Timo Sirainen wrote:
On Wed, 2009-04-08 at 14:38 -0700, Daniel L. Miller wrote:
Having a consistent name prefix for all the processes sounds nice - but
then you'd stick out as the exception to typical multi-process server
names (like Postfix's master, smtpd, cleanup, etc.). Is it a Good Thing
to deviate from accepted (poor) practices? Hmm....
Other tradeoffs...more space consumed in logfiles. More screen width
consumed during listings. Not necessarily a Good Thing - not
necessarily a Bad Thing. But something to ponder on.
No, I wouldn't write the dovecot- prefix to log files. So while the
binary names would be dovecot-lda, dovecot-imap-login, etc. the logs
would contain only lda(user), imap-login(user), etc.
So the process name won't match the name in the log file? That sounds
like a Bad Thing.
Another thing I was thinking about previously was that in process lists
they were prefixed with dovecot/. So the binary names could be lda,
imap-login, etc. but they'd show up in process lists as dovecot/lda,
dovecot/imap-login, etc.
That I like.
I would also consider the Dovecot architecture. As I (mis)understand
it, the "dovecot" process spawns the necessary imap, pop3, and login
daemons. So having a "dovecot.conf" file for controlling these is quite
appropriate. However, unless I've missed something (quite likely) -
"deliver" has nothing to do with the listening daemons. So having the
"lda" configuration in the dovecot.conf file might be inappropriate - I
would suggest splitting that off to a "dovecot-lda.conf" file (or
whatever you change the delivery agent name to).
But deliver also reads dovecot.conf and inside protocol lda {} it can
also override all settings from dovecot.conf. So I don't really like the
idea of splitting the configuration. Also in v1.3+ the only thing that
reads dovecot.conf is doveconf binary, master and deliver and everything
else get their configuration from it.
Told you I didn't know what I was talking about!
--
Daniel