Bug#445680: cyrus-imapd-2.2: mail lost: race condition with duplicates and shutdown?

[Ross Boylan]

Package: cyrus-imapd-2.2
Version: 2.2.13-13
Severity: normal

This problem appears to involve data loss.  A message was being
delivered while the server was shutting down.  It appears not to have
been delivered, but when the server came back up there was an entry
for it in the duplicates database.  So the retry (by exim) was tossed.
As a result, the mail was never saved by cyrus.

I speculate that the problem is that a message's id goes into the
duplicate database before the message is actually delivered.  It would
be safer to make the entry after delivery.

Oct  7 01:28:59 corn cyrus/master[22025]: about to exec /usr/lib/cyrus/bin/lmtpd
Oct  7 01:28:59 corn cyrus/lmtpunix[22025]: executed
Oct  7 01:29:00 corn cyrus/lmtpunix[22025]: accepted connection
Oct  7 01:29:00 corn cyrus/lmtpunix[22025]: lmtp connection preauth'd as postman
Oct  7 01:29:00 corn cyrus/lmtpunix[22025]: WARNING: sieve script /var/spool/sieve/r/ross/defaultbc doesn't exist: No such file or directory
Oct  7 01:29:00 corn cyrus/lmtpunix[22025]: duplicate_check: <E1IeRVL-0005jA-1K at corn.betterworld.us>  user.ross.comp.admin 0
Oct  7 01:29:02 corn cyrus/lmtpunix[22025]: duplicate_check: <E1IeRVL-0005jA-1K at corn.betterworld.us>  user.ross.comp.admin 0
Oct  7 01:29:02 corn cyrus/lmtpunix[22025]: mystore: starting txn 2147485087
Oct  7 01:29:02 corn cyrus/lmtpunix[22025]: mystore: committing txn 2147485087
Oct  7 01:29:02 corn cyrus/lmtpunix[22025]: duplicate_mark: <E1IeRVL-0005jA-1K at corn.betterworld.us>  user.ross.comp.admin 1191745740 134539227
Oct  7 01:29:03 corn cyrus/master[16669]: exiting on SIGTERM/SIGINT

Then, when it comes back up:
Oct  7 08:26:56 corn cyrus/master[8503]: about to exec /usr/lib/cyrus/bin/lmtpd
Oct  7 08:26:56 corn cyrus/lmtpunix[8503]: executed
Oct  7 08:26:56 corn cyrus/lmtpunix[8503]: accepted connection
Oct  7 08:26:56 corn cyrus/lmtpunix[8503]: lmtp connection preauth'd as postman
Oct  7 08:26:56 corn cyrus/lmtpunix[8503]: WARNING: sieve script /var/spool/sieve/r/ross/defaultbc doesn't exist: No such file or directory
Oct  7 08:26:56 corn cyrus/lmtpunix[8503]: duplicate_check: <E1IeRVL-0005jA-1K at corn.betterworld.us>  user.ross.comp.admin 1191745740
Oct  7 08:26:56 corn cyrus/lmtpunix[8503]: dupelim: eliminated duplicate message to user.ross.comp.admin id <E1IeRVL-0005jA-1K at corn.betterworld.us> (delivery)
Oct  7 08:27:57 corn cyrus/master[8338]: process 8503 exited, status 0


The corresponding exim logs show this
2007-10-07 01:28:58 1IeRVL-0005jA-1K <= root at localhost H=localhost (corn.betterworld.us) [127.0.0.1] P=smtp S=5266
2007-10-07 01:29:03 1IeRVL-0005jA-1K == ross+comp.admin at betterworld.us <root at localhost> R=imap_user T=cyrus defer (-1): LMTP connection closed after end of data

2007-10-07 08:26:56 1IeRVL-0005jA-1K => ross+comp.admin <root at localhost> R=imap_user T=cyrus
2007-10-07 08:26:56 1IeRVL-0005jA-1K Completed

The message that got lost was from one backup job; the next backup job
was for Cyrus and shut the server down as its first step.

It is possible the real problem is that cyrus did not fully commit the
original transaction.  I have a bdb backend.

The relevant spool directory has nothing timestamped around 1:29, so
the message, not just the index entry, seems to be missing.

-- System Information:
Debian Release: lenny/sid
  APT prefers testing
  APT policy: (990, 'testing'), (990, 'stable'), (50, 'unstable')
Architecture: i386 (i686)

Kernel: Linux 2.6.18-5-686 (SMP w/2 CPU cores)
Locale: LANG=en_US.UTF-8, LC_CTYPE=en_US.UTF-8 (charmap=UTF-8)
Shell: /bin/sh linked to /bin/bash

Versions of packages cyrus-imapd-2.2 depends on:
ii  cyrus-common-2.2         2.2.13-13       Cyrus mail system (common files)
ii  libasn1-6-heimdal        0.7.2.dfsg.1-10 Libraries for Heimdal Kerberos
ii  libc6                    2.6.1-1+b1      GNU C Library: Shared libraries
ii  libcomerr2               1.40.2-1        common error description library
ii  libdb4.2                 4.2.52+dfsg-4   Berkeley v4.2 Database Libraries [
ii  libgssapi4-heimdal       0.7.2.dfsg.1-10 Libraries for Heimdal Kerberos
ii  libkrb5-17-heimdal       0.7.2.dfsg.1-10 Libraries for Heimdal Kerberos
ii  libroken16-heimdal       0.7.2.dfsg.1-10 Libraries for Heimdal Kerberos
ii  libsasl2-2               2.1.22.dfsg1-14 Authentication abstraction library
ii  libssl0.9.8              0.9.8e-9        SSL shared libraries
ii  libwrap0                 7.6.dbs-14      Wietse Venema's TCP wrappers libra

cyrus-imapd-2.2 recommends no packages.

-- no debconf information