Worker SIGABRT takes down all workers?

Worker SIGABRT takes down all workers?

[Jonathan del Strother]

Hi there,
We're trying to figure out a problem we're having where Ruby 2.1.2 /
ImageMagick is causing a SIGABRT in a unicorn worker process.  When it
does so, I see this in unicorn.log :

I, [2014-08-21T11:40:44.282905 #65706]  INFO -- : reaped
#<Process::Status: pid 66787 exit 0> worker=0
I, [2014-08-21T11:40:44.283289 #65706]  INFO -- : reaped
#<Process::Status: pid 66801 exit 0> worker=7
I, [2014-08-21T11:40:44.283583 #65706]  INFO -- : reaped
#<Process::Status: pid 66790 exit 0> worker=3
I, [2014-08-21T11:40:44.283871 #65706]  INFO -- : reaped
#<Process::Status: pid 66793 exit 0> worker=4
I, [2014-08-21T11:40:44.284198 #65706]  INFO -- : reaped
#<Process::Status: pid 66794 exit 0> worker=5
I, [2014-08-21T11:40:44.284538 #65706]  INFO -- : reaped
#<Process::Status: pid 66798 exit 0> worker=6
I, [2014-08-21T11:40:44.284832 #65706]  INFO -- : reaped
#<Process::Status: pid 66788 exit 0> worker=1
E, [2014-08-21T11:40:44.385491 #65706] ERROR -- : reaped
#<Process::Status: pid 66789 SIGABRT (signal 6) (core dumped)>
worker=2
I, [2014-08-21T11:40:44.385951 #65706]  INFO -- : master complete
I, [2014-08-21T11:40:48.073191 #72528]  INFO -- : Refreshing Gem list
I, [2014-08-21T11:41:06.841582 #72528]  INFO -- : listening on
addr=0.0.0.0:8090 fd=17

So, as far as I can tell, the SIGABRT in a worker causes all siblings
workers to be reaped and restarted.   Is that intended/expected?
Shouldn't it just kill the single worker & restart that?

-Jonathan

Re: Worker SIGABRT takes down all workers?

[Eric Wong]

Jonathan del Strother <jon.delStrother@audioboo.fm> wrote:
> So, as far as I can tell, the SIGABRT in a worker causes all siblings
> workers to be reaped and restarted.   Is that intended/expected?
> Shouldn't it just kill the single worker & restart that?

Not intended or expected.  I cannot reproduce it by doing:

	kill -ABRT $WORKER_PID

Can you?
Can you setup a small test case which reproduces the issue?
Is anything else running that could be sending signals to all workers?

Thanks for any more info you can provide.

Re: Worker SIGABRT takes down all workers?

[Jonathan del Strother]

On 21 August 2014 21:32, Eric Wong <e@80x24.org> wrote:
> Jonathan del Strother <jon.delStrother@audioboo.fm> wrote:
>> So, as far as I can tell, the SIGABRT in a worker causes all siblings
>> workers to be reaped and restarted.   Is that intended/expected?
>> Shouldn't it just kill the single worker & restart that?
>
> Not intended or expected.  I cannot reproduce it by doing:
>
>         kill -ABRT $WORKER_PID
>

You're quite right, sorry.  I'm running via SMF, which is detecting
the child process's SIGABRT and deciding to restart the entire group.
I'll look into getting SMF to ignore those.