Why I chose Flask to build sr.ht's mini-services

Published 2019-01-30 on Drew DeVault's blog

sr.ht is a large, production-scale suite of web applications (I call them “mini-services”, as they strike a balance between microservices and monolithic applications) which are built in Python with Flask. David Lord, one of the maintainers of Flask, reached out to me when he heard about sr.ht and saw that it was built with Flask. At his urging, I’d like to share the rationale behind the decision and how it’s turned out in the long run.

I have a long history of writing web applications with Flask, so much so that I think I’ve lost count of them by now - at least 15, if not 20. Flask’s simplicity and flexibility is what keeps bringing me back. Frameworks like Django or Rails are much different: they are the kitchen sink, and then some. I generally don’t need the whole kitchen sink, and if I were given it, I would want to change some details. Flask is nice because it gives you the basics and lets you build what you need on top of it, and you’re never working around a cookie-cutter system which doesn’t cut your cookies in quite the way you need.

In sr.ht’s case in particular, though, I have chosen to extend Flask with a new module common to all sr.ht projects. After all, each service of sr.ht has a lot in common with the rest. Some of the things that live in this core module are:

  • Shared jinja2 templates and stylesheets
  • Shared rigging for SQLAlchemy (ORM)
  • Shared rigging for Alembic
  • A little validation module I’m very proud of
  • API behavior, webhooks, OAuth, etc, which are consistent throughout sr.ht

The mini-service-oriented architecture allows sr.ht services to be deployed ala-carte for users who only need a fraction of what we offer. This design requires a lot of custom code to integrate all of the services with each other - for example, all of the services use a single shared config file, which contains both shared config options and service-specific configuration. sr.ht also uses a novel approach to authentication, in which both user logins and API authentication is delegated to an external service, meta.sr.ht, requiring further custom code still. core.sr.ht additionally provides common SQLAlchemy mixins for things like user tables, which have many common properties, but for each service may have service-specific columns as well.

Django provides their own ORM, their own authentication, their own models, and more. In order to meet the design constraints of sr.ht, I’d have spent twice as long ripping out the rest of Django’s bits and fixing anything that broke in the resulting mess. With Flask, these bits were never written for me in the first place, which gives me the freedom to implement this design greenfield. Flask is small and what code it does bring to the table is highly pluggable.

Though it’s well suited to many of my needs, I don’t think Flask is perfect. A few things I dislike about it:

  • First-class jinja2 support is probably out of scope.
  • flask.Flask and flask.Blueprint should be the same thing.
  • I’m not a fan of Flask’s approach to configuration. I have a better(?) config module that I drag around to all of my projects.

And to summarize the good:

  • It provides a nice no-nonsense interface for requests, responses, and routing.
  • It has a lot of nice hooks for adding your own middleware.
  • It doesn’t do much more than that, which means you’re free to choose and compose other tools to make up the difference.

I think that on the whole it’s quite good. There are frameworks which are smaller still - but I think Flask hits a sweet spot. If you’re making a monolithic web app and can live within the on-rails Django experience, you might want to use it. But if you are making smaller apps or need to rig things up in a unique way - something I find myself doing almost every time - Flask is probably for you.

Have a comment on one of my posts? Start a discussion in my public inbox by sending an email to ~sircmpwn/public-inbox@lists.sr.ht [mailing list etiquette]

Articles from blogs I follow around the net

Status update, May 2019

This month is a little special: it’s been one whole year I’ve started writing status updates! My list of projects has been growing, little by little, and I now spend a pretty big slice of my “open-source contributions time” doing reviews. While this gives …

via emersion June 4, 2019

Updates in May 2019

This post gives an overview of the recent updates to the Writing an OS in Rust blog and to the used tools. I was quite busy with my master thesis this month, so I didn't have the time to create new content or major new features. However, there were qu…

via Writing an OS in Rust June 3, 2019

Giving up on wlroots-rs

Giving up on wlroots-rs

via Way Cooler April 29, 2019

Generated by openring