delta‑rs–generated checkpoint that Spark serverless cannot deserialize because serverless use restricted/optimized parquet reader(Serverless does not allow disabling the vectorized Parquet reader)! delta‑rs uses Apache Arrow + Rust Parquet writers, which Are 100% valid Parquet, but can emit encodings that Spark serverless doesn’t accept for Delta checkpoints

Parquet column cannot be converted” typically means reader‑side incompatibility, not corruption

If this is okay/possible: Disable checkpoint creation in delta‑rs (best workaround)
Otherwise, you might have to use Job Compute cluster instead of serverless