[HBASE-28330] TestUnknownServers.testListUnknownServers is flaky in branch-2

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 2.5.7
• Fix Version/s: 2.6.0, 2.5.8
• Component/s: test
• Assignee: Sun Xin
• Reporter: Sun Xin
• Created: 25/Jan/24 11:18
• Updated: 27/Jan/24 11:33
• Resolved: 26/Jan/24 07:00

Description

[ERROR] Tests run: 1, Failures: 1, Errors: 0, Skipped: 0, Time elapsed: 1.913 s <<< FAILURE! - in org.apache.hadoop.hbase.master.TestUnknownServers
[ERROR] org.apache.hadoop.hbase.master.TestUnknownServers.testListUnknownServers  Time elapsed: 0.204 s  <<< FAILURE!
java.lang.AssertionError: expected:<1> but was:<2> 

The value of TestUnknownServers.SLAVES is different between branch-2 and master.

It is 1 in master but 2 in branch-2.

The RegionServer marked UNKNOWN_SERVER is the one that holds regions but is not tracked by the ServerManager.

Please see HMaster.getUnknownServers

private List<ServerName> getUnknownServers() {
  if (serverManager != null) {
    final Set<ServerName> serverNames = getAssignmentManager().getRegionStates().getRegionStates()
      .stream().map(RegionState::getServerName).collect(Collectors.toSet());
    final List<ServerName> unknownServerNames = serverNames.stream()
      .filter(sn -> sn != null && serverManager.isServerUnknown(sn)).collect(Collectors.toList());
    return unknownServerNames;
  }
  return null;
} 

In UT TestUnknownServers.testListUnknownServers, we start a HBase cluster with 2 RegionServer, if all region are assigned to ONE server, then only that server is called UNKNOWN_SERVER, the UT will fail.

Issue Links

• links to: GitHub Pull Request #5650