[rabbitmq-discuss] Recovering from power failure

Tue Aug 4 20:05:52 BST 2009

> tsuraan wrote:
>> From the eof, I'm guessing that some file wasn't completely written,
>> so the persister is pretty angry.
>
> Actually, there may be a "*.previous" file in the same directory. When
> the persister is switching files, it makes sure not to delete the
> previous file until the new one is ready for use. You might find that
> restarting with the .previous file lets you get back in operation again.

Yeah, there is a .previous file.  it looks like the
rabbit_persister.LOG is 8 bytes, and the .previous file is ~140MB.
The contents of the .previous file look pretty good; at least the
message contents look reasonable from hexdump.  Should the persister
process know how to replace a corrupt log file with the .previous
file?  Also, if I'm using transactions everywhere, would that
.previous contain the queue states after the last successful commit?