mesos / mih Goto Github PK

MIH (Mesos-in-Hadoop) lets you launch a Mesos cluster as a Hadoop job, for
example to easily experiment with Mesos frameworks like Spark on an existing
Hadoop cluster.

MIH requires Hadoop 0.20.203 or later. It should also work with the Cloudera
distribution of Hadoop version 3 or later. On older Hadoops (e.g. 0.20.2),
MIH will run but Hadoop does not properly clean up processes spawned by a job
when the job finishes, so you risk having processes left on the nodes.

To build MIH, export HADOOP_HOME=<path to your Hadoop directory> and run make.

To run, export MESOS_HOME=<location where you installed Mesos on the slaves>
and use HADOOP_HOME/bin/hadoop mih.jar <number of slots>.

MIH will acquire the given number of slots from Hadoop (as a set of long
running map tasks) and launch a Mesos cluster inside them, then print out
the master URL for that cluster. To stop the cluster, kill the corresponding
Hadoop job with HADOOP_HOME/bin/hadoop job -kill.

You can also provide a second argument to mih.jar for the memory to use per
slot (in megabytes). By default, it's 1024 MB.