fix: platform admin authentication and UserContext completeness

Issues fixed:
- Platform selection returned LoginResponse requiring user timestamps,
  but UserContext doesn't have created_at/updated_at. Created dedicated
  PlatformSelectResponse that returns only token and platform info.

- UserContext was missing platform context fields (token_platform_id,
  token_platform_code). JWT token included them but they weren't
  extracted into UserContext, causing fallback warnings.

- admin_menu_config.py accessed admin_platforms (SQLAlchemy relationship)
  on UserContext (Pydantic schema). Changed to use accessible_platform_ids.

- Static file mount order in main.py caused 404 for locale files.
  More specific paths (/static/modules/X/locales) must be mounted
  before less specific paths (/static/modules/X).

Changes:
- models/schema/auth.py: Add PlatformSelectResponse, token_platform_id,
  token_platform_code, can_access_platform(), get_accessible_platform_ids()
- admin_auth.py: Use PlatformSelectResponse for select-platform endpoint
- admin_platform_service.py: Accept User | UserContext in validation
- admin_menu_config.py: Use accessible_platform_ids instead of admin_platforms
- main.py: Mount locales before static for correct path priority

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
This commit is contained in:
2026-02-02 22:31:35 +01:00
parent 9a0dd84035
commit b935592430
6 changed files with 344 additions and 20 deletions

