[HUDI-5384] Make sure predicates are appropriately pushed down to HoodieFileIndex when lazy listing

• Type: Bug
• Status: Closed
• Priority: Blocker
• Resolution: Fixed
• Affects Version/s: 0.13.0
• Fix Version/s: 0.13.0
• Component/s: spark, spark-sql
• Labels: pull-request-available
• Story Points: 1
• Epic Link: Hudi Spark Datasource
• Assignee: Alexey Kudinkin
• Reporter: Alexey Kudinkin

Reviewers

sivabalan narayanan

• Created: 13/Dec/22 21:36
• Updated: 20/Jan/23 21:52
• Resolved: 20/Jan/23 21:52

Description

After introduction of lazy-listing capability in HUDI-4812, it exposed an issue in Spark's design, where predicates are pushed-down into generic FileIndex implementations only during the execution phase.

This poses following issues:

1. HoodieFileIndex isn't listing the table until `listFiles` method is invoked
2. Listing would actually be performed only during actual execution in `FileSourceScanExac` node
3. Since listing isn't performed until the actual execution, table statistics are initialized w/ bogus values (of 1 byte) and Cost-based Optimizations (CBO) will be taking incorrect decisions based on that

Issue Links

• links to: GitHub Pull Request #7423