[HDFS-4941] TestNNThroughputBenchmark fails because of interrupted ReplicationMonitor

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Duplicate
• Affects Version/s: 3.0.0-alpha1
• Component/s: namenode, test
• Assignee: Unassigned
• Reporter: Andrew Wang
• Created: 27/Jun/13 23:44
• Updated: 12/May/16 18:13
• Resolved: 28/Jun/13 14:46

Description

This has been breaking trunk test runs pretty consistently.

HDFS-4840 introduced some new code to BlockManager that terminates the NN if the ReplicationMonitor thread is interrupted unexpectedly:

          if (!namesystem.isRunning()) {
            LOG.info("Stopping ReplicationMonitor.");
            if (!(t instanceof InterruptedException)) {
              LOG.info("ReplicationMonitor received an exception"
                  + " while shutting down.", t);
            }
            break;
          }
          LOG.fatal("ReplicationMonitor thread received Runtime exception. ", t);
          terminate(1, t);

Unfortunately, NNThroughputBenchmark interrupts in ReplicationStats#generateInputs:

      // stop replication monitor
      BlockManagerTestUtil.getReplicationThread(namesystem.getBlockManager())
          .interrupt();

Issue Links

• is broken by: HDFS-4840 ReplicationMonitor gets NPE during shutdown
• is duplicated by: HDFS-4887 TestNNThroughputBenchmark exits abruptly