Autor: W B Hacker Data: Para: exim users Assunto: Re: [exim] dnslists - how to set timeouts
Marten Lehmann wrote:
> Hello,
>
>
>>But there is already a very reasonable default that does just what you asked for.
>
>
> after which seconds does exim timeout DNS calls?
Not sure. I would have expected 30s, even if on dialup, to be enough to get the
info if it was *ever* going to be had.
But - I've got at least one logged as timing-out 3m 29s after the initial
connection was established.
Anyone know if the default has retries?
CAVEAT - this is on a 'production' server, not a 'fully instrumented' R&D box,
so I don't have a timestamp for the start of the actual callout vs start of
overall connection. Don't take that time as definitive!
On R&D boxes where I do have such timestamps, the 'normal' return is
sub-one-second when the callout leaves the box at all (i.e. not cached by Exim
itself). OTOH, external DNS cache is one or two hops away, and there are also
DNS on-box.
>
>
>>Entries such as those below may not show for many days, then run for an hour in
>>a burst of *many*, other days one to four only for the entire day, etc.
>>
>>2006-10-21 13:19:32 DNS list lookup defer (probably timeout) for
>>113.249.106.85.dnsbl.njabl.org: assumed not in list
>
>
> Well, I haven't seen any of them in my logs, but I'm still pretty sure
> that these dns calls are the reason for the client timeouts. I think
> that the smtp-client timeouts just occur before the exim timeout, so if
> the connection is closed before exim would have detected the timeout
> they don't actually appear and get logged.
No need to guess. They are logged, or can be made to be:
IF 'log_selector = +all'
AND connection not abandoned earlier, of course....
We run the callouts in the 'CONNECT' phase, but do no rejection until RCPT, so
early-disconnect occurs only when a caller attempts to pipeline (not advertised
here) or disregard smtp sync (enforced).
