This project has retired. For details please refer to its Attic page.

Chukwa Administration Guide

This chapter is the detailed configuration guide to Chukwa configuration.

Please read this chapter carefully and ensure that all requirements have been satisfied. Failure to do so will cause you (and us) grief debugging strange errors and/or data loss.

Chukwa uses the same configuration system as Hadoop. To configure a deploy, edit a file of environment variables in etc/chukwa/ -- this configuration is used mostly by the launcher shell scripts getting the cluster off the ground -- and then add configuration to an XML file to do things like override Chukwa defaults, tell Chukwa what Filesystem to use, or the location of the HBase configuration.

When running in distributed mode, after you make an edit to an Chukwa configuration, make sure you copy the content of the conf directory to all nodes of the cluster. Chukwa will not do this for you. Use rsync.


Chukwa should work on any POSIX platform, but GNU/Linux is the only production platform that has been tested extensively. Chukwa has also been used successfully on Mac OS X, which several members of the Chukwa team use for development.

The only absolute software requirements are Java 1.6 or better and Hadoop

HICC, the Chukwa visualization interface, requires HBase 0.90.4+.

The Chukwa cluster management scripts rely on ssh; these scripts, however, are not required if you have some alternate mechanism for starting and stopping daemons.

Installing Chukwa

A minimal Chukwa deployment has three components:

  • A Hadoop and HBase cluster on which Chukwa will process data (referred to as the Chukwa cluster).
  • A collector process, that writes collected data to HBase.
  • One or more agent processes, that send monitoring data to the collector. The nodes with active agent processes are referred to as the monitored source nodes.
  • Data analytics script, summarize Hadoop Cluster Health.
  • HICC, the Chukwa visualization tool.
Chukwa Components

First Steps

  • Obtain a copy of Chukwa. You can find the latest release on the Chukwa release page.
  • Un-tar the release, via tar xzf.
  • Make sure a copy of Chukwa is available on each node being monitored, and on each node that will run a collector.
  • We refer to the directory containing Chukwa as CHUKWA_HOME. It may be helpful to set CHUKWA_HOME explicitly in your environment, but Chukwa does not require that you do so.

General Configuration

Agents and collectors are configured differently, but part of the process is common to both.

  • Make sure that JAVA_HOME is set correctly and points to a Java 1.6 JRE. It's generally best to set this in etc/chukwa/
  • In etc/chukwa/, set CHUKWA_LOG_DIR and CHUKWA_PID_DIR to the directories where Chukwa should store its console logs and pid files. The pid directory must not be shared between different Chukwa instances: it should be local, not NFS-mounted.
  • Optionally, set CHUKWA_IDENT_STRING. This string is used to name Chukwa's own console log files.


Agents are the Chukwa processes that actually produce data. This section describes how to configure and run them. More details are available in the Agent configuration guide.


This section describes how to set up the agent process on the source nodes.

The one mandatory configuration step is to set up $CHUKWA_HOME/etc/chukwa/collectors. This file should contain a list of hosts that will run Chukwa collectors. Agents will pick a random collector from this list to try sending to, and will fail-over to another listed collector on error. The file should look something like:


Edit the CHUKWA_HOME/etc/chukwa/initial_adaptors configuration file. This is where you tell Chukwa what log files to monitor. See the adaptor configuration guide for a list of available adaptors.

There are a number of optional settings in $CHUKWA_HOME/etc/chukwa/chukwa-agent-conf.xml:

  • The most important of these is the cluster/group name that identifies the monitored source nodes. This value is stored in each Chunk of collected data; you can therefore use it to distinguish data coming from different groups of machines.
        <description>The cluster's name for this agent</description>
  • Another important option is chukwaAgent.checkpoint.dir. This is the directory Chukwa will use for its periodic checkpoints of running adaptors. It must not be a shared directory; use a local, not NFS-mount, directory.
  • Setting the option chukwaAgent.control.remote will disallow remote connections to the agent control socket.

Starting, Stopping, And Monitoring

To run an agent process on a single node, use bin/chukwa agent.

Typically, agents run as daemons. The script bin/ will ssh to each machine listed in etc/chukwa/agents and start an agent, running in the background. The script bin/ does the reverse.

You can, of course, use any other daemon-management system you like. For instance, tools/init.d includes init scripts for running Chukwa agents.

To check if an agent is working properly, you can telnet to the control port (9093 by default) and hit "enter". You will get a status message if the agent is running normally.

Configuring Hadoop For Monitoring

One of the key goals for Chukwa is to collect logs from Hadoop clusters. This section describes how to configure Hadoop to send its logs to Chukwa. Note that these directions require Hadoop Earlier versions of Hadoop do not have the hooks that Chukwa requires in order to grab MapReduce job logs.

The Hadoop configuration files are located in HADOOP_HOME/etc/hadoop. To setup Chukwa to collect logs from Hadoop, you need to change some of the Hadoop configuration files.

