grpc-mod

gRPC-Mod

gRPC client that let's you manipulate the response objects and returns promises or observables for streams.

Table of contents

• Installation
• API by example
  • Enhancing the client
    • GRPCModClient
    • mod
  • Promises and observables
    • Unary requests
    • Server stream requests
    • Client stream requests
    • Bidirectional stream requests
  • Manipulating the response
    • Provided mods
      • noDefaults
      • noNilProps
      • enforceNumber
    • Creating custom mods
      • getTypeSchema
      • deepApplyMod

Installation

npm install grpc-mod

API by example

Consider the following .proto file

syntax = "proto3";

package test;

message GetRequest {}

message GetOneRequest {
  string id = 1;
}

message GetPageRequest {
  int32 from = 1;
  int32 count = 2;
}

message SaveTodoRequest {
  string label = 1;
}

message TodoResponse {
  string id = 1;
  string label = 2;
  int64 created = 3;
  int64 completed = 4;
}

message TodoListResponse {
  repeated TodoResponse todos = 1;
}


service TestService {
  rpc getTodo (GetOneRequest) returns (TodoResponse) {}
  rpc getAllTodos (GetRequest) returns (stream TodoResponse) {}
  rpc saveTodos (stream SaveTodoRequest) returns (TodoListResponse) {}
  rpc getTodoPages (stream GetPageRequest) returns (stream TodoListResponse) {}
}

Enhancing the client

There are two ways to enhance the client

1. You either create a new client with GRPCModClient

const client = new GRPCModClient(config)

1. enhance an existing one with the mod method

mod(client)

GRPCModClient

new GRPCModClient(config: { protoPath, packageName, serviceName, serverAddress, credentials [, options] } [, options])

config: required - the configuration object containing information for instantiating the client. it takes the following props:

• protoPath: required - path to the .proto file
• packageName: required - name of the package
• serviceName: required - name of the service we want to instantiate the client from
• serverAddress: required - address of the server we want our client to connect to
• credentials: required - credentials needed for connecting to the server
• options: optional - options to apply to channel creation

options: optional - the GRPCMod client options

• noDefaults: boolean: Nullifies the default values added to the response object (default is false) - example
• noNilProps: boolean: Removes null or undefined values from the response object (default is false) - example
• enforceNumber: boolean: Turns all int64 types into numbers (default is false) - example
• mods: function | [function] : one or a list of custom mods

const path = require('path')
const { credentials, } = require('grpc')
const { GRPCModClient, } = require('grpc-mod')

const PROTO_PATH = path.resolve(__dirname, '../pb/messages.proto')
const client = new GRPCModClient({
  protoPath: PROTO_PATH,
  packageName: 'test',
  serviceName: 'TestService',
  serverAddress: 'localhost:8080',
  credentials: credentials.createInsecure(),
})

mod

mod(client [, schema] [, options])

const path = require('path')
const { load, credentials, } = require('grpc')
const { mod, } = require('grpc-mod')

const PROTO_PATH = path.resolve(__dirname, '../pb/messages.proto')
const proto = load(PROTO_PATH).test

const client = new proto.TestService('localhost:8080', credentials.createInsecure())
mod(client)

The mod method is most useful when you only want to make use of promises and observables. If you want to write mods that need to know about the type of any of the properties, you would need a schema. GRPCModClient takes care of generating a schema for you, but if you really want to just use mod, for whatever reason, and still write mods that depend on a schema, you'd have to provide one yourself. You can either use protocol-buffers-schema or just write it yourself but it has to resemble the model that protocol-buffers-schema returns

NOTE: Some of the provided mods also depend on a schema and if you're using the mod method to enhance your client and activate one of those mods without providing a schema, the method will throw an error.

Promises and observables

Unary requests

// Simple promise
let doc
client
  .getTodo({ id: '5a54caf61bec98000f59dcbe' })
  .then(response => {
    doc = response
  })
  .catch(err => {
    console.log(err)
  })

// Async/await

let doc
try {
  doc = await client.getTodo({ id: '5a54caf61bec98000f59dcbe' })
} catch (err) {
  console.log(err)
}

Server stream requests

// Using observables
const observable = client.getAllTodos({})

observable.subscribe({
  next: (data) => console.log(data),
  error: (err) => console.log(err),
  complete: () => console.log('complete')
})

// Converting to Promise
const docs = await client.getAllTodos({}).toPromise()
// docs will be an array containing all the responses

Client stream requests

const todosToAdd = [...]

const stream = client.saveTodos()
// the stream has a getPromise method which you can use to listen for the response
const promise = stream.getPromise()

for(const label in todosToAdd) {
  sream.write({ label })
}
stream.end()

const docs = await promise // list of the added todos maybe?

