zhihai xu created MAPREDUCE-6198:
------------------------------------
Summary: NPE from JobTracker#resolveAndAddToTopology in MR1 cause
initJob and heartbeat failure.
Key: MAPREDUCE-6198
URL: https://issues.apache.org/jira/browse/MAPREDUCE-6198
Project: Hadoop Map/Reduce
Issue Type: Bug
Components: mrv1
Reporter: zhihai xu
Assignee: zhihai xu
NPE from JobTracker#resolveAndAddToTopology in MR1 cause initJob and heartbeat
failure. The NPE is caused by dnsToSwitchMapping.resolve return null at the
following:
{code}
List <String> rNameList = dnsToSwitchMapping.resolve(tmpList);
String rName = rNameList.get(0);
{code}
I check the code in MR2, MR2 handle it correctly in coreResolve of
RackResolver.java
{code}
List <String> rNameList = dnsToSwitchMapping.resolve(tmpList);
String rName = null;
if (rNameList == null || rNameList.get(0) == null) {
rName = NetworkTopology.DEFAULT_RACK;
LOG.info("Couldn't resolve " + hostName + ". Falling back to "
+ NetworkTopology.DEFAULT_RACK);
} else {
rName = rNameList.get(0);
LOG.info("Resolved " + hostName + " to " + rName);
}
{code}
We should do the same in MR1, if dnsToSwitchMapping.resolve return null, use
NetworkTopology.DEFAULT_RACK.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
