Replication is great provided it works. Replication works well if it's understood and works even better if it's tested.
You need to have a plan for the objectives for each individual server in the cluster. Which standby server will be the failover target?
Switchover is a controlled switch from the master to the standby. If performed correctly, there will be no data loss. To be safe, simply shut down the master node cleanly, using either the smart or fast shutdown modes. Do not use the immediate mode shutdown because you will almost certainly lose data that way.
Failover is a forced switch from the master node to a standby because of the loss of the master. So, in that case, there is no action to perform on the master; we presume it is not there anymore.
Next, we need to promote one of the standby nodes to be the new master. A standby node can be triggered into becoming a master node in one of two ways:
pg_ctl promoteSuppose you originally...