View File

@@ -0,0 +1,280 @@
# Middleware Frontend Detection - Problem Statement
**Date**: 2026-02-02
**Status**: Pending
**Priority**: Medium
---
## Context
During the fix for admin API authentication (where `/api/v1/admin/*` routes returned 401), we identified architectural issues in how the middleware detects frontend context (admin/vendor/storefront/platform).
The immediate authentication issue was fixed by making `FrontendType` mandatory in `require_module_access()`. However, the middleware still has design issues that should be addressed.
---
## Current State
### Middleware Files Involved
- `middleware/platform_context.py` - Detects platform from host/domain/path
- `middleware/vendor_context.py` - Detects vendor from subdomain/domain/path
### Current Detection Logic
```python
# In both PlatformContextManager and VendorContextManager
def is_admin_request(request: Request) -> bool:
host = request.headers.get("host", "")
path = request.url.path
# Production: domain-based
if host.startswith("admin."):
return True
# Development: path-based
return path.startswith("/admin")
```
### Routing Modes
**Development (localhost, path-based)**:
```
localhost:9999/admin/* → Admin pages
localhost:9999/api/v1/admin/* → Admin API
localhost:9999/vendor/* → Vendor pages
localhost:9999/api/v1/vendor/* → Vendor API
localhost:9999/* → Platform/storefront
```
**Production (domain/subdomain-based)**:
```
admin.platform.com/* → Admin (all paths)
vendor.platform.com/* → Vendor portal
shop.mystore.com/* → Vendor custom domain
api.platform.com/v1/* → Shared API domain (?)
platform.com/* → Marketing/platform pages
```
---
## Problems Identified
### 1. Incomplete Path Detection for Development Mode
The middleware only checks `/admin` but not `/api/v1/admin/*`:
```python
return path.startswith("/admin") # Misses /api/v1/admin/*
```
**Impact**: In development, API routes like `/api/v1/admin/messages/unread-count` are not recognized as admin requests, causing incorrect context detection.
**Note**: This doesn't break authentication anymore (fixed via `require_module_access`), but may affect context detection (vendor/platform context might be incorrectly applied to admin API routes).
### 2. Code Duplication
Same `is_admin_request` logic exists in 3 places:
- `PlatformContextManager.is_admin_request()` (static method)
- `PlatformContextMiddleware._is_admin_request()` (instance method)
- `VendorContextManager.is_admin_request()` (static method)
### 3. Hardcoded Paths
Path patterns are hardcoded in multiple locations:
- Middleware: `/admin`, `/vendor`
- Routes discovery: `/api/v1/admin`, `/api/v1/vendor` (in `app/modules/routes.py`)
- API main: `/v1/admin`, `/v1/vendor` (in `app/api/main.py`)
### 4. No Single Source of Truth
There's no centralized configuration that defines:
- What domains/subdomains map to which frontend
- What path patterns map to which frontend
- Whether we're in dev mode (path-based) or prod mode (domain-based)
### 5. Incomplete Frontend Coverage
Only `is_admin_request()` exists. No equivalent methods for:
- `is_vendor_request()`
- `is_storefront_request()`
- `is_platform_request()`
### 6. Not Using FrontendType Enum
Middleware returns `bool` instead of using the `FrontendType` enum that exists in `app/modules/enums.py`.
---
## Production Deployment Scenarios to Consider
### Scenario A: Subdomain per Frontend
```
admin.platform.com → Admin
vendor.platform.com → Vendor portal
*.platform.com → Vendor shops (wildcard subdomain)
platform.com → Marketing site
```
### Scenario B: Shared Domain with Path Routing
```
platform.com/admin/* → Admin
platform.com/vendor/* → Vendor
platform.com/api/v1/admin/* → Admin API
platform.com/* → Marketing/storefront
```
### Scenario C: Separate API Domain
```
admin.platform.com/* → Admin pages
api.platform.com/v1/admin/* → Admin API (different domain!)
```
**Issue**: Host is `api.platform.com`, path is `/v1/admin/*` (no `/api` prefix)
### Scenario D: Multi-Platform (current architecture)
```
oms.platform.com/* → OMS platform
loyalty.platform.com/* → Loyalty platform
admin.platform.com/* → Global admin for all platforms
```
---
## Proposed Solution Options
### Option A: Centralized FrontendDetector
Create a single utility class that handles all frontend detection:
```python
# app/core/frontend_detector.py
from app.core.config import settings
from app.modules.enums import FrontendType
class FrontendDetector:
"""Centralized frontend detection for both dev and prod modes."""
# Configurable patterns
ADMIN_SUBDOMAINS = ["admin"]
VENDOR_SUBDOMAINS = ["vendor", "portal"]
@classmethod
def detect(cls, host: str, path: str) -> FrontendType | None:
"""
Detect frontend type from request host and path.
Priority:
1. Domain/subdomain check (production)
2. Path prefix check (development)
"""
host_without_port = host.split(":")[0] if ":" in host else host
# Production: subdomain-based
subdomain = cls._get_subdomain(host_without_port)
if subdomain:
if subdomain in cls.ADMIN_SUBDOMAINS:
return FrontendType.ADMIN
if subdomain in cls.VENDOR_SUBDOMAINS:
return FrontendType.VENDOR
# Development: path-based
return cls._detect_from_path(path)
@classmethod
def _detect_from_path(cls, path: str) -> FrontendType | None:
# Check both page routes and API routes
admin_patterns = ["/admin", f"{settings.API_PREFIX}/admin"]
vendor_patterns = ["/vendor", f"{settings.API_PREFIX}/vendor"]
storefront_patterns = [f"{settings.API_PREFIX}/storefront"]
platform_patterns = [f"{settings.API_PREFIX}/platform"]
for pattern in admin_patterns:
if path.startswith(pattern):
return FrontendType.ADMIN
# ... etc
```
### Option B: Configuration-Driven Detection
Define all patterns in settings:
```python
# app/core/config.py
class Settings:
API_PREFIX: str = "/api/v1"
FRONTEND_DETECTION = {
FrontendType.ADMIN: {
"subdomains": ["admin"],
"path_prefixes": ["/admin"], # API prefix added automatically
},
FrontendType.VENDOR: {
"subdomains": ["vendor", "portal"],
"path_prefixes": ["/vendor"],
},
# ...
}
```
### Option C: Route-Level State Setting
Have the router set `request.state.frontend_type` when the route matches, eliminating detection entirely:
```python
# In route discovery or middleware
@app.middleware("http")
async def set_frontend_type(request: Request, call_next):
# Determine from matched route's tags or prefix
if request.scope.get("route"):
route = request.scope["route"]
if "admin" in route.tags:
request.state.frontend_type = FrontendType.ADMIN
return await call_next(request)
```
---
## Questions to Answer Before Implementation
1. **What's the production deployment model?**
- Subdomains per frontend?
- Separate API domain?
- Path-based on shared domain?
2. **Should detection be configurable?**
- Environment-specific patterns?
- Runtime configuration vs build-time?
3. **What does the middleware actually need?**
- `PlatformContextMiddleware`: Needs to know "is this NOT a platform-specific request?"
- `VendorContextMiddleware`: Needs to know "should I detect vendor context?"
- Maybe they just need `is_global_admin_request()` not full frontend detection
4. **API domain considerations?**
- Will API be on separate domain in production?
- If so, what's the path structure (`/v1/admin` vs `/api/v1/admin`)?
---
## Files to Modify (when implementing)
- `app/core/config.py` - Add centralized path/domain configuration
- `app/core/frontend_detector.py` - New centralized detection utility
- `middleware/platform_context.py` - Use centralized detector
- `middleware/vendor_context.py` - Use centralized detector
- `app/modules/routes.py` - Use centralized path configuration
---
## Related Commits
- `9a0dd84` - fix: make FrontendType mandatory in require_module_access
- `01e7602` - fix: add missing db argument to get_admin_context calls
---
## Next Steps
1. Answer the deployment model questions above
2. Choose solution option (A, B, or C)
3. Implement centralized detection
4. Update middleware to use centralized detection
5. Add tests for both dev and prod modes