Just following up on my own question a bit...
I'm trying to figure out how many mysql hits a single incoming message can
generate to arrive at a max_connections setting for mysql.
Worst case, I assume it would look like this:
-initial chkusr check (is it a valid address?)
-smtp-auth (possible if it's a local user sending to another)
-vdelivermail (look up where to deliver)
-spamc (if user has spamassassin enabled, prefs fetched via mysql)
Am I missing anything on the delivery side?
These obviously aren't all concurrent, but the hits are all pretty
So if I am allowing a max of 100 concurrent qmail-smtpd's, what would be a
"safe" number of concurrent mysql connections when the box is maxed out?
Multiply 100 by 2, 3, 4?
For mail retrieval, I can measure a bit more easily...
On Tue, 11 May 2004, Charles Sprickman wrote:
> I've seen a few people complaining that their mail is getting bounced.
> Not good, needless to say. On examing the bounces, I see that it's the
> chkusr error message that the user does not exist.
> At this point my best guess, by looking at overall server activity is that
> this is happening when the machine really gets swamped (all incoming smtp
> connections chewed up with spam runs - this server and a backup server).
> I have mysql set to 500 max connections, and just bumped that up to 800.
> I was surprised a bit by how chkusr behaves if the mysql server goes away
> - rather than erring on the side of caution and trying to deliver the
> mail, it bounces it. I verified this on a test box.
> FWIW, this is the Shupp chkusr patch, there's no version info in the
> patch, but it looks like it was made 3/16/2003.
> Is this how chkusr is "supposed" to operate? If so, I need an
> alternative, as most other parts of vpopmail survive a brief database
> outage without dumping good mail.
> Are there any other tested "chkusr" type patches out there? Please share!