Scalability and Infrastructure of RPC in WhatsApp Using Erlang

Zeyu Zhang from WhatsApp examines the lifecycle of an RPC (Remote Procedure Call) within WhatsApp's infrastructure, which heavily relies on Erlang services. The talk covers scaling the cluster to over 40,000 nodes using Erlang Distribution, the service discovery and routing mechanisms, job dispatching and queuing models, bottlenecks, and overload protections. Specific insights include the disabling of the global server for better scalability, optimization of SSL encryption and decryption through kernel offloading, and managing more than 1 billion concurrent connections. Zhang also goes into the details of service discovery using the PG module, internal queuing systems, and methods for efficient request dispatching. Additionally, he addresses several common challenges such as service bottlenecks and high CPU utilization, offering practical solutions that WhatsApp has implemented to optimize performance.