Autobuilds now exceeding 5 hour limit

Andrew Bartlett abartlet at samba.org
Sun Mar 11 20:48:37 UTC 2018


On Mon, 2018-03-12 at 09:32 +1300, Douglas Bagnall via samba-technical
wrote:

> The last 4 failures, including this, seem to have failed because they
> exceeded the 5 hour timeout. This has never happened before, though we
> did start hitting the then 4 hour limit in January.
> 
> Why is it slower? I don't know. No doubt the combination of the ldb
> performance regression in ef240aaca0ef693a967 (committed 2017-12-20),
> meltdown/spectre mitigations, and the ever increasing number of tests
> don't help, but it seems as if something in the last few days has made
> it significantly worse.
> 
> Comparing the "TOP 10 slowest" list with one from December in a very
> unscientific manner, I see slowdowns everywhere, with the worst being
> in vlv, drs.link_conflicts, and ridalloc_exop.

I've been working to make the build more parallel, and I have patches
here:

https://github.com/abartlet/samba/commits/parallel-test-for-master

However it trades time vs disk space and we run out of space on the
Catalyst Cloud VMs, so we need to figure how to give that room to grow.

In the meantime I agree we need to investigate what made VLV regress so
badly and what changed just this week.

Thank you so much for the analysis!

Andrew Bartlett
-- 
Andrew Bartlett                       http://samba.org/~abartlet/
Authentication Developer, Samba Team  http://samba.org
Samba Developer, Catalyst IT          http://catalyst.net.nz/services/samba




More information about the samba-technical mailing list