[SPARK-12552][CORE] Correctly count the driver resource when recovering from failure for Master

Currently in Standalone HA mode, the resource usage of driver is not correctly counted in Master when recovering from failure, this will lead to some unexpected behaviors like negative value in UI.

So here fix this to also count the driver's resource usage.

Also changing the recovered app's state to `RUNNING` when fully recovered. Previously it will always be WAITING even fully recovered.

andrewor14 please help to review, thanks a lot.

Author: jerryshao <[email protected]>

Closes #10506 from jerryshao/SPARK-12552.

Author: jerryshao

core/src/main/scala/org/apache/spark/deploy/master/Master.scala

@@ -367,7 +367,7 @@ private[deploy] class Master(
             drivers.find(_.id == driverId).foreach { driver =>
               driver.worker = Some(worker)
               driver.state = DriverState.RUNNING
-              worker.drivers(driverId) = driver
+              worker.addDriver(driver)
             }
           }
         case None =>
@@ -547,6 +547,9 @@ private[deploy] class Master(
     workers.filter(_.state == WorkerState.UNKNOWN).foreach(removeWorker)
     apps.filter(_.state == ApplicationState.UNKNOWN).foreach(finishApplication)
 
+    // Update the state of recovered apps to RUNNING
+    apps.filter(_.state == ApplicationState.WAITING).foreach(_.state = ApplicationState.RUNNING)
+
     // Reschedule drivers which were not claimed by any workers
     drivers.filter(_.worker.isEmpty).foreach { d =>
       logWarning(s"Driver ${d.id} was not found after master recovery")

core/src/test/scala/org/apache/spark/deploy/master/MasterSuite.scala

@@ -21,20 +21,24 @@ import java.util.Date
 import java.util.concurrent.ConcurrentLinkedQueue
 
 import scala.collection.JavaConverters._
+import scala.collection.mutable.{HashMap, HashSet}
 import scala.concurrent.duration._
 import scala.io.Source
 import scala.language.postfixOps
+import scala.reflect.ClassTag
 
 import org.json4s._
 import org.json4s.jackson.JsonMethods._
+import org.mockito.Mockito.{mock, when}
 import org.scalatest.{BeforeAndAfter, Matchers, PrivateMethodTester}
 import org.scalatest.concurrent.Eventually
 import other.supplier.{CustomPersistenceEngine, CustomRecoveryModeFactory}
 
 import org.apache.spark.{SecurityManager, SparkConf, SparkFunSuite}
 import org.apache.spark.deploy._
 import org.apache.spark.deploy.DeployMessages._
-import org.apache.spark.rpc.{RpcAddress, RpcEndpoint, RpcEnv}
+import org.apache.spark.rpc.{RpcAddress, RpcEndpoint, RpcEndpointRef, RpcEnv}
+import org.apache.spark.serializer
 
 class MasterSuite extends SparkFunSuite
   with Matchers with Eventually with PrivateMethodTester with BeforeAndAfter {
@@ -134,6 +138,81 @@ class MasterSuite extends SparkFunSuite
     CustomRecoveryModeFactory.instantiationAttempts should be > instantiationAttempts
   }
 
+  test("master correctly recover the application") {
+    val conf = new SparkConf(loadDefaults = false)
+    conf.set("spark.deploy.recoveryMode", "CUSTOM")
+    conf.set("spark.deploy.recoveryMode.factory",
+      classOf[FakeRecoveryModeFactory].getCanonicalName)
+    conf.set("spark.master.rest.enabled", "false")
+
+    val fakeAppInfo = makeAppInfo(1024)
+    val fakeWorkerInfo = makeWorkerInfo(8192, 16)
+    val fakeDriverInfo = new DriverInfo(
+      startTime = 0,
+      id = "test_driver",
+      desc = new DriverDescription(
+        jarUrl = "",
+        mem = 1024,
+        cores = 1,
+        supervise = false,
+        command = new Command("", Nil, Map.empty, Nil, Nil, Nil)),
+      submitDate = new Date())
+
+    // Build the fake recovery data
+    FakeRecoveryModeFactory.persistentData.put(s"app_${fakeAppInfo.id}", fakeAppInfo)
+    FakeRecoveryModeFactory.persistentData.put(s"driver_${fakeDriverInfo.id}", fakeDriverInfo)
+    FakeRecoveryModeFactory.persistentData.put(s"worker_${fakeWorkerInfo.id}", fakeWorkerInfo)
+
+    var master: Master = null
+    try {
+      master = makeMaster(conf)
+      master.rpcEnv.setupEndpoint(Master.ENDPOINT_NAME, master)
+      // Wait until Master recover from checkpoint data.
+      eventually(timeout(5 seconds), interval(100 milliseconds)) {
+        master.idToApp.size should be(1)
+      }
+
+      master.idToApp.keySet should be(Set(fakeAppInfo.id))
+      getDrivers(master) should be(Set(fakeDriverInfo))
+      master.workers should be(Set(fakeWorkerInfo))
+
+      // Notify Master about the executor and driver info to make it correctly recovered.
+      val fakeExecutors = List(
+        new ExecutorDescription(fakeAppInfo.id, 0, 8, ExecutorState.RUNNING),
+        new ExecutorDescription(fakeAppInfo.id, 0, 7, ExecutorState.RUNNING))
+
+      fakeAppInfo.state should be(ApplicationState.UNKNOWN)
+      fakeWorkerInfo.coresFree should be(16)
+      fakeWorkerInfo.coresUsed should be(0)
+
+      master.self.send(MasterChangeAcknowledged(fakeAppInfo.id))
+      eventually(timeout(1 second), interval(10 milliseconds)) {
+        // Application state should be WAITING when "MasterChangeAcknowledged" event executed.
+        fakeAppInfo.state should be(ApplicationState.WAITING)
+      }
+
+      master.self.send(
+        WorkerSchedulerStateResponse(fakeWorkerInfo.id, fakeExecutors, Seq(fakeDriverInfo.id)))
+
+      eventually(timeout(5 seconds), interval(100 milliseconds)) {
+        getState(master) should be(RecoveryState.ALIVE)
+      }
+
+      // If driver's resource is also counted, free cores should 0
+      fakeWorkerInfo.coresFree should be(0)
+      fakeWorkerInfo.coresUsed should be(16)
+      // State of application should be RUNNING
+      fakeAppInfo.state should be(ApplicationState.RUNNING)
+    } finally {
+      if (master != null) {
+        master.rpcEnv.shutdown()
+        master.rpcEnv.awaitTermination()
+        master = null
+        FakeRecoveryModeFactory.persistentData.clear()
+      }
+    }
+  }
+
   test("master/worker web ui available") {
     implicit val formats = org.json4s.DefaultFormats
     val conf = new SparkConf()
@@ -394,6 +473,9 @@ class MasterSuite extends SparkFunSuite
   // ==========================================
 
