DuckDB on Azure: direct az:// vs S3-compatible proxy
Summary:
A reproducible benchmark comparing two ways of running DuckDB against Azure Blob Storage: the native azure extension (az://) versus fronting the same storage account with a properly tuned s3proxy and using httpfs (s3://). Results across four data sizes (5k → 2M rows) show that the s3proxy path wins clearly for the DuckLake-relevant patterns we care about most: partitioned writes, warm reopens, and larger cold scans. The main exception is selective reads with predicate pushdown, where direct Azure wins. A later rerun at 4M rows also showed a crossover where direct Azure became faster for one very large single-file write.