lrsb created FLINK-39814:
----------------------------
Summary: FlinkStateSnapshot-triggered savepoints ignore the
referenced FlinkSessionJob state.savepoints.dir
Key: FLINK-39814
URL: https://issues.apache.org/jira/browse/FLINK-39814
Project: Flink
Issue Type: Bug
Components: Kubernetes Operator
Affects Versions: 1.14.0, 1.13.0, 1.12.0, 1.11.0, 1.10.0
Reporter: lrsb
When a savepoint is triggered through the FlinkStateSnapshot path for a job
whose
jobReference.kind is FlinkSessionJob, the operator resolves the savepoint target
directory from the parent FlinkDeployment's observe config instead of the
referenced
FlinkSessionJob's merged config. As a result, a per-job
spec.flinkConfiguration["state.savepoints.dir"] override is silently dropped
and all
savepoints land flat under the cluster-level state.savepoints.dir.
This also affects all operator-triggered savepoints (upgrade savepoints,
savepoint-on-deletion, manual triggers) when
kubernetes.operator.snapshot.resource.enabled is true, because those are all
routed
through the FlinkStateSnapshot CRD.
Per-job state.checkpoints.dir is not affected as Flink bakes it into the job's
CheckpointStorage at submission time, so checkpoints correctly land in the
per-job dir.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
