joogy06 SaaS Developer Patterns
Reference Files
Detailed code examples, patterns, and configuration are in the reference files below. Read the relevant file when working on that area.
| File | Covers |
|---|---|
| api-jobs-webhooks.md | tenant-scoped API endpoints, background job processing per tenant, webhook delivery systems |
| email-storage-testing-deploy.md | email/notification systems, file storage isolation, caching strategies, testing multi-tenant code, and deployment patterns |
| tenant-context-database.md | tenant context middleware, multi-tenant database layer with SQLAlchemy, row-level security |
Anti-Patterns
| Don't | Do Instead |
|---|---|
Pass tenant_id as function parameter through 15 layers |
Use contextvars — set once in middleware, read anywhere |
Cache with keys like project:123 (no tenant prefix) |
Always tenant:{tid}:project:123 — prevents cross-tenant cache reads |
| Run Celery tasks without tenant context propagation | Use TenantTask base class that injects/restores tenant from headers |
| Generate pre-signed S3 URLs without validating key prefix | Always verify the key starts with tenant-{tenant_id}/ before signing |
Use KEYS * in Redis for tenant key operations |
Use SCAN with cursor — KEYS blocks Redis on large datasets |
| Run all tenant schema migrations serially | Parallelize with bounded concurrency (5-10 at a time) |
| Write tests that share a single hardcoded tenant_id | Each test gets a fresh uuid4() tenant_id via fixture |
| Deploy new features to all tenants simultaneously | Canary by tier: internal, then starter, then pro, then enterprise |
| Store uploaded files at user-chosen paths | Sanitize filenames, prepend UUID, enforce tenant prefix |
| Let one tenant's failed webhook retries consume the entire retry queue | Use per-tenant or per-endpoint retry budgets with circuit breakers |