[Core] Qualification Tool for Gluten

[srinivasst]

Description

The purpose of this enhancement is to create a qualification tool that analyzes customer event files to determine which workloads are suitable for execution with Gluten. This is crucial when onboarding new customers, as not all workloads benefit from Gluten's native acceleration—especially workloads with RDD operations, unsupported SQL operators or UDF workloads.

Proposed Solution:

Develop a Java program to analyze the event files, given a Hadoop file path as input. The program will generate two reports:

• Application Report:
  • Percentage of RDD usage
  • Percentage of Unsupported SQL operations
  • Percentage of supported SQL operations
  • Cumulative task time for each application
  • Recommendation to use Gluten Acceleration (Recommended if Percentage of supported SQL operations >= 70%)
• Unsupported Operator Report:
  • Unsupported SQL operators
  • Impact on cumulative CPU time

Requirements:

• Compatibility with Hadoop file paths that point to:
  • Single event files
  • Event directories with rolling event files
  • Deeply nested directories containing event files
  • Compressed event files

[srinivasst]

WIP - will upload approach and PR in a few days.

[xumingming]

Interesting idea, Looking forward to it.