gRPC vs REST: Wann was einsetzen

// REST: JSON over HTTP — human-readable, widely supported
// POST /api/users
// Content-Type: application/json
// { "name": "Alice", "email": "alice@example.com" }
 
// Response: 201 Created
// { "id": "user-123", "name": "Alice", "email": "alice@example.com" }
 
// gRPC: Protocol Buffers over HTTP/2 — binary, type-safe, fast
// Defined in a .proto file:

// user_service.proto
syntax = "proto3";
 
package users;
 
service UserService {
  rpc CreateUser(CreateUserRequest) returns (CreateUserResponse);
  rpc GetUser(GetUserRequest) returns (User);
  rpc ListUsers(ListUsersRequest) returns (stream User);  // Server streaming
  rpc UpdateUsers(stream UpdateUserRequest) returns (UpdateUsersResponse); // Client streaming
}
 
message CreateUserRequest {
  string name = 1;
  string email = 2;
}
 
message CreateUserResponse {
  string id = 1;
  string name = 2;
  string email = 3;
}
 
message User {
  string id = 1;
  string name = 2;
  string email = 3;
  int64 created_at = 4;  // Unix timestamp
}
 
message GetUserRequest {
  string id = 1;
}
 
message ListUsersRequest {
  int32 page_size = 1;
  string page_token = 2;
}
 
message UpdateUserRequest {
  string id = 1;
  string name = 2;
  string email = 3;
}
 
message UpdateUsersResponse {
  int32 updated_count = 1;
}

// JSON serialization vs Protocol Buffers
 
// A typical user object in JSON: ~150 bytes
const jsonPayload = {
  id: "user-123",
  name: "Alice Johnson",
  email: "alice@example.com",
  createdAt: 1648684800,
  roles: ["admin", "editor"],
};
// JSON.stringify: ~120 bytes
 
// Same data in Protocol Buffers: ~45 bytes
// Binary encoding: field numbers + wire types + values
// No field names transmitted — just indices
// 60-70% smaller on average for typical payloads
 
// Performance implications at scale:
// 10,000 requests/second × 150 bytes (JSON) = 1.5 MB/s network
// 10,000 requests/second × 45 bytes (Protobuf) = 0.45 MB/s network
// 3x less bandwidth, 2-5x faster serialization/deserialization

// ❌ Using REST/JSON for high-throughput internal services
interface InternalServiceCall {
  protocol: 'REST/JSON';
  serialization: '~5ms per large payload';
  networkOverhead: 'HTTP/1.1 head-of-line blocking, text headers';
  typesSafety: 'Manual — OpenAPI spec may drift from implementation';
  // At 50k req/s between services, this adds up significantly
}
 
// ✅ Using gRPC for high-throughput internal services
interface OptimizedServiceCall {
  protocol: 'gRPC/Protobuf';
  serialization: '~0.5ms per large payload';
  networkOverhead: 'HTTP/2 multiplexing, binary headers';
  typeSafety: 'Generated code — proto file IS the implementation contract';
  // 10x serialization improvement, multiplexed connections
}

// gRPC streaming patterns
 
// 1. Unary — standard request/response (like REST)
// Client sends one message, server sends one response
async function getUser(id: string): Promise<User> {
  return client.getUser({ id });
}
 
// 2. Server streaming — server sends a stream of responses
// Client sends one request, server sends many responses
async function* listUsers(pageSize: number): AsyncGenerator<User> {
  const stream = client.listUsers({ pageSize, pageToken: '' });
  for await (const user of stream) {
    yield user;
  }
}
 
// 3. Client streaming — client sends a stream of requests
// Client sends many messages, server sends one response
async function batchUpdateUsers(
  updates: AsyncIterable<UpdateUserRequest>
): Promise<UpdateUsersResponse> {
  return client.updateUsers(updates);
}
 
// 4. Bidirectional streaming — both sides stream simultaneously
// Client and server send messages independently
async function syncUsers(
  localChanges: AsyncIterable<UserChange>,
  onRemoteChange: (change: UserChange) => void
): Promise<void> {
  const stream = client.syncUsers();
 
  // Send local changes
  for await (const change of localChanges) {
    stream.write(change);
  }
 
  // Receive remote changes
  stream.on('data', onRemoteChange);
}

// REST developer experience
 
// Pros:
// - cURL-friendly: curl http://localhost:3000/api/users/123
// - Browser-testable: just open the URL
// - Massive ecosystem: Express, Fastify, every HTTP library
// - Human-readable payloads: JSON is text
// - Low barrier to entry: everyone knows HTTP + JSON
 
// Cons:
// - No generated types: you write clients manually
// - Schema drift: OpenAPI spec and code get out of sync
// - Verbose: HTTP headers, JSON keys repeated in every message
// - No native streaming: need WebSockets or SSE for real-time
 
// gRPC developer experience
 
// Pros:
// - Generated clients: proto file → TypeScript/Go/Python/etc.
// - Type safety: compiler catches mismatches between services
// - Built-in streaming: first-class support for all 4 patterns
// - Performance: binary serialization, HTTP/2 multiplexing
 
// Cons:
// - Not browser-native: need gRPC-Web proxy or Connect
// - Harder to debug: binary payloads aren't human-readable
// - Tooling complexity: protoc compiler, code generation pipeline
// - Smaller ecosystem: fewer middleware options than REST

// REST: easy to debug with standard tools
// $ curl -s http://localhost:3000/api/users/123 | jq
// {
//   "id": "user-123",
//   "name": "Alice",
//   "email": "alice@example.com"
// }
 
// gRPC: need specialized tools
// $ grpcurl -plaintext localhost:50051 users.UserService/GetUser
// Requires grpcurl, grpc_cli, or Postman gRPC support
// Binary payload must be decoded to be readable

// When to choose REST
const chooseREST = {
  publicAPI: true,               // Clients are external developers
  browserClients: true,          // Frontend applications consume the API
  simpleCRUD: true,              // Standard create/read/update/delete
  teamFamiliarity: 'REST',      // Team knows REST, not gRPC
  debuggability: 'important',    // Need to curl and inspect easily
  ecosystem: 'broad',           // Need extensive middleware options
};
 
// When to choose gRPC
const chooseGRPC = {
  serviceToService: true,        // Internal communication between services
  highThroughput: true,          // 10k+ requests/second between services
  streaming: true,               // Real-time data feeds, event streams
  polyglot: true,                // Services in different languages need type safety
  latencySensitive: true,        // Every millisecond matters
  strictContracts: true,         // Schema-first development with enforcement
};
 
// The hybrid approach: REST for external, gRPC for internal
const hybridArchitecture = {
  externalAPI: 'REST/JSON',      // Public-facing, browser-friendly
  apiGateway: 'translates',      // Converts REST ↔ gRPC at the boundary
  internalServices: 'gRPC',      // Type-safe, high-performance
  // Best of both worlds — but adds gateway complexity
};