How can documents be moved from one collection to another collection in MongoDB?? For example: I have lot of documents in collection A and I want to move all 1 month older documents to collection B (these 1 month older documents should not be in collection A).
Using aggregation we can do copy. But what I am trying to do is moving of documents.
What method can be used to move documents?
The bulk operations #markus-w-mahlberg showed (and #mark-mullin refined) are efficient but unsafe as written. If the bulkInsert fails, the bulkRemove will still continue. To make sure you don't lose any records when moving, use this instead:
function insertBatch(collection, documents) {
var bulkInsert = collection.initializeUnorderedBulkOp();
var insertedIds = [];
var id;
documents.forEach(function(doc) {
id = doc._id;
// Insert without raising an error for duplicates
bulkInsert.find({_id: id}).upsert().replaceOne(doc);
insertedIds.push(id);
});
bulkInsert.execute();
return insertedIds;
}
function deleteBatch(collection, documents) {
var bulkRemove = collection.initializeUnorderedBulkOp();
documents.forEach(function(doc) {
bulkRemove.find({_id: doc._id}).removeOne();
});
bulkRemove.execute();
}
function moveDocuments(sourceCollection, targetCollection, filter, batchSize) {
print("Moving " + sourceCollection.find(filter).count() + " documents from " + sourceCollection + " to " + targetCollection);
var count;
while ((count = sourceCollection.find(filter).count()) > 0) {
print(count + " documents remaining");
sourceDocs = sourceCollection.find(filter).limit(batchSize);
idsOfCopiedDocs = insertBatch(targetCollection, sourceDocs);
targetDocs = targetCollection.find({_id: {$in: idsOfCopiedDocs}});
deleteBatch(sourceCollection, targetDocs);
}
print("Done!")
}
Update 2
Please do NOT upvote this answer any more. As written #jasongarber's answer is better in any aspect.
Update
This answer by #jasongarber is a safer approach and should be used instead of mine.
Provided I got you right and you want to move all documents older than 1 month, and you use mongoDB 2.6, there is no reason not to use bulk operations, which are the most efficient way of doing multiple operations I am aware of:
> var bulkInsert = db.target.initializeUnorderedBulkOp()
> var bulkRemove = db.source.initializeUnorderedBulkOp()
> var date = new Date()
> date.setMonth(date.getMonth() -1)
> db.source.find({"yourDateField":{$lt: date}}).forEach(
function(doc){
bulkInsert.insert(doc);
bulkRemove.find({_id:doc._id}).removeOne();
}
)
> bulkInsert.execute()
> bulkRemove.execute()
This should be pretty fast and it has the advantage that in case something goes wrong during the bulk insert, the original data still exists.
Edit
In order to prevent too much memory to be utilized, you can execute the bulk operation on every x docs processed:
> var bulkInsert = db.target.initializeUnorderedBulkOp()
> var bulkRemove = db.source.initializeUnorderedBulkOp()
> var x = 10000
> var counter = 0
> var date = new Date()
> date.setMonth(date.getMonth() -1)
> db.source.find({"yourDateField":{$lt: date}}).forEach(
function(doc){
bulkInsert.insert(doc);
bulkRemove.find({_id:doc._id}).removeOne();
counter ++
if( counter % x == 0){
bulkInsert.execute()
bulkRemove.execute()
bulkInsert = db.target.initializeUnorderedBulkOp()
bulkRemove = db.source.initializeUnorderedBulkOp()
}
}
)
> bulkInsert.execute()
> bulkRemove.execute()
Insert and remove:
var documentsToMove = db.collectionA.find({});
documentsToMove.forEach(function(doc) {
db.collectionB.insert(doc);
db.collectionA.remove(doc);
});
note: this method might be quite slow for large collections or collections holding large documents.
$out is use to create the new collection with data , so use $out
db.oldCollection.aggregate([{$out : "newCollection"}])
then use drop
db.oldCollection.drop()
you can use range query to get data from sourceCollection and keep the cursor data in variable and loop on it and insert to target collection:
var doc = db.sourceCollection.find({
"Timestamp":{
$gte:ISODate("2014-09-01T00:00:00Z"),
$lt:ISODate("2014-10-01T00:00:00Z")
}
});
doc.forEach(function(doc){
db.targetCollection.insert(doc);
})
Hope so it helps!!
