[Dovecot] High load/corrupted indexes after moving mailboxes?
Hello,
we're experiencing an odd issue after combining two vDisks used for mail storage into one big new LVM. We've done this in the past successfully and didn't experience a lot of problems. We stopped all services (dovecot/postfix) to ensure that mailbox contents (Maildir) will not be changed, rsynch'ed the mail to the new disks and restarted everything.
When we restarted dovecot, we noticed a huge spike in load (average load in top and high cpu iowait) and isolated dovecot as the issue. At some point, the average load crept up to over 100 and crashed the server twice. However, over time the average load and cpu wait times went down slightly. I have noticed a spike in load after moving user's mail storage, but not like this. It almost looks like everyone is either re-indexing or re-caching their mail?
Any help would be appreciated. We're on dovecot 0.99.
-Pat
On Wed, 2009-03-11 at 19:15 -0400, Patrick A. Treptau wrote:
Hello,
we're experiencing an odd issue after combining two vDisks used for mail storage into one big new LVM. We've done this in the past successfully and didn't experience a lot of problems. We stopped all services (dovecot/postfix) to ensure that mailbox contents (Maildir) will not be changed, rsynch'ed the mail to the new disks and restarted everything.
When we restarted dovecot, we noticed a huge spike in load (average load in top and high cpu iowait) and isolated dovecot as the issue. At some point, the average load crept up to over 100 and crashed the server twice. However, over time the average load and cpu wait times went down slightly. I have noticed a spike in load after moving user's mail storage, but not like this. It almost looks like everyone is either re-indexing or re-caching their mail?
Any help would be appreciated. We're on dovecot 0.99.
I don't say this lightly - upgrade. Personally I HATE that answer, I'm a "If it ain't broke don't fix it" person, but I've seen so many improvements to the backend and indexing systems this looks like it's an appropriate answer for you.
Rick
on 3-11-2009 4:15 PM Patrick A. Treptau spake the following:
Hello,
we're experiencing an odd issue after combining two vDisks used for mail storage into one big new LVM. We've done this in the past successfully and didn't experience a lot of problems. We stopped all services (dovecot/postfix) to ensure that mailbox contents (Maildir) will not be changed, rsynch'ed the mail to the new disks and restarted everything.
When we restarted dovecot, we noticed a huge spike in load (average load in top and high cpu iowait) and isolated dovecot as the issue. At some point, the average load crept up to over 100 and crashed the server twice. However, over time the average load and cpu wait times went down slightly. I have noticed a spike in load after moving user's mail storage, but not like this. It almost looks like everyone is either re-indexing or re-caching their mail?
Any help would be appreciated. We're on dovecot 0.99.
-Pat
I would seriously consider upgrading 0.99. Although it was a HUGE improvement over UWimap, the newer versions run circles around it. Even a move to one of the 1.0 stables will show major improvements, 1.1 even more, and I am sure the new 1.2 base in beta will do it again.
I know that it is what came with your distro, but it is not the best choice for a big mail user. I think that the biggest reason that many of the older distros haven't pushed an upgrade is the config file changes that came in the newer versions, not because 0.99 is better.
-- MailScanner is like deodorant... You hope everybody uses it, and you notice quickly if they don't!!!!
participants (3)
-
Patrick A. Treptau
-
Rick Romero
-
Scott Silva