flink-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "ASF GitHub Bot (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (FLINK-2790) Add high availability support for Yarn
Date Wed, 07 Oct 2015 10:32:27 GMT

    [ https://issues.apache.org/jira/browse/FLINK-2790?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14946642#comment-14946642
] 

ASF GitHub Bot commented on FLINK-2790:
---------------------------------------

Github user uce commented on a diff in the pull request:

    https://github.com/apache/flink/pull/1213#discussion_r41374493
  
    --- Diff: flink-runtime/src/test/scala/org/apache/flink/runtime/testingUtils/TestingJobManagerLike.scala
---
    @@ -0,0 +1,367 @@
    +/*
    + * Licensed to the Apache Software Foundation (ASF) under one
    + * or more contributor license agreements.  See the NOTICE file
    + * distributed with this work for additional information
    + * regarding copyright ownership.  The ASF licenses this file
    + * to you under the Apache License, Version 2.0 (the
    + * "License"); you may not use this file except in compliance
    + * with the License.  You may obtain a copy of the License at
    + *
    + *     http://www.apache.org/licenses/LICENSE-2.0
    + *
    + * Unless required by applicable law or agreed to in writing, software
    + * distributed under the License is distributed on an "AS IS" BASIS,
    + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
    + * See the License for the specific language governing permissions and
    + * limitations under the License.
    + */
    +
    +package org.apache.flink.runtime.testingUtils
    +
    +import akka.actor.{Terminated, Cancellable, ActorRef}
    +import akka.pattern.{ask, pipe}
    +import org.apache.flink.api.common.JobID
    +import org.apache.flink.runtime.FlinkActor
    +import org.apache.flink.runtime.execution.ExecutionState
    +import org.apache.flink.runtime.jobgraph.JobStatus
    +import org.apache.flink.runtime.jobmanager.JobManager
    +import org.apache.flink.runtime.messages.ExecutionGraphMessages.JobStatusChanged
    +import org.apache.flink.runtime.messages.JobManagerMessages.GrantLeadership
    +import org.apache.flink.runtime.messages.Messages.{Acknowledge, Disconnect}
    +import org.apache.flink.runtime.messages.RegistrationMessages.RegisterTaskManager
    +import org.apache.flink.runtime.messages.TaskManagerMessages.Heartbeat
    +import org.apache.flink.runtime.testingUtils.TestingJobManagerMessages._
    +import org.apache.flink.runtime.testingUtils.TestingMessages.{DisableDisconnect,
    +CheckIfJobRemoved, Alive}
    +import org.apache.flink.runtime.testingUtils.TestingTaskManagerMessages.AccumulatorsChanged
    +
    +import scala.collection.mutable
    +import scala.concurrent.Future
    +import scala.concurrent.duration._
    +
    +import language.postfixOps
    +
    +/** This mixin can be used to decorate a JobManager with messages for testing purpose.
    +  *
    +  */
    +trait TestingJobManagerLike extends FlinkActor {
    +  that: JobManager =>
    +
    +  import scala.collection.JavaConverters._
    +  import context._
    +
    +  val waitForAllVerticesToBeRunning = scala.collection.mutable.HashMap[JobID, Set[ActorRef]]()
    +  val waitForTaskManagerToBeTerminated = scala.collection.mutable.HashMap[String, Set[ActorRef]]()
    +
    +  val waitForAllVerticesToBeRunningOrFinished =
    +    scala.collection.mutable.HashMap[JobID, Set[ActorRef]]()
    +
    +  var periodicCheck: Option[Cancellable] = None
    +
    +  val waitForJobStatus = scala.collection.mutable.HashMap[JobID,
    +    collection.mutable.HashMap[JobStatus, Set[ActorRef]]]()
    +
    +  val waitForAccumulatorUpdate = scala.collection.mutable.HashMap[JobID, (Boolean, Set[ActorRef])]()
    +
    +  val waitForLeader = scala.collection.mutable.HashSet[ActorRef]()
    +
    +  val waitForNumRegisteredTaskManagers = mutable.PriorityQueue.newBuilder(
