5

When defininng the to be returned fields (collection.find(q, fields)), does (how does) mongoDB support to compare the field names (and not the value)?

E.g. to select a range of fields based on their name without regard to their value.

Assume the following fields (whatever the value might be) in a document:

fieldA000 fieldA001 fieldA002 fieldB000 fieldB001 fieldC000 ...

and I want to restrict the returned fields for any field name matching fieldA000 to fieldA999 (without need to define explicitely any field name) or fieldA.* (reg ex) or similar.

P.S.: I am currently evaluating whether we could use mongoDB instead of Cassandra where the column-range/slice-select is provided in a very easy to use way.

gustavodidomenico
  • 4,640
  • 1
  • 34
  • 50
Christopher Frank
  • 111
  • 1
  • 3

2 Answers2

3

Mongo doesn't have a way to match documents using regex's on key names.

However, you could do this with $where which lets you use execute JavaScript to select each document. The disadvantage of $where is that it can't take advantage of indexes, so it requires deserializing every document in the collection which can be too slow for most applications.

The other way to do this in Mongo is with $exists and $or but it requires explicit key names.

A schema like this would be more efficient for this type of query, and also facilitates slicing on the server using $:

{
    dataFields: [
        { id: 'A000', value: 'whatevs' },
        { id: 'A001', value: 'whatevs' },
        { id: 'A002', value: 'whatevs' },
        { id: 'B000', value: 'whatevs' },
        { id: 'B001', value: 'whatevs' },
        { id: 'C000', value: 'whatevs' },
    ]
}

or:

{
    dataFields: [
        { lettter: 'A', number: 0, value: 'whatevs' },
        { lettter: 'A', number: 1, value: 'whatevs' },
        { lettter: 'A', number: 2, value: 'whatevs' },
        { lettter: 'B', number: 0, value: 'whatevs' },
        { lettter: 'B', number: 1, value: 'whatevs' },
        { lettter: 'C', number: 0, value: 'whatevs' },
    ]
}
paulmelnikow
  • 16,895
  • 8
  • 63
  • 114
  • 1
    By the way: The loopup speed could be greatly enhanced by creating indices on the fields letter and number of dataFields array. – Philipp Apr 17 '13 at 14:03
  • Thanks a lot. I was kind of hoping I don't need to generate a different structure (than we use on Cassandra). – Christopher Frank Apr 17 '13 at 14:12
0

MongoDB does allow you to provide a projection on every query operation, which specifies which fields are to be returned.

You cannot include or exclude fields based on a pattern. What you can do is mark fields to be included:

db.foo.find({}, {'A000': 1, 'A001': 1, 'B000': 1})

(This query will return the _id, A000, A001 and B000 fields).

Or, mark specific fields to be excluded:

db.foo.find({}, {'B000': 0})

(This query will return every fields except B000).

See this link for more information.

In general, using projection like this to simulate column-slicing isn't a great use case for a document database — while the data will not be returned from the db server to the client (saving network and client parsing overhead), it will still be read from disk (unless a covering index can be used to return all of the fields), and the document has to be parsed to determine which portions to exclude.

Sean Reilly
  • 21,526
  • 4
  • 48
  • 62
  • How does this relate to the question? – Sammaye Apr 17 '13 at 15:01
  • @Sammaye: "and I want to restrict the returned fields for any field name matching fieldA000 to fieldA999" — I read this as a question about how to restrict which fields are returned with the result, rather than about how to filter which documents are returned. Do you disagree? – Sean Reilly Apr 17 '13 at 16:16
  • Indeed it does project a certain amount out however he is unsure about how many fields, hence he wishes to use a regex to project all the fields that match a range, his problem might be simplified too. In other words this won't work. – Sammaye Apr 17 '13 at 16:23
  • Also projection can use an index – Sammaye Apr 17 '13 at 16:27
  • Projection can (and will) use indexes. But IO won't be saved unless the index is completely covers the projection — which is unlikely with a large number of fields being returned. I've added that to the answer. – Sean Reilly Apr 17 '13 at 16:45