Keeping the platform current.
WHY IT MATTERS
Versions fall out of support quickly, and unsupported means no security fixes.
WHAT THE CADENCE IS
Frequent, compared with operating systems.
WHAT THAT DEMANDS
A routine, rather than a project each time.
WHAT TO CHECK BEFORE UPGRADING
Removed or changed interfaces Whether your manifests use anything deprecated Whether add-ons support the target version
WHAT COMMONLY BREAKS
Manifests using versions that were removed Third-party components lagging behind Custom controllers
WHAT TO DO FIRST
Scan manifests for deprecated usage.
WHAT ORDER TO UPGRADE IN
Control plane, then nodes.
WHY
Newer control planes support older nodes, not the reverse.
WHAT TO NEVER DO
Skip versions, unless explicitly supported.
WHAT TO PRACTISE ON
A non-production cluster, first.
WHAT TO PREPARE
A way back, or acceptance that there is not one.
WHY THAT MATTERS
Control plane upgrades are frequently irreversible.
WHAT TO DO ABOUT NODES
Replace them rather than upgrading in place, where possible.
WHY
It is cleaner, and it tests that workloads reschedule.
WHAT TO VERIFY AFTERWARDS
Every workload running Ingress functioning Storage attached
