Asynchronous HTTP microservices


Keywords
micro, micri, service, microservice, API
License
MIT
Install
npm install micri@3.0.1

Documentation

Micri — Asynchronous HTTP microservices

micri is an archaic non-SI decimal metric prefix for 10−14. Its symbol was mc.

Wikipedia - Micri-

npm version Install Size

Features

  • Easy: Designed for usage with async and await (more)
  • Fast: Ultra-high performance (even JSON parsing is opt-in)
  • Micri: The whole project is ~500 lines of code
  • Agile: Super easy deployment and containerization
  • Simple: Oriented for single purpose modules (function)
  • Standard: Just HTTP!
  • Explicit: No middleware - modules declare all dependencies
  • Lightweight: Install Size

Usage

const { serve } = require('micri')

const sleep = (ms) => new Promise((r) => setTimeout(r, ms));

const server = serve(async (req, res) => {
  await sleep(500)
  return 'Hello world'
})

server.listen(3000)

And go to this URL: http://localhost:3000 - 🎉

async & await

Examples

Micri is built for usage with async/await. You can read more about async / await here

const sleep = (ms) => new Promise((r) => setTimeout(r, ms));

module.exports = async (req, res) => {
  await sleep(500);
  return 'Ready!';
}

Body parsing

Examples

For parsing the incoming request body we included an async functions buffer, text and json

const {buffer, text, json} = require('micri')

module.exports = async (req, res) => {
  const buf = await buffer(req)
  console.log(buf)
  // <Buffer 7b 22 70 72 69 63 65 22 3a 20 39 2e 39 39 7d>
  const txt = await text(req)
  console.log(txt)
  // '{"price": 9.99}'
  const js = await json(req)
  console.log(js.price)
  // 9.99
  return ''
}

Routing

Micri has a simple built-in function router. The idea is fairly simple, you can use it as a wrapper virtually anywhere where it will be called with (req, res, optionalArgs) and can return a promise as a response to micri().

Firstly you create a router by calling the router(...) function. The router function takes routes as arguments. Routes are created by calling functions under on map, and the functions are organized there by HTTP method name. These functions in turn take two arguments, a predicate and request handler functions.

A predicate function gets the usual arguments (req, res, opts?). A predicate function may return a truthy value if the handler function should take care of this request, or it may return a falsy value if the handler should not take this request.

Multiple predicates can be combined by using Router.everyPredicate(...) that takes predicate functions as arguments. The function returns true if every predicate function given as an argument returns true.

The order of the route arguments marks the priority order of the routes. Therefore if two routes would match to a request the one that was passed earlier in the arguments list to the router() function will handle the request.

otherwise() is a special route function that will always match and thus can be used as the last route rule for sending an error and avoid throwing an exception in case no other route predicate matches.

const { Router: { router } } = require('micri');

micri(router(
	on.get((req) => req.url === '/', (req, _res) => ({ message: 'Hello world!'})),
	on.post((req) => req.url === '/', (req) => text(req)),
	otherwise((req, res) => send(res, 400, 'Method Not Accepted'))))
	.listen(3000);

Worker Threads

Micri supports offloading computationally heavy request handlers to worker threads seamlessly. The offloading is configured per handler by wrapping the handler function with withWorker(). It works directly at the top-level or per route when using the router. See with-workerthreads for a couple of examples how to use it.

micri(withWorker(() => doSomethingCPUHeavy))

Offloading requests to a worker may improve the responsiveness of a busy API significantly, as it removes almost all blocking from the main thread. In the following examples we first try to find prime numbers and finally return one as a response. In both cases we do two concurrent HTTP GET requests using curl.

Finding prime numbers using the main thread:

~% time curl 127.0.0.1:3000/main
299993curl 127.0.0.1:3000/main  0.01s user 0.00s system 0% cpu 8.791 total
~% time curl 127.0.0.1:3000/main
299993curl 127.0.0.1:3000/main  0.00s user 0.00s system 0% cpu 16.547 total

Notice that the second curl needs to wait until the first request finishes.

