Skip to content

Latest commit

 

History

History
386 lines (300 loc) · 16.5 KB

README.md

File metadata and controls

386 lines (300 loc) · 16.5 KB

Node.js Lambda Connector

The Node.js Lambda connector allows you to expose TypeScript functions as NDC functions/procedures for use in your Hasura DDN subgraphs.

How to Use

Tip

The following instructions are just a quick summary of how to use the Node.js Lambda connector. To see it in use in a wider Hasura DDN project, and to understand the underlying DDN concepts, please check out the Hasura DDN Getting Started Guide. The Node.js Lambda connector is used in the business logic part of that tutorial.

First, ensure you have Node.js v20+ installed and an existing Hasura DDN project created (see the Hasura DDN Getting Started Guide). Then, initialize the connector into that project:

ddn connector init -i
  • Select hasura/nodejs from the list of connectors
  • Name it something descriptive. For this example, we'll call it my_ts
  • Choose a port (press enter to accept the default recommended by the CLI).

This will generate the necessary files into the my_subgraph/connector/my_ts directory. This creates a functions.ts file in which you will write your functions, and a package.json with the ndc-lambda-sdk installed into it.

Restore all the npm packages required to run the connector by running inside the connector's directory:

npm install

To run the connector with the required environment variables applied, you can run the following command:

ddn connector setenv --connector connector.yaml -- npm run watch

This starts the connector in watch mode, which watches for code changes and restarts the connector when they are detected. npm run start can be used instead to just start the connector without watching for changes. Both start and watch are defined in the package.json's scripts section and use the ndc-lambda-sdk to host your functions.ts file.

Once you have written your functions, use the following to generate the related metadata that will link together any functions in this functions.ts file and your API:

ddn connector introspect my_ts

Then you can generate all the necessary hml files with the Commands that expose your functions in the API by running the following:

ddn connector-link add-resources my_ts

To make a local build of your supergraph you can run:

ddn supergraph build local

You can then run that build locally for testing by starting the engine and other connectors in the DDN project using Docker Compose:

HASURA_DDN_PAT=$(ddn auth print-pat) docker compose --env-file .env up --build --watch

You can view and query that local instance using the Hasura Graphiql Explorer by running:

ddn console --local

Functions

Any functions exported from functions.ts are made available as NDC functions/procedures to use in your Hasura metadata and expose as GraphQL fields in queries or mutation.

If you write a function that performs a read-only operation, you should mark it with the @readonly JSDoc tag, and it will be exposed as an NDC function, which will ultimately show up as a GraphQL query field in Hasura.

/** @readonly */
export function add(x: number, y: number): number {
  return x + y;
}

Functions without the @readonly JSDoc tag are exposed as NDC procedures, which will ultimately show up as a GraphQL mutation field in Hasura.

Arguments to the function end up being field arguments in GraphQL and the return value is what the field will return when queried. Every function must return a value; void, null or undefined is not supported.

/** @readonly */
export function hello(name: string, year: number): string {
  return `Hello ${name}, welcome to ${year}`
}

Async functions are supported:

type HttpStatusResponse = {
  code: number
  description: string
}

export async function test(): Promise<string> {
  const result = await fetch("http://httpstat.us/200")
  const responseBody = await result.json() as HttpStatusResponse;
  return responseBody.description;
}

If you'd like to split your functions across multiple files, do so, then simply re-export them from functions.ts like so:

export * from "./another-file-1"
export * from "./another-file-2"

The ndc-lambda-sdk uses the TypeScript types specified for your function arguments and return types to automatically derive the NDC schema (function/procedure schema, scalar types and object types) for your functions. This makes the available in your Hasura DDN metadata, where they can be bound to Commands and thereby exposed in your GraphQL supergraph.

