npm-serve

File and npm package server.

TL;DR;

Server-side

Issue the following command to use it:

$ npx @padcom/npm-serve@latest .

or if you're a Docker fan:

$ docker run --rm -it \
  --name=npm-serve \
  -p 2998:2998 \
  -v $(pwd):/var/lib/npm-serve/static-files \
  padcom/npm-serve

Client-side

<script type="libraries">
{
  "libraries": [
    "[email protected]/dist/vue.runtime.esm-browser.js",
    {
      "library": "@scope/[email protected]/dist/index.js",
      "stylesheets": [ 'dist/styles.css' ]
    }
  ],
  "config": {
    "cdn": "/package/",
    "overrides": true,
    "polyfills": true
  }
}
</script>
<script src="https://unpkg.com/@padcom/npm-serve@2/docs/importmap.js"></script>

Use the following to import a library and its styles:

const lib = await import('@scope/package')
loadStylesheetsFromLibrary('@scope/package')

Use the following to unload styles loaded for a library:

unloadStylesheetsFromLibrary('@scope/package')

A bit of history

These days bundlers and sophisticated development tools for node.js and frontend are very much part the daily routine. Webpack, Rollup/Vite, Parcel - those are just a few examples of tools that a frontend developer can't live without.

But does it have to be like that?

Back in March 2021 a major feature landed in Chrome: importmap's. Importmaps are a way to let the browser know where to look for packages when those are imported in the browser using package name, e.g.

import { createApp } from 'vue'

Here's an example import map that will allow for the above syntax:

  <script type="importmap">
    {
      "imports": {
        "vue": "https://unpkg.com/[email protected]/dist/vue.runtime.esm-browser.prod.js"
      }
    }
  </script>

The problem that arises is as follows: npm hosting facilities have a rather hard time providing the latest information about packages. Unpkg seems to be freezing every 30s while it is fetching the latest information from https://registry.npmjs.org which is very much an undesired behavior. On top of that it would be handy if we could use the same application to serve both application files and packages from the same application.

This is why @padcom/npm-serve has been created. Upon start it will serve all local files on port 2998 unless that file starts with /package/ in which case the rest of the path will be treated as package coordinates, including scope, name, version/tag and path to the file. So if in your project you have an index.html that has reference to, let's say, vue.js then adding the following will allow for the application to serve that package:

  <script type="importmap">
    {
      "imports": {
        "vue": "/package/[email protected]/dist/vue.runtime.esm-browser.prod.js"
      }
    }
  </script>

Package coordinates

Let's look at an example, in this case @padcom/mf-test-library1:

  <script type="importmap">
    {
      "imports": {
        "library1": "/package/@padcom/[email protected]/dist/index.js"
      }
    }
  </script>

In this example:

• @padcom - package scope; required if package is scoped - otherwise leave empty; packages don't have to be scoped but if they are here's where you specify it
• mf-test-library1 - required; the name of the package
• 0.0.2 - optional; version or tag of the package; if not specified latest is used
• /dist/index.js - optional; file to use; defaults to package.json/main field

So to always get the default export from latest @padcom/mf-test-library1 use the following coordinates:

  "imports": {
    "library1": "/package/@padcom/mf-test-library1"
  }

or if you'd like to use the beta version:

  "imports": {
    "library2": "/package/@padcom/mf-test-library2@beta"
  }

or if you'd like to import another file from latest version:

  <link rel="stylesheet" href="/package/@padcom/mf-test-library3/dist/style.css">

Using client-side generation of importmap

At times it'd be great to be able to, for example, switch from one CDN to another. Let's say JSPM is out for some reason. All you'd expect to have to do would be to provide a URL to another location just once and that'd be applied to all imports.

This is where the importmap generator comes into play. It's a simple JSON object generator, nothing fancy at first glance. However, the devil's in the details.

Loading client-side script

The library needs a starting point to do its magic. Add that to your file after the libraries script:

<script src="https://unpkg.com/@padcom/npm-serve@2/docs/importmap.js"></script>

Specifying libraries

At this point it is worth mentioning that I'll be referring to anything deployed on NPM as a library. So if you'd like to use, for example Vue.js (which is a library) you'd have to give a bit more details to the system:

<script type="libraries">
{
  "libraries": [
    "[email protected]/dist/vue.runtime.esm-browser.js"
  ]
}
</script>

This will by default generate the following importmap:

<script type="importmap">
{
  "imports": {
    "vue": "https://unpkg.com/[email protected]/dist/vue.runtime.esm-browser.js"
  }
}
</script>

Again, nothing fancy, but there's a little bit of magic already happening. The system knows that all libraries are available at https://unpkg.com, so that's where it will get them from.

