Re: heavily contended lwlocks with long wait queues scale badly
Andres Freund <andres@anarazel.de>
From: Andres Freund <andres@anarazel.de>
To: Zhihong Yu <zyu@yugabyte.com>
Cc: Bharath Rupireddy <bharath.rupireddyforpostgres@gmail.com>, pgsql-hackers@postgresql.org, Thomas Munro <thomas.munro@gmail.com>, Yura Sokolov <y.sokolov@postgrespro.ru>
Date: 2022-11-01T00:19:34Z
Lists: pgsql-hackers
Hi, On 2022-10-31 17:17:03 -0700, Zhihong Yu wrote: > On Mon, Oct 31, 2022 at 4:51 PM Andres Freund <andres@anarazel.de> wrote: > > > Hi, > > > > On 2022-10-31 16:21:06 +0530, Bharath Rupireddy wrote: > > > BTW, I've seen a sporadic crash (SEGV) with the patch in bg writer > > > with the same set up [1], I'm not sure if it's really because of the > > > patch. I'm unable to reproduce it now and unfortunately I didn't > > > capture further details when it occurred. > > > > That's likely because the prototype patch I submitted in this thread missed > > updating LWLockUpdateVar(). > > > > Updated patch attached. > > > > Greetings, > > > > Andres Freund > > > > Hi, > Minor comment: > > + uint8 lwWaiting; /* see LWLockWaitState */ > > Why not declare `lwWaiting` of type LWLockWaitState ? Unfortunately C99 (*) doesn't allow to specify the width of an enum field. With most compilers we'd end up using 4 bytes. Greetings, Andres Freund (*) C++ has allowed specifying this for quite a few years now and I think C23 will support it too, but that doesn't help us at this point.
Commits
-
lwlock: Fix quadratic behavior with very long wait lists
- 81038228582a 12.18 landed
- dc9d424cf0cd 13.14 landed
- 5f6ec27a6477 14.11 landed
- f374fb4aab3e 15.6 landed
- a4adc31f6902 16.0 landed