Question

Suppose I have web application errors reported to mongo db. To fetch them (grouped by URLs) I use the following query:

db.errors.group({
    cond: {
        date:{
            $gt:new Date(2013,0,3, 6, 0, 0), 
            $lt:new Date(2013,0,3, 8, 0, 0)
        }
    },
    key: {
        url: 1
    },
    initial: {
        csum: 0
    },
    reduce:
        function(obj, prev) {
            prev.csum++;
        }
})

What should I do if I would like to count the errors by a site url (as opposed to a page url)? I mean, that http://www.mysite.com/page1 and http://www.mysite.com/page2 will be counted in the same bucket. Ideally, it would be a function in the key section of the query somehow…

Answer 1

To extend JohnnyHK’s answer:

db.errors.group({
    cond: {
        date:{
            $gt:new Date(2013,0,3, 6, 0, 0),
            $lt:new Date(2013,0,3, 8, 0, 0)
        }
    },
    keyf: function(doc)
    {
        return { site: doc.url.match(/.*:\/\/[^\/]+/i)[0]};
    },
    initial: {
        csum: 0
    },
    reduce:
        function(obj, prev) {
            prev.csum++;
        }
})

Returns:

[
    {
        "site" : "http://www.mysite.com",
        "csum" : 3
    },
    {
        "site" : "http://www.example.com",
        "csum" : 1
    }
]

With a large dataset it’s not going to be fast though. If you control the code that generates the errors it might be easier to add a domain field you could group on.