diff options
author | Michael Paquier <michael@paquier.xyz> | 2024-11-11 09:02:30 +0900 |
---|---|---|
committer | Michael Paquier <michael@paquier.xyz> | 2024-11-11 09:02:30 +0900 |
commit | 5d4298e75f252b162008ad0475f29349023573cb (patch) | |
tree | f878fbd8ff87b3e264139e85daf171f9c298cb79 /src/backend/access/transam/parallel.c | |
parent | 29d66b2d2f80dba0ac7fa5453d4da2dfbdd5ed2b (diff) | |
download | postgresql-5d4298e75f252b162008ad0475f29349023573cb.tar.gz postgresql-5d4298e75f252b162008ad0475f29349023573cb.zip |
pg_stat_statements: Avoid some locking during PGSS entry scans
A single PGSS entry's spinlock is used to be able to modify "counters"
without holding pgss->lock exclusively, as mentioned at the top of
pg_stat_statements.c and within pgssEntry.
Within a single pgssEntry, stats_since and minmax_stats_since are never
modified without holding pgss->lock exclusively, so there is no need to
hold an entry's spinlock when reading stats_since and
minmax_stats_since, as done when scanning all the PGSS entries for
function calls of pg_stat_statements().
This also restores the consistency between the code and the comments
about the entry's spinlock usage. This change is a performance
improvement (it can be argued that this is a logic bug), so there is no
need for a backpatch. This saves two instructions from being read while
holding an entry's spinlock.
Author: Karina Litskevich
Reviewed-by: Michael Paquier, wenhui qiu
Discussion: https://postgr.es/m/CACiT8ibhCmzbcOxM0v4pRLH3abk-95LPkt7_uC2JMP+miPjxsg@mail.gmail.com
Diffstat (limited to 'src/backend/access/transam/parallel.c')
0 files changed, 0 insertions, 0 deletions