RE: Active queue are growing

3 msg	relay config question
10 msg	Delivering spamtrap mail
4 msg	qmail-ldap migration to postfix
11 msg	SASL user authentication
3 msg	assp
3 msg	Postfix+dovecot+Kerberos
3 msg	smtpd_tls_auth_only = yes - what are the possib...
1 msg	Debian Etch - setting up SMTP AUTH using SASL -...
1 msg	Postfix SSL certificates
6 msg	Before-Queue Content Filter and Address Mappings.
2 msg	Question about Header_check
2 msg	how to set postfix not to flood destination SMT...
1 msg	[OT] Mailgraph modified 1.13 version
1 msg	MUA with proper client certificates handling (f...
4 msg	postfix+procmail mailbox corruption, locking is...
11 msg	Need to enable VRFY
9 msg	Issue with relay_recipient_maps
2 msg	finally - working
5 msg	finally got log file error

Active queue are growing
\ Rocco Scappatura (23 May 2007)
. \ (Wietse Venema) (23 May 2007)
. . \ Gary V (23 May 2007)
. . . \ Rocco Scappatura (23 May 2007)
. . \ Rocco Scappatura (23 May 2007)
. . \ Rocco Scappatura (23 May 2007)
. . . \ Ralf Hildebrandt (23 May 2007)
. . . . \ Victor Duchovni (23 May 2007)
. . . . . \ Ralf Hildebrandt (23 May 2007)
. . . . . . \ Victor Duchovni (23 May 2007)
. . . . . . . \ Ralf Hildebrandt (23 May 2007)
. . . . . . . \ Ralf Hildebrandt (23 May 2007)
. . . . . . \ Noel Jones (23 May 2007)
. . . . . . . \ Victor Duchovni (23 May 2007)
. . . . . . . . \ Mark Martinec (23 May 2007)
. . . . . . . . . \ Rocco Scappatura (24 May 2007)

Subject:	RE: Active queue are growing
Group:	Postfix-users
From:	Rocco Scappatura
Date:	24 May 2007

> > > My amavisd-new scan times are similar. Times go down
> significantly
> > > if I disable network tests, so I suspect the large number of RBL
> > > lookups is the cause.
>
> My times are similar too. Most of the time is spent in
> SpamAssassin, mainly in network tests and related DNS
> queries, some in bayes database lookups, FuzzyOCR image
> processing is a heavy penalty too, and of course regexp rules
> take their CPU toll too. Pyzor is quite wasteful on CPU, and
> should be disabled. The drag of all drags can be a
> command-line virus scanner, if somebody still avoids using a
> fast daemonized scanners.

I'm not using Pyzor and I have disabled uvscan.

> > If throughput is low due to high DNS latency, but CPU usage
> is ligth
> > to moderate, one can raise the amavisd concurrency.
>
> True, it pays off to do a little experimentation and find the
> number of content filtering processes where throughput
> approaches its peak.
> The actual number mostly depends on hw, the set of SA rules
> and a choice of a database for bayes and AWL.
>
> Although my presentation from 2005 is getting a bit out of
> date, it still illustrates well the effect on throughput that
> choosing a number of processes and the a set of SA tests can have:
>
>
> http://www.ijs.si/software/amavisd/amavisd-new-magdeburg-20050519.pdf
>
> In particular see slides 13, 23, 28 etc.

I seen it again.. Congratulation! it is great!

I don't think that the cause is the DNS latency.. Infact, CPU usage is
heavy.. With amavis-nanny I saw that all process was busy.. And active
queue was growing..

Since yesterday night the load is decreased.. and since then the load in
normal..

I thought that yesterday Amavisd-new has reached the maximal troughput
(but if so, the strange thing is why I don't see the bounce that the
content filter didn't accept other message..).

Finally, I remember that the deferred queue wasn't congestioned..

BTW, I hope to have added other detail that can help you (if you want..)
to give a real explanation to the phenomeanon that I have lived
yesterday..

Today, I have a free active queue, while I have about 2000 message in
deferred queue..

Notice that I have some unvalidated domain. So I think that the majority
of them are backscatters.

I would like to solve this problem.. I know that one MTA didn't have to
bounce a non local message.. But this rule is not respected by MS
Exchange..

If this is true, what is the solution?

BR,

rocsca