Leader election cluster is best advise in such cases:
This is achieved via some drawbacks:
Leader Election Asynchronous Cluster can guarantee continuity of service at no performance cost for the leader and in some specific conditions with "No Data Loss".
In the field, regular scenario is to have long periods of time between hardware crashes: it is a requirement to repair the replication stream as fast as possible, replication-manager not only helps to do this, it also track failover SLA (Service Level Availability) to give an historical view on the replication stream latency and allow day to day switchover to enable database maintenance.
It is not always desirable to perform automatic failover in an asynchronous cluster, replication-manager enforces some tunable settings to constraint the architecture state in which the failover can happen.
We can classify SLA for automatic failover scenario into 3 states:
Staying in sync
When the replication can be monitored in sync, the failover can be done without loss of data, provided that replication-manager waits for all replicated events to be applied to the elected replica, before re-opening traffic. In order to reach this state most of the time, we advise next section settings.