /* Ask the worker for the file. */
+ shared = (DecodingWorkerShared *) dsm_segment_address(decoding_worker->seg);
+ SpinLockAcquire(&shared->mutex);
+ shared->lsn_upto = end_of_wal;
+ shared->done = done;
+ SpinLockRelease(&shared->mutex);
+
+ /*
+ * The worker needs to finish processing of the current WAL record. Even
+ * if it's idle, it'll need to close the output file. Thus we're likely to
+ * wait, so prepare for sleep.
+ */
+ ConditionVariablePrepareToSleep(&shared->cv);
+ for (;;)
+ {
+ int last_exported;
- repack_decode_concurrent_changes(decoding_ctx, end_of_wal);
+ SpinLockAcquire(&shared->mutex);
+ last_exported = shared->last_exported;
+ SpinLockRelease(&shared->mutex);
- if (dstate->nchanges == 0)
- return;
+ /*
+ * Has the worker exported the file we are waiting for?
+ */
+ if (last_exported == dest->file_seq)
+ break;
- apply_concurrent_changes(dstate, dest);
+ ConditionVariableSleep(&shared->cv, WAIT_EVENT_REPACK_WORKER_EXPORT);
+ }
+ ConditionVariableCancelSleep();
+
+ /* Open the file. */
+ DecodingWorkerFileName(fname, shared->relid, dest->file_seq);
+ file = BufFileOpenFileSet(&shared->sfs.fs, fname, O_RDONLY, false);
+ apply_concurrent_changes(file, dest);
+
+ BufFileClose(file);
+
+ /* Get ready for the next file. */
+ dest->file_seq++;
}
/*
@@ -3102,15 +3270,10 @@ cleanup_logical_decoding(LogicalDecodingContext *ctx)
dstate = (RepackDecodingState *) ctx->output_writer_private;
- ExecDropSingleTupleTableSlot(dstate->tsslot);
- FreeTupleDesc(dstate->tupdesc_change);
FreeTupleDesc(dstate->tupdesc);
- tuplestore_end(dstate->tstore);
-
FreeDecodingContext(ctx);
- ReplicationSlotRelease();
- ReplicationSlotDrop(NameStr(dstate->slotname), false);
+ ReplicationSlotDropAcquired();
pfree(dstate);
}
@@ -3124,7 +3287,6 @@ cleanup_logical_decoding(LogicalDecodingContext *ctx)
*/
static void
rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
- LogicalDecodingContext *decoding_ctx,
TransactionId frozenXid,
MultiXactId cutoffMulti)
{
@@ -3205,6 +3367,7 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
&chgdst.ident_index);
chgdst.ident_key = build_identity_key(ident_idx_new, OldHeap,
&chgdst.ident_key_nentries);
+ chgdst.file_seq = WORKER_FILE_SNAPSHOT + 1;
/*
* During testing, wait for another backend to perform concurrent data
@@ -3226,7 +3389,7 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
* hold AccessExclusiveLock. (Quite some amount of WAL could have been
* written during the data copying and index creation.)
*/
- process_concurrent_changes(decoding_ctx, end_of_wal, &chgdst);
+ process_concurrent_changes(end_of_wal, &chgdst, false);
/*
* Acquire AccessExclusiveLock on the table, its TOAST relation (if there
@@ -3307,8 +3470,11 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
XLogFlush(wal_insert_ptr);
end_of_wal = GetFlushRecPtr(NULL);
- /* Apply the concurrent changes again. */
- process_concurrent_changes(decoding_ctx, end_of_wal, &chgdst);
+ /*
+ * Apply the concurrent changes again. Indicate that the decoding worker
+ * won't be needed anymore.
+ */
+ process_concurrent_changes(end_of_wal, &chgdst, true);
/* Remember info about rel before closing OldHeap */
relpersistence = OldHeap->rd_rel->relpersistence;
@@ -3420,3 +3586,510 @@ build_new_indexes(Relation NewHeap, Relation OldHeap, List *OldIndexes)
return result;
}
+
+/*
+ * Try to start a background worker to perform logical decoding of data
+ * changes applied to relation while REPACK CONCURRENTLY is copying its
+ * contents to a new table.
+ */
+static void
+start_decoding_worker(Oid relid)
+{
+ Size size;
+ dsm_segment *seg;
+ DecodingWorkerShared *shared;
+ shm_mq *mq;
+ shm_mq_handle *mqh;
+ BackgroundWorker bgw;
+
+ /* Setup shared memory. */
+ size = BUFFERALIGN(offsetof(DecodingWorkerShared, error_queue)) +
+ BUFFERALIGN(REPACK_ERROR_QUEUE_SIZE);
+ seg = dsm_create(size, 0);
+ shared = (DecodingWorkerShared *) dsm_segment_address(seg);
+ shared->lsn_upto = InvalidXLogRecPtr;
+ shared->done = false;
+ SharedFileSetInit(&shared->sfs, seg);
+ shared->last_exported = -1;
+ SpinLockInit(&shared->mutex);
+ shared->dbid = MyDatabaseId;
+
+ /*
+ * This is the UserId set in cluster_rel(). Security context shouldn't be
+ * needed for decoding worker.
+ */
+ shared->roleid = GetUserId();
+ shared->relid = relid;
+ ConditionVariableInit(&shared->cv);
+ shared->backend_proc = MyProc;
+ shared->backend_pid = MyProcPid;
+ shared->backend_proc_number = MyProcNumber;
+
+ mq = shm_mq_create((char *) BUFFERALIGN(shared->error_queue),
+ REPACK_ERROR_QUEUE_SIZE);
+ shm_mq_set_receiver(mq, MyProc);
+ mqh = shm_mq_attach(mq, seg, NULL);
+
+ memset(&bgw, 0, sizeof(bgw));
+ snprintf(bgw.bgw_name, BGW_MAXLEN,
+ "REPACK decoding worker for relation \"%s\"",
+ get_rel_name(relid));
+ snprintf(bgw.bgw_type, BGW_MAXLEN, "REPACK decoding worker");
+ bgw.bgw_flags = BGWORKER_SHMEM_ACCESS |
+ BGWORKER_BACKEND_DATABASE_CONNECTION;
+ bgw.bgw_start_time = BgWorkerStart_RecoveryFinished;
+ bgw.bgw_restart_time = BGW_NEVER_RESTART;
+ snprintf(bgw.bgw_library_name, MAXPGPATH, "postgres");
+ snprintf(bgw.bgw_function_name, BGW_MAXLEN, "RepackWorkerMain");
+ bgw.bgw_main_arg = UInt32GetDatum(dsm_segment_handle(seg));
+ bgw.bgw_notify_pid = MyProcPid;
+
+ decoding_worker = palloc0_object(DecodingWorker);
+ if (!RegisterDynamicBackgroundWorker(&bgw, &decoding_worker->handle))
+ ereport(ERROR,
+ (errcode(ERRCODE_CONFIGURATION_LIMIT_EXCEEDED),
+ errmsg("out of background worker slots"),
+ errhint("You might need to increase \"%s\".", "max_worker_processes")));
+
+ decoding_worker->seg = seg;
+ decoding_worker->error_mqh = mqh;
+
+ /*
+ * The decoding setup must be done before the caller can have XID assigned
+ * for any reason, otherwise the worker might end up in a deadlock,
+ * waiting for the caller's transaction to end. Therefore wait here until
+ * the worker indicates that it has the logical decoding initialized.
+ */
+ ConditionVariablePrepareToSleep(&shared->cv);
+ for (;;)
+ {
+ bool initialized;
+
+ SpinLockAcquire(&shared->mutex);
+ initialized = shared->initialized;
+ SpinLockRelease(&shared->mutex);
+
+ if (initialized)
+ break;
+
+ ConditionVariableSleep(&shared->cv, WAIT_EVENT_REPACK_WORKER_EXPORT);
+ }
+ ConditionVariableCancelSleep();
+}
+
+/*
+ * Stop the decoding worker and cleanup the related resources.
+ *
+ * The worker stops on its own when it knows there is no more work to do, but
+ * we need to stop it explicitly at least on ERROR in the launching backend.
+ */
+static void
+stop_decoding_worker(void)
+{
+ BgwHandleStatus status;
+
+ /* Haven't reached the worker startup? */
+ if (decoding_worker == NULL)
+ return;
+
+ /* Could not register the worker? */
+ if (decoding_worker->handle == NULL)
+ return;
+
+ TerminateBackgroundWorker(decoding_worker->handle);
+ /* The worker should really exit before the REPACK command does. */
+ HOLD_INTERRUPTS();
+ status = WaitForBackgroundWorkerShutdown(decoding_worker->handle);
+ RESUME_INTERRUPTS();
+
+ if (status == BGWH_POSTMASTER_DIED)
+ ereport(FATAL,
+ (errcode(ERRCODE_ADMIN_SHUTDOWN),
+ errmsg("postmaster exited during REPACK command")));
+
+ shm_mq_detach(decoding_worker->error_mqh);
+
+ /*
+ * If we could not cancel the current sleep due to ERROR, do that before
+ * we detach from the shared memory the condition variable is located in.
+ * If we did not, the bgworker ERROR handling code would try and fail
+ * badly.
+ */
+ ConditionVariableCancelSleep();
+
+ dsm_detach(decoding_worker->seg);
+ pfree(decoding_worker);
+ decoding_worker = NULL;
+}
+
+/* Is this process a REPACK worker? */
+static bool is_repack_worker = false;
+
+static pid_t backend_pid;
+static ProcNumber backend_proc_number;
+
+/*
+ * See ParallelWorkerShutdown for details.
+ */
+static void
+RepackWorkerShutdown(int code, Datum arg)
+{
+ SendProcSignal(backend_pid,
+ PROCSIG_REPACK_MESSAGE,
+ backend_proc_number);
+
+ dsm_detach((dsm_segment *) DatumGetPointer(arg));
+}
+
+/* REPACK decoding worker entry point */
+void
+RepackWorkerMain(Datum main_arg)
+{
+ dsm_segment *seg;
+ DecodingWorkerShared *shared;
+ shm_mq *mq;
+ shm_mq_handle *mqh;
+
+ is_repack_worker = true;
+
+ /*
+ * Override the default bgworker_die() with die() so we can use
+ * CHECK_FOR_INTERRUPTS().
+ */
+ pqsignal(SIGTERM, die);
+ BackgroundWorkerUnblockSignals();
+
+ seg = dsm_attach(DatumGetUInt32(main_arg));
+ if (seg == NULL)
+ ereport(ERROR,
+ (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
+ errmsg("could not map dynamic shared memory segment")));
+
+ shared = (DecodingWorkerShared *) dsm_segment_address(seg);
+
+ /* Arrange to signal the leader if we exit. */
+ backend_pid = shared->backend_pid;
+ backend_proc_number = shared->backend_proc_number;
+ before_shmem_exit(RepackWorkerShutdown, PointerGetDatum(seg));
+
+ /*
+ * Join locking group - see the comments around the call of
+ * start_decoding_worker().
+ */
+ if (!BecomeLockGroupMember(shared->backend_proc, backend_pid))
+ /* The leader is not running anymore. */
+ return;
+
+ /*
+ * Setup a queue to send error messages to the backend that launched this
+ * worker.
+ */
+ mq = (shm_mq *) (char *) BUFFERALIGN(shared->error_queue);
+ shm_mq_set_sender(mq, MyProc);
+ mqh = shm_mq_attach(mq, seg, NULL);
+ pq_redirect_to_shm_mq(seg, mqh);
+ pq_set_parallel_leader(shared->backend_pid,
+ shared->backend_proc_number);
+
+ /* Connect to the database. */
+ BackgroundWorkerInitializeConnectionByOid(shared->dbid, shared->roleid, 0);
+
+ repack_worker_internal(seg);
+}
+
+static void
+repack_worker_internal(dsm_segment *seg)
+{
+ DecodingWorkerShared *shared;
+ LogicalDecodingContext *decoding_ctx;
+ SharedFileSet *sfs;
+ Snapshot snapshot;
+
+ /*
+ * Transaction is needed to open relation, and it also provides us with a
+ * resource owner.
+ */
+ StartTransactionCommand();
+
+ shared = (DecodingWorkerShared *) dsm_segment_address(seg);
+
+ /*
+ * Not sure the spinlock is needed here - the backend should not change
+ * anything in the shared memory until we have serialized the snapshot.
+ */
+ SpinLockAcquire(&shared->mutex);
+ Assert(XLogRecPtrIsInvalid(shared->lsn_upto));
+ sfs = &shared->sfs;
+ SpinLockRelease(&shared->mutex);
+
+ SharedFileSetAttach(sfs, seg);
+
+ /*
+ * Prepare to capture the concurrent data changes ourselves.
+ */
+ decoding_ctx = setup_logical_decoding(shared->relid);
+
+ /* Announce that we're ready. */
+ SpinLockAcquire(&shared->mutex);
+ shared->initialized = true;
+ SpinLockRelease(&shared->mutex);
+ ConditionVariableSignal(&shared->cv);
+
+ /* Build the initial snapshot and export it. */
+ snapshot = SnapBuildInitialSnapshotForRepack(decoding_ctx->snapshot_builder);
+ export_initial_snapshot(snapshot, shared);
+
+ /*
+ * Only historic snapshots should be used now. Do not let us restrict the
+ * progress of xmin horizon.
+ */
+ InvalidateCatalogSnapshot();
+
+ while (!decode_concurrent_changes(decoding_ctx, shared))
+ ;
+
+ /* Cleanup. */
+ cleanup_logical_decoding(decoding_ctx);
+ CommitTransactionCommand();
+}
+
+/*
+ * Make snapshot available to the backend that launched the decoding worker.
+ */
+static void
+export_initial_snapshot(Snapshot snapshot, DecodingWorkerShared *shared)
+{
+ char fname[MAXPGPATH];
+ BufFile *file;
+ Size snap_size;
+ char *snap_space;
+
+ snap_size = EstimateSnapshotSpace(snapshot);
+ snap_space = (char *) palloc(snap_size);
+ SerializeSnapshot(snapshot, snap_space);
+ FreeSnapshot(snapshot);
+
+ DecodingWorkerFileName(fname, shared->relid, shared->last_exported + 1);
+ file = BufFileCreateFileSet(&shared->sfs.fs, fname);
+ /* To make restoration easier, write the snapshot size first. */
+ BufFileWrite(file, &snap_size, sizeof(snap_size));
+ BufFileWrite(file, snap_space, snap_size);
+ pfree(snap_space);
+ BufFileClose(file);
+
+ /* Increase the counter to tell the backend that the file is available. */
+ SpinLockAcquire(&shared->mutex);
+ shared->last_exported++;
+ SpinLockRelease(&shared->mutex);
+ ConditionVariableSignal(&shared->cv);
+}
+
+/*
+ * Get the initial snapshot from the decoding worker.
+ */
+static Snapshot
+get_initial_snapshot(DecodingWorker *worker)
+{
+ DecodingWorkerShared *shared;
+ char fname[MAXPGPATH];
+ BufFile *file;
+ Size snap_size;
+ char *snap_space;
+ Snapshot snapshot;
+
+ shared = (DecodingWorkerShared *) dsm_segment_address(worker->seg);
+
+ /*
+ * The worker needs to initialize the logical decoding, which usually
+ * takes some time. Therefore it makes sense to prepare for the sleep
+ * first.
+ */
+ ConditionVariablePrepareToSleep(&shared->cv);
+ for (;;)
+ {
+ int last_exported;
+
+ SpinLockAcquire(&shared->mutex);
+ last_exported = shared->last_exported;
+ SpinLockRelease(&shared->mutex);
+
+ /*
+ * Has the worker exported the file we are waiting for?
+ */
+ if (last_exported == WORKER_FILE_SNAPSHOT)
+ break;
+
+ ConditionVariableSleep(&shared->cv, WAIT_EVENT_REPACK_WORKER_EXPORT);
+ }
+ ConditionVariableCancelSleep();
+
+ /* Read the snapshot from a file. */
+ DecodingWorkerFileName(fname, shared->relid, WORKER_FILE_SNAPSHOT);
+ file = BufFileOpenFileSet(&shared->sfs.fs, fname, O_RDONLY, false);
+ BufFileReadExact(file, &snap_size, sizeof(snap_size));
+ snap_space = (char *) palloc(snap_size);
+ BufFileReadExact(file, snap_space, snap_size);
+ BufFileClose(file);
+
+ /* Restore it. */
+ snapshot = RestoreSnapshot(snap_space);
+ pfree(snap_space);
+
+ return snapshot;
+}
+
+bool
+IsRepackWorker(void)
+{
+ return is_repack_worker;
+}
+
+/*
+ * Handle receipt of an interrupt indicating a repack worker message.
+ *
+ * Note: this is called within a signal handler! All we can do is set
+ * a flag that will cause the next CHECK_FOR_INTERRUPTS() to invoke
+ * ProcessRepackMessages().
+ */
+void
+HandleRepackMessageInterrupt(void)
+{
+ InterruptPending = true;
+ RepackMessagePending = true;
+ SetLatch(MyLatch);
+}
+
+/*
+ * Process any queued protocol messages received from parallel workers.
+ */
+void
+ProcessRepackMessages(void)
+{
+ MemoryContext oldcontext;
+
+ static MemoryContext hpm_context = NULL;
+
+ /*
+ * Nothing to do if we haven't launched the worker yet or have already
+ * terminated it.
+ */
+ if (decoding_worker == NULL)
+ return;
+
+ /*
+ * This is invoked from ProcessInterrupts(), and since some of the
+ * functions it calls contain CHECK_FOR_INTERRUPTS(), there is a potential
+ * for recursive calls if more signals are received while this runs. It's
+ * unclear that recursive entry would be safe, and it doesn't seem useful
+ * even if it is safe, so let's block interrupts until done.
+ */
+ HOLD_INTERRUPTS();
+
+ /*
+ * Moreover, CurrentMemoryContext might be pointing almost anywhere. We
+ * don't want to risk leaking data into long-lived contexts, so let's do
+ * our work here in a private context that we can reset on each use.
+ */
+ if (hpm_context == NULL) /* first time through? */
+ hpm_context = AllocSetContextCreate(TopMemoryContext,
+ "ProcessRepackMessages",
+ ALLOCSET_DEFAULT_SIZES);
+ else
+ MemoryContextReset(hpm_context);
+
+ oldcontext = MemoryContextSwitchTo(hpm_context);
+
+ /* OK to process messages. Reset the flag saying there are more to do. */
+ RepackMessagePending = false;
+
+ /*
+ * Read as many messages as we can from each worker, but stop when no more
+ * messages can be read from the worker without blocking.
+ */
+ while (true)
+ {
+ shm_mq_result res;
+ Size nbytes;
+ void *data;
+
+ res = shm_mq_receive(decoding_worker->error_mqh, &nbytes,
+ &data, true);
+ if (res == SHM_MQ_WOULD_BLOCK)
+ break;
+ else if (res == SHM_MQ_SUCCESS)
+ {
+ StringInfoData msg;
+
+ initStringInfo(&msg);
+ appendBinaryStringInfo(&msg, data, nbytes);
+ ProcessRepackMessage(&msg);
+ pfree(msg.data);
+ }
+ else
+ {
+ /*
+ * The decoding worker is special in that it exits as soon as it
+ * has its work done. Thus the DETACHED result code is fine.
+ */
+ Assert(res == SHM_MQ_DETACHED);
+
+ break;
+ }
+ }
+
+ MemoryContextSwitchTo(oldcontext);
+
+ /* Might as well clear the context on our way out */
+ MemoryContextReset(hpm_context);
+
+ RESUME_INTERRUPTS();
+}
+
+/*
+ * Process a single protocol message received from a single parallel worker.
+ */
+static void
+ProcessRepackMessage(StringInfo msg)
+{
+ char msgtype;
+
+ msgtype = pq_getmsgbyte(msg);
+
+ switch (msgtype)
+ {
+ case PqMsg_ErrorResponse:
+ case PqMsg_NoticeResponse:
+ {
+ ErrorData edata;
+
+ /* Parse ErrorResponse or NoticeResponse. */
+ pq_parse_errornotice(msg, &edata);
+
+ /* Death of a worker isn't enough justification for suicide. */
+ edata.elevel = Min(edata.elevel, ERROR);
+
+ /*
+ * If desired, add a context line to show that this is a
+ * message propagated from a parallel worker. Otherwise, it
+ * can sometimes be confusing to understand what actually
+ * happened.
+ */
+ if (edata.context)
+ edata.context = psprintf("%s\n%s", edata.context,
+ _("decoding worker"));
+ else
+ edata.context = pstrdup(_("decoding worker"));
+
+ /* Rethrow error or print notice. */
+ ThrowErrorData(&edata);
+
+ break;
+ }
+
+ default:
+ {
+ elog(ERROR, "unrecognized message type received from decoding worker: %c (message length %d bytes)",
+ msgtype, msg->len);
+ }
+ }
+}
diff --git a/src/backend/libpq/pqmq.c b/src/backend/libpq/pqmq.c
index 6e4bbfb5aa1..42f6fa472c5 100644
--- a/src/backend/libpq/pqmq.c
+++ b/src/backend/libpq/pqmq.c
@@ -14,6 +14,7 @@
#include "postgres.h"
#include "access/parallel.h"
+#include "commands/cluster.h"
#include "libpq/libpq.h"
#include "libpq/pqformat.h"
#include "libpq/pqmq.h"
@@ -175,6 +176,10 @@ mq_putmessage(char msgtype, const char *s, size_t len)
SendProcSignal(pq_mq_parallel_leader_pid,
PROCSIG_PARALLEL_APPLY_MESSAGE,
pq_mq_parallel_leader_proc_number);
+ else if (IsRepackWorker())
+ SendProcSignal(pq_mq_parallel_leader_pid,
+ PROCSIG_REPACK_MESSAGE,
+ pq_mq_parallel_leader_proc_number);
else
{
Assert(IsParallelWorker());
diff --git a/src/backend/postmaster/bgworker.c b/src/backend/postmaster/bgworker.c
index 261ccd3f59c..09c58371e8e 100644
--- a/src/backend/postmaster/bgworker.c
+++ b/src/backend/postmaster/bgworker.c
@@ -13,6 +13,7 @@
#include "postgres.h"
#include "access/parallel.h"
+#include "commands/cluster.h"
#include "libpq/pqsignal.h"
#include "miscadmin.h"
#include "pgstat.h"
@@ -142,6 +143,10 @@ static const struct
{
.fn_name = "SequenceSyncWorkerMain",
.fn_addr = SequenceSyncWorkerMain
+ },
+ {
+ .fn_name = "RepackWorkerMain",
+ .fn_addr = RepackWorkerMain
}
};
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index 603a2b94d05..7651b187418 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -194,7 +194,11 @@ StartupDecodingContext(List *output_plugin_options,
ctx->slot = slot;
- ctx->reader = XLogReaderAllocate(wal_segment_size, NULL, xl_routine, ctx);
+ /*
+ * TODO A separate patch for PG core, unless there's really a reason to
+ * pass ctx for private_data (May extensions expect ctx?).