Finding prime numbers using a worker thread:

~% time curl 127.0.0.1:3000/worker
299993curl 127.0.0.1:3000/worker  0.00s user 0.00s system 0% cpu 9.025 total
~% time curl 127.0.0.1:3000/worker
299993curl 127.0.0.1:3000/worker  0.00s user 0.00s system 0% cpu 9.026 total

Note how both concurrently executed requests took the same time to finish.

API

buffer(req, { limit = '1mb', encoding = 'utf8' })
text(req, { limit = '1mb', encoding = 'utf8' })
json(req, { limit = '1mb', encoding = 'utf8' })
  • Buffers and parses the incoming body and returns it.
  • Exposes an async function that can be run with await.
  • Can be called multiple times, as it caches the raw request body the first time.
  • limit is how much data is aggregated before parsing at max. Otherwise, an Error is thrown with statusCode set to 413 (see Error Handling). It can be a Number of bytes or a string like '1mb'.
  • If JSON parsing fails, an Error is thrown with statusCode set to 400 (see Error Handling)

For other types of data check the examples

Sending a different status code

So far we have used return to send data to the client. return 'Hello World' is the equivalent of send(res, 200, 'Hello World').

const {send} = require('micri')

module.exports = async (req, res) => {
  const statusCode = 400
  const data = { error: 'Custom error message' }

  send(res, statusCode, data)
}
send(res, statusCode, data = null)
  • Use require('micri').send.
  • statusCode is a Number with the HTTP status code, and must always be supplied.
  • If data is supplied it is sent in the response. Different input types are processed appropriately, and Content-Type and Content-Length are automatically set.
    • Stream: data is piped as an octet-stream. Note: it is your responsibility to handle the error event in this case (usually, simply logging the error and aborting the response is enough).
    • Buffer: data is written as an octet-stream.
    • object: data is serialized as JSON.
    • string: data is written as-is.
  • If JSON serialization fails (for example, if a cyclical reference is found), a 400 error is thrown. See Error Handling.
micri(fn)
  • This function is exposed as the default export.
  • Use require('micri').
  • Returns a http.Server that uses the provided function as the request handler.
  • The supplied function is run with await. So it can be async
sendError(req, res, error)
  • Use require('micri').sendError.
  • Used as the default handler for errors thrown.
  • Automatically sets the status code of the response based on error.statusCode.
  • Sends the error.message as the body.
  • Stacks are printed out with console.error and during development (when NODE_ENV is set to 'development') also sent in responses.
  • Usually, you don't need to invoke this method yourself, as you can use the built-in error handling flow with throw.

Error Handling

Micri allows you to write robust microservices. This is accomplished primarily by bringing sanity back to error handling and avoiding callback soup.

If an error is thrown and not caught by you, the response will automatically be 500. Important: Error stacks will be printed as console.error and during development mode (if the env variable NODE_ENV is 'development'), they will also be included in the responses.

If the error object throw is an instance of MicriError the message, statusCode and code properties of the object are used for the HTTP response.

Let's say you want to write a rate limiting module:

const rateLimit = require('my-rate-limit')

micri((req, res) => {
  await rateLimit(req);
  // ... your code
}).listen(3000);

If the API endpoint is abused, it can throw a MicriError like so:

if (tooMany) {
  throw MicriError(429, 'rate_limited' 'Rate limit exceeded');
}

The nice thing about this model is that the statusCode is merely a suggestion. The user can override it:

try {
  await rateLimit(req)
} catch (err) {
  if (429 == err.statusCode) {
    // perhaps send 500 instead?
    send(res, 500);
  }
}

If the error is based on another error that Micri caught, like a JSON.parse exception, then originalError will point to it. If a generic error is caught, the status will be set to 500.

In order to set up your own error handling mechanism, you can use composition in your handler:

const {send} = require('micri');

const handleErrors = fn => async (req, res) => {
  try {
    return await fn(req, res)
  } catch (err) {
    console.log(err.stack)
    send(res, 500, 'My custom error!')
  }
}

micri(handleErrors(async (req, res) => {
  throw new Error('What happened here?')
})).listen(3000);

Contributing

  1. Fork this repository to your own GitHub account and then clone it to your local device
  2. Link the package to the global module directory: npm link
  3. Within the module you want to test your local development instance of Micri, just link it to the dependencies: npm link micri. Instead of the default one from npm, node will now use your clone of Micri!