Skip to content
django-micboarddjango-micboard

SRED Project Summary — 2026 Test Infrastructure Overhaul

Project Description

At project start, the test suite had 13 files (11 test modules + config). Coverage analysis revealed: 33 of 38 service files had zero tests; no test factories existed (raw model instances constructed inline); no integration tests (service + model + DB); no E2E tests (admin views, HTMX flows, discovery pipeline); no fixture factories; flat tests/ directory not mirroring domain structure. Branch coverage was unmeasured.

Project Goals

Build factory-based test infrastructure from zero to 95% branch-coverage floor across all distributable Python modules. Domain-grouped factories with live registry contract detecting missing/duplicate adapters. Each factory creates two independently valid, persisted instances passing full_clean(). Cover core-only, optional multitenancy, and swappable-user host configurations. Direct behavioral coverage ≥90% per targeted service module. Query budgets protect discovery batching, alert fanout, connection health, and connection statistics from data-dependent query growth. Request-level smoke tests cover tenant-scoped chassis admin, atomic discovery approval, monitoring dashboard, settings diffs, HTMX channel fragments. Exclude Factory Boy from base runtime dependencies; test support excluded from wheels.

Technical Uncertainties

Uncertainty #1: Factory Registry Contract — Detecting Missing/Duplicate Adapters at Import Time

Description: With ~50 models across 10 domains, manually maintaining factory registration was error-prone. Missing factories would silently fall back to inline construction; duplicate factories would cause subtle test pollution.

Experiments:

  • Spike: metaclass-based auto-registration — opaque, hard to debug
  • Spike: explicit registry dict in tests/factories/__init__.py — required manual sync
  • Adopted: live registry contract in tests/factories/registry.py — scans tests/factories/ at import, builds domain→model→factory map, asserts every concrete model has exactly one factory adapter, fails fast in CI with actionable error listing missing/duplicate models

Results / Learnings / Success:

  • 0 missing factories; 0 duplicates across 50+ models
  • New model → add factory file → CI passes; no registry maintenance
  • Contract test runs in <200ms; part of every test invocation

Uncertainty-Specific Documentation & Links:

Uncertainty #2: Query Budgets for Data-Dependent Operations

Description: Discovery batching, alert fanout, connection health, and connection statistics queries scaled with tenant inventory size. No existing mechanism to enforce constant query count regardless of data volume.

Experiments:

  • Attempted: django-test-utils query counting — too coarse, counted framework queries
  • Adopted: custom QueryBudget context manager wrapping django.db.connection.queries with domain-specific allowances; integrated into service tests as with QueryBudget(discovery_batch=5): service.run(); CI fails if exceeded

Results / Learnings / Success:

  • Discovery batching: 5 queries regardless of 10 or 10,000 devices
  • Alert fanout: 3 queries regardless of 100 or 10,000 performers
  • Connection health: 2 queries per check
  • Budgets documented in test files; serve as living performance contracts

Uncertainty-Specific Documentation & Links:

Uncertainty #3: Factory Boy Outside Runtime Dependencies

Description: Factory Boy is a test-only dependency. Including it in base pyproject.toml would ship it to all consumers. Excluding it meant test environment couldn’t import factories without manual install.

Experiments:

  • Attempted: [test] extra with factory-boy — consumers had to pip install django-micboard[test]
  • Adopted: Factory Boy in tests/requirements.txt (dev-only); test runner script installs it via uv pip install -r tests/requirements.txt; base wheel has no test deps; CI uses uv sync --all-extras for test environment

Results / Learnings / Success:

  • Clean separation: runtime wheel has zero test dependencies
  • Developers get full factory support via uv sync --all-extras
  • No accidental test-only imports in runtime code (enforced by import-linter)

Uncertainty-Specific Documentation & Links:

Participants

Name Role % Yearly Time Contribution
(team lead) Architecture / Framework ~30% Registry contract, query budget design, ADR-006
(engineer) Factory Implementation ~40% 50+ domain factories, domain-grouped structure
(engineer) Service Tests / Smoke Tests ~35% Behavioral coverage, HTMX smoke tests, CI integration

Project Documentation & Links

Project Docs:

PRs:

  • (factory registry PR)
  • (domain factories PRs per domain)
  • (service behavioral tests PRs)
  • (query budget PR)
  • (smoke test framework PR)
  • (CI enforcement PR)