How do I perform the SQL Join equivalent in MongoDB?
For example say you have two collections (users and comments) and I want to pull all the comments with pid=444 along with the user info for each.
comments
{ uid:12345, pid:444, comment="blah" }
{ uid:12345, pid:888, comment="asdf" }
{ uid:99999, pid:444, comment="qwer" }
users
{ uid:12345, name:"john" }
{ uid:99999, name:"mia" }
Is there a way to pull all the comments with a certain field (eg. ...find({pid:444}) ) and the user information associated with each comment in one go?
At the moment, I am first getting the comments which match my criteria, then figuring out all the uid's in that result set, getting the user objects, and merging them with the comment's results. Seems like I am doing it wrong.
As of Mongo 3.2 the answers to this question are mostly no longer correct. The new $lookup operator added to the aggregation pipeline is essentially identical to a left outer join:
https://docs.mongodb.org/master/reference/operator/aggregation/lookup/#pipe._S_lookup
From the docs:
{
$lookup:
{
from: <collection to join>,
localField: <field from the input documents>,
foreignField: <field from the documents of the "from" collection>,
as: <output array field>
}
}
Of course Mongo is not a relational database, and the devs are being careful to recommend specific use cases for $lookup, but at least as of 3.2 doing join is now possible with MongoDB.
We can merge/join all data inside only one collection with a easy function in few lines using the mongodb client console, and now we could be able of perform the desired query.
Below a complete example,
.- Authors:
db.authors.insert([
{
_id: 'a1',
name: { first: 'orlando', last: 'becerra' },
age: 27
},
{
_id: 'a2',
name: { first: 'mayra', last: 'sanchez' },
age: 21
}
]);
.- Categories:
db.categories.insert([
{
_id: 'c1',
name: 'sci-fi'
},
{
_id: 'c2',
name: 'romance'
}
]);
.- Books
db.books.insert([
{
_id: 'b1',
name: 'Groovy Book',
category: 'c1',
authors: ['a1']
},
{
_id: 'b2',
name: 'Java Book',
category: 'c2',
authors: ['a1','a2']
},
]);
.- Book lending
db.lendings.insert([
{
_id: 'l1',
book: 'b1',
date: new Date('01/01/11'),
lendingBy: 'jose'
},
{
_id: 'l2',
book: 'b1',
date: new Date('02/02/12'),
lendingBy: 'maria'
}
]);
.- The magic:
db.books.find().forEach(
function (newBook) {
newBook.category = db.categories.findOne( { "_id": newBook.category } );
newBook.lendings = db.lendings.find( { "book": newBook._id } ).toArray();
newBook.authors = db.authors.find( { "_id": { $in: newBook.authors } } ).toArray();
db.booksReloaded.insert(newBook);
}
);
.- Get the new collection data:
db.booksReloaded.find().pretty()
.- Response :)
{
"_id" : "b1",
"name" : "Groovy Book",
"category" : {
"_id" : "c1",
"name" : "sci-fi"
},
"authors" : [
{
"_id" : "a1",
"name" : {
"first" : "orlando",
"last" : "becerra"
},
"age" : 27
}
],
"lendings" : [
{
"_id" : "l1",
"book" : "b1",
"date" : ISODate("2011-01-01T00:00:00Z"),
"lendingBy" : "jose"
},
{
"_id" : "l2",
"book" : "b1",
"date" : ISODate("2012-02-02T00:00:00Z"),
"lendingBy" : "maria"
}
]
}
{
"_id" : "b2",
"name" : "Java Book",
"category" : {
"_id" : "c2",
"name" : "romance"
},
"authors" : [
{
"_id" : "a1",
"name" : {
"first" : "orlando",
"last" : "becerra"
},
"age" : 27
},
{
"_id" : "a2",
"name" : {
"first" : "mayra",
"last" : "sanchez"
},
"age" : 21
}
],
"lendings" : [ ]
}
I hope this lines can help you.
This page on the official mongodb site addresses exactly this question:
https://mongodb-documentation.readthedocs.io/en/latest/ecosystem/tutorial/model-data-for-ruby-on-rails.html
When we display our list of stories, we'll need to show the name of the user who posted the story. If we were using a relational database, we could perform a join on users and stores, and get all our objects in a single query. But MongoDB does not support joins and so, at times, requires bit of denormalization. Here, this means caching the 'username' attribute.
