Re: CI and test improvements - Mailing list pgsql-hackers

On Sun, Aug 28, 2022 at 02:28:02PM -0700, Andres Freund wrote:
> > > > @@ -71,8 +69,6 @@ task:
> > > >      fingerprint_key: ccache/freebsd
> > > >      reupload_on_changes: true
> > > >
> > > > -  # Workaround around performance issues due to 32KB block size
> > > > -  repartition_script: src/tools/ci/gcp_freebsd_repartition.sh
> > > >    create_user_script: |
> > > >      pw useradd postgres
> > > >      chown -R postgres:postgres .
> > > > --
> > >
> > > What's the story there - at some point that was important for performance
> > > because of the native block size triggering significant read-modify-write
> > > cycles with postres' writes. You didn't comment on it in the commit message.
> >
> > Well, I don't know the history, but it seems to be unneeded now.
> 
> It's possible it was mainly needed for testing with aio + dio. But also
> possible that an upgrade improved the situation since.

Maybe freebsd got faster as a result of the TAU CPUs?
https://mobile.twitter.com/cirrus_labs/status/1534982111568052240

I noticed because it's been *slower* the last ~24h since cirrusci
disabled TAU, as Thomas commit mentioned.
https://twitter.com/cirrus_labs/status/1572657320093712384

For example this CF entry:

https://cirrus-ci.com/github/postgresql-cfbot/postgresql/commitfest/39/3736
https://cirrus-ci.com/task/4670794365140992 5m36s - 4days ago
https://cirrus-ci.com/task/4974926233862144 5m25s - 3days ago
https://cirrus-ci.com/task/5561409034518528 5m29s - 2days ago
https://cirrus-ci.com/task/6432442008469504 9m19s - yesterday

CF_BOT's latest tasks seem to be fast again, since 1-2h ago.
https://cirrus-ci.com/build/5178906041909248 9m1s
https://cirrus-ci.com/build/4593160281128960 5m8s
https://cirrus-ci.com/build/4539845644124160 5m22s

The logs for July show when freebsd started "being fast":
https://cirrus-ci.com/github/postgresql-cfbot/postgresql/commitfest/38/3708
https://cirrus-ci.com/task/6316073015312384 10m25s Jul 13
https://cirrus-ci.com/task/5662878987452416 5m48s Jul 15

Maybe that changed in July rather than June because the TAU CPUs were
still not available in every region/zone (?)
https://cloud.google.com/compute/docs/regions-zones/

I have no idea if the TAU CPUs eliminate/mitigate the original
performance issue you had with AIO.  But they have such a large effect
on freebsd that it could now be the fastest task, if given more than 2
CPUs.

-- 
Justin