Re: WAL Insertion Lock Improvements

On Wed, Jul 26, 2023 at 1:27 AM Andres Freund <andres(at)anarazel(dot)de> wrote:
>
> > 0001 has been now applied. I have done more tests while looking at
> > this patch since yesterday and was surprised to see higher TPS numbers
> > on HEAD with the same tests as previously, and the patch was still
> > shining with more than 256 clients.
>
> Just a small heads up:
>
> I just rebased my aio tree over the commit and promptly, on the first run, saw
> a hang. I did some debugging on that. Unfortunately repeated runs haven't
> repeated that hang, despite quite a bit of trying.

Hm. Please share workload details, test scripts, system info and any
special settings for running in my setup.

> The symptom I was seeing is that all running backends were stuck in
> LWLockWaitForVar(), even though the value they're waiting for had
> changed. Which obviously "shouldn't be possible".

Were the backends stuck there indefinitely? IOW, did they get into a deadlock?

> It's of course possible that this is AIO specific, but I didn't see anything
> in stacks to suggest that.
>
> I do wonder if this possibly exposed an undocumented prior dependency on the
> value update always happening under the list lock.

I'm going through the other thread mentioned by Michael Paquier. I'm
wondering if the deadlock issue illustrated here
https://www.postgresql.org/message-id/55BB50D3.9000702%40iki.fi is
showing up again, because 71e4cc6b8e reduced the contention on
waitlist lock and made things *a bit* faster.

--
Bharath Rupireddy
PostgreSQL Contributors Team
RDS Open Source Databases
Amazon Web Services: https://aws.amazon.com