Migrating from MapReduce Version 1 to MapReduce Version 2
If you previously ran MRv1 jobs on your cluster, prepare your cluster to run YARN applications in Data Fabric 6.0 or later.
Configuring the MapReduce Mode
Client and cluster nodes submit MapReduce applications to the YARN framework
(yarn
mode) unless you configure them to use the classic framework
(classic
mode). Because MapReduce version 1 (MRv1) is deprecated for
Data Fabric 6.0 and later, you can no longer submit MRv1
jobs using classic
mode. If you previously configured your client and
cluster nodes to use classic
mode, you must prepare your cluster to run
YARN applications in Data Fabric 6.0 or later. Before
upgrading to Data Fabric 6.0 or later, ensure that the
MapReduce mode is set to yarn
in the environment variable, on client nodes,
and on the cluster.
Configure the MapReduce mode for the following components:
Component | How to Change the Mode to yarn : |
---|---|
Environment Variable |
Set the MapReduce mode in an environment variable:
|
Client | In the hadoop_version file on a Data Fabric
client node, verify the MapReduce mode is set to yarn .
|
Cluster | Run the following command to display the cluster-wide mode: maprcli
cluster mapreduce get . If the cluster is set to classic
mode, use the command line or the Control System to set it back to the
yarn default for all nodes in the cluster:
|
Uninstalling MRv1 Packages
yarn
, uninstall the packages associated with MRv1:mapr-jobtracker
mapr-tasktracker
mapr-metrics
Operating System | Uninstall Command |
---|---|
Ubuntu |
apt-get remove <package name> |
CentOS/RedHat: | yum remove <package name> |
SLES | zypper rm <package name> |