Planning complex Mongo (And possibly other NoSQL) Databases

Question

There is no "right way" to do anything, but there are "better" and "worse" ways, as well as true and tested ways.

I'm studying Mongo and considering it for a project. It's gonna start small but possibly grow a lot.

Now, I understand there are different ways to organize information in MongoDB. Since there are no joins, I either insert ids (Or arrays) and do multiple queries to get a related item (I.e. person for a company, company Id on people object or array of people on company table), or I insert objects directly inside... (I.e., people as objects in an array inside the company, as opposed to being "decoupled").

Now, my issue is: what if I need to change how I load this info?

A good example, taking the users idea, is if I wanna get all people (Of all companies) inactive for the last month. Now, would I need to look separately into one company at a time to query this information? Should I keep them separate and use ids as references in this case...?

What are the tradeoffs, the performance impacts, and which approach is considered "more appropriate" by the community?

Answer 1

As noted by John F. Miller on another question at SO (https://stackoverflow.com/a/5373969/1793074), MongoDB is great at nesting data, but not so grat to search this nested data.

Another great resource on this matter comes from the same answer: http://seanhess.github.io/2012/02/01/mongodb_relational.html

Basically, you relinquish control when you work with "documents inside documents". A lot of controle.

With that said, the suggested approach is to only insert a document inside another if the "inside-document" is absolutelly dependant on the outside one. Then, and only then, you can "skip collections".

You can, though, insert objects that reference other collections, using denormalization techniques to have performance gains.

What to choose is up to each project, but for data that has to be queried and used in different situations, the best is to have separate collections, and reference by id or dennormalizing for faster access.