Re: pgstat wait timeout (RE: contrib/cache_scan)

On Wed, Mar 12, 2014 at 7:42 AM, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:

> Kouhei Kaigai <kaigai(at)ak(dot)jp(dot)nec(dot)com> writes:
> > WARNING: pgstat wait timeout
> > WARNING: pgstat wait timeout
> > WARNING: pgstat wait timeout
> > WARNING: pgstat wait timeout
>
> > Once I got above messages, write performance is dramatically
> > degraded, even though I didn't take detailed investigation.
>
> > I could reproduce it on the latest master branch without my
> > enhancement, so I guess it is not a problem something special
> > to me.
> > One other strangeness is, right now, this problem is only
> > happen on my virtual machine environment - VMware ESXi 5.5.0.
> > I couldn't reproduce the problem on my physical environment
> > (Fedora20, core i5-4570S).
>
> We've seen sporadic reports of that sort of behavior for years, but no
> developer has ever been able to reproduce it reliably. Now that you've
> got a reproducible case, do you want to poke into it and see what's going
> on?
>

I didn't know we were trying to reproduce it, nor that it was a mystery.
Do anything that causes serious IO constipation, and you will probably see
that message. For example, turn off synchronous_commit and run the default
pgbench transaction at a large scale but that still comfortably fits in
RAM, and wait for a checkpoint sync phase to kick in.

The pgstat wait timeout is a symptom, not the cause.

Cheers,

Jeff