cshuo opened a new issue, #19372:
URL: https://github.com/apache/hudi/issues/19372

   ### Parent
   
   Sub-issue of #14310.
   
   ### Problem
   
   Flink metadata tables still use the default storage layout even when the 
data table is on table version 10 and can use the LSM/native-log path. As a 
result, metadata-table reads—especially frequent lookups against the `files` 
partition—continue to use the classic file-group reader instead of the sorted 
LSM reader.
   
   ### Proposed work
   
   - Persist `hoodie.table.storage.layout=lsm_tree` when creating a Flink 
metadata table for table version 10 or later.
   - Keep other engines, older table versions, and existing healthy metadata 
tables unchanged.
   - Use `HoodieLsmFileGroupReader` for LSM metadata-table file slices 
containing only native logs.
   - Fall back to `HoodieFileGroupReader` for the default layout or slices 
containing legacy inline logs.
   - Do not use the reusable record-buffer cache on the LSM path in this first 
phase; preserve the existing reusable-reader behavior for the classic path.
   
   ### Acceptance criteria
   
   - A newly created Flink v10 metadata table persists the `lsm_tree` layout 
and initializes its file group with a native HFile log.
   - `files` partition point and batch lookups return correct results for 
native LSM slices.
   - Enabling or disabling metadata-reader reuse does not change LSM query 
results, and the LSM path does not initialize the reusable cache.
   - Default-layout and legacy/mixed-log slices continue to use the classic 
reader.
   - No migration is introduced for existing metadata tables.
   


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to