HDFS-6533. TestBPOfferService#testBasicFunctionalitytest fails intermittently. (Contributed by Wei-Chiu Chuang)

hadoop-hdfs-project/hadoop-hdfs/CHANGES.txt

@@ -1550,6 +1550,9 @@ Release 2.8.0 - UNRELEASED
     HDFS-9336. deleteSnapshot throws NPE when snapshotname is null.
     (Brahma Reddy Battula via aajisaka)
 
+    HDFS-6533. TestBPOfferService#testBasicFunctionalitytest fails
+    intermittently. (Wei-Chiu Chuang via Arpit Agarwal)
+
 Release 2.7.3 - UNRELEASED
 
   INCOMPATIBLE CHANGES

hadoop-hdfs-project/hadoop-hdfs/src/main/java/org/apache/hadoop/hdfs/server/datanode/BPServiceActor.java

@@ -127,6 +127,10 @@ class BPServiceActor implements Runnable {
     scheduler = new Scheduler(dnConf.heartBeatInterval, dnConf.blockReportInterval);
   }
 
+  public DatanodeRegistration getBpRegistration() {
+    return bpRegistration;
+  }
+
   boolean isAlive() {
     if (!shouldServiceRun || !bpThread.isAlive()) {
       return false;

hadoop-hdfs-project/hadoop-hdfs/src/test/java/org/apache/hadoop/hdfs/server/datanode/TestBPOfferService.java

@@ -180,7 +180,7 @@ public class TestBPOfferService {
     BPOfferService bpos = setupBPOSForNNs(mockNN1, mockNN2);
     bpos.start();
     try {
-      waitForInitialization(bpos);
+      waitForBothActors(bpos);
       
       // The DN should have register to both NNs.
       Mockito.verify(mockNN1).registerDatanode(
@@ -205,6 +205,7 @@ public class TestBPOfferService {
 
     } finally {
       bpos.stop();
+      bpos.join();
     }
   }
 
@@ -235,6 +236,7 @@ public class TestBPOfferService {
 
     } finally {
       bpos.stop();
+      bpos.join();
     }
     
     // Should ignore the delete command from the standby
@@ -260,6 +262,7 @@ public class TestBPOfferService {
       waitForOneToFail(bpos);
     } finally {
       bpos.stop();
+      bpos.join();
     }
   }
   
@@ -307,6 +310,7 @@ public class TestBPOfferService {
 
     } finally {
       bpos.stop();
+      bpos.join();
     }
   }
 
@@ -349,6 +353,7 @@ public class TestBPOfferService {
       waitForBlockReport(mockNN1, mockNN2);
     } finally {
       bpos.stop();
+      bpos.join();
     }
   }
 
@@ -404,6 +409,27 @@ public class TestBPOfferService {
     }, 100, 10000);
   }
 
+  private void waitForBothActors(final BPOfferService bpos)
+      throws Exception {
+    GenericTestUtils.waitFor(new Supplier<Boolean>() {
+      @Override
+      public Boolean get() {
+        List<BPServiceActor> actors = bpos.getBPServiceActors();
+
+        return bpos.isAlive() && getRegisteredActors(actors) == 2;
+      }
+      private int getRegisteredActors(List<BPServiceActor> actors) {
+        int regActors = 0;
+        for (BPServiceActor actor : actors) {
+          if (actor.getBpRegistration() != null) {
+            regActors++;
+          }
+        }
+        return regActors;
+      }
+    }, 100, 10000);
+  }
+  
   private void waitForBlockReport(final DatanodeProtocolClientSideTranslatorPB mockNN)
       throws Exception {
     GenericTestUtils.waitFor(new Supplier<Boolean>() {
@@ -540,6 +566,7 @@ public class TestBPOfferService {
           difference < 5000);
     } finally {
       bpos.stop();
+      bpos.join();
     }
   }
 
@@ -579,6 +606,7 @@ public class TestBPOfferService {
           + " processing ", difference < 5000);
     } finally {
       bpos.stop();
+      bpos.join();
     }
   }
   /**
@@ -624,6 +652,7 @@ public class TestBPOfferService {
           + "when errorReport threw IOException", secondCallTime != 0);
     } finally {
       bpos.stop();
+      bpos.join();
     }
   } 
 
@@ -675,6 +704,7 @@ public class TestBPOfferService {
           .reportBadBlocks(Mockito.any(LocatedBlock[].class));
     } finally {
       bpos.stop();
+      bpos.join();
     }
   }
 }