Designing A Push Notification System(Mobile/Web Push like Firebase/APNs)

1) Problem Clarification / Làm rõ bài toán

EN

We need a distributed notification service that pushes alerts to mobile/web users.

Questions to clarify:

One-to-one push? broadcast? segmented push?
Do we need delivery guarantees?
Do we support silent pushes?
Platform targets (APNs, FCM, WebSockets)?
Do we allow scheduling?
Are retries required?

VI

Thiết kế hệ thống push thông báo đến mobile/web.

Cần hỏi rõ:

Push cá nhân? broadcast? theo segment?
Yêu cầu delivery guarantee không?
Có hỗ trợ silent push không?
Target: APNs, FCM, Web push?
Có scheduling không?
Có retry không?

2) Requirements / Yêu cầu hệ thống

EN – Functional

✔ Device registration
✔ Send push to users / segments
✔ Delivery tracking
✔ Retry on failure
✔ Priority types (high vs normal)

VI – Chức năng

✔ Đăng ký thiết bị
✔ Gửi push theo user/segment
✔ Theo dõi trạng thái gửi
✔ Retry khi lỗi
✔ Mức ưu tiên (high/normal)

EN – Non-Functional

✔ Extremely scalable
✔ Low latency fan-out
✔ High delivery reliability
✔ Cost efficiency

VI – Phi chức năng

✔ Scale cực lớn
✔ Fan-out trễ thấp
✔ Đảm bảo độ tin cậy
✔ Chi phí tối ưu

3) Scale Estimation / Ước lượng tải

EN

Assume:

50M active users
10M push/min peak for broadcasts
40% failure retry potential

VI

Giả định:

50M người dùng active
Broadcast peak 10M push/phút
40% retry tiềm năng

4) High-Level Architecture / Kiến trúc tổng quan

Client → Device Registry Service → Notification Service → Queue → Push Adapters → APNs/FCM/Web
                                              ↓
                                          Status Store

EN

Registry binds user IDs with device tokens
Notification service prepares payload
Queue buffers bursts
Adapter integrates with APNs/FCM

VI

Registry lưu token thiết bị
Service build payload
Queue chống burst
Adapter call APNs/FCM/Web push

5) Device Registry / Đăng ký device

EN

Store mapping:

DEVICE (
   user_id,
   device_token,
   platform ENUM(ios,android,web),
   status,
   last_activeTs
)

VI

Lưu mapping:

DEVICE (
   user_id,
   device_token,
   platform,
   status,
   last_activeTs
)

Index by user_id for multi-device delivery.

6) Queue-based fan-out / Fan-out dựa trên queue

EN

Peak broadcast traffic must not overload push gateways.

=> Use Kafka / SQS:

Partition keyed on segment or user shard
Consumer groups scale horizontally

VI

Broadcast peak không thể push trực tiếp.

=> Kafka/SQS:

Partition theo segment/user shard
Consumer scale linear

7) Delivery Semantics / Đảm bảo gửi

EN

We aim at-least-once delivery:

Retry if gateway fails
Deduplicate using message IDs

Status store tracks:

sent
delivered
opened

VI

Mục tiêu at least once:

Retry khi gateway fail
Dedupe theo msg_id

Status theo dõi:

sent
delivered
opened

8) Push Adapter Layer / Tầng giao tiếp APNs/FCM

EN

Adapter normalizes format:

APNs HTTP/2
FCM REST/gRPC
Web push (VAPID)

VI

Adapter chuyển payload:

APNs HTTP/2
FCM REST/gRPC
Web push với VAPID

Retry strategy:

exponential backoff
jitter
DLQ for poison payloads

9) Scheduling & Campaign Service / Lịch chiến dịch

EN

Optional: allow scheduled or recurring notification campaigns.

Offering:

Cron-based scheduler
Precomputed recipient list
Fan-out through queue

VI

Tùy chọn: scheduler cho marketing/campaign.

Gồm:

Cron scheduler
Tính sẵn recipient list
Fan-out qua queue

10) Rate Limiting / Giới hạn tốc độ

EN

Global limit per gateway provider
User-level rate limit to avoid spam
Segment-based throttling

VI

Limit theo gateway
Limit theo user
Throttling theo segment

11) Failure Handling / Xử lý lỗi

EN

Cases:

Token expired → remove token
APNs feedback list → token invalidation
Burst spike → queue buffer
Queue outage → degrade mode

VI

Các lỗi:

Token hết hạn → xóa mapping
APNs feedback nhận token chết → xóa token
Peak burst → đệm qua queue
Queue chết → degrade mode

12) Observability / Giám sát

EN

Monitor:

delivery success rate
queue lag
retry count
device invalidation rate
notification latency

VI

Giám sát:

delivery success rate
queue lag
số retry
rate invalid token
latency push