[HUDI-8544] Fix RLI and SI MDT record generation to be consistent w/ spark task retries

• Type: Improvement
• Status: Closed
• Priority: Blocker
• Resolution: Fixed
• Fix Version/s: 1.0.0
• Component/s: metadata, writer-core
• Labels: blocker
• Assignee: sivabalan narayanan
• Reporter: sivabalan narayanan
• Created: 18/Nov/24 19:34
• Updated: 28/Nov/24 05:48
• Resolved: 28/Nov/24 05:48

Description

Very rarely we could have some inconstencies when RLI and SI record generation since it relies on RDD<WriteStatus> while all other MDT partitions relies on List<HoodieWriteStat> which is already computed and is in drivers memory. 

 

So, wanted to fix RLI and SI for now. Eventually we wanted to go w/ a full dag rewrite to be streaming friendly. 

 

Sub-Tasks

• Fix RLI to do on-demand lookup of data files to generate RLI records for MDT: Closed
• Fix SI record generation to do on-demand lookup instead of relying on RDD<WriteStatus>: Closed
• Update Metadata table API to remove WriteStatus reference: Closed