MAPREDUCE-6831. Flaky test TestJobImpl.testKilledDuringKillAbort. Contributed by Peter Bacsko

hadoop-mapreduce-project/hadoop-mapreduce-client/hadoop-mapreduce-client-app/src/test/java/org/apache/hadoop/mapreduce/v2/app/job/impl/TestJobImpl.java

@@ -102,7 +102,6 @@ import org.apache.hadoop.yarn.event.InlineDispatcher;
 import org.apache.hadoop.yarn.exceptions.YarnRuntimeException;
 import org.apache.hadoop.yarn.state.StateMachine;
 import org.apache.hadoop.yarn.state.StateMachineFactory;
-import org.apache.hadoop.yarn.util.ConverterUtils;
 import org.apache.hadoop.yarn.util.Records;
 import org.apache.hadoop.yarn.util.SystemClock;
 import org.junit.Assert;
@@ -498,9 +497,10 @@ public class TestJobImpl {
   public void testKilledDuringKillAbort() throws Exception {
     Configuration conf = new Configuration();
     conf.set(MRJobConfig.MR_AM_STAGING_DIR, stagingDir);
+    // not initializing dispatcher to avoid potential race condition between
+    // the dispatcher thread & test thread - see MAPREDUCE-6831
     AsyncDispatcher dispatcher = new AsyncDispatcher();
-    dispatcher.init(conf);
-    dispatcher.start();
+
     OutputCommitter committer = new StubbedOutputCommitter() {
       @Override
       public synchronized void abortJob(JobContext jobContext, State state)