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.
It is designed to scale up from single servers to thousands of machines, each offering local computation and storage. Rather than rely on hardware to deliver high-availability, the library itself is designed to detect and handle failures at the application layer, so delivering a highly-available service on top of a cluster of computers, each of which may be prone to failures.
This bundle provides a complete deployment of the core components of the Apache Hadoop 2.7.1 platform to perform distributed data analytics at scale. These components include:
- NameNode (HDFS)
- ResourceManager (Yarn)
- Slaves (DataNode and NodeManager)
- Client (example and node for manually running jobs from)
- Plugin (collocated on the Client)
This bundle also includes the following services for monitoring an analytics:
- Ganglia (monitoring UI)
- Flume Syslog & Flume HDFS (collect logs and ingest into HDFS for analysis)
Deploying this bundle gives you a fully configured and connected Apache Hadoop cluster on any supported cloud, which can be easily scaled to meet workload demands.
In this deployment, the aforementioned components are deployed on separate units. To deploy this bundle, simply use:
juju quickstart apache-processing-mapreduce
See juju quickstart --help
for deployment options, including machine
constraints and how to deploy a locally modified version of bundle.yaml
.
The default bundle deploys three slave nodes and one node of each of the other services. To scale the cluster, use:
juju add-unit slave -n 2
This will add two additional slave nodes, for a total of five.
conjure-up
is an interactive, terminal UI deployment tool for Juju bundles.
After installing conjure-up, you can deploy the canonical-kubernetes bundle and tweak config values with one command:
sudo apt install conjure-up
conjure-up apache-processing-mapreduce
Refer to the conjure-up documentation to learn more.
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 charms for each master component (namenode, resourcemanager)
also each provide a smoke-test
action that can be used to verify that each
component is functioning as expected. You can run them all and then watch the
action status list:
juju action do namenode/0 smoke-test
juju action do resourcemanager/0 smoke-test
watch -n 0.5 juju action status
Eventually, all of the actions should settle to status: completed
. If
any go instead to status: failed
then it means that component is not working
as expected. You can get more information about that component's smoke test:
juju action fetch <action-id>
Monitoring is provided by the Ganglia, which is available at
http://<ganglia-host>/ganglia
where <ganglia-host>
can be discovered from
juju status
. The logs from all nodes are also ingested into HDFS and can
be analyzed using Yarn and map-reduce jobs.
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.
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.
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 a 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 that charm.
Setting this option will cause all resources required by the charm to be
downloaded from the configured URL.
You can fetch the resources for all of the Apache Hadoop charms
(apache-hadoop-namenode
, apache-hadoop-resourcemanager
,
apache-hadoop-slave
, apache-hadoop-plugin
, etc) into a single
directory and serve them all with a single juju-resources serve
instance.