[Dovecot] Just one user experiencing server timeouts
[dovecot-1.0-0.beta2.7 on Fedora Core 5, maildir formatted mail directories, various clients: thunderbird 1.5.0.4 from FC4 and WinXP, SquirrelMail 1.4.6-5.fc5 RPM, various recent versions of Pine from various Fedoras and Redhat 9]
I have just one mail user who experiences server timeouts when trying to delete a message (move it to Trash) or other common operations. This user has 4.1GB of mail spread over 200+ Maildir-formatted folders, so I immediately expected sheer volume as the problem - this user has by far the most mail. This user experiences the problem from multiple Thunderbird installs, from SquirrelMail, and from Pine.
However, I created a new user account using our standard account creation processes, and did a direct copy of olduser's Maildir to newuser's homedir. Then I did a chown -R newuser ~newuser/Maildir, set up an account in Thunderbird for newuser's account, and tried for an hour to get a server timeout doing all the same operations that cause the timeout on olduser's account, with no timeouts. I absolutely cannot duplicate the behavior, but when I watch olduser do this from any of the above named clients, I see the timeouts on about 10% of the operations.
I've been looking into this off and on for weeks, but I am baffled. I can't think of what is so specific about this one user's account that is causing the trip up. Both home directories are stored on the same filesystem and as I mentioned I made an exact duplicate of Maildir. I'm completely out of ideas, does anyone have any thoughts? Does Dovecot do any data tracking on a per user basis that could be confusing it, some cache I can clear? Dovecot has been restarted numerous times since the trouble started.
-- Fran Fabrizio Senior Systems Analyst Department of Computer and Information Sciences University of Alabama at Birmingham http://www.cis.uab.edu/ 205.934.0653
On Mon, 2006-06-26 at 17:24 -0500, Fran Fabrizio wrote:
However, I created a new user account using our standard account creation processes, and did a direct copy of olduser's Maildir to newuser's homedir. Then I did a chown -R newuser ~newuser/Maildir, set up an account in Thunderbird for newuser's account, and tried for an hour to get a server timeout doing all the same operations that cause the timeout on olduser's account, with no timeouts. I absolutely cannot duplicate the behavior, but when I watch olduser do this from any of the above named clients, I see the timeouts on about 10% of the operations.
If you can make him reproduce this, you could see what the strace looks like:
http://wiki.dovecot.org/Debugging/ProcessTracing
If you can't interactively do this, you could change mail_executable in config file to point to some script, which for normal users just does "exec /usr/libexec/dovecot/imap" but for this user does:
exec strace -tt -o imap.log.date +s
/usr/libexec/dovecot/imap
Or something.
I've been looking into this off and on for weeks, but I am baffled. I can't think of what is so specific about this one user's account that is causing the trip up. Both home directories are stored on the same filesystem and as I mentioned I made an exact duplicate of Maildir. I'm completely out of ideas, does anyone have any thoughts? Does Dovecot do any data tracking on a per user basis that could be confusing it, some cache I can clear? Dovecot has been restarted numerous times since the trouble started.
I guess you could try clearing the dovecot.index* files from the Trash mailbox.
participants (2)
-
Fran Fabrizio
-
Timo Sirainen