[zeromq-dev] Router sockets and lost wandering messages

[Noah Gibbs]

We've realized an odd property of our architecture, and I'm curious if
there's a standard solution.

Currently we have router processes using a bound ZMQ_ROUTER socket to
receive messages from client processes, each connected to multiple routers
via a ZMQ_DEALER socket.

If a client process reboots, it will reconnect using a new zmq identity for
its dealer sockets.

This means that any messages (acknowledgements, certain scheduling packets)
that are addressed to the old zmq identity from before the client rebooted
will never be sent.  I *think* that means they will pile up unsent in the
router socket.

In order to avoid piled-up messages on a slow router, we use a very low HWM
in most of our sockets - 1 by default.  This seems like a single errant
message could prevent a router from sending to any of its clients again
until the socket was closed and reopened (which means rebooting the router
for us right now).

Is there a standard solution to this problem?  I know ZMQ (intentionally)
doesn't let you ask "how many messages are queued right now on this
socket?".  But that means that genuinely stale messages seem like they
would be forever undetectable.  Is there any way to "expire" messages after
a given period of time?

-- 
*Noah Gibbs*
Software Engineer |
noah at ooyala.com | (510) 260-5409 (cell)
www.ooyala.com | blog <http://www.ooyala.com/blog> |
@ooyala<http://www.twitter.com/ooyala>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.zeromq.org/pipermail/zeromq-dev/attachments/20120206/edf5fc9b/attachment.htm>