您现在访问的是微软AZURE全睃版技术文档网站,若需覝访问由世纪互蝔违蝥的MICROSOFT AZURE中国区技术文档网站,请访问 https://docs.azure.cn.

缩放 Azure HDInsight 群集Scale Azure HDInsight clusters

HDInsight 提供弹性,可让你选择纵向扩展和纵向缩减群集中的工作器节点数。HDInsight provides elasticity with options to scale up and scale down the number of worker nodes in your clusters. 这种弹性允许你在若干小时后或者在周末收缩群集。This elasticity allows you to shrink a cluster after hours or on weekends. 还可以在业务需求高峰期扩展群集。And expand it during peak business demands.

在定期批处理之前纵向扩展群集,使群集拥有足够的资源。Scale up your cluster before periodic batch processing so the cluster has adequate resources.  在完成处理并且用量下降后,可将 HDInsight 群集纵向缩减为使用更少的工作器节点。After processing completes, and usage goes down, scale down the HDInsight cluster to fewer worker nodes.

可以使用下述方法之一手动缩放群集。You can scale a cluster manually using one of the methods outlined below. 也可使用自动缩放选项来自动纵向缩放以响应某些指标。You can also use autoscale options to automatically scale up and down in response to certain metrics.


只支持使用 HDInsight 3.1.3 或更高版本的群集。Only clusters with HDInsight version 3.1.3 or higher are supported. 如果不确定群集的版本,可以查看“属性”页面。If you are unsure of the version of your cluster, you can check the Properties page.

用来缩放群集的实用程序Utilities to scale clusters

Microsoft 提供以下实用程序来缩放群集:Microsoft provides the following utilities to scale clusters:

实用程序Utility 说明Description
PowerShell AzPowerShell Az Set-AzHDInsightClusterSize -ClusterName CLUSTERNAME -TargetInstanceCount NEWSIZESet-AzHDInsightClusterSize -ClusterName CLUSTERNAME -TargetInstanceCount NEWSIZE
PowerShell AzureRMPowerShell AzureRM Set-AzureRmHDInsightClusterSize -ClusterName CLUSTERNAME -TargetInstanceCount NEWSIZESet-AzureRmHDInsightClusterSize -ClusterName CLUSTERNAME -TargetInstanceCount NEWSIZE
Azure CLIAzure CLI az hdinsight resize --resource-group RESOURCEGROUP --name CLUSTERNAME --workernode-count NEWSIZEaz hdinsight resize --resource-group RESOURCEGROUP --name CLUSTERNAME --workernode-count NEWSIZE
Azure 经典 CLIAzure Classic CLI azure hdinsight cluster resize CLUSTERNAME NEWSIZE
Azure 门户Azure portal 打开 HDInsight 群集的窗格,在左侧菜单中选择“群集大小”,然后在“群集大小”窗格中键入工作节点数并选择“保存”。****Open your HDInsight cluster pane, select Cluster size on the left-hand menu, then on the Cluster size pane, type in the number of worker nodes, and select Save.

Azure 门户缩放群集选项

使用以下任一方法可在几分钟之内扩展或缩放 HDInsight 群集。Using any of these methods, you can scale your HDInsight cluster up or down within minutes.


  • Azure 经典 CLI 已弃用,只应与经典部署模型配合使用。The Azure classic CLI is deprecated and should only be used with the classic deployment model. 进行所有其他的部署时,请使用 Azure CLIFor all other deployments, use the Azure CLI.
  • PowerShell AzureRM 模块已弃用。The PowerShell AzureRM module is deprecated. 请尽可能使用 Az 模块Please use the Az module whenever possible.

缩放操作的影响Impact of scaling operations

将节点添加**** 到正在运行的 HDInsight 群集时(纵向扩展)时,作业不受影响。When you add nodes to your running HDInsight cluster (scale up), jobs won't be affected. 在运行缩放过程时,可以安全提交新作业。New jobs can be safely submitted while the scaling process is running. 如果缩放操作失败,此类失败会让群集保持正常运行状态。If the scaling operation fails, the failure will leave your cluster in a functional state.

如果删除**** 节点(纵向缩减),则在缩放操作完成时,挂起的或正在运行的作业将会失败。If you remove nodes (scale down), pending or running jobs will fail when the scaling operation completes. 此失败的原因是某些服务在缩放过程中重启。This failure is because of some of the services restarting during the scaling process. 在手动缩放操作过程中,群集可能会停滞在安全模式。Your cluster may get stuck in safe mode during a manual scaling operation.

