[FLINK-25514] YARN session tests scan unrelated logs for bugs

• Type: Bug
• Status: Open
• Priority: Not a Priority
• Resolution: Unresolved
• Affects Version/s: 1.13.5, 1.14.2, 1.15.0
• Component/s: Deployment / YARN
• Labels: auto-deprioritized-minor test-stability
• Assignee: Unassigned
• Reporter: Niklas Semmler
• Created: 04/Jan/22 09:47
• Updated: 12/Jul/22 22:37

Description

Many YARN tests that inherit from org.apache.flink.yarn.YarnTestBase deploy a Flink cluster via YARN. The class provides the method ensureNoProhibitedStringInLogFiles to scans the Flink cluster log files for exceptions.

At its current state the method cannot distinguish between log files from different test cases that are part of the same test class. (E.g., org.apache.flink.yarn.YARNSessionFIFOSecuredITCase#testDetachedModeSecureWithPreInstallKeytab and org.apache.flink.yarn.YARNSessionFIFOSecuredITCase#testDetachedMode) Thereby, test cases fail only due to a failure of an earlier test case. (E.g., testDetachedMode fails due to a failure of testDetachedModeSecureWithPreInstallKeytab)

The filtering method should be improved, to make failures better understandable. To this end, the filtering method needs to filter on the container suffix of the path. For the YARN container name container_1640947393046_0001_01_000001 it needs to map the 0001 part to the number of the test case considering only test cases that spawn clusters. Or/and the naming of the containers need to be adapted to make this easier. An alternative solution would be to start one YARN cluster per test case (instead of as per test class).