[rabbitmq-discuss] Message TTL working fine in 2.6.1, broken in 2.7.0?

Wed Nov 30 07:42:46 GMT 2011

Andrew,

On 30/11/11 03:46, Andrew Maizels wrote:
> 2.7 seems to ignore TTL limits.

TTLs in 2.7 certainly can't be broken completely; we have tests that 
check that.

In fact they should work *better* in 2.7 in that requeued messages 
retain their original TTL rather than getting a fresh one.

> We had a slow  consumer leave about 1.8 million 40KB messages on the
> queue, which filled up the available disk and brought RabbitMQ down.
> The queues are durable and bound to a single exchange; no special
> parameters apart from the TTL, which is typically 600000
> milliseconds.

Were these 1.8 million messages the result of a relatively steady 
inbound stream? Over what period of time?

Can a single message get routed to multiple queues?

What memory limit does rabbit think it has (check the log for an entry like:
=INFO REPORT==== 28-Nov-2011::12:14:40 ===
Memory limit set to 4815MB of 12037MB total.
)?

Do you still have the log files from the time it failed?

> We've reverted to 2.6.1 for now, and everything is working again.

Is there any chance you could go back to 2.7.0, perhaps in a test setup, 
and see whether you can replicate the problem? If you do then taking a 
snapshot of the database dir at the time, together with the output of 
'rabbitmqctl report' should give us enough data to figure out what's 
going wrong.

Regards,

Matthias.