对于 HDInsight 支持的每种类型的群集,更改数据节点数的影响有所不同:The impact of changing the number of data nodes varies for each type of cluster supported by HDInsight:

  • Apache HadoopApache Hadoop

    可无缝增加正在运行的 Hadoop 群集中的工作器节点数,而不会影响任何作业。You can seamlessly increase the number of worker nodes in a running Hadoop cluster without impacting any jobs. 也可在操作进行中提交新作业。New jobs can also be submitted while the operation is in progress. 缩放操作中的失败会得到妥善处理。Failures in a scaling operation are gracefully handled. 群集始终保持正常运行状态。The cluster is always left in a functional state.

    当 Hadoop 群集纵向缩减为使用较少的数据节点时,一些服务会重启。When a Hadoop cluster is scaled down with fewer data nodes, some services are restarted. 此行为会导致所有正在运行和挂起的作业在缩放操作完成时失败。This behavior causes all running and pending jobs to fail at the completion of the scaling operation. 但是,可在操作完成后重新提交这些作业。You can, however, resubmit the jobs once the operation is complete.

  • Apache HBaseApache HBase

    可在 HBase 群集运行时无缝添加或删除节点。You can seamlessly add or remove nodes to your HBase cluster while it's running. 完成缩放操作后的几分钟内,区域服务器自动平衡。Regional Servers are automatically balanced within a few minutes of completing the scaling operation. 不过,也可以手动平衡区域服务器。However, you can manually balance the regional servers. 登录到群集头节点并运行以下命令:Log in to the cluster headnode and run the following commands:

    pushd %HBASE_HOME%\bin
    hbase shell

    有关使用 HBase shell 的详细信息,请参阅 HDInsight 中的 Apache HBase 示例入门For more information on using the HBase shell, see Get started with an Apache HBase example in HDInsight.

  • Apache StormApache Storm

    可在 Storm 运行时无缝添加或删除数据节点。You can seamlessly add or remove data nodes while Storm is running. 但是,在缩放操作成功完成后,需要重新平衡拓扑。However, after a successful completion of the scaling operation, you'll need to rebalance the topology. 重新平衡允许拓扑根据群集中的新节点数重新调整并行度设置Rebalancing allows the topology to readjust parallelism settings based on the new number of nodes in the cluster. 若要重新平衡正在运行的拓扑,请使用下列选项之一:To rebalance running topologies, use one of the following options:

    • Storm Web UIStorm web UI

      使用以下步骤来重新平衡使用 Storm UI 的拓扑。Use the following steps to rebalance a topology using the Storm UI.

      1. 在 Web 浏览器中打开 https://CLUSTERNAME.azurehdinsight.net/stormui,其中 CLUSTERNAME 是 Storm 群集的名称。Open https://CLUSTERNAME.azurehdinsight.net/stormui in your web browser, where CLUSTERNAME is the name of your Storm cluster. 如果出现提示,请输入在创建 HDInsight 群集时指定的群集管理员用户名和密码。If prompted, enter the HDInsight cluster administrator (admin) name and password you specified when creating the cluster.

      2. 选择要重新平衡的拓扑,并选择“重新平衡”**** 按钮。Select the topology you wish to rebalance, then select the Rebalance button. 输入执行重新平衡操作前的延迟。Enter the delay before the rebalance operation is done.

        HDInsight Storm 缩放重新平衡

    • 命令行界面 (CLI) 工具Command-line interface (CLI) tool

      连接到服务器并使用以下命令来重新平衡拓扑:Connect to the server and use the following command to rebalance a topology:

      storm rebalance TOPOLOGYNAME

      还可以指定参数来替代拓扑原来提供的并行度提示。You can also specify parameters to override the parallelism hints originally provided by the topology. 例如,下方代码会将 mytopology 拓扑重新配置为 5 个工作进程,blue-spou 组件有 3 个执行程序,yellow-bolt 组件有 10 个执行程序。For example, the code below reconfigures the mytopology topology to 5 worker processes, 3 executors for the blue-spout component, and 10 executors for the yellow-bolt component.

      ## Reconfigure the topology "mytopology" to use 5 worker processes,
      ## the spout "blue-spout" to use 3 executors, and
      ## the bolt "yellow-bolt" to use 10 executors
      $ storm rebalance mytopology -n 5 -e blue-spout=3 -e yellow-bolt=10
  • KafkaKafka

    执行缩放操作后,应重新均衡分区副本。You should rebalance partition replicas after scaling operations. 有关详细信息,请参阅通过 Apache Kafka on HDInsight 实现数据的高可用性文档。For more information, see the High availability of data with Apache Kafka on HDInsight document.