What if we'd like to change the default CDN to something else?

<script type="libraries">
{
  "libraries": [
    "[email protected]/dist/vue.runtime.esm-browser.js"
  ],
  "config": {
    "cdn": "/package/"
  }
}
</script>

That will produce the following importmap:

<script type="importmap">
{
  "imports": {
    "vue": "/package/[email protected]/dist/vue.runtime.esm-browser.js"
  }
}
</script>

This is where @padcom/npm-serve comes in - just serving static files + npm packages.

Stylesheets

To load styles for a given library you need to do two things. First, you need to tell the system, that the library has stylesheet(s):

<script type="libraries">
{
  "libraries": [
    {
      "library": "@padcom/mf-test-library3",
      "stylesheets": [ "dist/style.css" ]
    }
  ],
  "config": {
    "cdn": "/package/"
  }
}
</script>

Then, use the loadStylesheetsFromLibrary(name) function and give it the full name of the library (so, with scope if it is scoped):

loadStylesheetsFromLibrary('@padcom/mf-test-library5')

There's also the unloadStylesheetsFromLibrary(name) version to remove any stylesheets loaded for that library.

Overrides

Suppose you're working on a site that's composed of, let's say, 200 possible libraries. Trying to spin that kind of an environment locally will surely end with everything just freezing. But there's a solution to the problem. First things first - we already have an environment. That's, for example, https://unpkg.com. We can host the website from there - no problem. But for development purposes we'd need a sort of surgical knife to tell the system that for this particular library you'd like it to be hosted locally, let's say http://localhost:3005/dist/index.js. You can do that by specifying an override in the configuration:

https://unpkg.com/@scope/package?@padcom/mf-test-library5=http://localhost:3005/dist/index.js

That's simple enough, right? You just tell the system that for a given package the CDN is somewhere else. Another option is to just specify a version, like 1.0.4 or tag like latest or beta. In this case it won't be the CDN that's changed but the version of the library. On top of that there's nothing standing in your way to develop that particular part using full development tools but it needs to be a separate application, a sort of dev playground.

That override behavior can be explicitly turned off (maybe someone wouldn't like to have that capability in production for whatever reason):

<script type="libraries">
{
  "libraries": [
    ...
  ],
  "config": {
    "overrides": false
  }
}
</script>

Quick note about deploying non-latest versions

When you want to deploy a version for a different tag (as npm calls them) you need to let npm know about it:

$ npm publish --tag beta

Polyfilling missing features

Currently, at the time of writing, most of the browsers based on webkit do support ES modules and importmaps. However, not all of them do. For that reason there is a polyfill that can be automatically loaded. That polyfill contains feature detection code so it will only be used in the event that the code runs in an environment that doesn't support importmaps natively.

To enable polyfills add "polyfills": true to the "config": { ... } section like so:

  <script type="libraries">
    {
      "libraries": [ ... ],
      "config": {
        ...
        "polyfills": true
      }
    }
  </script>

Alternatively, if you'd like to customize the default polyfill URL or add more polyfills the `"polyfills" key can take either a string (location of the polyfill) or an array of strings (array of URLs containing polyfills):

  <script type="libraries">
    {
      "libraries": [ ... ],
      "config": {
        ...
        "polyfills": [
          'https://unpkg.com/es-module-shims/dist/es-module-shims.js'
        ]
      }
    }
  </script>

Versioning

It'd be not very interesting if we couldn't serve multiple versions of the same package. Luckily we can omit the version all together, in which case @padcom/npm-server will take whatever the latest version is. If, however, the specified version is one of

  'latest', 'next', 'beta', 'alpha'

e.g. /package/@padcom/mf-test-library2@beta, the application will use that tag instead of latest allowing you to, for example, make a beta release of your module and switching between production and beta versions easily.

CLI Options

When starting @padcom/npm-serve there are several options that you can use to tell the application where to store cached packages or where to serve files from.

-s location

-s location tells the application where to store cached packages

-p port

-p port tells the application to listen on a given port (default: 2998)

Document root

If you want to serve files from another place than the current folder just append it as last command line parameter, e.g.:

$ npx @padcom/npm-serve ~/my-project

By default @padcom/npm-serve will use the current folder to look for files to serve.

