Question

I have a large dataset with documents that sometimes cross-reference each other, sometimes do not. Before I can mapreduce based on those cross references, I have to set the array of cross-references to be that same for every value in the cross reference.

I use this in the shell function to consolidate those arrays:

function fixArray2() {
var counter = 0;
// I only want the xref for each field, I don't even want the id
var cursor = db.catalog.find({}, {xref: true, _id: false});

// I don't want to init this inside the loop, worried about memory leaks
var consolidatedArray = [];
while (cursor.hasNext()) {
    var xref1 = cursor.next().xref;
    // first pass: create a consolidated array when the cross references match
    var limitedCursor1 = db.catalog.find({"name":{$in:xref1}});
    while (limitedCursor1.hasNext()) {
        var doc1 = limitedCursor1.next();
        consolidatedArray = consolidatedArray.concat(doc1.xref);
    }
    consolidatedArray = consolidatedArray.unique();
    // now that we have the consolidated array, reset the xref field of the object to it
    for (var i=0; i<consolidatedArray.length; i++) {
        db.catalog.update({name:consolidatedArray[i]},{$set:{xref: consolidatedArray}},false, true);
    }

    consolidatedArray.length = 0;

    counter++;
    if (counter % 1000 == 0) {
        print("Processed " + counter + " documents.");
    }
}

}

It works, but I have to run it fairly often. Can anyone suggest improvements?

Answer 1

If you do the work up front when writing the documents to the collection you may be able to avoid doing this map-reduce where you do the work at a later time.

Therefore, get the list of documents that should be cross referenced and write them with the document upon insertion. Update as needed, when a document is removed or no longer references the other for example.