Fix shutdown bug.

CommonCode/SchedulerHelper.cs

@@ -459,7 +459,7 @@ namespace Microsoft.Research.Dryad
 
         void ISchedulerHelper.FinishJob()
         {
-            m_appMaster.Finish();
+            
         }
 
         string ISchedulerHelper.GetVertexServiceBaseAddress(string nodename, int instanceId)
@@ -535,6 +535,8 @@ namespace Microsoft.Research.Dryad
                 }
             }
 
+            m_appMaster.Finish(true);
+
             if (wait)
             {
                 try
@@ -619,6 +621,7 @@ namespace Microsoft.Research.Dryad
             DryadLogger.LogInformation("QueueYarnUpdate", "Task {0} on node {2} is in state {3}", taskId, nodeName,
                                        taskState);
             // Set change event arguments
+            
             YarnTaskState yTaskState = (YarnTaskState)taskState;
             VertexTask v = new VertexTask(taskId, nodeName, yTaskState, int.MaxValue, DateTime.UtcNow);
             m_taskUpdateQueue.Add(v);

DryadYarnBridge/YarnAppMasterManaged.cpp

@@ -62,12 +62,12 @@ namespace Microsoft { namespace Research { namespace Dryad { namespace YarnBridg
         }
     }
 
-    void AMInstance::Finish()
+    void AMInstance::Finish(bool success)
     {
         if (m_instance != IntPtr::Zero)
         {
             AMNativeInstance *instance =  (AMNativeInstance *) m_instance.ToPointer();
-            instance->Shutdown();
+            instance->Shutdown(success);
         }
     }
 

DryadYarnBridge/YarnAppMasterManaged.h

@@ -37,7 +37,7 @@ namespace Microsoft { namespace Research { namespace Dryad { namespace YarnBridg
 
 
         void Close();
-        void Finish();
+        void Finish(bool success);
 
         int GetHealthyNodeCount();
 

DryadYarnBridge/YarnAppMasterNative.cpp

@@ -246,8 +246,8 @@ namespace DryadYarn
             return false;
         }
 
-		jmethodID midShutdown = m_env->e->GetMethodID(clsDryadAppMaster, "shutdown", "()V");
-        if (midSchProc == NULL)
+		jmethodID midShutdown = m_env->e->GetMethodID(clsDryadAppMaster, "shutdown", "(ZZ)V");
+        if (midShutdown == NULL)
         {
             jthrowable exc;
             exc = m_env->e->ExceptionOccurred();
@@ -288,7 +288,6 @@ namespace DryadYarn
 
         jstring jName = env->NewStringUTF(name);
         jstring jCmdLine = env->NewStringUTF(commandLine);
-
         env->CallVoidMethod(m_inst->m_obj, m_inst->m_midSchProc, vertexId, jName, jCmdLine);
 
         env->DeleteLocalRef(jName);
@@ -299,16 +298,25 @@ namespace DryadYarn
         return true;
     }
 
-	bool AMNativeInstance::Shutdown()
+	bool AMNativeInstance::Shutdown(bool success)
     {
         fprintf(stderr, "Shutting down AMNativeInstance\n");
         fflush(stderr);
         JNIEnv* env = AttachToJvm(); 
+		fprintf(stderr, "Calling Shutdown\n");
+        fflush(stderr);
 		
-        env->CallVoidMethod(m_inst->m_obj, m_inst->m_midShutdown);
+		jboolean jImmedShutdown = 0;
+		jboolean jSuccess = 0;
+		if (success)
+		{
+			jSuccess = 1;
+		}
+        env->CallVoidMethod(m_inst->m_obj, m_inst->m_midShutdown, jImmedShutdown, jSuccess);
 
         // detach here?
-
+		fprintf(stderr, "Finished Shutdown\n");
+        fflush(stderr);
         return true;
     }
 

DryadYarnBridge/YarnAppMasterNative.h

@@ -70,7 +70,7 @@ namespace DryadYarn
 	    char* GetExceptionMessage();
 	    
 	    bool ScheduleProcess(int vertexId, const char* name, const char* commandLine);
-		bool Shutdown();
+		bool Shutdown(bool success);
 
     private:
 	    //void* operator new( size_t );

