/middy

🛵 The stylish Node.js middleware engine for AWS Lambda

Primary LanguageJavaScriptMIT LicenseMIT

Middy logo

The stylish Node.js middleware engine for AWS Lambda

npm version npm install size GitHub Actions test status badge
Known Vulnerabilities Standard Code Style Language grade: JavaScript Core Infrastructure Initiative (CII) Best Practices
Chat on Gitter Ask questions on StackOverflow

⚠️ NOTE: if you are upgrading from Middy 1.x, check out the upgrade instructions!

What is Middy

Middy is a very simple middleware engine that allows you to simplify your AWS Lambda code when using Node.js.

If you have used web frameworks like Express, then you will be familiar with the concepts adopted in Middy and you will be able to get started very quickly.

A middleware engine allows you to focus on the strict business logic of your Lambda and then attach additional common elements like authentication, authorization, validation, serialization, etc. in a modular and reusable way by decorating the main business logic.

Install

To install middy, you can use NPM:

npm install --save @middy/core

If you are using TypeScript, you will also want to make sure that you have installed the @types/aws-lambda peer-dependency:

npm install --save-dev @types/aws-lambda

Quick example

Code is better than 10,000 words, so let's jump into an example. Let's assume you are building a JSON API to process a payment:

//# handler.js #

// import core
import middy from '@middy/core' // esm Node v14+
//const middy = require('@middy/core') // commonjs Node v12+

// import some middlewares
import jsonBodyParser from '@middy/http-json-body-parser'
import httpErrorHandler from '@middy/http-error-handler'
import validator from '@middy/validator'

// This is your common handler, in no way different than what you are used to doing every day in AWS Lambda
const baseHandler = async (event, context) => {
 // we don't need to deserialize the body ourself as a middleware will be used to do that
 const { creditCardNumber, expiryMonth, expiryYear, cvc, nameOnCard, amount } = event.body

 // do stuff with this data
 // ...

 const response = { result: 'success', message: 'payment processed correctly'}
 return {statusCode: 200, body: JSON.stringify(response)}
}

// Notice that in the handler you only added base business logic (no deserialization,
// validation or error handler), we will add the rest with middlewares

const inputSchema = {
 type: 'object',
 properties: {
   body: {
     type: 'object',
     properties: {
       creditCardNumber: { type: 'string', minLength: 12, maxLength: 19, pattern: '\\d+' },
       expiryMonth: { type: 'integer', minimum: 1, maximum: 12 },
       expiryYear: { type: 'integer', minimum: 2017, maximum: 2027 },
       cvc: { type: 'string', minLength: 3, maxLength: 4, pattern: '\\d+' },
       nameOnCard: { type: 'string' },
       amount: { type: 'number' }
     },
     required: ['creditCardNumber'] // Insert here all required event properties
   }
 }
}

// Let's "middyfy" our handler, then we will be able to attach middlewares to it
const handler = middy(baseHandler)
  .use(jsonBodyParser()) // parses the request body when it's a JSON and converts it to an object
  .use(validator({inputSchema})) // validates the input
  .use(httpErrorHandler()) // handles common http errors and returns proper responses

module.exports = { handler }

Why?

One of the main strengths of serverless and AWS Lambda is that, from a developer perspective, your focus is mostly shifted toward implementing business logic.

Anyway, when you are writing a handler, you still have to deal with some common technical concerns outside business logic, like input parsing and validation, output serialization, error handling, etc.

Very often, all this necessary code ends up polluting the pure business logic code in your handlers, making the code harder to read and to maintain.

In other contexts, like generic web frameworks (fastify, hapi, express, etc.), this problem has been solved using the middleware pattern.

This pattern allows developers to isolate these common technical concerns into "steps" that decorate the main business logic code. Middleware functions are generally written as independent modules and then plugged into the application in a configuration step, thus not polluting the main business logic code that remains clean, readable, and easy to maintain.

Since we couldn't find a similar approach for AWS Lambda handlers, we decided to create middy, our own middleware framework for serverless in AWS land.