Bidirectional stream requests

const stream = client.getTodoPages()

// the stream has a getObservable method
const observable = stream.getObservable()
observable.subscribe({
  next: data => renderTable(data.todos),
  err: err => console.log(err)
})

function onPaginationSelect(from, count) {
  stream.write({from, count })
}

Manipulating the response

You can manipulate the response by providing a configuration object as the second argument when creating or enhancing a client. You have access to a few mods provided by the library but you can always add your own. Keep in mind that when activating any of the provided mods, they will be executed before those you provide.

Provided mods

noDefaults

gRPC will add a default value for any of the props that are not present on the response object. In a client environment you might not expect a completed prop on a todo object that is not completed, but gRPC will actually set it to 0. Find out more about default values here.

The noDefaults flag will turn all the default values to null. You can further remove these props completely with noNilProps

...
const client = new GRPCModClient({
  protoPath: PROTO_PATH,
  packageName: 'test',
  serviceName: 'TestService',
  serverAddress: 'localhost:8080',
  credentials: credentials.createInsecure(),
}, {
  noDefaults: true
})

const doc = await client.getTodo({id: '5a54caf61bec98000f59dcbe'})

// noDefaults: false =>
// {
//  id: '5a54caf61bec98000f59dcbe',
//  label: 'Do something',
//  created: '1515862405277',
//  completed: 0
// }


// noDefaults: true =>
// {
//  id: '5a54caf61bec98000f59dcbe',
//  label: 'Do something',
//  created: '1515862405277',
//  completed: null
// }

noNilProps

Following the noDefaults example, we can add noNilProps: true to our configuration to remove the null/undefined props completely.

...
const client = new GRPCModClient({
  ...
}, {
  noDefaults: true,
  noNilProps: true
})

const doc = await client.getTodo({id: '5a54caf61bec98000f59dcbe'})

// noDefaults: true, noNilProps: true =>
// {
//  id: '5a54caf61bec98000f59dcbe',
//  label: 'Do something',
//  created: '1515862405277'
// }

NOTE: You need to first turn the defaults to null with noDefaults so they will get excluded by noNilProps*

enforceNumber

You might have noticed in the previous examples that created is a string. We asked for int64 and it should be a Long object but what we ultimately want is a number. Read about this design here

The enforceNumber flag will turn all int64 values to numbers. There might be more "issues" like this, but this is the one I bumped into and I seriously have no idea about any other data-types not suported by JS.

...
const client = new GRPCModClient({
  ...
}, {
  noDefaults: true,
  noNilProps: true,
  enforceNumber: true
})

const doc = await client.getTodo({id: '5a54caf61bec98000f59dcbe'})

// noDefaults: true, noNilProps: true =>
// {
//  id: '5a54caf61bec98000f59dcbe',
//  label: 'Do something',
//  created: 1515862405277
// }

Creating custom mods

The configuration object accepts a 'mods' prop where you can specify one or a list of mods that you build.

...
const client = new GRPCModClient({
  ...
}, {
  noDefaults: true,
  noNilProps: true,
  enforceNumber: true,
  mods: myMod // or a list: [myMod2, myMod1]
})

Custom mods get applied from right to left. So, if myMod2 expects a response type returned by myMod1, you'd have to list them in the right order [myMod2, myMod1]. Also, any of the provided mods will run before yours.

