The stylish Node.js middleware engine for AWS Lambda
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.
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
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 }
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.
As you might have already seen from our first example here, using middy is very simple and requires just few steps:
- Write your Lambda handlers as usual, focusing mostly on implementing the bare business logic for them.
- Import
middy
and all the middlewares you want to use. - 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. - 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.
Middy implements the classic onion-like middleware pattern, with some peculiar details.
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.
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.
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
})
)
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
A middleware is an object that should contain at least 1 of 3 possible keys:
before
: a function that is executed in the before phaseafter
: a function that is executed in the after phaseonError
: 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 currentevent
(request), theresponse
(in the after phase), anderror
(in case of an error).
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 }
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.
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
}
}
Check the code for existing middlewares to see more examples on how to write a middleware.
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.
Tips and tricks to ensure you don't hit any performance or security issues. Did we miss something? Let us know.
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
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'}
})
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.
If you're using serverless, checkout serverless-bundle
.
It's a wrapper around webpack, babel, and a bunch of other dependencies.
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.
There is a whole document on this, PROFILING.md.
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
.
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.
error-logger
: Logs errorsinput-output-logger
: Logs request and responsedo-not-wait-for-empty-event-loop
: Sets callbackWaitsForEmptyEventLoop property to falsecloudwatch-metrics
: Hydrates lambda'scontext.metrics
property with an instance of AWS MetricLoggerwarmup
: Used to pre-warm a lambda function
http-content-negotiation
: ParsesAccept-*
headers and provides utilities for content negotiation (charset, encoding, language and media type) for HTTP requestshttp-header-normalizer
: Normalizes HTTP header names to their canonical formathttp-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 ofhttpErrorHandler
.http-multipart-body-parser
: Automatically parses HTTP requests with content typemultipart/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 eventsvalidator
: Automatically validates incoming events and outgoing responses against custom schemas
http-cors
: Sets HTTP CORS headers on responsehttp-error-handler
: Creates a proper HTTP response for errors that are created with the http-errors module and represents proper HTTP errors.http-event-normalizer
: Normalizes HTTP events by adding an empty object forqueryStringParameters
,multiValueQueryStringParameters
orpathParameters
if they are missing.http-security-headers
: Applies best practice security headers to responses. It's a simplified port of HelmetJS.http-partial-response
: Filter response objects attributes based on query string parameters.http-response-serializer
: HTTP response serializer.sqs-partial-batch-failure
: handles partially failed SQS batches.
rds-signer
: Fetches token for connecting to RDS with IAM users.s3-object-response
: Gets and write S3 object response.secrets-manager
: Fetches parameters from AWS Secrets Manager.ssm
: Fetches parameters from AWS Systems Manager Parameter Store.sts
: Fetches credentials to assumes IAM roles for connection to other AWS services.
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.
- 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
orpg
- 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
- middy-redis: Redis connection middleware
- middy-extractor: Extracts data from events using expressions
- @keboola/middy-error-logger: middleware that catches thrown exceptions and rejected promises and logs them comprehensibly to the console
- @keboola/middy-event-validator: Joi powered event validation middleware
- middy-reroute: provides complex redirect, rewrite and proxying capabilities by simply placing a rules file into your S3 bucket
- middytohof: Convert Middy middleware plugins to higher-order functions returning lambda handlers
- wrap-ware: A middleware wrapper which works with promises / async
- middy-middleware-warmup: A middy plugin to help keep your Lambdas warm during Winter
- @sharecover-co/middy-aws-xray-tracing: AWS X-Ray Tracing Middleware
- @sharecover-co/middy-http-response-serializer: This middleware serializes the response to JSON and wraps it in a 200 HTTP response
- @seedrs/middyjs-middleware: Collection of useful middlewares
- middy-autoproxyresponse: A middleware that lets you return simple JavaScript objects from Lambda function handlers and converts them into LAMBDA_PROXY responses
- jwt-auth: JSON web token authorization middleware based on
express-jwt
- middy-mongoose-connector: MongoDB connection middleware for mongoose.js
- @ematipico/middy-request-response: a middleware that creates a pair of request/response objects
- @marcosantonocito/middy-cognito-permission: Authorization and roles permission management for the Middy framework that works with Amazon Cognito
- middy-env: Fetch, validate and type cast environment variables
- sqs-json-body-parser: Parse the SQS body to JSON
- middy-lesslog: Middleware for
lesslog
, a teeny-tiny and severless-ready logging utility
- .Net port Voxel.MiddyNet @vgaltes
- GoLang port Vesper
Have a similar project? Let us know.
- 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.
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.
Licensed under MIT License. Copyright (c) 2017-2021 Luciano Mammino, will Farrell and the Middy team.