How it Works
This section describes how the feature works.
This feature enables users to group servers into upgrade groups, which are similar to availability zones. The nodes in each upgrade group are upgraded in parallel (the maximum number of parallel nodes supported is 15).
The upgrade groups are upgraded in a sequential manner. For example, the control plane groups are not upgraded concurrently with the worker groups, but one at a time.
Requirements and Limitations
Some requirements and limitations associated with this feature are as follows.
-
If the feature is disabled, the SMI reverts to the previous method of performing consecutive upgrade for the nodes.
-
If the feature is enabled, the upgrade group configuration becomes mandatory for all nodes.
-
There must be a majority of control plane or etcd nodes running at all time (for example, two out of three control planes should be always running).
-
All the worker or KVM nodes should be distributed among the upgrade zone in a manner that ensures the majority of nodes never gets upgraded at the same time.
-
The upgrade groups feature applies to the control plane, worker, KVM, and etcd node types,but doesn't apply to the CM-HA nodes.
-
For the K8s clusters, the nodes include the upgrade group name as a new label. This label enables the NFs to use the affinity and anti-affinity rules to achieve proper HA and replication. The NFs can use the upgrade-zone provisioned node label or use custom defined ones to enable the application to align with the affinity rule.