public inbox for [email protected]
help / color / mirror / Atom feedFrom: Aya Iwata (Fujitsu) <[email protected]>
To: 'Michael Paquier' <[email protected]>
To: Chao Li <[email protected]>
Cc: 'Peter Smith' <[email protected]>
Cc: Hayato Kuroda (Fujitsu) <[email protected]>
Cc: pgsql-hackers <[email protected]>
Subject: RE: [PROPOSAL] Termination of Background Workers for ALTER/DROP DATABASE
Date: Thu, 16 Oct 2025 15:04:09 +0000
Message-ID: <OS7PR01MB119642C7C0E94E767DB7000C1EAE9A@OS7PR01MB11964.jpnprd01.prod.outlook.com> (raw)
In-Reply-To: <[email protected]>
References: <[email protected]>
<OSCPR01MB1496614832F8014EC16FB78D2F5EEA@OSCPR01MB14966.jpnprd01.prod.outlook.com>
<[email protected]>
<OSCPR01MB14966EC12277712131EB8EDF1F5EEA@OSCPR01MB14966.jpnprd01.prod.outlook.com>
<OS7PR01MB11964C8FE9CDCC0F4C9110988EAEEA@OS7PR01MB11964.jpnprd01.prod.outlook.com>
<[email protected]>
<OS7PR01MB11964C077A3E61F4887DD3D09EAEFA@OS7PR01MB11964.jpnprd01.prod.outlook.com>
<CAHut+PtbOP_80OPZXCUZO=-pBJSRTmHcQ2MnVTFov1meNbw18Q@mail.gmail.com>
<CAHut+Pt5BN0LDh7OzbNqh9+zqHBgsrLX+vh-gn+3FKYTFHMvhw@mail.gmail.com>
<TY3PR01MB11969CBD6DF3E0DB820AD4262EAE8A@TY3PR01MB11969.jpnprd01.prod.outlook.com>
<[email protected]>
Hi Chao-san, Michael san
Thank you for your comments! To accept your comment, I updated patch to v0008.
> From: Chao Li <[email protected]>
> Sent: Wednesday, October 15, 2025 12:37 PM
...
> By searching for “ByOid”, we can get some existing examples:
>
> ObjectAddress
> RefreshMatViewByOid(Oid matviewOid, bool is_create, bool skipData,
> bool concurrent, const char *queryString,
> QueryCompletion *qc)
>
> The function name clearly tells refresh MatView by Oid, so the oid in parameter is an old of mat view.
>
> ResultRelInfo *
> ExecLookupResultRelByOid(ModifyTableState *node, Oid resultoid,
> bool missing_ok, bool update_cache)
>
> The function name indicates ResultRel, so the oid is a result oid.
>
> AccessMethodInfo *
> findAccessMethodByOid(Oid oid)
>
> The function name tells to find access method, the the oid is an access method’s OID.
>
> You can find more …
>
> But in this patch, the function name only indeeds “terminate background workers”, while the oid is a database oid. Maybe we can rename the
> function to “TerminateDatabaseBgWorkersByOid()”.
Thank you. I changed the function name to "'TerminateBgWorkersByDbOid".
I prefer this name because there are not official terminology "Database background worker" and it's shorter.
> -----Original Message-----
> From: Michael Paquier <[email protected]>
> Sent: Thursday, October 16, 2025 12:55 PM
...
> On Wed, Oct 15, 2025 at 02:48:43AM +0000, Aya Iwata (Fujitsu) wrote:
> + * Exit the bgworker when its database is dropped, renamed, moved to a
> + * different tablespace, or used as a template for CREATE DATABASE.
>
> I don't think that we need to list all these operations in details
> here. We could just say "if its database is involved in a CREATE,
> ALTER or DROP database command". The docs should provide these
> details, of course.
Thank you. I fixed this .h file comment.
>
> +#define BGWORKER_EXIT_AT_DATABASE_CHANGE 0x0004
>
> Flag name works here.
Sorry, I cannot follow. Please tell me more details about this comment.
> # XXX This spends more than 5 seconds because the backend retries counting
> # number of connecting processes 50 times. See CountOtherDBBackends().
>
> And that's annoying. Let's activate what I call the cheat mode for
> this one: an injection point that, if defined, enforces a lower number
> of tries when we loop over the workers to stop. That would make the
> test much faster when using a worker that should not be stopped,
> without impacting the coverage.
I tried to implement your idea. Thanks Kuroda-san to help it.
