BUG #15791: WalSndCtl->sync_standbys_defined modified too late
The following bug has been logged on the website:
Bug reference: 15791
Logged by: DamionZ Zhao
Email address: zhq651@126.com
PostgreSQL version: 11.2
Operating system: linux
Description:
1.when commit transaction, we need to kwow it is sync or async . There is a
step : call function SyncRepWaitForLSN in function
RecordTransactionCommit.
void SyncRepWaitForLSN(XLogRecPtr lsn, bool commit)
{
....
if (!WalSndCtl->sync_standbys_defined ||
lsn <= WalSndCtl->lsn[mode] )
{
...
return;
}
}
2.when modify WalSndCtl->sync_standbys_defined?
2.1 in function SyncRepUpdateSyncStandbysDefined,
WalSndCtl->sync_standbys_defined will be modified.
----------------------------------------------------
void SyncRepUpdateSyncStandbysDefined(void)
{
bool sync_standbys_defined = SyncStandbysDefined();
if (sync_standbys_defined != WalSndCtl->sync_standbys_defined)
{
WalSndCtl->sync_standbys_defined = sync_standbys_defined;
LWLockRelease(SyncRepLock);
}
}
2.2 SyncRepUpdateSyncStandbysDefined call stack is:
------------------------------------------
(1)CheckpointerMain(start checkpoint)-
UpdateSharedMemoryConfig
SyncRepUpdateSyncStandbysDefined
(2)CheckPointGuts -- Flush all data in shared memory to disk, and fsync
CheckPointBuffers --Flush all dirty blocks in buffer pool to disk at
checkpoint time.
BufferSync -- Write out all dirty buffers in the pool.
CheckpointWriteDelay
UpdateSharedMemoryConfig
SyncRepUpdateSyncStandbysDefined
2.3 when any checkpoint situation happened, WalSndCtl->sync_standbys_defined
will be modified.
3.my question:
----------------
when modify synchronous_standby_names in postgres.conf, and no checkpoint
happenes, will synchronous_standby_names take effect?
Em dom, 5 de mai de 2019 às 05:57, PG Bug reporting form
<noreply@postgresql.org> escreveu:
when modify synchronous_standby_names in postgres.conf, and no checkpoint
happenes, will synchronous_standby_names take effect?
This is not a bug. This question is suitable for -hackers.
Short answer is: yes. When you modify synchronous_standby_names and
reload postgresql.conf, checkpointer get SIGHUP and it calls
UpdateSharedMemoryConfig() -- see CheckpointerMain() circa 'if
(got_SIGHUP)'.
--
Euler Taveira Timbira -
http://www.timbira.com.br/
PostgreSQL: Consultoria, Desenvolvimento, Suporte 24x7 e Treinamento
On Sun, May 05, 2019 at 10:16:39PM -0300, Euler Taveira wrote:
This is not a bug. This question is suitable for -hackers.
Short answer is: yes. When you modify synchronous_standby_names and
reload postgresql.conf, checkpointer get SIGHUP and it calls
UpdateSharedMemoryConfig() -- see CheckpointerMain() circa 'if
(got_SIGHUP)'.
There could be always some delay between the moments SIGHUP gets
processed across the different processes as we are talking mainly
about running ProcessConfigFile(OGC_SIGHUP) which can happen for
normal backends once every query, but the change will be effective.
--
Michael