MatrixOne BranchBench

Run five BranchBench macrobenchmark workflows against MatrixOne 4.2.4: software_dev, failure_repro, data_cleaning, mcts, and simulation. This MatrixOne-only adaptation is based on dengn/matrixone-branchbench commit 95c18b7bf7a51029e3344a65e807df07aa28a2ea.

The microbench/ directory is an internal dependency of the macrobenchmark runner. Generated protobuf modules are included; no compilation is needed.

Requirements

• Python 3.11+, a MySQL-compatible CLI, and a dedicated MatrixOne 4.2.4 instance
• Enough resources for the selected scale; full simulation uses 1,000 workers

python3 -m venv .venv
source .venv/bin/activate
python -m pip install -r requirements-mo-v424.txt

export MO_HOST=127.0.0.1 MO_PORT=6001 MO_USER=root
export MO_PASSWORD='<your MatrixOne password>'

The driver also accepts --host, --port, --user, and --password. Keep credentials out of the repository.

Functional smoke run

Use an isolated instance. For example, start the standalone image locally:

docker run -d --name mo-branchbench-smoke \
  -p 127.0.0.1:16001:6001 matrixorigin/matrixone:4.2.4
export MO_PORT=16001

Wait for the SQL endpoint, then set MO_PASSWORD to that instance's password (the standalone image used in our local test used 111). Load the small seed:

MYSQL_PWD="$MO_PASSWORD" mysql -h "$MO_HOST" -P "$MO_PORT" -u "$MO_USER" \
  -e 'CREATE DATABASE bb_seed_smoke;'
MYSQL_PWD="$MO_PASSWORD" mysql -h "$MO_HOST" -P "$MO_PORT" -u "$MO_USER" \
  bb_seed_smoke < scripts/mo_v424_seed/ch_benchmark_seed_mo_smoke.sql

python scripts/run_mo_v424.py --scale smoke --seed-db bb_seed_smoke

Smoke mode runs all five workflows with one step per worker and one or two workers. It checks real branch and SQL operations; it is not an SF100 or SF1000 performance result.

SF100 / SF1000

Load the matching seed SQL into a new database. For SF100:

MYSQL_PWD="$MO_PASSWORD" mysql -h "$MO_HOST" -P "$MO_PORT" -u "$MO_USER" \
  -e 'CREATE DATABASE bb_seed_sf100;'
MYSQL_PWD="$MO_PASSWORD" mysql -h "$MO_HOST" -P "$MO_PORT" -u "$MO_USER" \
  bb_seed_sf100 < scripts/mo_v424_seed/ch_benchmark_seed_mo_sf100.sql

python scripts/run_mo_v424.py --scale sf100 --seed-db bb_seed_sf100 --plan
python scripts/run_mo_v424.py --scale sf100 --seed-db bb_seed_sf100 --reps 1

For SF1000, use ch_benchmark_seed_mo_sf1000.sql, a separate seed database, and --scale sf1000. To select workflows, repetitions, runtime cap, and output:

python scripts/run_mo_v424.py --scale sf100 --seed-db bb_seed_sf100 \
  --workflows software_dev simulation --reps 3 --timeout 7200 \
  --output results/my-run

--plan prints generated configurations without connecting to MatrixOne. Full workflow parameters are in scripts/mo_v424_configs/; full simulation uses 1,000 workers and one step. Run full-scale workloads on an isolated host.

Validity and output

The driver checks that SELECT VERSION() contains 4.2.4 and verifies the seed's warehouse and order_line row counts:

Scale	Warehouses	order_line rows
smoke	10	10
SF100	100	30,000,000
SF1000	1,000	300,000,000

Each repetition clones the seed into a unique database. The runner normally removes that run's clone and branches afterward; it retains the seed. If a process is forcibly interrupted, inspect its run ID before cleaning up any leftovers.

Output defaults to results/mo_v424/. Each run has config.textproto, run.log, driver_result.json, end-to-end stats JSON, and Parquet operation records. The output root has runs.jsonl. A valid run requires all workers and steps to complete, zero operation errors, no timeout, successful cleanup, and a zero exit code. Invalid runs make the driver exit nonzero.

Verification scope

On an isolated local matrixorigin/matrixone:4.2.4 container, all five workflows passed a smoke run: 7 worker steps, 188 recorded operations, zero operation errors, and zero cloned databases left after cleanup. The trimmed repository was checked for runner import and configuration generation. Full SF100/SF1000 correctness and performance have not been rerun on 4.2.4.