  • Apache Hive LLAPApache Hive LLAP

    缩放到 N 个工作器节点后,HDInsight 将自动设置以下配置并重启 Hive。After scaling to N worker nodes, HDInsight will automatically set the following configurations and restart Hive.

    • 最大总并发查询数:hive.server2.tez.sessions.per.default.queue = min(N, 32)Maximum Total Concurrent Queries: hive.server2.tez.sessions.per.default.queue = min(N, 32)
    • Hive 的 LLAP 使用的节点数:num_llap_nodes = NNumber of nodes used by Hive's LLAP: num_llap_nodes = N
    • 运行 Hive LLAP 守护程序的节点数:num_llap_nodes_for_llap_daemons = NNumber of Node(s) for running Hive LLAP daemon: num_llap_nodes_for_llap_daemons = N

如何安全地纵向缩减群集How to safely scale down a cluster

通过运行的作业纵向缩减群集Scale down a cluster with running jobs

为了避免运行的作业在纵向缩减操作过程中失败,可以尝试三项操作:To avoid having your running jobs fail during a scale down operation, you can try three things:

  1. 等待作业完成之后再纵向缩减群集。Wait for the jobs to complete before scaling down your cluster.
  2. 手动结束作业。Manually end the jobs.
  3. 在缩放操作完成后重新提交这些作业。Resubmit the jobs after the scaling operation has concluded.

若要查看挂起和正在运行的作业的列表,可以使用 YARN 资源管理器 UI,操作步骤如下:To see a list of pending and running jobs, you can use the YARN Resource Manager UI, following these steps:

  1. Azure 门户中,选择群集。From the Azure portal, select your cluster. 群集会在新的门户页中打开。The cluster is opened in a new portal page.

  2. 在主视图中,导航到“群集仪表板”**** > ****“Ambari 主页”。From the main view, navigate to Cluster dashboards > Ambari home. 输入群集凭据。Enter your cluster credentials.

  3. 在 Ambari UI 的左侧菜单中的服务列表内选择“YARN”。****From the Ambari UI, select YARN on the list of services on the left-hand menu.

  4. 从 "YARN" 页上,选择 " 快速链接 " 并将鼠标悬停在活动头节点上,然后选择 " 资源管理器 UI"。From the YARN page, select Quick Links and hover over the active head node, then select Resource Manager UI.

    Apache Ambari 快速链接资源管理器 UI

您可以使用直接访问资源管理器 UI https://<HDInsightClusterName>.azurehdinsight.net/yarnui/hn/clusterYou may directly access the Resource Manager UI with https://<HDInsightClusterName>.azurehdinsight.net/yarnui/hn/cluster.

可以看到作业的列表及其当前状态。You see a list of jobs, along with their current state. 在屏幕截图中,当前有一个作业正在运行:In the screenshot, there's one job currently running:

资源管理器 UI 应用程序

若要手动终止正在运行的应用程序,请通过 SSH shell 执行以下命令:To manually kill that running application, execute the following command from the SSH shell:

yarn application -kill <application_id>

例如:For example:

yarn application -kill "application_1499348398273_0003"

停滞在安全模式下Getting stuck in safe mode

纵向缩减群集时,HDInsight 使用 Apache Ambari 管理接口先停用额外的工作器节点。When you scale down a cluster, HDInsight uses Apache Ambari management interfaces to first decommission the extra worker nodes. 节点将其 HDFS 块复制到其他联机工作器节点。The nodes replicate their HDFS blocks to other online worker nodes. 然后,HDInsight 安全地纵向缩减群集。After that, HDInsight safely scales the cluster down. HDFS 在缩放操作期间进入安全模式。HDFS goes into safe mode during the scaling operation. HDFS 应该在缩放完成后退出安全模式。HDFS is supposed to come out once the scaling is finished. 但在某些情况下,HDFS 会在缩放操作期间停滞在安全模式下,因为文件块复制数量不足。In some cases, however, HDFS gets stuck in safe mode during a scaling operation because of file block under-replication.

默认情况下,进行 HDFS 配置时,会将 dfs.replication 设置为 1,该项控制每个文件块的可用副本数。By default, HDFS is configured with a dfs.replication setting of 1, which controls how many copies of each file block are available. 文件块的每个副本存储在群集的不同节点上。Each copy of a file block is stored on a different node of the cluster.

