Skip to content

Opinionated Node.js microservices framework built on Fastify ⚡️

License

Notifications You must be signed in to change notification settings

47ng/fastify-micro

Repository files navigation

fastify-micro

NPM MIT License Continuous Integration Coverage Status

Opinionated Node.js microservices framework built on Fastify.

Features

  • Secure and useful logging
  • Auto-load routes & plugins from the filesystem (opt-in)
  • Built-in Sentry support for error reporting (opt-in)
  • Service health monitoring
  • Graceful exit
  • First class TypeScript support

Installation

$ yarn add fastify-micro
# or
$ npm i fastify-micro

Usage

Minimal example:

import { createServer, startServer } from 'fastify-micro'

const server = createServer()

startServer(server)

Documentation

Environment Variables

Details of the required and accepted (optional) environment variables are available in the .env.example file.

Listening port

You can provide the port number where the server will be listening as the second argument of startServer:

import { createServer, startServer } from 'fastify-micro'

const server = createServer()
startServer(server, 3000)

If omitted, the port number will be read from the PORT environment variable:

// process.env.PORT = 4000

import { createServer, startServer } from 'fastify-micro'

const server = createServer()
startServer(server)

// Server started on 0.0.0.0:4000

If no value is specified either via code or environment, the default port will be 3000.

Auto-loading plugins and routes from the filesystem

Plugins and routes can be loaded from the filesystem using fastify-autoload:

import path from 'path'
import { createServer } from 'fastify-micro'

createServer({
  plugins: {
    dir: path.join(__dirname, 'plugins')
  },
  routes: {
    dir: path.join(__dirname, 'routes')
  }
})

The plugins and routes options are fastify-autoload configuration objects.

As recommended by Fastify, plugins will be loaded first, then routes. Attach your external services, decorators and hooks as plugin files, so that they will be loaded when declaring your routes.

Printing Routes

In development, the server will log the route tree on startup. This can be configured:

createServer({
  printRoutes:
    | 'auto'    // default: `console` in development, silent in production.
    | 'console' // always pretty-print routes using `console.info` (for humans)
    | 'logger'  // always print as NDJSON as part of the app log stream (info level)
    | false     // disable route printing
})

Other default plugins

The following plugins are loaded by default:

Loading other plugins

The server returned by createServer is a Fastify instance, you can register any Fastify-compatible plugin onto it, and use the full Fastify API:

const server = createServer()

server.register(require('fastify-cors'))

server.get('/', () => 'Hello, world !')

Logging

Fastify already has a great logging story with pino, this builds upon it.

Logs Redaction

Logs should be safe: no accidental leaking of access tokens and other secrets through environment variables being logged. For this, redact-env is used.

By default, it will only redact the value of SENTRY_DSN (see Sentry for more details), but you can pass it additional environment variables to redact:

createServer({
  // The values of these environment variables
  // will be redacted in the logs:
  redactEnv: [
    'JWT_SECRET',
    'AWS_S3_TOKEN',
    'DATABASE_URI'
    // etc...
  ]
})

You can also redact log fields by passing Pino redact paths to the redactLogPaths option:

createServer({
  // The values of these headers
  // will be redacted in the logs:
  redactLogPaths: [
    'req.headers["x-myapp-client-secret"]',
    'res.headers["x-myapp-server-secret"]'
    // etc...
  ]
})

The following security headers will be redacted by default:

  • Request headers:
    • Cookie
    • Authorization
    • X-Secret-Token
    • X-CSRF-Token
  • Response headers:
    • Set-Cookie

Environment Context

In case you want to perform log aggregation across your services, it can be useful to know who generated a log entry.

For that, you can pass a name in the options. It will add a from field in the logs with that name:

const server = createServer({
  name: 'api'
})

// The `name` property is now available on your server:
server.log.info({ msg: `Hello, ${server.name}` })
// {"from":"api":"msg":"Hello, api",...}

To add more context to your logs, you can set the following optional environment variables:

Env Var Name Log Key Description
INSTANCE_ID instance An identifier for the machine that runs your application
COMMIT_ID commit The git SHA-1 of your code

Note: for both INSTANCE_ID and COMMIT_ID, only the first 8 characters will be logged or sent to Sentry.

Request ID

By default, Fastify uses an incremental integer for its request ID, which is fast but lacks context and immediate visual identification.

Instead, fastify-micro uses a request ID that looks like this:

To9hgCK4MvOmFRVM.oPoAOhj93kEgbIdV

It is made of two parts, separated by a dot '.':

  • To9hgCK4MvOmFRVM is the user fingerprint
  • oPoAOhj93kEgbIdV is a random identifier

The user fingerprint is a hash of the following elements:

  • The source IP address
  • The user-agent header
  • A salt used for anonymization

The second part of the request ID is a random string of base64 characters that will change for every request, but stay common across the lifetime of the request, making it easier to visualize which requests are linked in the logs:

// Other log fields removed for brievity
{"reqId":"To9hgCK4MvOmFRVM.psM5GNErJq4l6OD6","req":{"method":"GET","url":"/foo"}}
{"reqId":"To9hgCK4MvOmFRVM.psM5GNErJq4l6OD6","res":{"statusCode":200}}
{"reqId":"To9hgCK4MvOmFRVM.oPoAOhj93kEgbIdV","req":{"method":"POST","url":"/bar"}}
{"reqId":"To9hgCK4MvOmFRVM.oPoAOhj93kEgbIdV","res":{"statusCode":201}}
{"reqId":"KyGsnkFDdtKLQUaW.Jj6TgkSAYJ4hcxLR","req":{"method":"GET","url":"/egg"}}
{"reqId":"KyGsnkFDdtKLQUaW.Jj6TgkSAYJ4hcxLR","res":{"statusCode":200}}

