grpc-resource-server

README

gRPC Resource Server

In this example, a gRPC server backed by a SQL database handles concurrent requests from different clients, and is able to batch and deduplicate queries to reduce load on the database.

The underlying gRPC service is defined in proto/books.proto. If you make changes to that file, run make. This requires the Go gRPC compiler toolchain.

Running the Example

Open two terminal windows to this directory. In the first window, start the server:

go run cmd/grpc-server/main.go -log-level=DEBUG -seed

Note that the -seed argument should only be used once, or after erasing the database.

Setting the log level to DEBUG will show you the exact SQL that is executed.

In the second window, run the client:

go run cmd/grpc-client/main.go

You should see a log line that looks like this:

DEBUG SELECT ids="[urn:isbn:978-1098118730 urn:isbn:fake-book urn:isbn:978-0786965601 urn:isbn:978-1491973899 urn:isbn:978-1098149482]" query="SELECT id, title FROM books WHERE id IN (?,?,?,?,?)" args="[urn:isbn:978-1098118730 urn:isbn:fake-book urn:isbn:978-0786965601 urn:isbn:978-1491973899 urn:isbn:978-1098149482]"

The order of the IDs and args may be different.

Both commands accept the -help flag to see how they can be configured.

The Server

The server starts in cmd/grpc-server/main.go, but the implementation of the gRPC server is in server/server.go, and the database is implemented in fetcher/fetcher.go.

The server implements a single RPC, GetBook, which is a single-resource lookup. This follows Resource-Oriented Design for gRPC and is a common pattern for both gRPC and REST services.

All requests to this server are concurrent: they do not share anything and may be initiated by different clients. restdataloader is able to collapse these concurrent requests and deduplicate unique IDs before passing them into the fetcher, which uses a SQL SELECT ... WHERE id IN (?...) query to gather multiple results.

The Client

The client starts in cmd/grpc-client/main.go. It simulates several clients by making several parallel, independent requests, including multiple requests for the same resource. Other than the protocol definitions in proto/, the client is self-contained.