SRED Project Summary — 2026 Extract Business Logic from Models to Services

💡 Try to be concise with answers Each project submission has to be reduced to around 400 words

Project Description

Five Django models embedded orchestration logic in save(), clean(), and lifecycle hooks: WirelessChassis (709 lines — largest model file in the codebase), WirelessUnit (449L), RFChannel (347L), DiscoveredDevice (353L), and ManufacturerConfiguration (213L). Business logic included status transitions, sync triggers, duplicate detection, JSON schema validation, and network state management. Additionally, post_save signals duplicated logic already present in service-layer methods (hardware_lifecycle.py 381L, hardware.py 534L). The implicit chain model.save() → django-lifecycle hook → signal → service was untestable in isolation and made bulk operations impossible without side effects.

Project Goals

Extract all orchestration from model persistence methods into explicit domain service seams. Models retain only declarative structure, constraints, and structural validation. Each affected model gets a dedicated service method (e.g., HardwareLifecycleService.save_chassis()) with a context parameter carrying caller metadata. Route all chassis field persistence through one seam: WirelessChassisWrite DTO → wireless_chassis_persistence_service.py. Enforce via CI: no save() overrides with DB writes, external calls, or signal emission.

Technical Uncertainties

Uncertainty #1: WirelessChassis — Disentangling 5 Write Paths into One Seam

Description: WirelessChassis.save() was invoked by manufacturer sync, discovery approval, promotion, imports, refresh, realtime updates, lifecycle creation, and regulatory repair — each with different context and side-effect requirements. No single service method could express all variants without becoming a god function.

Experiments:

• Attempted: single save_chassis() with 12 boolean flags — unmaintainable
• Attempted: inheritance hierarchy of save strategies — created circular imports with lifecycle adapters
• Adopted: WirelessChassisWrite DTO + context dict — callers construct DTO with exact fields + metadata; persistence service applies field updates, lifecycle service handles transitions, regulatory service handles band-plan coverage

Results / Learnings / Success:

• 8 write paths → 1 persistence seam + 2 domain services (lifecycle, regulatory)
• context carries triggered_by, operation_scope, skip_broadcast — explicit, testable
• Bulk operations use suppress_lifecycle=True in context; reviewed and approved

Uncertainty-Specific Documentation & Links:

• ADR: ADR-002 Extract Business Logic from Models
• PRs: (implementation PRs per model)

Uncertainty #2: DiscoveredDevice — Duplicate Detection Without Model Coupling

Description: DiscoveredDevice.save() performed network-level duplicate detection (MAC/IP matching) and queued items for approval. Moving this to a service required access to the queryset before persistence — but the model’s save() had already committed the row by the time dedup ran.

Experiments:

• Attempted: pre-save signal — still had committed row on race
• Adopted: DiscoveredDeviceService.claim_or_queue(dto, context) — creates in transaction with select_for_update on candidate set; duplicate → returns existing; unique → creates + queues for approval

Results / Learnings / Success:

• Duplicate detection now atomic and testable with fake clock
• Approval workflow decoupled from persistence
• Service unit tests cover race conditions; model save() is now 3 lines

Uncertainty-Specific Documentation & Links:

• ADR: ADR-002 Extract Business Logic from Models

Uncertainty #3: ManufacturerConfiguration — JSON Schema Validation as Pure Function

Description: validate() (60 lines) embedded jsonschema validation with custom error collection. Moving to service required preserving error format for admin forms while making validation a pure, importable function.

Experiments:

• Extracted validate_manufacturer_config(config_dict, schema) -> list[ValidationError] to services/discovery/configuration/validation.py
• Model clean() delegates to pure function; service uses same function
• Admin forms catch ValidationError and render field-level messages

Results / Learnings / Success:

• Validation logic now unit-testable with 200+ schema test cases (no DB)
• Single source of truth for schema validation across admin, API, and sync

Uncertainty-Specific Documentation & Links:

• ADR: ADR-002 Extract Business Logic from Models

---

Participants

Name	Role	% Yearly Time	Contribution
(team lead)	Architecture / Implementation	~35%	DTO design, chassis persistence seam, ADR-002
(engineer)	Implementation / Testing	~30%	DiscoveredDevice service, ManufacturerConfiguration validation
(engineer)	Implementation	~25%	WirelessUnit, RFChannel services, CI rules

---

Project Documentation & Links

Project Docs:

• ADR-002 Extract Business Logic from Models
• CONTEXT.md (domain model reference)

PRs:

• (per-model extraction PRs: ManufacturerConfiguration, DiscoveredDevice, WirelessUnit, RFChannel, WirelessChassis)
• (CI enforcement PR)