Lohnt sich dein eigenes Svix?

Build a self-hosted outbound webhook delivery service. Single Node.js project, TypeScript, Fastify for HTTP, Postgres via Drizzle ORM, BullMQ on Redis for the delivery queue. Docker Compose for Postgres and Redis. No cloud accounts, no telemetry, no auth provider: a single static API key in .env guards the admin and ingest routes.

Data model:
- applications: id, name, created_at (one per tenant/customer of mine)
- endpoints: id, application_id, url, description, secret, enabled, event_types (text array, empty means all), created_at
- messages: id, application_id, event_type, payload jsonb, created_at
- attempts: id, message_id, endpoint_id, attempt_number, status_code, response_body_excerpt, error, duration_ms, created_at

HTTP API (all JSON, all behind the API key header):
- POST /api/applications, GET /api/applications
- POST /api/applications/:id/endpoints, GET, PATCH, DELETE
- POST /api/applications/:id/messages: body has event_type and payload. Persist the message, resolve matching enabled endpoints, enqueue one delivery job per endpoint, return the message id with 202.
- GET /api/messages/:id/attempts
- POST /api/attempts/:id/replay: re-enqueue that single delivery immediately.

Delivery worker:
- POST the raw JSON payload with headers: webhook-id, webhook-timestamp (unix seconds), webhook-signature as "v1," plus base64 HMAC-SHA256 over "{id}.{timestamp}.{body}" using the endpoint secret.
- 5 second connect timeout, 10 second total timeout.
- Success is any 2xx. Retry on everything else with backoff: 5s, 30s, 5m, 30m, 2h, 5h, then give up and mark the endpoint as failing.
- Record an attempt row for every try, truncating response bodies to 2KB.
- Rate limit per endpoint to 20 concurrent deliveries max using a BullMQ group or per-endpoint limiter.

Also build a minimal server-rendered admin UI at / using Fastify plus plain HTML templates and no frontend framework: list applications, list endpoints, list recent messages, drill into a message to see attempts and hit replay. Ugly is fine, tables and buttons only.

Out of scope: a customer-facing self-serve portal, per-customer login, multi-region, event type schema validation, inbound webhook receiving, transformations.

Include a README with docker compose up, migration command, and one curl example that creates an application, an endpoint pointing at a local sink, and sends a message. Add a small script that runs a throwaway HTTP sink on port 4000 that verifies the signature and prints the payload, so delivery can be tested end to end. Write integration tests for signature generation and for the retry backoff schedule.