RE: Runaway threads with opendkim-2.3.2

From: Murray S. Kucherawy <msk_at_cloudmark.com>
Date: Mon, 9 May 2011 11:21:00 -0700

> -----Original Message-----
> From: opendkim-dev-bounce_at_lists.opendkim.org [mailto:opendkim-dev-bounce_at_lists.opendkim.org] On Behalf Of Gary Mills
> Sent: Monday, May 09, 2011 11:09 AM
> To: opendkim-dev_at_lists.opendkim.org
> Subject: Runaway threads with opendkim-2.3.2
>
> The two types of traceback seem to alternate. What could be happening?
>
> This was run on a 16-core Intel server. At busy times it was handling
> over 10,000 SMTP sessions per hour. I suppose a cunningly-crafted e- mail
> message could have caused this mayhem. An unprotected variable shared
> between threads could also do it. Or, maybe it's a bug?

The trace shows three different threads attempting to get a mutex inside the library. Since this is a snapshot of a running instance, it could be that there wasn't a problem; one thread held the lock and would appropriately release it shortly thereafter. The other possibility is that it shows a deadlock caused by one thread having acquired the lock and then terminated without releasing it, or the more common deadlock scenario where two threads already hold different locks, and then each decides it wants the other.

I'm pretty sure during a previous report from you that I'd gone over this with a fine-toothed comb to make sure neither of these can happen, but I'll take another run at it.
Received on Mon May 09 2011 - 18:21:07 PST