Someone in your data center unplugs a slave node by accident. Users of the cluster notice via the
Hadoop Web UI that the cluster size has shrunken and express concerns about data loss and
HDFS performance. The replication factor of all the files in the cluster is unchanged from the
default of 3. What can you tell the users?
The HDFS filesystem is corrupt until the administrator re adds the DataNode to the cluster. The
warnings associated with the event should be reported.
After identifying the outage, the NameNode will naturally re-replicate the data and there will be
no data loss. The administrator can re-add the DataNode at any time. The client can disregard
warnings concerned with this event. Data will be under-replicated but will become properly
replicated over time.
The NameNode will re replicate the data after the administrator issues a special command. The
data is not lost but is underreplicated until the administrator issues this command.
The NameNode will identify the outage and re-replicate the data when the clients receive
connection failures to the DataNode, so the end users can disregard such warnings.