Xinyuan Liang
07/08/2025, 11:10 AMXinyuan Liang
07/08/2025, 11:16 AMk describe flinkdeployment)
Flink Version: v2_0
Image: docker-fs-identity-release.dr-uw2.adobeitc.com/identity-data-prep:140fe51148a3642b7d1525164181deb45e9f3372
Image Pull Policy: Always
Job:
Args:
identityDataPrepExecutionEnv
Entry Class: com.adobe.aep.fns.identity.data.prep.app.Main
Jar URI: local:///opt/flink/usrlib/all.jar
Parallelism: 1
State: running
Upgrade Mode: last-state
Job Manager:
Replicas: 1
Resource:
Cpu: 2
Memory: 8g
and
Last Reconciled Spec: {"spec":{"job":{"jarURI":"local:///opt/flink/usrlib/all.jar","parallelism":1,"entryClass":"com.adobe.aep.fns.identity.data.prep.app.Main","args":["identityDataPrepExecutionEnv"],"state":"suspended","savepointTriggerNonce":null,"initialSavepointPath":null,"checkpointTriggerNonce":null,"upgradeMode":"savepoint","allowNonRestoredState":null,"savepointRedeployNonce":null,"autoscalerResetNonce":null},"restartNonce":1751972266, .....}
Last Stable Spec: {"spec":{"job":{"jarURI":"local:///opt/flink/usrlib/all.jar","parallelism":1,"entryClass":"com.adobe.aep.fns.identity.data.prep.app.Main","args":["identityDataPrepExecutionEnv"],"state":"running","savepointTriggerNonce":null,"initialSavepointPath":null,"checkpointTriggerNonce":null,"upgradeMode":"savepoint","allowNonRestoredState":null,"savepointRedeployNonce":null,"autoscalerResetNonce":null},"restartNonce":1751969519, ....}
Finding 2 -- Savepoint/checkpoint timestamp/history are also missing
Job Status:
Checkpoint Info:
Last Periodic Checkpoint Timestamp: 0
Job Id: 42ab796df84427a35779ab94184d2430
Job Name: identity-data-prep
Savepoint Info:
Last Periodic Savepoint Timestamp: 0
Savepoint History:
Start Time: 1751969599559
State: CANCELED
Update Time: 1751972272559
Upgrade Savepoint Path: wasbs://streamingestflink@stageva7identity.blob.core.windows.net/_realtime_flink/dcs/42ab796df84427a35779ab94184d2430/chk-12
P.S.
We did notice that the Flink Operator 1.12 release notes mention a fix for a broken last-state upgrade mode — tracked in FLINK-37571.
We've already upgraded the operator, but are wondering if the strange behavior we observed (e.g., fallback to savepoint or stateless mode) was related to this bug.