Every day, I receive a stock of documents (an update). What I want to do is insert each item that does not already exist.
I am using the Python driver (pymongo).
What I currently do is (pseudo-code):
for each document in update:
existing_document = collection.find_one(document)
if not existing_document:
document['insertion_date'] = now
else:
document = existing_document
document['last_update_date'] = now
my_collection.save(document)
My problem is that it is very slow (40 mins for less than 100 000 records, and I have millions of them in the update). I am pretty sure there is something builtin for doing this, but the document for update() is mmmhhh.... a bit terse.... (http://www.mongodb.org/display/DOCS/Updating )
Can someone advise how to do it faster?
This question is related to
python
mongodb
bulkinsert
mongodb-query
Drawing from Van Nguyen's answer above, use update instead of save. This gives you access to the upsert option.
NOTE: This method overrides the entire document when found (From the docs)
var conditions = { name: 'borne' } , update = { $inc: { visits: 1 }} , options = { multi: true };
Model.update(conditions, update, options, callback);
function callback (err, numAffected) { // numAffected is the number of updated documents })
If you want to update a selection of the document, but not the whole thing, you can use the $set method with update. (again, From the docs)... So, if you want to set...
var query = { name: 'borne' }; Model.update(query, ***{ name: 'jason borne' }***, options, callback)
Send it as...
Model.update(query, ***{ $set: { name: 'jason borne' }}***, options, callback)
This helps prevent accidentally overwriting all of your document(s) with { name: 'jason borne' }
.
In general, using update is better in MongoDB as it will just create the document if it doesn't exist yet, though I'm not sure how to work that with your python adapter.
Second, if you only need to know whether or not that document exists, count() which returns only a number will be a better option than find_one which supposedly transfer the whole document from your MongoDB causing unnecessary traffic.
Summary
Note, I'm presuming PyMongo, change to suit your language of choice.
Instructions:
Create the collection with an index with unique=true so you don't get duplicate records.
Iterate over your input records, creating batches of them of 15,000 records or so. For each record in the batch, create a dict consisting of the data you want to insert, presuming each one is going to be a new record. Add the 'created' and 'updated' timestamps to these. Issue this as a batch insert command with the 'ContinueOnError' flag=true, so the insert of everything else happens even if there's a duplicate key in there (which it sounds like there will be). THIS WILL HAPPEN VERY FAST. Bulk inserts rock, I've gotten 15k/second performance levels. Further notes on ContinueOnError, see http://docs.mongodb.org/manual/core/write-operations/
Record inserts happen VERY fast, so you'll be done with those inserts in no time. Now, it's time to update the relevant records. Do this with a batch retrieval, much faster than one at a time.
Iterate over all your input records again, creating batches of 15K or so. Extract out the keys (best if there's one key, but can't be helped if there isn't). Retrieve this bunch of records from Mongo with a db.collectionNameBlah.find({ field : { $in : [ 1, 2,3 ...}) query. For each of these records, determine if there's an update, and if so, issue the update, including updating the 'updated' timestamp.
Unfortunately, we should note, MongoDB 2.4 and below do NOT include a bulk update operation. They're working on that.
Key Optimization Points:
I don't think mongodb supports this type of selective upserting. I have the same problem as LeMiz, and using update(criteria, newObj, upsert, multi) doesn't work right when dealing with both a 'created' and 'updated' timestamp. Given the following upsert statement:
update( { "name": "abc" },
{ $set: { "created": "2010-07-14 11:11:11",
"updated": "2010-07-14 11:11:11" }},
true, true )
Scenario #1 - document with 'name' of 'abc' does not exist: New document is created with 'name' = 'abc', 'created' = 2010-07-14 11:11:11, and 'updated' = 2010-07-14 11:11:11.
Scenario #2 - document with 'name' of 'abc' already exists with the following: 'name' = 'abc', 'created' = 2010-07-12 09:09:09, and 'updated' = 2010-07-13 10:10:10. After the upsert, the document would now be the same as the result in scenario #1. There's no way to specify in an upsert which fields be set if inserting, and which fields be left alone if updating.
My solution was to create a unique index on the critera fields, perform an insert, and immediately afterward perform an update just on the 'updated' field.
You may use Upsert with $setOnInsert operator.
db.Table.update({noExist: true}, {"$setOnInsert": {xxxYourDocumentxxx}}, {upsert: true})
You could always make a unique index, which causes MongoDB to reject a conflicting save. Consider the following done using the mongodb shell:
> db.getCollection("test").insert ({a:1, b:2, c:3})
> db.getCollection("test").find()
{ "_id" : ObjectId("50c8e35adde18a44f284e7ac"), "a" : 1, "b" : 2, "c" : 3 }
> db.getCollection("test").ensureIndex ({"a" : 1}, {unique: true})
> db.getCollection("test").insert({a:2, b:12, c:13}) # This works
> db.getCollection("test").insert({a:1, b:12, c:13}) # This fails
E11000 duplicate key error index: foo.test.$a_1 dup key: { : 1.0 }
As of MongoDB 2.4, you can use $setOnInsert (http://docs.mongodb.org/manual/reference/operator/setOnInsert/)
Set 'insertion_date' using $setOnInsert and 'last_update_date' using $set in your upsert command.
To turn your pseudocode into a working example:
now = datetime.utcnow()
for document in update:
collection.update_one(
{"_id": document["_id"]},
{
"$setOnInsert": {"insertion_date": now},
"$set": {"last_update_date": now},
},
upsert=True,
)
Source: Stackoverflow.com