On 11/28/2014 2:03 PM, Stefan Kaes wrote:
I'm a bit confused now: does this mean no multi threaded application
can use monitoring safely?
It appears that the order of events at the receive endpoint isn't
guaranteed with more than one iothtread on a multi-cpu system.
You can however have more than one iothtread on a multi-cpu system if
you assign the process to only one cpu with a taskset -p or -c command.
But this isn't probably what you are looking for.
Francis
--- stefan
This was in fact a scheduling issue, not a zeromq issue. The
number of
iothreads must be limited to one in order to process the
accepted/disconnected/... events in te required order.
Sorry for the noise.
Francis
On 11/20/2014 1:16 PM, Francis Le Bourse wrote:
Hi,
I have observed random crashs in zeromq, for example in
mailbox::recv() at assert(ok), and double disconnect sequences on
servers when monitoring is active
- event_accepted(fd, addr)
- event_disconnected(fd, addr)
- a second event_disconnectd(same fd, addr) sometimes later.
The attached test programs exhibits the issue after a few
minutes. The
most frequent error is the double disconnect but another
random crash
may occur.
Start:
./tserver
./tclient
Tested on Centos 6.5 and 6.6, zeromq-3.2.4, czmq-2.0.3
Cheers,
Francis
_______________________________________________
zeromq-dev mailing list
[email protected] <mailto:[email protected]>
http://lists.zeromq.org/mailman/listinfo/zeromq-dev
_______________________________________________
zeromq-dev mailing list
[email protected]
http://lists.zeromq.org/mailman/listinfo/zeromq-dev
_______________________________________________
zeromq-dev mailing list
[email protected]
http://lists.zeromq.org/mailman/listinfo/zeromq-dev