> I suspect that your new test 002_worker_terminate.pl has a race
> condition in run_db_command(): are you sure that the bgworker has
> enough time to be reported as stopped in the server logs once
> safe_psql() finishes to run the database command given by the caller?
> On very slow and/or loaded machines, particularly, that could hurt the
> stability. It seems to me that this should use a wait_for_log()
> instead of a log_contains(), waiting for the worker to be reported as
> stopped depending on the command executed.
I fixed this test to use wait_for_log() instead of log_contains().
> Shouldn't this test also check that worker 0 (the one that does not
> have the flag set) is still running at the end of the test? I assume
> that querying pg_stat_activity would be enough at the end of the
> script.
Added. I cannot find a good way to clarify the worker is "worker 0" from the pg_stat_activity,
backend_type does not have the information. Thus I used the string "worker_spi dynamic" as the key.
Best regards,
Aya Iwata
Fujitsu Limited
Attachments:
[application/octet-stream] v0008-0001-Allow-background-workers-to-be-terminated.patch (13.1K, 2-v0008-0001-Allow-background-workers-to-be-terminated.patch)
download | inline diff:
From bf6e3f9d70621017e52bda41ddee5d0a33895639 Mon Sep 17 00:00:00 2001
From: "iwata.aya" <[email protected]>
Date: Thu, 11 Sep 2025 21:16:51 +0900
Subject: [PATCH v0008] Allow background workers to be terminated at DROP
DATABASE
---
doc/src/sgml/bgworker.sgml | 19 +++
src/backend/postmaster/bgworker.c | 41 ++++++
src/backend/storage/ipc/procarray.c | 23 ++-
src/include/postmaster/bgworker.h | 8 +
src/test/modules/worker_spi/Makefile | 4 +
src/test/modules/worker_spi/meson.build | 4 +
.../worker_spi/t/002_worker_terminate.pl | 139 ++++++++++++++++++
.../modules/worker_spi/worker_spi--1.0.sql | 3 +-
src/test/modules/worker_spi/worker_spi.c | 5 +
9 files changed, 243 insertions(+), 3 deletions(-)
create mode 100644 src/test/modules/worker_spi/t/002_worker_terminate.pl
diff --git a/doc/src/sgml/bgworker.sgml b/doc/src/sgml/bgworker.sgml
index 2c393385a91..6f4fc57e3d9 100644
--- a/doc/src/sgml/bgworker.sgml
+++ b/doc/src/sgml/bgworker.sgml
@@ -108,6 +108,25 @@ typedef struct BackgroundWorker
</listitem>
</varlistentry>
+ <varlistentry>
+ <term><literal>BGWORKER_EXIT_AT_DATABASE_CHANGE</literal></term>
+ <listitem>
+ <para>
+ <indexterm><primary>BGWORKER_EXIT_AT_DATABASE_CHANGE</primary></indexterm>
+ Requests termination of the background worker when its connected database is
+ dropped, renamed, moved to a different tablespace, or used as a template for
+ <command>CREATE DATABASE</command>. Specifically, the postmaster sends a
+ termination signal when any of these commands affect the worker's database:
+ <command>DROP DATABASE</command>,
+ <command>ALTER DATABASE RENAME TO</command>,
+ <command>ALTER DATABASE SET TABLESPACE</command>, or
+ <command>CREATE DATABASE</command>.
+ Requires both <literal>BGWORKER_SHMEM_ACCESS</literal> and
+ <literal>BGWORKER_BACKEND_DATABASE_CONNECTION</literal>.
+ </para>
+ </listitem>
+ </varlistentry>
+
</variablelist>
</para>
diff --git a/src/backend/postmaster/bgworker.c b/src/backend/postmaster/bgworker.c
index 1ad65c237c3..73618a47991 100644
--- a/src/backend/postmaster/bgworker.c
+++ b/src/backend/postmaster/bgworker.c
@@ -26,6 +26,7 @@
#include "storage/lwlock.h"
#include "storage/pmsignal.h"
#include "storage/proc.h"
+#include "storage/procarray.h"
#include "storage/procsignal.h"
#include "storage/shmem.h"
#include "tcop/tcopprot.h"
@@ -1396,3 +1397,43 @@ GetBackgroundWorkerTypeByPid(pid_t pid)
return result;
}
+
+/*
+ * Terminate all background workers connected to the given database, if they
+ * had requested it.
+ */
+
+void
+TerminateBgWorkersByDbOid(Oid oid)
+{
+ bool signal_postmaster = false;
+
+ LWLockAcquire(BackgroundWorkerLock, LW_EXCLUSIVE);
+
+ /*
+ * Iterate through slots, looking for workers connected to the given
+ * database.
