[SPARK-34438] Python Driver is not correctly detected using presigned URLs

• Type: Bug
• Status: In Progress
• Priority: Minor
• Resolution: Unresolved
• Affects Version/s: 3.0.0, 3.0.1, 3.0.2, 3.1.0
• Component/s: Spark Submit
• Assignee: Unassigned
• Reporter: Julian Fleischer
• Created: 15/Feb/21 01:51
• Updated: 15/Feb/21 02:08

Description

In AWS one can generate so-called presigned URLs. spark-submit accepts URLs for the driver program, e.g. http://my-web-server/driver.py. Now a presigned URL has a query fragment http://my-web-server/driver.py?signature.

Now the check for whether the given URL is a python driver simply checks whether it ends in .py – which the presigned URL does not, as it ends in signature.

The relevant check is in SparkSubmit.scala, Line 1051 (commit tagged v3.0.1):

https://github.com/apache/spark/blob/v3.0.1/core/src/main/scala/org/apache/spark/deploy/SparkSubmit.scala#L1051 

Here is a more realistic example URL:

https://bucket-name.s3.us-east-1.amazonaws.com/driver.py?X-Amz-Algorithm=AWS4-HMAC-SHA256&X-Amz-Credential=AKIATBNPKWPCNUMWMLUR%2F20210214%2Fus-east-1%2Fs3%2Faws4_request&X-Amz-Date=20210214T062047Z&X-Amz-Expires=172800&X-Amz-SignedHeaders=host&X-Amz-Signature=49ef39b6bb7090001af9312692788892551916a6ac0ff6c961ce52efb9acc235

A fix could be to parse the the given path as a java.net.URI and look for the pathname to end in .py (as opposed to the whole thing).

To circumvent this issue I am currently appending a fragment to the query which makes it end in .py, i.e. http://my-web-server/driver.py?signature#.py which does work.

Issue Links

• links to: [Github] Pull Request #31565 (scravy)
• links to: [Github] Pull Request #31565 (scravy)