notify.md
· 1.7 KiB · Markdown
Raw
# System Design: A Notification Service (Push, Email, SMS)
"Design a notification system" tests queues, fan-out, retries and user preferences, all common follow-up areas.
## Requirements
- Send push, email and SMS notifications, triggered by other services or scheduled campaigns.
- Respect user preferences, opt-outs and quiet hours.
- At-least-once delivery, with idempotency so users don't get duplicates.
- Scale example: 10M notifications a day, with bursts during campaigns.
## High-level flow
1. **Notification API:** producers call `POST /notify {user_id, template, data, priority}`.
2. **Validation and preferences:** check opt-outs, rate limits per user, and quiet hours.
3. **Message queue per channel** (Kafka or SQS): separate queues so a slow SMS provider doesn't block push.
4. **Channel workers:** render the template and call the provider (APNs, FCM, an email service, an SMS gateway).
5. **Delivery log:** store status per message for retries, analytics and support.
## Reliability details
- **Retries** with exponential backoff and a dead-letter queue for permanent failures.
- **Idempotency key** per notification so a retried job doesn't send twice.
- **Priority lanes:** one-time passwords must not wait behind a marketing blast.
- **Provider failover:** keep a second SMS or email vendor ready.
## Data model
`users_devices(user_id, device_token, platform)`, `preferences(user_id, channel, enabled, quiet_hours)`, `notifications(id, user_id, channel, status, created_at)`.
## Good extras
Template versioning, per-tenant throttling, tracking opens and clicks for email, and batching low-priority messages into digests.
---
*Disclosure: I work on [Linkjob AI](https://www.linkjob.ai/), an AI interview assistant. This is general prep material.*
System Design: A Notification Service (Push, Email, SMS)
"Design a notification system" tests queues, fan-out, retries and user preferences, all common follow-up areas.
Requirements
- Send push, email and SMS notifications, triggered by other services or scheduled campaigns.
- Respect user preferences, opt-outs and quiet hours.
- At-least-once delivery, with idempotency so users don't get duplicates.
- Scale example: 10M notifications a day, with bursts during campaigns.
High-level flow
- Notification API: producers call
POST /notify {user_id, template, data, priority}. - Validation and preferences: check opt-outs, rate limits per user, and quiet hours.
- Message queue per channel (Kafka or SQS): separate queues so a slow SMS provider doesn't block push.
- Channel workers: render the template and call the provider (APNs, FCM, an email service, an SMS gateway).
- Delivery log: store status per message for retries, analytics and support.
Reliability details
- Retries with exponential backoff and a dead-letter queue for permanent failures.
- Idempotency key per notification so a retried job doesn't send twice.
- Priority lanes: one-time passwords must not wait behind a marketing blast.
- Provider failover: keep a second SMS or email vendor ready.
Data model
users_devices(user_id, device_token, platform), preferences(user_id, channel, enabled, quiet_hours), notifications(id, user_id, channel, status, created_at).
Good extras
Template versioning, per-tenant throttling, tracking opens and clicks for email, and batching low-priority messages into digests.
Disclosure: I work on Linkjob AI, an AI interview assistant. This is general prep material.