database/postgres
1
0
Fork 0
mirror of https://github.com/postgres/postgres.git synced 2025-07-28 23:42:10 +03:00
Code Activity

Back-patch b1ffe3ff into REL_13_STABLE.

Browse Source 
This is a cherry pick of 4c8e00ae from the 14 branch into the 13 branch.

It avoids an assertion failure when ForwardSyncRequest() tries to
allocate memory while trying to compact the queue, if run in a critical
section.  RelationTruncate() gained a critical section in 38c579b0, and
could fail in that way in the 13 branch.

b1ffe3ff originally fixed the same problem with TruncateMultiXact(), but
for that case it only needed to go back as far as 14, where SLRUs
started using the sync request queue.  It also fixed a related deadlock
risk that doesn't apply in this case (this case doesn't wait), but it
might exist in theory and it doesn't hurt to keep the code the same as
later branches.

Author: Heikki Linnakangas <heikki.linnakangas@iki.fi> (original commit)
Reviewed-by: Michael Paquier <michael@paquier.xyz> (in this new context)
Reported-by: Yura Sokolov <y.sokolov@postgrespro.ru>
Discussion: https://postgr.es/m/f98aaa79-80b5-47c9-832a-31416a3a528b%40postgrespro.ru
This commit is contained in:
Thomas Munro
2025-01-09 10:03:10 +13:00
parent 417d41c658
commit a722252ef1
2 changed files with 11 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
src/backend/access/transam/xlog.c
View File

@ -9139,6 +9139,12 @@ CreateCheckPoint(int flags)
{
do
{
/*
* Keep absorbing fsync requests while we wait. There could even
* be a deadlock if we don't, if the process that prevents the
* checkpoint is trying to add a request to the queue.
*/
AbsorbSyncRequests();
pg_usleep(10000L); /* wait for 10 msec */
} while (HaveVirtualXIDsDelayingChkpt(vxids, nvxids));
}
@ -9151,6 +9157,7 @@ CreateCheckPoint(int flags)
{
do
{
AbsorbSyncRequests();
pg_usleep(10000L); /* wait for 10 msec */
} while (HaveVirtualXIDsDelayingChkptEnd(vxids, nvxids));
}