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-4348) Implement communication from ResourceManager to TaskManager
Date Tue, 23 Aug 2016 02:05:20 GMT

    [ https://issues.apache.org/jira/browse/FLINK-4348?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15431984#comment-15431984
] 

ASF GitHub Bot commented on FLINK-4348:
---------------------------------------

Github user beyond1920 commented on a diff in the pull request:

    https://github.com/apache/flink/pull/2389#discussion_r75792397
  
    --- Diff: flink-runtime/src/main/java/org/apache/flink/runtime/rpc/resourcemanager/ResourceManagerToTaskExecutorHeartbeatScheduler.java
---
    @@ -0,0 +1,242 @@
    +/*
    + * 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.rpc.resourcemanager;
    +
    +import akka.dispatch.OnFailure;
    +import akka.dispatch.OnSuccess;
    +import org.apache.flink.runtime.clusterframework.types.ResourceID;
    +import org.apache.flink.runtime.rpc.taskexecutor.SlotReport;
    +import org.apache.flink.runtime.rpc.taskexecutor.TaskExecutorGateway;
    +import org.slf4j.Logger;
    +import scala.concurrent.Future;
    +import scala.concurrent.duration.FiniteDuration;
    +
    +import java.util.UUID;
    +import java.util.concurrent.TimeUnit;
    +import java.util.concurrent.TimeoutException;
    +
    +import static org.apache.flink.util.Preconditions.checkState;
    +
    +/**
    + * heartbeat between ResourceManager and TaskManager, it is responsible for schedule
heartbeat and handle
    + * heartbeat lost cases
    + */
    +public class ResourceManagerToTaskExecutorHeartbeatScheduler {
    +
    +	/** default heartbeat interval time in millisecond */
    +	private static final long INITIAL_HEARTBEAT_INTERVAL_MILLIS = 5000;
    +
    +	/** default heartbeat timeout in millisecond */
    +	private static final long INITIAL_HEARTBEAT_TIMEOUT_MILLIS = 200;
    +
    +	/** max heartbeat interval time in millisecond (which is used in retry heartbeat case)
*/
    +	private static final long MAX_HEARTBEAT_TIMEOUT_MILLIS = 30000;
    +
    +	/** if a failure except for timeout exception happened when trigger heartbeat from resourceManager
to taskManager , next attemp will start after  ERROR_HEARTBEAT_DELAY_MILLIS millisecond */
    +	private static final long ERROR_HEARTBEAT_DELAY_MILLIS = 2000;
    +
    +	/** max heartbeat retry times when lost heartbeat */
    +	private static final int MAX_ATTEMPT_TIMES = 8;
    +
    +	private final long heartbeatInterval;
    +
    +	private final long heartbeatTimeout;
    +
    +	private final long maxHeartbeatTimeout;
    +
    +	private final long delayOnError;
    +
    +	private final int maxAttempt;
    +
    +
    +	/** taskManagerGateway to receive the heartbeat and report slot allocation */
    +	private final TaskExecutorGateway taskExecutorGateway;
    +
    +	/** the taskManager address */
    +	private final String taskExecutorAddress;
    +
    +	/** identify the taskManager resourceID */
    +	private final ResourceID resourceID;
    +
    +	/** identify the resourceManager rpc endpoint */
    +	private final ResourceManager resourceManager;
    +
    +	private final UUID resourceManagerLeaderSessionID;
    +
    +	private final Logger log;
    +
    +	private volatile boolean closed;
    +
    +	/**
    +	 * ResourceManagerToTaskExecutorHeartbeatScheduler constructor
    +	 *
    +	 * @param resourceManager         resourceManager which handles heartbeat communication
with taskManager
    +	 * @param taskExecutorGateway     taskManager which receives heartbeat from resourceManager
and report its slot
    +	 *                                allocation to resourceManager
    +	 * @param taskExecutorAddress     taskManager's address
    +	 * @param taskExecutorResourceID  taskManager's resourceID
    +	 * @param log
    +	 */
    +	public ResourceManagerToTaskExecutorHeartbeatScheduler(
    +		ResourceManager resourceManager, UUID resourceManagerLeaderSessionID, TaskExecutorGateway
taskExecutorGateway,
    +		String taskExecutorAddress, ResourceID taskExecutorResourceID, Logger log) {
    +		this(resourceManager, resourceManagerLeaderSessionID, taskExecutorGateway, taskExecutorAddress,
taskExecutorResourceID,
    +			log, INITIAL_HEARTBEAT_INTERVAL_MILLIS, INITIAL_HEARTBEAT_TIMEOUT_MILLIS, MAX_HEARTBEAT_TIMEOUT_MILLIS,
    +			ERROR_HEARTBEAT_DELAY_MILLIS, MAX_ATTEMPT_TIMES);
    +	}
    +
    +	/**
    +	 * ResourceManagerToTaskExecutorHeartbeatScheduler constructor
    +	 *
    +	 * @param resourceManager
    +	 * @param taskExecutorGateway
    +	 * @param taskExecutorAddress
    +	 * @param taskExecutorResourceID
    +	 * @param log
    +	 * @param heartbeatInterval
    +	 * @param heartbeatTimeout
    +	 * @param maxHeartbeatTimeout
    +	 * @param delayOnError
    +	 * @param maxAttempt
    +	 */
    +	public ResourceManagerToTaskExecutorHeartbeatScheduler(
    +		ResourceManager resourceManager, UUID resourceManagerLeaderSessionID, TaskExecutorGateway
taskExecutorGateway,