+ */
+ for (int slotno = 0; slotno < BackgroundWorkerData->total_slots; ++slotno)
+ {
+ BackgroundWorkerSlot *slot = &BackgroundWorkerData->slot[slotno];
+
+ if (slot->in_use &&
+ (slot->worker.bgw_flags & BGWORKER_EXIT_AT_DATABASE_CHANGE))
+ {
+ PGPROC *proc = BackendPidGetProc(slot->pid);
+
+ if (proc && proc->databaseId == oid)
+ {
+ slot->terminate = true;
+ signal_postmaster = true;
+ }
+ }
+ }
+
+ LWLockRelease(BackgroundWorkerLock);
+
+ /* Make sure the postmaster notices the change to shared memory. */
+ if (signal_postmaster)
+ SendPostmasterSignal(PMSIGNAL_BACKGROUND_WORKER_CHANGE);
+}
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index 200f72c6e25..397c59b3f5f 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -56,11 +56,13 @@
#include "catalog/pg_authid.h"
#include "miscadmin.h"
#include "pgstat.h"
+#include "postmaster/bgworker.h"
#include "port/pg_lfind.h"
#include "storage/proc.h"
#include "storage/procarray.h"
#include "utils/acl.h"
#include "utils/builtins.h"
+#include "utils/injection_point.h"
#include "utils/lsyscache.h"
#include "utils/rel.h"
#include "utils/snapmgr.h"
@@ -3717,8 +3719,19 @@ CountOtherDBBackends(Oid databaseId, int *nbackends, int *nprepared)
int autovac_pids[MAXAUTOVACPIDS];
int tries;
- /* 50 tries with 100ms sleep between tries makes 5 sec total wait */
- for (tries = 0; tries < 50; tries++)
+ /*
+ * Usually, we try 50 times with 100ms sleep between tries, making 5 sec
+ * total wait. If requested, it would be reduced to 10 times to shorten the
+ * test time.
+ */
+ int ntries = 50;
+
+#ifdef USE_INJECTION_POINTS
+ if (IS_INJECTION_POINT_ATTACHED("reduce-ncounts"))
+ ntries = 10;
+#endif
+
+ for (tries = 0; tries < ntries; tries++)
{
int nautovacs = 0;
bool found = false;
@@ -3768,6 +3781,12 @@ CountOtherDBBackends(Oid databaseId, int *nbackends, int *nprepared)
for (index = 0; index < nautovacs; index++)
(void) kill(autovac_pids[index], SIGTERM); /* ignore any error */
+ /*
+ * Terminate all background workers for this database, if they had
+ * requested it (BGWORKER_EXIT_AT_DATABASE_CHANGE)
+ */
+ TerminateBgWorkersByDbOid(databaseId);
+
/* sleep, then try again */
pg_usleep(100 * 1000L); /* 100ms */
}
diff --git a/src/include/postmaster/bgworker.h b/src/include/postmaster/bgworker.h
index 058667a47a0..4bedb057fec 100644
--- a/src/include/postmaster/bgworker.h
+++ b/src/include/postmaster/bgworker.h
@@ -59,6 +59,13 @@
*/
#define BGWORKER_BACKEND_DATABASE_CONNECTION 0x0002
+/*
+ * Exit the bgworker if its database is involved in a CREATE, ALTER or DROP
+ * database command.
+ * Requires BGWORKER_SHMEM_ACCESS and BGWORKER_BACKEND_DATABASE_CONNECTION.
