[rabbitmq-discuss] flow control in rabbitmq

[Michael Klishin]

2013/12/30 sandeep kumar <sandy.sandeeep at gmail.com>

> where are the published messages are getting store when connections to
> queue are in blocked state
>

When RabbitMQ blocks a publisher, it stops reading from the socket. So,
some amount of in flight
data will be stored in OS caches. Once those fill up, publishing operations
should block (at least with
most clients).


> how to handle this kind of scenario gracefully
>

Allow RabbitMQ use more RAM, add more consumers so that they drain queues
and implement a feature in the client (similar to a write-ahead log) where
messages can go before being
published. You can also monitor message rates and server state over HTTP
API and throttle
your publishers using app-specific rules.

See also http://www.rabbitmq.com/connection-blocked.html
-- 
MK

http://github.com/michaelklishin
http://twitter.com/michaelklishin
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.rabbitmq.com/pipermail/rabbitmq-discuss/attachments/20131230/b49eccc9/attachment.html>