Next message: Jerome Nenert: "single instance store and replication"
Zitat von Andre Nathan <>:
> Hello
>
> We have a cyrus server that runs under heavy load, and a few days ago it
> started to show a behaviour where its lmtpd processes take a long time
> to deliver messages sent from postfix. Below is an example of a postfix
> log message. with the email address removed:
>
> Feb 28 17:14:11 mta13 postfix/lmtp[9633]: 63CBA700071C:
> to=<user=>,
> relay=box2.prv.f3.k8.com.br[10.8.5.49]:2003, conn_use=6, delay=1718,
> delays=0.15/1118/0/600, dsn=4.4.2, status=deferred (conversation with
> box2.prv.f3.k8.com.br[10.8.5.49] timed out while sending end of data --
> message may be sent more than once)
>
> The last field in the "delays" field shows that the time out occurred
> after 600s trying to send the message to cyrus. Even when a timeout does
> not occur, the time for the message to be sent is around 100-300s.
>
> We have some other more lightly loaded servers in which the data is sent
> to lmtpd very quickly, and at any given moment there are only a few
> lmtpd's running. On the heavily loaded servers, there are always a few
> thousand lmtpd processes.
???
A few thousand lmtpd would be *way* too much because they all use the
same I/O bottleneck (if you don't have partitions on different I/O
paths). For a single I/O path i would recommend not more than some 10
.. 20 concurrent lmtpd with the exception if you are having complex
sieve rules which adds to latency.
Regards
Andreas
----
Cyrus Home Page: http://cyrusimap.web.cmu.edu/
Cyrus Wiki/FAQ: http://cyrusimap.web.cmu.edu/twiki
List Archives/Info: http://asg.web.cmu.edu/cyrus/mailing-list.html