Supported types

The basic scalar types supported are:

  • string (NDC scalar type: String)
  • number (NDC scalar type: Float)
  • boolean (NDC scalar type: Boolean)
  • bigint (NDC scalar type: BigInt, represented as a string in JSON)
  • Date (NDC scalar type: DateTime, represented as an ISO formatted string in JSON)

You can also import JSONValue from the SDK and use it to accept and return arbitrary JSON. Note that the value must be serializable to JSON.

import * as sdk from "@hasura/ndc-lambda-sdk"

export function myFunc(json: sdk.JSONValue): sdk.JSONValue {
  const propValue =
    json.value instanceof Object && "prop" in json.value && typeof json.value.prop === "string"
      ? json.value.prop
      : "default value";
  return new sdk.JSONValue({prop: propValue});
}

null, undefined and optional arguments/properties are supported:

export function myFunc(name: string | null, age?: number): string {
  const greeting = name != null
    ? `hello ${name}`
    : "hello stranger";
  const ageStatement = age !== undefined
    ? `you are ${age}`
    : "I don't know your age";

  return `${greeting}, ${ageStatement}`;
}

However, any undefineds in the return type will be converted to nulls, as GraphQL does not have the concept of undefined.

Object types and interfaces are supported. The types of the properties defined on these must be supported types.

type FullName = {
  title: string
  firstName: string
  surname: string
}

interface Greeting {
  polite: string
  casual: string
}

export function greet(name: FullName): Greeting {
  return {
    polite: `Hello ${name.title} ${name.surname}`,
    casual: `G'day ${name.firstName}`
  }
}

Arrays are also supported, but can only contain a single type (tuple types are not supported):

export function sum(nums: number[]): number {
  return nums.reduce((prev, curr) => prev + curr, 0);
}

Anonymous types are supported, but will be automatically named after the first place they are used. It is recommended that you avoid using anonymous types. Instead, prefer to name all your types to ensure the type name does not change unexpectedly as you rename usage sites and re-order usages of the anonymous type.

export function greet(
  name: { firstName: string, surname: string } // This type will be automatically named greet_name
): string {
  return `Hello ${name.firstName} ${name.surname}`;
}

Unsupported types

These types are unsupported as function parameter types or return types for functions that you export for invocation from Hasura. You can use whatever types you like inside your function or in related code, however.

Relaxed Types

"Relaxed types" are types that are otherwise unsupported, but instead of being rejected are instead converted into opaque custom scalar types. These scalar types are entirely unvalidated when used as input (ie. the caller of the function can send arbitrary JSON values), making it incumbent on the function itself to ensure the incoming value for that relaxed type actually matches its type. Because relaxed types are represented as custom scalar types, in GraphQL you will be unable to select into the type, if it is an object, and will only be able to select the whole thing.

Relaxed types are designed to be an escape hatch to help people get up and running using existing code quickly, where their existing code uses types that are unsupported. They are not intended to be used long term. You should prefer to modify your code to use only supported types. To opt into using relaxed types, one must apply the @allowrelaxedtypes JSDoc tag to the function that will be using the unsupported types.

The following unsupported types are allowed when using relaxed types, and will be converted into opaque unvalidated scalar types:

  • Union types
  • Tuple types
  • Types with index signatures
  • The any and unknown types
  • Enum types

Here's an example of a function that uses some relaxed types:

/**
 * @allowrelaxedtypes
 * @readonly
 */
export function findEmptyRecords(record: Record<string, string>): { emptyKeys: string[] } | string {
  const emptyKeys: string[] = [];
  const entries = Object.entries(record);

  if (entries.length === 0)
    return "Error: record was empty";

  for (const [key, value] of entries) {
    if (value === "")
      emptyKeys.push(key);
  }

  return { emptyKeys };
}

Error handling

By default, unhandled errors thrown from functions are caught by the Lambda SDK host, and an InternalServerError is returned to Hasura. The details of the uncaught error (the message and stack trace) is captured and will be logged in the OpenTelemetry trace associated with the GraphQL request. However, the GraphQL API caller will receive a generic "internal error" response to their query. This is to ensure internal error details are not leaked to GraphQL API clients.

