Consumer Messaging Platform — WhatsApp-Class System · View 02 of 23

High-Level Architecture

The shape of the system in one picture: clients hold a persistent authenticated socket, the core accepts and orders, and fanout drives delivery per device.

Editable source SVG draw.io All views

Why this shape

Persistent sockets, not polling — a message reaches an online device in one hop from the gateway
Accept and fanout are separated so a slow recipient never blocks the sender's acknowledgement
Per-device inboxes make multi-device a data-model property rather than a special case

Design targets

p99 send-to-deliver under 500 ms within a region
60 million concurrent sockets at peak per region cluster
1 million message accepts per second at design peak

Deliberately omitted

Calls, status, backup and search each have their own view
Observability agents and the business API gateway are shown in views 19 and 23