CanonicalStore Benchmark Evidence
mysql-8.0.46-local.json is a retained, redacted run of the default
canonical-mysql-contention/v1 workload from harness commit
f289a8752af298a90a7eb48e521835f7ba3f70d5.
The local MySQL 8.0.46 run used 8 workers, 25 operations per worker, 25 pre-seeded documents per
tenant, 256-byte payloads, explicit READ COMMITTED CanonicalStore mutations, and
innodb_flush_log_at_trx_commit=1. All 800 timed operations across four scenarios completed with
no classified failure.
The one-hot-tenant write scenario produced 346.782 ms p95 and 44.834 commits/second while its serialized tenant state grew from 50,697 to 746,173 bytes. It therefore missed the starter 100 ms p95 and 50 commits/second budgets. The report mechanically recommends evaluating PostgreSQL’s granular layout or a dedicated projection. The result is local machine evidence, not a claim about MySQL 8.4, Cloud SQL, a production topology, or PostgreSQL performance. Operators must rerun the same harness with representative state, concurrency, networking, durability, and SLO budgets.
Release verification validates the receipt’s schema, commit ancestry, operation counts, durability, decision arithmetic, and absence of connection/tenant/payload fields:
scripts/validate-canonical-mysql-benchmark-report.shCanonical source: benchmarks/canonical-store/README.md at cb04cf7