slug: benchmark title: “Benchmarks: TPC-H / TPC-DS / SSB” description: “Generate a TPC dataset, load it into SQE as Iceberg tables, and run the suite’s queries with per-query timings. Uses sqe-bench against a Nessie catalog over RustFS, all in Docker.”

Benchmarks: TPC-H / TPC-DS / SSB

Generate a benchmark dataset, load it into SQE as Iceberg tables, and run the suite’s queries with per-query timings. Everything runs in Docker: a Nessie catalog over RustFS holds the tables, and sqe-bench drives generate, load, and test. The default is TPC-H at scale factor 0.01, which completes in seconds.

How it works

The run has three phases:

1. Generate — sqe-bench generate writes the suite’s tables as Parquet files to a shared Docker volume.
2. Load — sqe-bench load issues one CREATE TABLE … AS SELECT * FROM read_parquet(…) per table. The coordinator reads the Parquet from the shared volume; tables land in the Nessie catalog.
3. Test — sqe-bench test runs every .sql in the suite over Flight SQL and reports pass / fail / error plus a per-query timing table.

sqe-bench is a separate image built from Dockerfile.bench. Both images build from this repo on first run if absent.

What it demonstrates

• All three sqe-bench phases — generate, load, and test — running end to end.
• TPC-H (default, SF0.01): all 22 queries pass with per-query timings.
• Configurable suite (BENCH=ssb, BENCH=tpcds) and scale factor (SCALE=0.1, SCALE=1).
• sqe-bench test output: a pass/fail/diff/skip/error summary and a BENCH_SUMMARY: line for machine parsing.

Status: validated (2026-06-07).

Run it

Full config, docker compose, suite SQL, and captured output are in the repo:

→ quickstart/benchmark/

cd quickstart/benchmark
cp .env.example .env
./run.sh