On Wed, 2006-04-12 at 07:41, Reuben Farrelly wrote:
That's why there's "scheduled downtime" and "after hours" to do these things, in case stuff goes wrong, as some users of horde with dovecot have found out lately ;)
What's 'after hours' for a company with users all around the world? Interesting concept, though...
The same time you'd do routine maintenance on other systems and networks I guess.
For the network, the redundant backups cover it transparently.
I'm just not sure why people want to do upgrade surgery on critical systems running on dovecot at any time, leave it in a 95% working state and hope users don't notice, whereby upgrading any other service would be done in a more controlled manner at a planned date/time with a full clean restart of the service. It only takes 10-15s to shut the child processes down and restart them, and then you know that everything is running properly on the new version.
Upgrading apache/sendmail/postfix/bind/qmail all requires a full service restart of the master and child processes, I'm not sure why there are complaints when dovecot just takes on the same behaviour.
Those services are all short term connections with no state maintained. No one will notice if you restart them or even reconnect the next time to a different machine. IMAP clients get very unhappy when they lose their connection and most don't fix it transparently. Unfortunately since they don't disconnect on a regular basis it probably isn't a good idea to even try to wait for the client to finish like apache does on a graceful restart. You might as well just kill it and say 'oops...'. It might be worth some effort to let a pop connection finish, though.
-- Les Mikesell lesmikesell@gmail.com