+ */
+ ctx->reader = XLogReaderAllocate(wal_segment_size, NULL, xl_routine, NULL);
if (!ctx->reader)
ereport(ERROR,
(errcode(ERRCODE_OUT_OF_MEMORY),
diff --git a/src/backend/replication/pgoutput_repack/pgoutput_repack.c b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
index 601016716b3..d0c464a98d5 100644
--- a/src/backend/replication/pgoutput_repack/pgoutput_repack.c
+++ b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
@@ -168,17 +168,13 @@ store_change(LogicalDecodingContext *ctx, ConcurrentChangeKind kind,
HeapTuple tuple)
{
RepackDecodingState *dstate;
- char *change_raw;
- ConcurrentChange change;
+ char kind_byte = (char) kind;
bool flattened = false;
- Size size;
- Datum values[1];
- bool isnull[1];
- char *dst;
dstate = (RepackDecodingState *) ctx->output_writer_private;
- size = VARHDRSZ + SizeOfConcurrentChange;
+ /* Store the change kind. */
+ BufFileWrite(dstate->file, &kind_byte, 1);
/*
* ReorderBufferCommit() stores the TOAST chunks in its private memory
@@ -195,46 +191,12 @@ store_change(LogicalDecodingContext *ctx, ConcurrentChangeKind kind,
tuple = toast_flatten_tuple(tuple, dstate->tupdesc);
flattened = true;
}
+ /* Store the tuple size ... */
+ BufFileWrite(dstate->file, &tuple->t_len, sizeof(tuple->t_len));
+ /* ... and the tuple itself. */
+ BufFileWrite(dstate->file, tuple->t_data, tuple->t_len);
- size += tuple->t_len;
- if (size >= MaxAllocSize)
- elog(ERROR, "Change is too big.");
-
- /* Construct the change. */
- change_raw = (char *) palloc0(size);
- SET_VARSIZE(change_raw, size);
-
- /*
- * Since the varlena alignment might not be sufficient for the structure,
- * set the fields in a local instance and remember where it should
- * eventually be copied.
- */
- change.kind = kind;
- dst = (char *) VARDATA(change_raw);
-
- /*
- * Copy the tuple.
- *
- * Note: change->tup_data.t_data must be fixed on retrieval!
- */
- memcpy(&change.tup_data, tuple, sizeof(HeapTupleData));
- memcpy(dst, &change, SizeOfConcurrentChange);
- dst += SizeOfConcurrentChange;
- memcpy(dst, tuple->t_data, tuple->t_len);
-
- /* The data has been copied. */
+ /* Free the flat copy if created above. */
if (flattened)
pfree(tuple);
-
- /* Store as tuple of 1 bytea column. */
- values[0] = PointerGetDatum(change_raw);
- isnull[0] = false;
- tuplestore_putvalues(dstate->tstore, dstate->tupdesc_change,
- values, isnull);
-
- /* Accounting. */
- dstate->nchanges++;
-
- /* Cleanup. */
- pfree(change_raw);
}
diff --git a/src/backend/storage/ipc/procsignal.c b/src/backend/storage/ipc/procsignal.c
index 5d33559926a..080da1ee490 100644
--- a/src/backend/storage/ipc/procsignal.c
+++ b/src/backend/storage/ipc/procsignal.c
@@ -19,6 +19,7 @@
#include "access/parallel.h"
#include "commands/async.h"
+#include "commands/cluster.h"
#include "miscadmin.h"
#include "pgstat.h"
#include "port/pg_bitutils.h"
@@ -697,6 +698,9 @@ procsignal_sigusr1_handler(SIGNAL_ARGS)
if (CheckProcSignal(PROCSIG_PARALLEL_APPLY_MESSAGE))
HandleParallelApplyMessageInterrupt();
+ if (CheckProcSignal(PROCSIG_REPACK_MESSAGE))
+ HandleRepackMessageInterrupt();
+
if (CheckProcSignal(PROCSIG_RECOVERY_CONFLICT))
HandleRecoveryConflictInterrupt();
diff --git a/src/backend/tcop/postgres.c b/src/backend/tcop/postgres.c
index 21de158adbb..2f667200d0c 100644
--- a/src/backend/tcop/postgres.c
+++ b/src/backend/tcop/postgres.c
@@ -36,6 +36,7 @@
#include "access/xact.h"
#include "catalog/pg_type.h"
#include "commands/async.h"
+#include "commands/cluster.h"
#include "commands/event_trigger.h"
#include "commands/explain_state.h"
#include "commands/prepare.h"
@@ -3575,6 +3576,9 @@ ProcessInterrupts(void)
if (ParallelApplyMessagePending)
ProcessParallelApplyMessages();
+
+ if (RepackMessagePending)
+ ProcessRepackMessages();
}
/*
diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt
index 4aa864fe3c3..b00bd794759 100644
--- a/src/backend/utils/activity/wait_event_names.txt
+++ b/src/backend/utils/activity/wait_event_names.txt
@@ -154,6 +154,7 @@ RECOVERY_CONFLICT_SNAPSHOT "Waiting for recovery conflict resolution for a vacuu
RECOVERY_CONFLICT_TABLESPACE "Waiting for recovery conflict resolution for dropping a tablespace."
RECOVERY_END_COMMAND "Waiting for <xref linkend="guc-recovery-end-command"/> to complete."
RECOVERY_PAUSE "Waiting for recovery to be resumed."
+REPACK_WORKER_EXPORT "Waiting for decoding worker to export a new output file."
REPLICATION_ORIGIN_DROP "Waiting for a replication origin to become inactive so it can be dropped."
REPLICATION_SLOT_DROP "Waiting for a replication slot to become inactive so it can be dropped."
RESTORE_COMMAND "Waiting for <xref linkend="guc-restore-command"/> to complete."
diff --git a/src/include/access/tableam.h b/src/include/access/tableam.h
index 1776d0dafdf..a9874cb1309 100644
--- a/src/include/access/tableam.h
+++ b/src/include/access/tableam.h
@@ -22,7 +22,6 @@
#include "access/xact.h"
#include "commands/vacuum.h"
#include "executor/tuptable.h"
-#include "replication/logical.h"
#include "storage/read_stream.h"
#include "utils/rel.h"
#include "utils/snapshot.h"
@@ -631,7 +630,6 @@ typedef struct TableAmRoutine
bool use_sort,
TransactionId OldestXmin,
Snapshot snapshot,
- LogicalDecodingContext *decoding_ctx,
TransactionId *xid_cutoff,
MultiXactId *multi_cutoff,
double *num_tuples,
@@ -1663,8 +1661,6 @@ table_relation_copy_data(Relation rel, const RelFileLocator *newrlocator)
* - *multi_cutoff - ditto
* - snapshot - if != NULL, ignore data changes done by transactions that this
* (MVCC) snapshot considers still in-progress or in the future.
- * - decoding_ctx - logical decoding context, to capture concurrent data
- * changes.
*
* Output parameters:
* - *xid_cutoff - rel's new relfrozenxid value, may be invalid
@@ -1678,7 +1674,6 @@ table_relation_copy_for_cluster(Relation OldTable, Relation NewTable,
bool use_sort,
TransactionId OldestXmin,
Snapshot snapshot,
- LogicalDecodingContext *decoding_ctx,
TransactionId *xid_cutoff,
MultiXactId *multi_cutoff,
double *num_tuples,
@@ -1687,7 +1682,7 @@ table_relation_copy_for_cluster(Relation OldTable, Relation NewTable,
{
OldTable->rd_tableam->relation_copy_for_cluster(OldTable, NewTable, OldIndex,
use_sort, OldestXmin,
- snapshot, decoding_ctx,
+ snapshot,
xid_cutoff, multi_cutoff,
num_tuples, tups_vacuumed,
tups_recently_dead);
diff --git a/src/include/commands/cluster.h b/src/include/commands/cluster.h
index 6a5c476294a..1b05d5d418b 100644
--- a/src/include/commands/cluster.h
+++ b/src/include/commands/cluster.h
@@ -17,11 +17,13 @@
#include "nodes/parsenodes.h"
#include "parser/parse_node.h"
#include "replication/decode.h"
+#include "postmaster/bgworker.h"
#include "replication/logical.h"
+#include "storage/buffile.h"
#include "storage/lock.h"
+#include "storage/shm_mq.h"
#include "utils/relcache.h"
#include "utils/resowner.h"
-#include "utils/tuplestore.h"
/* flag bits for ClusterParams->options */
@@ -44,6 +46,9 @@ typedef struct ClusterParams
* The following definitions are used by REPACK CONCURRENTLY.
*/
+/*
+ * Stored as a single byte in the output file.
+ */
typedef enum
{
CHANGE_INSERT,
@@ -52,68 +57,30 @@ typedef enum
CHANGE_DELETE
} ConcurrentChangeKind;
-typedef struct ConcurrentChange
-{
- /* See the enum above. */
- ConcurrentChangeKind kind;
-
- /*
- * The actual tuple.
- *
- * The tuple data follows the ConcurrentChange structure. Before use make
- * sure the tuple is correctly aligned (ConcurrentChange can be stored as
- * bytea) and that tuple->t_data is fixed.
- */
- HeapTupleData tup_data;
-} ConcurrentChange;
-
-#define SizeOfConcurrentChange (offsetof(ConcurrentChange, tup_data) + \
- sizeof(HeapTupleData))
-
/*
* Logical decoding state.
*
- * Here we store the data changes that we decode from WAL while the table
- * contents is being copied to a new storage. Also the necessary metadata
- * needed to apply these changes to the table is stored here.
+ * The output plugin uses it to store the data changes that it decodes from
+ * WAL while the table contents is being copied to a new storage.
*/
typedef struct RepackDecodingState
{
/* The relation whose changes we're decoding. */
Oid relid;
- /* Replication slot name. */
- NameData slotname;
-
- /*
- * Decoded changes are stored here. Although we try to avoid excessive
- * batches, it can happen that the changes need to be stored to disk. The
- * tuplestore does this transparently.
- */
- Tuplestorestate *tstore;
-
- /* The current number of changes in tstore. */
- double nchanges;
-
- /*
- * Descriptor to store the ConcurrentChange structure serialized (bytea).
- * We can't store the tuple directly because tuplestore only supports
- * minimum tuple and we may need to transfer OID system column from the
- * output plugin. Also we need to transfer the change kind, so it's better
- * to put everything in the structure than to use 2 tuplestores "in
- * parallel".
- */
- TupleDesc tupdesc_change;
-
- /* Tuple descriptor needed to update indexes. */
+ /* Tuple descriptor of the relation being processed. */
TupleDesc tupdesc;
- /* Slot to retrieve data from tstore. */
- TupleTableSlot *tsslot;
-
- ResourceOwner resowner;
+ /* The current output file. */
+ BufFile *file;
} RepackDecodingState;
+extern PGDLLIMPORT volatile sig_atomic_t RepackMessagePending;
+
+extern bool IsRepackWorker(void);
+extern void HandleRepackMessageInterrupt(void);
+extern void ProcessRepackMessages(void);
+
extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel);
extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid,
@@ -136,6 +103,6 @@ extern void finish_heap_swap(Oid OIDOldHeap, Oid OIDNewHeap,
extern bool am_decoding_for_repack(void);
extern bool change_useless_for_repack(XLogRecordBuffer *buf);
-extern void repack_decode_concurrent_changes(LogicalDecodingContext *ctx,
- XLogRecPtr end_of_wal);
+
+extern void RepackWorkerMain(Datum main_arg);
#endif /* CLUSTER_H */
diff --git a/src/include/storage/procsignal.h b/src/include/storage/procsignal.h
index 348fba53a93..a944ee0d211 100644
--- a/src/include/storage/procsignal.h
+++ b/src/include/storage/procsignal.h
@@ -36,6 +36,7 @@ typedef enum
PROCSIG_BARRIER, /* global barrier interrupt */
PROCSIG_LOG_MEMORY_CONTEXT, /* ask backend to log the memory contexts */
PROCSIG_PARALLEL_APPLY_MESSAGE, /* Message from parallel apply workers */
+ PROCSIG_REPACK_MESSAGE, /* Message from repack worker */
PROCSIG_RECOVERY_CONFLICT, /* backend is blocking recovery, check
* PGPROC->pendingRecoveryConflicts for the
* reason */
diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list
index 71a0802187e..b35d4e318fc 100644
--- a/src/tools/pgindent/typedefs.list
+++ b/src/tools/pgindent/typedefs.list
@@ -499,7 +499,6 @@ CompressFileHandle
CompressionLocation
CompressorState
ComputeXidHorizonsResult
-ConcurrentChange
ConcurrentChangeKind
ConditionVariable
ConditionVariableMinimallyPadded
@@ -639,6 +638,8 @@ DeclareCursorStmt
DecodedBkpBlock
DecodedXLogRecord
DecodingOutputState
+DecodingWorker
+DecodingWorkerShared
DefElem
DefElemAction
DefaultACLInfo
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 55+ messages in thread
* [PATCH 5/5] Use background worker to do logical decoding.
@ 2026-02-16 19:49 Antonin Houska <[email protected]>
0 siblings, 0 replies; 55+ messages in thread
From: Antonin Houska @ 2026-02-16 19:49 UTC (permalink / raw)
If the backend performing REPACK (CONCURRENTLY) does both data copying and
logical decoding, it has to "travel in time" back and forth and therefore it
has to invalidate system caches quite a few times. (The copying and the
decoding work with different catalog snapshots.) As the decoding worker has
separate caches, the switching is not necessary.
Without the worker, it'd also be difficult to switch between potentially long
running tasks like index build and WAL decoding. (No decoding during that time
at all can suspend archiving / recycling of WAL segments for some time, which
in turn may result in full disk.)
Another problem is that, after having acquired AccessExclusiveLock (in order
to swap the files), the backend needs to both decode and apply the data
changes that took place while it was waiting for the lock. With the decoding
worker, the decoding runs all the time, so the backend only needs to apply the
changes. This can reduce the time the exclusive lock is held for.
Note that the code added in order to handle ERRORs in the background worker
almost duplicates the existing code that does the same for other types of
workers (See ProcessParallelMessages() and
ProcessParallelApplyMessages()). Refactoring of the existing code might be
useful, to reduce the duplication.
---
src/backend/access/heap/heapam_handler.c | 44 -
src/backend/commands/cluster.c | 1175 +++++++++++++----
src/backend/libpq/pqmq.c | 5 +
src/backend/postmaster/bgworker.c | 5 +
src/backend/replication/logical/logical.c | 6 +-
.../pgoutput_repack/pgoutput_repack.c | 54 +-
src/backend/storage/ipc/procsignal.c | 4 +
src/backend/tcop/postgres.c | 4 +
.../utils/activity/wait_event_names.txt | 1 +
src/include/access/tableam.h | 7 +-
src/include/commands/cluster.h | 71 +-
src/include/storage/procsignal.h | 1 +
src/tools/pgindent/typedefs.list | 3 +-
13 files changed, 979 insertions(+), 401 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index 48b66823523..e55a4bf727e 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -33,7 +33,6 @@
#include "catalog/index.h"
#include "catalog/storage.h"
#include "catalog/storage_xlog.h"
-#include "commands/cluster.h"
#include "commands/progress.h"
#include "executor/executor.h"
#include "miscadmin.h"
@@ -688,7 +687,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
Relation OldIndex, bool use_sort,
TransactionId OldestXmin,
Snapshot snapshot,
- LogicalDecodingContext *decoding_ctx,
TransactionId *xid_cutoff,
MultiXactId *multi_cutoff,
double *num_tuples,
@@ -710,7 +708,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
BufferHeapTupleTableSlot *hslot;
BlockNumber prev_cblock = InvalidBlockNumber;
bool concurrent = snapshot != NULL;
- XLogRecPtr end_of_wal_prev = GetFlushRecPtr(NULL);
/* Remember if it's a system catalog */
is_system_catalog = IsSystemRelation(OldHeap);
@@ -972,31 +969,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
ct_val[1] = *num_tuples;
pgstat_progress_update_multi_param(2, ct_index, ct_val);
}
-
- /*
- * Process the WAL produced by the load, as well as by other
- * transactions, so that the replication slot can advance and WAL does
- * not pile up. Use wal_segment_size as a threshold so that we do not
- * introduce the decoding overhead too often.
- *
- * Of course, we must not apply the changes until the initial load has
- * completed.
- *
- * Note that our insertions into the new table should not be decoded
- * as we (intentionally) do not write the logical decoding specific
- * information to WAL.
- */
- if (concurrent)
- {
- XLogRecPtr end_of_wal;
-
- end_of_wal = GetFlushRecPtr(NULL);
- if ((end_of_wal - end_of_wal_prev) > wal_segment_size)
- {
- repack_decode_concurrent_changes(decoding_ctx, end_of_wal);
- end_of_wal_prev = end_of_wal;
- }
- }
}
if (indexScan != NULL)
@@ -1042,22 +1014,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
-
- /*
- * Try to keep the amount of not-yet-decoded WAL small, like
- * above.
- */
- if (concurrent)
- {
- XLogRecPtr end_of_wal;
-
- end_of_wal = GetFlushRecPtr(NULL);
- if ((end_of_wal - end_of_wal_prev) > wal_segment_size)
- {
- repack_decode_concurrent_changes(decoding_ctx, end_of_wal);
- end_of_wal_prev = end_of_wal;
- }
- }
}
tuplesort_end(tuplesort);
diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c
index b60cda202d2..7c9945846e2 100644
--- a/src/backend/commands/cluster.c
+++ b/src/backend/commands/cluster.c
@@ -12,12 +12,13 @@
* In concurrent mode, we lock the table with only ShareUpdateExclusiveLock,
* then do an initial copy as above. However, while the tuples are being
* copied, concurrent transactions could modify the table. To cope with those
- * changes, we rely on logical decoding to obtain them from WAL. The changes
- * are accumulated in a tuplestore. Once the initial copy is complete, we
- * read the changes from the tuplestore and re-apply them on the new heap.
- * Then we upgrade our ShareUpdateExclusiveLock to AccessExclusiveLock and
- * swap the relfilenodes. This way, the time we hold a strong lock on the
- * table is much reduced, and the bloat is eliminated.
+ * changes, we rely on logical decoding to obtain them from WAL. A bgworker
+ * consumes WAL while the initial copy is ongoing (to prevent excessive WAL
+ * from being reserved), and accumulates the changes in a file. Once the
+ * initial copy is complete, we read the changes from the file and re-apply
+ * them on the new heap. Then we upgrade our ShareUpdateExclusiveLock to
+ * AccessExclusiveLock and swap the relfilenodes. This way, the time we hold
+ * a strong lock on the table is much reduced, and the bloat is eliminated.
*
* There is hardly anything left of Paul Brown's original implementation...
*
@@ -45,6 +46,7 @@
#include "access/xlog_internal.h"
#include "access/xloginsert.h"
#include "access/xlogutils.h"
+#include "access/xlogwait.h"
#include "catalog/catalog.h"
#include "catalog/dependency.h"
#include "catalog/heap.h"
@@ -61,6 +63,8 @@
#include "commands/tablecmds.h"
#include "commands/vacuum.h"
#include "executor/executor.h"
+#include "libpq/pqformat.h"
+#include "libpq/pqmq.h"
#include "miscadmin.h"
#include "optimizer/optimizer.h"
#include "pgstat.h"
@@ -71,6 +75,8 @@
#include "storage/ipc.h"
#include "storage/lmgr.h"
#include "storage/predicate.h"
+#include "storage/procsignal.h"
+#include "tcop/tcopprot.h"
#include "utils/acl.h"
#include "utils/fmgroids.h"
#include "utils/guc.h"
@@ -117,6 +123,12 @@ typedef struct IndexInsertState
/* The WAL segment being decoded. */
static XLogSegNo repack_current_segment = 0;
+/*
+ * The first file exported by the decoding worker must contain a snapshot, the
+ * following ones contain the data changes.
+ */
+#define WORKER_FILE_SNAPSHOT 0
+
/*
* Information needed to apply concurrent data changes.
*/
@@ -136,8 +148,110 @@ typedef struct ChangeDest
/* Needed to update indexes of rel_dst. */
IndexInsertState *iistate;
+
+ /*
+ * Sequential number of the file containing the changes.
+ *
+ * TODO This field makes the structure name less descriptive. Should we
+ * rename it, e.g. to ChangeApplyInfo?
+ */
+ int file_seq;
} ChangeDest;
+/*
+ * Layout of shared memory used for communication between backend and the
+ * worker that performs logical decoding of data changes
+ */
+typedef struct DecodingWorkerShared
+{
+ /* Is the decoding initialized? */
+ bool initialized;
+
+ /*
+ * Once the worker has reached this LSN, it should close the current
+ * output file and either create a new one or exit, according to the field
+ * 'done'. If the value is InvalidXLogRecPtr, the worker should decode all
+ * the WAL available and keep checking this field. It is ok if the worker
+ * had already decoded records whose LSN is >= lsn_upto before this field
+ * has been set.
+ */
+ XLogRecPtr lsn_upto;
+
+ /* Exit after closing the current file? */
+ bool done;
+
+ /* The output is stored here. */
+ SharedFileSet sfs;
+
+ /* Number of the last file exported by the worker. */
+ int last_exported;
+
+ /* Synchronize access to the fields above. */
+ slock_t mutex;
+
+ /* Database to connect to. */
+ Oid dbid;
+
+ /* Role to connect as. */
+ Oid roleid;
+
+ /* Decode data changes of this relation. */
+ Oid relid;
+
+ /* The backend uses this to wait for the worker. */
+ ConditionVariable cv;
+
+ /* Info to signal the backend. */
+ PGPROC *backend_proc;
+ pid_t backend_pid;
+ ProcNumber backend_proc_number;
+
+ /*
+ * Memory the queue is located in.
+ *
+ * For considerations on the value see the comments of
+ * PARALLEL_ERROR_QUEUE_SIZE.
+ */
+#define REPACK_ERROR_QUEUE_SIZE 16384
+ char error_queue[FLEXIBLE_ARRAY_MEMBER];
+} DecodingWorkerShared;
+
+/*
+ * Generate worker's output file name. If relations of the same 'relid' happen
+ * to be processed at the same time, they must be from different databases and
+ * therefore different backends must be involved. (PID is already present in
+ * the fileset name.)
+ */
+static inline void
+DecodingWorkerFileName(char *fname, Oid relid, uint32 seq)
+{
+ snprintf(fname, MAXPGPATH, "%u-%u", relid, seq);
+}
+
+/*
+ * Backend-local information to control the decoding worker.
+ */
+typedef struct DecodingWorker
+{
+ /* The worker. */
+ BackgroundWorkerHandle *handle;
+
+ /* DecodingWorkerShared is in this segment. */
+ dsm_segment *seg;
+
+ /* Handle of the error queue. */
+ shm_mq_handle *error_mqh;
+} DecodingWorker;
+
+/* Pointer to currently running decoding worker. */
+static DecodingWorker *decoding_worker = NULL;
+
+/*
+ * Is there a message sent by a repack worker that the backend needs to
+ * receive?
+ */
+volatile sig_atomic_t RepackMessagePending = false;
+
static bool cluster_rel_recheck(RepackCommand cmd, Relation OldHeap,
Oid indexOid, Oid userid, LOCKMODE lmode,
int options);
@@ -145,7 +259,7 @@ static void check_repack_concurrently_requirements(Relation rel);
static void rebuild_relation(Relation OldHeap, Relation index, bool verbose,
bool concurrent);
static void copy_table_data(Relation NewHeap, Relation OldHeap, Relation OldIndex,
- Snapshot snapshot, LogicalDecodingContext *decoding_ctx,
+ Snapshot snapshot,
bool verbose,
bool *pSwapToastByContent,
TransactionId *pFreezeXid,
@@ -158,12 +272,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd,
static bool cluster_is_permitted_for_relation(RepackCommand cmd,
Oid relid, Oid userid);
-static void begin_concurrent_repack(Relation rel);
-static void end_concurrent_repack(void);
static LogicalDecodingContext *setup_logical_decoding(Oid relid);
-static HeapTuple get_changed_tuple(char *change);
-static void apply_concurrent_changes(RepackDecodingState *dstate,
- ChangeDest *dest);
+static bool decode_concurrent_changes(LogicalDecodingContext *ctx,
+ DecodingWorkerShared *shared);
+static void apply_concurrent_changes(BufFile *file, ChangeDest *dest);
static void apply_concurrent_insert(Relation rel, HeapTuple tup,
IndexInsertState *iistate,
TupleTableSlot *index_slot);
@@ -175,9 +287,9 @@ static void apply_concurrent_delete(Relation rel, HeapTuple tup_target);
static HeapTuple find_target_tuple(Relation rel, ChangeDest *dest,
HeapTuple tup_key,
TupleTableSlot *ident_slot);
-static void process_concurrent_changes(LogicalDecodingContext *decoding_ctx,
- XLogRecPtr end_of_wal,
- ChangeDest *dest);
+static void process_concurrent_changes(XLogRecPtr end_of_wal,
+ ChangeDest *dest,
+ bool done);
static IndexInsertState *get_index_insert_state(Relation relation,
Oid ident_index_id,
Relation *ident_index_p);
@@ -186,7 +298,6 @@ static ScanKey build_identity_key(Oid ident_idx_oid, Relation rel_src,
static void free_index_insert_state(IndexInsertState *iistate);
static void cleanup_logical_decoding(LogicalDecodingContext *ctx);
static void rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
- LogicalDecodingContext *decoding_ctx,
TransactionId frozenXid,
MultiXactId cutoffMulti);
static List *build_new_indexes(Relation NewHeap, Relation OldHeap, List *OldIndexes);
@@ -196,6 +307,13 @@ static Relation process_single_relation(RepackStmt *stmt,
ClusterParams *params);
static Oid determine_clustered_index(Relation rel, bool usingindex,
const char *indexname);
+static void start_decoding_worker(Oid relid);
+static void stop_decoding_worker(void);
+static void repack_worker_internal(dsm_segment *seg);
+static void export_initial_snapshot(Snapshot snapshot,
+ DecodingWorkerShared *shared);
+static Snapshot get_initial_snapshot(DecodingWorker *worker);
+static void ProcessRepackMessage(StringInfo msg);
static const char *RepackCommandAsString(RepackCommand cmd);
@@ -619,20 +737,20 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid,
/* rebuild_relation does all the dirty work */
PG_TRY();
{
- /*
- * For concurrent processing, make sure that our logical decoding
- * ignores data changes of other tables than the one we are
- * processing.
