Description

Hadoop is a software platform that lets one easily write and run applications that process vast amounts of data.
This charm manages the YARN master node (ResourceManager).

Overview

The Apache Hadoop software library is a framework that allows for the
distributed processing of large data sets across clusters of computers
using a simple programming model.

This charm deploys a node running the ResourceManager component of
Apache Hadoop 2.4.1,
which manages the computation resources and job execution for the platform.

Usage

This charm is intended to be deployed via one of the
apache bundles.
For example:

juju quickstart apache-analytics-sql

This will deploy the Apache Hadoop platform with Apache Hive available to
perform SQL-like queries against your data.

You can also manually load and run map-reduce jobs:

juju scp my-job.jar resourcemanager/0:
juju ssh resourcemanager/0
hadoop jar my-job.jar

Status and Smoke Test

The services provide extended status reporting to indicate when they are ready:

juju status --format=tabular

This is particularly useful when combined with watch to track the on-going
progress of the deployment:

watch -n 0.5 juju status --format=tabular

The message for each unit will provide information about that unit's state.
Once they all indicate that they are ready, you can perform a "smoke test"
to verify that Yarn is working as expected using the built-in smoke-test
action:

juju action do resourcemanager/0 smoke-test

After a few seconds or so, you can check the results of the smoke test:

juju action status

You will see status: completed if the smoke test was successful, or
status: failed if it was not. You can get more information on why it failed
via:

juju action fetch <action-id>

Benchmarking

This charm provides several benchmarks to gauge the performance of your
environment.

The easiest way to run the benchmarks on this service is to relate it to the
Benchmark GUI. You will likely also want to relate it to the
Benchmark Collector to have machine-level information collected during the
benchmark, for a more complete picture of how the machine performed.

However, each benchmark is also an action that can be called manually:

    $ juju action do resourcemanager/0 terasort
    Action queued with id: cbd981e8-3400-4c8f-8df1-c39c55a7eae6
    $ juju action fetch --wait 0 cbd981e8-3400-4c8f-8df1-c39c55a7eae6
    results:
      meta:
        composite:
          direction: asc
          units: ms
          value: "206676"
      results:
        raw: '{"Total vcore-seconds taken by all map tasks": "439783", "Spilled Records":
          "30000000", "WRONG_LENGTH": "0", "Reduce output records": "10000000", "HDFS:
          Number of bytes read": "1000001024", "Total vcore-seconds taken by all reduce
          tasks": "50275", "Reduce input groups": "10000000", "Shuffled Maps ": "8", "FILE:
          Number of bytes written": "3128977482", "Input split bytes": "1024", "Total
          time spent by all reduce tasks (ms)": "50275", "FILE: Number of large read operations":
          "0", "Bytes Read": "1000000000", "Virtual memory (bytes) snapshot": "7688794112",
          "Launched map tasks": "8", "GC time elapsed (ms)": "11656", "Bytes Written":
          "1000000000", "FILE: Number of read operations": "0", "HDFS: Number of write
          operations": "2", "Total megabyte-seconds taken by all reduce tasks": "51481600",
          "Combine output records": "0", "HDFS: Number of bytes written": "1000000000",
          "Total time spent by all map tasks (ms)": "439783", "Map output records": "10000000",
          "Physical memory (bytes) snapshot": "2329722880", "FILE: Number of write operations":
          "0", "Launched reduce tasks": "1", "Reduce input records": "10000000", "Total
          megabyte-seconds taken by all map tasks": "450337792", "WRONG_REDUCE": "0",
          "HDFS: Number of read operations": "27", "Reduce shuffle bytes": "1040000048",
          "Map input records": "10000000", "Map output materialized bytes": "1040000048",
          "CPU time spent (ms)": "195020", "Merged Map outputs": "8", "FILE: Number of
          bytes read": "2080000144", "Failed Shuffles": "0", "Total time spent by all
          maps in occupied slots (ms)": "439783", "WRONG_MAP": "0", "BAD_ID": "0", "Rack-local
          map tasks": "2", "IO_ERROR": "0", "Combine input records": "0", "Map output
          bytes": "1020000000", "CONNECTION": "0", "HDFS: Number of large read operations":
          "0", "Total committed heap usage (bytes)": "1755840512", "Data-local map tasks":
          "6", "Total time spent by all reduces in occupied slots (ms)": "50275"}'
    status: completed
    timing:
      completed: 2015-05-28 20:55:50 +0000 UTC
      enqueued: 2015-05-28 20:53:41 +0000 UTC
      started: 2015-05-28 20:53:44 +0000 UTC

Monitoring

This charm supports monitoring via Ganglia. To enable monitoring, you must
do both of the following (the order does not matter):

  • Add a relation to the Ganglia charm via the :master relation
  • Enable the ganglia_metrics config option

For example:

juju add-relation yarn-master ganglia:master
juju set yarn-master ganglia_metrics=true

Enabling monitoring will issue restart the ResourceManager and all NodeManager
components on all of the related compute-slaves. Take care to ensure that there
are no running jobs when enabling monitoring.

Deploying in Network-Restricted Environments

The Apache Hadoop charms can be deployed in environments with limited network
access. To deploy in this environment, you will need a local mirror to serve
the packages and resources required by these charms.

Mirroring Packages

You can setup a local mirror for apt packages using squid-deb-proxy.
For instructions on configuring juju to use this, see the
Juju Proxy Documentation.

Mirroring Resources

In addition to apt packages, the Apache Hadoop charms require a few binary
resources, which are normally hosted on Launchpad. If access to Launchpad
is not available, the jujuresources library makes it easy to create a mirror
of these resources:

sudo pip install jujuresources
juju-resources fetch --all /path/to/resources.yaml -d /tmp/resources
juju-resources serve -d /tmp/resources

This will fetch all of the resources needed by this charm and serve them via a
simple HTTP server. The output from juju-resources serve will give you a
URL that you can set as the resources_mirror config option for this charm.
Setting this option will cause all resources required by this charm to be
downloaded from the configured URL.

You can fetch the resources for all of the Apache Hadoop charms
(apache-hadoop-hdfs-master, apache-hadoop-yarn-master,
apache-hadoop-hdfs-secondary, apache-hadoop-plugin, etc) into a single
directory and serve them all with a single juju-resources serve instance.

Contact Information

Hadoop

Configuration

ganglia_metrics
(boolean) Enable metrics using Ganglia. Note that enabling this option will have no effect if the service is not related to a ganglia service via the ganglia:master relation. Enabling this option with the relation will issue a restart to the ResourceManager and all NodeManager components on all related compute-slaves. See the README for more information.
resources_mirror
(string) URL from which to fetch resources (e.g., Hadoop binaries) instead of Launchpad.
hadoop_version
(string) Hadoop version
2.7.2