    +    new Ordering[(Int, ActorRef)] {
    +      override def compare(x: (Int, ActorRef), y: (Int, ActorRef)): Int = y._1 - x._1
    +    })
    +
    +  var disconnectDisabled = false
    +
    +  abstract override def handleMessage: Receive = {
    +    handleTestingMessage orElse super.handleMessage
    +  }
    +
    +  def handleTestingMessage: Receive = {
    +    case Alive => sender() ! Acknowledge
    +
    +    case RequestExecutionGraph(jobID) =>
    +      currentJobs.get(jobID) match {
    +        case Some((executionGraph, jobInfo)) => sender() ! decorateMessage(
    +          ExecutionGraphFound(
    +            jobID,
    +            executionGraph)
    +        )
    +
    +        case None => archive.tell(decorateMessage(RequestExecutionGraph(jobID)), sender())
    +      }
    +
    +    case WaitForAllVerticesToBeRunning(jobID) =>
    +      if(checkIfAllVerticesRunning(jobID)){
    +        sender() ! decorateMessage(AllVerticesRunning(jobID))
    +      }else{
    +        val waiting = waitForAllVerticesToBeRunning.getOrElse(jobID, Set[ActorRef]())
    +        waitForAllVerticesToBeRunning += jobID -> (waiting + sender())
    +
    +        if(periodicCheck.isEmpty){
    +          periodicCheck =
    +            Some(
    +              context.system.scheduler.schedule(
    +                0 seconds,
    +                200 millis,
    +                self,
    +                decorateMessage(NotifyListeners)
    +              )
    +            )
    +        }
    +      }
    +    case WaitForAllVerticesToBeRunningOrFinished(jobID) =>
    +      if(checkIfAllVerticesRunningOrFinished(jobID)){
    +        sender() ! decorateMessage(AllVerticesRunning(jobID))
    +      }else{
    +        val waiting = waitForAllVerticesToBeRunningOrFinished.getOrElse(jobID, Set[ActorRef]())
    +        waitForAllVerticesToBeRunningOrFinished += jobID -> (waiting + sender())
    +
    +        if(periodicCheck.isEmpty){
    +          periodicCheck =
    +            Some(
    +              context.system.scheduler.schedule(
    +                0 seconds,
    +                200 millis,
    +                self,
    +                decorateMessage(NotifyListeners)
    +              )
    +            )
    +        }
    +      }
    +
    +    case NotifyListeners =>
    +      for(jobID <- currentJobs.keySet){
    +        notifyListeners(jobID)
    +      }
    +
    +      if(waitForAllVerticesToBeRunning.isEmpty && waitForAllVerticesToBeRunningOrFinished.isEmpty)
{
    +        periodicCheck foreach { _.cancel() }
    +        periodicCheck = None
    +      }
    +
    +
    +    case NotifyWhenJobRemoved(jobID) =>
    +      val gateways = instanceManager.getAllRegisteredInstances.asScala.map(_.getActorGateway)
    +
    +      val responses = gateways.map{
    +        gateway => gateway.ask(NotifyWhenJobRemoved(jobID), timeout).mapTo[Boolean]
    +      }
    +
    +      val jobRemovedOnJobManager = (self ? CheckIfJobRemoved(jobID))(timeout).mapTo[Boolean]
    +
    +      val allFutures = responses ++ Seq(jobRemovedOnJobManager)
    +
    +      import context.dispatcher
    +      Future.fold(allFutures)(true)(_ & _) map(decorateMessage(_)) pipeTo sender()
    +
    +    case CheckIfJobRemoved(jobID) =>
    +      if(currentJobs.contains(jobID)) {
    +        context.system.scheduler.scheduleOnce(
    +          200 milliseconds,
    +          self,
    +          decorateMessage(CheckIfJobRemoved(jobID))
    +        )(context.dispatcher, sender())
    +      } else {
    +        sender() ! decorateMessage(true)
    +      }
    +
    +    case NotifyWhenTaskManagerTerminated(taskManager) =>
    +      val waiting = waitForTaskManagerToBeTerminated.getOrElse(taskManager.path.name,
Set())
    +      waitForTaskManagerToBeTerminated += taskManager.path.name -> (waiting + sender)
    +
    +    case msg@Terminated(taskManager) =>
