Re: [PATCH] Implement timestamp'ed output on "make check"

[Pedro Franco de Carvalho <pedromfc at linux dot ibm dot com>]

Sergio Durigan Junior <sergiodj@redhat.com> writes:

> It is unfortunately not uncommon to have tests hanging on some of the
> BuildBot workers.  For example, the ppc64be/ppc64le+gdbserver builders
> are especially in a bad state when it comes to testing GDB/gdbserver,
> and we can have builds that take an absurd amount of time to
> finish (almost 1 week for one single build, for example).
>
> It may be hard to diagnose these failures, because sometimes we don't
> have access to the faulty systems, and other times we're just too busy
> to wait and check which test is actually hanging.  During one of our
> conversations about the topic, someone proposed that it would be a
> good idea to have a timestamp put together with stdout output, so that
> we can come back later and examine which tests are taking too long to
> complete.

Hello Sergio,

I've looked into the ppc64le native-gdbserver/native-extended-gdbserver
buildbot workers, it seems that they started getting slower with this
patch:

f19c7ff839d7a32ebb48482ae7d318fb46ca823d
GDBSERVER: Listen on a unix domain (instead of TCP) socket if requested.

I believe this patch has been reverted, but the builders haven't reached
the reverted point yet.  Would it make sense to make the workers resume
after the reverted patch?  This would be at commit
80e24d09860dbeba7d435b4a4f0990f85dbc084e.

They did also get even slower after this patch:

fe1a5cad302b5535030cdf62895e79512713d738
[gdb/testsuite] Log wait status on process no longer exists error

But I think it only happens in combination with the patch above (see
https://sourceware.org/ml/gdb-patches/2018-12/msg00060.html for a more
detailed explanation).

Thanks!!

--
Pedro Franco de Carvalho