First option (Using mongo dump)
1.Get a dump from collection
mongodump -d db -c source_collection
2.Restore from collection
mongorestore -d db -c target_collection dir=dump/db_name/source_collection.bson
Second Option
Running aggregate
db.getCollection('source_collection').aggregate([ { $match: {"emailAddress" : "apitester#mailinator.com"} }, { $out: "target_collection" } ])
Third Option (Slowest)
Running a through for loop
db.getCollection('source_collection').find().forEach(function(docs){ db.getCollection('target_collection').insert(docs); }) print("Rolleback Completed!");
May be from the performance point of view it's better to remove a lot of documents using one command(especially if you have indexes for query part) rather than deleting them one-by-one.
For example:
db.source.find({$gte: start, $lt: end}).forEach(function(doc){
db.target.insert(doc);
});
db.source.remove({$gte: start, $lt: end});
This is a restatement of #Markus W Mahlberg
Returning the favor - as a function
function moveDocuments(sourceCollection,targetCollection,filter) {
var bulkInsert = targetCollection.initializeUnorderedBulkOp();
var bulkRemove = sourceCollection.initializeUnorderedBulkOp();
sourceCollection.find(filter)
.forEach(function(doc) {
bulkInsert.insert(doc);
bulkRemove.find({_id:doc._id}).removeOne();
}
)
bulkInsert.execute();
bulkRemove.execute();
}
An example use
var x = {dsid:{$exists: true}};
moveDocuments(db.pictures,db.artifacts,x)
to move all documents that have top level element dsid from the pictures to the artifacts collection
Here's an update to #jasongarber's answer which uses the more recent mongo 'bulkWrite' operation (Read docs here), and also keeps the whole process asynchronous so you can run it as part of a wider script which depends on its' completion.
async function moveDocuments (sourceCollection, targetCollection, filter) {
const sourceDocs = await sourceCollection.find(filter)
console.log(`Moving ${await sourceDocs.count()} documents from ${sourceCollection.collectionName} to ${targetCollection.collectionName}`)
const idsOfCopiedDocs = await insertDocuments(targetCollection, sourceDocs)
const targetDocs = await targetCollection.find({_id: {$in: idsOfCopiedDocs}})
await deleteDocuments(sourceCollection, targetDocs)
console.log('Done!')
}
async function insertDocuments (collection, documents) {
const insertedIds = []
const bulkWrites = []
await documents.forEach(doc => {
const {_id} = doc
insertedIds.push(_id)
bulkWrites.push({
replaceOne: {
filter: {_id},
replacement: doc,
upsert: true,
},
})
})
if (bulkWrites.length) await collection.bulkWrite(bulkWrites, {ordered: false})
return insertedIds
}
async function deleteDocuments (collection, documents) {
const bulkWrites = []
await documents.forEach(({_id}) => {
bulkWrites.push({
deleteOne: {
filter: {_id},
},
})
})
if (bulkWrites.length) await collection.bulkWrite(bulkWrites, {ordered: false})
}
From MongoDB 3.0 up, you can use the copyTo command with the following syntax:
db.source_collection.copyTo("target_collection")
Then you can use the drop command to remove the old collection:
db.source_collection.drop()
I do like the response from #markus-w-mahlberg, however at times, I have seen the need to keep it a bit simpler for people. As such I have a couple of functions that are below. You could naturally wrap thing here with bulk operators as he did, but this code works with new and old Mongo systems equally.
function parseNS(ns){
//Expects we are forcing people to not violate the rules and not doing "foodb.foocollection.month.day.year" if they do they need to use an array.
