reiabreu commented on issue #9077:
URL: https://github.com/apache/storm/issues/9077#issuecomment-6713066842

   *Generated with the help of an LLM.*
   
   **Moving this to 3.3.0; it stays open.**
   
   Part of this shipped in 3.2.0:
   - #9094 added `TupleDeserializationException`, thrown by 
`KryoTupleDeserializer` for tuples from an unknown task or with an unknown 
stream id.
   - #9094 and #9159 added `topology.tuple.deserialization.strict.enable`. When 
it is set, any deserialization failure, including one wrapping an 
`IOException`, kills the receiving worker instead of being dropped.
   
   Still open:
   - **One intentional type in `recv()`.** `DeserializingConnectionCallback` 
still matches 8 exception classes along the cause chain. 
`TupleDeserializationException` extends `IllegalArgumentException`, which is in 
that list, so it is dropped because of its parent class. `KryoException`, 
`BufferUnderflowException` and `NegativeArraySizeException` are not wrapped at 
the source.
   - **The wrap boundary.** In the default mode, an `IllegalArgumentException` 
or `ClassCastException` thrown by a serializer registered through 
`topology.kryo.register` is dropped like a corrupt frame. It gets an ERROR log 
and the `deserializationFailures` metric, but not a fatal error. Making those 
types fatal again would bring back the remote worker-kill that #9074 fixed.
   - **Context in the log.** The drop log has the byte count and destination 
task. It has no source task, stream or input position to tell a framework 
decode error from a serializer error.
   
   One direction that might solve the boundary: wrap each user-registered 
serializer at registration time (in `SerializationFactory`) so its failures are 
rethrown as a distinct type. This has not been tried, and it needs a check that 
it works with Kryo.
   
   Changing which failures are fatal needs its own review, so it is not planned 
for 3.2.0.
   


-- 
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