70k mqueue.in but load under 1 ??

[Glenn Steen]

On 28/11/06, Tony Johansson <tony.johansson at svenskakyrkan.se> wrote:
> I've added sbl-xbl.spamhaus.org, dnsbl.sorbs.net and relays.ordb.org at
> the MTA level as well as upped greet_pause to 10 seconds.
>
> The queue problem is now gone (for the moment) but I still dont get how
> a mailscanner installation could drop to under 1 in load and just sit
> there watching the queue build.
>
> Would it be possible do define some dynamic variables? Load under 1,
> queue over 1k - start another mailscanner child (or whatever appropriate)
>
When MailScanner "detects" there are more than "Max Normal Queue Size"
it'll stop trying to process mails in a FIFO manner and switch to just
grabbing the "Max Unsafe Messages Per Scan" number of messages
(unsorted... Probably makes it more like LIFO). So in a way, it
already tries to do something like that.

As to why... Well, the processes are likely waiting for something...
Most likely some network resource, or something equally slow, since
you see only a modest wait in sar (and some Idle...). What did vmstat
have to say during the same period? Not swapping are you? iostat? Any
"hot disks"?
One can also wonder if a --debug --debug-sa didn't show "perceptible"
pauses, so that one could make a betrter guess at _what_ was having a
slow time of it.

Tjena
-- 
-- Glenn
email: glenn < dot > steen < at > gmail < dot > com
work: glenn < dot > steen < at > ap1 < dot > se