Load Balance on the Client in gRPC

Distributed Services with Go — by Travis Jeffery (63 / 84)

👈 Three Load-Balancing Strategies | TOC | Make Servers Discoverable 👉

Though the ideas we’ll talk about in this chapter can apply to any client and server, because our service is a gRPC service, we’ll use those terms. gRPC separates server discovery, load balancing, and client requests and response handling — often the only code you’ll write is the latter. In gRPC, resolvers discover servers and pickers load balance by picking what server will handle the current request. gRPC also has balancers that manage subconnections but defer the load balancing to the pickers. gRPC provides an API (base.NewBalancerBuilderV2) to create a base balancer, but you probably won’t have to write your own balancer.

When you call grpc.Dial, gRPC takes the address and passes it on to the resolver, and the resolver discovers the servers. gRPC’s default resolver is the DNS resolver. If the address you give to gRPC has multiple DNS records associated with it, gRPC will balance the requests across each of those records’ servers. You can write your own resolvers and use resolvers written by the community. For example, Kuberesolver[52] resolves the servers by fetching the endpoints from Kubernetes’ API.

gRPC uses round-robin load balancing by default. The round-robin algorithm works by sending the first call to the first server, the second call to the second server, and so on. After the last server, it…