The purpose is to detect the scenario that a failover has happened and the connection to the old master (now replica or no longer part of the cluster) doesn't necessarily mean that we have lost the connection to a master.
Thus we postpone signaling cluster_not_ok until the node down timeout has fired. This also means that reconnect failures don't immediately trigger cluster_not_ok.
A few test cases are affected by this change. Additionally some refactoring of test code is done to be able to reuse parts of it.
The purpose is to detect the scenario that a failover has happened and the connection to the old master (now replica or no longer part of the cluster) doesn't necessarily mean that we have lost the connection to a master.
Thus we postpone signaling
cluster_not_ok
until the node down timeout has fired. This also means that reconnect failures don't immediately triggercluster_not_ok
.A few test cases are affected by this change. Additionally some refactoring of test code is done to be able to reuse parts of it.