Caching strategy

Caching is an interesting topic in this context. Normally you'd want all the files to be as fresh as possible without doing any roundtrips to the server. That can't be accomplished 100% but we can get pretty close to it. For example, by providing Cache-Control', 'max-age=30 we'll tell the browser to only make requests every 30 minutes and in between to use whatever content it (the browser) has currently cached. After that time the browser will make a request but it'd be pointless to transfer that same file again. This is where the etag comes into play. If the request contains If-None-Match header with a value that value will be compared to a calculated value (based currently on the timestamp of the cached package archive) which will make the browser use the cached file again.

However, in the meantime, when someone deploys a new version, we'd like to get that update. To achieve this the @padcom/npm-serve employs the last good deploy approach serving whatever has been already cached but if the last update was more than 30s ago it will schedule an asynchronous update of the package. That way the user eventually gets whatever was up-to-date but each time a request is made it is processed with maximum speed.

Strategies for dependency management

There are at least 3 main strategies when building applications that fetch modules from remote locations:

Package everything

When building an application that uses some external packages, if you're using a bundler such as webpack or rollup all the dependencies that the application is using will be inlined in the resulting module definition. Those inlined packages will be a private copy thereof for use only with that particular module.

Don't package common libraries

In this scenario main packages, such as Vue.js, React or Angular runtime will be provided by the host. The imported modules will be smaller, but will depend on the browser to know where to find them. Everything else can be inlined, making those other packages private to that module.

In the event when you export sort of a bootstrap function and delegate the initialization to the module this is a great option allowing you to host an application written in different frameworks. Each module is then treated as a microfrontend.

Don't use a bundler

In this scenario every single imported module is listed in the importmap. No bundling required.

Do's and Don'ts

Initially the idea of microfrontends, created by Michael Geers, assumed that the rigth abstraction to implement microfrontends will be custom elements. However, besides few examples (some extremely prominent, such as Google Chrome settings page) not many projects use it in in favor of module composition. This package, in essence, implements the technical capability to use modules that are deployed to NPM registry - just without the use of npm install.

• If you don't have a very specific case don't use custom components as wrappers for microfrontends
• If you're all set on one framework (and that better be Vue.js :D) then export components from microfrontends. Vue has some exciting features for async components!
• If you really must be specific use a label (latest, beta) instead of a specific version
• If you want to limit the number of possible versions (e.g. all lower than 3.0 but higer than 2.0) omit the part of version that you want the service to fill in automatically
• If speed is what you have the need for then use full version with full path to the exported files; it'll limit the number of 302 responses

Workflows

So, you have a pretty good understanding of how the @padcom/npm-serve package works and you can use it for a simple project, but how would you put together a massive project with multiple teams and tens of microfrontends?

First of all, since multiple teams and, most probably, multiple parts of the organization will be involved in the process it is important to allow for separation of accidental commonalities. Accidental commonalities (as opposed to essential commonalities) are things that are common for two or more teams but none of that common code is owned by any of those teams. For example, let's say we have the Dragons team that takes care of the shopping cart module and then another team, say the Pengiuns, that focuses 100% on ads. In a monorepo or monolith setup we'd have a single repository organized into subfolders (either top-level with monorepo or maybe as subfolder of src in a monolith) which means that every single time anybody changes anything in any of the modules then everybody will get it. That's how git works or rather that's how version control works. You may not notice any changes in your files but something new has arrived into your .git folder. That is completely unnecessary and in point of fact massively violates the Dependency Inversion Principle from the SOLID bunch.

With that we have the following projects, each living in their own, separate repository:

• main
• ads
• shopping-cart
• product
• product-list
• navigation
• search
• promotions

Let's say that each project will be deployed under the @example scope, which gives us the following packages and their default exports in their 0th version:

• @example/[email protected]/index.html
• @example/[email protected]/dist/index.js exports { start }
• @example/[email protected]/dist/index.js exports { start }
• @example/[email protected]/dist/index.js exports { start }
• @example/[email protected]/dist/index.js exports { start }
• @example/[email protected]/dist/index.js exports { start }
• @example/[email protected]/dist/index.js exports { start }
• @example/[email protected]/dist/index.js exports { start }

The @example/main, because it's package.json/main field will point to index.html that's the file that will be served by CDNs such as @padcom/npm-serve or unpkg.com. This gives us automatically the following versioning capabilities:

• declaring accessing host in any version, even partial or tag-based names would work
• declaring submodules to take the most up-to-date version up to certain limits (following semver)

All that needs to happen is that each of the modules is published to npm registry. But wait! Those are "my" things, the company won't allow others to use "my" code! Am I doomed?

