Closed r7vme closed 6 years ago
Since this is a rather ugly workaround I hope we can improve this later on. Is there some movement in your k8s-client issue? I would prefer getting the design right with the circuit breaker idea and create a new client to get new TCP connections on each reconciliation.
there is a silence https://github.com/kubernetes/client-go/issues/342
Fixes: https://github.com/giantswarm/kvm-operator-node-controller/issues/5
Tested, now node-controller restarted if master dies.
1) count errors 2) healthz will response with http 500 error count is not 0