- */
- if (concurrent)
- begin_concurrent_repack(OldHeap);
-
rebuild_relation(OldHeap, index, verbose, concurrent);
}
PG_FINALLY();
{
if (concurrent)
- end_concurrent_repack();
+ {
+ /*
+ * Since during normal operation the worker was already asked to
+ * exit, stopping it explicitly is especially important on ERROR.
+ * However it still seems a good practice to make sure that the
+ * worker never survives the REPACK command.
+ */
+ stop_decoding_worker();
+ }
}
PG_END_TRY();
@@ -929,7 +1047,6 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
bool swap_toast_by_content;
TransactionId frozenXid;
MultiXactId cutoffMulti;
- LogicalDecodingContext *decoding_ctx = NULL;
Snapshot snapshot = NULL;
#if USE_ASSERT_CHECKING
LOCKMODE lmode;
@@ -943,19 +1060,36 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
if (concurrent)
{
/*
- * Prepare to capture the concurrent data changes.
+ * The worker needs to be member of the locking group we're the leader
+ * of. We ought to become the leader before the worker starts. The
+ * worker will join the group as soon as it starts.
*
- * Note that this call waits for all transactions with XID already
- * assigned to finish. If some of those transactions is waiting for a
- * lock conflicting with ShareUpdateExclusiveLock on our table (e.g.
- * it runs CREATE INDEX), we can end up in a deadlock. Not sure this
- * risk is worth unlocking/locking the table (and its clustering
- * index) and checking again if its still eligible for REPACK
- * CONCURRENTLY.
+ * This is to make sure that the deadlock described below is
+ * detectable by deadlock.c: if the worker waits for a transaction to
+ * complete and we are waiting for the worker output, then effectively
+ * we (i.e. this backend) are waiting for that transaction.
*/
- decoding_ctx = setup_logical_decoding(tableOid);
+ BecomeLockGroupLeader();
+
+ /*
+ * Start the worker that decodes data changes applied while we're
+ * copying the table contents.
+ *
+ * Note that the worker has to wait for all transactions with XID
+ * already assigned to finish. If some of those transactions is
+ * waiting for a lock conflicting with ShareUpdateExclusiveLock on our
+ * table (e.g. it runs CREATE INDEX), we can end up in a deadlock.
+ * Not sure this risk is worth unlocking/locking the table (and its
+ * clustering index) and checking again if it's still eligible for
+ * REPACK CONCURRENTLY.
+ */
+ start_decoding_worker(tableOid);
+
+ /*
+ * Wait until the worker has the initial snapshot and retrieve it.
+ */
+ snapshot = get_initial_snapshot(decoding_worker);
- snapshot = SnapBuildInitialSnapshotForRepack(decoding_ctx->snapshot_builder);
PushActiveSnapshot(snapshot);
}
@@ -980,7 +1114,7 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
NewHeap = table_open(OIDNewHeap, NoLock);
/* Copy the heap data into the new table in the desired order */
- copy_table_data(NewHeap, OldHeap, index, snapshot, decoding_ctx, verbose,
+ copy_table_data(NewHeap, OldHeap, index, snapshot, verbose,
&swap_toast_by_content, &frozenXid, &cutoffMulti);
/* The historic snapshot won't be needed anymore. */
@@ -1001,14 +1135,11 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
if (index)
index_close(index, NoLock);
- rebuild_relation_finish_concurrent(NewHeap, OldHeap, decoding_ctx,
- frozenXid, cutoffMulti);
+ rebuild_relation_finish_concurrent(NewHeap, OldHeap, frozenXid,
+ cutoffMulti);
pgstat_progress_update_param(PROGRESS_REPACK_PHASE,
PROGRESS_REPACK_PHASE_FINAL_CLEANUP);
-
- /* Done with decoding. */
- cleanup_logical_decoding(decoding_ctx);
}
else
{
@@ -1180,8 +1311,7 @@ make_new_heap(Oid OIDOldHeap, Oid NewTableSpace, Oid NewAccessMethod,
*/
static void
copy_table_data(Relation NewHeap, Relation OldHeap, Relation OldIndex,
- Snapshot snapshot, LogicalDecodingContext *decoding_ctx,
- bool verbose, bool *pSwapToastByContent,
+ Snapshot snapshot, bool verbose, bool *pSwapToastByContent,
TransactionId *pFreezeXid, MultiXactId *pCutoffMulti)
{
Relation relRelation;
@@ -1342,7 +1472,6 @@ copy_table_data(Relation NewHeap, Relation OldHeap, Relation OldIndex,
*/
table_relation_copy_for_cluster(OldHeap, NewHeap, OldIndex, use_sort,
cutoffs.OldestXmin, snapshot,
- decoding_ctx,
&cutoffs.FreezeLimit,
&cutoffs.MultiXactCutoff,
&num_tuples, &tups_vacuumed,
@@ -2376,58 +2505,6 @@ RepackCommandAsString(RepackCommand cmd)
}
-/*
- * Call this function before REPACK CONCURRENTLY starts to setup logical
- * decoding. It makes sure that other users of the table put enough
- * information into WAL.
- *
- * The point is that at various places we expect that the table we're
- * processing is treated like a system catalog. For example, we need to be
- * able to scan it using a "historic snapshot" anytime during the processing
- * (as opposed to scanning only at the start point of the decoding, as logical
- * replication does during initial table synchronization), in order to apply
- * concurrent UPDATE / DELETE commands.
- *
- * Note that TOAST table needs no attention here as it's not scanned using
- * historic snapshot.
- */
-static void
-begin_concurrent_repack(Relation rel)
-{
- Oid toastrelid;
-
- /*
- * Avoid logical decoding of other relations by this backend. The lock we
- * have guarantees that the actual locator cannot be changed concurrently:
- * TRUNCATE needs AccessExclusiveLock.
- */
- Assert(CheckRelationLockedByMe(rel, ShareUpdateExclusiveLock, false));
- repacked_rel_locator = rel->rd_locator;
- toastrelid = rel->rd_rel->reltoastrelid;
- if (OidIsValid(toastrelid))
- {
- Relation toastrel;
-
- /* Avoid logical decoding of other TOAST relations. */
- toastrel = table_open(toastrelid, AccessShareLock);
- repacked_rel_toast_locator = toastrel->rd_locator;
- table_close(toastrel, AccessShareLock);
- }
-}
-
-/*
- * Call this when done with REPACK CONCURRENTLY.
- */
-static void
-end_concurrent_repack(void)
-{
- /*
- * Restore normal function of (future) logical decoding for this backend.
- */
- repacked_rel_locator.relNumber = InvalidOid;
- repacked_rel_toast_locator.relNumber = InvalidOid;
-}
-
/*
* Is this backend performing logical decoding on behalf of REPACK
* (CONCURRENTLY) ?
@@ -2492,9 +2569,10 @@ static LogicalDecodingContext *
setup_logical_decoding(Oid relid)
{
Relation rel;
- TupleDesc tupdesc;
+ Oid toastrelid;
LogicalDecodingContext *ctx;
- RepackDecodingState *dstate = palloc0_object(RepackDecodingState);
+ NameData slotname;
+ RepackDecodingState *dstate;
/*
* REPACK CONCURRENTLY is not allowed in a transaction block, so this
@@ -2502,20 +2580,22 @@ setup_logical_decoding(Oid relid)
*/
Assert(!TransactionIdIsValid(GetTopTransactionIdIfAny()));
- /*
- * A single backend should not execute multiple REPACK commands at a time,
- * so use PID to make the slot unique.
- */
- snprintf(NameStr(dstate->slotname), NAMEDATALEN, "repack_%d", MyProcPid);
-
/*
* Make sure we can use logical decoding.
*/
CheckSlotPermissions();
CheckLogicalDecodingRequirements();
- /* RS_TEMPORARY so that the slot gets cleaned up on ERROR. */
- ReplicationSlotCreate(NameStr(dstate->slotname), true, RS_TEMPORARY,
- false, false, false);
+
+ /*
+ * A single backend should not execute multiple REPACK commands at a time,
+ * so use PID to make the slot unique.
+ *
+ * RS_TEMPORARY so that the slot gets cleaned up on ERROR.
+ */
+ snprintf(NameStr(slotname), NAMEDATALEN, "repack_%d", MyProcPid);
+ ReplicationSlotCreate(NameStr(slotname), true, RS_TEMPORARY, false, false,
+ false);
+
EnsureLogicalDecodingEnabled();
/*
@@ -2538,104 +2618,109 @@ setup_logical_decoding(Oid relid)
DecodingContextFindStartpoint(ctx);
+ /*
+ * decode_concurrent_changes() needs non-blocking callback.
+ */
+ ctx->reader->routine.page_read = read_local_xlog_page_no_wait;
+
+ /*
+ * read_local_xlog_page_no_wait() needs to be able to indicate the end of
+ * WAL.
+ */
+ ctx->reader->private_data = MemoryContextAllocZero(ctx->context,
+ sizeof(ReadLocalXLogPageNoWaitPrivate));
+
+
/* Some WAL records should have been read. */
Assert(ctx->reader->EndRecPtr != InvalidXLogRecPtr);
+ /*
+ * Initialize repack_current_segment so that we can notice WAL segment
+ * boundaries.
+ */
XLByteToSeg(ctx->reader->EndRecPtr, repack_current_segment,
wal_segment_size);
- /*
- * Setup structures to store decoded changes.
- */
+ dstate = palloc0_object(RepackDecodingState);
dstate->relid = relid;
- dstate->tstore = tuplestore_begin_heap(false, false,
- maintenance_work_mem);
- /* Caller should already have the table locked. */
- rel = table_open(relid, NoLock);
- tupdesc = CreateTupleDescCopy(RelationGetDescr(rel));
- dstate->tupdesc = tupdesc;
- table_close(rel, NoLock);
+ /*
+ * Tuple descriptor may be needed to flatten a tuple before we write it to
+ * a file. A copy is needed because the decoding worker invalidates system
+ * caches before it starts to do the actual work.
+ */
+ rel = table_open(relid, AccessShareLock);
+ dstate->tupdesc = CreateTupleDescCopy(RelationGetDescr(rel));
- /* Initialize the descriptor to store the changes ... */
- dstate->tupdesc_change = CreateTemplateTupleDesc(1);
+ /* Avoid logical decoding of other relations. */
+ repacked_rel_locator = rel->rd_locator;
+ toastrelid = rel->rd_rel->reltoastrelid;
+ if (OidIsValid(toastrelid))
+ {
+ Relation toastrel;
- TupleDescInitEntry(dstate->tupdesc_change, 1, NULL, BYTEAOID, -1, 0);
- /* ... as well as the corresponding slot. */
- dstate->tsslot = MakeSingleTupleTableSlot(dstate->tupdesc_change,
- &TTSOpsMinimalTuple);
+ /* Avoid logical decoding of other TOAST relations. */
+ toastrel = table_open(toastrelid, AccessShareLock);
+ repacked_rel_toast_locator = toastrel->rd_locator;
+ table_close(toastrel, AccessShareLock);
+ }
+ table_close(rel, AccessShareLock);
- dstate->resowner = ResourceOwnerCreate(CurrentResourceOwner,
- "logical decoding");
+ /* The file will be set as soon as we have it opened. */
+ dstate->file = NULL;
ctx->output_writer_private = dstate;
+
return ctx;
}
/*
- * Retrieve tuple from ConcurrentChange structure.
+ * Decode logical changes from the WAL sequence and store them to a file.
*
- * The input data starts with the structure but it might not be appropriately
- * aligned.
- */
-static HeapTuple
-get_changed_tuple(char *change)
-{
- HeapTupleData tup_data;
- HeapTuple result;
- char *src;
-
- /*
- * Ensure alignment before accessing the fields. (This is why we can't use
- * heap_copytuple() instead of this function.)
- */
- src = change + offsetof(ConcurrentChange, tup_data);
- memcpy(&tup_data, src, sizeof(HeapTupleData));
-
- result = (HeapTuple) palloc(HEAPTUPLESIZE + tup_data.t_len);
- memcpy(result, &tup_data, sizeof(HeapTupleData));
- result->t_data = (HeapTupleHeader) ((char *) result + HEAPTUPLESIZE);
- src = change + SizeOfConcurrentChange;
- memcpy(result->t_data, src, result->t_len);
-
- return result;
-}
-
-/*
- * Decode logical changes from the WAL sequence up to end_of_wal.
+ * If true is returned, there is no more work for the worker.
*/
-void
-repack_decode_concurrent_changes(LogicalDecodingContext *ctx,
- XLogRecPtr end_of_wal)
+static bool
+decode_concurrent_changes(LogicalDecodingContext *ctx,
+ DecodingWorkerShared *shared)
{
RepackDecodingState *dstate;
- ResourceOwner resowner_old;
+ XLogRecPtr lsn_upto;
+ bool done;
+ char fname[MAXPGPATH];
dstate = (RepackDecodingState *) ctx->output_writer_private;
- resowner_old = CurrentResourceOwner;
- CurrentResourceOwner = dstate->resowner;
- PG_TRY();
+ /* Open the output file. */
+ DecodingWorkerFileName(fname, shared->relid, shared->last_exported + 1);
+ dstate->file = BufFileCreateFileSet(&shared->sfs.fs, fname);
+
+ SpinLockAcquire(&shared->mutex);
+ lsn_upto = shared->lsn_upto;
+ done = shared->done;
+ SpinLockRelease(&shared->mutex);
+
+ while (true)
{
- while (ctx->reader->EndRecPtr < end_of_wal)
- {
- XLogRecord *record;
- XLogSegNo segno_new;
- char *errm = NULL;
- XLogRecPtr end_lsn;
+ XLogRecord *record;
+ XLogSegNo segno_new;
+ char *errm = NULL;
+ XLogRecPtr end_lsn;
- record = XLogReadRecord(ctx->reader, &errm);
- if (errm)
- elog(ERROR, "%s", errm);
+ CHECK_FOR_INTERRUPTS();
- if (record != NULL)
- LogicalDecodingProcessRecord(ctx, ctx->reader);
+ record = XLogReadRecord(ctx->reader, &errm);
+ if (record)
+ {
+ LogicalDecodingProcessRecord(ctx, ctx->reader);
/*
* If WAL segment boundary has been crossed, inform the decoding
- * system that the catalog_xmin can advance. (We can confirm more
- * often, but a filling a single WAL segment should not take much
- * time.)
+ * system that the catalog_xmin can advance.
+ *
+ * TODO Does it make sense to confirm more often? Segment size
+ * seems appropriate for restart_lsn (because less than a segment
+ * cannot be recycled anyway), however more frequent checks might
+ * be beneficial for catalog_xmin.
*/
end_lsn = ctx->reader->EndRecPtr;
XLByteToSeg(end_lsn, segno_new, wal_segment_size);
@@ -2646,80 +2731,137 @@ repack_decode_concurrent_changes(LogicalDecodingContext *ctx,
(uint32) (end_lsn >> 32), (uint32) end_lsn);
repack_current_segment = segno_new;
}
+ }
+ else
+ {
+ ReadLocalXLogPageNoWaitPrivate *priv;
+
+ if (errm)
+ ereport(ERROR, (errmsg("%s", errm)));
- CHECK_FOR_INTERRUPTS();
+ /*
+ * In the decoding loop we do not want to get blocked when there
+ * is no more WAL available, otherwise the loop would become
+ * uninterruptible.
+ */
+ priv = (ReadLocalXLogPageNoWaitPrivate *)
+ ctx->reader->private_data;
+ if (priv->end_of_wal)
+ /* Do not miss the end of WAL condition next time. */
+ priv->end_of_wal = false;
+ else
+ ereport(ERROR, (errmsg("could not read WAL record")));
+ }
+
+ /*
+ * Whether we could read new record or not, keep checking if
+ * 'lsn_upto' was specified.
+ */
+ if (XLogRecPtrIsInvalid(lsn_upto))
+ {
+ SpinLockAcquire(&shared->mutex);
+ lsn_upto = shared->lsn_upto;
+ /* 'done' should be set at the same time as 'lsn_upto' */
+ done = shared->done;
+ SpinLockRelease(&shared->mutex);
+ }
+ if (!XLogRecPtrIsInvalid(lsn_upto) &&
+ ctx->reader->EndRecPtr >= lsn_upto)
+ break;
+
+ if (record == NULL)
+ {
+ int64 timeout = 0;
+ WaitLSNResult res;
+
+ /*
+ * Before we retry reading, wait until new WAL is flushed.
+ *
+ * There is a race condition such that the backend executing
+ * REPACK determines 'lsn_upto', but before it sets the shared
+ * variable, we reach the end of WAL. In that case we'd need to
+ * wait until the next WAL flush (unrelated to REPACK). Although
+ * that should not be a problem in a busy system, it might be
+ * noticeable in other cases, including regression tests (which
+ * are not necessarily executed in parallel). Therefore it makes
+ * sense to use timeout.
+ *
+ * If lsn_upto is valid, WAL records having LSN lower than that
+ * should already have been flushed to disk.
+ */
+ if (XLogRecPtrIsInvalid(lsn_upto))
+ timeout = 100L;
+ res = WaitForLSN(WAIT_LSN_TYPE_PRIMARY_FLUSH,
+ ctx->reader->EndRecPtr + 1,
+ timeout);
+ if (res != WAIT_LSN_RESULT_SUCCESS &&
+ res != WAIT_LSN_RESULT_TIMEOUT)
+ ereport(ERROR, (errmsg("waiting for WAL failed")));
}
- InvalidateSystemCaches();
- CurrentResourceOwner = resowner_old;
- }
- PG_CATCH();
- {
- /* clear all timetravel entries */
- InvalidateSystemCaches();
- CurrentResourceOwner = resowner_old;
- PG_RE_THROW();
}
- PG_END_TRY();
+
+ /*
+ * Close the file so we can make it available to the backend.
+ */
+ BufFileClose(dstate->file);
+ dstate->file = NULL;
+ SpinLockAcquire(&shared->mutex);
+ shared->lsn_upto = InvalidXLogRecPtr;
+ shared->last_exported++;
+ SpinLockRelease(&shared->mutex);
+ ConditionVariableSignal(&shared->cv);
+
+ return done;
}
/*
* Apply changes stored in 'file'.
*/
static void
-apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
+apply_concurrent_changes(BufFile *file, ChangeDest *dest)
{
+ char kind;
+ uint32 t_len;
Relation rel = dest->rel;
TupleTableSlot *index_slot,
*ident_slot;
HeapTuple tup_old = NULL;
- if (dstate->nchanges == 0)
- return;
-
/* TupleTableSlot is needed to pass the tuple to ExecInsertIndexTuples(). */
- index_slot = MakeSingleTupleTableSlot(dstate->tupdesc, &TTSOpsHeapTuple);
+ index_slot = MakeSingleTupleTableSlot(RelationGetDescr(rel),
+ &TTSOpsHeapTuple);
/* A slot to fetch tuples from identity index. */
ident_slot = table_slot_create(rel, NULL);
- while (tuplestore_gettupleslot(dstate->tstore, true, false,
- dstate->tsslot))
+ while (true)
{
- bool shouldFree;
- HeapTuple tup_change,
- tup,
+ size_t nread;
+ HeapTuple tup,
tup_exist;
- char *change_raw,
- *src;
- ConcurrentChange change;
- bool isnull[1];
- Datum values[1];
CHECK_FOR_INTERRUPTS();
- /* Get the change from the single-column tuple. */
- tup_change = ExecFetchSlotHeapTuple(dstate->tsslot, false, &shouldFree);
- heap_deform_tuple(tup_change, dstate->tupdesc_change, values, isnull);
- Assert(!isnull[0]);
-
- /* Make sure we access aligned data. */
- change_raw = (char *) DatumGetByteaP(values[0]);
- src = (char *) VARDATA(change_raw);
- memcpy(&change, src, SizeOfConcurrentChange);
+ nread = BufFileReadMaybeEOF(file, &kind, 1, true);
+ /* Are we done with the file? */
+ if (nread == 0)
+ break;
- /*
- * Extract the tuple from the change. The tuple is copied here because
- * it might be assigned to 'tup_old', in which case it needs to
- * survive into the next iteration.
- */
- tup = get_changed_tuple(src);
+ /* Read the tuple. */
+ BufFileReadExact(file, &t_len, sizeof(t_len));
+ tup = (HeapTuple) palloc(HEAPTUPLESIZE + t_len);
+ tup->t_data = (HeapTupleHeader) ((char *) tup + HEAPTUPLESIZE);
+ BufFileReadExact(file, tup->t_data, t_len);
+ tup->t_len = t_len;
+ ItemPointerSetInvalid(&tup->t_self);
+ tup->t_tableOid = RelationGetRelid(dest->rel);
- if (change.kind == CHANGE_UPDATE_OLD)
+ if (kind == CHANGE_UPDATE_OLD)
{
Assert(tup_old == NULL);
tup_old = tup;
}
- else if (change.kind == CHANGE_INSERT)
+ else if (kind == CHANGE_INSERT)
{
Assert(tup_old == NULL);
@@ -2727,12 +2869,11 @@ apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
pfree(tup);
}
- else if (change.kind == CHANGE_UPDATE_NEW ||
- change.kind == CHANGE_DELETE)
+ else if (kind == CHANGE_UPDATE_NEW || kind == CHANGE_DELETE)
{
HeapTuple tup_key;
- if (change.kind == CHANGE_UPDATE_NEW)
+ if (kind == CHANGE_UPDATE_NEW)
{
tup_key = tup_old != NULL ? tup_old : tup;
}
@@ -2749,7 +2890,7 @@ apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
if (tup_exist == NULL)
elog(ERROR, "failed to find target tuple");
- if (change.kind == CHANGE_UPDATE_NEW)
+ if (kind == CHANGE_UPDATE_NEW)
apply_concurrent_update(rel, tup, tup_exist, dest->iistate,
index_slot);
else
@@ -2764,26 +2905,19 @@ apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
pfree(tup);
}
else
- elog(ERROR, "unrecognized kind of change: %d", change.kind);
+ elog(ERROR, "unrecognized kind of change: %d", kind);
/*
* If a change was applied now, increment CID for next writes and
* update the snapshot so it sees the changes we've applied so far.
