[zeromq-dev] Heartbeats and/or keep-alives

[Marcelo Cantos]

I've been looking around for information on how (if?) ZeroMQ deals with
flaky networks, and have found very little beyond an inconclusive
exchange<http://lists.zeromq.org/pipermail/zeromq-dev/2010-May/003437.html>
on
the subject of keep-alives.

I am very keen to try a migration from RabbitMQ to ZeroMQ, but this is quite
a sticking point, since we have found that our network tends to deaden
long-idle connections — they don't die, they just stop sending data. I'm
guessing this is due to our use of el-cheapo networking hardware, but
haven't investigated in any depth. In RabbitMQ, we managed to rid ourselves
of the problem by setting the RequestedHeartbeat field on each client
connection to a reasonable value (five minutes seems to do the trick).

I have confirmed that ZeroMQ also suffers from deadened idle connections on
our network, for which the only fix seems to be to restart both peers. Has
any progress been made (or at least further discussion had) on how to deal
with this? Is there a solution that is as simple and unobtrusive as what we
did with RabbitMQ? Is there any solution at all that doesn't require manual
restarts?


Sincerely,
Marcelo Cantos
Melbourne, Australia
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.zeromq.org/pipermail/zeromq-dev/attachments/20101013/47755ebe/attachment.htm>