[zeromq-dev] Extremely strange PUB/SUB behavior
Calvin de Vries
devries.calvin at gmail.com
Wed Feb 22 20:14:44 CET 2012
More color -
I was actually able to "cause" this issue now by taking down a client that
had a similar subscription list. The same subscription that had stopped
receiving messages earlier has stopped receiving messages again. I'm
thinking there's some issue with sending to the wrong pipe in the trie
after a connection has been closed?
I'll try my best to create a test case.
On Wed, Feb 22, 2012 at 2:01 PM, Calvin de Vries
<devries.calvin at gmail.com>wrote:
> I'm seeing some extremely strange behavior in production with ZeroMQ. It
> seems as if I stop receiving messages for a certain subscription and am not
> sure why. I've been using my current pattern in production for months and
> this has occurred twice in that time frame (once today and once last week)
> and any time I try to reproduce it, I cannot.
> Basically, I have a Linux C++ server using a PUB socket to send out
> messages for many different subscriptions (about 3000 different
> subscriptions each is a char* 1-5 letters long) once every 5 seconds to
> Windows C++ clients using a SUB socket. It sends out a message for every
> single subscription in that time frame. Any given client may be subscribed
> for anywhere between 10-100 unique subscriptions. One of my clients (the
> same one actually out of around 30-35) stopped receiving messages for just
> one of the subscriptions. Another client that has the same subscription
> list does not stop receiving any messages for any subscription. The first
> time, this behavior had eventually somehow "fixed" itself, the client
> started receiving messages for all subscriptions once again. Today when it
> happened, the client was restarted and received all subscriptions again as
> Since both times another client with the same subscription had still been
> receiving messages, I know that the publisher is still sending out the
> message. It's not a socket becoming "closed" because it is only one socket
> and still receiving messages for other subscriptions. It seems that this
> is extremely rare, though so every time I try to document with a test case,
> I cannot do it. My best guesses to the cause of this problem sound
> ridiculous - the subscription somehow becoming "cleaned up" or
> "unsubscribed" for no reason. I just can't justify it and don't see how
> that could possibly be happening.
> I'm curious if anyone has ever noticed anything like this, or if there's
> perhaps some sort of flaw in my system. It seems very straightforward to me
> and am still baffled how this is occurring.
> Any help, insight or suggestions would be greatly appreciated.
> Thank you,
> Calvin de Vries
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the zeromq-dev