[SPARK-23460] PySpark concurrency python egg cache directory

• Type: Question
• Status: Resolved
• Priority: Trivial
• Resolution: Incomplete
• Affects Version/s: 2.1.2
• Component/s: PySpark
• Labels: bulk-closed pyFiles

Environment

YARN last

• Assignee: Unassigned
• Reporter: Dmitiry
• Created: 18/Feb/18 06:24
• Updated: 21/May/19 04:14
• Resolved: 21/May/19 04:14

Description

We are experiencing intermittent failures when running task on pyspark while installing dependencies through --py-files with python egg. We set (else permission denied on egg cache):

--conf "spark.executorEnv.PYTHON_EGG_CACHE=./.python-eggs"

 

Error:

INFO - File "build/bdist.linux-x86_64/egg/ua_parser/user_agent_parser.py", line 409, in <module>
INFO - File "/usr/lib/python2.7/dist-packages/pkg_resources.py", line 904, in resource_filename
INFO - self, resource_name
INFO - File "/usr/lib/python2.7/dist-packages/pkg_resources.py", line 1380, in get_resource_filename
INFO - return self._extract_resource(manager, zip_path)
INFO - File "/usr/lib/python2.7/dist-packages/pkg_resources.py", line 1405, in _extract_resource
INFO - self.egg_name, self._parts(zip_path)
INFO - File "/usr/lib/python2.7/dist-packages/pkg_resources.py", line 984, in get_cache_path
INFO - self.extraction_error()
INFO - File "/usr/lib/python2.7/dist-packages/pkg_resources.py", line 950, in extraction_error
INFO - raise err
INFO - ExtractionError: Can't extract file(s) to egg cache
INFO - 
INFO - The following error occurred while trying to extract file(s) to the Python egg
INFO - cache:
INFO - 
INFO - [Errno 17] File exists: './.python-eggs'
INFO - 
INFO - The Python egg cache directory is currently set to:
INFO - 
INFO - ./.python-eggs/
INFO - 
INFO - Perhaps your account does not have write access to this directory? You can
INFO - change the cache directory by setting the PYTHON_EGG_CACHE environment
INFO - variable to point to an accessible directory.

 

We create a package with an option `safe_zip=False`. But pyspark whatever use egg cache directory.

Is there any way around this?