I recently updated from 1.0.4 to 2.0.5. Since then, streaming jobs have
been failing to launch due to what seems like an incorrect staging path:
# /opt/hadoop2/bin/hadoop jar
/opt/hadoop2/share/hadoop/tools/lib/hadoop-streaming-2.0.5-alpha.jar
-input foo -output bar -mapper baz -reducer foobar
13/08/01 10:43:50 WARN util.NativeCodeLoader: Unable to load
native-hadoop library for your platform... using builtin-java classes
where applicable
13/08/01 10:43:50 WARN conf.Configuration: session.id is deprecated.
Instead, use dfs.metrics.session-id
13/08/01 10:43:50 INFO jvm.JvmMetrics: Initializing JVM Metrics with
processName=JobTracker, sessionId=
13/08/01 10:43:50 INFO jvm.JvmMetrics: Cannot initialize JVM Metrics
with processName=JobTracker, sessionId= - already initialized
13/08/01 10:43:50 INFO mapreduce.JobSubmitter: Cleaning up the staging
area *file:/user/myuser1544460269/.staging/job_local1544460269_0001*
13/08/01 10:43:50 ERROR security.UserGroupInformation:
PriviledgedActionException as:myuser (auth:SIMPLE)
cause:org.apache.hadoop.util.Shell$ExitCodeException: chmod: cannot
access `*/user/myuser1544460269/.staging/job_local1544460269_0001*': No
such file or directory
13/08/01 10:43:50 ERROR streaming.StreamJob: Error Launching job :
chmod: cannot access
`/user/myuser1544460269/.staging/job_local1544460269_0001': No such file
or directory
Streaming Command Failed!
This is for a job launched as "myuser". Given that
mapreduce.jobtracker.staging.root.dir is set to /user, I would expect
the staging area to be in /user/myuser/.staging/job_local$jobid.
Instead, it is in /user/myuser$jobid/.staging/job_local$jobid, which
fails since /user/myuser$jobid/ doesn't exist. Has the way
staging.root.dir is used changed in 2.x?
Thank you,
Pierre
