02-19-2015 12:47 PM
Assuming you have multiple data nodes and a replication factor greater than 1, you should not experience any outage if a single DN restarts. Redundancy and data availability are built into the HDFS Architecture. The caveat to this is that really small test clusters, like 2 or 3 datanodes, or if you have HDFS block replication set to 1 or 2, you could be subject to data loss if a single DN goes down. If you correctly architect your cluster, a single DN dropping or restarting is just another day in normal operation for HDFS.