steFaiz opened a new pull request, #8380:
URL: https://github.com/apache/paimon/pull/8380

   ### Purpose
   The first part of https://github.com/apache/paimon/issues/8322
   
   As discussed in  https://github.com/apache/paimon/pull/8344, this PR tries 
an another way: use the oldest file as the anchor file to lookup DVs. This 
approach shifts the complicity from core module to engines module e.g. 
Spark/Flink/Ray
   
   Key design:
   1. During `DataEvolutionFileStoreScan`, in post filter, keep the anchor file 
even it's not needed.
   2. Still use `_ROW_ID` to filter out deleted rows. (Not returned positions) 
. Convert the Deletion Files to Map<Range, DeletionVector> in `DataSplit`. This 
is because:
      a. Following the `IndexedSplitRecordReader` implementation.
      b. returned positions is the concept of a single file. For blobs and 
vectors, many files are concated (for blobs there even may be placeholders and 
row range holes). Maintaining returned positions across many concated (and 
maybe uncontagious) files is complicated and break its semantic (returned 
position should be within a single file)
   
   ### Tests
   ITCases only.


-- 
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