在预期的块副本数不可用时,HDFS 会进入安全模式,此时 Ambari 会生成警报。When the expected number of block copies aren't available, HDFS enters safe mode and Ambari generates alerts. HDFS 可能会进入安全模式以进行缩放操作。HDFS may enter safe mode for a scaling operation. 如果未检测到复制所需的节点数,则群集可能会停滞在安全模式。The cluster may get stuck in safe mode if the required number of nodes aren't detected for replication.

启用安全模式时的错误示例Example errors when safe mode is turned on

org.apache.hadoop.hdfs.server.namenode.SafeModeException: Cannot create directory /tmp/hive/hive/819c215c-6d87-4311-97c8-4f0b9d2adcf0. Name node is in safe mode.
org.apache.http.conn.HttpHostConnectException: Connect to active-headnode-name.servername.internal.cloudapp.net:10001 [active-headnode-name.servername. internal.cloudapp.net/] failed: Connection refused

可以查看 /var/log/hadoop/hdfs/ 文件夹中的名称节点日志,以了解缩放群集时群集进入安全模式的大致时间。You can review the name node logs from the /var/log/hadoop/hdfs/ folder, near the time when the cluster was scaled, to see when it entered safe mode. 日志文件命名为 Hadoop-hdfs-namenode-<active-headnode-name>.*The log files are named Hadoop-hdfs-namenode-<active-headnode-name>.*.

根本原因是 Hive 在运行查询时依赖于 HDFS 中的临时文件。The root cause was that Hive depends on temporary files in HDFS while running queries. 当 HDFS 进入安全模式时,Hive 无法运行查询,因为它无法写入 HDFS。When HDFS enters safe mode, Hive can't run queries because it can't write to HDFS. HDFS 中的临时文件位于装载到各个工作器节点 VM 的本地驱动器中。Temp files in HDFS are located in the local drive mounted to the individual worker node VMs. 这些文件将在其他工作器节点之间复制,并至少提供三个副本。The files are replicated among other worker nodes at three replicas, minimum.

如何防止 HDInsight 停滞在安全模式下How to prevent HDInsight from getting stuck in safe mode

可通过多种方法防止 HDInsight 保留在安全模式:There are several ways to prevent HDInsight from being left in safe mode:

  • 在缩减 HDInsight 之前停止所有 Hive 作业。Stop all Hive jobs before scaling down HDInsight. 或者,计划好纵向缩减进程,以避免与运行中的 Hive 作业冲突。Alternately, schedule the scale down process to avoid conflicting with running Hive jobs.
  • 执行缩减操作之前,在 HDFS 中手动清理 Hive 的 scratch tmp 目录文件。Manually clean up Hive's scratch tmp directory files in HDFS before scaling down.
  • 只将 HDInsight 纵向缩减为三个工作节点(最少数量)。Only scale down HDInsight to three worker nodes, minimum. 避免将工作节点数减少至一个。Avoid going as low as one worker node.
  • 根据需要运行命令来退出安全模式。Run the command to leave safe mode, if needed.

以下部分将介绍这些选项。The following sections describe these options.

停止所有 Hive 作业Stop all Hive jobs

在缩减至一个工作节点之前停止所有 Hive 作业。Stop all Hive jobs before scaling down to one worker node. 如果已计划工作负荷,请在完成 Hive 工作后执行缩减。If your workload is scheduled, then execute your scale-down after Hive work is done.

在缩放之前停止 Hive 作业有助于将临时文件夹中的 scratch 文件(如果有)的数目减至最少。Stopping the Hive jobs before scaling, helps minimize the number of scratch files in the tmp folder (if any).

手动清理 Hive 的 scratch 文件Manually clean up Hive's scratch files

如果 Hive 遗留了临时文件,可以在缩减之前手动清理这些文件,以避免进入安全模式。If Hive has left behind temporary files, then you can manually clean up those files before scaling down to avoid safe mode.

  1. 通过查看 hive.exec.scratchdir 配置属性,了解用于 Hive 临时文件的具体位置。Check which location is being used for Hive temporary files by looking at the hive.exec.scratchdir configuration property. 此参数在 /etc/hive/conf/hive-site.xml 中设置:This parameter is set within /etc/hive/conf/hive-site.xml:

  2. 停止 Hive 服务,并确保所有查询和作业都已完成。Stop Hive services and be sure all queries and jobs are completed.

