Honestly this is how all development should be done. Treat the web server as yet another interface into your application to keep it as slim as possible. Need to change web servers, support thrift, avro, etc? Just write another interface that hooks into your application. Easy!

There are plenty of examples in every language where a web framework turns into a web service + caching + session management + ORM + whatever else you can think of then you use all of those features and now you're inflexible, can't easily add new interfaces without seriously rewriting the old and good luck if you need to move away from that web framework for whatever reason.

YES! Your username should be binaryfullofwisdom.

I will take that under advisement :)

"Flask is then a thin wrapper that instantiates the app and then exposes the views to talk to it."

Why did you decide to go this way? Is it because the user-facing parts of the app (the bits where Flask helps) are only a small part?

I mean, if I'm developing a CRUD app with small amounts of supporting non-interactive code, I can't see a reason not to use Django (and put the other stuff in manage.py commands).

On the other hand, if I were to write something like an analytics system, which had a web UI for users to inspect and play with results, I can see that I'd feel constrained by Django when writing the non-web parts.

As someone who loved getting started with Flask, but then switched to Django after getting stuck without a good structure, I'd really like to understand whether the choice of Django vs. Flask should be made based on individual preferences (comfort of an imposed structure vs. freedom without it) or based on the type of project (mainly web, or mainly non-web)?

Django is fine. And honestly, when I'm starting out I just have a single file with Flask, a connection to the db and a few routes. If that's all you need, why complicate your life?

I love SQLAlchemy. Maybe things have changed (I'm a jaded Django 0.96 refugee), but as soon as you take away the Django ORM, you lose the admin and then all your advantages just sort of fall away. But really, that's my personal preference with a lot of bias from being there in the early Django days. I think the Django Rest Framework is absolutely awesome, and it almost pulled me back to Django but it just took my first morning of screwing around with some config to remind me of why I prefer the simple frameworks.

And actually, I could already feel with feature creep there. I wanted to have a remote filestore, so there's a Django plugin for doing it. But a) why is my framework in charge of how I store the files for my application (that means I need Django loaded in ipython!) and b) it didn't work quite how I needed and I could see storm clouds brewing.

In terms of general structure, if you know you're building something bigger, I'd suggest at least splitting the web bit and everything else as soon as possible. For me, leaving the web app to handling routing / validation / marshalling / sessions / templating etc feels like the right abstraction.

I look back on any apps I've built / worked on that a) started off as a CRUD system and b) are still going and most of them are a mess. Even if they're clean, the core business logic is mixed up with the standard framework structure. If you really work at it you can separate things out in the code, but my feeling is that by starting with an app that doesn't know about the web, you'll make it much easier to discover where that abstraction lies.

TLDR; use Django if it works for you. Whatever you choose, try to keep the framework out of your application.

Thanks. Your comments (and whalesalad's) helped make this clearer for me.

A simple (and I think common) situation may illustrate what I took away from your comments:

- You create a CRUD app in Django with two types of users: staff (who use the admin interface to modify anything) and regular (can create/modify their own data, but only read other people's data)

- The staff make a mess via the admin interface, so you decide to add some type of undo thing. You look around for an existing addon that does model versioning, but people still spend time reverting changes back.

- You decide you need some business logic (changes going through some review/approval process) and are scratching your head about how to do that in Django.

I can see then why "I look back on any apps I've built / worked on that a) started off as a CRUD system and b) are still going and most of them are a mess."

Hey, original author here.

In regards to your last question, here are my subjective ways of thinking about this. Firstly, I don't think there are many pre-defined or general rules with software development. I've been lucky to work on a pretty diverse set of projects in my career, and I can tell you success or failure almost always boils down to fundamentals not tools - how well do you understand the space you're working in (for web development - how well do you understand databases, caching, concurrency, scaling strategies) and how well do you understand the abstraction tools provided by your language? If your decision making is grounded in these fundamentals, then the tools you choose are not going to make that big an impact. From this point, tool choice is really about economizing the work you'll have to do and managing risk. If you're gonna be building a bunch of CRUD stuff and non-technical users need to view and modify data - then Django and its admin system makes a ton of sense. If, on the other hand, you're developing JSON APIs that you'll be deploying into high-volume conditions that you'll have to react quickly to, you don't want to pin your ability to scale and move quickly on fundamental decisions made in a large underlying library with many unknowns, in this scenario Flask makes a lot of sense because it's not overly prescriptive and plays nicely with almost everything. For this particular project none of us were very familiar with Django, so there were far too many unknowns for us to be comfortable staking a lot on it.

