[HDFS-15147] LazyPersistTestCase wait logic is error-prone

• Type: Bug
• Status: Resolved
• Priority: Minor
• Resolution: Fixed
• Fix Version/s: 3.3.0, 3.1.4, 3.2.2, 2.10.1
• Hadoop Flags: Reviewed
• Assignee: Ahmed Hussein
• Reporter: Ahmed Hussein
• Created: 27/Jan/20 23:54
• Updated: 27/Feb/20 20:51
• Resolved: 27/Feb/20 16:00

Description

LazyPersistTestCase has some issues hat lead to inconsistent result of the test cases:

• the wait periods to change of status is too long. It reaches 10 secs in some cases.
• triggerBlockReport() only triggers FBR of DN with index 0. This is counter intuitive because the JUnit tests restart the DN assuming that the restarted DN will send a FBR. However, this never happens because the DN will get a new index post restart.

  protected final void triggerBlockReport()
      throws IOException, InterruptedException {
    // Trigger block report to NN
    DataNodeTestUtils.triggerBlockReport(cluster.getDataNodes().get(0));
    Thread.sleep(10 * 1000);
  }

inigoiri suggested that we propagate the findings and fixes from HDFS-13179 and HDFS-15144 into LazyPersistTestCase.java. This will eventually reduce the runtime and make the test cases more stable.