Relational purists may be feeling uneasy already, as if we were violating some universal law. But let’s bear in mind that MongoDB collections are not equivalent to relational tables; each serves a unique design objective. A normalized table provides an atomic, isolated chunk of data. A document, however, more closely represents an object as a whole. In the case of a social news site, it can be argued that a username is intrinsic to the story being posted.
You have to do it the way you described. MongoDB is a non-relational database and doesn't support joins.
With right combination of $lookup, $project and $match, you can join mutiple tables on multiple parameters. This is because they can be chained multiple times.
Suppose we want to do following (reference)
SELECT S.* FROM LeftTable S
LEFT JOIN RightTable R ON S.ID = R.ID AND S.MID = R.MID
WHERE R.TIM > 0 AND S.MOB IS NOT NULL
Step 1: Link all tables
you can $lookup as many tables as you want.
$lookup - one for each table in query
$unwind - correctly denormalises data , else it'd be wrapped in arrays
Python code..
db.LeftTable.aggregate([
# connect all tables
{"$lookup": {
"from": "RightTable",
"localField": "ID",
"foreignField": "ID",
"as": "R"
}},
{"$unwind": "R"}
])
Step 2: Define all conditionals
$project : define all conditional statements here, plus all the variables you'd like to select.
Python Code..
db.LeftTable.aggregate([
# connect all tables
{"$lookup": {
"from": "RightTable",
"localField": "ID",
"foreignField": "ID",
"as": "R"
}},
{"$unwind": "R"},
# define conditionals + variables
{"$project": {
"midEq": {"$eq": ["$MID", "$R.MID"]},
"ID": 1, "MOB": 1, "MID": 1
}}
])
Step 3: Join all the conditionals
$match - join all conditions using OR or AND etc. There can be multiples of these.
$project: undefine all conditionals
Complete Python Code..
db.LeftTable.aggregate([
# connect all tables
{"$lookup": {
"from": "RightTable",
"localField": "ID",
"foreignField": "ID",
"as": "R"
}},
{"$unwind": "$R"},
# define conditionals + variables
{"$project": {
"midEq": {"$eq": ["$MID", "$R.MID"]},
"ID": 1, "MOB": 1, "MID": 1
}},
# join all conditionals
{"$match": {
"$and": [
{"R.TIM": {"$gt": 0}},
{"MOB": {"$exists": True}},
{"midEq": {"$eq": True}}
]}},
# undefine conditionals
{"$project": {
"midEq": 0
}}
])
Pretty much any combination of tables, conditionals and joins can be done in this manner.
You can join two collection in Mongo by using lookup which is offered in 3.2 version. In your case the query would be
db.comments.aggregate({
$lookup:{
from:"users",
localField:"uid",
foreignField:"uid",
as:"users_comments"
}
})
or you can also join with respect to users then there will be a little change as given below.
db.users.aggregate({
$lookup:{
from:"comments",
localField:"uid",
foreignField:"uid",
as:"users_comments"
}
})
It will work just same as left and right join in SQL.
As others have pointed out you are trying to create a relational database from none relational database which you really don't want to do but anyways, if you have a case that you have to do this here is a solution you can use. We first do a foreach find on collection A( or in your case users) and then we get each item as an object then we use object property (in your case uid) to lookup in our second collection (in your case comments) if we can find it then we have a match and we can print or do something with it.
Hope this helps you and good luck :)
db.users.find().forEach(
function (object) {
var commonInBoth=db.comments.findOne({ "uid": object.uid} );
if (commonInBoth != null) {
printjson(commonInBoth) ;
printjson(object) ;
}else {
// did not match so we don't care in this case
}
});
Here's an example of a "join" * Actors and Movies collections:
https://github.com/mongodb/cookbook/blob/master/content/patterns/pivot.txt
It makes use of .mapReduce() method
* join - an alternative to join in document-oriented databases
$lookup (aggregation)
Performs a left outer join to an unsharded collection in the same database to filter in documents from the “joined” collection for processing. To each input document, the $lookup stage adds a new array field whose elements are the matching documents from the “joined” collection. The $lookup stage passes these reshaped documents to the next stage.
