Known Issues Fixed In CDH4.2.1
For a list of the upstream fixes applied in CDH4.2.1, see What's New in CDH4.2.1
— Jobs fail in MRv2 (YARN) when run on medium or larger-sized data sets
A bug in the merge phase causes Reduce tasks to fail on YARN whenever their input data exceeds memory and must spill to disk.
Resolution: Fixed in CDH4.2.1
— With replication enabled, Region Server may abort
When Replication is enabled, a Region Server could abort during a failover of an aborted Region Server.
Resolution:Fixed in CDH4.2.1.
Workaround:To avoid this problem, set hbase.zookeeper.useMulti to false in the conf/hbase-site.xml, as follows:
<property> <name>hbase.zookeeper.useMulti</name> <value>false</value> </property>