Hadoop is a software platform that lets one easily write and run applications that process vast amounts of data.
This charm manages a dedicated client node as a place to run mapreduce jobs. However, its main purpose is to serve as a base layer for other client charms, such as Gobblin' or Spark.
This is the base layer for charms that wish to connect to the core Apache Hadoop cluster. Including this layer provides and manages the relation to hadoop-plugin, and all your reactive charm needs to do is respond to one or more of the states listed below.
The plugin services provides the appropriate Hadoop libraries for the cluster,
and sets up the standard Hadoop config files in
To create a charm layer using this base layer, you need only include it in
This will fetch this layer from interfaces.juju.solutions and incorporate
it into your charm layer. You can then add handlers under the
directory. Note that any file under
reactive/ will be expected to
contain handlers, whether as Python decorated functions or executables
using the external handler protocol.
This layer, via the hadoop-plugin interface, will set the following states:
hadoop.hdfs.readyThe Hadoop cluster has indicated that HDFS is ready to store data. Handlers reacting to this state will be passed an instance of the hadoop-plugin class, and can use the following methods to access information about HDFS:
hadoop.namenodes()A list of one or two NameNode addresses.
hadoop.hdfs_port()The port on which the NameNodes are listening.
hadoop.yarn.readyThe Hadoop cluster has indicated that Yarn is ready to process data. Handlers reacting to this state will be passed an instance of the hadoop-plugin class, and can use the following methods to access information about Yarn:
hadoop.resourcemanagers()A list of one or two ResourceManager addresses.
hadoop.yarn_port()The port on which the ResourceManagers are listening.
hadoop.yarn_hs_ipc_port()The IPC port for the Yarn HistoryServer.
hadoop.yarn_hs_http_port()The HTTP port for the Yarn HistoryServer.
hadoop.readyThe Hadoop cluster has indicated that both HDFS and Yarn are ready. This is a combination of the previous two states, and also provides an instance upon which any of the previously mentioned methods can be called.
An example using these states would be:
@when('hadoop.ready') def configure_service(hadoop): update_config( hadoop.namenodes(), hadoop.hdfs_port(), hadoop.resourcemanagers(), hadoop.yarn_port()) restart_service()
This layer supports the following options, which can be set in
packages A list of system packages to be installed when Hadoop is being installed.
groups A list of system groups to be created when Hadoop is being configured.
users A list of system users to be created when Hadoop is being configured.
dirs A mapping of directories to be created when Hadoop is being configured. Each entry should contain the following keys:
- path Absolute directory path.
- perms Octal permissions.
- owner User name to own directory.
- group Name of group for directory.
layer.yaml using these options might be:
includes: ['layer:hadoop-client'] options: hadoop-client: groups: [spark] users: [spark] dirs: spark_home: path: /var/lib/spark perms: 0755 owner: spark group: spark
This layer currently does not define any actions.