All db sync should be done only at primary controller

[Bogdan Dobrelya]

Related bug https://bugs.launchpad.net/fuel/+bug/1330875

Fuel deploys controllers in parallel, and since non-blocking SST was introduced for galera cluster deployment in Fuel, race conditions are *very* likely to appear for all DB entities.

To fix it we should ensure all db sync operations would be issued only for primary controller node.

Another point is, that in order to run patching (or upgrading) smoothly, all db syncing should be done with refreshonly = false

[Vladimir Kuklin (vkuklin)]

This is almost not needed as db_sync for already migrated database will be instant and not wasting any resources.

[Vladimir Kuklin (vkuklin)]

I will set this bug as confirmed but severity seems medium to me unless proven otherwise.

[Bogdan Dobrelya (bogdando)]

Another point is, that in order to run patching (or upgrading) smoothly, all db syncing should be done with refreshonly = false, but we use true (and puppet upstream do manifests the same) - see https://review.openstack.org/#/c/117217 for example

[Matthew Mosesohn (raytrac3r)]

Package updates should refresh db_sync, but it is probably a good idea to force refreshonly = false to be sure that the DB schema is updated. We should either do patching upgrades one at a time or keep refreshonly = true for secondary controllers.

[Bogdan Dobrelya (bogdando)]

Raised to high due to this issue relates to many changes as well

[Bogdan Dobrelya (bogdando)]

Looks like the comment #1 is absolutely correct, no need for that change.

[Bogdan Dobrelya (bogdando)]

I set this bug as a dup of the newer bug #1330875 as the latter looks better described and has all of the required milestone in place.