Second Jammy Jellyfish test rebuild

[Colin Watson]

On Thu, Mar 24, 2022 at 07:44:11AM -0700, Erich Eickmeyer wrote:
> I'm seeing a lot of failed arm64 builds without build logs. In the
> past when I've seen this, I've merely retried the build and it has
> succeeded. This shows a pattern in that it's likely we've got a buggy
> build server.

Yeah, this has been reported to us (Launchpad), but for the record we
have no handle on a fix at present.  The basic symptom is that
buildd-manager timed out trying to poll the builder (probably over
several attempts).  This is either because the builder crashed or hung
or something like that (possible under load, and sometimes particular
builds will repeatedly OOM a builder or similar), or because of a
network failure that meant we lost contact for too long, or because of a
buildd-manager bug.  Unfortunately it's extremely difficult even for us
to tell which at present.  Ubuntu test rebuilds tend to exacerbate this
sort of thing, because the whole system is under significantly more load
than usual.

It's something that we'll keep trying to debug, but it's not as simple
as a single buggy build machine, and so I want to set expectations that
this may not be something we can fix quickly.

-- 
Colin Watson (he/him)                              [cjwatson at ubuntu.com]