Aaron Glahe created SPARK-9235:
----------------------------------
Summary: PYSPARK_DRIVER_PYTHON env variable is not set on the YARN
Node manager acting as driver when yarn-cluster mode
Key: SPARK-9235
URL: https://issues.apache.org/jira/browse/SPARK-9235
Project: Spark
Issue Type: Bug
Components: PySpark
Affects Versions: 1.4.1, 1.5.0
Environment: CentOS 6.6, python 2.7, Spark 1.4.1 tagged version, YARN
Cluster Manager, CDH 5.4.1 (Hadoop 2.6.0++), Java 1.7
Reporter: Aaron Glahe
Priority: Minor
Relates to SPARK-9229
Env: Spark on YARN, Java 1.7, Centos 6.6, CDH 5.4.1 (Hadoop 2.6.0++), Anaconda
Python 2.7.10 "installed" in /srv/software directory
On a client/submitting machine, we set the PYSPARK_DRIVER_PYTHON env var in
spark-env.sh that pointed the anaconda python executable, which was on every
YARN node:
export PYSPARK_DRIVER_PYTHON='/srv/software/anaconda/bin/python'
side note, export PYSPARK_PYTHON='/srv/software/anaconda/bin/python' was set as
well in the spark-env.sh.
run the command:
spark-submit test.py --master yarn --deploy-mode cluster
It appears as though the Node Manager with the DRIVER does not use the
PYSPARK_DRIVER_PYTHON env python, but instead uses the CentOS system default
(which in this case is python 2.6).
Workaround appears to setting the python path in the SPARK_YARN_USER_ENV
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