if (ns instanceof Array){
database = ns[0];
collection = ns[1];
}
else{
tNS = ns.split(".");
if (tNS.length > 2){
print('ERROR: NS had more than 1 period in it, please pass as an [ "dbname","coll.name.with.dots"] !');
return false;
}
database = tNS[0];
collection = tNS[1];
}
return {database: database,collection: collection};
}
function insertFromCollection( sourceNS, destNS, query, batchSize, pauseMS){
//Parse and check namespaces
srcNS = parseNS(sourceNS);
destNS = parseNS(destNS);
if ( srcNS == false || destNS == false){return false;}
batchBucket = new Array();
totalToProcess = db.getDB(srcNS.database).getCollection(srcNS.collection).find(query,{_id:1}).count();
currentCount = 0;
print("Processed "+currentCount+"/"+totalToProcess+"...");
db.getDB(srcNS.database).getCollection(srcNS.collection).find(query).addOption(DBQuery.Option.noTimeout).forEach(function(doc){
batchBucket.push(doc);
if ( batchBucket.length > batchSize){
db.getDB(destNS.database).getCollection(destNS.collection)insert(batchBucket);
currentCount += batchBucket.length;
batchBucket = [];
sleep (pauseMS);
print("Processed "+currentCount+"/"+totalToProcess+"...");
}
}
print("Completed");
}
/** Example Usage:
insertFromCollection("foo.bar","foo2.bar",{"type":"archive"},1000,20);
You could obviously add a db.getSiblingDB(srcNS.database).getCollection(srcNS.collection).remove(query,true)
If you wanted to also remove the records after they are copied to the new location. The code can easily be built like that to make it restartable.
I had 2297 collection for 15 million of documents but some collection was empty.
Using only copyTo the script failed, but with this script optimization:
db.getCollectionNames().forEach(function(collname) {
var c = db.getCollection(collname).count();
if(c!==0){
db.getCollection(collname).copyTo('master-collection');
print('Copied collection ' + collname);
}
});
all works fine for me.
NB: copyTo is deprecated because it block the read/write operation: so I think is fine if you know that the database is not usable during this operation.
In my case for each didn't work. So I had to make some changes.
var kittySchema = new mongoose.Schema({
name: String
});
var Kitten = mongoose.model('Kitten', kittySchema);
var catSchema = new mongoose.Schema({
name: String
});
var Cat = mongoose.model('Cat', catSchema);
This is Model for both the collection
`function Recursion(){
Kitten.findOne().lean().exec(function(error, results){
if(!error){
var objectResponse = results;
var RequiredId = objectResponse._id;
delete objectResponse._id;
var swap = new Cat(objectResponse);
swap.save(function (err) {
if (err) {
return err;
}
else {
console.log("SUCCESSFULL");
Kitten.deleteOne({ _id: RequiredId }, function(err) {
if (!err) {
console.log('notification!');
}
else {
return err;
}
});
Recursion();
}
});
}
if (err) {
console.log("No object found");
// return err;
}
})
}`
I planned to arhieve 1000 records at a time using bulkinsert and bulkdelete methods of pymongo.
For both source and target
create mongodb objects to connect to the database.
instantiate the bulk objects. Note: I created a backup of bulk objects too. This will help me to rollback the insertion or removal when an error occurs.
example:
For source
// replace this with mongodb object creation logic
source_db_obj = db_help.create_db_obj(source_db, source_col)
source_bulk = source_db_obj.initialize_ordered_bulk_op()
source_bulk_bak = source_db_obj.initialize_ordered_bulk_op()
For target
// replace this with mogodb object creation logic
target_db_obj = db_help.create_db_obj(target_db, target_col)
target_bulk = target_db_obj.initialize_ordered_bulk_op()
target_bulk_bak = target_db_obj.initialize_ordered_bulk_op()
Obtain the source records that matches the filter criteria
source_find_results = source_db_obj.find(filter)
Loop through the source records
create target and source bulk operations
Append archived_at field with the current datetime to the target collection
//replace this with the logic to obtain the UTCtime.
doc['archived_at'] = db_help.getUTCTime()
target_bulk.insert(document)
source_bulk.remove(document)
for rollback in case of any errors or exceptions, create target_bulk_bak and source_bulk_bak operations.
