Hi,
I've got a problem with dovecot on an Intel Mac OSX box. Basically, I
was running an older verion (0.8.?, I think), and everything was fine.
Then, when I upgraded to 1.0.0 via MacPorts, I found that periodically
the system gets into a state where it simply stops dealing with IMAP
requests.
This tends to manifest itself as Thunderbird simply stalling, not
fetching any new messages or folder contents. Sometimes quitting and
restarting Thunderbird will help, but at other times the problem
persists even afterwards. It seems to be linked to specific mailboxes
(or messages?) - often one "problem folder" will reproduce the effect
100% whilst operating on the rest of the mailbox is fine.
The problem /seems/ to often start as a result of trying to move
messages between folders. I'm not certain if that's a root cause,
though, or simply a consequence of "opening" an affected folder to
perform the move. I have a spam folder with a large number (>10000) of
messages in it, and trying to file stuff in there is a relatively
surefire way to generate the problem.
Once it's happened once, it inevitably escalates - restarting the
client helps a bit, as does restarting dovecot. However, the only way to
"fix" it and restore full functionality seems to be to delete all of
the
index and dovecot* files from the mailbox and folders... it generally
takes a day or so after that before things start to go wrong again.
All of this seems to point to corruption of the index files in some
way, but the log is universally unhelpful - there are no error messages
of any kind showing up there. The actual dovecot instance doesn't appear
to crash or even "hang" properly - you can still make inbound IMAP
connections OK, it's just that they don't ever give a response to
commands that touch the affected mailboxes.
Sometimes, I end up with temporary files lying around in the folders
in the mailbox - apparently mails which were in the process of being
moved when the server "gave up". I'm not sure if this is actually
part
of the problem, though, or just a symptom.
My configuration (from dovecot -n) is:
# /opt/local/etc/dovecot/dovecot.conf
ssl_disable: yes
disable_plaintext_auth: no
login_dir: /opt/local/var/run/dovecot/login
login_executable: /opt/local/libexec/dovecot/imap-login
mail_location: maildir:~/Library/Mail/IMAP
auth default:
passdb:
driver: pam
args: *
userdb:
driver: passwd
I'm running this on my personal machine, so there's only one user on the
system, which I have set up using MailDir (on HFS). There's a fair bit
of mail in many of my folders, but I doubt that's the problem as the
previous version didn't seem to have any issues...
I'm doing mail delivery using postfix directly into the maildir folders.
That doesn't seem to be affected at all by this, though - mail still
arrives fine even when Dovecot is failing to serve the folder contents
to IMAP clients... you just can't see it ^-^
Has anyone seen an effect like this? Or does anyone have any pointers as
to what I could try to debug it?
Any suggestions would be very greatly appreciated!
Thanks!
--
Ben Carter - ben at gunk.demon.co.uk / ben at saillune.net (preferred)
are you using the dovecot lda, or postfix, or procmail? i've been using dovecot on a PPC mac os x box for quite some time, but had to abandon the dovecot lda because of exactly the problems you describe. i switched to postfix, then procmail, and have not had any problems since. -SM- On May 6, 2007, at 5:49 AM, Ben Carter wrote:> > Hi, > > I've got a problem with dovecot on an Intel Mac OSX box. > Basically, I was running an older verion (0.8.?, I think), and > everything was fine. Then, when I upgraded to 1.0.0 via MacPorts, I > found that periodically the system gets into a state where it > simply stops dealing with IMAP requests. > > This tends to manifest itself as Thunderbird simply stalling, not > fetching any new messages or folder contents. Sometimes quitting > and restarting Thunderbird will help, but at other times the > problem persists even afterwards. It seems to be linked to specific > mailboxes (or messages?) - often one "problem folder" will > reproduce the effect 100% whilst operating on the rest of the > mailbox is fine. > > The problem /seems/ to often start as a result of trying to move > messages between folders. I'm not certain if that's a root cause, > though, or simply a consequence of "opening" an affected folder to > perform the move. I have a spam folder with a large number (>10000) > of messages in it, and trying to file stuff in there is a > relatively surefire way to generate the problem. > > Once it's happened once, it inevitably escalates - restarting the > client helps a bit, as does restarting dovecot. However, the only > way to "fix" it and restore full functionality seems to be to > delete all of the index and dovecot* files from the mailbox and > folders... it generally takes a day or so after that before