Most efficient way to change a string field value to its substring - mongodb

Question

I have a collection filled with documents that look like this:
{
data: 11,
version: "0.0.32"
}
and some have a test suffix to version:
{
data: 55,
version: "0.0.42-test"
}
The version field has different values but it always conforms to the pattern: 0.0.XXX. I would like to update all the documents to look like this:
{
data: 11,
version: 32
}
and the suffixed version (for test documents - version should be negative):
{
data: 55,
version: -42
}
The collection with these documents is used by our critical system, that needs to be turned off while updating the data - so I want the update/change to be as fast as possible. There are about 66_000_000 documents in this collection, and it's about 100GB in size.
Which type of mongodb operation would be the most efficient one?

Answer 1

The most efficient way to do this is in the upcoming release of MongoDB as of this writing using the $split operator to split our string as shown here then assign the last element in the array to a variable using the $let variable operator and the $arrayElemAt operators.
Next, we use the $switch operator to perform a logical condition processing or case statement against that variable.
The condition here is $gt which returns true if the value contains "test", and in which case in the in expression we split that string and simply return the $concatenated value of the first element in the newly computed array and the -. If the condition evaluates to false, we just return the variable.
Of course in our case statement, we use the $indexOfCP which returns -1 if there were no occurrences of "test".
let cursor = db.collection.aggregate(
[
{ "$project": {
"data": 1,
"version": {
"$let": {
"vars": {
"v": {
"$arrayElemAt": [
{ "$split": [ "$version", "." ] },
-1
]
}
},
"in": {
"$switch": {
"branches": [
{
"case": {
"$gt": [
{ "$indexOfCP": [ "$$v", "test" ] },
-1
]
},
"then": {
"$concat": [
"-",
"",
{ "$arrayElemAt": [
{ "$split": [ "$$v", "-" ] },
0
]}
]
}
}
],
"default": "$$v"
}
}
}
}
}}
]
)
The aggregation query produces something like this:
{ "_id" : ObjectId("57a98773cbbd42a2156260d8"), "data" : 11, "version" : "32" }
{ "_id" : ObjectId("57a98773cbbd42a2156260d9"), "data" : 55, "version" : "-42" }
As you can see, the "version" field data are string. If the data type for that field does not matter then, you can simply use the $out aggregation pipeline stage operator to write the result into a new collection or replace your collection.
{ "out": "collection" }
If you need to convert your data to floating point number then, the only way to do this, simply because MongoDB doesn't not provides a way to do type conversion out of the box except for integer to string, is to iterate the aggregation Cursor object and convert your value using parseFloat or Number then update your documents using the $set operator and the bulkWrite() method for maximum efficiency.
let requests = [];
cursor.forEach(doc => {
requests.push({
"updateOne": {
"filter": { "_id": doc._id },
"update": {
"$set": {
"data": doc.data,
"version": parseFloat(doc.version)
},
"$unset": { "person": " " }
}
}
});
if ( requests.length === 1000 ) {
// Execute per 1000 ops and re-init
db.collection.bulkWrite(requests);
requests = [];
}}
);
// Clean up queues
if(requests.length > 0) {
db.coll.bulkWrite(requests);
}
While the aggregation query will perfectly work in MongoDB 3.4 or newer our best bet from MongoDB 3.2 backwards is mapReduce with the bulkWrite() method.
var results = db.collection.mapReduce(
function() {
var v = this.version.split(".")[2];
emit(this._id, v.indexOf("-") > -1 ? "-"+v.replace(/\D+/g, '') : v)
},
function(key, value) {},
{ "out": { "inline": 1 } }
)["results"];
results looks like this:
[
{
"_id" : ObjectId("57a98773cbbd42a2156260d8"),
"value" : "32"
},
{
"_id" : ObjectId("57a98773cbbd42a2156260d9"),
"value" : "-42"
}
]
From here you use the previous .forEach loop to update your documents.
From MongoDB 2.6 to 3.0 you will need to use the now deprecated Bulk() API and it associated method as show in my answer here.