target_bulk_bak.find({'_id':doc['_id']}).remove_one()
source_bulk_bak.insert(doc)
//remove the extra column
doc.pop('archieved_at', None)
When the record count to 1000, execute the target - bulk insertion and source - bulk removal. Note: this method takes target_bulk and source_bulk objects for execution.
execute_bulk_insert_remove(source_bulk, target_bulk)
When exception occurs, execute the target_bulk_bak removal and source_bulk_bak inesertions. This would rollback the changes. Since mongodb doesn't have rollback, I came up with this hack
execute_bulk_insert_remove(source_bulk_bak, target_bulk_bak)
Finally re-initialize the source and target bulk and bulk_bak objects. This is necessary because you can use them only once.
Complete code
def execute_bulk_insert_remove(source_bulk, target_bulk):
try:
target_bulk.execute()
source_bulk.execute()
except BulkWriteError as bwe:
raise Exception(
"could not archive document, reason: {}".format(bwe.details))
def archive_bulk_immediate(filter, source_db, source_col, target_db, target_col):
"""
filter: filter criteria for backup
source_db: source database name
source_col: source collection name
target_db: target database name
target_col: target collection name
"""
count = 0
bulk_count = 1000
source_db_obj = db_help.create_db_obj(source_db, source_col)
source_bulk = source_db_obj.initialize_ordered_bulk_op()
source_bulk_bak = source_db_obj.initialize_ordered_bulk_op()
target_db_obj = db_help.create_db_obj(target_db, target_col)
target_bulk = target_db_obj.initialize_ordered_bulk_op()
target_bulk_bak = target_db_obj.initialize_ordered_bulk_op()
source_find_results = source_db_obj.find(filter)
start = datetime.now()
for doc in source_find_results:
doc['archived_at'] = db_help.getUTCTime()
target_bulk.insert(doc)
source_bulk.find({'_id': doc['_id']}).remove_one()
target_bulk_bak.find({'_id': doc['_id']}).remove_one()
doc.pop('archieved_at', None)
source_bulk_bak.insert(doc)
count += 1
if count % 1000 == 0:
logger.info("count: {}".format(count))
try:
execute_bulk_insert_remove(source_bulk, target_bulk)
except BulkWriteError as bwe:
execute_bulk_insert_remove(source_bulk_bak, target_bulk_bak)
logger.info("Bulk Write Error: {}".format(bwe.details))
raise
source_bulk = source_db_obj.initialize_ordered_bulk_op()
source_bulk_bak = source_db_obj.initialize_ordered_bulk_op()
target_bulk = target_db_obj.initialize_ordered_bulk_op()
target_bulk_bak = target_db_obj.initialize_ordered_bulk_op()
end = datetime.now()
logger.info("archived {} documents to {} in ms.".format(
count, target_col, (end - start)))
I'm using meteor 0.3.7 in Win7(32) and trying to create a simple logging system using 2 MongoDB collections to store data that are linked by DBRef.
The current pseudo schema is :
Users {
username : String,
password : String,
created : Timestamp,
}
Logs {
user_id : DBRef {$id, $ref}
message : String
}
I use server methods to insert the logs so I can do some upserts on the clients collection.
Now I want to do an old "left join" and display a list of the last n logs with the embedded User name.
I don't want to embed the Logs in Users because the most used operation is getting the last n logs. Embedding in my opinion was going to have a big impact in performance.
What is the best approach to achieve this?
Next it was great if possible to edit the User name and all items change theis name
Regards
Playing around with Cursor.observe answered my question. It may not be the most effective way of doing this, but solves my future problems of derefering DBRefs "links"
So for the server we need to publish a special collection. One that can enumerate the cursor and for each document search for the corresponding DBRef.
Bare in mind this implementation is hardcoded and should be done as a package like UnRefCollection.