Usage

As you might have already seen from our first example here, using middy is very simple and requires just few steps:

  1. Write your Lambda handlers as usual, focusing mostly on implementing the bare business logic for them.
  2. Import middy and all the middlewares you want to use.
  3. Wrap your handler in the middy() factory function. This will return a new enhanced instance of your original handler, to which you will be able to attach the middlewares you need.
  4. Attach all the middlewares you need using the function .use(somemiddleware())

Example:

import middy from '@middy/core'
import middleware1 from 'sample-middleware1'
import middleware2 from 'sample-middleware2'
import middleware3 from 'sample-middleware3'

const baseHandler = (event, context) => {
  /* your business logic */
}

const handler = middy(baseHandler)

handler
  .use(middleware1())
  .use(middleware2())
  .use(middleware3())

module.exports = { handler }

.use() takes a single middleware or an array of middlewares, so you can attach multiple middlewares in a single call:

import middy from "@middy/core";
import middleware1 from "sample-middleware1";
import middleware2 from "sample-middleware2";
import middleware3 from "sample-middleware3";
const middlewares = [middleware1(), middleware2(), middleware3()]

const baseHandler = (event, context) => {
  /* your business logic */
};

const handler = middy(baseHandler);

handler.use(middlewares)

module.exports = { handler };

You can also attach inline middlewares by using the functions .before, .after and .onError.

For a more detailed use case and examples check the Writing a middleware section.

How it works

Middy implements the classic onion-like middleware pattern, with some peculiar details.

Middy middleware engine diagram

When you attach a new middleware this will wrap the business logic contained in the handler in two separate steps.

When another middleware is attached this will wrap the handler again and it will be wrapped by all the previously added middlewares in order, creating multiple layers for interacting with the request (event) and the response.

This way the request-response cycle flows through all the middlewares, the handler and all the middlewares again, giving the opportunity within every step to modify or enrich the current request, context, or the response.

Execution order

Middlewares have two phases: before and after.

The before phase, happens before the handler is executed. In this code the response is not created yet, so you will have access only to the request.

The after phase, happens after the handler is executed. In this code you will have access to both the request and the response.

If you have three middlewares attached (as in the image above), this is the expected order of execution:

  • middleware1 (before)
  • middleware2 (before)
  • middleware3 (before)
  • handler
  • middleware3 (after)
  • middleware2 (after)
  • middleware1 (after)

Notice that in the after phase, middlewares are executed in inverted order, this way the first handler attached is the one with the highest priority as it will be the first able to change the request and last able to modify the response before it gets sent to the user.

Interrupt middleware execution early

Some middlewares might need to stop the whole execution flow and return a response immediately.

If you want to do this you can invoke return response in your middleware.

Note: this will totally stop the execution of successive middlewares in any phase (before, after, onError) and returns an early response (or an error) directly at the Lambda level. If your middlewares do a specific task on every request like output serialization or error handling, these won't be invoked in this case.

In this example, we can use this capability for building a sample caching middleware:

// some function that calculates the cache id based on the current event
const calculateCacheId = event => {
  /* ... */
}
const storage = {}

// middleware
const cacheMiddleware = options => {
  let cacheKey

  const cacheMiddlewareBefore = async (request) => {
    cacheKey = options.calculateCacheId(request.event)
    if (options.storage.hasOwnProperty(cacheKey)) {
      // exits early and returns the value from the cache if it's already there
      return options.storage[cacheKey]
    }
  }

  const cacheMiddlewareAfter = async (request) => {
    // stores the calculated response in the cache
    options.storage[cacheKey] = request.response
  }

  return {
    before: cacheMiddlewareBefore,
    after: cacheMiddlewareAfter
  }
}

// sample usage
const handler = middy((event, context) => {
  /* ... */
}).use(
  cacheMiddleware({
    calculateCacheId,
    storage
  })
)

Handling errors

But, what happens when there is an error?

When there is an error, the regular control flow is stopped and the execution is moved back to all the middlewares that implemented a special phase called onError, following the order they have been attached.

Every onError middleware can decide to handle the error and create a proper response or to delegate the error to the next middleware.

