andygrove opened a new issue, #6140: URL: https://github.com/apache/datafusion-comet/issues/6140
### Describe the bug The native Iceberg write gate and the native storage factory read the scheme of a data location differently, so some locations pass the gate and then fail at execution. - The gate (`CometIcebergNativeWrite.storageScheme`, `CometIcebergNativeWrite.scala:310-315`) takes the text before `://`, and treats a location without `://` as `file`. - The native side (`scheme_of`, `iceberg_common.rs:229`) takes the text before the first `:`. Hadoop normalises `hdfs:///warehouse/t` to `hdfs:/warehouse/t`. For that location the gate sees no `://`, calls it `file` and admits it. The native writer reads `hdfs`, which has no storage backend, and every task fails with `Unsupported storage scheme: hdfs`. The query fails instead of falling back to iceberg-java. The same applies to any `scheme:/path` form. ### Steps to reproduce Create an Iceberg table whose data location is written `hdfs:/path` (for example through `LOCATION 'hdfs:/...'`, or a Hadoop catalog whose warehouse was normalised), enable `spark.comet.iceberg.write.enabled`, and insert into it. ### Expected behavior The write falls back to iceberg-java with an `unsupported storage scheme: hdfs` reason. The gate should use the same scheme rule as `scheme_of`, ideally one shared implementation. ### Additional context #6065 (closes #5541) moves the supported-scheme lists to the native factory and touches `storageScheme`, but it keeps the `://` split, so it does not fix this. It would be a natural place to align the parsing too. Found in an audit of the native Iceberg write path before enabling it by default. Part of #5649. -- 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]
