This topic describes how to run Flume on a Gateway node to synchronize data based on Alibaba Cloud E-MapReduce (EMR) V3.17.0 and later versions.


EMR has supported Apache Flume since V3.16.0 and has supported default monitoring since V3.17.0.

  • Basic data flows
    Running Flume on Gateway nodes avoids the impact on EMR Hadoop clusters. Basic data flows that are streamed through Flume agents installed on Gateway nodes are shown in the following figure.
    Basic data flows

Prepare the environment

The test is performed using EMR that is deployed in the China East 1 (Hangzhou) region. The version of EMR is V3.17.0. The components required for this test are as follows.
  • Flume: 1.8.0

You can use EMR to automatically create a Hadoop cluster. For more information, see Create a cluster.

  • Click Create Cluster, click Flume for the cluster type, and select Flume from Optional Services.

  • Create a Gateway node and associate it to the Hadoop cluster created in the previous step.


  • Run Flume
    • The default path of Flume configuration files is /etc/ecm/flume-conf. See Use Flume for modifying the configuration file for Flume agents. After the modification, use the following command to run a Flume agent.
      nohup flume-ng agent -n a1 -f &
    • You can use the -c flag or the --conf flag to replace the default configuration file with a custom file. For example:
      nohup flume-ng agent -n a1 -f -c path-to-flume-conf &
    Notice For more information, see Use Flume. You need to add the zookeeperQuorum configuration item to the configuration file when the Flume agents installed on Gateways use sinks to write data to HBase. For example:
    The hostname of the ZooKeeper cluster emr-header-1.cluster-46349 is the value of the hbase.zookeeper.quorum configuration item in the /etc/ecm/hbase-conf/hbase-site.xml file.
  • View monitoring information
    Monitoring data of Flume agents is displayed in the cluster console by default. On the Clusters and Services page, click FLUME to jump to the cluster console as shown in the following figure.
    The FLUME service page

    Monitoring data is classified by the components (sources, channels, or sinks) of Flume agents. For example, CHANNEL.channel1 represents monitoring data of the channel1 channel component. Note: Avoid using the same component name when configuring different agents.

    Refer to the official Flume website for viewing monitoring data of Flume agents using Ganglia by creating proper configurations. After doing this, monitoring data of Flume agents will not be displayed in the console.

  • View logs
    By default, the log path of a Flume agent is /mnt/disk1/log/flume/${flume-agent-name}/flume.log. You can modify the /etc/ecm/flume-conf/ configuration file to change the log path. However, we recommend that you do not change the default log path.
    Notice A log path contains a Flume agent name. Provide a unique name for each agent to avoid logs of different agents to be stored in the same directory.