When a middleware handles the error and creates a response, the execution is still propagated to all the other error middlewares and they have a chance to update or replace the response as needed. At the end of the error middlewares sequence, the response is returned to the user.

If no middleware manages the error, the Lambda execution fails reporting the unmanaged error.

// Initialize response
request.response = request.response ?? {}

// Add to response
request.response.add = 'more'

// Override an error
request.error = new Error('...')

// handle the error
return request.response

Writing a middleware

A middleware is an object that should contain at least 1 of 3 possible keys:

  1. before: a function that is executed in the before phase
  2. after: a function that is executed in the after phase
  3. onError: a function that is executed in case of errors

before, after and onError functions need to have the following signature:

async (request) => {
  // ...
}

Where:

  • request: is a reference to the current context and allows access to (and modification of) the current event (request), the response (in the after phase), and error (in case of an error).

Configurable middlewares

In order to make middlewares configurable, they are generally exported as a function that accepts a configuration object. This function should then return the middleware object with before, after, and onError as keys.

E.g.

// customMiddleware.js

const defaults = {}

module.exports = (opts = {}) => {
  const options = { ...defaults, ...opts }

  const customMiddlewareBefore = async (request) => {
    // might read options
  }
  const customMiddlewareAfter = async (request) => {
    // might read options
  }
  const customMiddlewareOnError = async (request) => {
    // might read options
  }

  return {
    // Having descriptive function names will allow for easier tracking of performance bottlenecks using @middy/core/profiler
    before: customMiddlewareBefore,
    after: customMiddlewareAfter,
    onError: customMiddlewareOnError
  }
}

With this convention in mind, using a middleware will always look like the following example:

import middy  from '@middy/core'
import customMiddleware from 'customMiddleware.js'

const handler = middy(async (event, context) => {
  // do stuff
  return {}
})

handler.use(
  customMiddleware({
    option1: 'foo',
    option2: 'bar'
  })
)

module.exports = { handler }

Inline middlewares

Sometimes you want to create handlers that serve a very small need and that are not necessarily re-usable. In such cases, you probably will need to hook only into one of the different phases (before, after or onError).

In these cases you can use inline middlewares which are shortcut functions to hook logic into Middy's control flow.

Let's see how inline middlewares work with a simple example:

import middy from '@middy/core'

const handler = middy((event, context) => {
  // do stuff
})

handler.before(async (request) => {
  // do something in the before phase
})

handler.after(async (request) => {
  // do something in the after phase
})

handler.onError(async (request) => {
  // do something in the on error phase
})

module.exports = { handler }

As you can see above, a middy instance also exposes the before, after and onError methods to allow you to quickly hook in simple inline middlewares.

Request caching & Internal storage

The handler also contains an internal object that can be used to store values securely between middlewares that expires when the event ends. To compliment this there is also a cache where middleware can store request promises. During before these promises can be stored into internal then resolved only when needed. This pattern is useful to take advantage of the async nature of node especially when you have multiple middleware that require reaching out the external APIs.

Here is a middleware boilerplate using this pattern:

import { canPrefetch, getInternal, processCache } from '@middy/util'

const defaults = {
  fetchData: {}, // { internalKey: params }
  disablePrefetch: false,
  cacheKey: 'custom',
  cacheExpiry: -1,
  setToContext: false
}

module.exports = (opts = {}) => {
  const options = { ...defaults, ...opts }

  const fetch = () => {
    const values = {}
    // Start your custom fetch
    for (const internalKey of Object.keys(options.fetchData)) {
      values[internalKey] = fetch('...', options.fetchData[internalKey]).then(res => res.text())
    }
    // End your custom fetch
    return values
  }

  let prefetch, client, init
  if (canPrefetch(options)) {
    init = true
    prefetch = processCache(options, fetch)
  }

  const customMiddlewareBefore = async (request) => {
    let cached
    if (init) {
      cached = prefetch
    } else {
      cached = processCache(options, fetch, request)
    }

    Object.assign(request.internal, cached)
    if (options.setToContext) Object.assign(request.context, await getInternal(Object.keys(options.fetchData), request))

    else init = false
  }

  return {
    before: customMiddlewareBefore
  }
}

More details on creating middlewares

Check the code for existing middlewares to see more examples on how to write a middleware.

