This interface layer handles the communication with HDFS via the dfs interface protocol. It is intended for internal use within the Hadoop cluster charms. For typical usage, [interface-hadoop-plugin][] should be used instead.

Charms requiring this interface can be clients, DataNodes, or SecondaryNameNodes. Clients simply depend on HDFS to serve as a distributed file system to them, while DataNodes and SecondaryNameNodes register to provide additional services back.

This interface layer will set the following states, as appropriate:

• {relation_name}.joined The relation is established, but HDFS may not yet have provided any connection or service information.

• {relation_name}.ready HDFS has provided its connection and service information, and is ready to serve as a distributed file system. The provided information can be accessed via the following methods:

  • hosts_map()
  • port()
  • webhdfs_port()

For example, a typical client would respond to {relation_name}.ready:

@when('flume.installed', 'hdfs.ready')
def hdfs_ready(hdfs):
    flume.configure(hdfs)
    flume.start()

A charm providing this interface is providing the HDFS service.

This interface layer will set the following states, as appropriate:

• {relation_name}.clients One or more clients of any type have been related. The charm should call the following methods to provide the appropriate information to the clients:
  • send_spec(spec)
  • send_hosts_map(hosts)
  • send_ports(port, webhdfs_port)
  • send_ready(ready)

Example:

@when('namenode.clients')
@when('hdfs.ready')
def serve_client(client):
    client.send_spec(utils.build_spec())
    client.send_ports(dist_config.get('port'), dist_config.get('webhdfs_port'))
    client.send_ready(True)

@when('namenode.clients')
@when_not('hdfs.ready')
def check_ready(client):
    client.send_ready(False)

• [email protected]

• Apache Hadoop home page
• Apache Hadoop bug trackers
• Apache Hadoop mailing lists
• Apache Hadoop Juju Charm