schenksj opened a new issue, #4520: URL: https://github.com/apache/datafusion-comet/issues/4520
### Why this is needed (surfaced by the Delta integration) When a Delta table is opened with custom Hadoop filesystem options — `DeltaTable.forPath(spark, path, fsOptions)` against a custom scheme such as `fake://` — Delta internally reads its own `_delta_log/*.checkpoint.parquet` files via ordinary V1 parquet scans. `CometScanRule` claims those scans for native execution, but Comet's native reader goes through `object_store`, which doesn't know the custom scheme, so it crashes at execution with `Generic URL error: Unable to recognise URL "fake:///..."`. There is no graceful recovery once native execution has started — the scan must be declined at planning time so Spark's Hadoop-FS-aware reader handles it. Concretely this blocks `DeltaTableSuite "dropFeatureSupport - with filesystem options"`. ### Also a baseline fix (standalone) This isn't Delta-specific in mechanism: *any* V1 parquet scan on a filesystem `object_store` can't parse crashes Comet's native reader the same way on `main` today, rather than falling back. So the fix stands on its own as general robustness for custom Hadoop `FileSystem` schemes. ### Proposed implementation Decline the scan at planning when its root-path scheme isn't natively readable. Rather than hardcode the object_store-supported scheme set in the planner (a mirror that drifts — e.g. `gcs`/`oss` aren't object_store schemes, while `http`/`az`/`memory` are), answer from object_store itself via a small JNI helper (`NativeBase.isObjectStoreSchemeSupported`, backed by `ObjectStoreScheme::parse` — the same path `prepare_object_store_with_configs` uses). Union in the user's libhdfs scheme config (`spark.hadoop.fs.comet.libhdfs.schemes`) on the JVM side; cache per scheme; assume supported if native can't be consulted. ### Priority Standalone, but **required by the in-progress `contrib/delta` integration** (which carries this change today). Extracting it lets it land independently and the Delta work rebase onto it — please prioritize accordingly. ### Tests - `ParquetReadFromFakeHadoopFsSuite` (libhdfs scheme regression guard) passes. - Add a custom-scheme decline test (`fake://` → falls back to Spark, not a crash). -- 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]