Here we can quickly see that:

  • There are two users interacting with the service
  • User To9hgCK4MvOmFRVM made two requests:
    • psM5GNErJq4l6OD6 - GET /foo -> 200
    • oPoAOhj93kEgbIdV - POST /bar -> 201
  • User KyGsnkFDdtKLQUaW made one request:
    • Jj6TgkSAYJ4hcxLR - GET /egg -> 200

Anonymising Request ID Fingerprints

By default, request ID fingerprints are rotated every time an app is built (when createServer is called). This will most likely correspond to when your app starts, and would make it impossible to track users across restarts of your app, or across multiple instances when scaling up. While it's good for privacy (while keeping a good debugging value per-service), it will be a pain for distributed systems.

If you need reproducibility in the fingerprint, you can set the LOG_FINGERPRINT_SALT environment variable to a constant across your services / instances.

Sentry

Built-in support for Sentry is provided, and can be activated by setting the SENTRY_DSN environment variable to the DSN that is found in your project settings.

Sentry will receive any unhandled errors (5xx) thrown by your application. 4xx errors are considered "handled" errors and will not be reported.

You can manually report an error, at the server or request level:

// Anywhere you have access to the server object:
const error = new Error('Manual error report')
server.sentry.report(error)

// In a route:
const exampleRoute = (req, res) => {
  const error = new Error('Error from a route')
  // This will add request context to the error:
  req.sentry.report(error)
}

Enriching error reports

You can enrich your error reports by defining two async callbacks:

  • getUser, to retrieve user information to pass to Sentry
  • getExtra, to add two kinds of extra key:value information:
    • tags: tags are searchable string-based key/value pairs, useful for filtering issues/events.
    • context: extra data to display in issues/events, not searchable.

Example:

import { createServer } from 'fastify-micro'

createServer({
  sentry: {
    getUser: async (server, req) => {
      // Example: fetch user from database
      const user = await server.db.findUser(req.auth.userID)
      return user
    },
    getExtra: async (server, req) => {
      // Req may be undefined here
      return {
        tags: {
          foo: 'bar' // Can search/filter issues by `foo`
        },
        context: {
          egg: 'spam'
        }
      }
    }
  }
})

ProTip: if you're returning Personally Identifiable Information in your enrichment callbacks, don't forget to mention it in your privacy policy 🙂

You can also enrich manually-reported errors:

const exampleRoute = (req, res) => {
  const error = new Error('Error from a route')
  // Add extra data to the error
  req.sentry.report(error, {
    tags: {
      projectID: req.params.projectID
    },
    context: {
      performance: 42
    }
  })
}

Note: v2 to v3 migration

in versions <= 2.x.x, the request object was passed as the second argument to the report function.

To migrate to version 3.x.x, you can remove this argument and use the sentry decoration on the request instead:

const exampleRoute = (req, res) => {
  const error = new Error('Error from a route')

  // version 2.x.x
  server.sentry.report(error, req, {
    // Extra context
  })

  // version 3.x.x
  req.sentry.report(error, {
    // Extra context
  })
}

Sentry Releases

There are two ways to tell Sentry about which Release to use when reporting errors:

  • Via the SENTRY_RELEASE environment variable
  • Via the options:
import { createServer } from 'fastify-micro'

createServer({
  sentry: {
    release: 'foo'
  }
})

A value passed in the options will take precedence over a value passed by the environment variable.

Graceful exit

When receiving SIGINT or SIGTERM, Fastify applications quit instantly, potentially leaking file descriptors or open resources.

To clean up before exiting, add a cleanupOnExit callback in the options:

createServer({
  cleanupOnExit: async app => {
    // Release external resources
    await app.database.close()
  }
})

This uses the Fastify onClose hook, which will be called when receiving a termination signal. If the onClose hooks take too long to resolve, the process will perform a hard-exit after a timeout.

You can specify the list of signals to handle gracefully, along with a few other options:

createServer({
  gracefulShutdown: {
    signals: ['SIGINT', 'SIGTERM', 'SIGQUIT', 'SIGTSTP'],

    // How long to wait for the onClose hooks to resolve
    // before perfoming a hard-exit of the process (default 10s):
    timeoutMs: 20_000,

    // The exit code to use when hard-exiting (default 1)
    hardExitCode: 123
  }
})

Service availability monitoring & health check

under-pressure is used to monitor the health of the service, and expose a health check route at /_health.

Default configuration:

  • Max event loop delay: 1 second
  • Health check interval: 5 seconds

Options for under-pressure can be provided under the underPressure key in the server options:

createServer({
  underPressure: {
    // Custom health check for testing attached services' health:
    healthCheck: async server => {
      try {
        const databaseOk = Boolean(await server.db.checkConnection())
        // Returned data will show up in the endpoint's response:
        return {
          databaseOk
        }
      } catch (error) {
        server.sentry.report(error)
        return false
      }
    },

    // You can also pass anything accepted by under-pressure options:
    maxEventLoopDelay: 3000
  }
})

If for some reason you wish to disable service health monitoring, you can set the FASTIFY_MICRO_DISABLE_SERVICE_HEALTH_MONITORING environment variable to true.

Deprecated APIs

  • configure (will be removed in v4.x): Use plugins with full fastify-autoload options.
  • routesDir (will be removed in v4.x): Use routes with full fastify-autoload options.

License

MIT - Made with ❤️ by François Best

Using this package at work ? Sponsor me to help with support and maintenance.