TypeScript

Middy can be used with TypeScript with typings built in in every official package.

Here's an example of how you might be using Middy with TypeScript for a Lambda receiving events from API Gateway:

import middy from '@middy/core'
import { APIGatewayProxyEvent, APIGatewayProxyResult } from 'aws-lambda'

async function baseHandler (event: APIGatewayProxyEvent): Promise<APIGatewayProxyResult> {
  // the returned response will be checked against the type `APIGatewayProxyResult`
  return {
    statusCode: 200,
    body: `Hello from ${event.path}`
  }
}

let handler = middy(baseHandler)
handler
  .use(someMiddleware)
  .use(someOtherMiddleware)

export default handler

And here's an example of how you can write a custom middleware for a Lambda receiving events from API Gateway:

import middy from '@middy/core'
import { APIGatewayProxyEvent, APIGatewayProxyResult } from 'aws-lambda'

const middleware = (): middy.MiddlewareObj<APIGatewayProxyEvent, APIGatewayProxyResult> => {
  const before: middy.MiddlewareFn<APIGatewayProxyEvent, APIGatewayProxyResult> = async (
    request
  ): Promise<void> => {
    // Your middleware logic
  }

  const after: middy.MiddlewareFn<APIGatewayProxyEvent, APIGatewayProxyResult> = async (
    request
  ): Promise<void> => {
    // Your middleware logic
  }

  return {
    before,
    after
  }
}

export default middleware

Note: The Middy core team does not use TypeScript often and we can't certainly claim that we are TypeScript experts. We tried our best to come up with type definitions that should give TypeScript users a good experience. There is certainly room for improvement, so we would be more than happy to receive contributions 😊

See devDependencies for each middleware for list of dependencies that may be required with transpiling TypeScript.

Common Patterns and Best Practice

Tips and tricks to ensure you don't hit any performance or security issues. Did we miss something? Let us know.

ENV variables

Be sure to set AWS_NODEJS_CONNECTION_REUSE_ENABLED=1 when connecting to AWS services. This allows you to reuse the first connection established. See Reusing Connections with Keep-Alive in Node.js

Adding internal values to context

When all of your middlewares are done, and you need a value or two for your handler, this is how you get them:

import {getInternal} from '@middy/util'

middy(baseHandler)
  // Incase you want to add values on to internal directly
  .before((async (request) => {
    request.internal = {
      env: process.env.NODE_ENV
    }
  }))
  .use(sts(...))
  .use(ssm(...))
  .use(rdsSigner(...))
  .use(secretsManager(...))
  .before(async (request) => {
    // internal == { key: 'value' }

    // Map with same name
    Object.assign(request.context, await getInternal(['key'], request)) // context == { key: 'value'}

    // Map to new name
    Object.assign(request.context, await getInternal({'newKey':'key'}, request)) // context == { newKey: 'value'}

    // get all the values, only if you really need to, but you should only request what you need for the handler
    Object.assign(request.context, await getInternal(true, request)) // context == { key: 'value'}
  })

Connecting to RDS securely

First, you need to pass in a password. In order from most secure to least: RDS.Signer, SecretsManager, SSM using SecureString. SSM can be considered equally secure to SecretsManager if you have your own password rotation system.

Additionally, you will want to verify the RDS certificate and the domain of your connection. You can use this sudo code to get you started:

import tls from 'tls'
const ca = `-----BEGIN CERTIFICATE----- ...` // https://docs.aws.amazon.com/AmazonRDS/latest/UserGuide/UsingWithRDS.SSL.html

connectionOptions = {
  ...,
  ssl: {
    rejectUnauthorized: true,
      ca,
      checkServerIdentity: (host, cert) => {
      const error = tls.checkServerIdentity(host, cert)
      if (
        error &&
        !cert.subject.CN.endsWith('.rds.amazonaws.com')
      ) {
        return error
      }
    }
  }
}

Corresponding RDS.ParameterGroups values should be set to enforce TLS connections.

