关于Hortonworks边缘节点使用及6节点HA生产集群的技术咨询

Hey there! Let’s dive into how to make the most of edge nodes in your 6-node Hortonworks HA cluster (2 masters + 4 slaves) based on the topology you shared. Your master nodes are already running a ton of critical services (Active NameNode, ResourceManager, HiveServer2, etc.), so edge nodes are perfect for offloading client-side work without bogging down the core cluster.

Core Use Cases for Edge Nodes in Your Setup #

Edge nodes act as the "user-facing gateway" to your cluster—here’s how they fit your specific environment:

• Centralized Client Tool Hosting: Install Hive CLI/Beeline, HBase Shell, Spark CLI, Sqoop, and Flume clients here instead of scattering them across master/slave nodes. This makes client management way easier.
• Job Submission Offload: Submit Spark, MapReduce, or Oozie jobs from the edge node instead of your master nodes. Your Master 1 is already running Active ResourceManager and other heavy services—this takes pressure off it.
• Ad-Hoc Querying: Run Hive or HBase ad-hoc queries via Beeline or HBase Shell. You can even install a secondary Zeppelin instance here for teams that need their own notebook environment (since your Master 2 already has one).
• Data Ingestion Hub: Deploy Flume agents or Sqoop jobs to pull external data into HDFS/HBase. Edge nodes are ideal as a safe, isolated entry point for external data.
• Dev/Test Sandbox: Give developers a dedicated environment to test cluster interactions without letting them touch your core master/slave nodes.

Setting Up Your Edge Node: Critical Steps #

Since you’re using Hortonworks, leverage Ambari for smooth configuration—here’s what to focus on:

• Install Matching Client Packages: Add the edge node as a new host in Ambari, then run the Install Client wizard. Make sure to check all clients relevant to your cluster (HDFS, YARN, Hive, HBase, Spark, etc.) to guarantee version compatibility.
• Sync HA Configuration Files: Ambari should automatically sync HA-specific configs (like hdfs-site.xml with JournalNode/NameNode HA details, yarn-site.xml with ResourceManager HA settings) to the edge node’s /etc/hadoop/conf, /etc/hive/conf, etc. If you make manual changes later, don’t forget to re-sync!
• Kerberos Integration (If Enabled): If your cluster uses Kerberos, join the edge node to your Kerberos domain, copy over krb5.conf, and set up keytabs for users who need to authenticate to cluster services.
• Network Access: Ensure the edge node can reach all cluster service ports—like HDFS 8020, YARN 8032, HiveServer2 10000, and ZooKeeper 2181. Adjust firewalls accordingly to avoid connection failures.

Scenario-Specific Examples for Your Cluster #

Let’s put this into action with real commands tailored to your setup:

• Submitting a Spark Job to YARN:

  spark-submit --class com.yourcompany.YourJob \
    --master yarn \
    --deploy-mode cluster \
    /path/to/your-job.jar \
    arg1 arg2
  

  This sends the job to your Active ResourceManager (Master 1) while keeping the edge node only responsible for submission, not execution.
• Querying Hive via Beeline:

  # With Kerberos authentication
  beeline -u "jdbc:hive2://master1.yourdomain.com:10000/default;principal=hive/_HOST@YOURDOMAIN.COM"
  # Without Kerberos
  beeline -u "jdbc:hive2://master1.yourdomain.com:10000/default"
  

• Ingesting Logs to HDFS with Flume:
  Create a flume-log-ingest.conf file:

  agent.sources = tailSource
  agent.sinks = hdfsSink
  agent.channels = memoryChannel
  
  agent.sources.tailSource.type = exec
  agent.sources.tailSource.command = tail -F /var/log/your-app.log
  agent.sources.tailSource.channels = memoryChannel
  
  agent.channels.memoryChannel.type = memory
  agent.channels.memoryChannel.capacity = 1000
  
  agent.sinks.hdfsSink.type = hdfs
  agent.sinks.hdfsSink.hdfs.path = hdfs://nameservice1/user/flume/app-logs/%Y%m%d
  agent.sinks.hdfsSink.hdfs.filePrefix = app-log
  agent.sinks.hdfsSink.channel = memoryChannel
  

  Start the Flume agent:

  flume-ng agent --conf /etc/flume/conf --conf-file /path/to/flume-log-ingest.conf --name agent -Dflume.root.logger=INFO,console
  

  Note: nameservice1 is your HA HDFS service name—match it to your cluster’s config.

Best Practices to Keep Things Running Smoothly #

• Don’t Run Core Cluster Services on Edge Nodes: Keep it lightweight—only clients and ingestion tools belong here.
• Limit Resource Usage: If multiple teams use the edge node, use Linux cgroups or ulimit to prevent one user from hogging CPU/memory.
• Sync Configs Regularly: Whenever you update cluster settings (like HA service endpoints), use Ambari to push changes to the edge node.
• Monitor the Edge Node: Install the same Metrics Monitor you have on Master 1 to track CPU, memory, and disk usage—this helps catch bottlenecks early.
• Consider Multiple Edge Nodes: If you have separate teams or workloads, deploy multiple edge nodes to avoid conflicts and improve isolation.

内容的提问来源于stack exchange，提问作者Nabil