ObserveGeneration and lastStableGeneration for FlinkBlueGreenDeployment

[james-kan-shopify]

What is the purpose of the change

This pull request makes FlinkBlueGreenDeployment status generation-aware so external deploy tooling can
distinguish stale parent status from status that belongs to the currently applied spec generation.

Before this change, a BlueGreen parent could still report ACTIVE_BLUE / ACTIVE_GREEN and RUNNING from the
previous generation after Kubernetes had already accepted a new spec/image. This could allow kapp to mark a
deploy successful before the new BlueGreen generation had actually completed, especially around savepoint
handoff where the parent can temporarily return to ACTIVE_* / RUNNING before the new color is live.

The change adds two parent status fields with separate semantics:

• observedGeneration: the latest metadata.generation observed by the operator.
• lastStableGeneration: the latest metadata.generation that reached a stable BlueGreen terminal state.

observedGeneration provides the standard Kubernetes freshness signal used by kapp’s supportsObservedGeneration
behavior. lastStableGeneration provides the BlueGreen-specific completion signal needed because ACTIVE_* /
RUNNING alone is not always proof that the current parent generation is complete.

Brief change log

• Added status.observedGeneration to FlinkBlueGreenDeploymentStatus.
• Added status.lastStableGeneration to FlinkBlueGreenDeploymentStatus.
• Updated the BlueGreen CRD schema and generated custom-resource reference docs.
• Centralized generation stamping in BlueGreenDeploymentService.patchStatusUpdateControl(...).
  • Every status patch records observedGeneration = metadata.generation.
  • lastStableGeneration advances only when the parent status is written as stable terminal status (RUNNING
    or SUSPENDED).
• Added controller no-op backfill for existing stable BlueGreen resources missing the new fields.
• Added tests covering stale generation, stable generation, savepoint handoff, failure, existing-resource
  backfill, initial suspended deploys, and final active stability.

Verifying this change

This change added tests and can be verified as follows:

• Added focused controller tests for:
  • observedGeneration updating when a new spec generation is observed.
  • Existing stable resources backfilling missing generation status fields.
  • lastStableGeneration not advancing when a transition fails before start.
  • lastStableGeneration not advancing during savepoint handoff while parent status is ACTIVE_* / RUNNING.
  • Initial suspended BlueGreen deploys being marked stable for the current generation.
  • Final active BlueGreen deployments being marked stable for the current generation.

Verified locally with:

  mvn -pl flink-kubernetes-operator -am \                                                                      
    -DskipITs -DskipE2E -DskipRat \                                                                            
    -Dcheckstyle.skip -Dspotless.check.skip=true -Dspotless.apply.skip=true \                                  
    -DfailIfNoTests=false \                                                                                    
    -Dtest=FlinkBlueGreenDeploymentControllerTest test                                                         

Result:

  Tests run: 76, Failures: 0, Errors: 0, Skipped: 0                                                            

Also verified formatting with Spotless check for the touched module.

###Does this pull request potentially affect one of the following parts:

• Dependencies (does it add or upgrade a dependency): no
• The public API, i.e., is any changes to the CustomResourceDescriptors: yes
• Core observer or reconciler logic that is regularly executed: yes

Documentation

• Does this pull request introduce a new feature? yes
• If yes, how is the feature documented? docs

[drossos]

Ok wrapped my head around things and seems to make sense. Have some structure changes and a q.

We also need to tophat to ensure would work as expected in deployed env

[drossos]

LGTM after review + discussions. Will merge in after summit and roll-out 👍