On Tue, 2009-01-06 at 17:06 -0600, Mike Abbott wrote:
Patch #11 adds a few dtrace providers to key points in the code. We
are still validating the correct placement and usefulness of these.
One cool thing about dtrace is that when not in use the hooks
literally are just a couple of no-op instructions.
I don't see dtrace-dovecot.h included in it.
Anyway I think the current code is too invasive. For example:
#ifdef APPLE_OS_X_SERVER if (DOVECOT_OD_LOOKUP_START_ENABLED()) DOVECOT_OD_LOOKUP_START(in_od_info, (char *) in_user_name); #endif
Couldn't the DOVECOT_OD_LOOKUP_START() macro be changed so that the whole code would become simply:
DOVECOT_OD_LOOKUP_START(in_od_info, (char *) in_user_name);And it would internally have the "if (DOVECOT_OD_LOOKUP_START_ENABLED())" part. That also makes it simple to get rid of the #ifdefs in .c files.
Patch #12 increases many listen queue sizes to handle high loads
better. The new values are not necessarily the best possible values
but they do allow more throughput.
I guess those can be changed. I've really no idea what the tradeoffs here are. I suppose it uses some more memory, but is it relevant with today's memory sizes? Could they be increased even higher than your values?
Patch #13 makes dovecot adapt to dynamic host identity changes upon
config reload.
Since it's not necessary to limit the domain name length, it could be just i_strdup()ed. (I should have done this in the mbox code too.)
Patch #14 covers three miscellaneous changes, to support more file
descriptors
I guess this is the same as running ulimit -n before starting dovecot? Is it useful to have this code inside Dovecot?
and an "unlimited" number of mail processes,
- /* APPLE - treat as "unlimited" */
- set->max_mail_processes = 1000000;
Hmm. Is it really a good idea to allow unlimited? Anyway -1U would be "more" unlimited. :)
and a different default mail user.
"uid=98", /* _dovecot's uid */"gid=6", /* _dovecot's gid */
Why does it matter which uid/gid is used for dump-capability? Maybe it could be just run as root.