Notification Service

Designing multi-channel notifications: email, SMS, push, in-app with templates and preferences.

Requirements

• Templates with variables; localization
• User preferences; channel fallbacks and quiet hours
• Deduplication and idempotency

High-Level Design

• API to enqueue notifications; outbox pattern from producers
• Router selects channels based on preferences and urgency
• Per-channel workers with provider retries and DLQs

Capacity & Sizing

• Notifications/sec with burst tolerance; channel throughput limits
• Provider quotas and rate limits per account/region
• DLQ sizing and reprocessing policies

Key Components

• Enqueue API and outbox processors
• Router with preference evaluation and channel selection
• Channel workers with provider adapters

Data Model

• templates (template_id PK, name, locale, body)
• notifications (notif_id PK, user_id, template_id, status, created_at)
• preferences (user_id PK, email, sms, push, quiet_hours)
• deliveries (delivery_id PK, notif_id, channel, provider, status, attempts, last_error)

APIs

• Enqueue: POST /api/notifications with body { userId, templateId, data }
• Preferences: PUT /api/preferences/:userId with body { email, sms, push, quietHours }
• Status: GET /api/notifications/:id

Hot Path

1. Enqueue → persist → publish → route channels
2. Deliver via provider → update delivery status

Delivery Flow

1. Enqueue notification; persist and publish event
2. Router expands into channel-specific deliveries
3. Workers call providers with retries/backoff; DLQ on failure

Scaling

• Partition queues by channel; autoscale workers
• Use provider pools and rate limits; fallback providers
• Batch sends when supported (email)

Caching & TTL

• Cache compiled templates per locale with TTL
• Short TTL cache for user preferences; invalidate on update

Failure Modes & Mitigations

• Provider outage → fallback route and DLQ
• Spam/blocks → feedback loops and suppression lists
• Duplicate sends → idempotency keys per notif/user

Observability

• Delivery rates, failure reasons, and time-to-send
• End-to-end tracing across router and workers