On Mon, Sep 06, 2021 at 10:02:11PM +0200, Mark Wielaard wrote:
> This doesn't really explain the above failures, but there were still
> two tests that didn't use their own debuginfod client cache. Also one
> set a useless, but confusing DEBUGINFOD_URLS.
> 
> Still looking for explanations of the other FAILS (which I cannot
> reproduce locally).

As expected there are still several failures on the buildbot workers,
but I did manage to replicate one failure locally. It happened because
PORT1 and PORT2 were equal.

    ss -atn | fgrep ":$PORT2" && $PORT1 -ne $PORT2 || break

Does actually break when PORT1 == PORT2. The && -ne should have been
|| -eq. But just simplify the port selection a bit to use
non-overlapping ranges (split the 100 possible ports in 0-49 for port1
and 50-99 for port2).

Also because of a typo on error only the metrics of port1 were
shown. Fix the typo so we can better diagnose the remaining failures.

Cheers,

Mark