[SPARK-1850] Bad exception if multiple jars exist when running PySpark

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 1.0.0
• Fix Version/s: 1.0.1
• Component/s: PySpark
• Assignee: Unassigned
• Reporter: Andrew Or
• Created: 15/May/14 19:39
• Updated: 05/Nov/14 10:43
• Resolved: 02/Jul/14 16:53

Description

Found multiple Spark assembly jars in /Users/andrew/Documents/dev/andrew-spark/assembly/target/scala-2.10:
Traceback (most recent call last):
  File "/Users/andrew/Documents/dev/andrew-spark/python/pyspark/shell.py", line 43, in <module>
    sc = SparkContext(os.environ.get("MASTER", "local[*]"), "PySparkShell", pyFiles=add_files)
  File "/Users/andrew/Documents/dev/andrew-spark/python/pyspark/context.py", line 94, in __init__
    SparkContext._ensure_initialized(self, gateway=gateway)
  File "/Users/andrew/Documents/dev/andrew-spark/python/pyspark/context.py", line 180, in _ensure_initialized
    SparkContext._gateway = gateway or launch_gateway()
  File "/Users/andrew/Documents/dev/andrew-spark/python/pyspark/java_gateway.py", line 49, in launch_gateway
    gateway_port = int(proc.stdout.readline())
ValueError: invalid literal for int() with base 10: 'spark-assembly-1.0.0-SNAPSHOT-hadoop1.0.4-deps.jar\n'

It's trying to read the Java gateway port as an int from the sub-process' STDOUT. However, what it read was an error message, which is clearly not an int. We should differentiate between these cases and just propagate the original message if it's not an int. Right now, this exception is not very helpful.

Issue Links

• is part of: SPARK-2242 Running sc.parallelize(..).count() hangs pyspark