[KAFKA-9605] EOS Producer could throw illegal state if trying to complete a failed batch after fatal error

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 2.4.0, 2.5.0
• Fix Version/s: 2.6.0
• Assignee: Boyang Chen
• Reporter: Boyang Chen
• Created: 25/Feb/20 18:12
• Updated: 11/Mar/20 22:24
• Resolved: 11/Mar/20 22:24

Description

In the Producer we could see network client hits fatal exception while trying to complete the batches after Txn manager hits fatal fenced error:

 
[2020-02-24T13:23:29-08:00] (streams-soak-trunk-eos_soak_i-02ea56d369c55eec2_streamslog) [2020-02-24 21:23:28,673] ERROR [kafka-producer-network-thread | stream-soak-test-5e16fa60-12a3-4c4f-9900-c75f7d10859f-StreamThread-3-1_0-producer] [Producer clientId=stream-soak-test-5e16fa60-12a3-4c4f-9900-c75f7d10859f-StreamThread-3-1_0-producer, transactionalId=stream-soak-test-1_0] Aborting producer batches due to fatal error (org.apache.kafka.clients.producer.internals.Sender)
[2020-02-24T13:23:29-08:00] (streams-soak-trunk-eos_soak_i-02ea56d369c55eec2_streamslog) org.apache.kafka.common.errors.ProducerFencedException: Producer attempted an operation with an old epoch. Either there is a newer producer with the same transactionalId, or the producer's transaction has been expired by the broker.
[2020-02-24T13:23:29-08:00] (streams-soak-trunk-eos_soak_i-02ea56d369c55eec2_streamslog) [2020-02-24 21:23:28,674] INFO [stream-soak-test-5e16fa60-12a3-4c4f-9900-c75f7d10859f-StreamThread-3] [Producer clientId=stream-soak-test-5e16fa60-12a3-4c4f-9900-c75f7d10859f-StreamThread-3-0_0-producer, transactionalId=stream-soak-test-0_0] Closing the Kafka producer with timeoutMillis = 9223372036854775807 ms. (org.apache.kafka.clients.producer.KafkaProducer)
[2020-02-24T13:23:29-08:00] (streams-soak-trunk-eos_soak_i-02ea56d369c55eec2_streamslog) [2020-02-24 21:23:28,684] INFO [kafka-producer-network-thread | stream-soak-test-5e16fa60-12a3-4c4f-9900-c75f7d10859f-StreamThread-3-1_0-producer] [Producer clientId=stream-soak-test-5e16fa60-12a3-4c4f-9900-c75f7d10859f-StreamThread-3-1_0-producer, transactionalId=stream-soak-test-1_0] Resetting sequence number of batch with current sequence 354277 for partition windowed-node-counts-0 to 354276 (org.apache.kafka.clients.producer.internals.TransactionManager)
[2020-02-24T13:23:29-08:00] (streams-soak-trunk-eos_soak_i-02ea56d369c55eec2_streamslog) [2020-02-24 21:23:28,684] INFO [kafka-producer-network-thread | stream-soak-test-5e16fa60-12a3-4c4f-9900-c75f7d10859f-StreamThread-3-1_0-producer] Resetting sequence number of batch with current sequence 354277 for partition windowed-node-counts-0 to 354276 (org.apache.kafka.clients.producer.internals.ProducerBatch)
[2020-02-24T13:23:29-08:00] (streams-soak-trunk-eos_soak_i-02ea56d369c55eec2_streamslog) [2020-02-24 21:23:28,685] ERROR [kafka-producer-network-thread | stream-soak-test-5e16fa60-12a3-4c4f-9900-c75f7d10859f-StreamThread-3-1_0-producer] [Producer clientId=stream-soak-test-5e16fa60-12a3-4c4f-9900-c75f7d10859f-StreamThread-3-1_0-producer, transactionalId=stream-soak-test-1_0] Uncaught error in request completion: (org.apache.kafka.clients.NetworkClient)
[2020-02-24T13:23:29-08:00] (streams-soak-trunk-eos_soak_i-02ea56d369c55eec2_streamslog) java.lang.IllegalStateException: Should not reopen a batch which is already aborted.
        at org.apache.kafka.common.record.MemoryRecordsBuilder.reopenAndRewriteProducerState(MemoryRecordsBuilder.java:295)
        at org.apache.kafka.clients.producer.internals.ProducerBatch.resetProducerState(ProducerBatch.java:395)
        at org.apache.kafka.clients.producer.internals.TransactionManager.lambda$adjustSequencesDueToFailedBatch$4(TransactionManager.java:770)
        at org.apache.kafka.clients.producer.internals.TransactionManager$TopicPartitionEntry.resetSequenceNumbers(TransactionManager.java:180)
        at org.apache.kafka.clients.producer.internals.TransactionManager.adjustSequencesDueToFailedBatch(TransactionManager.java:760)
        at org.apache.kafka.clients.producer.internals.TransactionManager.handleFailedBatch(TransactionManager.java:735)
        at org.apache.kafka.clients.producer.internals.Sender.failBatch(Sender.java:671)
        at org.apache.kafka.clients.producer.internals.Sender.failBatch(Sender.java:662)
        at org.apache.kafka.clients.producer.internals.Sender.completeBatch(Sender.java:620)
        at org.apache.kafka.clients.producer.internals.Sender.handleProduceResponse(Sender.java:554)
        at org.apache.kafka.clients.producer.internals.Sender.access$100(Sender.java:69)
        at org.apache.kafka.clients.producer.internals.Sender$1.onComplete(Sender.java:745)
        at org.apache.kafka.clients.ClientResponse.onComplete(ClientResponse.java:109)
        at org.apache.kafka.clients.NetworkClient.completeResponses(NetworkClient.java:571)
        at org.apache.kafka.clients.NetworkClient.poll(NetworkClient.java:563)
        at org.apache.kafka.clients.producer.internals.Sender.runOnce(Sender.java:304)
        at org.apache.kafka.clients.producer.internals.Sender.run(Sender.java:239)
        at java.lang.Thread.run(Thread.java:748)

The proper fix is to add a check for handle failed batch in txn manager.

Issue Links

• links to: GitHub Pull Request #8177