[rabbitmq-discuss] Mnesia db integrity

[Matthias Radestock]

Valentino,

Valentino Volonghi wrote:
> After a simulated crash (basically just shutdown the system without
> stopping services) and a restart rabbitmq reacts by crashing on the
> mnesia directory.

Mnesia should generally recover from this without assistance.

Next time this happens, could you please zip up the mnesia dir and send 
it to me?

Also, can you reproduce the problem by, say, killing the rabbit process?

> I thought that by saving rabbitmq log files and removing the mnesia 
> files rabbitmq would have picked up the undelivered messages

That's not quite enough. Removing the mnesia files will remove all the 
durable exchanges and queues, so there is no place for the persisted 
messages to go. See
 
http://lists.rabbitmq.com/pipermail/rabbitmq-discuss/2009-January/002974.html

for a sequence of steps that might work.

> apparently all I get from the logs is a:
> 
> Rolling persister log to "/mnt/ebs/adrollserver/mnesia/adserverrmq/ 
> rabbit_persister.LOG.previous"
> 
> Which basically means that it discards my messages that won't be 
> transmitted again.

The above does not indicate that the messages have been discarded 
(though for the reasons I mentioned above they will have been), only 
that the persister has created a new snapshot of the persisted data.


Matthias.