If you want to return specific error details to the GraphQL API client, you can deliberately throw one of the below error classes (these error types correspond with the error status codes in the NDC Specification):

Error Class Used When
sdk.Forbidden A permission check failed - for example, a mutation might fail because a check constraint was not met.
sdk.Conflict A conflicting state would be created for the data source - for example, a mutation might fail because a foreign key constraint was not met.
sdk.UnprocessableContent There was something semantically incorrect in the request. For example, an invalid value for a function argument was received.
import * as sdk from "@hasura/ndc-lambda-sdk"

/** @readonly */
export function divide(x: number, y: number): number {
  if (y === 0) {
    throw new sdk.UnprocessableContent("Cannot divide by zero", { myErrorMetadata: "stuff", x, y })
  }
  return x / y;
}

The GraphQL API will return the error from the API looking similar to this:

{
  "data": null,
  "errors": [
    {
      "message": "ndc: Cannot divide by zero",
      "path": ["divide"],
      "extensions": {
        "details": {
          "myErrorMetadata": "stuff",
          "x": 10,
          "y": 0
        }
      }
    }
  ]
}

If you must include stack traces in the GraphQL API response, you can collect and add them to the error details yourself using a helper function (sdk.getErrorDetails). However, it is not recommended to expose stack traces to API end users. Instead, API administrators can look in the GraphQL API tracing to find the stack traces logged.

import * as sdk from "@hasura/ndc-lambda-sdk"

/** @readonly */
export function getStrings(): Promise<string[]> {
  try {
    return await queryForStrings();
  } catch (e) {
    const details = e instanceof Error
      ? sdk.getErrorDetails(e) // Returns { message: string, stack: string }
      : {};
    throw new sdk.UnprocessableContent("Something went wrong :/", details)
  }
}

Parallel execution

If functions are involved remote relationships in your Hasura metadata, then they may be queried in a batch-based fashion. In this situation, any async functions that are marked with the @readonly JSDoc tag may be executed in parallel. The default degree of parallelism per query request to the connector is 10, but you may customise this by using the @paralleldegree JSDoc tag on your function.

/**
 * This function will only run up to 5 http requests in parallel per query
 *
 * @readonly
 * @paralleldegree 5
 */
export async function test(statusCode: number): Promise<string> {
  const result = await fetch("http://httpstat.us/${statusCode}")
  const responseBody = await result.json() as any;
  return responseBody.description;
}

Non-readonly functions are not invoked in parallel within the same mutation request to the connector, so it is invalid to use the @paralleldegree JSDoc tag on those functions.

Documentation

JSDoc comments on your functions and types are used to provide descriptions for types exposed in your GraphQL schema. For example:

/** Different types of greetings */
interface Greeting {
  /** A greeting if you want to be polite */
  polite: string
  /** A casual-toned greeting */
  casual: string
}

/**
 * Creates a greeting string using the specified name
 *
 * @param title The person's title, for example, Mr or Mrs
 * @param firstName The person's first name
 * @param lastName The person's last name (surname)
 * @readonly
 */
export function greet(title: string, firstName: string, lastName: string): Greeting {
  return {
    polite: `Hello ${title} ${lastName}`,
    casual: `G'day ${firstName}`
  }
}

Descriptions are collected for:

  • Functions
  • Function parameters
  • Types
  • Type properties

Tracing

Your functions are automatically instrumented with OpenTelemetry traces that Hasura will capture.

By default, the following spans are emitted:

  • handleQuery/handleMutation - wraps the request from the Hasura DDN to the connector
  • prepare arguments - wraps the process of preparing arguments to pass to your function
  • function invocation - wraps a (potentially) parallel function invocation during a query
  • Function: <function name> - wraps the invocation of your function
  • reshape result - wraps the projection of the function result to match the requirements of the GraphQL selection set

If you want to add additional spans around your own code, you can do so by using the OpenTelemetry SDK and withActiveSpan from ndc-lambda-sdk:

import opentelemetry from '@opentelemetry/api';
import * as sdk from "@hasura/ndc-lambda-sdk"

const tracer = opentelemetry.trace.getTracer("my functions"); // Name your functions service here

export async function doSomething(): Promise<string> {
  const spanAttributes = { myAttribute: "value" };
  return await sdk.withActiveSpan(tracer, "my span name", async () => {
    return await doSomethingExpensive();
  }, spanAttributes);
}

The span will be wrapped around the function you pass to sdk.withActiveSpan. The function can optionally be an async function that returns a Promise, and if so, the span will be ended when the Promise resolves.