Details
-
Type:
Bug
-
Status: Resolved
-
Priority:
Minor
-
Resolution: Won't Fix
-
Affects Version/s: None
-
Fix Version/s: None
-
Component/s: None
-
Labels:None
-
Environment:
Mac OS X 10.7, Java 1.6.0_26
Description
When starting Hadoop on OS X 10.7 ("Lion") using start-all.sh, Hadoop logs the following errors:
2011-07-28 11:45:31.469 java[77427:1a03] Unable to load realm info from SCDynamicStore
Hadoop does seem to function properly despite this.
Issue Links
- is cloned by
-
HADOOP-8719
Workaround for kerberos-related log errors upon running any hadoop command on OSX
-
- Resolved
-
- is related to
-
PIG-2740
get rid of "java[77427:1a03] Unable to load realm info from SCDynamicStore" log lines when running pig tests
-
- Closed
-
Activity
- All
- Comments
- Work Log
- History
- Activity
- Transitions
Having exactly the same problem on the OS X 10.7.1/JDK 1.6.0_26, Hadoop 0.20.204.0, pseudo-distributed mode.
NameNode daemon seems fail to start - run "jps" to verify. Also "stop_all.sh" outputs "no namenode to stop"
Please disregard previous comment about NameNode daemon not starting - it was a separate issue not related to this specific problem.
I think the OS X JDK is trying to get some information from configd which is failing that causes the whole LoginContent to break.
This can be worked around by specifying kerberos config on the command line:
-Djava.security.krb5.realm=OX.AC.UK -Djava.security.krb5.kdc=kdc0.ox.ac.uk:kdc1.ox.ac.uk
What realm do we use for a standalone machine?
For a standalone machine, you can just set the values to nothing.
Adding the following to your environment or hadoop-env.sh should suppress the errors:
export HADOOP_OPTS="-Djava.security.krb5.realm= -Djava.security.krb5.kdc="
Java has its own internal version of Kerberos. That version is very, very stupid in 1.6 and earlier when it comes to using naming services for auto-discovery of the realm and KDC information. You'll see similar weirdness even on non-OS X boxes when the krb5.conf doesn't explicitly list the realm information. The same configuration fix mentioned here applies there as well. This has been fixed in JRE 1.7. Allegedly.
I'm closing this as Won't Fix since this is a known JRE bug and not particularly anything we can do about it in Hadoop.
I'm also seeing this on another project (on OS X 10.7) and it's happening when I attempt to perform a login using a keytab file. No idea why yet.