+ */
+#define BGWORKER_EXIT_AT_DATABASE_CHANGE 0x0004
+
/*
* This class is used internally for parallel queries, to keep track of the
* number of active parallel workers and make sure we never launch more than
@@ -128,6 +135,7 @@ extern const char *GetBackgroundWorkerTypeByPid(pid_t pid);
/* Terminate a bgworker */
extern void TerminateBackgroundWorker(BackgroundWorkerHandle *handle);
+extern void TerminateBgWorkersByDbOid(Oid oid);
/* This is valid in a running worker */
extern PGDLLIMPORT BackgroundWorker *MyBgworkerEntry;
diff --git a/src/test/modules/worker_spi/Makefile b/src/test/modules/worker_spi/Makefile
index 024b34cdbb3..e7c5c059e32 100644
--- a/src/test/modules/worker_spi/Makefile
+++ b/src/test/modules/worker_spi/Makefile
@@ -6,6 +6,10 @@ EXTENSION = worker_spi
DATA = worker_spi--1.0.sql
PGFILEDESC = "worker_spi - background worker example"
+EXTRA_INSTALL = src/test/modules/injection_points
+
+export enable_injection_points
+
TAP_TESTS = 1
ifdef USE_PGXS
diff --git a/src/test/modules/worker_spi/meson.build b/src/test/modules/worker_spi/meson.build
index d673ece48a0..5ba66051396 100644
--- a/src/test/modules/worker_spi/meson.build
+++ b/src/test/modules/worker_spi/meson.build
@@ -26,8 +26,12 @@ tests += {
'sd': meson.current_source_dir(),
'bd': meson.current_build_dir(),
'tap': {
+ 'env': {
+ 'enable_injection_points': get_option('injection_points') ? 'yes' : 'no',
+ },
'tests': [
't/001_worker_spi.pl',
+ 't/002_worker_terminate.pl'
],
},
}
diff --git a/src/test/modules/worker_spi/t/002_worker_terminate.pl b/src/test/modules/worker_spi/t/002_worker_terminate.pl
new file mode 100644
index 00000000000..20987ff3a3b
--- /dev/null
+++ b/src/test/modules/worker_spi/t/002_worker_terminate.pl
@@ -0,0 +1,139 @@
+# Copyright (c) 2025, PostgreSQL Global Development Group
+
+# Test background workers can be terminated by db commands
+
+use strict;
+use warnings FATAL => 'all';
+use PostgreSQL::Test::Cluster;
+use PostgreSQL::Test::Utils;
+use Test::More;
+
+# This test depends on injection points to detect whether background workers
+# remain.
+if ($ENV{enable_injection_points} ne 'yes')
+{
+ plan skip_all => 'Injection points not supported by this build';
+}
+
+# Ensure the worker_spi dynamic worker is launched on the specified database
+sub launch_bgworker
+{
+ my ($node, $database, $testcase, $request_terminate) = @_;
+ my $offset = -s $node->logfile;
+
+ # Launch a background worker on the given database
+ my $result = $node->safe_psql(
+ $database, qq(
+ SELECT worker_spi_launch($testcase, oid, 0, '{}', $request_terminate) IS NOT NULL
+ FROM pg_database WHERE datname = '$database';
+ ));
+ is($result, 't', "dynamic bgworker launched");
+
+ # Check the worker is surely initialized
+ $node->wait_for_log(
+ qr/LOG: worker_spi dynamic worker $testcase initialized with .*\..*/,
+ $offset);
+}
+
+# Run the given query and verify the background worker can be terminated
+sub run_db_command
+{
+ my ($node, $command, $testname) = @_;
+ my $offset = -s $node->logfile;
+
+ $node->safe_psql('postgres', $command);
+
+ $node->wait_for_log(
+ qr/terminating background worker \"worker_spi dynamic\" due to administrator command/,
+ $offset);
+
+ note("background worker can be terminated at $testname");
+}
+
+my $node = PostgreSQL::Test::Cluster->new('mynode');
+$node->init;
+$node->start;
+
+# Check if the extension injection_points is available, as it may be
+# possible that this script is run with installcheck, where the module
+# would not be installed by default.
+if (!$node->check_extension('injection_points'))
+{
+ plan skip_all => 'Extension injection_points not installed';
+}
+
+$node->safe_psql('postgres', 'CREATE EXTENSION worker_spi;');
+
+# Launch a background worker without BGWORKER_EXIT_AT_DATABASE_CHANGE
+launch_bgworker($node, 'postgres', 0, "false");
+
+# Ensure CREATE DATABASE WITH TEMPLATE fails because background worker retains
+
+# Firstly register an injection point to make the test faster. Normally, it
+# spends more than 5 seconds because the backend retries, counting the number
+# of connecting processes 50 times, but now the counting would be done only 10
+# times. See CountOtherDBBackends().
+$node->safe_psql('postgres', "CREATE EXTENSION injection_points;");
+$node->safe_psql('postgres',
+ "SELECT injection_points_attach('reduce-ncounts', 'error');");
+
+my $stderr;
+
+$node->psql(
+ 'postgres',
+ "CREATE DATABASE testdb WITH TEMPLATE postgres",
+ stderr => \$stderr);
+ok( $stderr =~
+ "source database \"postgres\" is being accessed by other users",
+ "background worker blocked the database creation");
+
+# Confirm a background worker is still running
+$node->safe_psql(
+ "postgres", qq(
+ SELECT count(1) FROM pg_stat_activity
+ WHERE backend_type = 'worker_spi dynamic';));
+
+# Terminate the worker for upcoming tests
+$node->safe_psql(
+ "postgres", qq(
+ SELECT pg_terminate_backend(pid)
+ FROM pg_stat_activity WHERE backend_type = 'worker_spi dynamic';));
+
+# The injection point won't be used anymore, release it.
