Logger should be got by oslo.log getLogger instead of logging package

[Yang Youseok]

I was trying to off log message in oslo.messaging package using `default_log_level` in config file, but debug message from oslo.messaging keep emitting.

After I dig the source, some modules like _drivers/amqpdriver.py did not use getLogger in oslo.log package, so user could not set log level for the module.

Åt the same time, module like _drivers/impl_rabbit.py do use oslo.log getLogger so that I can set the log level normally.

I'm not sure it's intentional, but it seems to be bug since user expect the `default_log_level` affect the entire library package.

[Ken Giusti (kgiusti)]

Your observations are correct.

I talked a bit with Doug Hellman regarding what you're seeing. Way back we 'de-namespaced' the modules such that "oslo.messaging" became "oslo_messaging". This created problems for folks that had "oslo.messaging=<WHATEVER>" in their default_log_levels setting since the module name changed to 'oslo_messaging' and their log setting were being ignored.

To help fix that a workaround was added to oslo.log:

http://git.openstack.org/cgit/openstack/oslo.log/tree/oslo_log/log.py#n446

So whenever oslo_log's getLogger is called it converts the name "oslo_messaging" to "oslo.messaging". However as you noticed some of the oslo.messaging files use the python library's getLogger, which doesn't do the translation and will not filter based on "oslo.messaging=<WHATEVER>"

So this is indeed a bug in oslo.messaging (as far as I'm concerned).

Workaround: include both "oslo.messaging=<WHATEVER>" and "oslo_messaging=<WHATEVER>" in your default_log_levels (note the '_' vs '.')

[Yang Youseok (ileixe)]

Yes, I found the related code in oslo.log and expected maybe there was historical reason for the code. :)

Imho, before the oslo.log changed the default value of default_log_level config to use "oslo_messaging", it's better to use logging in oslo.log for the other modules.

I found oslo.cache which is specified in "default_log_level" also only use the logger got from oslo.log.

How about that?

[Ken Giusti (kgiusti)]

Adding in Doug who has more context on the oslo.log change.

Background:

The original bug:
https://bugs.launchpad.net/mos/+bug/1484501

See Doug's comment here:
https://review.openstack.org/#/c/213795/2/oslo_log/log.py

Looks like "oslo_messaging" was added to the defaults by this patch:

https://review.openstack.org/#/c/458069/

From Doug's comment on log.py it appears that the "oslo.<xxx>" format is the desired format.

The addition of "oslo_messaging" to the defaults appears to be a work around to oslo.messaging's inconsistent use of logging.

IMHO all oslo.<whatever> modules that have a dependency on oslo.log should use oslo.log for all logging. Using both oslo.log and py logging by the same module should be a bug (unless there's a specific use case requirement - like notification logging).

So if oslo.cache is using both oslo.log and py logging it should be fixed to use only oslo.log.

Doug - what say ye?