We've been running 1.1 rc3 now for about a month, and overall it's been
very fast and stable.
There is however an Index-related problem that keeps cropping up which
prevents a user from accessing their mailbox until their Indexes are
manually removed.
Somehow, an Index becomes problematic, and Dovecot hangs forever when
the mailbox is accessed. It manifests when a user logs in and issues a
"SELECT INBOX", or "GETQUOTAROOT" when "maildirsize" needs rebuilding.
When Dovecot hangs, no data is returned and no further commands are
accepted, and the daemon must be killed. We see no errors in the logs,
but have yet to catch the problem with "mail_debug" enabled.
We're using NFS with Maildir's and these related options:
mail_location = maildir:/mail/%Lu/Maildir:INDEX=/mail/%Lu/Cache
mmap_disable = yes
mail_nfs_storage = yes
mail_nfs_index = yes
mail_drop_priv_before_exec = yes
mail_cache_min_mail_count = 128
mailbox_idle_check_interval = 5
maildir_stat_dirs = yes
maildir_copy_with_hardlinks = yes
maildir_copy_preserve_filename = no
Has anyone else run across the need to purge Indexes manually? Could
this be unrecoverable Index corruption, or a locking issue?
We've made a copy of the mailbox when the problem occurs. However, when
we test it later, Dovecot re-builds the "dovecot.index" file itself just
fine, and functions normally.
We plan to "mail_debug" when this happens next to see if we can get more
information.
Thanks!
-Rich
