You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Worker nodes also respect some kind of upgrade strategy such that not all of them are updated simultaneously. It would include maxUnavailable count, default to 1 for control plane/etcd nodes.
max_unavailable_worker: 10% is available by default on a cluster
With default values upgrade_strategy, the sequence of how the upgrade happens in the cluster is given below
- etcd nodes - will upgrade one by one with the etcd components upgrading
- control plane nodes - will upgrade one by one - will be cordoned when this is happening. Control plane components - sidekick, kube-apiserver, kube-controller-manager, kube-scheduler and worker plane components - kubelet, kubeproxy
- etcd nodes will be cordoned - when the worker plane components - kubelet, kubeproxy are getting upgraded
- currently in rancher - worker plane nodes and add ons will get executed/upgraded in parallel.
- Default max_unavailable_worker: 10% so 1 worker node gets upgraded at a time.
If Drain is enabled, the nodes are drained.
Else they go into cordoned state. and then Upgrade and to Active.
State transition wrt to worker nodes from UI
When Drain : false
- Active --> Cordoned (Upgrading message) --> Active
When Drain: True
- Active --> Cordoned (Upgrading message) (might not be visible as it might happen really fast --> Draining --> Drained (Upgrading message) --> Active
Worker nodes also respect some kind of upgrade strategy such that not all of them are updated simultaneously. It would include maxUnavailable count, default to 1 for control plane/etcd nodes.
#23777
The text was updated successfully, but these errors were encountered: