[9fans] Weird timeout thing in smtpd?

[9fans] Weird timeout thing in smtpd?

[Dan Cross]

Is anyone else using upas/smtpd to send mail from a `normal' [sic] client
like the Apple Mail.app program?  I get weird timeout's sending email with
an error message of `(null)' and then my recipients get duplicates.  I'm
trying to understand why, as it's starting to get annoying for them.

	- Dan C.

Re: [9fans] Weird timeout thing in smtpd?

[Russ Cox]

> Is anyone else using upas/smtpd to send mail from a `normal' [sic] client
> like the Apple Mail.app program?  I get weird timeout's sending email with
> an error message of `(null)' and then my recipients get duplicates.  I'm
> trying to understand why, as it's starting to get annoying for them.

is your client connecting from a trusted ip address?
if not, it's subject to /mail/lib/validatesender, which
may be taking too long for your client.

russ

Re: [9fans] Weird timeout thing in smtpd?

[Dan Cross]

Russ Cox <russcox@gmail.com> writes:
> is your client connecting from a trusted ip address?
> if not, it's subject to /mail/lib/validatesender, which
> may be taking too long for your client.

Maybe that's it.  I'm connected from whatever IP address my DSL
provider gives me, which is certainly not trusted in any way that I
know about.  However, I'm authenticating to the server, which I thought
would turn off such things.

	- Dan C.

Re: [9fans] Weird timeout thing in smtpd?

[Dan Cross]

Dan Cross <cross@math.psu.edu> writes:
> 
> Russ Cox <russcox@gmail.com> writes:
> > is your client connecting from a trusted ip address?
> > if not, it's subject to /mail/lib/validatesender, which
> > may be taking too long for your client.
> 
> Maybe that's it.  I'm connected from whatever IP address my DSL
> provider gives me, which is certainly not trusted in any way that I
> know about.  However, I'm authenticating to the server, which I thought
> would turn off such things.

Hmm, a little more digging shows that it's hanging on reading the
output of /bin/upas/send in data() in smtpd.c.  I worked around that by
only reading send's stderr output if status is non-zero.  Moving
beyond that, it seems to hang forever in proc_wait(), which is sort
of strange.  That would be, I think, waiting for send's exit status.
This is a bit complex for me as I'm debugging this over SSH using
ed, so the more productivly enhancing tools are unavailable at the
moment.

	- Dan C.

Re: [9fans] Weird timeout thing in smtpd?

[Dan Cross]

Dan Cross <cross@math.psu.edu> writes:
> Hmm, a little more digging shows that it's hanging on reading the
> output of /bin/upas/send in data() in smtpd.c.  I worked around that by
> only reading send's stderr output if status is non-zero.  Moving
> beyond that, it seems to hang forever in proc_wait(), which is sort
> of strange.  That would be, I think, waiting for send's exit status.
> This is a bit complex for me as I'm debugging this over SSH using
> ed, so the more productivly enhancing tools are unavailable at the
> moment.

Well, this is starting to make some sense.  Send is taking a *really*
long time to do anything.  I think what's happening is that send is
running so long that the client times out and spits out an error
message and resolves to retry later.  However, on the server side,
send is still running, and eventually the mail gets delivered.  At
some point, the client retries sending the mail, and send works
very quickly at that point, so the client is happy but a duplicate
message gets sent.

So, the question is: what would cause send to take a *really* long
time?  I mean, on the order of seconds....  And, more to the point,
how can I fix that?

	- Dan C.

Re: [9fans] Weird timeout thing in smtpd?

[Russ Cox]

> So, the question is: what would cause send to take a *really* long
> time?  I mean, on the order of seconds....  And, more to the point,
> how can I fix that?

run send by hand and then run acid on it and look around.  ;-)

Re: [9fans] Weird timeout thing in smtpd?

[Dan Cross]

Russ Cox <russcox@gmail.com> writes:
> 
> > So, the question is: what would cause send to take a *really* long
> > time?  I mean, on the order of seconds....  And, more to the point,
> > how can I fix that?
> 
> run send by hand and then run acid on it and look around.  ;-)

Following Russ's suggestion, we discovered that send was hanging
out waiting for /mail/lib/qmail to finish, which was in turn waiting
on runq.  On systems with backlogged mail, runq can take a long time
to complete.  Running runq in the background makes the problem go
away.  The fix is now on sources.

	- Dan C.