RabbitMQ Flow Control.

[rajesh...@gmail.com]

Hi All, 

Our RabbitMQ cluster is a 6 node 16cpu and 16GB memory cluster.

We currently do 20k messages per second on a single queue bound to direct exchange in rabbitMQ.

Recently we did split the queues based on routing key bindings to direct exchange in to 4 different queues we are trying to scale our capacity to 40k messages per second.

We have 2 node ha policy for each queue we are currently seeing backpressure and flow control in publisher connections and channels even though the consumers are very quick to remove messages from queue and there are 0 mesages backlogged or pending in Queue. we are not touching the cpu, memory and disk limits everything looks green on the cluster, Still we are seeing flow control on publisher connections and channels.

What can be the reason for this and did any one face a similar problem?

[Luke Bakken]

Hello,

Please start by letting us know what version of Erlang and RabbitMQ you are using.

Do you have any monitoring information you can share?

Also, we recommend an odd number of nodes in a cluster. This way if a partition happens a majority can be established. This is not related to your issue, however.

Thanks,

Luke

[rajesh...@gmail.com]

Thank you for your response Luke, We are currently using  RabbitMQ 3.8.2 and Erlang 22.