Things that cannot simply be replaced.
WHAT MAKES AN APPLICATION STATEFUL
Data it holds that must survive An identity other components depend on Ordering requirements between instances
WHAT EXAMPLES LOOK LIKE
Databases Message brokers Anything with a local cache that matters
WHAT ORCHESTRATORS PROVIDE FOR THESE
Stable names Storage that follows each instance Ordered startup and shutdown
WHY ORDERING MATTERS
Clustered software frequently requires members to start in sequence.
WHAT REMAINS YOUR PROBLEM
Backups Upgrades Failover behaviour of the software itself
WHY THAT NEEDS SAYING
The orchestrator manages containers, not data correctness.
WHAT TO BE CAREFUL WITH
Scaling down, which may delete storage depending on policy Automatic rescheduling moving an instance away from its data Storage that cannot attach to more than one node
WHAT TO SET
A reclaim policy retaining storage.
WHAT TO TEST DELIBERATELY
Losing a node holding an instance Restoring from backup Scaling up and down
WHAT TO CONSIDER
Whether the application belongs in the cluster at all.
WHY
A managed database removes most of this.
WHAT TO PREFER WHEN UNSURE
Stateless workloads in the cluster, state outside it.