The $lookup stage has the following syntaxes:
Equality Match
To perform an equality match between a field from the input documents with a field from the documents of the “joined” collection, the $lookup stage has the following syntax:
{
$lookup:
{
from: <collection to join>,
localField: <field from the input documents>,
foreignField: <field from the documents of the "from" collection>,
as: <output array field>
}
}
The operation would correspond to the following pseudo-SQL statement:
SELECT *, <output array field>
FROM collection
WHERE <output array field> IN (SELECT <documents as determined from the pipeline>
FROM <collection to join>
WHERE <pipeline> );
Mongo URL
It depends on what you're trying to do.
You currently have it set up as a normalized database, which is fine, and the way you are doing it is appropriate.
However, there are other ways of doing it.
You could have a posts collection that has imbedded comments for each post with references to the users that you can iteratively query to get. You could store the user's name with the comments, you could store them all in one document.
The thing with NoSQL is it's designed for flexible schemas and very fast reading and writing. In a typical Big Data farm the database is the biggest bottleneck, you have fewer database engines than you do application and front end servers...they're more expensive but more powerful, also hard drive space is very cheap comparatively. Normalization comes from the concept of trying to save space, but it comes with a cost at making your databases perform complicated Joins and verifying the integrity of relationships, performing cascading operations. All of which saves the developers some headaches if they designed the database properly.
With NoSQL, if you accept that redundancy and storage space aren't issues because of their cost (both in processor time required to do updates and hard drive costs to store extra data), denormalizing isn't an issue (for embedded arrays that become hundreds of thousands of items it can be a performance issue, but most of the time that's not a problem). Additionally you'll have several application and front end servers for every database cluster. Have them do the heavy lifting of the joins and let the database servers stick to reading and writing.
TL;DR: What you're doing is fine, and there are other ways of doing it. Check out the mongodb documentation's data model patterns for some great examples. http://docs.mongodb.org/manual/data-modeling/
There is a specification that a lot of drivers support that's called DBRef.
DBRef is a more formal specification for creating references between documents. DBRefs (generally) include a collection name as well as an object id. Most developers only use DBRefs if the collection can change from one document to the next. If your referenced collection will always be the same, the manual references outlined above are more efficient.
Taken from MongoDB Documentation: Data Models > Data Model Reference >
Database References
Before 3.2.6, Mongodb does not support join query as like mysql. below solution which works for you.
db.getCollection('comments').aggregate([
{$match : {pid : 444}},
{$lookup: {from: "users",localField: "uid",foreignField: "uid",as: "userData"}},
])
You can run SQL queries including join on MongoDB with mongo_fdw from Postgres.
MongoDB does not allow joins, but you can use plugins to handle that. Check the mongo-join plugin. It's the best and I have already used it. You can install it using npm directly like this npm install mongo-join. You can check out the full documentation with examples.
(++) really helpful tool when we need to join (N) collections
(--) we can apply conditions just on the top level of the query
Example
var Join = require('mongo-join').Join, mongodb = require('mongodb'), Db = mongodb.Db, Server = mongodb.Server;
db.open(function (err, Database) {
Database.collection('Appoint', function (err, Appoints) {
/* we can put conditions just on the top level */
Appoints.find({_id_Doctor: id_doctor ,full_date :{ $gte: start_date },
full_date :{ $lte: end_date }}, function (err, cursor) {
var join = new Join(Database).on({
field: '_id_Doctor', // <- field in Appoints document
to: '_id', // <- field in User doc. treated as ObjectID automatically.
from: 'User' // <- collection name for User doc
}).on({
field: '_id_Patient', // <- field in Appoints doc
to: '_id', // <- field in User doc. treated as ObjectID automatically.
from: 'User' // <- collection name for User doc
})
join.toArray(cursor, function (err, joinedDocs) {
/* do what ever you want here */
/* you can fetch the table and apply your own conditions */
.....
.....
.....
resp.status(200);
resp.json({
"status": 200,
"message": "success",
"Appoints_Range": joinedDocs,
});
return resp;
});
});
You can do it using the aggregation pipeline, but it's a pain to write it yourself.
You can use mongo-join-query to create the aggregation pipeline automatically from your query.
This is how your query would look like:
const mongoose = require("mongoose");
const joinQuery = require("mongo-join-query");
joinQuery(
mongoose.models.Comment,
{
find: { pid:444 },
populate: ["uid"]
},
(err, res) => (err ? console.log("Error:", err) : console.log("Success:", res.results))
);
Your result would have the user object in the uid field and you can link as many levels deep as you want. You can populate the reference to the user, which makes reference to a Team, which makes reference to something else, etc..
