[rabbitmq-discuss] Flow control issue when shovelling messages between data centres
simon at rabbitmq.com
Wed May 15 15:37:30 BST 2013
On 15/05/13 10:48, SimonT wrote:
> I'd be interested though to understand better the relationship is
> between down-stream queue size / consumption rate and flow control.
A connection will go into flow control if it is trying to enqueue
messages faster than the queue(s) can accept them....
> example, if the consumer is not keeping up and the queue is growing,
> will this increase the chance of flow control at the publisher.
...and the amount of work done to accept a message is much higher if the
message must go to disc. So if you're publishing persistent messages the
cost should be constant, but if you're publishing transient messages
there's a big step change in cost to accept a message when a queue gets
big enough that RabbitMQ starts to page it out.
> what is the effect of TTL on the queue - e.g. will a shorter TTL result
> in less up-streamm flow control?
Only to the extent that it keeps such a queue shorter.
More information about the rabbitmq-discuss