Not at all, it's actually a pretty standard requirement to be able to publish npm packages to company-hosted (or SaaS-based) registry. What you can do here is to setup another proxy npm registry that will collect packages from multiple sources (npm and your private registry) and provides them all under one, common registry.

That, in fact, is what you should do anyways, because https://registry.npmjs.org can get really slow at times not to mention it can go offline at which point nothing works anymore. Having a caching proxy for npm packages will shield you and your company from those kinds of outages and speed up package installation in general.

Back to the story of Dragons and Penguins.

Today was a busy day for Penguins. The project manager brought to everyone's attention that notorious bug that the CEO has been talking about during Yesterday's all-hands meeting. The stakes are high, atmosphere is thick, everyone's on the edge and wants to get shit done ASAP.

John is our designated driver for the adventure. He opens up a new terminal, navigates to the ads repository and starts his development environment. He can work either on a locally-served stripped down version of the lib or, assuming that the files are available somehow via http, hook them up directly to the dev server by navigating to https://example-super.app?@example/ads=http://localhost:1234 where 1234 is the port where your files are served. And best of all, the rest of the application is served as-is in production! It just works!

The bug was actually easy to find and fix, it's just one of those things that never come up during testing or development and the users are furious that it doesn't work. John fixes it on a branch and pushes it to the remote repository.

Upon creating the PR a build system kicks in and runs all the tests and builds the package to verify John didn't break anything.

That PR gets immediatelly (yeah, right?) reviewed, last-minute changes are applied and reviewed again, all builds are green and you're ready to merge. But before you do that you deploy your code to the registry with the @example/[email protected]:

$ npm publish --tag beta

Then John navigates to https://example-super.app?@example/ads=0.0.1-beta to see his new code pulled in from npm and served. Ups! That's am... I thought that wasn't doing that... Are you sure this is ok?... So, next change, PR, review, automated tests and builds, and a few minutes later John publishes @example/[email protected]. Darn - John forgot to update the version number, but no worries - npm is ready for those silly mistakes and won't let you override anything by mistake (nor if that was your intention!). He then navigates to https://example-super.app?@example/ads=0.0.1-beta and after a while, when all the caches have been refreshed, he can finally see the hard work he and his collegues did that day.

His best buddy, Lucifer the Tester, will do that same thing to check if your fix does what you say it does. All is good, the branch is merged by Lucifer as a sort of rubber stamp that it's been verified and he signs it with blood. That mistuque act spins off another automated build that will increment the main patch version, build and publish the final product to the registry as @example/[email protected]. A hot tee later everyone in the world gets the fix, and... then there is cacke :D

But, at the same time as the whole shabang with dudes at Penguins was going on, Marry has been working with the other Dragons on implementing a cool new feature for the shopping cart. They have been deploying 3 times Today already under @example/[email protected]. The feature is still somewhat incomplete but it already shows where it will be situated in the shopping cart landscape. There are multiple endpoints for the shopping cart, after all users do love a quick overview of how much money they'll gladly transfer to our accounts. Jerry, the project manager, was just told that an update to the alpha version has been deployed that adds that green icon to the left of that thing. Jerry then navigates to https://example-super.app?@example/shopping-cart=0.0.1-aplha. He receives the application with @example/[email protected] (yes, Marry and the other Dragons have been busy lately!) and enjoys the progress that's been made. Yet another boring day at the office...

At any time everyone can run both versions at the same time to see the added green icon there with that annoying bug in ads fixed.

A few months before Marry, John and Lucifer started working on the application, Sasha was setting up the deployment of the system. He knew to setup a proxy npm registry that will shield everyone from npmjs.org outages and has added the relevant information to How to configure your setup.txt document that every new person joining the team has to get familiar with. Now everone enjoys fast npm installs and at the same time the source for Cloudfront won't have to fail while retrieving updates. He does that by using the padcom/npm-serve docker image, exposing port 2998 and mounting /var/lib/npm-serve/static-files to where the content of @example/host@latest/dist is stored and forwards that traffic via a local instance of *nginx to the outside world (IDK, it seems like the right thing to do to introduce some kind of request caching between the clients and a node.js application).

And that's it! That's how not only developers, but also managers and infrastructure can benefit from using the microfrontends.

Because I think that happy developers make for happy customers which turns into money in everyone's pocket!

Examples

Microfrontends with Vue and React Multiple external libraries, common library, host.

To use this example cd to the host folder and issue the following command:

$ npx @padcom/npm-serve .