[KAFKA-9802] Fix flaky system test `TransactionsTest.test_transactions`

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 2.5.1
• Assignee: Jason Gustafson
• Reporter: Jason Gustafson
• Created: 01/Apr/20 22:28
• Updated: 28/May/20 03:59
• Resolved: 28/May/20 03:59

Description

Found when investigating a test failure in `kafkatest.tests.core.transactions_test.TransactionsTest.test_transactions`. The failure was caused by an unexpected transaction timeout. Looking into the logs, I found that the timeout was due to a Produce request which took just over 10s to complete. The test case that failed involved a hard kill of the broker, so it is certainly possible for some produce requests to reach up to the request timeout.

The problem in this case is that `request.timeout.ms` was set to 30s while the `transaction.timeout.ms` was set to only 10s. There is no benefit to allowing a larger request timeout since the coordinator is just going to abort the transaction. So we should fix this test case to set these timeouts consistently. It might also be worth logging a warning to the user when the request timeout is larger than the transaction timeout.

Issue Links

• links to: GitHub Pull Request #8736