Re: relayd: random abnormal termination on daemon reload

Hi Rafael

I have a theory on the second issue. I'm not super-familiar with the
inner-workings of relayd but it would appear fatalx() is used to kill
the child process with an abnormal exit. This kicked off a sequence
where all the other relayd processes sequentially shut down.

However the error codes do not propagate to the parent process.

parent_shutdown() is used to perform the final exit and it always
exits with error code of 0.

This is why rcctl reported OK on the aborted reload.

Regards
Lloyd

Rafael Sadowski wrote:

> Yet another reload race issue.
> 
> Thanks for the report. The problem is caused by a faulty IMSG in the
> IMSG dispatcher that terminates with fatalx(). This happens when
> processes send IMGS during a reload. That would be okay if we didn't
> have races.
> 
> A quick fix would be to replace the fatalx with log_debug after a
> careful audit, and then, in a second step, fundamentally redesign the
> reload concept.
> 
> I'll take care of the first one for the release.
> 
> Rafael
>