*/
- if (change.kind != CHANGE_UPDATE_OLD)
+ if (kind != CHANGE_UPDATE_OLD)
{
CommandCounterIncrement();
UpdateActiveSnapshotCommandId();
}
-
- /* TTSOpsMinimalTuple has .get_heap_tuple==NULL. */
- Assert(shouldFree);
- pfree(tup_change);
}
- tuplestore_clear(dstate->tstore);
- dstate->nchanges = 0;
-
/* Cleanup. */
ExecDropSingleTupleTableSlot(index_slot);
ExecDropSingleTupleTableSlot(ident_slot);
@@ -2958,25 +3092,59 @@ find_target_tuple(Relation rel, ChangeDest *dest, HeapTuple tup_key,
}
/*
- * Decode and apply concurrent changes.
+ * Decode and apply concurrent changes, up to (and including) the record whose
+ * LSN is 'end_of_wal'.
*/
static void
-process_concurrent_changes(LogicalDecodingContext *decoding_ctx,
- XLogRecPtr end_of_wal, ChangeDest *dest)
+process_concurrent_changes(XLogRecPtr end_of_wal, ChangeDest *dest, bool done)
{
- RepackDecodingState *dstate;
+ DecodingWorkerShared *shared;
+ char fname[MAXPGPATH];
+ BufFile *file;
pgstat_progress_update_param(PROGRESS_REPACK_PHASE,
PROGRESS_REPACK_PHASE_CATCH_UP);
- dstate = (RepackDecodingState *) decoding_ctx->output_writer_private;
+ /* Ask the worker for the file. */
+ shared = (DecodingWorkerShared *) dsm_segment_address(decoding_worker->seg);
+ SpinLockAcquire(&shared->mutex);
+ shared->lsn_upto = end_of_wal;
+ shared->done = done;
+ SpinLockRelease(&shared->mutex);
+
+ /*
+ * The worker needs to finish processing of the current WAL record. Even
+ * if it's idle, it'll need to close the output file. Thus we're likely to
+ * wait, so prepare for sleep.
+ */
+ ConditionVariablePrepareToSleep(&shared->cv);
+ for (;;)
+ {
+ int last_exported;
- repack_decode_concurrent_changes(decoding_ctx, end_of_wal);
+ SpinLockAcquire(&shared->mutex);
+ last_exported = shared->last_exported;
+ SpinLockRelease(&shared->mutex);
- if (dstate->nchanges == 0)
- return;
+ /*
+ * Has the worker exported the file we are waiting for?
+ */
+ if (last_exported == dest->file_seq)
+ break;
- apply_concurrent_changes(dstate, dest);
+ ConditionVariableSleep(&shared->cv, WAIT_EVENT_REPACK_WORKER_EXPORT);
+ }
+ ConditionVariableCancelSleep();
+
+ /* Open the file. */
+ DecodingWorkerFileName(fname, shared->relid, dest->file_seq);
+ file = BufFileOpenFileSet(&shared->sfs.fs, fname, O_RDONLY, false);
+ apply_concurrent_changes(file, dest);
+
+ BufFileClose(file);
+
+ /* Get ready for the next file. */
+ dest->file_seq++;
}
/*
@@ -3102,15 +3270,10 @@ cleanup_logical_decoding(LogicalDecodingContext *ctx)
dstate = (RepackDecodingState *) ctx->output_writer_private;
- ExecDropSingleTupleTableSlot(dstate->tsslot);
- FreeTupleDesc(dstate->tupdesc_change);
FreeTupleDesc(dstate->tupdesc);
- tuplestore_end(dstate->tstore);
-
FreeDecodingContext(ctx);
- ReplicationSlotRelease();
- ReplicationSlotDrop(NameStr(dstate->slotname), false);
+ ReplicationSlotDropAcquired();
pfree(dstate);
}
@@ -3124,7 +3287,6 @@ cleanup_logical_decoding(LogicalDecodingContext *ctx)
*/
static void
rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
- LogicalDecodingContext *decoding_ctx,
TransactionId frozenXid,
MultiXactId cutoffMulti)
{
@@ -3205,6 +3367,7 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
&chgdst.ident_index);
chgdst.ident_key = build_identity_key(ident_idx_new, OldHeap,
&chgdst.ident_key_nentries);
+ chgdst.file_seq = WORKER_FILE_SNAPSHOT + 1;
/*
* During testing, wait for another backend to perform concurrent data
@@ -3226,7 +3389,7 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
* hold AccessExclusiveLock. (Quite some amount of WAL could have been
* written during the data copying and index creation.)
*/
- process_concurrent_changes(decoding_ctx, end_of_wal, &chgdst);
+ process_concurrent_changes(end_of_wal, &chgdst, false);
/*
* Acquire AccessExclusiveLock on the table, its TOAST relation (if there
@@ -3307,8 +3470,11 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
XLogFlush(wal_insert_ptr);
end_of_wal = GetFlushRecPtr(NULL);
- /* Apply the concurrent changes again. */
- process_concurrent_changes(decoding_ctx, end_of_wal, &chgdst);
+ /*
+ * Apply the concurrent changes again. Indicate that the decoding worker
+ * won't be needed anymore.
+ */
+ process_concurrent_changes(end_of_wal, &chgdst, true);
/* Remember info about rel before closing OldHeap */
relpersistence = OldHeap->rd_rel->relpersistence;
@@ -3420,3 +3586,510 @@ build_new_indexes(Relation NewHeap, Relation OldHeap, List *OldIndexes)
return result;
}
+
+/*
+ * Try to start a background worker to perform logical decoding of data
+ * changes applied to relation while REPACK CONCURRENTLY is copying its
+ * contents to a new table.
+ */
+static void
+start_decoding_worker(Oid relid)
+{
+ Size size;
+ dsm_segment *seg;
+ DecodingWorkerShared *shared;
+ shm_mq *mq;
+ shm_mq_handle *mqh;
+ BackgroundWorker bgw;
+
+ /* Setup shared memory. */
+ size = BUFFERALIGN(offsetof(DecodingWorkerShared, error_queue)) +
+ BUFFERALIGN(REPACK_ERROR_QUEUE_SIZE);
+ seg = dsm_create(size, 0);
+ shared = (DecodingWorkerShared *) dsm_segment_address(seg);
+ shared->lsn_upto = InvalidXLogRecPtr;
+ shared->done = false;
+ SharedFileSetInit(&shared->sfs, seg);
+ shared->last_exported = -1;
+ SpinLockInit(&shared->mutex);
+ shared->dbid = MyDatabaseId;
+
+ /*
+ * This is the UserId set in cluster_rel(). Security context shouldn't be
+ * needed for decoding worker.
+ */
+ shared->roleid = GetUserId();
+ shared->relid = relid;
+ ConditionVariableInit(&shared->cv);
+ shared->backend_proc = MyProc;
+ shared->backend_pid = MyProcPid;
+ shared->backend_proc_number = MyProcNumber;
+
+ mq = shm_mq_create((char *) BUFFERALIGN(shared->error_queue),
+ REPACK_ERROR_QUEUE_SIZE);
+ shm_mq_set_receiver(mq, MyProc);
+ mqh = shm_mq_attach(mq, seg, NULL);
+
+ memset(&bgw, 0, sizeof(bgw));
+ snprintf(bgw.bgw_name, BGW_MAXLEN,
+ "REPACK decoding worker for relation \"%s\"",
+ get_rel_name(relid));
+ snprintf(bgw.bgw_type, BGW_MAXLEN, "REPACK decoding worker");
+ bgw.bgw_flags = BGWORKER_SHMEM_ACCESS |
+ BGWORKER_BACKEND_DATABASE_CONNECTION;
+ bgw.bgw_start_time = BgWorkerStart_RecoveryFinished;
+ bgw.bgw_restart_time = BGW_NEVER_RESTART;
+ snprintf(bgw.bgw_library_name, MAXPGPATH, "postgres");
+ snprintf(bgw.bgw_function_name, BGW_MAXLEN, "RepackWorkerMain");
+ bgw.bgw_main_arg = UInt32GetDatum(dsm_segment_handle(seg));
+ bgw.bgw_notify_pid = MyProcPid;
+
+ decoding_worker = palloc0_object(DecodingWorker);
+ if (!RegisterDynamicBackgroundWorker(&bgw, &decoding_worker->handle))
+ ereport(ERROR,
+ (errcode(ERRCODE_CONFIGURATION_LIMIT_EXCEEDED),
+ errmsg("out of background worker slots"),
+ errhint("You might need to increase \"%s\".", "max_worker_processes")));
+
+ decoding_worker->seg = seg;
+ decoding_worker->error_mqh = mqh;
+
+ /*
+ * The decoding setup must be done before the caller can have XID assigned
+ * for any reason, otherwise the worker might end up in a deadlock,
+ * waiting for the caller's transaction to end. Therefore wait here until
+ * the worker indicates that it has the logical decoding initialized.
+ */
+ ConditionVariablePrepareToSleep(&shared->cv);
+ for (;;)
+ {
+ bool initialized;
+
+ SpinLockAcquire(&shared->mutex);
+ initialized = shared->initialized;
+ SpinLockRelease(&shared->mutex);
+
+ if (initialized)
+ break;
+
+ ConditionVariableSleep(&shared->cv, WAIT_EVENT_REPACK_WORKER_EXPORT);
+ }
+ ConditionVariableCancelSleep();
+}
+
+/*
+ * Stop the decoding worker and cleanup the related resources.
+ *
+ * The worker stops on its own when it knows there is no more work to do, but
+ * we need to stop it explicitly at least on ERROR in the launching backend.
+ */
+static void
+stop_decoding_worker(void)
+{
+ BgwHandleStatus status;
+
+ /* Haven't reached the worker startup? */
+ if (decoding_worker == NULL)
+ return;
+
+ /* Could not register the worker? */
+ if (decoding_worker->handle == NULL)
+ return;
+
+ TerminateBackgroundWorker(decoding_worker->handle);
+ /* The worker should really exit before the REPACK command does. */
+ HOLD_INTERRUPTS();
+ status = WaitForBackgroundWorkerShutdown(decoding_worker->handle);
+ RESUME_INTERRUPTS();
+
+ if (status == BGWH_POSTMASTER_DIED)
+ ereport(FATAL,
+ (errcode(ERRCODE_ADMIN_SHUTDOWN),
+ errmsg("postmaster exited during REPACK command")));
+
+ shm_mq_detach(decoding_worker->error_mqh);
+
+ /*
+ * If we could not cancel the current sleep due to ERROR, do that before
+ * we detach from the shared memory the condition variable is located in.
+ * If we did not, the bgworker ERROR handling code would try and fail
+ * badly.
+ */
+ ConditionVariableCancelSleep();
+
+ dsm_detach(decoding_worker->seg);
+ pfree(decoding_worker);
+ decoding_worker = NULL;
+}
+
+/* Is this process a REPACK worker? */
+static bool is_repack_worker = false;
+
+static pid_t backend_pid;
+static ProcNumber backend_proc_number;
+
+/*
+ * See ParallelWorkerShutdown for details.
+ */
+static void
+RepackWorkerShutdown(int code, Datum arg)
+{
+ SendProcSignal(backend_pid,
+ PROCSIG_REPACK_MESSAGE,
+ backend_proc_number);
+
+ dsm_detach((dsm_segment *) DatumGetPointer(arg));
+}
+
+/* REPACK decoding worker entry point */
+void
+RepackWorkerMain(Datum main_arg)
+{
+ dsm_segment *seg;
+ DecodingWorkerShared *shared;
+ shm_mq *mq;
+ shm_mq_handle *mqh;
+
+ is_repack_worker = true;
+
+ /*
+ * Override the default bgworker_die() with die() so we can use
+ * CHECK_FOR_INTERRUPTS().
+ */
+ pqsignal(SIGTERM, die);
+ BackgroundWorkerUnblockSignals();
+
+ seg = dsm_attach(DatumGetUInt32(main_arg));
+ if (seg == NULL)
+ ereport(ERROR,
+ (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
+ errmsg("could not map dynamic shared memory segment")));
+
+ shared = (DecodingWorkerShared *) dsm_segment_address(seg);
+
+ /* Arrange to signal the leader if we exit. */
+ backend_pid = shared->backend_pid;
+ backend_proc_number = shared->backend_proc_number;
+ before_shmem_exit(RepackWorkerShutdown, PointerGetDatum(seg));
+
+ /*
+ * Join locking group - see the comments around the call of
+ * start_decoding_worker().
+ */
+ if (!BecomeLockGroupMember(shared->backend_proc, backend_pid))
+ /* The leader is not running anymore. */
+ return;
+
+ /*
+ * Setup a queue to send error messages to the backend that launched this
+ * worker.
+ */
+ mq = (shm_mq *) (char *) BUFFERALIGN(shared->error_queue);
+ shm_mq_set_sender(mq, MyProc);
+ mqh = shm_mq_attach(mq, seg, NULL);
+ pq_redirect_to_shm_mq(seg, mqh);
+ pq_set_parallel_leader(shared->backend_pid,
+ shared->backend_proc_number);
+
+ /* Connect to the database. */
+ BackgroundWorkerInitializeConnectionByOid(shared->dbid, shared->roleid, 0);
+
+ repack_worker_internal(seg);
+}
+
+static void
+repack_worker_internal(dsm_segment *seg)
+{
+ DecodingWorkerShared *shared;
+ LogicalDecodingContext *decoding_ctx;
+ SharedFileSet *sfs;
+ Snapshot snapshot;
+
+ /*
+ * Transaction is needed to open relation, and it also provides us with a
+ * resource owner.
+ */
+ StartTransactionCommand();
+
+ shared = (DecodingWorkerShared *) dsm_segment_address(seg);
+
+ /*
+ * Not sure the spinlock is needed here - the backend should not change
+ * anything in the shared memory until we have serialized the snapshot.
+ */
+ SpinLockAcquire(&shared->mutex);
+ Assert(XLogRecPtrIsInvalid(shared->lsn_upto));
+ sfs = &shared->sfs;
+ SpinLockRelease(&shared->mutex);
+
+ SharedFileSetAttach(sfs, seg);
+
+ /*
+ * Prepare to capture the concurrent data changes ourselves.
+ */
+ decoding_ctx = setup_logical_decoding(shared->relid);
+
+ /* Announce that we're ready. */
+ SpinLockAcquire(&shared->mutex);
+ shared->initialized = true;
+ SpinLockRelease(&shared->mutex);
+ ConditionVariableSignal(&shared->cv);
+
+ /* Build the initial snapshot and export it. */
+ snapshot = SnapBuildInitialSnapshotForRepack(decoding_ctx->snapshot_builder);
+ export_initial_snapshot(snapshot, shared);
+
+ /*
+ * Only historic snapshots should be used now. Do not let us restrict the
+ * progress of xmin horizon.
+ */
+ InvalidateCatalogSnapshot();
+
+ while (!decode_concurrent_changes(decoding_ctx, shared))
+ ;
+
+ /* Cleanup. */
+ cleanup_logical_decoding(decoding_ctx);
+ CommitTransactionCommand();
+}
+
+/*
+ * Make snapshot available to the backend that launched the decoding worker.
+ */
+static void
+export_initial_snapshot(Snapshot snapshot, DecodingWorkerShared *shared)
+{
+ char fname[MAXPGPATH];
+ BufFile *file;
+ Size snap_size;
+ char *snap_space;
+
+ snap_size = EstimateSnapshotSpace(snapshot);
+ snap_space = (char *) palloc(snap_size);
+ SerializeSnapshot(snapshot, snap_space);
+ FreeSnapshot(snapshot);
+
+ DecodingWorkerFileName(fname, shared->relid, shared->last_exported + 1);
+ file = BufFileCreateFileSet(&shared->sfs.fs, fname);
+ /* To make restoration easier, write the snapshot size first. */
+ BufFileWrite(file, &snap_size, sizeof(snap_size));
+ BufFileWrite(file, snap_space, snap_size);
+ pfree(snap_space);
+ BufFileClose(file);
+
+ /* Increase the counter to tell the backend that the file is available. */
+ SpinLockAcquire(&shared->mutex);
+ shared->last_exported++;
+ SpinLockRelease(&shared->mutex);
+ ConditionVariableSignal(&shared->cv);
+}
+
+/*
+ * Get the initial snapshot from the decoding worker.
+ */
+static Snapshot
+get_initial_snapshot(DecodingWorker *worker)
+{
+ DecodingWorkerShared *shared;
+ char fname[MAXPGPATH];
+ BufFile *file;
+ Size snap_size;
+ char *snap_space;
+ Snapshot snapshot;
+
+ shared = (DecodingWorkerShared *) dsm_segment_address(worker->seg);
+
+ /*
+ * The worker needs to initialize the logical decoding, which usually
+ * takes some time. Therefore it makes sense to prepare for the sleep
+ * first.
+ */
+ ConditionVariablePrepareToSleep(&shared->cv);
+ for (;;)
+ {
+ int last_exported;
+
+ SpinLockAcquire(&shared->mutex);
+ last_exported = shared->last_exported;
+ SpinLockRelease(&shared->mutex);
+
+ /*
+ * Has the worker exported the file we are waiting for?
+ */
+ if (last_exported == WORKER_FILE_SNAPSHOT)
+ break;
+
+ ConditionVariableSleep(&shared->cv, WAIT_EVENT_REPACK_WORKER_EXPORT);
+ }
+ ConditionVariableCancelSleep();
+
+ /* Read the snapshot from a file. */
+ DecodingWorkerFileName(fname, shared->relid, WORKER_FILE_SNAPSHOT);
+ file = BufFileOpenFileSet(&shared->sfs.fs, fname, O_RDONLY, false);
+ BufFileReadExact(file, &snap_size, sizeof(snap_size));
+ snap_space = (char *) palloc(snap_size);
+ BufFileReadExact(file, snap_space, snap_size);
+ BufFileClose(file);
+
+ /* Restore it. */
+ snapshot = RestoreSnapshot(snap_space);
+ pfree(snap_space);
+
+ return snapshot;
+}
+
+bool
+IsRepackWorker(void)
+{
+ return is_repack_worker;
+}
+
+/*
+ * Handle receipt of an interrupt indicating a repack worker message.
+ *
+ * Note: this is called within a signal handler! All we can do is set
+ * a flag that will cause the next CHECK_FOR_INTERRUPTS() to invoke
+ * ProcessRepackMessages().
+ */
+void
+HandleRepackMessageInterrupt(void)
+{
+ InterruptPending = true;
+ RepackMessagePending = true;
+ SetLatch(MyLatch);
+}
+
+/*
+ * Process any queued protocol messages received from parallel workers.
+ */
+void
+ProcessRepackMessages(void)
+{
+ MemoryContext oldcontext;
+
+ static MemoryContext hpm_context = NULL;
+
+ /*
+ * Nothing to do if we haven't launched the worker yet or have already
+ * terminated it.
+ */
+ if (decoding_worker == NULL)
+ return;
+
+ /*
+ * This is invoked from ProcessInterrupts(), and since some of the
+ * functions it calls contain CHECK_FOR_INTERRUPTS(), there is a potential
+ * for recursive calls if more signals are received while this runs. It's
+ * unclear that recursive entry would be safe, and it doesn't seem useful
+ * even if it is safe, so let's block interrupts until done.
+ */
+ HOLD_INTERRUPTS();
+
+ /*
+ * Moreover, CurrentMemoryContext might be pointing almost anywhere. We
+ * don't want to risk leaking data into long-lived contexts, so let's do
+ * our work here in a private context that we can reset on each use.
+ */
+ if (hpm_context == NULL) /* first time through? */
+ hpm_context = AllocSetContextCreate(TopMemoryContext,
+ "ProcessRepackMessages",
+ ALLOCSET_DEFAULT_SIZES);
+ else
+ MemoryContextReset(hpm_context);
+
+ oldcontext = MemoryContextSwitchTo(hpm_context);
+
+ /* OK to process messages. Reset the flag saying there are more to do. */
+ RepackMessagePending = false;
+
+ /*
+ * Read as many messages as we can from each worker, but stop when no more
+ * messages can be read from the worker without blocking.
+ */
+ while (true)
+ {
+ shm_mq_result res;
+ Size nbytes;
+ void *data;
+
+ res = shm_mq_receive(decoding_worker->error_mqh, &nbytes,
+ &data, true);
+ if (res == SHM_MQ_WOULD_BLOCK)
+ break;
+ else if (res == SHM_MQ_SUCCESS)
+ {
+ StringInfoData msg;
+
+ initStringInfo(&msg);
+ appendBinaryStringInfo(&msg, data, nbytes);
+ ProcessRepackMessage(&msg);
+ pfree(msg.data);
+ }
+ else
+ {
+ /*
+ * The decoding worker is special in that it exits as soon as it
+ * has its work done. Thus the DETACHED result code is fine.
+ */
+ Assert(res == SHM_MQ_DETACHED);
+
+ break;
+ }
+ }
+
+ MemoryContextSwitchTo(oldcontext);
+
+ /* Might as well clear the context on our way out */
+ MemoryContextReset(hpm_context);
+
+ RESUME_INTERRUPTS();
+}
+
+/*
+ * Process a single protocol message received from a single parallel worker.
+ */
+static void
+ProcessRepackMessage(StringInfo msg)
+{
+ char msgtype;
+
+ msgtype = pq_getmsgbyte(msg);
+
+ switch (msgtype)
+ {
+ case PqMsg_ErrorResponse:
+ case PqMsg_NoticeResponse:
+ {
+ ErrorData edata;
+
+ /* Parse ErrorResponse or NoticeResponse. */
+ pq_parse_errornotice(msg, &edata);
+
+ /* Death of a worker isn't enough justification for suicide. */
+ edata.elevel = Min(edata.elevel, ERROR);
+
+ /*
+ * If desired, add a context line to show that this is a
+ * message propagated from a parallel worker. Otherwise, it
+ * can sometimes be confusing to understand what actually
+ * happened.
+ */
+ if (edata.context)
+ edata.context = psprintf("%s\n%s", edata.context,
+ _("decoding worker"));
+ else
+ edata.context = pstrdup(_("decoding worker"));
+
+ /* Rethrow error or print notice. */
+ ThrowErrorData(&edata);
+
+ break;
+ }
+
+ default:
+ {
+ elog(ERROR, "unrecognized message type received from decoding worker: %c (message length %d bytes)",
+ msgtype, msg->len);
+ }
+ }
+}
diff --git a/src/backend/libpq/pqmq.c b/src/backend/libpq/pqmq.c
index 6e4bbfb5aa1..42f6fa472c5 100644
--- a/src/backend/libpq/pqmq.c
+++ b/src/backend/libpq/pqmq.c
@@ -14,6 +14,7 @@
#include "postgres.h"
#include "access/parallel.h"
+#include "commands/cluster.h"
#include "libpq/libpq.h"
#include "libpq/pqformat.h"
#include "libpq/pqmq.h"
@@ -175,6 +176,10 @@ mq_putmessage(char msgtype, const char *s, size_t len)
SendProcSignal(pq_mq_parallel_leader_pid,
PROCSIG_PARALLEL_APPLY_MESSAGE,
pq_mq_parallel_leader_proc_number);
+ else if (IsRepackWorker())
+ SendProcSignal(pq_mq_parallel_leader_pid,
+ PROCSIG_REPACK_MESSAGE,
+ pq_mq_parallel_leader_proc_number);
else
{
Assert(IsParallelWorker());
diff --git a/src/backend/postmaster/bgworker.c b/src/backend/postmaster/bgworker.c
index 261ccd3f59c..09c58371e8e 100644
--- a/src/backend/postmaster/bgworker.c
+++ b/src/backend/postmaster/bgworker.c
@@ -13,6 +13,7 @@
#include "postgres.h"
#include "access/parallel.h"
+#include "commands/cluster.h"
#include "libpq/pqsignal.h"
#include "miscadmin.h"
#include "pgstat.h"
@@ -142,6 +143,10 @@ static const struct
{
.fn_name = "SequenceSyncWorkerMain",
.fn_addr = SequenceSyncWorkerMain
+ },
+ {
+ .fn_name = "RepackWorkerMain",
+ .fn_addr = RepackWorkerMain
}
};
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index 603a2b94d05..7651b187418 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -194,7 +194,11 @@ StartupDecodingContext(List *output_plugin_options,
ctx->slot = slot;
- ctx->reader = XLogReaderAllocate(wal_segment_size, NULL, xl_routine, ctx);
+ /*
+ * TODO A separate patch for PG core, unless there's really a reason to
+ * pass ctx for private_data (May extensions expect ctx?).