+$node->safe_psql('postgres',
+ "SELECT injection_points_detach('reduce-ncounts');");
+
+# Ensure BGWORKER_EXIT_AT_DATABASE_CHANGE allows background workers to be
+# terminated at some database manipulations.
+#
+# Testcase 1: CREATE DATABASE WITH TEMPLATE
+launch_bgworker($node, 'postgres', 1, "true");
+run_db_command(
+ $node,
+ "CREATE DATABASE testdb WITH TEMPLATE postgres",
+ "CREATE DATABASE WITH TEMPLATE");
+
+# Testcase 2: ALTER DATABASE RENAME
+launch_bgworker($node, 'testdb', 2, "true");
+run_db_command(
+ $node,
+ "ALTER DATABASE testdb RENAME TO renameddb",
+ "ALTER DATABASE RENAME");
+
+# Preparation for the next test; create another tablespace
+my $tablespace = PostgreSQL::Test::Utils::tempdir;
+$node->safe_psql('postgres',
+ "CREATE TABLESPACE test_tablespace LOCATION '$tablespace'");
+
+# Testcase 3: ALTER DATABASE SET TABLESPACE
+launch_bgworker($node, 'renameddb', 3, "true");
+run_db_command(
+ $node,
+ "ALTER DATABASE renameddb SET TABLESPACE test_tablespace",
+ "ALTER DATABASE SET TABLESPACE");
+
+# Testcase 4: DROP DATABASE
+launch_bgworker($node, 'renameddb', 4, "true");
+run_db_command($node, "DROP DATABASE renameddb", "DROP DATABASE");
+
+done_testing();
diff --git a/src/test/modules/worker_spi/worker_spi--1.0.sql b/src/test/modules/worker_spi/worker_spi--1.0.sql
index 84deb6199f6..3d12de37bea 100644
--- a/src/test/modules/worker_spi/worker_spi--1.0.sql
+++ b/src/test/modules/worker_spi/worker_spi--1.0.sql
@@ -7,7 +7,8 @@
CREATE FUNCTION worker_spi_launch(index int4,
dboid oid DEFAULT 0,
roleoid oid DEFAULT 0,
- flags text[] DEFAULT '{}')
+ flags text[] DEFAULT '{}',
+ request_termination boolean DEFAULT false)
RETURNS pg_catalog.int4 STRICT
AS 'MODULE_PATHNAME'
LANGUAGE C;
diff --git a/src/test/modules/worker_spi/worker_spi.c b/src/test/modules/worker_spi/worker_spi.c
index bea8339f464..2912abe6cce 100644
--- a/src/test/modules/worker_spi/worker_spi.c
+++ b/src/test/modules/worker_spi/worker_spi.c
@@ -404,10 +404,15 @@ worker_spi_launch(PG_FUNCTION_ARGS)
Size ndim;
int nelems;
Datum *datum_flags;
+ bool request_termination = PG_GETARG_BOOL(4);
memset(&worker, 0, sizeof(worker));
worker.bgw_flags = BGWORKER_SHMEM_ACCESS |
BGWORKER_BACKEND_DATABASE_CONNECTION;
+
+ if (request_termination)
+ worker.bgw_flags |= BGWORKER_EXIT_AT_DATABASE_CHANGE;
+
worker.bgw_start_time = BgWorkerStart_RecoveryFinished;
worker.bgw_restart_time = BGW_NEVER_RESTART;
sprintf(worker.bgw_library_name, "worker_spi");
--
2.39.3
view thread (67+ messages) latest in thread
reply
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Reply to all the recipients using the --to and --cc options:
reply via email
To: [email protected]
Cc: [email protected], [email protected], [email protected], [email protected], [email protected]
Subject: RE: [PROPOSAL] Termination of Background Workers for ALTER/DROP DATABASE
In-Reply-To: <OS7PR01MB119642C7C0E94E767DB7000C1EAE9A@OS7PR01MB11964.jpnprd01.prod.outlook.com>
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
This inbox is served by agora; see mirroring instructions
for how to clone and mirror all data and code used for this inbox