bug#41625: Sporadic guix-offload crashes due to EOF errors

[Ludovic Courtès]

Hi,

Marius Bakke <marius@gnu.org> skribis:

> Marius Bakke <marius@gnu.org> writes:
>
>> 'guix offload test' passes without problems.
>
> Not so fast, running it in a loop reveals the crash.
>
> There is a trace file in /root/offloadtest.trace on Berlin with such an
> occurence.  It looks like a timeout is reached shortly before the EOF
> error:
>
> 10139 poll([{fd=14, events=POLLIN|POLLOUT}], 1, 0) = 1 ([{fd=14, 
> revents=POLLOUT}])
> 10139 poll([{fd=14, events=POLLIN}], 1, 15000) = 0 (Timeout)
> 10139 write(2, "Backtrace:\n", 11)      = 11
>
> This seems to be from a different node than the one reported previously,
> as the preceding connect() was to this machine:
>
> 10139 connect(44, {sa_family=AF_INET, sin_port=htons(22), 
> sin_addr=inet_addr("141.80.167.186")}, 16) = -1 EINPROGRESS (Operation now in 
> progress)

So it looks like ‘connect’ fails and eventually we get an EOF object.
However, I don’t see where that EOF comes from because the return value
of ‘connect!’ (the Guile-SSH procedure) is properly checked.

Ludo’.