Dennis-Mircea Ciupitu created FLINK-40535:
---------------------------------------------
Summary: Standalone session clusters are redeployed instead of
scaled in-place
Key: FLINK-40535
URL: https://issues.apache.org/jira/browse/FLINK-40535
Project: Flink
Issue Type: Bug
Components: Kubernetes Operator
Reporter: Dennis-Mircea Ciupitu
Currently, the {{StandaloneFlinkService#scale}} exposes 2 issues that break the
TaskManager in-place rescaling:
# The JobSpec is not correctly extracted. (this one was introduced in 1.6.0
operator version)
# The replica comparison is performed on an Integer object reference instead
of the boxed value.
Those 2 code issues outline the following real problems:
- Changing {{spec.taskManager.replicas}} on a standalone session cluster
produces a full redeploy instead of in-place scale.
- For 128 replicas or more, two equal counts are distinct objects, so the
operator issues a redundant scale call on every reconcile loop and the "not
scaling" branch becomes unreachable.
Those 2 problems were not seen because the UTs that were backing them were well
implemented and structured.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
