On Wed, 2005-08-17 at 13:05 -0300, eu wrote:
> Hello, i need help...
>
> i have a postgresql 8.0.3 database running on the backend of a postfix
> setup ( i didn't trust Mysql for the job ) on Linux kernel 2.6.8.1,
> serving email to a dozen different virtual domains ( including this one
> i'm using right now )...
>
> however, this setup takes a whole lot of simultaneous connections to the
> database
> ( postfix, amavis-new, clamav + spamassassin, apache+squirrelmail,
> courier-authlib and courier-imap ), all those services were configured
> to use Unix Domain Sockets, instead of TCP ( netstat -anvp showed too
> many short-lived connections were kept on CLOSE_WAIT for too long while
> connecting via TCP ), and, despite i have made max_connections on the
> postgresql.conf as high as 500, after ( around ) 64 simultaneous
> connections i start having a lot of "sorry too many clients already"
> errors on my postgresql logfile...
> then, my users go nuts complaining about how slow the server is and/or
> having to retype their passwords too many times ( of course, since
> authlib can't pick a connection to authenticate against the database ).
>
> max_connections, shared_buffers, shmmax were tweaked, but the server
> does not seems to respect the 500 max_connections...
> i *know* i'm missing something obvious, but, what could be wrong ?...
> i'm lost... any help would be most appreciated... please.
>
Can you post the relevent portions of your postgresql.conf file? Do you
see any error messsages when starting the database up (perhaps about
shared memory or such)? We also run a postfix mailserver (with maildrop,
courier-imap/vauthchkpw, etc) against a postgresql database with
max_connections set to 512 (FreeBSD 5.x machine). On the delivery end of
things we pool the connections from the postfix server using proxymap(8)
(which helped ease the connections load).
Sven