+ */
+ ctx->reader = XLogReaderAllocate(wal_segment_size, NULL, xl_routine, NULL);
if (!ctx->reader)
ereport(ERROR,
(errcode(ERRCODE_OUT_OF_MEMORY),
diff --git a/src/backend/replication/pgoutput_repack/pgoutput_repack.c b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
index 601016716b3..d0c464a98d5 100644
--- a/src/backend/replication/pgoutput_repack/pgoutput_repack.c
+++ b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
@@ -168,17 +168,13 @@ store_change(LogicalDecodingContext *ctx, ConcurrentChangeKind kind,
HeapTuple tuple)
{
RepackDecodingState *dstate;
- char *change_raw;
- ConcurrentChange change;
+ char kind_byte = (char) kind;
bool flattened = false;
- Size size;
- Datum values[1];
- bool isnull[1];
- char *dst;
dstate = (RepackDecodingState *) ctx->output_writer_private;
- size = VARHDRSZ + SizeOfConcurrentChange;
+ /* Store the change kind. */
+ BufFileWrite(dstate->file, &kind_byte, 1);
/*
* ReorderBufferCommit() stores the TOAST chunks in its private memory
@@ -195,46 +191,12 @@ store_change(LogicalDecodingContext *ctx, ConcurrentChangeKind kind,
tuple = toast_flatten_tuple(tuple, dstate->tupdesc);
flattened = true;
}
+ /* Store the tuple size ... */
+ BufFileWrite(dstate->file, &tuple->t_len, sizeof(tuple->t_len));
+ /* ... and the tuple itself. */
+ BufFileWrite(dstate->file, tuple->t_data, tuple->t_len);
- size += tuple->t_len;
- if (size >= MaxAllocSize)
- elog(ERROR, "Change is too big.");
-
- /* Construct the change. */
- change_raw = (char *) palloc0(size);
- SET_VARSIZE(change_raw, size);
-
- /*
- * Since the varlena alignment might not be sufficient for the structure,
- * set the fields in a local instance and remember where it should
- * eventually be copied.
- */
- change.kind = kind;
- dst = (char *) VARDATA(change_raw);
-
- /*
- * Copy the tuple.
- *
- * Note: change->tup_data.t_data must be fixed on retrieval!
- */
- memcpy(&change.tup_data, tuple, sizeof(HeapTupleData));
- memcpy(dst, &change, SizeOfConcurrentChange);
- dst += SizeOfConcurrentChange;
- memcpy(dst, tuple->t_data, tuple->t_len);
-
- /* The data has been copied. */
+ /* Free the flat copy if created above. */
if (flattened)
pfree(tuple);
-
- /* Store as tuple of 1 bytea column. */
- values[0] = PointerGetDatum(change_raw);
- isnull[0] = false;
- tuplestore_putvalues(dstate->tstore, dstate->tupdesc_change,
- values, isnull);
-
- /* Accounting. */
- dstate->nchanges++;
-
- /* Cleanup. */
- pfree(change_raw);
}
diff --git a/src/backend/storage/ipc/procsignal.c b/src/backend/storage/ipc/procsignal.c
index 5d33559926a..080da1ee490 100644
--- a/src/backend/storage/ipc/procsignal.c
+++ b/src/backend/storage/ipc/procsignal.c
@@ -19,6 +19,7 @@
#include "access/parallel.h"
#include "commands/async.h"
+#include "commands/cluster.h"
#include "miscadmin.h"
#include "pgstat.h"
#include "port/pg_bitutils.h"
@@ -697,6 +698,9 @@ procsignal_sigusr1_handler(SIGNAL_ARGS)
if (CheckProcSignal(PROCSIG_PARALLEL_APPLY_MESSAGE))
HandleParallelApplyMessageInterrupt();
+ if (CheckProcSignal(PROCSIG_REPACK_MESSAGE))
+ HandleRepackMessageInterrupt();
+
if (CheckProcSignal(PROCSIG_RECOVERY_CONFLICT))
HandleRecoveryConflictInterrupt();
diff --git a/src/backend/tcop/postgres.c b/src/backend/tcop/postgres.c
index 21de158adbb..2f667200d0c 100644
--- a/src/backend/tcop/postgres.c
+++ b/src/backend/tcop/postgres.c
@@ -36,6 +36,7 @@
#include "access/xact.h"
#include "catalog/pg_type.h"
#include "commands/async.h"
+#include "commands/cluster.h"
#include "commands/event_trigger.h"
#include "commands/explain_state.h"
#include "commands/prepare.h"
@@ -3575,6 +3576,9 @@ ProcessInterrupts(void)
if (ParallelApplyMessagePending)
ProcessParallelApplyMessages();
+
+ if (RepackMessagePending)
+ ProcessRepackMessages();
}
/*
diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt
index 4aa864fe3c3..b00bd794759 100644
--- a/src/backend/utils/activity/wait_event_names.txt
+++ b/src/backend/utils/activity/wait_event_names.txt
@@ -154,6 +154,7 @@ RECOVERY_CONFLICT_SNAPSHOT "Waiting for recovery conflict resolution for a vacuu
RECOVERY_CONFLICT_TABLESPACE "Waiting for recovery conflict resolution for dropping a tablespace."
RECOVERY_END_COMMAND "Waiting for <xref linkend="guc-recovery-end-command"/> to complete."
RECOVERY_PAUSE "Waiting for recovery to be resumed."
+REPACK_WORKER_EXPORT "Waiting for decoding worker to export a new output file."
REPLICATION_ORIGIN_DROP "Waiting for a replication origin to become inactive so it can be dropped."
REPLICATION_SLOT_DROP "Waiting for a replication slot to become inactive so it can be dropped."
RESTORE_COMMAND "Waiting for <xref linkend="guc-restore-command"/> to complete."
diff --git a/src/include/access/tableam.h b/src/include/access/tableam.h
index 1776d0dafdf..a9874cb1309 100644
--- a/src/include/access/tableam.h
+++ b/src/include/access/tableam.h
@@ -22,7 +22,6 @@
#include "access/xact.h"
#include "commands/vacuum.h"
#include "executor/tuptable.h"
-#include "replication/logical.h"
#include "storage/read_stream.h"
#include "utils/rel.h"
#include "utils/snapshot.h"
@@ -631,7 +630,6 @@ typedef struct TableAmRoutine
bool use_sort,
TransactionId OldestXmin,
Snapshot snapshot,
- LogicalDecodingContext *decoding_ctx,
TransactionId *xid_cutoff,
MultiXactId *multi_cutoff,
double *num_tuples,
@@ -1663,8 +1661,6 @@ table_relation_copy_data(Relation rel, const RelFileLocator *newrlocator)
* - *multi_cutoff - ditto
* - snapshot - if != NULL, ignore data changes done by transactions that this
* (MVCC) snapshot considers still in-progress or in the future.
- * - decoding_ctx - logical decoding context, to capture concurrent data
- * changes.
*
* Output parameters:
* - *xid_cutoff - rel's new relfrozenxid value, may be invalid
@@ -1678,7 +1674,6 @@ table_relation_copy_for_cluster(Relation OldTable, Relation NewTable,
bool use_sort,
TransactionId OldestXmin,
Snapshot snapshot,
- LogicalDecodingContext *decoding_ctx,
TransactionId *xid_cutoff,
MultiXactId *multi_cutoff,
double *num_tuples,
@@ -1687,7 +1682,7 @@ table_relation_copy_for_cluster(Relation OldTable, Relation NewTable,
{
OldTable->rd_tableam->relation_copy_for_cluster(OldTable, NewTable, OldIndex,
use_sort, OldestXmin,
- snapshot, decoding_ctx,
+ snapshot,
xid_cutoff, multi_cutoff,
num_tuples, tups_vacuumed,
tups_recently_dead);
diff --git a/src/include/commands/cluster.h b/src/include/commands/cluster.h
index 6a5c476294a..1b05d5d418b 100644
--- a/src/include/commands/cluster.h
+++ b/src/include/commands/cluster.h
@@ -17,11 +17,13 @@
#include "nodes/parsenodes.h"
#include "parser/parse_node.h"
#include "replication/decode.h"
+#include "postmaster/bgworker.h"
#include "replication/logical.h"
+#include "storage/buffile.h"
#include "storage/lock.h"
+#include "storage/shm_mq.h"
#include "utils/relcache.h"
#include "utils/resowner.h"
-#include "utils/tuplestore.h"
/* flag bits for ClusterParams->options */
@@ -44,6 +46,9 @@ typedef struct ClusterParams
* The following definitions are used by REPACK CONCURRENTLY.
*/
+/*
+ * Stored as a single byte in the output file.
+ */
typedef enum
{
CHANGE_INSERT,
@@ -52,68 +57,30 @@ typedef enum
CHANGE_DELETE
} ConcurrentChangeKind;
-typedef struct ConcurrentChange
-{
- /* See the enum above. */
- ConcurrentChangeKind kind;
-
- /*
- * The actual tuple.
- *
- * The tuple data follows the ConcurrentChange structure. Before use make
- * sure the tuple is correctly aligned (ConcurrentChange can be stored as
- * bytea) and that tuple->t_data is fixed.
- */
- HeapTupleData tup_data;
-} ConcurrentChange;
-
-#define SizeOfConcurrentChange (offsetof(ConcurrentChange, tup_data) + \
- sizeof(HeapTupleData))
-
/*
* Logical decoding state.
*
- * Here we store the data changes that we decode from WAL while the table
- * contents is being copied to a new storage. Also the necessary metadata
- * needed to apply these changes to the table is stored here.
+ * The output plugin uses it to store the data changes that it decodes from
+ * WAL while the table contents is being copied to a new storage.
*/
typedef struct RepackDecodingState
{
/* The relation whose changes we're decoding. */
Oid relid;
- /* Replication slot name. */
- NameData slotname;
-
- /*
- * Decoded changes are stored here. Although we try to avoid excessive
- * batches, it can happen that the changes need to be stored to disk. The
- * tuplestore does this transparently.
- */
- Tuplestorestate *tstore;
-
- /* The current number of changes in tstore. */
- double nchanges;
-
- /*
- * Descriptor to store the ConcurrentChange structure serialized (bytea).
- * We can't store the tuple directly because tuplestore only supports
- * minimum tuple and we may need to transfer OID system column from the
- * output plugin. Also we need to transfer the change kind, so it's better
- * to put everything in the structure than to use 2 tuplestores "in
- * parallel".
- */
- TupleDesc tupdesc_change;
-
- /* Tuple descriptor needed to update indexes. */
+ /* Tuple descriptor of the relation being processed. */
TupleDesc tupdesc;
- /* Slot to retrieve data from tstore. */
- TupleTableSlot *tsslot;
-
- ResourceOwner resowner;
+ /* The current output file. */
+ BufFile *file;
} RepackDecodingState;
+extern PGDLLIMPORT volatile sig_atomic_t RepackMessagePending;
+
+extern bool IsRepackWorker(void);
+extern void HandleRepackMessageInterrupt(void);
+extern void ProcessRepackMessages(void);
+
extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel);
extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid,
@@ -136,6 +103,6 @@ extern void finish_heap_swap(Oid OIDOldHeap, Oid OIDNewHeap,
extern bool am_decoding_for_repack(void);
extern bool change_useless_for_repack(XLogRecordBuffer *buf);
-extern void repack_decode_concurrent_changes(LogicalDecodingContext *ctx,
- XLogRecPtr end_of_wal);
+
+extern void RepackWorkerMain(Datum main_arg);
#endif /* CLUSTER_H */
diff --git a/src/include/storage/procsignal.h b/src/include/storage/procsignal.h
index 348fba53a93..a944ee0d211 100644
--- a/src/include/storage/procsignal.h
+++ b/src/include/storage/procsignal.h
@@ -36,6 +36,7 @@ typedef enum
PROCSIG_BARRIER, /* global barrier interrupt */
PROCSIG_LOG_MEMORY_CONTEXT, /* ask backend to log the memory contexts */
PROCSIG_PARALLEL_APPLY_MESSAGE, /* Message from parallel apply workers */
+ PROCSIG_REPACK_MESSAGE, /* Message from repack worker */
PROCSIG_RECOVERY_CONFLICT, /* backend is blocking recovery, check
* PGPROC->pendingRecoveryConflicts for the
* reason */
diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list
index 71a0802187e..b35d4e318fc 100644
--- a/src/tools/pgindent/typedefs.list
+++ b/src/tools/pgindent/typedefs.list
@@ -499,7 +499,6 @@ CompressFileHandle
CompressionLocation
CompressorState
ComputeXidHorizonsResult
-ConcurrentChange
ConcurrentChangeKind
ConditionVariable
ConditionVariableMinimallyPadded
@@ -639,6 +638,8 @@ DeclareCursorStmt
DecodedBkpBlock
DecodedXLogRecord
DecodingOutputState
+DecodingWorker
+DecodingWorkerShared
DefElem
DefElemAction
DefaultACLInfo
--
2.47.3
--=-=-=--
^ permalink raw reply [nested|flat] 55+ messages in thread
* [PATCH v35 5/5] Use background worker to do logical decoding.
@ 2026-02-16 19:49 Antonin Houska <[email protected]>
0 siblings, 0 replies; 55+ messages in thread
From: Antonin Houska @ 2026-02-16 19:49 UTC (permalink / raw)
If the backend performing REPACK (CONCURRENTLY) does both data copying and
logical decoding, it has to "travel in time" back and forth and therefore it
has to invalidate system caches quite a few times. (The copying and the
decoding work with different catalog snapshots.) As the decoding worker has
separate caches, the switching is not necessary.
Without the worker, it'd also be difficult to switch between potentially long
running tasks like index build and WAL decoding. (No decoding during that time
at all can suspend archiving / recycling of WAL segments for some time, which
in turn may result in full disk.)
Another problem is that, after having acquired AccessExclusiveLock (in order
to swap the files), the backend needs to both decode and apply the data
changes that took place while it was waiting for the lock. With the decoding
worker, the decoding runs all the time, so the backend only needs to apply the
changes. This can reduce the time the exclusive lock is held for.
Note that the code added in order to handle ERRORs in the background worker
almost duplicates the existing code that does the same for other types of
workers (See ProcessParallelMessages() and
ProcessParallelApplyMessages()). Refactoring of the existing code might be
useful, to reduce the duplication.
---
src/backend/access/heap/heapam_handler.c | 44 -
src/backend/commands/cluster.c | 1179 +++++++++++++----
src/backend/libpq/pqmq.c | 5 +
src/backend/postmaster/bgworker.c | 5 +
src/backend/replication/logical/logical.c | 6 +-
.../pgoutput_repack/pgoutput_repack.c | 54 +-
src/backend/storage/ipc/procsignal.c | 4 +
src/backend/tcop/postgres.c | 4 +
.../utils/activity/wait_event_names.txt | 1 +
src/include/access/tableam.h | 7 +-
src/include/commands/cluster.h | 71 +-
src/include/storage/procsignal.h | 1 +
src/tools/pgindent/typedefs.list | 3 +-
13 files changed, 981 insertions(+), 403 deletions(-)
diff --git a/src/backend/access/heap/heapam_handler.c b/src/backend/access/heap/heapam_handler.c
index 41c52046e14..4c503a63832 100644
--- a/src/backend/access/heap/heapam_handler.c
+++ b/src/backend/access/heap/heapam_handler.c
@@ -33,7 +33,6 @@
#include "catalog/index.h"
#include "catalog/storage.h"
#include "catalog/storage_xlog.h"
-#include "commands/cluster.h"
#include "commands/progress.h"
#include "executor/executor.h"
#include "miscadmin.h"
@@ -688,7 +687,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
Relation OldIndex, bool use_sort,
TransactionId OldestXmin,
Snapshot snapshot,
- LogicalDecodingContext *decoding_ctx,
TransactionId *xid_cutoff,
MultiXactId *multi_cutoff,
double *num_tuples,
@@ -710,7 +708,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
BufferHeapTupleTableSlot *hslot;
BlockNumber prev_cblock = InvalidBlockNumber;
bool concurrent = snapshot != NULL;
- XLogRecPtr end_of_wal_prev = GetFlushRecPtr(NULL);
/* Remember if it's a system catalog */
is_system_catalog = IsSystemRelation(OldHeap);
@@ -972,31 +969,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
ct_val[1] = *num_tuples;
pgstat_progress_update_multi_param(2, ct_index, ct_val);
}
-
- /*
- * Process the WAL produced by the load, as well as by other
- * transactions, so that the replication slot can advance and WAL does
- * not pile up. Use wal_segment_size as a threshold so that we do not
- * introduce the decoding overhead too often.
- *
- * Of course, we must not apply the changes until the initial load has
- * completed.
- *
- * Note that our insertions into the new table should not be decoded
- * as we (intentionally) do not write the logical decoding specific
- * information to WAL.
- */
- if (concurrent)
- {
- XLogRecPtr end_of_wal;
-
- end_of_wal = GetFlushRecPtr(NULL);
- if ((end_of_wal - end_of_wal_prev) > wal_segment_size)
- {
- repack_decode_concurrent_changes(decoding_ctx, end_of_wal);
- end_of_wal_prev = end_of_wal;
- }
- }
}
if (indexScan != NULL)
@@ -1042,22 +1014,6 @@ heapam_relation_copy_for_cluster(Relation OldHeap, Relation NewHeap,
/* Report n_tuples */
pgstat_progress_update_param(PROGRESS_REPACK_HEAP_TUPLES_INSERTED,
n_tuples);
-
- /*
- * Try to keep the amount of not-yet-decoded WAL small, like
- * above.
- */
- if (concurrent)
- {
- XLogRecPtr end_of_wal;
-
- end_of_wal = GetFlushRecPtr(NULL);
- if ((end_of_wal - end_of_wal_prev) > wal_segment_size)
- {
- repack_decode_concurrent_changes(decoding_ctx, end_of_wal);
- end_of_wal_prev = end_of_wal;
- }
- }
}
tuplesort_end(tuplesort);
diff --git a/src/backend/commands/cluster.c b/src/backend/commands/cluster.c
index 6111f7985ab..86c0d44f2d1 100644
--- a/src/backend/commands/cluster.c
+++ b/src/backend/commands/cluster.c
@@ -12,12 +12,13 @@
* In concurrent mode, we lock the table with only ShareUpdateExclusiveLock,
* then do an initial copy as above. However, while the tuples are being
* copied, concurrent transactions could modify the table. To cope with those
- * changes, we rely on logical decoding to obtain them from WAL. The changes
- * are accumulated in a tuplestore. Once the initial copy is complete, we
- * read the changes from the tuplestore and re-apply them on the new heap.
- * Then we upgrade our ShareUpdateExclusiveLock to AccessExclusiveLock and
- * swap the relfilenodes. This way, the time we hold a strong lock on the
- * table is much reduced, and the bloat is eliminated.
+ * changes, we rely on logical decoding to obtain them from WAL. A bgworker
+ * consumes WAL while the initial copy is ongoing (to prevent excessive WAL
+ * from being reserved), and accumulates the changes in a file. Once the
+ * initial copy is complete, we read the changes from the file and re-apply
+ * them on the new heap. Then we upgrade our ShareUpdateExclusiveLock to
+ * AccessExclusiveLock and swap the relfilenodes. This way, the time we hold
+ * a strong lock on the table is much reduced, and the bloat is eliminated.
*
* There is hardly anything left of Paul Brown's original implementation...
*
@@ -45,6 +46,7 @@
#include "access/xlog_internal.h"
#include "access/xloginsert.h"
#include "access/xlogutils.h"
+#include "access/xlogwait.h"
#include "catalog/catalog.h"
#include "catalog/dependency.h"
#include "catalog/heap.h"
@@ -61,6 +63,8 @@
#include "commands/tablecmds.h"
#include "commands/vacuum.h"
#include "executor/executor.h"
+#include "libpq/pqformat.h"
+#include "libpq/pqmq.h"
#include "miscadmin.h"
#include "optimizer/optimizer.h"
#include "pgstat.h"
@@ -71,6 +75,8 @@
#include "storage/ipc.h"
#include "storage/lmgr.h"
#include "storage/predicate.h"
+#include "storage/procsignal.h"
+#include "tcop/tcopprot.h"
#include "utils/acl.h"
#include "utils/fmgroids.h"
#include "utils/guc.h"
@@ -117,6 +123,12 @@ typedef struct IndexInsertState
/* The WAL segment being decoded. */
static XLogSegNo repack_current_segment = 0;
+/*
+ * The first file exported by the decoding worker must contain a snapshot, the
+ * following ones contain the data changes.
+ */
+#define WORKER_FILE_SNAPSHOT 0
+
/*
* Information needed to apply concurrent data changes.
*/
@@ -136,8 +148,110 @@ typedef struct ChangeDest
/* Needed to update indexes of rel_dst. */
IndexInsertState *iistate;
+
+ /*
+ * Sequential number of the file containing the changes.
+ *
+ * TODO This field makes the structure name less descriptive. Should we
+ * rename it, e.g. to ChangeApplyInfo?
+ */
+ int file_seq;
} ChangeDest;
+/*
+ * Layout of shared memory used for communication between backend and the
+ * worker that performs logical decoding of data changes
+ */
+typedef struct DecodingWorkerShared
+{
+ /* Is the decoding initialized? */
+ bool initialized;
+
+ /*
+ * Once the worker has reached this LSN, it should close the current
+ * output file and either create a new one or exit, according to the field
+ * 'done'. If the value is InvalidXLogRecPtr, the worker should decode all
+ * the WAL available and keep checking this field. It is ok if the worker
+ * had already decoded records whose LSN is >= lsn_upto before this field
+ * has been set.
+ */
+ XLogRecPtr lsn_upto;
+
+ /* Exit after closing the current file? */
+ bool done;
+
+ /* The output is stored here. */
+ SharedFileSet sfs;
+
+ /* Number of the last file exported by the worker. */
+ int last_exported;
+
+ /* Synchronize access to the fields above. */
+ slock_t mutex;
+
+ /* Database to connect to. */
+ Oid dbid;
+
+ /* Role to connect as. */
+ Oid roleid;
+
+ /* Decode data changes of this relation. */
+ Oid relid;
+
+ /* The backend uses this to wait for the worker. */
+ ConditionVariable cv;
+
+ /* Info to signal the backend. */
+ PGPROC *backend_proc;
+ pid_t backend_pid;
+ ProcNumber backend_proc_number;
+
+ /*
+ * Memory the queue is located in.
+ *
+ * For considerations on the value see the comments of
+ * PARALLEL_ERROR_QUEUE_SIZE.
+ */
+#define REPACK_ERROR_QUEUE_SIZE 16384
+ char error_queue[FLEXIBLE_ARRAY_MEMBER];
+} DecodingWorkerShared;
+
+/*
+ * Generate worker's output file name. If relations of the same 'relid' happen
+ * to be processed at the same time, they must be from different databases and
+ * therefore different backends must be involved. (PID is already present in
+ * the fileset name.)
+ */
+static inline void
+DecodingWorkerFileName(char *fname, Oid relid, uint32 seq)
+{
+ snprintf(fname, MAXPGPATH, "%u-%u", relid, seq);
+}
+
+/*
+ * Backend-local information to control the decoding worker.
+ */
+typedef struct DecodingWorker
+{
+ /* The worker. */
+ BackgroundWorkerHandle *handle;
+
+ /* DecodingWorkerShared is in this segment. */
+ dsm_segment *seg;
+
+ /* Handle of the error queue. */
+ shm_mq_handle *error_mqh;
+} DecodingWorker;
+
+/* Pointer to currently running decoding worker. */
+static DecodingWorker *decoding_worker = NULL;
+
+/*
+ * Is there a message sent by a repack worker that the backend needs to
+ * receive?