Mods are simple functions that take 2 arguments: the response object and a schema object. The response object is exactly what it says, the response you get from the server. The schema object is not the entire schema generated from the .proto file, but only the part relevant to the response and only the top level. E.g.:

function myMod(response, schema) {
  console.log(reponse)
  // {
  //  id: '5a54caf61bec98000f59dcbe',
  //  label: 'Do something',
  //  created: 1515862405277
  // }

  console.log(schema)
  // { id:
  //    { name: 'id',
  //      type: 'string',
  //      tag: 1,
  //      map: null,
  //      oneof: null,
  //      required: false,
  //      repeated: false,
  //      options: {}, },
  // label:
  //    { name: 'label',
  //      type: 'string',
  //      tag: 2,
  //      map: null,
  //      oneof: null,
  //      required: false,
  //      repeated: false,
  //      options: {}, },
  // created:
  //    { name: 'created',
  //      type: 'int64',
  //      tag: 3,
  //      map: null,
  //      oneof: null,
  //      required: false,
  //      repeated: false,
  //      options: {}, },
  // completed:
  //    { name: 'completed',
  //      type: 'int64',
  //      tag: 4,
  //      map: null,
  //      oneof: null,
  //      required: false,
  //      repeated: false,
  //      options: {},
  //    },
  // }
}

The schema is generated from the .proto file with protocol-buffers-schema and is only relevant to the top level of the response object, meaning that if you have property of a custom type, the schema of that type will not be available. This prevents the need of generating deeply nested schemas and eventual circular dependencies, like in the case of a User type that has a field called friends being a repeated User type. Those friends would also be of type User and also have friends of their own and so on.

Our saveTodos method is a client-stream request ant it resolves with a TodoListResponse. The TodoListResponse is an object with a todos property of type repeated TodoResponse (an array of todos).

function myMod(response, schema) {
  console.log(schema)
  // { todos:
  //    { name: 'todos',
  //      type: 'TodoResponse',
  //      tag: 1,
  //      map: null,
  //      oneof: null,
  //      required: false,
  //      repeated: true,
  //      options: {},
  //    },
  // }
}

getTypeSchema

If we want to modify all the todos in that list, we'd have to map over data. This is pretty trivial if we don't need to check the type of the todo's properties, but if we want to manipulate a value based on the type of that property, we'd need access to the TodoResponse schema.

There is, however, a method to retrieve the schema of a property's type, getTypeSchema. This method is part of the schema object of every type and we'll see how it works by exploring how you'd implement a mod similar to enforceNumber.

function enforceNumber(response, schema){
  // response is { todos: [{...}, {...}, {...}] }

  const todoSchema = schema['todos'].getTypeSchema()

  const moddedList = response.todos.map(todo => {
    // reduce the todo entries and create a new object
    return Object.entries(todo).reduce((acc, [key, value]) => {
      // if the current property is of type `int64` we parse it into a number
      return {
        ...acc,
        [key]: todoSchema[key].type === 'int64' ? parseInt(value) : value
      }
    }, {})
  })

  // Return the modified response object,
  // or you can just return the list, but I recommend you stick with the contract
  return { todos: moddedList }
}

NOTE: This is not how enforceNumber is written. The actual implementation recursively handles nested response objects and are partially applied with the schema argument. You can read through the source if you're interested how it all works.

deepApplyMod

deepApplyMod(response [, schema], transformFn [, predicate])

This is a helper function, provided by 'grpc-mod' which you can use to apply mods to nested response objects.

Let's say that we want to transform created and completed props from unix to date strings.

function unixToDateString(response) {
  return {
    ...response,
    created: new Date(response.created).toString(),
    completed: new Date(response.completed).toString()
  }
}

And then we add this function to the mods. Easy, right? This is enough if we only ever get a TodoResponse, but in the case of a TodoListResponse, it won't get applied. Remember, TodoListResponse looks like:

{ data: [TodoResponse, TodoResponse, TodoResponse, ...]}

so there will be no 'created' or 'completed' props. In fact, with the above mod, those would get added alongside data. We'll use deepApplyMod to solve this issue, providing a transformation function and a predicate to only apply it to the created and completed keys.

The transformation function takes a single argument, and it's the value of the currently iterated property. The predicate is optional (default true), and it's a function that takes 2 arguments, the key and the value of the property.

import { deepApplyMod } from 'grpc-mod'

// This is the transformation function that will get applied to the properties
const convertToDateString =(value) => new Date(value).toString()

// We want to also provide a predicate to apply that transformation only to the `created` and `completed` keys
const isDateProp = (key, value) => ['created', 'completed'].includes(key)

function unixToDateString(response, schema) {
  return deepApplyMod(response, null, convertToDateString, isDateProp)
}