Juno glance spins with use_syslog=True

[Mark Goddard]

With syslog enabled on Juno (use_syslog=True in glance-api.conf), glance spins on startup, consuming all available CPU cycles.

With some carefully placed calls to exit(), the line in glance causing the problem was determined to be https://github.com/openstack/glance/blob/stable/juno/glance/openstack/common/log.py#L523. After this point the code gets lost in logging.handlers.SyslogHandler().

I noticed that the <component>.openstack.common.log source file was somewhat different in glance compared to several other OpenStack components. An update to glance.openstack.common missed the juno branch by a couple of commits https://github.com/openstack/glance/commit/d4f890798a0588a80e21fad044d4df5d0964dc70. Cherry picking this commit fixed the issue for me.

[Louis Taylor (kragniz)]

This should be fixed when glance moves to oslo_log. Work in progress patch: https://review.openstack.org/#/c/149344/

[Mark Goddard (mgoddard)]

Ok, that's good to hear. Will the change be applied to stable/juno? If not, should we apply the commit listed above as a cherry pick from master?

[Mark Goddard (mgoddard)]

Apologies, the actual config option is use_syslog, rather than enable_syslog.

[Ian Cordasco (icordasc)]

The switch to oslo_log from glance/openstack/common/log.py is not going to be backported to stable/juno, no. This is definitely something I have heard of people encountering before (and I'm frankly surprised there wasn't already a bug for it).