Bundling Lambda packages

If you're using serverless, checkout serverless-bundle. It's a wrapper around webpack, babel, and a bunch of other dependencies.

Keeping Lambda node_modules small

Using a bundler is the optimal solution, but can be complex depending on your setup. In this case you should remove excess files from your node_modules directory to ensure it doesn't have anything excess shipped to AWS. We put together a .yarnclean file you can check out and use as part of your CI/CD process.

Keeping your middlewares fast

There is a whole document on this, PROFILING.md.

FAQ

My lambda keep timing out without responding, what do I do?

Likely your event loop is not empty. This happens when you have a database connect still open for example. Checkout @middy/do-not-wait-for-empty-event-loop.

Available middlewares

These middleware focus on common use cases when using Lambda with other AWS services. Each middleware should do a single task. We try to balance each to be as performant as possible while meeting the majority of developer needs.

Misc

Request Transformation

  • http-content-negotiation: Parses Accept-* headers and provides utilities for content negotiation (charset, encoding, language and media type) for HTTP requests
  • http-header-normalizer: Normalizes HTTP header names to their canonical format
  • http-json-body-parser: Automatically parses HTTP requests with JSON body and converts the body into an object. Also handles gracefully broken JSON if used in combination of httpErrorHandler.
  • http-multipart-body-parser: Automatically parses HTTP requests with content type multipart/form-data and converts the body into an object.
  • http-urlencode-body-parser: Automatically parses HTTP requests with URL encoded body (typically the result of a form submit).
  • http-urlencode-path-parser: Automatically parses HTTP requests with URL encoded path.
  • s3-key-normalizer: Normalizes key names in s3 events.
  • sqs-json-body-parser: Parse body from SQS events
  • validator: Automatically validates incoming events and outgoing responses against custom schemas

Response Transformation

Fetch Data

Community generated middleware

The following middlewares are created and maintained outside this project. We cannot guarantee for its functionality. If your middleware is missing, feel free to open a Pull Request.

Version 2.x

  • middy-ajv: AJV validator optimized for performance
  • middy-sparks-joi: Joi validator
  • middy-idempotent: idempotency middleware for middy
  • middy-jsonapi: JSONAPI middleware for middy
  • middy-lesslog: Middleware for lesslog, a teeny-tiny and severless-ready logging utility
  • middy-rds: Creates RDS connection using knex or pg
  • middy-recaptcha: reCAPTCHA validation middleware
  • middy-event-loop-tracer: Middleware for dumping active tasks with their stacktraces in the event queue just before AWS Lambda function timeouts. So you can understand what was going on in the function when timeout happens.
  • middy-console-logger: Middleware for filtering logs printed over console logging methods. If the level of the console logging method is equal or bigger than configured level, the log is printed, Otherwise, it is ignored.
  • middy-invocation: Middleware for accessing current AWS Lambda invocation event and context from anywhere without need to passing event and context as arguments through your code.
  • middy-profiler: Middleware for profiling CPU on AWS Lambda during invocation and shows what methods/modules consume what percent of CPU time

Version 1.x

Middy's Influence

Have a similar project? Let us know.

A brief history of Middy

  • Middy was started in the early beginning of AWS Lambda.
  • 2017-08-03: First commit
  • 2017-09-04: v0.2.1 First release
  • 2018-05-20: v1.0.0-alpha
  • 2020-01-09: v1.0.0-beta
  • 2020-04-25: v1.0.0 Released
  • 2020 Review from @lmammino
  • 2020 Review from @willfarrell
  • 2021: v2.0.0 Coming soon
  • 2021-01-24: v2.0.0-alpha
  • 2021-03-12: v2.0.0-beta
  • 2021-04-01: v2.0.0

Fun Fact: The adding of the emoji-icon was the 2nd commit to the project.

Contributing

In the spirit of Open Source Software, everyone is very welcome to contribute to this repository. Feel free to raise issues or to submit Pull Requests.

Before contributing to the project, make sure to have a look at our Code of Conduct.

License

Licensed under MIT License. Copyright (c) 2017-2021 Luciano Mammino, will Farrell and the Middy team.

FOSSA Status