This is the documentation for CDH 4.7.0.
Documentation for other versions is available at Cloudera Documentation.

Known Issues Fixed in CDH4.3.0

Apache MapReduce

— In CDH4 MRv1, the mapreduce.jobtracker.split.metainfo.maxsize property may not get applied properly at the JobTracker

In CDH4 MRv1, the JobTracker provides the mapreduce.jobtracker.split.metainfo.maxsize property that controls the maximum split-metadata size of jobs users can submit to prevent memory crashes at the JobTracker. Users who have overridden this property's defaults for various limit-changing purposes (expansion or lowering) may need to make a further change to enforce this limit in CDH4 MRv1 because a bug in the JobTracker may allow its value to be overridden by the job's own configuration.

Bug: None

Severity: Low

Resolution: Fixed in CDH4.3

Workaround: Add the following property to the JobTracker's mapred-site.xml, and make sure it is marked final. Cloudera Manager users can add the same property in the JobTracker Safety Valve for mapred-site.xml:
 <property>
 <name>mapreduce.jobtracker.split.metainfo.maxsize</name>
 <value>(Administrator Desired Value)</value>
 <final>true</final>
</property> 

Apache HBase

— Old zero-byte replication log files are not deleted

In all CDH4.2.x releases, old zero-byte replication log files are not deleted.

Bug: None

Severity: Low

Resolution: Fixed in CDH4.3

Workaround: None

Hue

— Beeswax authentication fails in a secure cluster if the domain set in hive.metastore.uris does not match the Kerberos realm's domain

When Hue is configured to use a remote metastore, Beeswax looks at hive-site.xml to understand how to start Beeswax and communicate with the metastore. In a secure cluster, if the hive.metastore.uris parameter is set to anything other than the fully qualified domain name (FQDN) that the Kerberos realm is assigned to, authentication will fail.

Bug: HUE-1045

Severity: Low

Resolution: Fixed in CDH4.3

Workaround: Change the hive.metastore.uris configuration parameter in hive-site.xml to use the fully qualified domain name. Make sure the Kerberos realm that the Hive Kerberos principal belongs to is assigned to the same fully qualified domain name.

Apache Flume

—Avro sink property incorrectly named

In the Flume User Guide, the Avro sink property reset-connection-interval was incorrectly referred to as connection-reset-interval.

Bug: FLUME-1979

Severity: Low

Resolution: Fixed in CDH4.3

Workaround: None

Apache Oozie

— Oozie may show jobs as RUNNING when they are not

Restarting the JobTracker while a workflow is running two or more actions in a fork should cause the workflow and all of its running actions to be listed as FAILED, but in fact only the workflow and one of the actions are listed as FAILED; the rest show as RUNNING even though they are not actually running.

Bug: OOZIE-1205

Severity: Low

Resolution: Fixed in CDH4.3.0

Workaround: The false RUNNING status should be harmless, but killing the FAILED workflows will change any RUNNING status to KILLED.