Timo Sirainen wrote:
Failing that, an alternative kill signal sounds like the way to go.
Now that I think of it, that won't work. The imap/pop3 processes kill themselves when they notice that the stderr pipe gets closed, so I can't think of a way for master to notify the imap/pop3 processes on-the-fly that they shouldn't die. Guess I'll add a setting instead.
What about the alternate kill signal to the master just stopping listening for new connections? (release the *:143 and *:110 handles).
If logging by syslog you can happily have two dovecot masters sending log lines ... won't work in other cases I suppose (suppose that is what I'm missing?)
With both a new master and an old master you can then also choose to go and do a proper kill on your old master 24 hours later or something like that to ensure nothing runs on the old binaries for longer than 24 hours (as a proper kill on the master now kills all the relevant children).
Rob.