การกำหนดบริการและข้อความด้วย Protobuf IDL

In a gRPC microservice, the contract comes before the code. You describe your data and your RPC methods in a language-neutral Interface Definition Language (IDL) called .proto, then generate type-safe stubs for Node.js, Go, Java, and more from that single file.

• One .proto file is the single source of truth shared by client and server.
• Protocol Buffers (protobuf) is both the IDL and the binary wire format.
• You never hand-write the serialization code — the compiler does it.

This lesson shows how to write .proto contracts and turn them into JavaScript client and server stubs.

Every modern .proto file starts by declaring the syntax version and a package. The package namespaces your symbols so two services can both define a User without colliding.

• syntax = "proto3"; — always use proto3 for new services.
• package — logical namespace, maps to a JS object path after generation.
• The file groups message (data shapes) and service (RPC methods).

A message is a record of typed fields. The number after the = is the field tag, not a default value. Tags are how protobuf identifies each field on the binary wire — they must be unique within a message and must never change once deployed.

• Field names can be renamed freely; tags must stay stable for backward compatibility.
• Tags 1-15 use a single byte, so reserve them for your most frequent fields.
• Scalar types: string, bool, int32, int64, double, bytes.

When you load a .proto with @grpc/proto-loader, each protobuf type maps to a JavaScript value. Knowing the mapping avoids surprises at runtime.

• string, bool, int32, float, double map to JS string, boolean, number.
• int64 / uint64 map to a string by default in JS (numbers can exceed Number.MAX_SAFE_INTEGER).
• bytes becomes a Buffer.
• Unset proto3 scalars come back as their zero value ("", 0, false) — they are never undefined on the wire.

A service block lists the RPC methods clients can call. Each rpc takes exactly one request message and returns exactly one response message. Wrapping request/response in dedicated messages (rather than passing bare scalars) lets you add fields later without breaking the contract.

• Method names are PascalCase by convention.
• Always define a request and a response message per method, even if one is empty.

The stream keyword on either side of an rpc declaration controls the streaming model. The same IDL keyword decides whether your generated handler receives a single value or a stream.

• Unary: rpc Get(Req) returns (Res) — one in, one out.
• Server streaming: returns (stream Res) — server pushes many.
• Client streaming: (stream Req) — client uploads many.
• Bidirectional: (stream Req) returns (stream Res).

Beyond scalars, protobuf gives you composite shapes for real-world data.

• enum — a closed set of values; the first member must be 0 and is the default.
• repeated — an ordered list; decodes to a JS Array.
• Nested messages model structured sub-records.

Prefix enum members to avoid name clashes, since enum values share the enclosing scope.

In Node you generate stubs at runtime with @grpc/proto-loader plus @grpc/grpc-js. The loader parses the .proto and loadPackageDefinition turns it into a navigable JS object keyed by your package path.

• keepCase: true preserves field names exactly as written in the proto.
• longs: String keeps 64-bit ints safe as strings.
• The package path users.v1 becomes proto.users.v1.

The generated UserService gives you a service definition you register handlers against. Each unary handler receives (call, callback); call.request is your decoded request message, and you reply via the Node-style callback(err, response).

• Return a gRPC status by passing an error with a code from grpc.status.
• The response object must match the response message fields.

The same generated UserService is also a client constructor. You instantiate it with a target address and credentials, then call methods directly. Each unary call takes the request object and a Node-style callback.

• credentials.createInsecure() for local/dev; use TLS in production.
• The request and response are plain JS objects matching the proto messages.

A contract is only useful if it can change without breaking deployed clients. proto3 makes additive evolution safe when you follow a few rules.

• Add new fields with new, never-before-used tag numbers — old clients ignore them.
• Never reuse or renumber tags; mark removed tags with reserved.
• Renaming a field is wire-compatible (tag is what matters), but breaks JSON/text usage.
• Use reserved for both removed tag numbers and names to prevent accidental reuse.

You need to remove the deprecated phone field (tag 4) from a deployed User message. What is the correct, backward-compatible way to do it?

You learned to design and consume gRPC contracts with Protobuf IDL:

• A .proto file with syntax = "proto3" and a package is the single source of truth for client and server.
• message defines typed records; the number after each field is its stable wire tag, not a default.
• service + rpc declare methods; the stream keyword selects unary, server-, client-, or bidirectional streaming.
• enum, repeated, and nested messages model richer data; int64 decodes to a JS string.
• In Node, @grpc/proto-loader + grpc.loadPackageDefinition generate both the server service and the client constructor.
• Evolve contracts additively and protect retired tags/names with reserved.