How WeChat Guarantees Real‑Time, Lossless Messaging: Architecture & Sequence Mechanism

Background

WeChat, with a daily active user base of 1.34 billion in 2023, originated from the QQMail team and initially inherited an email‑style storage‑forward model for message handling. The core requirements for an instant‑messaging platform are (1) near‑real‑time delivery and (2) no message loss.

Message Sending Architecture

The sending process is divided into two main parts. First, the mobile client (Phone A) sends a request to the access layer ConnectSvr, which forwards it to the logic layer SendSvr. After anti‑spam checks, blacklist filtering, and other business logic, SendSvr stores the message in MsgStore. Second, the server notifies the recipient (Phone B) via a push flow: SendSvr triggers PushSvr, which either pushes through the long‑connection ConnectSvr (steps 5.1‑6) or falls back to platform‑specific push services such as APNs, WPPush, or BBPush (step 5.2) when the long‑connection is unavailable.

Message Receiving Architecture

When Phone B receives the push notification, it initiates a three‑step retrieval process: (1) request to ConnectSvr, (2) forwarding to ReceiveSvr, and (3) fetching the pending messages from MsgStore. This flow ensures that the recipient can obtain the message within roughly 100 ms after the sender’s request.

Sequence‑Based Loss‑Prevention Mechanism

To avoid message loss, each user is allocated a 32‑bit sequence space (≈4.2 billion values). Every outgoing message receives a unique, monotonically increasing sequence number. The server records the highest sequence assigned, while the client stores the highest sequence it has successfully processed.

The client periodically reports its current sequence ( Seq_cli) to the server. The server compares this with its own highest sequence ( Seq_svr) and delivers any missing messages in the range (Seq_cli + 1) … Seq_svr. This incremental delivery works even if acknowledgments are lost or the network is weak, because the client can re‑request the missing range after a timeout.

When a user logs in from multiple devices, each device reports its own Seq_cli. The server’s stored maximum sequence ensures that already‑confirmed messages are not resent, while unconfirmed messages are still delivered to any device that needs them.

Conclusion

The described architecture satisfies WeChat’s two fundamental messaging requirements: near‑real‑time delivery and zero loss. Although the system has evolved since its 2014 version, the core principles of layered servers, push fallback mechanisms, and sequence‑based synchronization remain valuable references for designing large‑scale instant‑messaging services.