peterxcli opened a new pull request, #5888: URL: https://github.com/apache/datafusion-comet/pull/5888
## Which issue does this PR close? Closes #4992. ## Rationale for this change Iceberg append micro-batches and batch changelog queries cannot currently use Comet throughout their data-processing path. This adds experimental native execution while reusing Spark/Iceberg planning and recovery. ## What changes are included in this PR? - Read Spark 4.x Iceberg micro-batches using the source's offset-bounded tasks, preserving progress and checkpoint restart. Enable compatible native aggregate computation around Spark state-store boundaries. - Read `.changes` tasks with the existing native Iceberg reader and process carry-over removal, update images, and net changes in Rust through `create_changelog_view`. - Add opt-in configuration, documentation, and CI coverage. Retain fallback for unsupported scans, state buffers, CDC value types, and procedure closure layouts. Spark retains offsets, persistent state, and sink commits. Streaming joins/deduplication remain in Spark. Batch CDC retains Iceberg's restriction on MOR changes involving delete manifests; this does not add a streaming CDC source. ## How are these changes tested? 17 integration tests passed against Polaris and RustFS with Spark 4.1.3 / Iceberg 1.11.0. They assert native execution and Spark-equivalent results for COW V1/V2/V3 CDC, snapshot/timestamp bounds, integrity queries, and streaming checkpoint/retry behavior. The same 17 tests passed with the local catalog after rebasing onto current main. Three Rust CDC tests, Clippy, Scala formatting, and style checks also passed. -- 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]