+ */
+volatile sig_atomic_t RepackMessagePending = false;
+
static bool cluster_rel_recheck(RepackCommand cmd, Relation OldHeap,
Oid indexOid, Oid userid, LOCKMODE lmode,
int options);
@@ -145,7 +259,7 @@ static void check_repack_concurrently_requirements(Relation rel);
static void rebuild_relation(Relation OldHeap, Relation index, bool verbose,
bool concurrent);
static void copy_table_data(Relation NewHeap, Relation OldHeap, Relation OldIndex,
- Snapshot snapshot, LogicalDecodingContext *decoding_ctx,
+ Snapshot snapshot,
bool verbose,
bool *pSwapToastByContent,
TransactionId *pFreezeXid,
@@ -158,12 +272,10 @@ static List *get_tables_to_repack_partitioned(RepackCommand cmd,
static bool cluster_is_permitted_for_relation(RepackCommand cmd,
Oid relid, Oid userid);
-static void begin_concurrent_repack(Relation rel);
-static void end_concurrent_repack(void);
static LogicalDecodingContext *setup_logical_decoding(Oid relid);
-static HeapTuple get_changed_tuple(char *change);
-static void apply_concurrent_changes(RepackDecodingState *dstate,
- ChangeDest *dest);
+static bool decode_concurrent_changes(LogicalDecodingContext *ctx,
+ DecodingWorkerShared *shared);
+static void apply_concurrent_changes(BufFile *file, ChangeDest *dest);
static void apply_concurrent_insert(Relation rel, HeapTuple tup,
IndexInsertState *iistate,
TupleTableSlot *index_slot);
@@ -175,9 +287,9 @@ static void apply_concurrent_delete(Relation rel, HeapTuple tup_target);
static HeapTuple find_target_tuple(Relation rel, ChangeDest *dest,
HeapTuple tup_key,
TupleTableSlot *ident_slot);
-static void process_concurrent_changes(LogicalDecodingContext *decoding_ctx,
- XLogRecPtr end_of_wal,
- ChangeDest *dest);
+static void process_concurrent_changes(XLogRecPtr end_of_wal,
+ ChangeDest *dest,
+ bool done);
static IndexInsertState *get_index_insert_state(Relation relation,
Oid ident_index_id,
Relation *ident_index_p);
@@ -186,7 +298,6 @@ static ScanKey build_identity_key(Oid ident_idx_oid, Relation rel_src,
static void free_index_insert_state(IndexInsertState *iistate);
static void cleanup_logical_decoding(LogicalDecodingContext *ctx);
static void rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
- LogicalDecodingContext *decoding_ctx,
TransactionId frozenXid,
MultiXactId cutoffMulti);
static List *build_new_indexes(Relation NewHeap, Relation OldHeap, List *OldIndexes);
@@ -196,6 +307,13 @@ static Relation process_single_relation(RepackStmt *stmt,
ClusterParams *params);
static Oid determine_clustered_index(Relation rel, bool usingindex,
const char *indexname);
+static void start_decoding_worker(Oid relid);
+static void stop_decoding_worker(void);
+static void repack_worker_internal(dsm_segment *seg);
+static void export_initial_snapshot(Snapshot snapshot,
+ DecodingWorkerShared *shared);
+static Snapshot get_initial_snapshot(DecodingWorker *worker);
+static void ProcessRepackMessage(StringInfo msg);
static const char *RepackCommandAsString(RepackCommand cmd);
@@ -619,20 +737,20 @@ cluster_rel(RepackCommand cmd, Relation OldHeap, Oid indexOid,
/* rebuild_relation does all the dirty work */
PG_TRY();
{
- /*
- * For concurrent processing, make sure that our logical decoding
- * ignores data changes of other tables than the one we are
- * processing.
- */
- if (concurrent)
- begin_concurrent_repack(OldHeap);
-
rebuild_relation(OldHeap, index, verbose, concurrent);
}
PG_FINALLY();
{
if (concurrent)
- end_concurrent_repack();
+ {
+ /*
+ * Since during normal operation the worker was already asked to
+ * exit, stopping it explicitly is especially important on ERROR.
+ * However it still seems a good practice to make sure that the
+ * worker never survives the REPACK command.
+ */
+ stop_decoding_worker();
+ }
}
PG_END_TRY();
@@ -929,7 +1047,6 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
bool swap_toast_by_content;
TransactionId frozenXid;
MultiXactId cutoffMulti;
- LogicalDecodingContext *decoding_ctx = NULL;
Snapshot snapshot = NULL;
#if USE_ASSERT_CHECKING
LOCKMODE lmode;
@@ -943,19 +1060,36 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
if (concurrent)
{
/*
- * Prepare to capture the concurrent data changes.
+ * The worker needs to be member of the locking group we're the leader
+ * of. We ought to become the leader before the worker starts. The
+ * worker will join the group as soon as it starts.
*
- * Note that this call waits for all transactions with XID already
- * assigned to finish. If some of those transactions is waiting for a
- * lock conflicting with ShareUpdateExclusiveLock on our table (e.g.
- * it runs CREATE INDEX), we can end up in a deadlock. Not sure this
- * risk is worth unlocking/locking the table (and its clustering
- * index) and checking again if its still eligible for REPACK
- * CONCURRENTLY.
+ * This is to make sure that the deadlock described below is
+ * detectable by deadlock.c: if the worker waits for a transaction to
+ * complete and we are waiting for the worker output, then effectively
+ * we (i.e. this backend) are waiting for that transaction.
*/
- decoding_ctx = setup_logical_decoding(tableOid);
+ BecomeLockGroupLeader();
+
+ /*
+ * Start the worker that decodes data changes applied while we're
+ * copying the table contents.
+ *
+ * Note that the worker has to wait for all transactions with XID
+ * already assigned to finish. If some of those transactions is
+ * waiting for a lock conflicting with ShareUpdateExclusiveLock on our
+ * table (e.g. it runs CREATE INDEX), we can end up in a deadlock.
+ * Not sure this risk is worth unlocking/locking the table (and its
+ * clustering index) and checking again if it's still eligible for
+ * REPACK CONCURRENTLY.
+ */
+ start_decoding_worker(tableOid);
+
+ /*
+ * Wait until the worker has the initial snapshot and retrieve it.
+ */
+ snapshot = get_initial_snapshot(decoding_worker);
- snapshot = SnapBuildInitialSnapshotForRepack(decoding_ctx->snapshot_builder);
PushActiveSnapshot(snapshot);
}
@@ -980,7 +1114,7 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
NewHeap = table_open(OIDNewHeap, NoLock);
/* Copy the heap data into the new table in the desired order */
- copy_table_data(NewHeap, OldHeap, index, snapshot, decoding_ctx, verbose,
+ copy_table_data(NewHeap, OldHeap, index, snapshot, verbose,
&swap_toast_by_content, &frozenXid, &cutoffMulti);
/* The historic snapshot won't be needed anymore. */
@@ -1001,14 +1135,11 @@ rebuild_relation(Relation OldHeap, Relation index, bool verbose, bool concurrent
if (index)
index_close(index, NoLock);
- rebuild_relation_finish_concurrent(NewHeap, OldHeap, decoding_ctx,
- frozenXid, cutoffMulti);
+ rebuild_relation_finish_concurrent(NewHeap, OldHeap, frozenXid,
+ cutoffMulti);
pgstat_progress_update_param(PROGRESS_REPACK_PHASE,
PROGRESS_REPACK_PHASE_FINAL_CLEANUP);
-
- /* Done with decoding. */
- cleanup_logical_decoding(decoding_ctx);
}
else
{
@@ -1180,8 +1311,7 @@ make_new_heap(Oid OIDOldHeap, Oid NewTableSpace, Oid NewAccessMethod,
*/
static void
copy_table_data(Relation NewHeap, Relation OldHeap, Relation OldIndex,
- Snapshot snapshot, LogicalDecodingContext *decoding_ctx,
- bool verbose, bool *pSwapToastByContent,
+ Snapshot snapshot, bool verbose, bool *pSwapToastByContent,
TransactionId *pFreezeXid, MultiXactId *pCutoffMulti)
{
Relation relRelation;
@@ -1342,7 +1472,6 @@ copy_table_data(Relation NewHeap, Relation OldHeap, Relation OldIndex,
*/
table_relation_copy_for_cluster(OldHeap, NewHeap, OldIndex, use_sort,
cutoffs.OldestXmin, snapshot,
- decoding_ctx,
&cutoffs.FreezeLimit,
&cutoffs.MultiXactCutoff,
&num_tuples, &tups_vacuumed,
@@ -2376,58 +2505,6 @@ RepackCommandAsString(RepackCommand cmd)
}
-/*
- * Call this function before REPACK CONCURRENTLY starts to setup logical
- * decoding. It makes sure that other users of the table put enough
- * information into WAL.
- *
- * The point is that at various places we expect that the table we're
- * processing is treated like a system catalog. For example, we need to be
- * able to scan it using a "historic snapshot" anytime during the processing
- * (as opposed to scanning only at the start point of the decoding, as logical
- * replication does during initial table synchronization), in order to apply
- * concurrent UPDATE / DELETE commands.
- *
- * Note that TOAST table needs no attention here as it's not scanned using
- * historic snapshot.
- */
-static void
-begin_concurrent_repack(Relation rel)
-{
- Oid toastrelid;
-
- /*
- * Avoid logical decoding of other relations by this backend. The lock we
- * have guarantees that the actual locator cannot be changed concurrently:
- * TRUNCATE needs AccessExclusiveLock.
- */
- Assert(CheckRelationLockedByMe(rel, ShareUpdateExclusiveLock, false));
- repacked_rel_locator = rel->rd_locator;
- toastrelid = rel->rd_rel->reltoastrelid;
- if (OidIsValid(toastrelid))
- {
- Relation toastrel;
-
- /* Avoid logical decoding of other TOAST relations. */
- toastrel = table_open(toastrelid, AccessShareLock);
- repacked_rel_toast_locator = toastrel->rd_locator;
- table_close(toastrel, AccessShareLock);
- }
-}
-
-/*
- * Call this when done with REPACK CONCURRENTLY.
- */
-static void
-end_concurrent_repack(void)
-{
- /*
- * Restore normal function of (future) logical decoding for this backend.
- */
- repacked_rel_locator.relNumber = InvalidOid;
- repacked_rel_toast_locator.relNumber = InvalidOid;
-}
-
/*
* Is this backend performing logical decoding on behalf of REPACK
* (CONCURRENTLY) ?
@@ -2492,9 +2569,10 @@ static LogicalDecodingContext *
setup_logical_decoding(Oid relid)
{
Relation rel;
- TupleDesc tupdesc;
+ Oid toastrelid;
LogicalDecodingContext *ctx;
- RepackDecodingState *dstate = palloc0_object(RepackDecodingState);
+ NameData slotname;
+ RepackDecodingState *dstate;
/*
* REPACK CONCURRENTLY is not allowed in a transaction block, so this
@@ -2502,20 +2580,22 @@ setup_logical_decoding(Oid relid)
*/
Assert(!TransactionIdIsValid(GetTopTransactionIdIfAny()));
- /*
- * A single backend should not execute multiple REPACK commands at a time,
- * so use PID to make the slot unique.
- */
- snprintf(NameStr(dstate->slotname), NAMEDATALEN, "repack_%d", MyProcPid);
-
/*
* Make sure we can use logical decoding.
*/
CheckSlotPermissions();
CheckLogicalDecodingRequirements();
- /* RS_TEMPORARY so that the slot gets cleaned up on ERROR. */
- ReplicationSlotCreate(NameStr(dstate->slotname), true, RS_TEMPORARY,
- false, false, false);
+
+ /*
+ * A single backend should not execute multiple REPACK commands at a time,
+ * so use PID to make the slot unique.
+ *
+ * RS_TEMPORARY so that the slot gets cleaned up on ERROR.
+ */
+ snprintf(NameStr(slotname), NAMEDATALEN, "repack_%d", MyProcPid);
+ ReplicationSlotCreate(NameStr(slotname), true, RS_TEMPORARY, false, false,
+ false);
+
EnsureLogicalDecodingEnabled();
/*
@@ -2538,104 +2618,109 @@ setup_logical_decoding(Oid relid)
DecodingContextFindStartpoint(ctx);
+ /*
+ * decode_concurrent_changes() needs non-blocking callback.
+ */
+ ctx->reader->routine.page_read = read_local_xlog_page_no_wait;
+
+ /*
+ * read_local_xlog_page_no_wait() needs to be able to indicate the end of
+ * WAL.
+ */
+ ctx->reader->private_data = MemoryContextAllocZero(ctx->context,
+ sizeof(ReadLocalXLogPageNoWaitPrivate));
+
+
/* Some WAL records should have been read. */
Assert(ctx->reader->EndRecPtr != InvalidXLogRecPtr);
+ /*
+ * Initialize repack_current_segment so that we can notice WAL segment
+ * boundaries.
+ */
XLByteToSeg(ctx->reader->EndRecPtr, repack_current_segment,
wal_segment_size);
- /*
- * Setup structures to store decoded changes.
- */
+ dstate = palloc0_object(RepackDecodingState);
dstate->relid = relid;
- dstate->tstore = tuplestore_begin_heap(false, false,
- maintenance_work_mem);
- /* Caller should already have the table locked. */
- rel = table_open(relid, NoLock);
- tupdesc = CreateTupleDescCopy(RelationGetDescr(rel));
- dstate->tupdesc = tupdesc;
- table_close(rel, NoLock);
+ /*
+ * Tuple descriptor may be needed to flatten a tuple before we write it to
+ * a file. A copy is needed because the decoding worker invalidates system
+ * caches before it starts to do the actual work.
+ */
+ rel = table_open(relid, AccessShareLock);
+ dstate->tupdesc = CreateTupleDescCopy(RelationGetDescr(rel));
- /* Initialize the descriptor to store the changes ... */
- dstate->tupdesc_change = CreateTemplateTupleDesc(1);
+ /* Avoid logical decoding of other relations. */
+ repacked_rel_locator = rel->rd_locator;
+ toastrelid = rel->rd_rel->reltoastrelid;
+ if (OidIsValid(toastrelid))
+ {
+ Relation toastrel;
- TupleDescInitEntry(dstate->tupdesc_change, 1, NULL, BYTEAOID, -1, 0);
- /* ... as well as the corresponding slot. */
- dstate->tsslot = MakeSingleTupleTableSlot(dstate->tupdesc_change,
- &TTSOpsMinimalTuple);
+ /* Avoid logical decoding of other TOAST relations. */
+ toastrel = table_open(toastrelid, AccessShareLock);
+ repacked_rel_toast_locator = toastrel->rd_locator;
+ table_close(toastrel, AccessShareLock);
+ }
+ table_close(rel, AccessShareLock);
- dstate->resowner = ResourceOwnerCreate(CurrentResourceOwner,
- "logical decoding");
+ /* The file will be set as soon as we have it opened. */
+ dstate->file = NULL;
ctx->output_writer_private = dstate;
+
return ctx;
}
/*
- * Retrieve tuple from ConcurrentChange structure.
+ * Decode logical changes from the WAL sequence and store them to a file.
*
- * The input data starts with the structure but it might not be appropriately
- * aligned.
+ * If true is returned, there is no more work for the worker.
*/
-static HeapTuple
-get_changed_tuple(char *change)
-{
- HeapTupleData tup_data;
- HeapTuple result;
- char *src;
-
- /*
- * Ensure alignment before accessing the fields. (This is why we can't use
- * heap_copytuple() instead of this function.)
- */
- src = change + offsetof(ConcurrentChange, tup_data);
- memcpy(&tup_data, src, sizeof(HeapTupleData));
-
- result = (HeapTuple) palloc(HEAPTUPLESIZE + tup_data.t_len);
- memcpy(result, &tup_data, sizeof(HeapTupleData));
- result->t_data = (HeapTupleHeader) ((char *) result + HEAPTUPLESIZE);
- src = change + SizeOfConcurrentChange;
- memcpy(result->t_data, src, result->t_len);
-
- return result;
-}
-
-/*
- * Decode logical changes from the WAL sequence up to end_of_wal.
- */
-void
-repack_decode_concurrent_changes(LogicalDecodingContext *ctx,
- XLogRecPtr end_of_wal)
+static bool
+decode_concurrent_changes(LogicalDecodingContext *ctx,
+ DecodingWorkerShared *shared)
{
RepackDecodingState *dstate;
- ResourceOwner resowner_old;
+ XLogRecPtr lsn_upto;
+ bool done;
+ char fname[MAXPGPATH];
dstate = (RepackDecodingState *) ctx->output_writer_private;
- resowner_old = CurrentResourceOwner;
- CurrentResourceOwner = dstate->resowner;
- PG_TRY();
+ /* Open the output file. */
+ DecodingWorkerFileName(fname, shared->relid, shared->last_exported + 1);
+ dstate->file = BufFileCreateFileSet(&shared->sfs.fs, fname);
+
+ SpinLockAcquire(&shared->mutex);
+ lsn_upto = shared->lsn_upto;
+ done = shared->done;
+ SpinLockRelease(&shared->mutex);
+
+ while (true)
{
- while (ctx->reader->EndRecPtr < end_of_wal)
+ XLogRecord *record;
+ XLogSegNo segno_new;
+ char *errm = NULL;
+ XLogRecPtr end_lsn;
+
+ CHECK_FOR_INTERRUPTS();
+
+ record = XLogReadRecord(ctx->reader, &errm);
+ if (record)
{
- XLogRecord *record;
- XLogSegNo segno_new;
- char *errm = NULL;
- XLogRecPtr end_lsn;
-
- record = XLogReadRecord(ctx->reader, &errm);
- if (errm)
- elog(ERROR, "%s", errm);
-
- if (record != NULL)
- LogicalDecodingProcessRecord(ctx, ctx->reader);
+ LogicalDecodingProcessRecord(ctx, ctx->reader);
/*
* If WAL segment boundary has been crossed, inform the decoding
- * system that the catalog_xmin can advance. (We can confirm more
- * often, but a filling a single WAL segment should not take much
- * time.)
+ * system that the catalog_xmin can advance.
+ *
+ * TODO Does it make sense to confirm more often? Segment size
+ * seems appropriate for restart_lsn (because less than a segment
+ * cannot be recycled anyway), however more frequent checks might
+ * be beneficial for catalog_xmin.
*/
end_lsn = ctx->reader->EndRecPtr;
XLByteToSeg(end_lsn, segno_new, wal_segment_size);
@@ -2646,80 +2731,137 @@ repack_decode_concurrent_changes(LogicalDecodingContext *ctx,
(uint32) (end_lsn >> 32), (uint32) end_lsn);
repack_current_segment = segno_new;
}
-
- CHECK_FOR_INTERRUPTS();
}
- InvalidateSystemCaches();
- CurrentResourceOwner = resowner_old;
+ else
+ {
+ ReadLocalXLogPageNoWaitPrivate *priv;
+
+ if (errm)
+ ereport(ERROR, (errmsg("%s", errm)));
+
+ /*
+ * In the decoding loop we do not want to get blocked when there
+ * is no more WAL available, otherwise the loop would become
+ * uninterruptible.
+ */
+ priv = (ReadLocalXLogPageNoWaitPrivate *)
+ ctx->reader->private_data;
+ if (priv->end_of_wal)
+ /* Do not miss the end of WAL condition next time. */
+ priv->end_of_wal = false;
+ else
+ ereport(ERROR, (errmsg("could not read WAL record")));
+ }
+
+ /*
+ * Whether we could read new record or not, keep checking if
+ * 'lsn_upto' was specified.
+ */
+ if (XLogRecPtrIsInvalid(lsn_upto))
+ {
+ SpinLockAcquire(&shared->mutex);
+ lsn_upto = shared->lsn_upto;
+ /* 'done' should be set at the same time as 'lsn_upto' */
+ done = shared->done;
+ SpinLockRelease(&shared->mutex);
+ }
+ if (!XLogRecPtrIsInvalid(lsn_upto) &&
+ ctx->reader->EndRecPtr >= lsn_upto)
+ break;
+
+ if (record == NULL)
+ {
+ int64 timeout = 0;
+ WaitLSNResult res;
+
+ /*
+ * Before we retry reading, wait until new WAL is flushed.
+ *
+ * There is a race condition such that the backend executing
+ * REPACK determines 'lsn_upto', but before it sets the shared
+ * variable, we reach the end of WAL. In that case we'd need to
+ * wait until the next WAL flush (unrelated to REPACK). Although
+ * that should not be a problem in a busy system, it might be
+ * noticeable in other cases, including regression tests (which
+ * are not necessarily executed in parallel). Therefore it makes
+ * sense to use timeout.
+ *
+ * If lsn_upto is valid, WAL records having LSN lower than that
+ * should already have been flushed to disk.
+ */
+ if (XLogRecPtrIsInvalid(lsn_upto))
+ timeout = 100L;
+ res = WaitForLSN(WAIT_LSN_TYPE_PRIMARY_FLUSH,
+ ctx->reader->EndRecPtr + 1,
+ timeout);
+ if (res != WAIT_LSN_RESULT_SUCCESS &&
+ res != WAIT_LSN_RESULT_TIMEOUT)
+ ereport(ERROR, (errmsg("waiting for WAL failed")));
+ }
}
- PG_CATCH();
- {
- /* clear all timetravel entries */
- InvalidateSystemCaches();
- CurrentResourceOwner = resowner_old;
- PG_RE_THROW();
- }
- PG_END_TRY();
+
+ /*
+ * Close the file so we can make it available to the backend.
+ */
+ BufFileClose(dstate->file);
+ dstate->file = NULL;
+ SpinLockAcquire(&shared->mutex);
+ shared->lsn_upto = InvalidXLogRecPtr;
+ shared->last_exported++;
+ SpinLockRelease(&shared->mutex);
+ ConditionVariableSignal(&shared->cv);
+
+ return done;
}
/*
* Apply changes stored in 'file'.
*/
static void
-apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
+apply_concurrent_changes(BufFile *file, ChangeDest *dest)
{
+ char kind;
+ uint32 t_len;
Relation rel = dest->rel;
TupleTableSlot *index_slot,
*ident_slot;
HeapTuple tup_old = NULL;
- if (dstate->nchanges == 0)
- return;
-
/* TupleTableSlot is needed to pass the tuple to ExecInsertIndexTuples(). */
- index_slot = MakeSingleTupleTableSlot(dstate->tupdesc, &TTSOpsHeapTuple);
+ index_slot = MakeSingleTupleTableSlot(RelationGetDescr(rel),
+ &TTSOpsHeapTuple);
/* A slot to fetch tuples from identity index. */
ident_slot = table_slot_create(rel, NULL);
- while (tuplestore_gettupleslot(dstate->tstore, true, false,
- dstate->tsslot))
+ while (true)
{
- bool shouldFree;
- HeapTuple tup_change,
- tup,
+ size_t nread;
+ HeapTuple tup,
tup_exist;
- char *change_raw,
- *src;
- ConcurrentChange change;
- bool isnull[1];
- Datum values[1];
CHECK_FOR_INTERRUPTS();
- /* Get the change from the single-column tuple. */
- tup_change = ExecFetchSlotHeapTuple(dstate->tsslot, false, &shouldFree);
- heap_deform_tuple(tup_change, dstate->tupdesc_change, values, isnull);
- Assert(!isnull[0]);
+ nread = BufFileReadMaybeEOF(file, &kind, 1, true);
+ /* Are we done with the file? */
+ if (nread == 0)
+ break;
- /* Make sure we access aligned data. */
- change_raw = (char *) DatumGetByteaP(values[0]);
- src = (char *) VARDATA(change_raw);
- memcpy(&change, src, SizeOfConcurrentChange);
+ /* Read the tuple. */
+ BufFileReadExact(file, &t_len, sizeof(t_len));
+ tup = (HeapTuple) palloc(HEAPTUPLESIZE + t_len);
+ tup->t_data = (HeapTupleHeader) ((char *) tup + HEAPTUPLESIZE);
+ BufFileReadExact(file, tup->t_data, t_len);
+ tup->t_len = t_len;
+ ItemPointerSetInvalid(&tup->t_self);
+ tup->t_tableOid = RelationGetRelid(dest->rel);
- /*
- * Extract the tuple from the change. The tuple is copied here because
- * it might be assigned to 'tup_old', in which case it needs to
- * survive into the next iteration.
