« Scott Nonnenberg


Private Node.js modules: A journey

2016 Jul 12

One of the best benefits of Node.js is the ease of extracting code into its own new project. But you probably won’t want to make that code fully public.

It took me quite a while to get to a solution I was happy with for my private modules. Let’s start at the beginning.

Clumsy beginnings: git submodules

I built my first Node.js app in 2012, using a single git repository. Having come from Rails, this was very comfortable. As we got closer to that first demo with customers, the project gained new dependencies, new application-specific code, but also a good amount of code I could already tell would be useful across apps. So, as I worked on my next few Node.js apps later that year, I extracted this utility code into a separate library called thehelp.

I didn’t want to make my utility project available publicly, so the official npm registry was out. I was probably a bit scarred from my experience with Ruby gems, because I also shied away from anything like a ‘real’ dependency.

After mulling it over, I realized that I had the tools right in front of me to share code between projects: git! I used git submodules to share the code across my projects. I made the experience as nice as possible with tags and a custom shell script to check for and upgrade to new versions.

But all dependencies needed by the submodule had to be included in the parent project’s package.json. It was an ugly, error-prone, manual process.

At least I can laugh about it now.

Evolution: git + npm

As thehelp grew, it became clear that I needed to split it up. But I didn’t want to use the same mechanism as before - the manual package.json updates were already painful with just one package.

Again I considered the tools I had in front of me. I was a lot more comfortable with npm now, and this time I discovered that npm and git could play together nicely. With a git+ssh dependency in my package.json I could install dependencies directly from a git repository. Simple and easy! Especially since I was already good at using tags for each release.

The great thehelp split happened in late 2013, and it was an easy transition. It was so much nicer having smaller, more focused projects. And there were no sad manual steps! I streamlined things by writing some code to pull git+ssh paths out of my package.json and tell me if I was on the latest version. Here’s the core tag-processing algorithm, from my never-released git-latest-version project:

Latest.prototype.parseTags = function(tags) {
  var tagRegex = /^v?\d+\.\d+\.\d+$/;

  return _(tags)
    .filter(function(tag) {
      return tagRegex.test(tag.name);
    })
    .map(function(tag) {
      return tag.name;
    })
    .sort(function(left, right) {
      return semver.compare(left, right);
    })
    .reverse()
    .value();
};

Scaling issues

This worked very well for me until the spring of 2014. The first issue I ran into was intermittent ‘connection reset by peer’ errors when I ran an npm install. It wasn’t an urgent problem because it would usually work the second time. After a while I finally figured it out: for an empty project with a lot of new dependencies to install, npm opened a lot of concurrent connections. And most servers reject that many parallel SSH sessions.

The second issue was more than an intermittent annoyance. Like a good Node.js citizen, I upgraded to npm@1.4.x. But I discovered that a ‘correctness change’ now caused all of my git+ssh dependencies to be reinstalled during every npm install. Because of thehelp-project this meant a re-download of phantomjs. It meant more connections to remote git servers, and more frequent ‘connection reset by peer’ errors.

npm install was now a long, error-prone process, even if I just added one new dependency or changed one tag.

So, I reverted to an old version of npm and started looking for alternatives. I had lived on the fringe for too long with git submodules and git+ssh.

Biting the bullet: sinopia

With this update to my techniques, I needed to move beyond the tools already in hand. And so, I researched open-source tools to give me a local npm registry. I could run a full copy of the npm registry, but that was too heavyweight. I had a gut sense for the level of effort I was ready to put into this. sinopia fit the bill.

It was great! sinopia is a lightweight little Node.js server, proxying my connection to the npm registry, caching node module packages and metadata to disk. It made all of my private packages look just like modules straight from the npm registry.

It had a nice unexpected benefit, too! With it installed on my laptop, I could now run an npm install without an internet connection! All previously-installed node modules would always ready to be installed again. For almost two years, I had sinopia configured to start when I logged into my laptop, always proxying in the background. Its storage directory, full of .tgz files, got up to 543 megabytes.

Sadly, the project has lost steam and hasn’t seen updates in quite some time. Whenever I thought about it, I realized I was getting more and more nervous about it. When would it break completely?

Now: npm private modules

I got my answer when I tried to use an npm private module with sinopia. I could download a public scoped package with a config change, but sinopia doesn’t understand how to send your credentials to npm. Once you start using npm private modules, you can no longer use sinopia.

That’s mostly okay. After all, you have a new place for your private modules now!

However, you will lose those offline installs without another solution. And I’ve got one for you! I did some searching and found npm_lazy, a simple little caching npm proxy. I quickly discovered that it didn’t support npm private modules either, so I submitted a pull request to add it. With it now merged you can forge ahead without fear!

I’m feeling good. I like npm private modules, even though I periodically run into lack of support for them. I’ve released a number of scoped modules now by just flipping the switch from private to public. Easy!

I’m thinking this setup will last at least another year or two. :0)

The landscape

For a long time, if you wanted private node modules, there were either costly private ‘enterprise’ solutions, or lightweight open-source solutions. Slowly, as competition increased, services started to claim the middle ground, and then npm itself released private module support in April 2015.

npm plans allow unlimited packages at $7 per developer, per month. No server management hassles, no strange configurations. Just normal npm and your normal npm user, the same high-availability infrastructure we know and love.

But perhaps you don’t want to store your private code offsite. npm enterprise was first released in late 2014 and gives you a seamless, easily-installed, on-premises registry experience. If need support for other types of private modules, Sonatype’s Nexus (among others) might be a better fit.

You’ll have to decide if your organization is willing to store its private code offsite. If you already use Github, you’ve already made your decision. :0)

Lots of small projects

Learn from my mistakes! Get used to frequently creating lots of little projects, and take advantage of the systems that make it easy. Maybe install via git+ssh during high-churn development, then pay the $7/mo. for npm private modules.

There are great emergent benefits to small, well-factored projects: better testing of what’s in each project, momentum towards Single Responsibility Principle (SRP), practice in creating good API surface area and documentation, greater comfort with npm and node modules, and so on.

Reduce the friction of doing this with a private registry of some kind. You’ll thank me.

I won't share your email with anyone. See previous emails.

NEXT:

Better changelogs, strings, and paths 2016 Jul 19

I’m always on the lookout for ways to do Node.js and Javascript development better, but I haven’t found a good vehicle for these kinds of discoveries yet. I briefly mentioned a few in a recent post... Read more »

PREVIOUS:

This blog is now open source! 2016 Jul 05

I just open-sourced this blog, based on the Gatsby static site generator. Very meta! You could be reading this on GitHub! I thought I’d walk through some of the more interesting details, and... Read more »


Hi, I'm Scott. I've written both server and client code in many languages for many employers and clients. I've also got a bit of an unusual perspective, since I've spent time in roles outside the pure 'software developer.' You can find me on Mastodon.