It's tremendously freeing to have your application live as what amounts to vanilla python (or ruby, or php, or clojure, or insert-your-favorite-language-here). Once you've abstracted your core business logic from the HTTP layer, a lot of problems become easier to solve. If you want to replace Flask entirely it becomes very simple. If you want to replace REST entirely with a CQRS-style system, it's much easier. Testing gets easier. Wanna play with your app without firing up a web server? Build it so that you can use it from an ipython shell.

Vanilla is the holy grail. You mention manage.py commands but why not if __name__ == '__main__': do the thing. Or the multitude of tools like invoke (https://github.com/pyinvoke/invoke) and click (http://click.pocoo.org/5/). I digress...

Django is great for a monolith. Stack Django REST Framework on top of it and you're even bigger and nastier. I would never use it again because the amount of time spent bending it to my will often exceeds the time I'd spend building my own system. Fair warning, I have been using Django in production since version 0.96 and have built many companies on it. After ten years of building web apps, using every MVC under the sun, and working with every size/shape team under the sun you really start to appreciate simple things that compose well together. They're easier to reason about. They're easier to replace. They're easier to pull out and turn into a standalone library or shared service. Are you hitting the wall with one tiny part of your app? Replace it with something written in an entirely different language and communicate via JSON, or over a message bus. Not everything can be mapped to a WSGI request/response.

I am so glad aidos has the top comment on this post because his/her style is what we should all be aiming for from the get-go.

I'd definitely suggest reading up on the SOLID –https://en.wikipedia.org/wiki/SOLID_(object-oriented_design) – principles. Furthermore, the adapter concept is often explained with scary Java code samples, but the core of it is wonderful. Essentially you should consider your API as a vanilla system in your language of choice and you build adapters which then allow for external services to communicate. One adapter could be HTTP (via Flask perhaps) for a front-end website. One adapter could be HTTP but one that only speaks JSON, which you might serve under api.yourdomain.com. Another adapter might be via a queue system which allows for arbitrary commands from an external source to get written to a durable queue, so that they can be replayed later.

I took some melatonin so at this point I feel like I am just babbling and about to fall asleep.

tl;dr build things without frameworks, use your language well, be vanilla, compose tiny things to build big things.

I think you are biased to pick a "micro framework" as a starting point. Because of your experience, you have probably seen all the common patterns in building a web app. As a result, you probably have a solution, or know exactly where to look for a solution to any requirement you come across. However, less experienced team would need a lot more guidance and proven best practices to follow. An analogy is a driver in a foreign country. A GPS map might not provide her with the shortest or most enjoyable route, it gets the job done. Django is great for inexperienced team. It's also great for solo consultant.

I feel it's also a lot more about temperament than it is about experience. Some developers prefer to build from scratch and some prefer to maximise reuse.

I fall mostly into the second camp. For non-trivial functionality I will (in order of preference) 1. Find a good existing framework or library 2. Refactor something I've already written to be reusable 3. Develop something specifically for the task at hand.

Even when I plump for 3 I'll be thinking about how it could be made reusable - I just postpone actually putting any effort in that direction on the grounds of YAGNI.

Great points! I've worked with Flask for a while, as I've gotten to be a better developer. I used to have a lot of my logic in the application itself. Now, I ensure my logic lives in its own modules. Flask is really just either a way to serve my templates or API.