things > start to go wrong again. > > All of this seems to point to corruption of the index files in > some way, but the log is universally unhelpful - there are no error > messages of any kind showing up there. The actual dovecot instance > doesn't appear to crash or even "hang" properly - you can still > make inbound IMAP connections OK, it's just that they don't ever > give a response to commands that touch the affected mailboxes. > > Sometimes, I end up with temporary files lying around in the > folders in the mailbox - apparently mails which were in the process > of being moved when the server "gave up". I'm not sure if this is > actually part of the problem, though, or just a symptom. > > My configuration (from dovecot -n) is: > > # /opt/local/etc/dovecot/dovecot.conf > ssl_disable: yes > disable_plaintext_auth: no > login_dir: /opt/local/var/run/dovecot/login > login_executable: /opt/local/libexec/dovecot/imap-login > mail_location: maildir:~/Library/Mail/IMAP > auth default: > passdb: > driver: pam > args: * > userdb: > driver: passwd > > I'm running this on my personal machine, so there's only one user > on the system, which I have set up using MailDir (on HFS). There's > a fair bit of mail in many of my folders, but I doubt that's the > problem as the previous version didn't seem to have any issues... > > I'm doing mail delivery using postfix directly into the maildir > folders. That doesn't seem to be affected at all by this, though - > mail still arrives fine even when Dovecot is failing to serve the > folder contents to IMAP clients... you just can't see it ^-^ > > Has anyone seen an effect like this? Or does anyone have any > pointers as to what I could try to debug it? > > Any suggestions would be very greatly appreciated! > > Thanks! > -- > Ben Carter - ben at gunk.demon.co.uk / ben at saillune.net (preferred) >
On Sun, 2007-05-06 at 13:49 +0100, Ben Carter wrote:> The problem /seems/ to often start as a result of trying to move > messages between folders. I'm not certain if that's a root cause, > though, or simply a consequence of "opening" an affected folder to > perform the move. I have a spam folder with a large number (>10000) of > messages in it, and trying to file stuff in there is a relatively > surefire way to generate the problem.Try if dotlock_use_excl=yes helps. That setting was made to fix hard linking problems with HFS. Otherwise http://wiki.dovecot.org/Debugging/ProcessTracing might show up something useful when looking at what imap process is doing at the time it seems to be hanging. -------------- next part -------------- A non-text attachment was scrubbed... Name: signature.asc Type: application/pgp-signature Size: 189 bytes Desc: This is a digitally signed message part URL: <http://dovecot.org/pipermail/dovecot/attachments/20070509/e4a92ba1/attachment.bin>
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 On May 9, 2007, at 2:35 AM, Timo Sirainen wrote:> * PGP Signed by an unknown key > > On Sun, 2007-05-06 at 13:49 +0100, Ben Carter wrote: >> The problem /seems/ to often start as a result of trying to move >> messages between folders. I'm not certain if that's a root cause, >> though, or simply a consequence of "opening" an affected folder to >> perform the move. I have a spam folder with a large number >> (>10000) of >> messages in it, and trying to file stuff in there is a relatively >> surefire way to generate the problem. > > Try if dotlock_use_excl=yes helps. That setting was made to fix hard > linking problems with HFS. > > Otherwise http://wiki.dovecot.org/Debugging/ProcessTracing might > show up > something useful when looking at what imap process is doing at the > time > it seems to be hanging.Is this a setting in dovecot.conf? I'm using all the defaults, there, myself. Jon -----BEGIN PGP SIGNATURE----- Version: PGP Universal 2.6.1 Charset: US-ASCII wj8DBQFGQbp6sTedWZOD3gYRAtP1AJ9ly++iRdNDFWs+dHm+zzkB8WRIswCgwZoI NrQmAuBDvMhF65FxLI2e6FY=8qPy -----END PGP SIGNATURE-----
Timo Sirainen wrote:> On Sun, 2007-05-06 at 13:49 +0100, Ben Carter wrote: >> The problem /seems/ to often start as a result of trying to move >> messages between folders. I'm not certain if that's a root cause, >> though, or simply a consequence of "opening" an affected folder to >> perform the move. I have a spam folder with a large number (>10000) of >> messages in it, and trying to file stuff in there is a relatively >> surefire way to generate the problem. > > Try if dotlock_use_excl=yes helps. That setting was made to fix hard > linking problems with HFS.After a week or so of intensive testing, it appears that we have a winner! That does indeed seem to resolve the problem - whilst it was sporadic to begin with, it generally only took a couple of days to manifest itself, so I'm reasonably confident that it's gone away. Many thanks! -- Ben Carter - ben at gunk.demon.co.uk / ben at saillune.net (preferred)