[zeromq-dev] Pub/sub, inproc shutdown Assert failure

[hp010170]

Marc, Martin:

I am not sure if this is related, however, I have observed that:

1. tests/test_shutdown_stress.cpp seg-faults on seemingly random runs
2. test_shutdown_stress uses PUB/SUB
3. test_shutdown_stress does uses tcp:// instead of inproc://

To illustrate:

--x--x--
hp $ for i in {1..20}; do ./tests/test_shutdown_stress; done
test_shutdown_stress running...
Segmentation fault
test_shutdown_stress running...
Segmentation fault
test_shutdown_stress running...
test_shutdown_stress running...
Segmentation fault
test_shutdown_stress running...
test_shutdown_stress running...
test_shutdown_stress running...
test_shutdown_stress running...
test_shutdown_stress running...
Segmentation fault
test_shutdown_stress running...
test_shutdown_stress running...
test_shutdown_stress running...
test_shutdown_stress running...
test_shutdown_stress running...
Segmentation fault
test_shutdown_stress running...
Segmentation fault
test_shutdown_stress running...
Segmentation fault
test_shutdown_stress running...
Segmentation fault
test_shutdown_stress running...
test_shutdown_stress running...
test_shutdown_stress running...
Segmentation fault
--x--x--

As you can see, I am not able to make out a pattern from a precursory 
glance.  I haven't plugged in the debugging tools yet, this was just 
from regularly building the code, I noticed the issue.

The above is from the latest libzmq in the repository.

Let me know if I can provide any more information.

-HP

On 10/06/2012 18:39, Martin Hurton wrote:
> Hi Marc,
>
> 1) could you create an issue
> 2) could you put together minimal C program reproducing this bug and
> make pull request so it finds its way into issues repo
>
> Thanks, martin
>
> On Sat, Jun 9, 2012 at 9:19 PM, Marc Criley <mcriley.1979 at gmail.com> wrote:
>> I'm getting:
>>
>> Assertion failed: ok (mailbox.cpp:79)
>>
>> when trying to shut down my application. After having searched the archives
>> and whitepapers I'm still at a loss. Here's the structure of what I have and
>> am trying to do (version 2.1.9-1, distributed with Ubuntu):
>>
>> - One main thread establishes a PUB socket using the 'inproc' transport.
>> - Four separate threads each open a SUB socket.
>> - 'inproc' requires the pub and sub sockets to use the same context, so that
>> is done.
>> - Each subscriber socket waits on 'recv()' for something to arrive, which is
>> then processed, and returns back to waiting for the next message.
>>
>> Everything runs fine in the application. However, at shutdown:
>>
>> - There is no pending traffic, it has all been cleared.
>> - The main thread closes its PUB socket.
>> - The main thread invokes zmq_term, which blocks.
>> - This unblocks the four subscriber threads waiting on recv().
>> - Each subscriber thread closes its socket and the thread terminates.
>>
>> At this point I expect the main thread call of zmq_term() to complete.
>> That's not what happens, instead I get:
>>
>> Assertion failed: ok (mailbox.cpp:79)
>>
>> I believe I did this in accordance with the 0MQ Termination whitepaper
>> (http://www.zeromq.org/whitepapers:0mq-termination), and this is the
>> sequence I use to shutdown a tcp transport client thread, which works
>> without problem.
>>
>> Deferring closing the PUB socket until after zmq_term doesn't work, as
>> zmq_term blocks until the socket is closed.
>>
>> Any suggestions would be appreciated.
>>
>> Marc C
>>
>>
>> _______________________________________________
>> zeromq-dev mailing list
>> zeromq-dev at lists.zeromq.org
>> http://lists.zeromq.org/mailman/listinfo/zeromq-dev
>>