We use Dovecot on a cluster (10'000 domains, 30'000 account) without any problem with pop/imap. We use Dovecot 1.0 RC10
Now we use virtual transport from postfix for mail delivery.
We need the Dovecot LDA for sieve support.
We test this solution on some domains with success.
But if we change the configuration of all our domains, we have a lots of bounce with this error in postfix log:
postfix/pipe[24573]: 4F6D7CF11: to=info@XXXX.com, relay=dovecot, delay=1200, status=bounced (Command time limit exceeded: "/usr/local/dovecot/libexec/dovecot/deliver")
If I watch the deliver activity with ps/awk/watch, I see that if deliver duration time exceed some seconds, the deliver process stay in memory but do nothing. Here a strace output of this case:
Process 349 attached - interrupt to quit gettimeofday({1161965973, 768478}, {0, 0}) = 0 poll(
Strace output nothing before the SIGKILL from postfix (command time exceed 1200 second), the end of the strace:
[{fd=5, events=POLLIN|POLLPRI|POLLERR|POLLHUP|POLLNVAL}, {fd=9, events=POLLIN|POLLPRI|POLLERR|POLLHUP|POLLNVAL}], 2, 2147483647) = -1 EINTR (Interrupted system call) +++ killed by SIGKILL +++
I'm not sure is this problem has a relation with our deliver problem. But in the deliver log /var/log/dovecot/deliver, I have a lots of message:
deliver(info@labomex.com): 2006.10.27 18:29:28 Error: file_dotlock_replace(/var/mail/labomex.com/mails/info/.dovecot.lda-dupes) failed: No such file or directory deliver(info@labomex.com): 2006.10.27 18:29:28 Error: rename(/var/mail/labomex.com/mails/info/.dovecot.lda-dupes.lock, /var/mail/labomex.com/mails/info/.dovecot.lda-dupes) failed: No such file or directory
Our dovecot.conf (lda parts):
protocol lda { postmaster_address = postmaster@clm.net4all.ch #hostname = mail_plugins = cmusieve mail_plugin_dir = /usr/local/dovecot/lib/dovecot/lda sendmail_path = /usr/sbin/sendmail auth_socket_path = /var/run/dovecot/auth-master log_timestamp = %Y.%m.%d %H:%M:%S%t log_path = /var/log/dovecot/deliver info_log_path = /var/log/dovecot/deliver.info }
We use on UID/GID per domain so every account of a domain use the same UID/GID.
The mail storage is on a Debian Sarge NFS server with 2.4.27-3-686-smp. Indexes are stored localy on each POP/IMAP/LDA server.
I don't understand why we have this problem. If somebody can help us, thanks a lots.
Dominique Feyer