At my work, we use Oracle for our database. Which works great. I am not the main db admin, but I do work with it. One thing I like is that the DB has a built in logic layer using PL/SQL which ca handle logic related to saving the data and retrieve it. I really like this because it allows our MVC application (PHP/Zend Framework) to be lighter, and makes it easier to tie in another platform into the data, such as desktop or mobile.
Although, I have a personal project where I want to use couchdb or mongodb, and I want to try and accomplish a similar goal. outside of the mvc/framework, I want to have an API layer that the main applications talk to. they dont actually talk directly to the database. They specify the design document (couchdb) or something similar for mongo, to get the results. And that API layer will validate the incoming data and make sure that data itself is saved and updated properly. Such as saving a new user, in the framework I only need to send a json obejct with the keys/values that need to be saved and the api layer saves the data in the proper places where needed.
This API would probably have a UI, but only for administrative purposes and to make my life easier. In general it will always reply with json strings, or pre-rendered/cached html in some cases. Since each api layer would be specific to the application anyways.
I was wondering if anyone has done anything like this, or had any tips on nethods I could accomplish this. I am currently looking to write my application in python, and the front end will likely be something like Angularjs. Although I am also looking at node.js for a back end.
We do this exact thing at my current job. We have MongoDB on the back end, a RESTful API on top of it and then PHP/Zend on the front end.
Most of our data is read only, so we import that data into MongoDB and then the RESTful API (in Java) just serves it up.
Some things to think about with this approach:
Write generic sorting/paging logic in your API. You'll need this for lists of data. The user can pass in things like http://yourapi.com/entity/1?pageSize=10&page=3.
Make sure to create appropriate indexes in Mongo to match what people will query on. Imagine you are storing users. Make an index in Mongo on the user id field, or just use the _id field that is already indexed in all your calls.
Make sure to include all relevant data in a given document. Mongo doesn't do joins like you're used to in Oracle. Just keep in mind modeling data is very different with a document database.
You seem to want to write a layer (the middle tier API) that is database agnostic. That's a good goal. Just be careful not to let Mongo specific terminology creep into your exposed API. Mongo has specific operators/concepts that you'll need to mask with more generic terms. For example, they have a $set operator. Don't expose that directly.
Finally after having a decent amount of experience with CouchDB and Mongo, I'd definitely go with Mongo.