[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Compute loop + no processing with spamass-milter

From: Dan Nelson
Subject: Re: Compute loop + no processing with spamass-milter
Date: Sun, 18 Aug 2002 16:55:57 -0500
User-agent: Mutt/1.5.1i

In the last episode (Aug 17), Terry Kennedy said:
>   I'm trying to install the milter on BSD/OS 4.3. I started with the
> 0.1.1 kit, but I get "timeout before data read" from sendmail.
>   Here's the log. At the point where the logging stops, the milter is
> chewing all available CPU.
> Aug 10 07:05:37 test2 spamass-milter[11970]: mlfi_eoh: enter
> Aug 10 07:09:37 test2 sm-msp-queue[11971]: g7AB5bdX011971: Milter 
> (spamassassin): timeout before data read
>   At this point the milter is still chewing all available CPU:
> 11970  p0  S      0:00.08 spamass-milter -p /var/run/spamass.sock -d 2
> 11972  p0  R      5:00.34 (spamass-milter)
>   later...
> 11970  p0  S      0:00.16 spamass-milter -p /var/run/spamass.sock -d 2
> 11972  p0  R     13:07.12 (spamass-milter)
>   and it will continue doing that until it is killed. I don't think this is
> related to the poll() emulation as this is the same behavior I was seeing 
> with the 0.1.1 version, which used select().

This is wort of weird.  There's really nothing in mlfi_eoh.  All it
does it write a blank line so spamc knows where the headers stop and
the body starts.  Does BSD/OS have strace/truss/ktrace?  Try tracing
the milter while it's hung like that and see what syscalls it's
calling.  A gdb stack trace might be interesting too.

>   If I take a test mail message and do "spamc < test > output", I get
> the expected output, so I'm pretty sure it is in the milter and not
> spamc/spamd. And I tried this with the exact same message that the
> milter wedged on, so that shouldn't be the problem either.

Does it process any emails at all, or is there a special set of killer
messages?  If it doesn't do anything, there might be an interaction
with your poll emulation library, especially since the milter is
threaded.  I can convert the poll to select easy enough and send you a
patch for testing.  Poll is just a nicer interface to use, which is why
I started with it.

        Dan Nelson

reply via email to

[Prev in Thread] Current Thread [Next in Thread]