Server Side
CC.Logs = new Meteor.Collection("logs");
CC.Users = new Meteor.Collection("users");
Meteor.publish('logsAndUsers', function (page, size) {
var self = this;
var startup = true;
var startupList = [], uniqArr = [];
page = page || 1;
size = size || 100;
var skip = (page - 1) * size;
var cursor = CC.Logs.find({}, {limit : size, skip : skip});
var handle = cursor.observe({
added : function(doc, idx){
var clone = _.clone(doc);
var refId = clone.user_id.oid; // showld search DBRefs
if (startup){
startupList.push(clone);
if (!_.contains(uniqArr, refId))
uniqArr.push(refId);
} else {
// Clients added logs
var deref = CC.Users.findOne({_id : refid});
clone.user = deref;
self.set('logsAndUsers', clone._id, clone);
self.flush();
}
},
removed : function(doc, idx){
self.unset('logsAndUsers', doc._id, _.keys(doc));
self.flush();
},
changed : function(new_document, idx, old_document){
var set = {};
_.each(new_document, function (v, k) {
if (!_.isEqual(v, old_document[k]))
set[k] = v;
});
self.set('logsAndUsers', new_document._id, set);
var dead_keys = _.difference(_.keys(old_document), _.keys(new_document));
self.unset('logsAndUsers', new_document._id, dead_keys);
self.flush();
},
moved : function(document, old_index, new_index){
// Not used
}
});
self.onStop(function(){
handle.stop();
});
// Deref on first Run
var derefs = CC.Users.find({_id : {$in : uniqArr} }).fetch();
_.forEach(startupList, function (item){
_.forEach(derefs, function(ditems){
if (item["user_id"].oid === ditems._id){
item.user = ditems;
return false;
}
});
self.set('logsAndUsers', item._id, item);
});
delete derefs; // Not needed anymore
startup = false;
self.complete();
self.flush();
});
For each added logs document it'll search the users collection and try to add to the logs collection the missing information.
The added function is called for each document in the logs collection in the first run I created a startupList and an array of unique users ids so for the first run it'll query the db only once. Its a good idea to put a paging mechanism to speed up things.
Client Side
On the client, subscribe to the logsAndUsers collection, if you want to make changes do it directly to the Logs collection.
LogsAndUsers = new Meteor.collection('logsAndUser');
Logs = new Meteor.colection('logs'); // Changes here are observed in the LogsAndUsers collection
Meteor.autosubscribe(function () {
var page = Session.get('page') || 1;
Meteor.subscribe('logsAndUsers', page);
});
Why not just also store the username in the logs collection as well?
Then you can query on them directly without needing any kind of "join"
If for some reason you need to be able to handle that username change, you just fetch the user object by name, then query on Logs with { user_id : user._id }
I'm trying to use as much of the OOTB sync and RESTful functionality in Backbone. I have a Web API set up for basic CRUD for my models. I have:
var SearchModel = Backbone.Model.extend({});
var SearchMappingModel = Backbone.Model.extend({});
var SearchComponentModel = Backbone.Model.extend({});
var SearchCollection = Backbone.Collection.extend({});
var SearchMappingCollection = Backbone.Collection.extend({});
var SearchComponentCollection = Backbone.Collection.extend({});
For every Search there is 1-to-many SearchMappings, and for every SearchMapping, there are 1-to-many SearchComponents. My URLs for sync would be something like, "/search" for the Search collection, "'/searchmapping/' + searchId" for the SearchMapping collection, and "'/searchcomponent/' + mappingId" for the SearchComponent collection.
My question is, since each collection is dependent on the previous one, is there a way I can make a cascading relationship in backbone to minimize my code and use as much of the basic sync functionality that's already there?
My initial thought is to create a collection within a collection and write my own .fetch() to first fetch the parent collection and on its success then fetch the child, which will then also get its child after its own success, like this:
var SearchCollection = Backbone.Collection.extend({
model: SearchModel,
initialize: function (data) {
this.url = baseURL + "/search";
this.data = data;
this.SearchMappingCollection = new SearchMappingCollection();
},
fetchData: function () {
this.fetch({
success: _.bind(function (results) {
this.fetchListSuccess(results);
}, this)
});
},
fetchListSuccess: function (results) {
this.SearchMappingCollection.fetchData(results);
}
The same would be done on a .save(). This may be a good way of doing it, but wanted to get feedback from anyone else that's done something similar.