This type of development style (http://martinfowler.com/eaaCatalog/serviceLayer.html) is quite well-known in the Java world yet unfortunately people always think that the technique is overkill (in reality, it's really really not overkill, it's only overkill if you force it to be).

Thanks to javax Most things could be switched out easily

I like this approach. How do you handle routes? Do you just initialize a wrapper app that is composed of your own application, wrap your views in Flask views objects and then call `add_url_rule` on them in the wrapper setup code?

If I understand this correctly, how do you define something like a redirect or abort within your own views?

I've tried to explain my structure a bit better in a reply to voltagex_.

For your specific question - all routing is done from within the top / Flask layer (as is redirecting / aborting). My App layer does not have any Flask stuff in, at all, ever.

This is probably a pretty stupid question, but can you point me in the direction of some recommended reading or documentation regarding how i might go about using a web framework as a wrapper for a separate app? I'm just getting started in python web dev, coming from PHP-MVC-framework-land and this idea is fascinating to me but i'm having a hard time conceptualizing how you would do it.

edit: to be clear, i've read Fowler and understand the idea of a multi-layered application i'm just having a hard time understanding how to achieve it technically.

Not a stupid question, especially since I just made it up as I went along - so there's not much to go on. Will try to put together a little mini example on github (but that will require me actually doing a bit of dissecting).

I've also looked through the Fowler stuff (as well as almost everything else on the subject!). One thing I struggled with a lot was how to handle the DB layer. Fowler has that as the layer right in the middle of the system but I really didn't like that model. All the web frameworks are set up so the db layer is very entangled with the framework and it's quite hard to escape.

Let me put something up on github. Send me an email (in my profile) and then I can let you know once I have something to look at.

Maybe I'm just not thinking clearly today but how do you turn Flask into a wrapper? I thought you had to put enough logic in to a class that inherits from Flask to return views and such.

I think I need to explain a bit better. I have a few different layers. From the outside in:

Flask/Web

Deal with sessions, authentication, caching, templating, logging etc here. Fires up an instance of the App and keeps it around for the duration of the life of the Web instance. I actually have an API in here that looks like what you'd expect in most other web apps. It does the validation / error handling etc. In terms of manipluating the data, I'm not a total purist - the API is allowed to load up the SQLA models and make changes. If there's a more complex operation going on though, it will live inside the next layer down.

App

Deal with remote services / data models / db connection. I guess on the outside you might call this a "service layer". A surprisingly large amount of this is data persistance management. The model I use is to put the data access stuff just off to the side at the top level of the app. Nothing in here knows anything about things like web frameworks. This is important. In other places in the system, this same App is used by RQ (worker instances taking on work via a redis queue). There's another web app that uses it for administrative tasks. I can boot it up to work with it in ipython (I do this a lot). Oh, and I don't manage the db life-cycle here. Whichever layer sits above this determines when it's ok to commit transactions. If something goes wrong here we raise app specific expections - the layer above can handle those and relay them as appropriate.

Core

In the very core I have "pure" algorithmic stuff (for context - http://www.countfire.com is effectively an image recognition app). Really this is just one component that the App happens to be using. If the App wants to talk to this, it needs to create the data structures that the core defines. At various borders I often have mappings from the objects in the current local world to versions for the created to talk to the layer below (ps I like namedtuples).

It's a lot more complex than all this, each of these layers has more sub layers etc. And really, what I've called the Core is a sublayer of the App (in my case there's a well definied algorithmic chunk that I think of as the core so it's in its own package - it stands alone as its own library).

There's no real way of knowing where you'll draw lines between the layers when you start but I have a few rules. Any bit of code should be given the bare minimum of data required to do its work. At borders map local data to data for the next layer. Make sure the stuff going into the current layer is really there because it needs to be there. Be a pragmatist - things will change, you'll pull stuff out into submodules. By following the other rules hopefully refactoring won't be too painful.

I'm curious, is this an open-source project? I would love to have a look into how you built this. If not, well, no worries.

No, sorry - it's very much commercial code (it's our startup). Though it's 90% unicorns and dragons :-)

Edit http://www.countfire.com