Purushottam Sinha created FLINK-39963:
-----------------------------------------

             Summary: Runtime: 
ExecutionTimeBasedSlowTaskDetectorTest.testBalancedInput flaky
                 Key: FLINK-39963
                 URL: https://issues.apache.org/jira/browse/FLINK-39963
             Project: Flink
          Issue Type: Bug
          Components: Runtime / Coordination
            Reporter: Purushottam Sinha


Problem
The test fails intermittently in CI with AssertionError: Expected size: 2 but 
was: 0, failing the core test module. It is a test-only timing race, not a 
production defect.

Evidence
  - Assertion fails at ExecutionTimeBasedSlowTaskDetectorTest.java:269 
(assertThat(slowTasks).hasSize(2)).
  - findSlowTasks reads System.currentTimeMillis() 
(ExecutionTimeBasedSlowTaskDetector.java:148) and marks a task slow only if its 
execution time is strictly greater than the baseline (median × multiplier).
  With createSlowTaskDetector(0.3, 1, 0) and equal input bytes, the two running 
tasks must out-age the one finished task.
  - The test relies on real elapsed time; when setup, markFinished(), and 
findSlowTasks() run within one millisecond tick, running tasks are not strictly 
greater than the baseline → 0 detected.
  - Observed: Java 11 / Test (core), commit 4902753. CI: 
https://github.com/apache/flink/actions/runs/27802684638/job/82280284780

Proposed fix
  - Set explicit execution-state timestamps so running tasks deterministically 
exceed the finished baseline, instead of relying on wall-clock granularity.



--
This message was sent by Atlassian Jira
(v8.20.10#820010)

Reply via email to