pnowojski commented on issue #7405: [FLINK-11249] FlinkKafkaProducer011 can not 
be migrated to FlinkKafkaProducer
URL: https://github.com/apache/flink/pull/7405#issuecomment-458530867
 
 
   Thanks again for the progress in this issue. I think there might be couple 
of issues:
   1. You call snapshot with the same ID twice
   ```
                testHarness.snapshot(1L, 3L);
   
                OperatorSubtaskState snapshot = testHarness.snapshot(1L, 3L);
   ```
   
   2. all 3 of your tests seem to run against the same Kafka brokers, ZooKeeper 
instances. You shouldn't be recovering to the same savepoint twice, without 
reseting external state in Kafka brokers and ZooKeeper. Otherwise second (and 
subsequent) recoveries will be working against Kafka/ZooKeeper state as it was 
left by the first recovery (pending transactions committed/aborted etc).
   
   3. Kafka has a transaction timeout for pending transactions, after which the 
transactions are being aborted. If the time between writing down the 
Kafka/ZooKeeper state and restoring from it is greater then the said timeout, 
probably transactions will be aborted. I'm afraid that this might prove 
impossible to test migration with pending transactions from previous Flink 
versions :/ The only thing that we could test is to write down the 
Kafka/ZooKeeper state with pending transactions with current code version, 
restart Kafka (possibly with newer version) and try to recover from savepoint...

----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
 
For queries about this service, please contact Infrastructure at:
[email protected]


With regards,
Apache Git Services

Reply via email to