Sieve failure on deliverdb corruption...

[Rob Mueller]

> I'm not sure running recovery on a potentially in-use database is a good
> idea.  But then I'm not a berkeley DB expert.
>
> I'm sure its an even worse idea to potentially have multiple processes
> running recovery on it at once.

I'm sure you could create a lock file to ensure only one is run at a time,
though I bet it probably does that itself...

What are the possibilities:

1. Any attempts to open the DB fail. That's what's happening anyway, so no
change...
2. Any attempts to open the DB crash. That's probably OK too, the lmtp
delivery will fail, and the MTA will try again later
3. Any attempts to open the DB freeze. Worst case, probably not traumatic
though

Well, seems it would be worth trying to run the recovery from my point of
view. At least, that's all I did, and I didn't stop cyrus, and it seemed to
work fine...

Rob