WhatsApp

Messaging service

Design a mobile messaging service for one-to-one and group chats that delivers every message exactly once to every device, even after days offline, with end-to-end encryption.

Overview

A messaging service moves small, private messages between phones. Unlike a feed, every message has specific recipients, and each one must arrive: not lost, not shown twice, and in the order it was sent. Recipients are often offline, so the server has to hold messages until their devices come back.

Three things make it hard at scale. Every online phone keeps a connection open to the service, so the service holds hundreds of millions of connections at once. Delivery must survive dropped connections, retries and crashes without duplicates. And with end-to-end encryption, the server routes messages it cannot read, which moves work such as encrypting for each device and each group member onto the phones.

Questions to ask first

How many users and messages? — 500 million daily active users; 40 messages sent per user a day
Groups? How big? — Yes, up to about 1,000 members
Several devices per account? — Yes: one phone plus a few linked devices
Does the server keep chat history? — No: it holds a message only until every device has it
How long can a device be offline? — Days or weeks; undelivered messages expire after 30 days
Encryption? — End to end; the server sees who talks to whom, not what they say

What you will be able to do

Size and route hundreds of millions of persistent connections.
Explain store-and-forward delivery, acknowledgments and receipts.
Get exactly-once display from at-least-once delivery with message IDs.
Compare client fan-out and server fan-out for devices and groups under end-to-end encryption.
Say what the server stores, for how long, and what it can and cannot see.

Key concepts

Persistent connections
Session registry
Store and forward
Delivery receipts
At-least-once plus dedup
Client fan-out
Sender Keys
Encrypted attachments
Presence

Functional requirements

Send a text message to a person or a group
Deliver it at once to online devices, and later to offline ones
Show sent, delivered and read status to the sender
Deliver every message to all of a user's linked devices
Send photos and videos as attachments
Show whether a contact is online and typing

Non-functional requirements

Messages to online devices arrive in under 500 ms at p99 within a region (a target we chose)
No message is lost once the server has acknowledged it
Each message is shown once, in the order sent within a chat
Chat servers can fail or be redeployed without losing messages
The server cannot read message content

Interview questions

How does a message reach a user who is offline?
How do you make sure a message is shown exactly once?
How do you scale to hundreds of millions of open connections?
How do group messages work with end-to-end encryption?
How do sent, delivered and read receipts work?
How do you guarantee ordering within a chat?
Where do photos and videos go?
What happens when a chat server crashes?