2 ms·
The last time I checked (~4 years ago) RabbitMQ and others didn't have good contention management. I am not sure if we can blame users all the time. I remember
by wslh 12y ago
The last time I checked (~4 years ago) RabbitMQ and others didn't have good contention management. I am not sure if we can blame users all the time. I remember being involved in this thread: http://lists.rabbitmq.com/pipermail/rabbitmq-discuss/2010-May/007252.html http://lists.rabbitmq.com/pipermail/rabbitmq-discuss/2010-Ma... and this proposal: http://lists.w3.org/Archives/Public/www-ws-arch/2002Oct/0401.html http://lists.w3.org/Archives/Public/www-ws-arch/2002Oct/0401...
- cheald 12y agoRMQ deals with this fairly decently now via flow control (as of 2.8.0, I think). Basically, if it determines that the queue isn't keeping up with publishers, it'll throttle publishers (by blocking on publishes) to slow down publication to a manageable speed. It also uses this mechanism to block publishers if free memory/free disk space fall below preconfigured thresholds. When a publisher is blocked, RMQ notifies it that is has been flow controlled, so the publisher can optionally take some action to help mitigate whatever circumstances caused the flow control. It's not perfect, but it's actually pretty good at mitigating the problem of RMQ causing an RMQ-oriented system to eat itself.