I ended up not using a cascading format. It seems that it was adding more complexity and giving nothing in return. All 3 collections now sit on the controller level, and I just load the next collection after each collection is loaded on each "reset" event.
I'm using meteor 0.3.7 in Win7(32) and trying to create a simple logging system using 2 MongoDB collections to store data that are linked by DBRef.
The current pseudo schema is :
Users {
username : String,
password : String,
created : Timestamp,
}
Logs {
user_id : DBRef {$id, $ref}
message : String
}
I use server methods to insert the logs so I can do some upserts on the clients collection.
Now I want to do an old "left join" and display a list of the last n logs with the embedded User name.
I don't want to embed the Logs in Users because the most used operation is getting the last n logs. Embedding in my opinion was going to have a big impact in performance.
What is the best approach to achieve this?
Next it was great if possible to edit the User name and all items change theis name
Regards
Playing around with Cursor.observe answered my question. It may not be the most effective way of doing this, but solves my future problems of derefering DBRefs "links"
So for the server we need to publish a special collection. One that can enumerate the cursor and for each document search for the corresponding DBRef.
Bare in mind this implementation is hardcoded and should be done as a package like UnRefCollection.
Server Side
CC.Logs = new Meteor.Collection("logs");
CC.Users = new Meteor.Collection("users");
Meteor.publish('logsAndUsers', function (page, size) {
var self = this;
var startup = true;
var startupList = [], uniqArr = [];
page = page || 1;
size = size || 100;
var skip = (page - 1) * size;
var cursor = CC.Logs.find({}, {limit : size, skip : skip});
var handle = cursor.observe({
added : function(doc, idx){
var clone = _.clone(doc);
var refId = clone.user_id.oid; // showld search DBRefs
if (startup){
startupList.push(clone);
if (!_.contains(uniqArr, refId))
uniqArr.push(refId);
} else {
// Clients added logs
var deref = CC.Users.findOne({_id : refid});
clone.user = deref;
self.set('logsAndUsers', clone._id, clone);
self.flush();
}
},
removed : function(doc, idx){
self.unset('logsAndUsers', doc._id, _.keys(doc));
self.flush();
},
changed : function(new_document, idx, old_document){
var set = {};
_.each(new_document, function (v, k) {
if (!_.isEqual(v, old_document[k]))
set[k] = v;
});
self.set('logsAndUsers', new_document._id, set);
var dead_keys = _.difference(_.keys(old_document), _.keys(new_document));
self.unset('logsAndUsers', new_document._id, dead_keys);
self.flush();
},
moved : function(document, old_index, new_index){
// Not used
}
});
self.onStop(function(){
handle.stop();
});
// Deref on first Run
var derefs = CC.Users.find({_id : {$in : uniqArr} }).fetch();
_.forEach(startupList, function (item){
_.forEach(derefs, function(ditems){
if (item["user_id"].oid === ditems._id){
item.user = ditems;
return false;
}
});
self.set('logsAndUsers', item._id, item);
});
delete derefs; // Not needed anymore
startup = false;
self.complete();
self.flush();
});
For each added logs document it'll search the users collection and try to add to the logs collection the missing information.
The added function is called for each document in the logs collection in the first run I created a startupList and an array of unique users ids so for the first run it'll query the db only once. Its a good idea to put a paging mechanism to speed up things.
Client Side
On the client, subscribe to the logsAndUsers collection, if you want to make changes do it directly to the Logs collection.
LogsAndUsers = new Meteor.collection('logsAndUser');
Logs = new Meteor.colection('logs'); // Changes here are observed in the LogsAndUsers collection
Meteor.autosubscribe(function () {
var page = Session.get('page') || 1;
Meteor.subscribe('logsAndUsers', page);
});
Why not just also store the username in the logs collection as well?
Then you can query on them directly without needing any kind of "join"
If for some reason you need to be able to handle that username change, you just fetch the user object by name, then query on Logs with { user_id : user._id }