[openstack-dev] Proposed Logging Standards

Sean Dague sean at dague.net
Mon Jan 27 14:59:15 UTC 2014

On 01/27/2014 09:44 AM, Macdonald-Wallace, Matthew wrote:
>> -----Original Message-----
>> From: Sean Dague [mailto:sean at dague.net]
>> Sent: 27 January 2014 14:26
>> To: OpenStack Development Mailing List (not for usage questions)
>> Subject: Re: [openstack-dev] Proposed Logging Standards
>> On 01/27/2014 09:07 AM, Macdonald-Wallace, Matthew wrote:
>>> Hi Sean,
>>> I'm currently working on moving away from the "built-in" logging to use
>> log_config=<filename> and the python logging framework so that we can start
>> shipping to logstash/sentry/<insert other useful tool here>.
>>> I'd be very interested in getting involved in this, especially from a "why do we
>> have log messages that are split across multiple lines" perspective!
>> Do we have many that aren't either DEBUG or TRACE? I thought we were pretty
>> clean there.
> True, most (all?!!) are DEBUG/TRACE and mainly from calling out to other clients (Neutron/Glance/Cinder/etc), but if you're sending DEBUG somewhere useful for future processing then trying to glue the split-lines back together again can be "interesting".
> At the moment we are assuming that anything that doesn't start with a date-stamp is associated with the line above it.  This is probably OK for now, however if anything changes in future that negates this rule we won't catch it until it's too late!
>>> P.S. FWIW, I'd also welcome details on what the "Audit" level gives us
>>> that the others don't... :)
>> Well as far as I can tell the AUDIT level was a prior drive by contribution that's
>> not being actively maintained. Honestly, I think we should probably rip it out,
>> because I don't see any in tree tooling to use it, and it's horribly inconsistent.
> +1 for this, I wondered if it was something to do with Ceilometer but I'm guessing probably not from your comment here.
> I've also noticed just now that we appear to be "re-inventing" some parts of the logging framework (openstack.common.log.WriteableLogger for example appears to be a "catchall" when we should just be handing off to the default logger and letting the python logging framework decide what to do IMHO).

WriteableLogger exists for a very specific reason: eventlet. Eventlet
assumes a file object for logging, not a python logger.

I've proposed a change for that -
https://github.com/eventlet/eventlet/pull/75 - but it's not yet upstream.


Sean Dague
Samsung Research America
sean at dague.net / sean.dague at samsung.com

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 547 bytes
Desc: OpenPGP digital signature
URL: <http://lists.openstack.org/pipermail/openstack-dev/attachments/20140127/1f12a514/attachment.pgp>

More information about the OpenStack-dev mailing list