CryoThrust opened a new pull request, #13294:
URL: https://github.com/apache/seatunnel/pull/13294

   ### Purpose of this pull request
   
   Yes, I am contributing to Apache SeaTunnel.
   
   Fixes #13293. This is the separate `TaskReportStatusOperation` follow-up 
suggested in the review of #12748; it does not include that PR's two readers.
   
   Hazelcast's `readObject(TaskLocation.class)` overrides object construction 
rather than validating the class ID in the message. A registered `Progress` ID 
with TaskLocation-shaped fields is therefore accepted as a task location by the 
current reader. Decode the wire-selected object first, then reject a 
non-TaskLocation with an `IOException` naming the expected and actual types.
   
   ### Does this PR introduce _any_ user-facing change?
   
   No configuration or normal task-execution changes. Serializer IDs, write 
order, status handling and retry behavior are unchanged. Invalid location 
values now fail explicitly at deserialization. Null was already rejected by 
Hazelcast and remains rejected, with a clearer diagnostic.
   
   ### How was this patch tested?
   
   Local results on JDK 11:
   
   - **26 unit tests passed**, with no failures, errors or skips. The four new 
tests use real Hazelcast serialization, including the operation envelope and 
production factories: all task states round-trip with the original location and 
identical bytes; a same-factory/different-class fixture, an unrelated value and 
null are rejected. The other 22 tests cover neighboring task-state and record 
serializers.
   - **The updated three-node failover IT passed.** Before switching masters, 
it now verifies that the worker reports contain exactly the planned task IDs. 
The existing checks still verify idle-reader closure, a newly completed 
checkpoint after failover, unchanged output rows and job cancellation.
   - Scoped Spotless formatting/check and both modules' skip-tests verification 
passed.
   
   Before the reader change, the new wrong-class test fails because no 
exception is thrown. This is a controlled malformed-wire regression, not a 
claim that an ordinary Progress writer emits TaskLocation fields.
   
   Equivalent commands from the repository root, with reactor dependencies 
built:
   
   ```bash
   ./mvnw -pl :seatunnel-engine-server \
     
-Dtest=TaskReportStatusOperationSerializationTest,TaskStateSerializationTest,RecordSerializerTest,ResourceDataSerializerHookTest,CleanLogOperationSerializationTest
 test
   
   ./mvnw -pl :seatunnel-engine-server,:connector-seatunnel-e2e-base \
     -DskipUT=true -DskipIT=false -DfailIfNoTests=false \
     
-Dit.test=CheckpointCoordinatorFailoverIT#testCheckpointCompletesAfterMasterFailoverWithIdleClosedReader
 verify
   
   ./mvnw -pl :seatunnel-engine-server,:connector-seatunnel-e2e-base 
spotless:apply
   ./mvnw -pl :seatunnel-engine-server,:connector-seatunnel-e2e-base 
-DskipTests verify
   ```
   
   The local runs capped test JVMs at two processors and 2 GiB heap. These are 
scoped checks, not a full reactor or connector-matrix result.
   
   ### Check list
   
   - [x] Regression tests and the existing engine IT updated.
   - [x] No new dependencies, binary packages, configuration options or 
incompatible valid-wire changes; documentation and connector registration 
updates are not applicable.
   


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to