Fix clearing of candidates from previous iterations
During initialization the CandidateRegistry wasnt clearing
candidates from the previous instance in a reliable way.
Event with ReadMajority if the cluster members havent been updated
yet, it acted as if there were no candidates present so it did
not clear them.
Introduce a mechansim that goes through the akka singleton that
picks owners, so it only executes the clearing of candidates once
the cluster is in a healthy state.
JIRA: CONTROLLER-2025
Change-Id: I5a43908e45273adfea325621189f633ce493ca87
Signed-off-by: Tomas Cere <tomas.cere@pantheon.tech>
Signed-off-by: Robert Varga <robert.varga@pantheon.tech>
(cherry picked from commit
df7335957f40d6b905b7591b2dc32ffb6093b671)