Hi, I'm trying to debug (without great results, so far) a very weird behaviour in my environment. Sorry if this came out in another form before, but I've not been able to find out anything resembling what I'm experiencing, so far...
OS is a debian wheezy and dovecot is the version provided by the OS:
# 2.1.7: /etc/dovecot/dovecot.conf # OS: Linux 3.2.0-4-amd64 x86_64 Debian 7.4
The system uses postfix with virtual users, credentials and maildir locations stored on mysql, local delivery through LMTP and sieve+managesieve available to users (although managesieve only listens on localhost and is only usable through the webmail interface).
FWIW, I've been running 10-20 systems just like this one (mostly with earlier dovecot versions and using LDA when LMTP wasn't available yet) for many years, and this is the very first time I'm so puzzled I end up searching for help here :-|
Even though the system has been working for months (more or less 1 year now) without big changes, a strange behaviour appeared (or became evident enough to be noticed) in the last few days, apparently involving sieve and some sort of "default/hidden" action triggered by something. Or at least that's what I ended up supposing.
The problem in short:
Messages I expected to find in certain folders were simply not there, although sieve logs reported them as delivered in those same folders. By manually inspecting the maildir, I actually found the messages, but they were all flagged as \\Seen + \\Deleted and therefore not displayed by the MUA (ThunderBird in my case).
I initially supposed this to be a TB issue, 'cause nothing in my sieve script ever flags messages that way, but I could not find anything on my client explaining that.
Then I received a notification from one of my users saying he was experiencing exactly the same, except he's using mutt as IMAP client; that restricted the issue to server-side interactions only, but nothing useful could be found in logfiles...
This is the "relevant" portion of a "dovecot -n" output:
[...] plugin { sieve = ~/.dovecot.sieve sieve_dir = ~/sieve sieve_extensions = +vacation-seconds +editheader sieve_vacation_default_period = 2d sieve_vacation_dont_check_recipient = yes sieve_vacation_max_period = 30d sieve_vacation_min_period = 2s sieve_vacation_use_original_recipient = no }
[...] protocol lmtp { mail_plugins = sieve hostname = alternativename.bofhland.org }
There is no sieve_(before|after) script in addition to what users create for themselves.
The mailboxes are relatively low-traffic; most of the traffic is caused by a single role address translated into multiple-recipients alias going to many of the users (100-300 messages per day).
Almost all the users using sieve rely on it to store traffic for that role-address in a dedicated folder, including the mutt user and myself. Not that's all that important, but means that most of the misbehaviours observed (but not all) are observed in these folders.
I enabled mail_debug, but there's nothing unusual in there AFAICT:
it just says:
- using the following location for user's Sieve script: /blabla/.dovecot.sieve
- opening script /blabla/.dovecot.sieve
- script binary /blabla/.dovecot.svbin successfully loaded
- not saving binary /blabla/.dovecot.svbin, because it is already stored
- executing script from /blabla/.dovecot.svbin
- stored mail into mailbox 'DedicatedFolder'
And it does, except the message happens to be flagged as seen and deleted...
In order to try restricting what may be causing the issue, I added...
removeflag "\\Deleted";
removeflag "\\Seen";...to the rule sending messages to the dedicated folder cited above for my mailbox. The problem apparently disappeared for me as long as I kept those in place.
This should mean (I suppose) that the decision of flagging messages that way is taken before the users' sieve scripts are applied (sort of "hidden sieve_before").
The problem is hard to replicate as it's apparently not deterministic: the same message is being flagged for one user but not for the other, different messages between the same people and with the same thread are sometimes flagged and sometimes not.
Something is surely triggering whatever is causing this, but it's not evident enough for me to point a finger and say "ha ha!" (https://www.youtube.com/watch?v=rX7wtNOkuHo)
So, a question for people knowing pigeonhole's internals: does it ever choose to mark messages as described above for any reason? If yes, I may be triggering this behavior with something I recently added/modified, and at least I'll have something to focus my investigation on. :-|
Any other suggestion is obviously welcome...