CryoThrust opened a new issue, #13293:
URL: https://github.com/apache/seatunnel/issues/13293

   ### Search before asking
   
   Searched open/closed issues and PRs for `TaskReportStatusOperation` and 
task-status wire validation, and checked the file lists of 400 open PRs. 
#12747/#12748 cover `TaskOperation` and `AssignSplitOperation`, not this 
status-report reader. The #12748 review explicitly suggested this separate 
follow-up. No open PR in that scan changes `TaskReportStatusOperation.java`.
   
   ### What happened
   
   `TaskReportStatusOperation.readInternal` still uses:
   
   ```java
   location = in.readObject(TaskLocation.class);
   status = in.readObject();
   ```
   
   Hazelcast's class argument overrides construction rather than validating the 
serialized class ID. With a controlled payload carrying the registered TaskData 
factory's `PROGRESS_TYPE` ID and TaskLocation-compatible fields, the reader 
accepts a `TaskLocation` and then reads the following `RUNNING` state 
successfully. The wire-selected class was not a TaskLocation.
   
   This is a real-Hazelcast, in-process serialization diagnostic, not a claim 
that an ordinary Progress writer produces TaskLocation-compatible fields or 
that malformed data was sent to a running cluster. Valid location/status round 
trips preserve their bytes. Null is already rejected by Hazelcast's typed 
deserializer; null acceptance is not claimed.
   
   ### SeaTunnel Version
   
   Current dev `50e8a0d30db53b6be77509fee8873a2439716171`. The tested local 
server classes were built previously at `6abd36f7d`; 
`TaskReportStatusOperation`, `TracingOperation` and `TaskLocation` have no 
source differences between those revisions. `javap` confirms the tested reader 
invokes the typed overload. This is not a fresh full-engine build result on dev.
   
   ### SeaTunnel Config
   
   Not applicable: the probe serializes/deserializes operation bodies using 
`DefaultSerializationServiceBuilder`, without a cluster or job.
   
   ### Running Command
   
   [Standalone probe and portable reproduction 
instructions](https://gist.github.com/CryoThrust/801bbd5c6e1fd7cc303371b981b4f99c).
   
   ### Error Exception
   
   The wrong-class fixture is incorrectly accepted, so there is no exception 
for that case:
   
   ```text
   PASS valid TaskReportStatusOperation location/status and byte-identical 
round trip
   REPRODUCED: registered PROGRESS_TYPE payload accepted as TaskLocation; 
following RUNNING status intact
   CONTROL: null is already rejected by Hazelcast typed deserialization: 
HazelcastSerializationException
   ```
   
   ### Zeta or Flink or Spark Version
   
   Zeta operation reader; no engine E2E run.
   
   ### Java or Scala Version
   
   Amazon Corretto 11.0.25, macOS ARM64.
   
   ### Proposed scope / Are you willing to submit PR?
   
   Yes, I can take this as a narrow follow-up: decode the wire-selected object, 
reject a non-TaskLocation at the reader with an informative IOException, and 
retain explicit null rejection. Add real Hazelcast same-factory/different-class 
and valid/null controls, preserving the following task state, serializer IDs 
and valid wire layout. No change to task-status transition logic, retries or 
the two readers already covered by #12748.
   
   ### Code of Conduct
   
   - [x] I agree to follow this project's [Code of 
Conduct](https://www.apache.org/foundation/policies/conduct).
   


-- 
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