- */
- tup = get_changed_tuple(src);
-
- if (change.kind == CHANGE_UPDATE_OLD)
+ if (kind == CHANGE_UPDATE_OLD)
{
Assert(tup_old == NULL);
tup_old = tup;
}
- else if (change.kind == CHANGE_INSERT)
+ else if (kind == CHANGE_INSERT)
{
Assert(tup_old == NULL);
@@ -2727,12 +2869,11 @@ apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
pfree(tup);
}
- else if (change.kind == CHANGE_UPDATE_NEW ||
- change.kind == CHANGE_DELETE)
+ else if (kind == CHANGE_UPDATE_NEW || kind == CHANGE_DELETE)
{
HeapTuple tup_key;
- if (change.kind == CHANGE_UPDATE_NEW)
+ if (kind == CHANGE_UPDATE_NEW)
{
tup_key = tup_old != NULL ? tup_old : tup;
}
@@ -2749,7 +2890,7 @@ apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
if (tup_exist == NULL)
elog(ERROR, "failed to find target tuple");
- if (change.kind == CHANGE_UPDATE_NEW)
+ if (kind == CHANGE_UPDATE_NEW)
apply_concurrent_update(rel, tup, tup_exist, dest->iistate,
index_slot);
else
@@ -2764,26 +2905,19 @@ apply_concurrent_changes(RepackDecodingState *dstate, ChangeDest *dest)
pfree(tup);
}
else
- elog(ERROR, "unrecognized kind of change: %d", change.kind);
+ elog(ERROR, "unrecognized kind of change: %d", kind);
/*
* If a change was applied now, increment CID for next writes and
* update the snapshot so it sees the changes we've applied so far.
*/
- if (change.kind != CHANGE_UPDATE_OLD)
+ if (kind != CHANGE_UPDATE_OLD)
{
CommandCounterIncrement();
UpdateActiveSnapshotCommandId();
}
-
- /* TTSOpsMinimalTuple has .get_heap_tuple==NULL. */
- Assert(shouldFree);
- pfree(tup_change);
}
- tuplestore_clear(dstate->tstore);
- dstate->nchanges = 0;
-
/* Cleanup. */
ExecDropSingleTupleTableSlot(index_slot);
ExecDropSingleTupleTableSlot(ident_slot);
@@ -2954,25 +3088,59 @@ find_target_tuple(Relation rel, ChangeDest *dest, HeapTuple tup_key,
}
/*
- * Decode and apply concurrent changes.
+ * Decode and apply concurrent changes, up to (and including) the record whose
+ * LSN is 'end_of_wal'.
*/
static void
-process_concurrent_changes(LogicalDecodingContext *decoding_ctx,
- XLogRecPtr end_of_wal, ChangeDest *dest)
+process_concurrent_changes(XLogRecPtr end_of_wal, ChangeDest *dest, bool done)
{
- RepackDecodingState *dstate;
+ DecodingWorkerShared *shared;
+ char fname[MAXPGPATH];
+ BufFile *file;
pgstat_progress_update_param(PROGRESS_REPACK_PHASE,
PROGRESS_REPACK_PHASE_CATCH_UP);
- dstate = (RepackDecodingState *) decoding_ctx->output_writer_private;
+ /* Ask the worker for the file. */
+ shared = (DecodingWorkerShared *) dsm_segment_address(decoding_worker->seg);
+ SpinLockAcquire(&shared->mutex);
+ shared->lsn_upto = end_of_wal;
+ shared->done = done;
+ SpinLockRelease(&shared->mutex);
- repack_decode_concurrent_changes(decoding_ctx, end_of_wal);
+ /*
+ * The worker needs to finish processing of the current WAL record. Even
+ * if it's idle, it'll need to close the output file. Thus we're likely to
+ * wait, so prepare for sleep.
+ */
+ ConditionVariablePrepareToSleep(&shared->cv);
+ for (;;)
+ {
+ int last_exported;
- if (dstate->nchanges == 0)
- return;
+ SpinLockAcquire(&shared->mutex);
+ last_exported = shared->last_exported;
+ SpinLockRelease(&shared->mutex);
- apply_concurrent_changes(dstate, dest);
+ /*
+ * Has the worker exported the file we are waiting for?
+ */
+ if (last_exported == dest->file_seq)
+ break;
+
+ ConditionVariableSleep(&shared->cv, WAIT_EVENT_REPACK_WORKER_EXPORT);
+ }
+ ConditionVariableCancelSleep();
+
+ /* Open the file. */
+ DecodingWorkerFileName(fname, shared->relid, dest->file_seq);
+ file = BufFileOpenFileSet(&shared->sfs.fs, fname, O_RDONLY, false);
+ apply_concurrent_changes(file, dest);
+
+ BufFileClose(file);
+
+ /* Get ready for the next file. */
+ dest->file_seq++;
}
/*
@@ -3098,15 +3266,10 @@ cleanup_logical_decoding(LogicalDecodingContext *ctx)
dstate = (RepackDecodingState *) ctx->output_writer_private;
- ExecDropSingleTupleTableSlot(dstate->tsslot);
- FreeTupleDesc(dstate->tupdesc_change);
FreeTupleDesc(dstate->tupdesc);
- tuplestore_end(dstate->tstore);
-
FreeDecodingContext(ctx);
- ReplicationSlotRelease();
- ReplicationSlotDrop(NameStr(dstate->slotname), false);
+ ReplicationSlotDropAcquired();
pfree(dstate);
}
@@ -3120,7 +3283,6 @@ cleanup_logical_decoding(LogicalDecodingContext *ctx)
*/
static void
rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
- LogicalDecodingContext *decoding_ctx,
TransactionId frozenXid,
MultiXactId cutoffMulti)
{
@@ -3201,6 +3363,7 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
&chgdst.ident_index);
chgdst.ident_key = build_identity_key(ident_idx_new, OldHeap,
&chgdst.ident_key_nentries);
+ chgdst.file_seq = WORKER_FILE_SNAPSHOT + 1;
/*
* During testing, wait for another backend to perform concurrent data
@@ -3222,7 +3385,7 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
* hold AccessExclusiveLock. (Quite some amount of WAL could have been
* written during the data copying and index creation.)
*/
- process_concurrent_changes(decoding_ctx, end_of_wal, &chgdst);
+ process_concurrent_changes(end_of_wal, &chgdst, false);
/*
* Acquire AccessExclusiveLock on the table, its TOAST relation (if there
@@ -3303,8 +3466,11 @@ rebuild_relation_finish_concurrent(Relation NewHeap, Relation OldHeap,
XLogFlush(wal_insert_ptr);
end_of_wal = GetFlushRecPtr(NULL);
- /* Apply the concurrent changes again. */
- process_concurrent_changes(decoding_ctx, end_of_wal, &chgdst);
+ /*
+ * Apply the concurrent changes again. Indicate that the decoding worker
+ * won't be needed anymore.
+ */
+ process_concurrent_changes(end_of_wal, &chgdst, true);
/* Remember info about rel before closing OldHeap */
relpersistence = OldHeap->rd_rel->relpersistence;
@@ -3416,3 +3582,510 @@ build_new_indexes(Relation NewHeap, Relation OldHeap, List *OldIndexes)
return result;
}
+
+/*
+ * Try to start a background worker to perform logical decoding of data
+ * changes applied to relation while REPACK CONCURRENTLY is copying its
+ * contents to a new table.
+ */
+static void
+start_decoding_worker(Oid relid)
+{
+ Size size;
+ dsm_segment *seg;
+ DecodingWorkerShared *shared;
+ shm_mq *mq;
+ shm_mq_handle *mqh;
+ BackgroundWorker bgw;
+
+ /* Setup shared memory. */
+ size = BUFFERALIGN(offsetof(DecodingWorkerShared, error_queue)) +
+ BUFFERALIGN(REPACK_ERROR_QUEUE_SIZE);
+ seg = dsm_create(size, 0);
+ shared = (DecodingWorkerShared *) dsm_segment_address(seg);
+ shared->lsn_upto = InvalidXLogRecPtr;
+ shared->done = false;
+ SharedFileSetInit(&shared->sfs, seg);
+ shared->last_exported = -1;
+ SpinLockInit(&shared->mutex);
+ shared->dbid = MyDatabaseId;
+
+ /*
+ * This is the UserId set in cluster_rel(). Security context shouldn't be
+ * needed for decoding worker.
+ */
+ shared->roleid = GetUserId();
+ shared->relid = relid;
+ ConditionVariableInit(&shared->cv);
+ shared->backend_proc = MyProc;
+ shared->backend_pid = MyProcPid;
+ shared->backend_proc_number = MyProcNumber;
+
+ mq = shm_mq_create((char *) BUFFERALIGN(shared->error_queue),
+ REPACK_ERROR_QUEUE_SIZE);
+ shm_mq_set_receiver(mq, MyProc);
+ mqh = shm_mq_attach(mq, seg, NULL);
+
+ memset(&bgw, 0, sizeof(bgw));
+ snprintf(bgw.bgw_name, BGW_MAXLEN,
+ "REPACK decoding worker for relation \"%s\"",
+ get_rel_name(relid));
+ snprintf(bgw.bgw_type, BGW_MAXLEN, "REPACK decoding worker");
+ bgw.bgw_flags = BGWORKER_SHMEM_ACCESS |
+ BGWORKER_BACKEND_DATABASE_CONNECTION;
+ bgw.bgw_start_time = BgWorkerStart_RecoveryFinished;
+ bgw.bgw_restart_time = BGW_NEVER_RESTART;
+ snprintf(bgw.bgw_library_name, MAXPGPATH, "postgres");
+ snprintf(bgw.bgw_function_name, BGW_MAXLEN, "RepackWorkerMain");
+ bgw.bgw_main_arg = UInt32GetDatum(dsm_segment_handle(seg));
+ bgw.bgw_notify_pid = MyProcPid;
+
+ decoding_worker = palloc0_object(DecodingWorker);
+ if (!RegisterDynamicBackgroundWorker(&bgw, &decoding_worker->handle))
+ ereport(ERROR,
+ (errcode(ERRCODE_CONFIGURATION_LIMIT_EXCEEDED),
+ errmsg("out of background worker slots"),
+ errhint("You might need to increase \"%s\".", "max_worker_processes")));
+
+ decoding_worker->seg = seg;
+ decoding_worker->error_mqh = mqh;
+
+ /*
+ * The decoding setup must be done before the caller can have XID assigned
+ * for any reason, otherwise the worker might end up in a deadlock,
+ * waiting for the caller's transaction to end. Therefore wait here until
+ * the worker indicates that it has the logical decoding initialized.
+ */
+ ConditionVariablePrepareToSleep(&shared->cv);
+ for (;;)
+ {
+ bool initialized;
+
+ SpinLockAcquire(&shared->mutex);
+ initialized = shared->initialized;
+ SpinLockRelease(&shared->mutex);
+
+ if (initialized)
+ break;
+
+ ConditionVariableSleep(&shared->cv, WAIT_EVENT_REPACK_WORKER_EXPORT);
+ }
+ ConditionVariableCancelSleep();
+}
+
+/*
+ * Stop the decoding worker and cleanup the related resources.
+ *
+ * The worker stops on its own when it knows there is no more work to do, but
+ * we need to stop it explicitly at least on ERROR in the launching backend.
+ */
+static void
+stop_decoding_worker(void)
+{
+ BgwHandleStatus status;
+
+ /* Haven't reached the worker startup? */
+ if (decoding_worker == NULL)
+ return;
+
+ /* Could not register the worker? */
+ if (decoding_worker->handle == NULL)
+ return;
+
+ TerminateBackgroundWorker(decoding_worker->handle);
+ /* The worker should really exit before the REPACK command does. */
+ HOLD_INTERRUPTS();
+ status = WaitForBackgroundWorkerShutdown(decoding_worker->handle);
+ RESUME_INTERRUPTS();
+
+ if (status == BGWH_POSTMASTER_DIED)
+ ereport(FATAL,
+ (errcode(ERRCODE_ADMIN_SHUTDOWN),
+ errmsg("postmaster exited during REPACK command")));
+
+ shm_mq_detach(decoding_worker->error_mqh);
+
+ /*
+ * If we could not cancel the current sleep due to ERROR, do that before
+ * we detach from the shared memory the condition variable is located in.
+ * If we did not, the bgworker ERROR handling code would try and fail
+ * badly.
+ */
+ ConditionVariableCancelSleep();
+
+ dsm_detach(decoding_worker->seg);
+ pfree(decoding_worker);
+ decoding_worker = NULL;
+}
+
+/* Is this process a REPACK worker? */
+static bool is_repack_worker = false;
+
+static pid_t backend_pid;
+static ProcNumber backend_proc_number;
+
+/*
+ * See ParallelWorkerShutdown for details.
+ */
+static void
+RepackWorkerShutdown(int code, Datum arg)
+{
+ SendProcSignal(backend_pid,
+ PROCSIG_REPACK_MESSAGE,
+ backend_proc_number);
+
+ dsm_detach((dsm_segment *) DatumGetPointer(arg));
+}
+
+/* REPACK decoding worker entry point */
+void
+RepackWorkerMain(Datum main_arg)
+{
+ dsm_segment *seg;
+ DecodingWorkerShared *shared;
+ shm_mq *mq;
+ shm_mq_handle *mqh;
+
+ is_repack_worker = true;
+
+ /*
+ * Override the default bgworker_die() with die() so we can use
+ * CHECK_FOR_INTERRUPTS().
+ */
+ pqsignal(SIGTERM, die);
+ BackgroundWorkerUnblockSignals();
+
+ seg = dsm_attach(DatumGetUInt32(main_arg));
+ if (seg == NULL)
+ ereport(ERROR,
+ (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
+ errmsg("could not map dynamic shared memory segment")));
+
+ shared = (DecodingWorkerShared *) dsm_segment_address(seg);
+
+ /* Arrange to signal the leader if we exit. */
+ backend_pid = shared->backend_pid;
+ backend_proc_number = shared->backend_proc_number;
+ before_shmem_exit(RepackWorkerShutdown, PointerGetDatum(seg));
+
+ /*
+ * Join locking group - see the comments around the call of
+ * start_decoding_worker().
+ */
+ if (!BecomeLockGroupMember(shared->backend_proc, backend_pid))
+ /* The leader is not running anymore. */
+ return;
+
+ /*
+ * Setup a queue to send error messages to the backend that launched this
+ * worker.
+ */
+ mq = (shm_mq *) (char *) BUFFERALIGN(shared->error_queue);
+ shm_mq_set_sender(mq, MyProc);
+ mqh = shm_mq_attach(mq, seg, NULL);
+ pq_redirect_to_shm_mq(seg, mqh);
+ pq_set_parallel_leader(shared->backend_pid,
+ shared->backend_proc_number);
+
+ /* Connect to the database. */
+ BackgroundWorkerInitializeConnectionByOid(shared->dbid, shared->roleid, 0);
+
+ repack_worker_internal(seg);
+}
+
+static void
+repack_worker_internal(dsm_segment *seg)
+{
+ DecodingWorkerShared *shared;
+ LogicalDecodingContext *decoding_ctx;
+ SharedFileSet *sfs;
+ Snapshot snapshot;
+
+ /*
+ * Transaction is needed to open relation, and it also provides us with a
+ * resource owner.
+ */
+ StartTransactionCommand();
+
+ shared = (DecodingWorkerShared *) dsm_segment_address(seg);
+
+ /*
+ * Not sure the spinlock is needed here - the backend should not change
+ * anything in the shared memory until we have serialized the snapshot.
+ */
+ SpinLockAcquire(&shared->mutex);
+ Assert(XLogRecPtrIsInvalid(shared->lsn_upto));
+ sfs = &shared->sfs;
+ SpinLockRelease(&shared->mutex);
+
+ SharedFileSetAttach(sfs, seg);
+
+ /*
+ * Prepare to capture the concurrent data changes ourselves.
+ */
+ decoding_ctx = setup_logical_decoding(shared->relid);
+
+ /* Announce that we're ready. */
+ SpinLockAcquire(&shared->mutex);
+ shared->initialized = true;
+ SpinLockRelease(&shared->mutex);
+ ConditionVariableSignal(&shared->cv);
+
+ /* Build the initial snapshot and export it. */
+ snapshot = SnapBuildInitialSnapshotForRepack(decoding_ctx->snapshot_builder);
+ export_initial_snapshot(snapshot, shared);
+
+ /*
+ * Only historic snapshots should be used now. Do not let us restrict the
+ * progress of xmin horizon.
+ */
+ InvalidateCatalogSnapshot();
+
+ while (!decode_concurrent_changes(decoding_ctx, shared))
+ ;
+
+ /* Cleanup. */
+ cleanup_logical_decoding(decoding_ctx);
+ CommitTransactionCommand();
+}
+
+/*
+ * Make snapshot available to the backend that launched the decoding worker.
+ */
+static void
+export_initial_snapshot(Snapshot snapshot, DecodingWorkerShared *shared)
+{
+ char fname[MAXPGPATH];
+ BufFile *file;
+ Size snap_size;
+ char *snap_space;
+
+ snap_size = EstimateSnapshotSpace(snapshot);
+ snap_space = (char *) palloc(snap_size);
+ SerializeSnapshot(snapshot, snap_space);
+ FreeSnapshot(snapshot);
+
+ DecodingWorkerFileName(fname, shared->relid, shared->last_exported + 1);
+ file = BufFileCreateFileSet(&shared->sfs.fs, fname);
+ /* To make restoration easier, write the snapshot size first. */
+ BufFileWrite(file, &snap_size, sizeof(snap_size));
+ BufFileWrite(file, snap_space, snap_size);
+ pfree(snap_space);
+ BufFileClose(file);
+
+ /* Increase the counter to tell the backend that the file is available. */
+ SpinLockAcquire(&shared->mutex);
+ shared->last_exported++;
+ SpinLockRelease(&shared->mutex);
+ ConditionVariableSignal(&shared->cv);
+}
+
+/*
+ * Get the initial snapshot from the decoding worker.
+ */
+static Snapshot
+get_initial_snapshot(DecodingWorker *worker)
+{
+ DecodingWorkerShared *shared;
+ char fname[MAXPGPATH];
+ BufFile *file;
+ Size snap_size;
+ char *snap_space;
+ Snapshot snapshot;
+
+ shared = (DecodingWorkerShared *) dsm_segment_address(worker->seg);
+
+ /*
+ * The worker needs to initialize the logical decoding, which usually
+ * takes some time. Therefore it makes sense to prepare for the sleep
+ * first.
+ */
+ ConditionVariablePrepareToSleep(&shared->cv);
+ for (;;)
+ {
+ int last_exported;
+
+ SpinLockAcquire(&shared->mutex);
+ last_exported = shared->last_exported;
+ SpinLockRelease(&shared->mutex);
+
+ /*
+ * Has the worker exported the file we are waiting for?
+ */
+ if (last_exported == WORKER_FILE_SNAPSHOT)
+ break;
+
+ ConditionVariableSleep(&shared->cv, WAIT_EVENT_REPACK_WORKER_EXPORT);
+ }
+ ConditionVariableCancelSleep();
+
+ /* Read the snapshot from a file. */
+ DecodingWorkerFileName(fname, shared->relid, WORKER_FILE_SNAPSHOT);
+ file = BufFileOpenFileSet(&shared->sfs.fs, fname, O_RDONLY, false);
+ BufFileReadExact(file, &snap_size, sizeof(snap_size));
+ snap_space = (char *) palloc(snap_size);
+ BufFileReadExact(file, snap_space, snap_size);
+ BufFileClose(file);
+
+ /* Restore it. */
+ snapshot = RestoreSnapshot(snap_space);
+ pfree(snap_space);
+
+ return snapshot;
+}
+
+bool
+IsRepackWorker(void)
+{
+ return is_repack_worker;
+}
+
+/*
+ * Handle receipt of an interrupt indicating a repack worker message.
+ *
+ * Note: this is called within a signal handler! All we can do is set
+ * a flag that will cause the next CHECK_FOR_INTERRUPTS() to invoke
+ * ProcessRepackMessages().
+ */
+void
+HandleRepackMessageInterrupt(void)
+{
+ InterruptPending = true;
+ RepackMessagePending = true;
+ SetLatch(MyLatch);
+}
+
+/*
+ * Process any queued protocol messages received from parallel workers.
+ */
+void
+ProcessRepackMessages(void)
+{
+ MemoryContext oldcontext;
+
+ static MemoryContext hpm_context = NULL;
+
+ /*
+ * Nothing to do if we haven't launched the worker yet or have already
+ * terminated it.
+ */
+ if (decoding_worker == NULL)
+ return;
+
+ /*
+ * This is invoked from ProcessInterrupts(), and since some of the
+ * functions it calls contain CHECK_FOR_INTERRUPTS(), there is a potential
+ * for recursive calls if more signals are received while this runs. It's
+ * unclear that recursive entry would be safe, and it doesn't seem useful
+ * even if it is safe, so let's block interrupts until done.
+ */
+ HOLD_INTERRUPTS();
+
+ /*
+ * Moreover, CurrentMemoryContext might be pointing almost anywhere. We
+ * don't want to risk leaking data into long-lived contexts, so let's do
+ * our work here in a private context that we can reset on each use.
+ */
+ if (hpm_context == NULL) /* first time through? */
+ hpm_context = AllocSetContextCreate(TopMemoryContext,
+ "ProcessRepackMessages",
+ ALLOCSET_DEFAULT_SIZES);
+ else
+ MemoryContextReset(hpm_context);
+
+ oldcontext = MemoryContextSwitchTo(hpm_context);
+
+ /* OK to process messages. Reset the flag saying there are more to do. */
+ RepackMessagePending = false;
+
+ /*
+ * Read as many messages as we can from each worker, but stop when no more
+ * messages can be read from the worker without blocking.
+ */
+ while (true)
+ {
+ shm_mq_result res;
+ Size nbytes;
+ void *data;
+
+ res = shm_mq_receive(decoding_worker->error_mqh, &nbytes,
+ &data, true);
+ if (res == SHM_MQ_WOULD_BLOCK)
+ break;
+ else if (res == SHM_MQ_SUCCESS)
+ {
+ StringInfoData msg;
+
+ initStringInfo(&msg);
+ appendBinaryStringInfo(&msg, data, nbytes);
+ ProcessRepackMessage(&msg);
+ pfree(msg.data);
+ }
+ else
+ {
+ /*
+ * The decoding worker is special in that it exits as soon as it
+ * has its work done. Thus the DETACHED result code is fine.
+ */
+ Assert(res == SHM_MQ_DETACHED);
+
+ break;
+ }
+ }
+
+ MemoryContextSwitchTo(oldcontext);
+
+ /* Might as well clear the context on our way out */
+ MemoryContextReset(hpm_context);
+
+ RESUME_INTERRUPTS();
+}
+
+/*
+ * Process a single protocol message received from a single parallel worker.
+ */
+static void
+ProcessRepackMessage(StringInfo msg)
+{
+ char msgtype;
+
+ msgtype = pq_getmsgbyte(msg);
+
+ switch (msgtype)
+ {
+ case PqMsg_ErrorResponse:
+ case PqMsg_NoticeResponse:
+ {
+ ErrorData edata;
+
+ /* Parse ErrorResponse or NoticeResponse. */
+ pq_parse_errornotice(msg, &edata);
+
+ /* Death of a worker isn't enough justification for suicide. */
+ edata.elevel = Min(edata.elevel, ERROR);
+
+ /*
+ * If desired, add a context line to show that this is a
+ * message propagated from a parallel worker. Otherwise, it
+ * can sometimes be confusing to understand what actually
+ * happened.