  3. 列出在上面找到的 scratch 目录 hdfs://mycluster/tmp/hive/ 的内容,看其是否包含任何文件:List the contents of the scratch directory found above, hdfs://mycluster/tmp/hive/ to see if it contains any files:

    hadoop fs -ls -R hdfs://mycluster/tmp/hive/hive

    下面是存在文件时的示例输出:Here is a sample output when files exist:

    sshuser@scalin:~$ hadoop fs -ls -R hdfs://mycluster/tmp/hive/hive
    drwx------   - hive hdfs          0 2017-07-06 13:40 hdfs://mycluster/tmp/hive/hive/4f3f4253-e6d0-42ac-88bc-90f0ea03602c
    drwx------   - hive hdfs          0 2017-07-06 13:40 hdfs://mycluster/tmp/hive/hive/4f3f4253-e6d0-42ac-88bc-90f0ea03602c/_tmp_space.db
    -rw-r--r--   3 hive hdfs         27 2017-07-06 13:40 hdfs://mycluster/tmp/hive/hive/4f3f4253-e6d0-42ac-88bc-90f0ea03602c/inuse.info
    -rw-r--r--   3 hive hdfs          0 2017-07-06 13:40 hdfs://mycluster/tmp/hive/hive/4f3f4253-e6d0-42ac-88bc-90f0ea03602c/inuse.lck
    drwx------   - hive hdfs          0 2017-07-06 20:30 hdfs://mycluster/tmp/hive/hive/c108f1c2-453e-400f-ac3e-e3a9b0d22699
    -rw-r--r--   3 hive hdfs         26 2017-07-06 20:30 hdfs://mycluster/tmp/hive/hive/c108f1c2-453e-400f-ac3e-e3a9b0d22699/inuse.info
  4. 如果知道 Hive 已处理这些文件,则可以删除这些文件。If you know Hive is done with these files, you can remove them. 在 Yarn 资源管理器 UI 页中查看,确保配置单元没有任何正在运行的查询。Be sure that Hive doesn't have any queries running by looking in the Yarn Resource Manager UI page.

    用于从 HDFS 中删除文件的示例命令行:Example command line to remove files from HDFS:

    hadoop fs -rm -r -skipTrash hdfs://mycluster/tmp/hive/

缩减 HDInsight 时保持三个或更多个工作器节点Scale HDInsight to three or more worker nodes

如果将群集纵向缩减到少于三个工作器节点,群集通常会停滞在安全模式,因此,请至少保留三个工作器节点。If your clusters get stuck in safe mode frequently when scaling down to fewer than three worker nodes, then keep at least three worker nodes.

与纵向缩减到仅一个工作器节点相比,拥有三个工作器节点的成本更高。Having three worker nodes is more costly than scaling down to only one worker node. 但是,此操作将防止群集停滞在安全模式。However, this action will prevent your cluster from getting stuck in safe mode.

将 HDInsight 缩减到一个工作器节点Scale HDInsight down to one worker node

即使群集纵向缩减到 1 个节点,工作器节点 0 仍将继续存在。Even when the cluster is scaled down to one node, worker node 0 will still survive. 永远不能停用工作器节点 0。Worker node 0 can never be decommissioned.

运行命令来退出安全模式。Run the command to leave safe mode

最后一种做法是执行退出安全模式的命令。The final option is to execute the leave safe mode command. 如果 HDFS 进入安全模式的原因是 Hive 文件复制数量不足,请执行以下命令退出安全模式:If HDFS entered safe mode because of Hive file under-replication, execute the following command to leave safe mode:

hdfs dfsadmin -D 'fs.default.name=hdfs://mycluster/' -safemode leave

纵向缩减 Apache HBase 群集Scale down an Apache HBase cluster

在完成缩放操作后的几分钟内,区域服务器会自动进行均衡。Region servers are automatically balanced within a few minutes after completing a scaling operation. 若要手动均衡区域服务器,请完成以下步骤:To manually balance region servers, complete the following steps:

  1. 使用 SSH 连接到 HDInsight 群集。Connect to the HDInsight cluster using SSH. 有关详细信息,请参阅 将 SSH 与 HDInsight 配合使用For more information, see Use SSH with HDInsight.

  2. 启动 HBase shell:Start the HBase shell:

    hbase shell
  3. 使用以下命令手动均衡区域服务器:Use the following command to manually balance the region servers:


后续步骤Next steps

有关缩放 HDInsight 群集的特定信息,请参阅:For specific information on scaling your HDInsight cluster, see: