Calico Shop created FLINK-40119:
-----------------------------------
Summary: Blue/Green: after an aborted transition, the next deploy
deletes Blue immediately instead of waiting the deletion delay
Key: FLINK-40119
URL: https://issues.apache.org/jira/browse/FLINK-40119
Project: Flink
Issue Type: Bug
Components: Kubernetes Operator
Affects Versions: kubernetes-operator-1.15.0
Reporter: Calico Shop
When a Blue/Green transition is aborted, the deployment rolls back to
ACTIVE_BLUE but `deploymentReadyTimestamp` is not cleared - it keeps the value
from the aborted transition.
On the next deployment, the delete-Blue gate computes `deletionTimestamp =
deploymentReadyTimestamp + deletionDelay`. Because the timestamp is stale (from
the previous transition), this value is already in the past, so Blue is deleted
the instant Green first reports RUNNING 0 skipping the deletion delay entirely.
Steps to reproduce:
1. Deploy a Blue/Green deployment (ACTIVE_BLUE).
2. Trigger a transition to Green and let Green be marked ready (sets
`deploymentReadyTimestamp`).
3. Cause the transition to abort (rolls back to ACTIVE_BLUE).
4. Trigger another deployment.
5. Observe Blue is deleted immediately once Green reports RUNNING, without
honoring the deletion delay.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
