[DNM] benchmarks against object storage

[a10y]

I'm using this PR as a space to collect some info about running the TPC-H queries against object storage. Goals are to compare

• Vortex
• Parquet

Against storage backends

• S3
• S3 Express One Zone
• Google Cloud Storage

Changes

This PR creates a new binary that runs every TPC-H query while logging IOs in our objectstore reader, allowing us to examine both request sizes and request counts for each query.

Parquet and Vortex are each selectable, and the bucket is also configurable.

To run the test that uses S3 Express One Zone, you need to set AWS_S3_EXPRESS=true in your .env or directly in your shell environment

[a10y]

Initial results

Attaching two zips, one with TRACE-level logs of executing all TPC-H queries (except q15) using the Vortex Datafusion provider.

s3express_vortex.zip s3_vortex.zip

Some interesting bits:

Total number of IO's to perform each query:

Total time to execute the query (not including table registration)

S3 normal:

S3 Express One: