Re: thread leakage

From: Murray S. Kucherawy <msk_at_blackops.org>
Date: Sun, 20 Mar 2011 23:36:35 -0700 (PDT)

Thanks to Daniel letting me poke around inside the running process, there
does indeed appear to be an issue with timeouts getting ignored inside
libopendkim during certain ADSP-based checks. This means a query whose
DNS answer never comes results in a thread that enters an infinite loop,
but since it spends most of its time asleep waiting for I/O it wouldn't
appear as a tight loop causing the CPU to spike, but rather just take up a
thread slot and associated resources.

I also discovered a slight inconsistency between the libar and libunbound
interfaces that had needed to be handled properly, and wasn't. This is
why the symptoms appeared to be specific to libunbound.

While investigating this, I also found that the DNS callback interval
feature of libopendkim wasn't being used in all the places it should be.

Working on a patch, to be included in v2.3.1. As a result, that release
may be delayed a day or two.

-MSK
Received on Mon Mar 21 2011 - 06:36:54 PST