Ceilometer backend can hang if started before mongo cluster starts

[Gregory Elkinbard]

In this environment Controllers and Mongo nodes are on separate compute elements (VMs for appliance).
If the controllers are started before the mongo nodes come online. Ceilometer backend may hang and not recover.

It should be possible to adjust the pacemaker dependencies list to wait until mongo is operational and then start ceilometer.
Rather then to force the user to do this by hand.

[Vladimir Kuklin (vkuklin)]

I think, this should be fixed in Ceilometer code with connection timeouts for Mongo.

[Igor Degtiarov (idegtiarov)]

Actually database reconnection already implemented and could be configured in ceilometer.conf file in [database] group.

Two parameter could be set there:
* times of reconnection attempts --- `max_retries` default value is 10 and
* timestamp between reconnection attempts --- `retry_interval` (10 sec by default).

Bug should be set to invalid.