Utility to enforce current replication policy on already exitsting blocks
-------------------------------------------------------------------------
Key: HADOOP-3211
URL: https://issues.apache.org/jira/browse/HADOOP-3211
Project: Hadoop Core
Issue Type: New Feature
Reporter: Robert Chansler
If the replica placement policy changes, or if the cluster topology changes,
existing block replicas will not be deliberately moved to implement the new
policy or to reflect the new topology. The re-replicator and the re-balancer
always implement the current policy, but a healthy block may never attract the
attention of either of those facilities.
In practice, all replicas of a block have been found within a single rack (as
allowed by the previous _random rack_ placement policy. The {{fsck}} facility
reports such examples.
The re-placer utility could follow the model of the re-balancer. Or even be
part of the re-balancer.
An alternative implementation would scan the {{fsck}} report for curious
blocks, and rename a replica block file on the data node, so as to hide it. The
re-replicator would be expected to create a new replica properly placed. (This
is a fix up that any administrator could implement.)
(1761414)
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
