2026-04-23 Postgres backend report

2026-04-23 Postgres backend report

Test delta

• Default SQLite backend suite: 311 passed via python -m pytest -p no:schemathesis app/backend/tests/ -q --basetemp .pytest-tmp
• New Postgres-focused smoke coverage: 6 passed via python -m pytest -p no:schemathesis app/backend/tests/test_postgres_backend.py -q --basetemp .pytest-tmp
• Scope added in this change:
  • backend factory and wrapper selection
  • Postgres CRUD/query smoke coverage
  • concurrent write smoke coverage
  • startup/snapshot gating for Postgres
  • Postgres readiness checks

Benchmarks

Local p95 latency, single-user case, 200 seeded projects:

Operation	SQLite p95	Postgres p95
get_project	3.312 ms	4.605 ms
list_projects	11.010 ms	11.078 ms
query_projects(metric_type="binary")	7.249 ms	8.484 ms

Result:

• Postgres stayed within the acceptance guard for single-user usage
• worst observed regression in this local run was ~1.39x on get_project
• list_projects was effectively flat between backends in the local benchmark

CI runtime impact

• Existing SQLite backend suite runtime in local verification: 208.92 s
• New Postgres smoke suite runtime in local verification: 29.70 s
• Expected GitHub Actions impact:
  • roughly +30-90 s for the new job after warm image cache
  • potentially +1-2 min on cold runners because of Postgres image/bootstrap overhead

Notes

• SQLite remains the default path when AB_DATABASE_URL is unset
• HF snapshot sync is intentionally skipped for Postgres runtimes
• CI still keeps the original default verification job for SQLite; Postgres is covered by a dedicated Ubuntu smoke job