upstream/postgresql
1
0
Fork 0
mirror of https://github.com/postgres/postgres.git synced 2026-06-09 00:32:10 -04:00
Code Issues Projects Releases Packages Wiki Activity Actions

Fix checksum state transition during promotion

Browse source 
When a standby is promoted to primary during checksum enabling when the
state is inprogress-on, the standby shall revert the state to off since
checksums weren't fully enabled at the time of the crash.  Consider the
following scenario:

  1) primary/standby cluster has checksums off
  2) primary starts enabling checksums
  3) primary moves to inprogress-on
  4) standby receives that and moves to inprogress-on too
  5) primary crashes
  6) standby gets promoted, and does the StartupXLOG thing
  7) standby moves from inprogress-on back to off

Any processes in the standby need to be informed at step 6 to change
state with a procsignalbarrier, else they will stay in inprogress-on
while new backends will see the state as off.  StartupXLOG failed to
emit a procsignalbarrier which caused inconsistent state in the node
promoted to primary.

Fixed by emitting a procsignalbarrier during promotion, and adding a
new test for this scenario.

Author: Daniel Gustafsson <daniel@yesql.se>
Reported-by: Tomas Vondra <tomas@vondra.me>
Discussion: https://postgr.es/m/f1281cf3-89a3-4936-9bc5-2a5a6291229f@vondra.me
This commit is contained in:
Daniel Gustafsson 2026-05-29 21:26:18 +02:00
parent 38470c2c1e
commit 5fee7cab1b
2 changed files with 62 additions and 1 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

5
src/backend/access/transam/xlog.c
View file

@ -6610,6 +6610,7 @@ StartupXLOG(void)
SetLocalDataChecksumState(XLogCtl->data_checksum_version);
SpinLockRelease(&XLogCtl->info_lck);
EmitAndWaitDataChecksumsBarrier(PG_DATA_CHECKSUM_OFF);
ereport(WARNING,
errmsg("enabling data checksums was interrupted"),
errhint("Data checksum processing must be manually restarted for checksums to be enabled."));
@ -6621,7 +6622,7 @@ StartupXLOG(void)
* checksums and we can move to off instead of prompting the user to
* perform any action.
*/
if (XLogCtl->data_checksum_version == PG_DATA_CHECKSUM_INPROGRESS_OFF)
else if (XLogCtl->data_checksum_version == PG_DATA_CHECKSUM_INPROGRESS_OFF)
{
XLogChecksums(PG_DATA_CHECKSUM_OFF);
@ -6629,6 +6630,8 @@ StartupXLOG(void)
XLogCtl->data_checksum_version = PG_DATA_CHECKSUM_OFF;
SetLocalDataChecksumState(XLogCtl->data_checksum_version);
SpinLockRelease(&XLogCtl->info_lck);
EmitAndWaitDataChecksumsBarrier(PG_DATA_CHECKSUM_OFF);
}
/*

58
src/test/modules/test_checksums/t/003_standby_restarts.pl
View file

@ -226,4 +226,62 @@ unlike(
qr/page verification failed,.+\d$/m,
"no checksum validation errors in standby log");
# ---------------------------------------------------------------------------
# Test that enforced state transitions during promotion (via StartupXLOG) are
# performed as expected. When the primary crashes during inprogress-on the
# standby should revert to off at promotion. In order to check the transition
# the test will keep an open psql session with the standby during promotion.
# The cluster is currently broken down from the previous test. Start up the
# primary as primary, disable checksums and create a new standby from that
# state.
$node_standby->clean_node();
$node_primary->start();
disable_data_checksums($node_primary, wait => 'off');
# Re-create a new streaming standby linking to primary. The replication slot
# name is reused from earlier but a fresh backup is taken
$backup_name = 'my_new_backup';
$node_primary->backup($backup_name);
$node_standby = PostgreSQL::Test::Cluster->new('standby_restarts_standby');
$node_standby->init_from_backup($node_primary, $backup_name,
has_streaming => 1);
$node_standby->append_conf(
'postgresql.conf', qq[
primary_slot_name = '$slotname'
]);
$node_standby->start;
$node_primary->wait_for_catchup($node_standby, 'replay');
# Open a background psql connection on the primary and inject a barrier to
# block progress on to keep the state from advancing past inprogress-on
my $node_primary_bpsql = $node_primary->background_psql('postgres');
$node_primary_bpsql->query_safe('CREATE TEMPORARY TABLE tt (a integer);');
# Also open a background psql connection to the standby to make sure we have
# an active backend during promotion.
my $node_standby_bpsql = $node_standby->background_psql('postgres');
# Start to enable checksums and wait until both primary and standby have moved
# to the inprogress-on state. Processing will block here as the temporary rel
# barrier will block the primary from finishing.
enable_data_checksums($node_primary, wait => 'inprogress-on');
$node_primary->wait_for_catchup($node_standby, 'replay');
test_checksum_state($node_standby, 'inprogress-on');
# Crash the primary before checksums are enabled and promote the standby. The
# new primary node will now revert the state of 'off' since checksums weren't
# fully enabled during the crash.
$node_primary->teardown_node();
$node_standby->promote;
wait_for_checksum_state($node_standby, 'off');
# Ensure that the any backend which was active before, and during, promotion
# sees the new state.
$result = $node_standby_bpsql->query_safe("SHOW data_checksums;");
is($result, 'off',
'ensure checksums are set to off after promotion during inprogress-on');
$node_standby_bpsql->quit;
$node_standby->stop;
done_testing();