There is no service in the experiment infra that can initialize a ClickHouse schema and bulk-load a dataset (ClickBench, H2O) from a prepared file as part of an experiment run. Currently this is done manually via benchmark/export_to_database.py outside the experiment runner.