  • Copy CHUKWA_HOME/etc/chukwa/ file to HADOOP_CONF_DIR/
  • Copy CHUKWA_HOME/etc/chukwa/ file to HADOOP_CONF_DIR/
  • Edit HADOOP_HOME/etc/hadoop/ file and change $CHUKWA_LOG_DIR to your actual CHUKWA log dirctory (ie, CHUKWA_HOME/var/log)

Setup HBase Table

Chukwa is moving towards a model of using HBase to store metrics data to allow real-time charting. This section describes how to configure HBase and HICC to work together.

  • Presently, we support HBase 0.90.4+. If you have HBase 0.89 jars anywhere, they will cause linkage errors. Check for and remove them.
  • Setting up tables:
    /path/to/hbase-0.90.4/bin/hbase shell < etc/chukwa/hbase.schema


This section describes how to set up the Chukwa collectors. For more details, see the collector configuration guide.


First, edit $CHUKWA_HOME/etc/chukwa/ In addition to the general directions given above, you should set HADOOP_CONF_DIR and HBASE_CONF_DIR. This should be the Hadoop deployment Chukwa will use to store collected data. You will get a version mismatch error if this is configured incorrectly.

Next, edit $CHUKWA_HOME/etc/chukwa/chukwa-collector-conf.xml.

Use HBase For Data Storage

  • Configuring the collector: set HBaseWriter as your writer, or add it to the pipeline if you are using

Use HDFS For Data Storage

The one mandatory configuration parameter is writer.hdfs.filesystem. This should be set to the HDFS root URL on which Chukwa will store data. Various optional configuration options are described in the collector configuration guide and in the collector configuration file itself.

Starting, Stopping, And Monitoring

To run a collector process on a single node, use bin/chukwa collector.

Typically, collectors run as daemons. The script bin/ will ssh to each collector listed in etc/chukwa/collectors and start a collector, running in the background. The script bin/ does the reverse.

You can, of course, use any other daemon-management system you like. For instance, tools/init.d includes init scripts for running Chukwa collectors.

To check if a collector is working properly, you can simply access http://collectorhost:collectorport/chukwa?ping=true with a web browser. If the collector is running, you should see a status page with a handful of statistics.

ETL Processes (Optional)

For storing data to HDFS, the archive and demux mapreduce jobs can be started by running:

CHUKWA_HOME/bin/chukwa archive

Demux mapreduce jobs can be started by rnning:

CHUKWA_HOME/bin/chukwa demux

Setup Cluster Aggregation Script

For data analytics with Apache Pig, there are some additional environment setup. Apache Pig does not use the same environment variable name as Hadoop, therefore make sure the following environment are setup correctly:

  1. Download and setup Apache Pig 0.9.1.
  2. Define Apache Pig class path:
  3. Create a jar file of HBASE_CONF_DIR, run:
    jar cf $CHUKWA_HOME/hbase-env.jar $HBASE_CONF_DIR
  4. Setup a cron job or Hudson job for analytics script to run periodically:
    pig -Dpig.additional.jars=${HBASE_HOME}/hbase-0.90.4.jar:${HBASE_HOME}/lib/zookeeper-3.3.2.jar:${PIG_PATH}/pig.jar:${CHUKWA_HOME}/hbase-env.jar ${CHUKWA_HOME}/script/pig/ClusterSummary.pig



Edit etc/chukwa/auth.conf and add authorized user to the list.

Starting, Stopping, And Monitoring

The Hadoop Infrastructure Care Center (HICC) is the Chukwa web user interface. HICC is started by invoking

bin/chukwa hicc

Once the webcontainer with HICC has been started, point your favorite browser to:


Troubleshooting Tips

UNIX Processes For Chukwa Agents

The Chukwa agent process name is identified by:


Command line to use to search for the process name:

ps ax | grep org.apache.hadoop.chukwa.datacollection.agent.ChukwaAgent

UNIX Processes For Chukwa Collectors

Chukwa Collector name is identified by:


UNIX Processes For Chukwa Data Processes

Chukwa Data Processors are identified by:


The processes are scheduled execution, therefore they are not always visible from the process list.

Checks For Disk Full

If anything is wrong, use /etc/init.d/chukwa-agent and CHUKWA_HOME/tools/init.d/chukwa-system-metrics stop to shutdown Chukwa. Look at agent.log and collector.log file to determine the problems.

The most common problem is the log files are growing unbounded. Set up a cron job to remove old log files:

 0 12 * * * CHUKWA_HOME/tools/ 10 $CHUKWA_HOME/var/log nowait

This will set up the log file expiration for CHUKWA_HOME/var/log for log files older than 10 days.

Emergency Shutdown Procedure

If the system is not functioning properly and you cannot find an answer in the Administration Guide, execute the kill command. The current state of the java process will be written to the log files. You can analyze these files to determine the cause of the problem.

kill -3 <pid>