   private val _scheduleExecutorsOnWorkers = PrivateMethod[Array[Int]]('scheduleExecutorsOnWorkers)
+  private val _drivers = PrivateMethod[HashSet[DriverInfo]]('drivers)
+  private val _state = PrivateMethod[RecoveryState.Value]('state)
+
   private val workerInfo = makeWorkerInfo(4096, 10)
   private val workerInfos = Array(workerInfo, workerInfo, workerInfo)
 
@@ -412,12 +494,18 @@ class MasterSuite extends SparkFunSuite
     val desc = new ApplicationDescription(
       "test", maxCores, memoryPerExecutorMb, null, "", None, None, coresPerExecutor)
     val appId = System.currentTimeMillis.toString
-    new ApplicationInfo(0, appId, desc, new Date, null, Int.MaxValue)
+    val endpointRef = mock(classOf[RpcEndpointRef])
+    val mockAddress = mock(classOf[RpcAddress])
+    when(endpointRef.address).thenReturn(mockAddress)
+    new ApplicationInfo(0, appId, desc, new Date, endpointRef, Int.MaxValue)
   }
 
   private def makeWorkerInfo(memoryMb: Int, cores: Int): WorkerInfo = {
     val workerId = System.currentTimeMillis.toString
-    new WorkerInfo(workerId, "host", 100, cores, memoryMb, null, "http://localhost:80")
+    val endpointRef = mock(classOf[RpcEndpointRef])
+    val mockAddress = mock(classOf[RpcAddress])
+    when(endpointRef.address).thenReturn(mockAddress)
+    new WorkerInfo(workerId, "host", 100, cores, memoryMb, endpointRef, "http://localhost:80")
   }
 
   private def scheduleExecutorsOnWorkers(
@@ -499,4 +587,40 @@ class MasterSuite extends SparkFunSuite
       assert(receivedMasterAddress === RpcAddress("localhost2", 10000))
     }
   }
+
+  private def getDrivers(master: Master): HashSet[DriverInfo] = {
+    master.invokePrivate(_drivers())
+  }
+
+  private def getState(master: Master): RecoveryState.Value = {
+    master.invokePrivate(_state())
+  }
+}
+
+private class FakeRecoveryModeFactory(conf: SparkConf, ser: serializer.Serializer)
+    extends StandaloneRecoveryModeFactory(conf, ser) {
+  import FakeRecoveryModeFactory.persistentData
+
+  override def createPersistenceEngine(): PersistenceEngine = new PersistenceEngine {
+
+    override def unpersist(name: String): Unit = {
+      persistentData.remove(name)
+    }
+
+    override def persist(name: String, obj: Object): Unit = {
+      persistentData(name) = obj
+    }
+
+    override def read[T: ClassTag](prefix: String): Seq[T] = {
+      persistentData.filter(_._1.startsWith(prefix)).map(_._2.asInstanceOf[T]).toSeq
+    }
+  }
+
+  override def createLeaderElectionAgent(master: LeaderElectable): LeaderElectionAgent = {
+    new MonarchyLeaderAgent(master)
+  }
+}
+
+private object FakeRecoveryModeFactory {
+  val persistentData = new HashMap[String, Object]()
 }