AnsweredAssumed Answered

Recovery time when a node in MapR cluster is failed

Question asked by jennie on May 28, 2012
Latest reply on May 29, 2012 by tomer
Our team built up MapR M5 cluster using 6 nodes.

I tested how cluster respond when any node is failed in cluster.

For testing this, we used NFS and made IO operation between client and Hadoop cluster.

During this IO operation, I turned off name node or data node .
I found out IO  operation is suspended.
(In name node was about 6 minutes , in data node was about 16 minutes) 

When a name node or  a data node is failed, is Hadoop cluster possible  not to be influenced ? 
Even being influenced, is there any solution to reduce recovery time ?

Our detailed test configuration like below.
[1] Client : desktop which installed centos
[2] Hadoop cluster
    [2.1] node 1 : CLDB, zookeeper, job tracker, nfs, task tracker, file server, web server,
    [2.2] node 2, 3 : CLDB, zookeeper, job tracker, nfs, task tracker, file server
    [2.3] node 4, 5, 6 : nfs, task tracker, file server

Outcomes