    +      super.handleMessage(msg)
    +
    +      waitForTaskManagerToBeTerminated.remove(taskManager.path.name) foreach {
    +        _ foreach {
    +          listener =>
    +            listener ! decorateMessage(TaskManagerTerminated(taskManager))
    +        }
    +      }
    +
    +    case NotifyWhenAccumulatorChange(jobID) =>
    +
    +      val (updated, registered) = waitForAccumulatorUpdate.
    +        getOrElse(jobID, (false, Set[ActorRef]()))
    +      waitForAccumulatorUpdate += jobID -> (updated, registered + sender)
    +      sender ! true
    +
    +    /**
    +     * Notification from the task manager that changed accumulator are transferred on
next
    +     * Hearbeat. We need to keep this state to notify the listeners on next Heartbeat
report.
    +     */
    +    case AccumulatorsChanged(jobID: JobID) =>
    +      waitForAccumulatorUpdate.get(jobID) match {
    +        case Some((updated, registered)) =>
    +          waitForAccumulatorUpdate.put(jobID, (true, registered))
    +        case None =>
    +      }
    +
    +    /**
    +     * Disabled async processing of accumulator values and send accumulators to the listeners
if
    +     * we previously received an [[AccumulatorsChanged]] message.
    +     */
    +    case msg : Heartbeat =>
    +      super.handleMessage(msg)
    +
    +      waitForAccumulatorUpdate foreach {
    +        case (jobID, (updated, actors)) if updated =>
    +          currentJobs.get(jobID) match {
    +            case Some((graph, jobInfo)) =>
    +              val flinkAccumulators = graph.getFlinkAccumulators
    +              val userAccumulators = graph.aggregateUserAccumulators
    +              actors foreach {
    +                actor => actor ! UpdatedAccumulators(jobID, flinkAccumulators, userAccumulators)
    +              }
    +            case None =>
    +          }
    +          waitForAccumulatorUpdate.put(jobID, (false, actors))
    +        case _ =>
    +      }
    +
    +    case RequestWorkingTaskManager(jobID) =>
    +      currentJobs.get(jobID) match {
    +        case Some((eg, _)) =>
    +          if(eg.getAllExecutionVertices.asScala.isEmpty){
    +            sender ! decorateMessage(WorkingTaskManager(None))
    +          } else {
    +            val resource = eg.getAllExecutionVertices.asScala.head.getCurrentAssignedResource
    +
    +            if(resource == null){
    +              sender ! decorateMessage(WorkingTaskManager(None))
    +            } else {
    +              sender ! decorateMessage(
    +                WorkingTaskManager(
    +                  Some(resource.getInstance().getActorGateway)
    +                )
    +              )
    +            }
    +          }
    +        case None => sender ! decorateMessage(WorkingTaskManager(None))
    +      }
    +
    --- End diff --
    
    empty line


> Add high availability support for Yarn
> --------------------------------------
>
>                 Key: FLINK-2790
>                 URL: https://issues.apache.org/jira/browse/FLINK-2790
>             Project: Flink
>          Issue Type: Sub-task
>          Components: JobManager, TaskManager
>            Reporter: Till Rohrmann
>             Fix For: 0.10
>
>
> Add master high availability support for Yarn. The idea is to let Yarn restart a failed
application master in a new container. For that, we set the number of application retries
to something greater than 1. 
> From version 2.4.0 onwards, it is possible to reuse already started containers for the
TaskManagers, thus, avoiding unnecessary restart delays.
> From version 2.6.0 onwards, it is possible to specify an interval in which the number
of application attempts have to be exceeded in order to fail the job. This will prevent long
running jobs from eventually depleting all available application attempts.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Mime
View raw message