Tag Archives: nosql

Design Notes on Database Application Development

I recently took a brief look at web2py since its Database Abstraction Layer (DAL) sounded interesting. After following the tutorials in the Overview chapter, I ended up not delving into DAL.  Instead, I revisited Ruby on Rails which I had last looked circa 2005 when its support for PostgreSQL was rather weak. Although that has improved, the basic approach of Rails hasn’t and is very similar to web2py’s as well as Django.

The essence of these full stack frameworks is that the developer —or should I say programmer— knows how to design classes and that that design can be transformed into a relational database design mechanically and flawlessly. Maybe I’m biased because my first involvements in software engineering were in the areas of IT system design, but the latter premise seems like a very weak foundation for developing robust systems.

Maybe this also explains, in part, why NoSQL solutions have cropped up and found a following. It’s much easier —albeit not reliable or sound— to design a system when you “liberate” yourself from the “straitjacket” constraints of a database design (aka schema). However, in my experience, most systems beyond the blog/wiki/toy database examples used in tutorials benefit from specialized data analysis and modeling that result in a formal database design.

If the database precedes the application development, it appears the Rails/Django/web2py approach is back asswards. If the database is available during development, doesn’t it make more sense to re-use the database design effort in creating the application rather than try to repeat it with Python or Ruby classes?

The Pyrseas project, in its quest to rescue Andromeda, plans to implement a utility to help in database application development (tentatively named dbappgen but that may change). The utility will start off by connecting to a database and will allow the developer to create the application based on the tables and other objects present in it, not the other way around. If database changes are needed, they can be made there first, not by imposing some application-induced “migration” (as web2py does).

Since Pyrseas’ dbtoyaml utility already outputs a YAML/JSON description of a database, it may even be possible to conduct an application development exercise, at least in part, without connecting to the database. For example, if the developer invokes dbappgen moviesdb, the utility could first check whether a file named moviesdb.yaml (or with similar extensions) exists in the current directory and if so, read it and recreate in memory the PostgreSQL catalogs, thus allowing “offline” development.

In other news, due to the recent discussions regarding Python 3, my hopes that Werkzeug will be ported soon have been dashed. It appears it’s either use WebOb or write my own framework (which ironically is what Armin recommended).