replication ,eject-replica-source ,validate_can_perform_action

[gang]

In replication, when a master down, call eject_replica_source to eject the old master and promote a slave as a new master.
but in func eject_relica_source in trove/instances/models.py it call self.validate_can_perform_action() to test master not in "not active" state ,that is contradictory

[Amrith Kumar (amrith)]

Please elaborate on why you think this is contradictory, what problem(s) do you face?

[Morgan Jones (6-morgan)]

I believe the code is correct - you should only be able to eject a master if it is active, because if there is a detectable issue with the master, it should be corrected rather than ejected. The reason to eject a master is that it is not running or reachable, but that would not cause it to not be marked ACTIVE (since we would have no way to tell if it was indeed active, but simply not reachable from the controller).

[gang (zhg0517)]

In my understanding，a master can be ejected when it is out of service, if it healthy we can use promote_to_replica_source to exchange master and slave.

I test eject function in the follow steps:
1. stop mysql service on master, the vm is active and mysql is down, but I met an error
"if last_heartbeat_delta < agent_expiry_interval:
            raise exception.BadRequest(_("Replica Source %s cannot be ejected"
                                         " as it has a current heartbeat")
                                       % self.id)
"
2. shut down the vm,and the status of database instance is SHUTDOWN, then I get an error from self.validate_can_perform_action()
"msg = (_("Instance %(instance_id)s is not currently available for an "
                 "action to be performed (status was %(action_status)s).") %
               {'instance_id': self.id, 'action_status': status})
"

I think the above two condictions could use eject operation, but I met errors.
Could you please list one condiction the use the eject function correctlly, thanks.

[Amrith Kumar (amrith)]

Just verified and code behaves as expected. reduce your heartbeat timeout and try again.

[Amrith Kumar (amrith)]

Morgan, maybe the thing that makes your description confusing is the sentence, "The reason to eject a master is that it is not running or reachable, but that would not cause it to not be marked ACTIVE (since we would have no way to tell if it was indeed active, but simply not reachable from the controller)."

three negatives with an interesting nesting make it a little harder to parse than one would expect.