Horton

Re: stress test for parallel workers

Tom Lane <tgl@sss.pgh.pa.us>

View thread View on postgr.es
From: Tom Lane <tgl@sss.pgh.pa.us>
To: Thomas Munro <thomas.munro@gmail.com>
Cc: Andrew Dunstan <andrew.dunstan@2ndquadrant.com>, Mark Wong <mark@2ndquadrant.com>, Justin Pryzby <pryzby@telsasoft.com>, Andres Freund <andres@anarazel.de>, pgsql-hackers <pgsql-hackers@postgresql.org>
Date: 2019-10-11T20:13:46Z
Lists: pgsql-hackers 
Thomas Munro <thomas.munro@gmail.com> writes:
> Yeah, I don't know anything about this stuff, but I was also beginning
> to wonder if something is busted in the arch-specific fault.c code
> that checks if stack expansion is valid[1], in a way that fails with a
> rapidly growing stack, well timed incoming signals, and perhaps
> Docker/LXC (that's on Mark's systems IIUC, not sure about the ARM
> boxes that failed or if it could be relevant here).  Perhaps the
> arbitrary tolerances mentioned in that comment are relevant.
> [1] https://github.com/torvalds/linux/blob/master/arch/powerpc/mm/fault.c#L244

Hm, the bit about "we'll allow up to 1MB unconditionally" sure seems
to match up with the observations here.  I also wonder about the
arbitrary definition of "a long way" as 2KB.  Could it be that that
misbehaves in the face of a userland function with more than 2KB of
local variables?

It's not very clear how those things would lead to an intermittent
failure though.  In the case of the postmaster crashes, we now see
that timing of signal receipts is relevant.  For infinite_recurse,
maybe it only fails if an sinval interrupt happens at the wrong time?
(This theory would predict that commit 798070ec0 made the problem
way more prevalent than it had been ... need to go see if the
buildfarm history supports that.)

			regards, tom lane

Commits

  1. In the postmaster, rely on the signal infrastructure to block signals.

  2. Paper over regression failures in infinite_recurse() on PPC64 Linux.

  3. Hack pg_ctl to report postmaster's exit status.

  4. Re-order some regression test scripts for more parallelism.