An important concept of high availability is node fencing, or physically disabling or separating a node from the cluster under specific circumstances. A somewhat common related method of accomplishing this is to shoot the other node in the head (STONITH).
We technically utilize an aspect of this in other recipes when we connect to the VIP in order to destroy it before establishing it on the local node. One critical flaw in STONITH is the assumption that it's possible to contact the remote system in order to disable it. This may not be possible in cases of network disruption, leaving us with no assurance that the remote primary is actually offline or otherwise disabled.
This is why version 4.4 of repmgr introduced a workaround for this scenario. It is now possible to specify a certain amount of connected child nodes (either standby or witness) must...
