Question

I have a really simple Mongodb collection e.g.

{ "_id" : ObjectId("objid"), "url" : "http://mydomain.com", "datestamp" : ISODate("2012-03-17T02:00:45.119Z"), "totalcount" : 1 }

I’m trying to query the collection to return items in descending order as follows:

@globallinks = Mongo::Connection.new.db("mydb").collection("mycollect")
@toplinks = @globallinks.find({}, :sort => ["totalcount", Mongo::DESCENDING], :limit => 100)

This query does not return items in desc order. I know for a fact by scanning through the result set that records look to be unsorted.

Does anyone have any ideas?

Answer 1

What you’re doing works fine for me and it should work.

However, if you only have a limited number of totalcount values then you will get a result set that looks random. Consider a collection where most totalcount values are 2:

> show = ->(o) { puts "totalcount = #{o['totalcount'].to_i}, other = #{o['other'].to_i}" }
> t.find().each(&show)
totalcount = 1, other = 1
totalcount = 1, other = 4
totalcount = 2, other = 3
totalcount = 2, other = 2
totalcount = 2, other = 1
totalcount = 2, other = 4

Now we sort it by totalcount (and only by totalcount):

> t.find({}, :sort => ['totalcount', Mongo::DESCENDING]).each(&show)
totalcount = 2, other = 3
totalcount = 2, other = 2
totalcount = 2, other = 1
totalcount = 2, other = 4
totalcount = 1, other = 1
totalcount = 1, other = 4

Now we apply a :limit that just happens to leave us with only totalcount == 2 values:

> t.find({}, :sort => ['totalcount', Mongo::DESCENDING], :limit => 4).each(&show)
totalcount = 2, other = 3
totalcount = 2, other = 2
totalcount = 2, other = 1
totalcount = 2, other = 4

Do you notice how the other ordering looks random? But, if we add a secondary sort key, we’ll get something that looks sorted:

> t.find({}, :sort => [['totalcount', Mongo::DESCENDING], ['other', Mongo::ASCENDING]], :limit => 4).show(&each)
totalcount = 2, other = 1
totalcount = 2, other = 2
totalcount = 2, other = 3
totalcount = 2, other = 4

totalcount sounds like something that wouldn’t have that many values and would at the very least contain many duplicate values. The :limit is applied after the :sort so it can easily select results with only one or two values for totalcount. Without a secondary sort key, you can get results that look random even though they are (partially) sorted.

Any time you use a sort key that can have duplicated values you usually want to include a secondary sort key to sensibly sort things within the primary groups.