From: | Ashutosh Bapat <ashutosh(dot)bapat(dot)oss(at)gmail(dot)com> |
---|---|
To: | "Michael J(dot) Baars" <mjbaars1977(dot)pgsql(at)gmail(dot)com> |
Cc: | pgsql-hackers(at)lists(dot)postgresql(dot)org |
Subject: | Re: Upgrade from Fedora 40 to Fedora 42, or from PostgreSQL 16.3 to PostgreSQL 16.9 |
Date: | 2025-07-18 12:17:17 |
Message-ID: | CAExHW5uDOt1qpEy-BhKCh8sw=jPS=SrsiKdbVHjRpswtFGo+ew@mail.gmail.com |
Views: | Whole Thread | Raw Message | Download mbox | Resend email |
Thread: | |
Lists: | pgsql-hackers |
Hi Michael,
On Fri, Jul 18, 2025 at 3:01 PM Michael J. Baars
<mjbaars1977(dot)pgsql(at)gmail(dot)com> wrote:
>
> Hello fellow PostgreSQL users and developers,
>
> I installed a new Fedora release last week and ran into a peculiar problem:
>
> Each night I need to run computations through about 70.000 rows and using release 16.3 that took about 4 hours to complete, but using release 16.9 the same computations now take about 8 hours to complete.
>
> I've checked the unix sockets, inet sockets and the NVMe drive for deviations between the different kernels, but found none. Configuration files are exactly identical as well. Somewhere in between release 16.3 and release 16.9, changes must have been implemented that make the execution engine about two times slower than it was.
>
> Two particular queries, that run about 70.000 times each night, gave the following results:
>
> Query 1 over unix sockets:
>
> release 16.3:
> Time: 194.093 ms
> Time: 169.714 ms
> Time: 174.424 ms
> Time: 170.285 ms
>
> release 16.9:
> Time: 281.554 ms
> Time: 290.257 ms
> Time: 290.886 ms
> Time: 287.528 ms
>
> Query 2 over unix sockets:
>
> release 16.3:
> Time: 132.500 ms
> Time: 133.182 ms
> Time: 134.511 ms
> Time: 130.818 ms
>
> release 16.9:
> Time: 251.238 ms
> Time: 247.749 ms
> Time: 246.862 ms
> Time: 248.541 ms
>
> These differences add up to about 4 hours additional computation time each night. Apparently I have to scale down on the amount of work that can be done in one night.
Can you please provide "explain analyze" output of both the queries
across both releases for investigation.
--
Best Wishes,
Ashutosh Bapat
From | Date | Subject | |
---|---|---|---|
Next Message | Bertrand Drouvot | 2025-07-18 13:39:15 | Verify predefined LWLocks tranches have entries in wait_event_names.txt |
Previous Message | Zhijie Hou (Fujitsu) | 2025-07-18 12:02:50 | RE: Conflict detection for update_deleted in logical replication |