GraphManager/filesystem/DrHdfsClient.cpp

@@ -339,11 +339,11 @@ DrString DrHdfsOutputStream::GetURIForWrite(int partitionIndex,
                                             DrMetaDataRef /*metaData */)
 {
     DrAssert(m_hdfsInstance != DrNull);
-    DrString fileName;
-    fileName.Set(m_baseUri);
-    //String^ fileName = m_baseUri + "-tmp/part-" + partitionIndex.ToString("D8") + "." + version;
-    fileName.AppendF("-tmp/part-%8d.%d", partitionIndex, version);
-    return fileName;
+    String^ fileName = m_baseUri + "-tmp/part-" + partitionIndex.ToString("D8") + "." + version;
+	
+	//DrLogI("HDFS GetURIForWrite returning '%s'", fileName);  // DCF HDFS debug
+
+    return DrString(fileName);
 }
 
 void DrHdfsOutputStream::DiscardUnusedPartition(int partitionIndex, 

Java/DryadAppMaster.java

@@ -134,7 +134,7 @@ public class DryadAppMaster
     {
 	log = LogFactory.getLog("DryadAppMaster");
 	log.info("In DryadAppMaster constructor");
-	shuttingDown = new AtomicBoolean();
+	shuttingDown = new AtomicBoolean(false);
 	scheduleProcesses = new AtomicBoolean(true);
 	responseId = new AtomicInteger();
 	nextVertexId = new AtomicInteger(2); //first vertex id is 2 to map to Dryad Vertex Scheduler
@@ -233,19 +233,20 @@ public class DryadAppMaster
     {
 	// check to see if we should cancel the heartbeat
 	if (shuttingDown.get()) {
+	    log.info("Cancelling heartbeat"); 
 	    heartbeatHandle.cancel(true);
-	}
-	log.info("Sending heartbeat to the RM"); 
-	AllocateResponse response = sendAllocateRequest();
-	if (response != null) {
-	    int oldNodeCount = clusterNodeCount;
-	    clusterNodeCount = response.getNumClusterNodes();
-	    if (clusterNodeCount != oldNodeCount) {
-		log.info("There are now " + clusterNodeCount + " available nodes on the cluster.");
+	} else {
+	    log.info("Sending heartbeat to the RM"); 
+	    AllocateResponse response = sendAllocateRequest();
+	    if (response != null) {
+		int oldNodeCount = clusterNodeCount;
+		clusterNodeCount = response.getNumClusterNodes();
+		if (clusterNodeCount != oldNodeCount) {
+		    log.info("There are now " + clusterNodeCount + " available nodes on the cluster.");
+		}
+		processResponse(response);
 	    }
-	    processResponse(response);
 	}
-	
     }
     
     private void launchContainer(Container container, ContainerManager cm)
@@ -317,7 +318,9 @@ public class DryadAppMaster
     {
 	// is this the first allocation?	
 	if (scheduleProcesses.compareAndSet(true, false)) {
-	    int numProcessesToStart = Math.max(response.getNumClusterNodes() - 1, maxNodes); //don't schedule a process where the graph manager is running
+	    //don't schedule a process where the graph manager is running
+	    int numProcessesToStart = Math.max(response.getNumClusterNodes() - 1, maxNodes);
+	    log.info("There are " + response.getNumClusterNodes() + " nodes in the cluster. maxNodes = " + maxNodes);
 	    scheduleProcess(numProcessesToStart);
 	}
 	
@@ -339,7 +342,8 @@ public class DryadAppMaster
 	    
 	    // Need to notify graph manager of current state
 	    VertexInfo vi = runningContainers.remove(cid);
-	    if (vi != null) {
+	    //only send events up the stack when we are not shutting down
+	    if (vi != null && !shuttingDown.get()) {  
 		int containerState = 0;
 		if (containerStatus.getState() == ContainerState.COMPLETE) {
 		    if (containerStatus.getExitStatus() == 0) {
@@ -426,7 +430,7 @@ public class DryadAppMaster
 	}
     }
 
-    public void shutdown(boolean immediateShutdown)
+    public void shutdown(boolean immediateShutdown, boolean success)
     { 
 	shuttingDown.set(true);
 	heartbeatHandle.cancel(immediateShutdown); // if we are shutting down, we can just interrupt the running thread, if necessary
@@ -435,13 +439,18 @@ public class DryadAppMaster
 	// send the shutdown message to the RM
 	FinishApplicationMasterRequest request = Records.newRecord(FinishApplicationMasterRequest.class);
 	request.setAppAttemptId(appAttemptID);
-	request.setFinishApplicationStatus(FinalApplicationStatus.SUCCEEDED);  // NYI - determine success
+	if (success) {
+	    request.setFinishApplicationStatus(FinalApplicationStatus.SUCCEEDED);  
+	} else {
+	    request.setFinishApplicationStatus(FinalApplicationStatus.FAILED);
+	}
 	try {
 	    //response is currently an empty class
 	    FinishApplicationMasterResponse response = resourceManager. finishApplicationMaster(request);  
 	} catch (YarnRemoteException|IOException e) {
 	    log.error("Error communicating with RM: " + e.getMessage() , e);
 	}
+	log.info("FinishApplicationMasterRequest sent");
     }
 
     private void startContainers(List<Container> newContainers)