Race condition in postmap?
in [Postfix]
|
From: Eray Aslan on 12 Feb 2010 08:07 On 12.02.2010 14:47, Richard Cooper wrote: > On 12 Feb 2010, at 12:12, Eray Aslan wrote: >> On 12.02.2010 13:25, Richard Cooper wrote: >>> Feb 12 00:41:24 mail1 postfix/smtpd[24782]: NOQUEUE: reject: RCPT from unknown[111.111.111.111]: 550 5.1.1 <recipient(a)example.com>: Recipient address rejected: User unknown in virtual alias table; from=<sender(a)example.net> to=< recipient(a)example.com > proto=SMTP helo=<example.org> >>> >>> This is a very intermittent and short lived error. Emails to recipient(a)example.com were working before the error and start working again a few seconds after it. >> >> http://www.postfix.org/DATABASE_README.html#safe_db > > > I'm not sure that apples to my case. That page says "If the update fails in the middle [because the disk is full or because something else happens] then you have no usable database, and Postfix will stop working". In my case the update completes without error, correctly writes virtual_aliases.db and postfix continues working. The only visible error is that during the update Postfix "forgets" some of the lookup table for a short period of time. You might also want to try CDB. Its updates are atomic. Recommended instead of Berkeley DB. http://www.postfix.org/CDB_README.html -- Eray
From: Victor Duchovni on 12 Feb 2010 14:47
On Fri, Feb 12, 2010 at 11:25:05AM +0000, Richard Cooper wrote: > Based on my debugging it seems that this error is related to me running postmap to rebuild the virtual_alias table This is despite the fact that the recipient(a)example.com address is correctly configured in both the old and new virtual_aliases. Here a log of what was happening at the same time as the above error: The original Berkeley DB (version 1.8x) which was available at the time that "hash" and "btree" table support were added to Postfix was a simple indexed file format and library. In that version of Berkeley DB there were no memory mapped page pools, transaction logs, ... If are using Berkeley DB on a BSD system with version 1.8x, then postmap is race-free due to the Postfix locking protocol for Berkeley DB files. Newer much more feature-full versions of Berkeley DB are no longer race-free with the Postfix locking protocol, and you need to atomically create/rename a newly built table. I don't use Berkeley DB for multi-reader tables, I strongly recommend CDB for that purpose. I only use Berkeley DB for single reader/writer tables such as TLS session caches, address verification caches, ... -- Viktor. P.S. Morgan Stanley is looking for a New York City based, Senior Unix system/email administrator to architect and sustain our perimeter email environment. If you are interested, please drop me a note. |