When you shut down a cluster, follow this sequence to preserve your data and replication:
- Verify that recent data has finished processing.
- Shut down any NFS servers.
- Shut down any ecosystem components that are running.
- Shut down the job and task trackers.
- Shut down the ResourceManager and NodeManager services.
- Shut down Warden on all nodes that are not running CLDB.
- Shut down Warden on the CLDB nodes.
- Shut down ZooKeeper on the ZooKeeper nodes.
To shut down the cluster:
Before you start, make sure there are no active MapReduce or HBase processes, and that no data is being loaded to the cluster or being persisted within the cluster.
- Change to the
rootuser (or use
sudofor the following commands).
- Before shutting down the cluster, you will need a list of NFS nodes, CLDB nodes, and all remaining nodes. Once the CLDB is shut down, you cannot retrieve a list of nodes; it is important to obtain this information at the beginning of the process. Use the
node listcommand as follows:
Determine which nodes are running the NFS gateway. Example:
Determine which nodes are running the CLDB. Example:
List all non-CLDB nodes. Example:
Shut down all NFS instances. Example:
- If your cluster is running any ecosystem components, shut down those components on all nodes.
Shut down all JobTracker and TaskTracker services on all nodes. Example:
If you are running in YARN mode, shut down all ResourceManager and NodeManager services on all nodes. To shut down ResourceManager and NodeManager services, specifiy the
maprcli node servicescommand with the
nameparameter and either the
SSH into each node that is not running CLDB and stop Warden. Example:
SSH into each CLDB node and stop Warden. Example:
SSH into each Zookeeper node and stop Zookeeper. Example:
- (Optional) Shut down the nodes using the Linux