[LIVY-859] Start PySpark application Failed via sparkmagic with Python 3.7.6

• Type: Question
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Fix Version/s: 0.9.0
• Environment: CDH: 6.3.2 ; Spark: 2.4.0 ; Python: 3.7.6 ; Livy: 0.7.0-incubating
• Assignee: Unassigned
• Reporter: Eric Luo
• Created: 21/May/21 02:28
• Updated: 12/Nov/22 22:10

Description

I have two CDH clusters (stage and prod) with the same environment, the same Livy service installed on them, a Python 3.7 environment configured via Ansible, and I write PySpark by calling Livy via sparkmagic in jupyter-lab. code in jupyter-lab, the stage environment works fine, but the prod environment gives an error.

Error log:

 

21/05/20 14:43:45 INFO driver.SparkEntries: Created Spark session (with Hive support).
21/05/20 14:43:50 ERROR repl.PythonInterpreter: Process has died with 134
21/05/20 14:43:50 ERROR repl.PythonInterpreter: Fatal Python error: initfsencoding: unable to load the file system codec
ModuleNotFoundError: No module named 'encodings'

 

On a prod environment machine, I actually have PYSPARK_PYTHON (/opt/miniconda/bin/python) configured to ``import encodings``, and I can run pyspark SHELL directly without any problems.

My environment configuration.

CDH: 6.3.2
Spark: 2.4.0
Python: 3.7.6

In CDH's "Spark Service Advanced Configuration Snippet (Safety Valve) for spark-conf/spark-env.sh" configuration, I configured PYSPARK_PYTHON and PYSPARK_ DRIVER_PYTHON

```
export PYSPARK_PYTHON=${PYSPARK_PYTHON:-/opt/miniconda/bin/python}
export PYSPARK_DRIVER_PYTHON=${PYSPARK_DRIVER_PYTHON:-/opt/miniconda/bin/python}
```

livy-env.sh also configures PYSPARK_PYTHON and PYSPARK_DRIVER_PYTHON

```
PYSPARK_PYTHON=/opt/miniconda/bin/python
PYSPARK_DRIVER_PYTHON=/opt/miniconda/bin/python

JAVA_HOME=/usr/java/jdk1.8.0_181-cloudera/
HADOOP_HOME=/opt/cloudera/parcels/CDH/lib/hadoop
HADOOP_CONF_DIR=/etc/hadoop/conf
SPARK_HOME=/opt/cloudera/parcels/CDH/lib/spark
```