    +		String taskExecutorAddress, ResourceID taskExecutorResourceID, Logger log, long heartbeatInterval,
    +		long heartbeatTimeout, long maxHeartbeatTimeout, long delayOnError, int maxAttempt)
{
    +		this.resourceManager = resourceManager;
    +		this.resourceManagerLeaderSessionID = resourceManagerLeaderSessionID;
    +		this.taskExecutorGateway = taskExecutorGateway;
    +		this.taskExecutorAddress = taskExecutorAddress;
    +		this.resourceID = taskExecutorResourceID;
    +		this.log = log;
    +		this.heartbeatInterval = heartbeatInterval;
    +		this.heartbeatTimeout = heartbeatTimeout;
    +		this.maxHeartbeatTimeout = maxHeartbeatTimeout;
    +		this.delayOnError = delayOnError;
    +		this.maxAttempt = maxAttempt;
    +	}
    +
    +	/**
    +	 * start to schedule next heartbeat
    +	 */
    +	public void start() {
    +		checkState(!closed, "The heartbeat connection is already closed");
    +		sendHeartbeatToTaskManagerLater(1, heartbeatTimeout, heartbeatInterval);
    +	}
    +
    +	/**
    +	 * Checks if the heartbeat schedule was closed.
    +	 *
    +	 * @return True if the heartbeat schedule was closed, false otherwise.
    +	 */
    +	public boolean isClosed() {
    +		return closed;
    +	}
    +
    +	/**
    +	 * stop to schedule heartbeat
    +	 */
    +	public void close() {
    +		closed = true;
    +	}
    +
    +	/**
    +	 * get the heartbeat interval
    +	 *
    +	 * @return heartbeat interval
    +	 */
    +	public long getHeartbeatInterval() {
    +		return heartbeatInterval;
    +	}
    +
    +	/**
    +	 * send a heartbeat attempt to taskManager, receive the slotReport from TaskManager
or failed depends on the future result.
    +	 *
    +	 * @param attempt
    +	 * @param timeoutMillis
    +	 */
    +	private void sendHeartbeatToTaskManager(final int attempt, final long timeoutMillis)
{
    +		// eager check for closed to avoid some unnecessary work
    +		if (closed) {
    +			return;
    +		}
    +		FiniteDuration timeout = new FiniteDuration(timeoutMillis, TimeUnit.MILLISECONDS);
    +		Future<SlotReport> heartbeatResponse = taskExecutorGateway.triggerHeartbeatToResourceManager(resourceManagerLeaderSessionID,
timeout);
    +
    +		heartbeatResponse.onSuccess(new OnSuccess<SlotReport>() {
    +
    +			@Override
    +			public void onSuccess(SlotReport result) throws Throwable {
    +				if (!isClosed()) {
    +					// notify the slotManager and trigger next time heartbeat
    +					resourceManager.handleSlotReportFromTaskManager(result);
    +					sendHeartbeatToTaskManagerLater(1, heartbeatTimeout, heartbeatInterval);
    +				}
    +			}
    +		}, resourceManager.getRpcService().getExecutionContext());
    +
    +		// upon failure, retry
    +		heartbeatResponse.onFailure(new OnFailure() {
    +			@Override
    +			public void onFailure(Throwable failure) {
    +				if (!isClosed()) {
    +					if (attempt == maxAttempt) {
    +						log.error("ResourceManager {} fail to keep heartbeat with taskManager {} after
{} attempts",
    +							resourceManager.getAddress(), taskExecutorAddress, attempt);
    +						closed = true;
    +						// mark TaskManager as failed after heartbeat interaction attempts failed for many
times
    +						resourceManager.getSelf().notifyResourceFailure(resourceID);
    +					} else {
    +						// upon timeout exception, exponential increase timeout
    +						if (failure instanceof TimeoutException) {
    +							if (log.isDebugEnabled()) {
    +								log.debug("ResourceManager {} lost heartbeat in {} ms with taskManager {} at
{} attempts ",
    +									resourceManager.getAddress(), timeoutMillis, attempt);
    +							}
    +
    +							long retryTimeoutMillis = Math.min(2 * timeoutMillis, maxHeartbeatTimeout);
    +							sendHeartbeatToTaskManager(attempt + 1, retryTimeoutMillis);
    +						} else {
    +							log.error(
    +								"ResourceManager {} fail to keep heartbeat with taskManager due to an error",
    +								failure);
    +						}
    +						sendHeartbeatToTaskManagerLater(attempt + 1, timeoutMillis, delayOnError);
    --- End diff --
    
    modified


> Implement communication from ResourceManager to TaskManager
> -----------------------------------------------------------
>
>                 Key: FLINK-4348
>                 URL: https://issues.apache.org/jira/browse/FLINK-4348
>             Project: Flink
>          Issue Type: Sub-task
>          Components: Cluster Management
>            Reporter: Kurt Young
>            Assignee: zhangjing
>
> There are mainly 3 logics initiated from RM to TM:
> * Heartbeat, RM use heartbeat to sync with TM's slot status
> * SlotRequest, when RM decides to assign slot to JM, should first try to send request
to TM for slot. TM can either accept or reject this request.
> * FailureNotify, in some corner cases, TM will be marked as invalid by cluster manager
master(e.g. yarn master), but TM itself does not realize. RM should send failure notify to
TM and TM can terminate itself



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

Mime
View raw message