date: 2019-08-18
- The machine-controller does not delete unhealthy nodes if the machine are present at the cloud provider.
- The machine-controller will delete & recreate a cloud provider instance if it does not join the cluster within a certain amount of time.
- Also the machine-controller will create a new cloud provider instance if the node turns unhealthy (on the init process) and the actual cloud provider instance does not exist anymore.
If e.g. a worker node get in trouble and kublet doesn't respond anymore. We would detect this, but wouldn't do an action on it. Similar to GKE node auto repair we cloud give an end user the option to trigger automatic reprovisioning of the node. Currently he needs to delete the machine object by hand to trigger a new provisioning.
Kubermatic issue: kubermatic/kubermatic#4101