Refactor API servers into one modular server

[eemeli]

Fixes #57

This PR effects a number of related changes:

• kansa is renamed as server (database schema is still kansa, mind), and becomes the only remaining front-facing express.js server. Relatively few code changes except for the loading of optional modules:

  Object.keys(config.modules).forEach(name => {
    const mc = config.modules[name]
    if (mc) {
      const module = require(`./modules/${name}`)
      app.use(`/${name}`, module(pgp, mc))
    }
  })

• hugo and raami are each refactored as modules, and moved under modules/
• nginx is renamed as proxy
• A modules section is added to config/kansa.yaml

Actual implementation and functionality should not be affected by any of these changes, and there are no changes to the database. A few things that should belong to individual modules don't yet; specifically the tests for Hugo nomination endpoints are still under integration-tests/.

[eemeli]

Updates:

• The Slack inviter is now also a module slack which takes & requires a few config parameters
• npm install is run for enabled modules
• Error objects are now under /common/errors/ and are published as @kansa/errors

[offbyone]

I'm late to the party, but I wanted to add that I actually really prefer the microserver architecture that is in there now to this. It gives the freedom to break out parts of the functionality of the system and replace them with other tools in a modular fashion behind the main API facade. That's a really powerful pattern, and I'd rather keep it.

[offbyone]

I want to be quite specific here, I want the ability to plug in different API frameworks, all the way up to language runtimes, under the main /api facade.

[eemeli]

This change should in fact make it easier to implement services with other language runtimes, as it's starting a process of breaking apart the functionalities provided in particular by what was kansa into smaller modules, which may then in turn be enabled or disabled. No changes are made to e.g. persistence.

As a specific example, stripe is now a module alongside hugo and raami because the overhead involved in splitting into a module (mostly in terms of complexity) is now sufficiently low that it makes sense. To now use an alternative implementation for it, it suffices to disabled the express.js module from the kansa.yaml config, add the alternative to your container orchestration, and configure your reverse proxy for it. Previously doing the same would have been somewhat messier.

[offbyone]

Ah! I misunderstood the level at which the split was occurring. 👍🏻🤟🏻 On Sat, Sep 8, 2018 at 11:15 AM Eemeli Aro notifications@github.com wrote:

This change should in fact make it easier to implement services with other language runtimes, as it's starting a process of breaking apart the functionalities provided in particular by what was kansa into smaller modules, which may then in turn be enabled or disabled. No changes are made to e.g. persistence.

As a specific example, stripe is now a module alongside hugo and raami because the overhead involved in splitting into a module (mostly in terms of complexity) is now sufficiently low that it makes sense. To now use an alternative implementation for it, it suffices to disabled the express.js module from the kansa.yaml config, add the alternative to your container orchestration, and configure your reverse proxy for it. Previously doing the same would have been somewhat messier.

— You are receiving this because your review was requested.

Reply to this email directly, view it on GitHub https://github.com/maailma/kansa/pull/60#issuecomment-419662795, or mute the thread https://github.com/notifications/unsubscribe-auth/AALFvWjVVHhxPfLY4LJpN7AcxQRKx21Xks5uZAkkgaJpZM4Wf4IX .

--

Chris R offby1 on Twitter, offbyone on github Not to be taken literally, internally, or seriously

[eemeli]

More updates:

• The common utilities are now published as @kansa/common rather than separately. That package also includes most notably auth-user, which allows for much cleaner and clearer user authentication that's now mostly handled at the router level.
• The common dependency is installed on the server via npm pack (Build rules added to Makefile and .travis.yml)
• While working through the above, I've refactored quite a few functions to use ES7 async/await for additional clarity.
• Added a bunch of tests. Unfortunately the ones for badges & barcodes need to expect for tarra to fail in CI, as required fonts are missing there.

[eemeli]

Hopefully I can let this rest for a little while after this:

• Moved to common: config, log-entry, mail, trueish
• Added auth.admin_roles to config (not sure about this one, it feels a bit hacky)
• Separated /admin paths into their own module

[eemeli]

So as it turns out I didn't let this rest for a while.

• Server file structure is pretty thoroughly refactored
• The public module is added
• Module init gets a new shared config parameter
• @kansa/common version 1.1.0 is published on npm
• Commit 9a7abfa fixes a parameter length limit in the Postgres function name_match()

I'm going to now merge this into master, and effectively continue development in a new dev branch that builds on top of this (including PRs #61 and #63) and includes changes to the API endpoints.