Our load balancer health checks (and other folks too) just load the main glance URL and look for an http status of 300 to determine if glance is okay. Starting I think in Kilo, glance changed and now logs a warning. This is highly unnecessary and ends up generating gigs of useless logs which make diagnosing real issues more difficult.
At the least this should be an INFO, but ideally there's no point in logging this at all.
2015-08-04 17:42:43.058 24075 WARNING glance.api.middleware.version_negotiation [-] Unknown version. Returning version choices.
2015-08-04 17:42:43.577 24071 WARNING glance.api.middleware.version_negotiation [-] Unknown version. Returning version choices.
2015-08-04 17:42:45.083 24076 WARNING glance.api.middleware.version_negotiation [-] Unknown version. Returning version choices.
2015-08-04 17:42:45.317 24064 WARNING glance.api.middleware.version_negotiation [-] Unknown version. Returning version choices.
2015-08-04 17:42:47.092 24074 WARNING glance.api.middleware.version_negotiation [-] Unknown version. Returning version choices.
I've met this before.
It seems that your health checkers returns request versions with the following path: '/'. Starting from kilo, version negotiator considers this as multiple choices request cause it doesn't know which version you are requesting.
In the current implementation glance have a special request '/versions' to return all available versions in glance.
As a possible workaround I would recommend to change the health check request to /versions.