NoSQL optimization strategy

Question

Looking for some input on the following. Hopefully this isn't too subjective for the moderators.

Just started playing with deployd.com BaaS API and have the following scenario.

If an mobile app was to have a Group object and a User, where a User could belong to many Groups and Groups have many Users (many-to-many relationship) I could design this several ways with two that I'm considering:

A)

Users [{
  id: 1,
  groups : {1,2,3,4}
}]

Groups [{
  id: 1,
  users : {1,2,3,4}
}]

And

B)

Users [{
   id: 1
}]

Groups [{
   id: 1
}]

UserGroups [{
   id: 1,
   group: 1,
   user: 1,
 },{
   id: 2
   group: 1,
   user: 2,
}]

I'm leaning towards B as I can store meta-data (date a user joined the group, etc) but this seems to be more a RDBMS method and I'm wondering if I will lose any of the benefits of NoSQL by trying to create such relations.

Hypothetically speaking this mobile app would be used by thousands of mobile users simultaneously and hence choosing NoSQL versus RDBMS in the first place.

@Philipp All groups of a user is certainly going to get a lot more queries. — Andrew Grothe, Dec 08 '13 at 19:02

Philipp · Accepted Answer · 2013-12-08T18:58:34.050

1

B) is not a good idea. MongoDB doesn't do joins, so any operation which needs multiple collections means multiple subsequent queries, which is much slower than a database-internal JOIN in a relational database. That means you should store relations in the documents themself.

When you want to store meta-information, keep in mind that arrays can't just store primitive values. They can also store objects. Example:

{
    id:1,
    name:"Bob",
    groups: [
        { name: "nice people",
          position: "member",
          joined: ISODate(2013, 12, 3)
        } ,
        { name: "evil people",
          position: "admin",
          joined: ISODate(2012, 11, 22)
        }

    ]
}

It might be a good idea to store enough meta-information so that many common queries can be fulfilled without querying the referenced objects. When you want to show a list of groups when someone looks up a users profile, and you store all the group information you need in that overview in the users-document, you avoid having to do a second query to retrieve the group-documents.

edited Dec 08 '13 at 18:58

answered Dec 08 '13 at 18:53

Philipp

67,764
9
118
153

So larger documents and less queries will benefit performance. – Andrew Grothe Dec 08 '13 at 18:59
@agrothe yes. The most expensive operations are those which require multiple collections, because MongoDB has absoulutely no support for operations which span multiple collections. – Philipp Dec 08 '13 at 19:01
But updating the name of a group wouldn't be a disaster? – Meena Alfons Jun 19 '17 at 22:26
@MeenaAlfons A long-running multi-update query? Sure. But a disaster? Not necessarily. Keep in mind that updates like that are a rare use-case and can be performed in the background. Few people will notice if a group rename needs a few minutes before it's visible to everyone. – Philipp Jun 20 '17 at 08:09

NoSQL optimization strategy

1 Answers1