270922 (3) [Avatar] Offline
#1
Under topic "ENSURING HIGH AVAILABILITY OF THE CONTROLLERS AND THE SCHEDULER" it is said:
"The Controller Manager and Scheduler can run collocated with the API server and etcd, or they
can run on separate machines. When collocated, they can talk to the local API server directly,
otherwise they connect to the API servers through the load balancer."

If Controller Manager and Scheduler talks to the local API server, isn't there a risk of system halting if the local API server goes down?
Marko Lukša (49) [Avatar] Offline
#2
No. Let's say the Controller Manager running on master A is the active Controller Manager in the cluster (it was elected as the leader). If the API server on master A dies, the Controller Manager will no longer be able to update the resource that designates who the leader is. This will allow another Controller Manager on a different master node to take the leader position.