[rabbitmq-discuss] RDQ files filling the disk.
tim at rabbitmq.com
Wed May 22 10:57:49 BST 2013
On 05/22/2013 10:56 AM, Simon MacMullen wrote:
> On 22/05/13 09:38, Tim Watson wrote:
>>>> Meanwhile, Rabbit2 reportedly have been receiving the messages
>>>> from Rabbit1. I also saw no shovel connection related errors in
>>>> rabbit.log, etc.
>>>> Q1. Does this sound plausible? If so, how could this happen?
>> Sounds like rabbit1 is paging due to memory pressure. Do you see a
>> memory alarm in the logs? Even if you don't, it may be paging
> The memory alarm is a bit of a red herring - the idea of preemptive
> paging is to prevent the memory alarm going off!
Good point - I wasn't thinking about that in quite the right terms there.
>>>> Q2. I'm thinking the rdq files are meant to get deleted at some
>>>> point. What would cause the files to*not* get deleted?
>> The paged out messages will need to be delivered and acked and, at
>> the same time, memory use will need to drop back so that eventually
>> the broker stops paging deliberately.
> Again, don't really look at memory use, but just look at the total
> number of messages in the broker...
Makes sense. So once the # messages starts to drop, you'd expect the
broker to stop paging and eventually (if everything fits nicely into
memory) that file was become obsolete and get removed.
More information about the rabbitmq-discuss