tom-s-powell opened a new pull request, #18011:
URL: https://github.com/apache/iceberg/pull/18011

   ## Summary
   
   Avoid eagerly caching and counting the orphan-file dataset when 
`PrefixMismatchMode` is `IGNORE` or `DELETE`. The default `ERROR` behaviour 
remains unchanged.
   
   ## Why
   
   `findOrphanFiles` previously cached and counted the complete joined dataset 
for every prefix mismatch mode. This eager Spark job is required for `ERROR`, 
where all scheme and authority conflicts must be collected before any files are 
deleted.
   
   `IGNORE` and `DELETE` do not inspect the conflict accumulator. Eagerly 
materializing the dataset adds unnecessary work and cache pressure, which is 
particularly expensive for tables with very large file listings.
   
   ## How it works
   
   For `IGNORE` and `DELETE`, `findOrphanFiles` now returns the lazy dataset 
immediately after constructing the join and partition mapping.
   
   For `ERROR`, it retains the existing cache and count operation so conflicts 
are detected before deletion begins.
   
   Regression tests use a Spark accumulator to verify that `IGNORE` and 
`DELETE` do not evaluate the input dataset until the result is consumed. The 
change is applied to Spark 3.5, 4.0, 4.1, and 4.2.
   
   ## Testing
   
   - Ran the lazy evaluation and prefix conflict tests on all four Spark 
versions: 18/18 passed per version.
   - Ran the complete Spark 4.0 orphan-file action test class: 175 passed and 
23 skipped.
   - Ran `spotlessCheck` on all affected modules.
   


-- 
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]


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to