Matthew Wong created KAFKA-9624:
-----------------------------------
Summary: test_throttled_reassignment as EndToEndTest
Key: KAFKA-9624
URL: https://issues.apache.org/jira/browse/KAFKA-9624
Project: Kafka
Issue Type: Bug
Components: system tests
Affects Versions: 2.4.1
Reporter: Matthew Wong
Fix For: 2.4.1
The test_throttled_reassignment test fails because the consumer that is used to
validate reassignment does not start on time to consume all messages. This does
not seem like an issue with the throttling of the reassignment, since
increasing the timeout allowed the test to pass multiple consecutive runs
locally. This test seemed to rely on the default JmxTool for the console
consumer that was removed in this commit:
[{{179d0d7}}|https://github.com/apache/kafka/commit/179d0d73d65ab2c3eb8bc79c70b9893f07038447]
The console consumer would check to see if it had partitions assigned to it
before beginning to consume. Although the test occasionally failed with the
JmxTool, it began to fail much more after the removal. Error messages of
failures followed the below format with varying numbers of missed messages.
They are the first messages by the producer.
```535 acked message did not make it to the Consumer. They are: 0, 1, 2, 3, 4,
5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19...plus 515 more. Total
Acked: 192792, Total Consumed: 192259. We validated that the first 535 of these
missing messages correctly made it into Kafka's data files. This suggests they
were lost on their way to the consumer.```
In the scope of the test, this error suggests that the test is falling into the
race condition described in produce_consume_validate.py, which has the timeout
to prevent the consumer from missing initial messages. Rewriting this test as
an EndToEndTest allows to use its verifiable consumer that can await partition
assignment, addressing the race condition.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
