Re: Problems with stats

From: Graham Murray <graham_at_gmurray.org.uk>
Date: Sat, 04 Sep 2010 08:37:20 +0100

"Murray S. Kucherawy" <msk_at_blackops.org> writes:

> On Sat, 4 Sep 2010, Graham Murray wrote:
>> 1) If opendkim-stats (or some other program) has the stats file open
>> when the opendkim wants to write a stats record, the open in append
>> mode fails and not only is that entry not made but the opendkim stops
>> recording stats.
>
> That's odd, because the new code does an fopen() and doesn't do any
> locking at all. What sorts of errors other than permissions errors or
> disk full, etc. could cause such a problem? Whatever they are, they
> should be logged.
>
> The freezing of stats recording is a hold-over from the previous
> version when it was a hash table instead of a flat file. I've removed
> that for the next Beta.

I was wrong about the reason for stats stopping being added to the
file. Looking at the logs it seems that opendkim was segfaulting and had
reached its restart limit. I have now rebuilt with debugging and turned
on EnableCoredumps and will analyse the core file if it crashes
again. Just before each crash syslog has lines such as the following

Sep 3 19:08:17 newton opendkim[21276]: o83I8Bo8021681: ADSP query: `_adsp._domainkey.ymail.com' query error
Sep 3 19:08:17 newton sm-mta[21681]: o83I8Bo8021681: Milter insert (1): header: Authentication-Results: newton.gmurray.org.uk; dkim=pass (768-bit key)\n\theader.i=_at_yahoogroups.com header.b=PFeLykK3; dkim-adsp=temperror
Sep 3 19:08:17 newton sm-mta[21681]: o83I8Bo8021681: milter_sys_read(dkim-filter): cmd read returned 0, expecting 5
Sep 3 19:08:17 newton sm-mta[21681]: o83I8Bo8021681: Milter (dkim-filter): to error state
Sep 3 19:08:17 newton opendkim[17328]: terminated with signal 11, restarting


>> 2) README lists 20 fields in signature records, opendkim-stats expects
>> 19 but there are only 18 in the outputted record. This causes
>> opendkim-stats to give an "an expected field count" error for every
>> stats line.
>
> What are your build-time options? This is a bug with
> "--enable-diffheaders" that I also found earlier today, but it wasn't
> happening before I turned that on. With that compiled, the output was
> one field short. I'll attach a patch that fixes this. If you're not
> using that then something else must be wrong.

My build options are --enable-diffheaders --enable-vbr --enable-stats --enable-report_intervals --enable-dkim_reputation --with-lua
Received on Sat Sep 04 2010 - 07:37:32 PST