database/postgres
1
0
Fork 0
mirror of https://github.com/postgres/postgres.git synced 2025-07-31 22:04:40 +03:00
Code Activity

Speedup truncations of relation forks.

Browse Source 
When a relation is truncated, shared_buffers needs to be scanned
so that any buffers for the relation forks are invalidated in it.
Previously, shared_buffers was scanned for each relation forks, i.e.,
MAIN, FSM and VM, when VACUUM truncated off any empty pages
at the end of relation or TRUNCATE truncated the relation in place.
Since shared_buffers needed to be scanned multiple times,
it could take a long time to finish those commands especially
when shared_buffers was large.

This commit changes the logic so that shared_buffers is scanned only
one time for those three relation forks.

Author: Kirk Jamison
Reviewed-by: Masahiko Sawada, Thomas Munro, Alvaro Herrera, Takayuki Tsunakawa and Fujii Masao
Discussion: https://postgr.es/m/D09B13F772D2274BB348A310EE3027C64E2067@g01jpexmbkw24
This commit is contained in:
Fujii Masao
2019-09-24 17:31:26 +09:00
parent 2e5c83acbb
commit 6d05086c0a
10 changed files with 172 additions and 86 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
contrib/pg_visibility/pg_visibility.c
View File

@ -383,6 +383,8 @@ pg_truncate_visibility_map(PG_FUNCTION_ARGS)
{
Oid relid = PG_GETARG_OID(0);
Relation rel;
ForkNumber fork;
BlockNumber block;
rel = relation_open(relid, AccessExclusiveLock);
@ -392,7 +394,12 @@ pg_truncate_visibility_map(PG_FUNCTION_ARGS)
RelationOpenSmgr(rel);
rel->rd_smgr->smgr_vm_nblocks = InvalidBlockNumber;
visibilitymap_truncate(rel, 0);
block = visibilitymap_prepare_truncate(rel, 0);
if (BlockNumberIsValid(block))
{
fork = VISIBILITYMAP_FORKNUM;
smgrtruncate(rel->rd_smgr, &fork, 1, &block);
}
if (RelationNeedsWAL(rel))
{
@ -418,7 +425,7 @@ pg_truncate_visibility_map(PG_FUNCTION_ARGS)
* here and when we sent the messages at our eventual commit. However,
* we're currently only sending a non-transactional smgr invalidation,
* which will have been posted to shared memory immediately from within
* visibilitymap_truncate. Therefore, there should be no race here.
* smgr_truncate. Therefore, there should be no race here.
*
* The reason why it's desirable to release the lock early here is because
* of the possibility that someone will need to use this to blow away many