[KAFKA-8688] Upgrade system tests fail due to data loss with older message format

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 2.4.0
• Component/s: system tests
• Assignee: Rajini Sivaram
• Reporter: Rajini Sivaram
• Reviewer: Ismael Juma
• Created: 19/Jul/19 12:45
• Updated: 31/Jul/19 15:24
• Resolved: 31/Jul/19 15:24

Description

System test failure for TestUpgrade/test_upgrade: from_kafka_version=0.9.0.1, to_message_format_version=0.9.0.1, compression_types=.lz4

3 acked message did not make it to the Consumer. They are: [33906, 33900, 33903]. The first 3 missing messages were validated to ensure they are in Kafka's data files. 3 were missing. This suggests data loss. Here are some of the messages not found in the data files: [33906, 33900, 33903]

Traceback (most recent call last):
  File "/home/jenkins/workspace/system-test-kafka_5.3.x/kafka/venv/local/lib/python2.7/site-packages/ducktape-0.7.5-py2.7.egg/ducktape/tests/runner_client.py", line 132, in run
    data = self.run_test()
  File "/home/jenkins/workspace/system-test-kafka_5.3.x/kafka/venv/local/lib/python2.7/site-packages/ducktape-0.7.5-py2.7.egg/ducktape/tests/runner_client.py", line 189, in run_test
    return self.test_context.function(self.test)
  File "/home/jenkins/workspace/system-test-kafka_5.3.x/kafka/venv/local/lib/python2.7/site-packages/ducktape-0.7.5-py2.7.egg/ducktape/mark/_mark.py", line 428, in wrapper
    return functools.partial(f, *args, **kwargs)(*w_args, **w_kwargs)
  File "/home/jenkins/workspace/system-test-kafka_5.3.x/kafka/tests/kafkatest/tests/core/upgrade_test.py", line 136, in test_upgrade
    self.run_produce_consume_validate(core_test_action=lambda: self.perform_upgrade(from_kafka_version,
  File "/home/jenkins/workspace/system-test-kafka_5.3.x/kafka/tests/kafkatest/tests/produce_consume_validate.py", line 112, in run_produce_consume_validate
    self.validate()
  File "/home/jenkins/workspace/system-test-kafka_5.3.x/kafka/tests/kafkatest/tests/produce_consume_validate.py", line 135, in validate
    assert succeeded, error_msg
AssertionError: 3 acked message did not make it to the Consumer. They are: [33906, 33900, 33903]. The first 3 missing messages were validated to ensure they are in Kafka's data files. 3 were missing. This suggests data loss. Here are some of the messages not found in the data files: [33906, 33900, 33903]

Logs show:

1. Broker 1 is leader of partition
2. Broker 2 successfully fetches from offset 10947 and processes request
3. Broker 2 sends fetch request to broker 1 for offset 10950
4. Broker 1 sets is HW to 10950, acknowledges produce requests up to HW
5. Broker 2 is elected leader
6. Broker 2 truncates to its local HW of 10947 - 3 messages are lost

This data loss is a known issue that was fixed under KIP-101. But since this can still happen with older messages formats, we should update upgrade tests to cope with some data loss.