On Wed, 2011-12-07 at 20:38 +0200, Mark Zealey wrote:
Following on from this thread: http://www.dovecot.org/list/dovecot/2010-September/052704.html https://owa2010.webfusion.com/owa/redir.aspx?C=a447887e3d204944a8c766142df41453&URL=http%3a%2f%2fwww.dovecot.org%2flist%2fdovecot%2f2010-September%2f052704.html in 2.0.16 I am seeing the same issues. I believe this would be related to some timeouts set on our mysql servers which are pretty low (30 or 60 seconds or so to drop idle connections).
Yes, most likely. But it only happens if the quota isn't enforced (e.g. unlimited quota), because otherwise the previous "current quota usage" lookup would have automatically reconnected.
This is happening for imap, pop and lmtp processes when they try to update quotas stored in a mysql database; presumably it is the dict process that actually has these issues - the lmtp delivery isn't lasting longer than 1 second so shouldn't be seeing any timeouts there. This is always followed by the "Error: dict quota: Quota update failed, it's now desynced" line; presumably it is telling the truth but I'm not sure how I could tell.
Yes. All sql dict updates go through dict process, which keeps a long running sql connection open.
We are also seeing this "MySQL server has gone away" issue in login processes but it isn't causing an issue there as presumably dovecot transparently retries rather than just giving up as it seems to in the quota updates. Any thoughts or advice would be appreciated.
You mean auth processes. In case of auth the lookup is retried though.
Anyway, v2.1 now retries the commit: http://hg.dovecot.org/dovecot-2.1/rev/782f09d13ece