+ */
+ if (edata.context)
+ edata.context = psprintf("%s\n%s", edata.context,
+ _("decoding worker"));
+ else
+ edata.context = pstrdup(_("decoding worker"));
+
+ /* Rethrow error or print notice. */
+ ThrowErrorData(&edata);
+
+ break;
+ }
+
+ default:
+ {
+ elog(ERROR, "unrecognized message type received from decoding worker: %c (message length %d bytes)",
+ msgtype, msg->len);
+ }
+ }
+}
diff --git a/src/backend/libpq/pqmq.c b/src/backend/libpq/pqmq.c
index 6e4bbfb5aa1..42f6fa472c5 100644
--- a/src/backend/libpq/pqmq.c
+++ b/src/backend/libpq/pqmq.c
@@ -14,6 +14,7 @@
#include "postgres.h"
#include "access/parallel.h"
+#include "commands/cluster.h"
#include "libpq/libpq.h"
#include "libpq/pqformat.h"
#include "libpq/pqmq.h"
@@ -175,6 +176,10 @@ mq_putmessage(char msgtype, const char *s, size_t len)
SendProcSignal(pq_mq_parallel_leader_pid,
PROCSIG_PARALLEL_APPLY_MESSAGE,
pq_mq_parallel_leader_proc_number);
+ else if (IsRepackWorker())
+ SendProcSignal(pq_mq_parallel_leader_pid,
+ PROCSIG_REPACK_MESSAGE,
+ pq_mq_parallel_leader_proc_number);
else
{
Assert(IsParallelWorker());
diff --git a/src/backend/postmaster/bgworker.c b/src/backend/postmaster/bgworker.c
index 8678ea4e139..1edb77ba431 100644
--- a/src/backend/postmaster/bgworker.c
+++ b/src/backend/postmaster/bgworker.c
@@ -13,6 +13,7 @@
#include "postgres.h"
#include "access/parallel.h"
+#include "commands/cluster.h"
#include "libpq/pqsignal.h"
#include "miscadmin.h"
#include "pgstat.h"
@@ -142,6 +143,10 @@ static const struct
{
.fn_name = "SequenceSyncWorkerMain",
.fn_addr = SequenceSyncWorkerMain
+ },
+ {
+ .fn_name = "RepackWorkerMain",
+ .fn_addr = RepackWorkerMain
}
};
diff --git a/src/backend/replication/logical/logical.c b/src/backend/replication/logical/logical.c
index 603a2b94d05..7651b187418 100644
--- a/src/backend/replication/logical/logical.c
+++ b/src/backend/replication/logical/logical.c
@@ -194,7 +194,11 @@ StartupDecodingContext(List *output_plugin_options,
ctx->slot = slot;
- ctx->reader = XLogReaderAllocate(wal_segment_size, NULL, xl_routine, ctx);
+ /*
+ * TODO A separate patch for PG core, unless there's really a reason to
+ * pass ctx for private_data (May extensions expect ctx?).
+ */
+ ctx->reader = XLogReaderAllocate(wal_segment_size, NULL, xl_routine, NULL);
if (!ctx->reader)
ereport(ERROR,
(errcode(ERRCODE_OUT_OF_MEMORY),
diff --git a/src/backend/replication/pgoutput_repack/pgoutput_repack.c b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
index 601016716b3..d0c464a98d5 100644
--- a/src/backend/replication/pgoutput_repack/pgoutput_repack.c
+++ b/src/backend/replication/pgoutput_repack/pgoutput_repack.c
@@ -168,17 +168,13 @@ store_change(LogicalDecodingContext *ctx, ConcurrentChangeKind kind,
HeapTuple tuple)
{
RepackDecodingState *dstate;
- char *change_raw;
- ConcurrentChange change;
+ char kind_byte = (char) kind;
bool flattened = false;
- Size size;
- Datum values[1];
- bool isnull[1];
- char *dst;
dstate = (RepackDecodingState *) ctx->output_writer_private;
- size = VARHDRSZ + SizeOfConcurrentChange;
+ /* Store the change kind. */
+ BufFileWrite(dstate->file, &kind_byte, 1);
/*
* ReorderBufferCommit() stores the TOAST chunks in its private memory
@@ -195,46 +191,12 @@ store_change(LogicalDecodingContext *ctx, ConcurrentChangeKind kind,
tuple = toast_flatten_tuple(tuple, dstate->tupdesc);
flattened = true;
}
+ /* Store the tuple size ... */
+ BufFileWrite(dstate->file, &tuple->t_len, sizeof(tuple->t_len));
+ /* ... and the tuple itself. */
+ BufFileWrite(dstate->file, tuple->t_data, tuple->t_len);
- size += tuple->t_len;
- if (size >= MaxAllocSize)
- elog(ERROR, "Change is too big.");
-
- /* Construct the change. */
- change_raw = (char *) palloc0(size);
- SET_VARSIZE(change_raw, size);
-
- /*
- * Since the varlena alignment might not be sufficient for the structure,
- * set the fields in a local instance and remember where it should
- * eventually be copied.
- */
- change.kind = kind;
- dst = (char *) VARDATA(change_raw);
-
- /*
- * Copy the tuple.
- *
- * Note: change->tup_data.t_data must be fixed on retrieval!
- */
- memcpy(&change.tup_data, tuple, sizeof(HeapTupleData));
- memcpy(dst, &change, SizeOfConcurrentChange);
- dst += SizeOfConcurrentChange;
- memcpy(dst, tuple->t_data, tuple->t_len);
-
- /* The data has been copied. */
+ /* Free the flat copy if created above. */
if (flattened)
pfree(tuple);
-
- /* Store as tuple of 1 bytea column. */
- values[0] = PointerGetDatum(change_raw);
- isnull[0] = false;
- tuplestore_putvalues(dstate->tstore, dstate->tupdesc_change,
- values, isnull);
-
- /* Accounting. */
- dstate->nchanges++;
-
- /* Cleanup. */
- pfree(change_raw);
}
diff --git a/src/backend/storage/ipc/procsignal.c b/src/backend/storage/ipc/procsignal.c
index 7505c9d3a37..7c2d637bd33 100644
--- a/src/backend/storage/ipc/procsignal.c
+++ b/src/backend/storage/ipc/procsignal.c
@@ -19,6 +19,7 @@
#include "access/parallel.h"
#include "commands/async.h"
+#include "commands/cluster.h"
#include "miscadmin.h"
#include "pgstat.h"
#include "port/pg_bitutils.h"
@@ -698,6 +699,9 @@ procsignal_sigusr1_handler(SIGNAL_ARGS)
if (CheckProcSignal(PROCSIG_PARALLEL_APPLY_MESSAGE))
HandleParallelApplyMessageInterrupt();
+ if (CheckProcSignal(PROCSIG_REPACK_MESSAGE))
+ HandleRepackMessageInterrupt();
+
if (CheckProcSignal(PROCSIG_RECOVERY_CONFLICT))
HandleRecoveryConflictInterrupt();
diff --git a/src/backend/tcop/postgres.c b/src/backend/tcop/postgres.c
index d01a09dd0c4..9711abaf2eb 100644
--- a/src/backend/tcop/postgres.c
+++ b/src/backend/tcop/postgres.c
@@ -36,6 +36,7 @@
#include "access/xact.h"
#include "catalog/pg_type.h"
#include "commands/async.h"
+#include "commands/cluster.h"
#include "commands/event_trigger.h"
#include "commands/explain_state.h"
#include "commands/prepare.h"
@@ -3575,6 +3576,9 @@ ProcessInterrupts(void)
if (ParallelApplyMessagePending)
ProcessParallelApplyMessages();
+
+ if (RepackMessagePending)
+ ProcessRepackMessages();
}
/*
diff --git a/src/backend/utils/activity/wait_event_names.txt b/src/backend/utils/activity/wait_event_names.txt
index 4aa864fe3c3..b00bd794759 100644
--- a/src/backend/utils/activity/wait_event_names.txt
+++ b/src/backend/utils/activity/wait_event_names.txt
@@ -154,6 +154,7 @@ RECOVERY_CONFLICT_SNAPSHOT "Waiting for recovery conflict resolution for a vacuu
RECOVERY_CONFLICT_TABLESPACE "Waiting for recovery conflict resolution for dropping a tablespace."
RECOVERY_END_COMMAND "Waiting for <xref linkend="guc-recovery-end-command"/> to complete."
RECOVERY_PAUSE "Waiting for recovery to be resumed."
+REPACK_WORKER_EXPORT "Waiting for decoding worker to export a new output file."
REPLICATION_ORIGIN_DROP "Waiting for a replication origin to become inactive so it can be dropped."
REPLICATION_SLOT_DROP "Waiting for a replication slot to become inactive so it can be dropped."
RESTORE_COMMAND "Waiting for <xref linkend="guc-restore-command"/> to complete."
diff --git a/src/include/access/tableam.h b/src/include/access/tableam.h
index 82aa258f72d..ce72e7e017d 100644
--- a/src/include/access/tableam.h
+++ b/src/include/access/tableam.h
@@ -22,7 +22,6 @@
#include "access/xact.h"
#include "commands/vacuum.h"
#include "executor/tuptable.h"
-#include "replication/logical.h"
#include "storage/read_stream.h"
#include "utils/rel.h"
#include "utils/snapshot.h"
@@ -631,7 +630,6 @@ typedef struct TableAmRoutine
bool use_sort,
TransactionId OldestXmin,
Snapshot snapshot,
- LogicalDecodingContext *decoding_ctx,
TransactionId *xid_cutoff,
MultiXactId *multi_cutoff,
double *num_tuples,
@@ -1663,8 +1661,6 @@ table_relation_copy_data(Relation rel, const RelFileLocator *newrlocator)
* - *multi_cutoff - ditto
* - snapshot - if != NULL, ignore data changes done by transactions that this
* (MVCC) snapshot considers still in-progress or in the future.
- * - decoding_ctx - logical decoding context, to capture concurrent data
- * changes.
*
* Output parameters:
* - *xid_cutoff - rel's new relfrozenxid value, may be invalid
@@ -1678,7 +1674,6 @@ table_relation_copy_for_cluster(Relation OldTable, Relation NewTable,
bool use_sort,
TransactionId OldestXmin,
Snapshot snapshot,
- LogicalDecodingContext *decoding_ctx,
TransactionId *xid_cutoff,
MultiXactId *multi_cutoff,
double *num_tuples,
@@ -1687,7 +1682,7 @@ table_relation_copy_for_cluster(Relation OldTable, Relation NewTable,
{
OldTable->rd_tableam->relation_copy_for_cluster(OldTable, NewTable, OldIndex,
use_sort, OldestXmin,
- snapshot, decoding_ctx,
+ snapshot,
xid_cutoff, multi_cutoff,
num_tuples, tups_vacuumed,
tups_recently_dead);
diff --git a/src/include/commands/cluster.h b/src/include/commands/cluster.h
index 6a5c476294a..1b05d5d418b 100644
--- a/src/include/commands/cluster.h
+++ b/src/include/commands/cluster.h
@@ -17,11 +17,13 @@
#include "nodes/parsenodes.h"
#include "parser/parse_node.h"
#include "replication/decode.h"
+#include "postmaster/bgworker.h"
#include "replication/logical.h"
+#include "storage/buffile.h"
#include "storage/lock.h"
+#include "storage/shm_mq.h"
#include "utils/relcache.h"
#include "utils/resowner.h"
-#include "utils/tuplestore.h"
/* flag bits for ClusterParams->options */
@@ -44,6 +46,9 @@ typedef struct ClusterParams
* The following definitions are used by REPACK CONCURRENTLY.
*/
+/*
+ * Stored as a single byte in the output file.
+ */
typedef enum
{
CHANGE_INSERT,
@@ -52,68 +57,30 @@ typedef enum
CHANGE_DELETE
} ConcurrentChangeKind;
-typedef struct ConcurrentChange
-{
- /* See the enum above. */
- ConcurrentChangeKind kind;
-
- /*
- * The actual tuple.
- *
- * The tuple data follows the ConcurrentChange structure. Before use make
- * sure the tuple is correctly aligned (ConcurrentChange can be stored as
- * bytea) and that tuple->t_data is fixed.
- */
- HeapTupleData tup_data;
-} ConcurrentChange;
-
-#define SizeOfConcurrentChange (offsetof(ConcurrentChange, tup_data) + \
- sizeof(HeapTupleData))
-
/*
* Logical decoding state.
*
- * Here we store the data changes that we decode from WAL while the table
- * contents is being copied to a new storage. Also the necessary metadata
- * needed to apply these changes to the table is stored here.
+ * The output plugin uses it to store the data changes that it decodes from
+ * WAL while the table contents is being copied to a new storage.
*/
typedef struct RepackDecodingState
{
/* The relation whose changes we're decoding. */
Oid relid;
- /* Replication slot name. */
- NameData slotname;
-
- /*
- * Decoded changes are stored here. Although we try to avoid excessive
- * batches, it can happen that the changes need to be stored to disk. The
- * tuplestore does this transparently.
- */
- Tuplestorestate *tstore;
-
- /* The current number of changes in tstore. */
- double nchanges;
-
- /*
- * Descriptor to store the ConcurrentChange structure serialized (bytea).
- * We can't store the tuple directly because tuplestore only supports
- * minimum tuple and we may need to transfer OID system column from the
- * output plugin. Also we need to transfer the change kind, so it's better
- * to put everything in the structure than to use 2 tuplestores "in
- * parallel".
- */
- TupleDesc tupdesc_change;
-
- /* Tuple descriptor needed to update indexes. */
+ /* Tuple descriptor of the relation being processed. */
TupleDesc tupdesc;
- /* Slot to retrieve data from tstore. */
- TupleTableSlot *tsslot;
-
- ResourceOwner resowner;
+ /* The current output file. */
+ BufFile *file;
} RepackDecodingState;
+extern PGDLLIMPORT volatile sig_atomic_t RepackMessagePending;
+
+extern bool IsRepackWorker(void);
+extern void HandleRepackMessageInterrupt(void);
+extern void ProcessRepackMessages(void);
+
extern void ExecRepack(ParseState *pstate, RepackStmt *stmt, bool isTopLevel);
extern void cluster_rel(RepackCommand command, Relation OldHeap, Oid indexOid,
@@ -136,6 +103,6 @@ extern void finish_heap_swap(Oid OIDOldHeap, Oid OIDNewHeap,
extern bool am_decoding_for_repack(void);
extern bool change_useless_for_repack(XLogRecordBuffer *buf);
-extern void repack_decode_concurrent_changes(LogicalDecodingContext *ctx,
- XLogRecPtr end_of_wal);
+
+extern void RepackWorkerMain(Datum main_arg);
#endif /* CLUSTER_H */
diff --git a/src/include/storage/procsignal.h b/src/include/storage/procsignal.h
index 348fba53a93..a944ee0d211 100644
--- a/src/include/storage/procsignal.h
+++ b/src/include/storage/procsignal.h
@@ -36,6 +36,7 @@ typedef enum
PROCSIG_BARRIER, /* global barrier interrupt */
PROCSIG_LOG_MEMORY_CONTEXT, /* ask backend to log the memory contexts */
PROCSIG_PARALLEL_APPLY_MESSAGE, /* Message from parallel apply workers */
+ PROCSIG_REPACK_MESSAGE, /* Message from repack worker */
PROCSIG_RECOVERY_CONFLICT, /* backend is blocking recovery, check
* PGPROC->pendingRecoveryConflicts for the
* reason */
diff --git a/src/tools/pgindent/typedefs.list b/src/tools/pgindent/typedefs.list
index 71a0802187e..b35d4e318fc 100644
--- a/src/tools/pgindent/typedefs.list
+++ b/src/tools/pgindent/typedefs.list
@@ -499,7 +499,6 @@ CompressFileHandle
CompressionLocation
CompressorState
ComputeXidHorizonsResult
-ConcurrentChange
ConcurrentChangeKind
ConditionVariable
ConditionVariableMinimallyPadded
@@ -639,6 +638,8 @@ DeclareCursorStmt
DecodedBkpBlock
DecodedXLogRecord
DecodingOutputState
+DecodingWorker
+DecodingWorkerShared
DefElem
DefElemAction
DefaultACLInfo
--
2.47.3
--kmtyk4ln3767qrma--
^ permalink raw reply [nested|flat] 55+ messages in thread
end of thread, other threads:[~2026-02-16 19:49 UTC | newest]
Thread overview: 55+ messages (download: mbox mbox.gz follow: Atom feed)
-- links below jump to the message on this page --
2019-08-28 05:12 [PATCH v23 5/5] Additional test for new GUC setting. Kyotaro Horiguchi <[email protected]>
2019-08-28 05:12 [PATCH v21 5/5] Additional test for new GUC setting. Kyotaro Horiguchi <[email protected]>
2019-08-28 05:12 [PATCH 4/4] Additional test for new GUC setting. Kyotaro Horiguchi <[email protected]>
2019-08-28 05:12 [PATCH v23 5/5] Additional test for new GUC setting. Kyotaro Horiguchi <[email protected]>
2022-04-05 08:18 Re: standby recovery fails (tablespace related) (tentative patch and discussion) Kyotaro Horiguchi <[email protected]>
2022-07-13 16:43 ` Re: standby recovery fails (tablespace related) (tentative patch and discussion) Alvaro Herrera <[email protected]>
2022-07-14 21:47 ` Re: standby recovery fails (tablespace related) (tentative patch and discussion) Alvaro Herrera <[email protected]>
2022-07-15 07:30 ` Re: standby recovery fails (tablespace related) (tentative patch and discussion) Kyotaro Horiguchi <[email protected]>
2022-07-15 07:56 ` Re: standby recovery fails (tablespace related) (tentative patch and discussion) Alvaro Herrera <[email protected]>
2022-07-15 10:58 ` Re: standby recovery fails (tablespace related) (tentative patch and discussion) Alvaro Herrera <[email protected]>
2022-07-15 12:03 ` Re: standby recovery fails (tablespace related) (tentative patch and discussion) Alvaro Herrera <[email protected]>
2022-07-20 10:50 ` Re: standby recovery fails (tablespace related) (tentative patch and discussion) Alvaro Herrera <[email protected]>
2022-07-20 15:25 ` Re: standby recovery fails (tablespace related) (tentative patch and discussion) Alvaro Herrera <[email protected]>
2022-07-20 16:34 ` Re: standby recovery fails (tablespace related) (tentative patch and discussion) Alvaro Herrera <[email protected]>
2022-07-21 11:01 ` Re: standby recovery fails (tablespace related) (tentative patch and discussion) Alvaro Herrera <[email protected]>
2022-07-21 11:14 ` Re: standby recovery fails (tablespace related) (tentative patch and discussion) Thomas Munro <[email protected]>
2022-07-21 11:20 ` Re: standby recovery fails (tablespace related) (tentative patch and discussion) Alvaro Herrera <[email protected]>
2022-07-21 11:25 ` Re: standby recovery fails (tablespace related) (tentative patch and discussion) Alvaro Herrera <[email protected]>
2022-07-22 01:02 ` Re: standby recovery fails (tablespace related) (tentative patch and discussion) Kyotaro Horiguchi <[email protected]>
2022-07-22 00:20 ` Re: standby recovery fails (tablespace related) (tentative patch and discussion) Kyotaro Horiguchi <[email protected]>
2022-07-22 08:18 ` Re: standby recovery fails (tablespace related) (tentative patch and discussion) Alvaro Herrera <[email protected]>
2022-07-22 08:49 ` Re: standby recovery fails (tablespace related) (tentative patch and discussion) Kyotaro Horiguchi <[email protected]>
2022-07-27 06:07 ` Re: standby recovery fails (tablespace related) (tentative patch and discussion) Alvaro Herrera <[email protected]>
2022-07-22 08:53 ` Re: standby recovery fails (tablespace related) (tentative patch and discussion) Thomas Munro <[email protected]>
2022-07-21 11:11 ` Re: standby recovery fails (tablespace related) (tentative patch and discussion) Thomas Munro <[email protected]>
2022-07-21 11:17 ` Re: standby recovery fails (tablespace related) (tentative patch and discussion) Alvaro Herrera <[email protected]>
2022-07-27 18:54 ` Re: standby recovery fails (tablespace related) (tentative patch and discussion) Alvaro Herrera <[email protected]>
2022-07-28 18:04 ` Re: standby recovery fails (tablespace related) (tentative patch and discussion) Matthias van de Meent <[email protected]>
2022-07-28 21:57 ` Re: standby recovery fails (tablespace related) (tentative patch and discussion) Tom Lane <[email protected]>
2022-07-28 23:27 ` Re: standby recovery fails (tablespace related) (tentative patch and discussion) Thomas Munro <[email protected]>
2022-07-29 05:20 ` Re: standby recovery fails (tablespace related) (tentative patch and discussion) Kyotaro Horiguchi <[email protected]>
2022-07-29 10:59 ` Re: standby recovery fails (tablespace related) (tentative patch and discussion) Alvaro Herrera <[email protected]>
2022-07-30 14:37 ` Re: standby recovery fails (tablespace related) (tentative patch and discussion) Tom Lane <[email protected]>
2022-07-30 16:51 ` Re: standby recovery fails (tablespace related) (tentative patch and discussion) Tom Lane <[email protected]>
2022-07-30 23:08 ` Re: standby recovery fails (tablespace related) (tentative patch and discussion) Thomas Munro <[email protected]>
2022-07-30 23:17 ` Re: standby recovery fails (tablespace related) (tentative patch and discussion) Tom Lane <[email protected]>
2022-07-31 03:46 ` Re: standby recovery fails (tablespace related) (tentative patch and discussion) Thomas Munro <[email protected]>
2025-12-09 18:44 [PATCH 5/5] Use background worker to do logical decoding. Antonin Houska <[email protected]>
2025-12-09 18:44 [PATCH 5/5] Use background worker to do logical decoding. Antonin Houska <[email protected]>
2025-12-13 18:27 [PATCH 5/6] Use background worker to do logical decoding. Antonin Houska <[email protected]>
2025-12-13 18:27 [PATCH 5/6] Use background worker to do logical decoding. Antonin Houska <[email protected]>
2026-01-08 16:47 [PATCH 5/6] Use background worker to do logical decoding. Antonin Houska <[email protected]>
2026-01-08 16:47 [PATCH 5/6] Use background worker to do logical decoding. Antonin Houska <[email protected]>
2026-01-12 16:30 [PATCH 5/6] Use background worker to do logical decoding. Antonin Houska <[email protected]>
2026-01-12 16:30 [PATCH 5/6] Use background worker to do logical decoding. Antonin Houska <[email protected]>
2026-01-22 08:23 [PATCH 5/6] Use background worker to do logical decoding. Antonin Houska <[email protected]>
2026-01-22 08:23 [PATCH 5/6] Use background worker to do logical decoding. Antonin Houska <[email protected]>
2026-01-27 10:48 [PATCH 5/6] Use background worker to do logical decoding. Antonin Houska <[email protected]>
2026-01-27 10:48 [PATCH v33 5/5] Use background worker to do logical decoding. Antonin Houska <[email protected]>
2026-01-27 10:48 [PATCH 5/6] Use background worker to do logical decoding. Antonin Houska <[email protected]>
2026-01-27 10:48 [PATCH v33 5/5] Use background worker to do logical decoding. Antonin Houska <[email protected]>
2026-02-16 19:49 [PATCH v35 5/5] Use background worker to do logical decoding. Antonin Houska <[email protected]>
2026-02-16 19:49 [PATCH 5/5] Use background worker to do logical decoding. Antonin Houska <[email protected]>
2026-02-16 19:49 [PATCH 5/5] Use background worker to do logical decoding. Antonin Houska <[email protected]>
2026-02-16 19:49 [PATCH v35 5/5] Use background worker to do logical decoding. Antonin Houska <[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