Skip to main content
This runbook turns the jdbc-endpoint config-connect seam into a cloud product evaluation: OLTP throughput/goodput/latency from the recognized BenchBase TPC-C implementation, plus the price/performance composite, against a managed database you operate.
The connect chain (target → adapter → suite → evaluator) is covered by this repo’s CI (unit + mock). The live steps below need your managed database and a BenchBase build — clousight-bench never provisions the database on this path. The audited tpmC is not claimed; the numbers are BenchBase’s, unaudited.

​
0. What you need

  • A managed PostgreSQL/MySQL-compatible instance (Aliyun RDS/PolarDB, AWS RDS/Aurora, …) with a benchbase database created and network access from the driver machine.
  • A BenchBase build (Java 17+): put benchbase on PATH or set BENCHBASE_HOME.
  • pip install clousight-bench.

​
1. Where to run the driver

Same rule as every connect-mode evaluation: run the driver in-region (same VPC as the instance). OLTP tail latency over the public internet measures your ISP, not the database.

​
2. Connect profile

Start from the committed example (examples/cloud-connect/aliyun-rds-postgres.yaml / aws-aurora-postgres.yaml):
Credentials come from BENCHBASE_DB_USER / BENCHBASE_DB_PASSWORD env vars (BenchBase reads its config XML; the harness writes it owner-only 0600 into a 0700 work dir and removes it after the run).

​
3. Preflight, then run

csbench doctor now probes for real: a TCP reachability probe of the JDBC endpoint plus BenchBase launcher + Java ≥ 17 checks. A wrong address, a closed allowlist or a missing/old tool surfaces here with a remediation hint — not as a mid-run stack trace. The same checks gate csbench run at its PREFLIGHT stage.
You get tpc-c.throughput_req_per_sec, tpc-c.goodput_req_per_sec, tpc-c.p99_latency_us, … — environmental measurements from unmodified BenchBase.

​
4. Price/performance

CLOUSIGHT_PRICING_DATA=feed.json csbench run … adds $ per (req/sec) under extensions.pricing.price_performance[] — unaudited, from your own price input, never able to change the verdict. (Official price/tpmC needs an audited tpmC; we deliberately do not claim either.)

​
5. Honest caveats

  • scalefactor (warehouses) must scale with the instance class — an undersized warehouse count measures lock contention, not the service.
  • Load phase (--create --load) is part of the run and can take long at high scale factors; size time so the measured window dwarfs warmup effects.
  • Compare same dbtype, same scalefactor/terminals, same region locality only.