Disclaimer: I wrote mongo-join-query to tackle this exact problem.
playORM can do it for you using S-SQL(Scalable SQL) which just adds partitioning such that you can do joins within partitions.
Nope, it doesn't seem like you're doing it wrong. MongoDB joins are "client side". Pretty much like you said:
At the moment, I am first getting the comments which match my criteria, then figuring out all the uid's in that result set, getting the user objects, and merging them with the comment's results. Seems like I am doing it wrong.
1) Select from the collection you're interested in.
2) From that collection pull out ID's you need
3) Select from other collections
4) Decorate your original results.
It's not a "real" join, but it's actually alot more useful than a SQL join because you don't have to deal with duplicate rows for "many" sided joins, instead your decorating the originally selected set.
There is alot of nonsense and FUD on this page. Turns out 5 years later MongoDB is still a thing.
I think, if You need normalized data tables - You need to try some other database solutions.
But I've foun that sollution for MOngo on Git
By the way, in inserts code - it has movie's name, but noi movie's ID.
Problem
You have a collection of Actors with an array of the Movies they've done.
You want to generate a collection of Movies with an array of Actors in each.
Some sample data
db.actors.insert( { actor: "Richard Gere", movies: ['Pretty Woman', 'Runaway Bride', 'Chicago'] });
db.actors.insert( { actor: "Julia Roberts", movies: ['Pretty Woman', 'Runaway Bride', 'Erin Brockovich'] });
Solution
We need to loop through each movie in the Actor document and emit each Movie individually.
The catch here is in the reduce phase. We cannot emit an array from the reduce phase, so we must build an Actors array inside of the "value" document that is returned.
The code
map = function() {
for(var i in this.movies){
key = { movie: this.movies[i] };
value = { actors: [ this.actor ] };
emit(key, value);
}
}
reduce = function(key, values) {
actor_list = { actors: [] };
for(var i in values) {
actor_list.actors = values[i].actors.concat(actor_list.actors);
}
return actor_list;
}
Notice how actor_list is actually a javascript object that contains an array. Also notice that map emits the same structure.
Run the following to execute the map / reduce, output it to the "pivot" collection and print the result:
printjson(db.actors.mapReduce(map, reduce, "pivot"));
db.pivot.find().forEach(printjson);
Here is the sample output, note that "Pretty Woman" and "Runaway Bride" have both "Richard Gere" and "Julia Roberts".
{ "_id" : { "movie" : "Chicago" }, "value" : { "actors" : [ "Richard Gere" ] } }
{ "_id" : { "movie" : "Erin Brockovich" }, "value" : { "actors" : [ "Julia Roberts" ] } }
{ "_id" : { "movie" : "Pretty Woman" }, "value" : { "actors" : [ "Richard Gere", "Julia Roberts" ] } }
{ "_id" : { "movie" : "Runaway Bride" }, "value" : { "actors" : [ "Richard Gere", "Julia Roberts" ] } }
We can merge two collection by using mongoDB sub query. Here is example,
Commentss--
`db.commentss.insert([
{ uid:12345, pid:444, comment:"blah" },
{ uid:12345, pid:888, comment:"asdf" },
{ uid:99999, pid:444, comment:"qwer" }])`
Userss--
db.userss.insert([
{ uid:12345, name:"john" },
{ uid:99999, name:"mia" }])
MongoDB sub query for JOIN--
`db.commentss.find().forEach(
function (newComments) {
newComments.userss = db.userss.find( { "uid": newComments.uid } ).toArray();
db.newCommentUsers.insert(newComments);
}
);`
Get result from newly generated Collection--
db.newCommentUsers.find().pretty()
Result--
`{
"_id" : ObjectId("5511236e29709afa03f226ef"),
"uid" : 12345,
"pid" : 444,
"comment" : "blah",
"userss" : [
{
"_id" : ObjectId("5511238129709afa03f226f2"),
"uid" : 12345,
"name" : "john"
}
]
}
{
"_id" : ObjectId("5511236e29709afa03f226f0"),
"uid" : 12345,
"pid" : 888,
"comment" : "asdf",
"userss" : [
{
"_id" : ObjectId("5511238129709afa03f226f2"),
"uid" : 12345,
"name" : "john"
}
]
}
{
"_id" : ObjectId("5511236e29709afa03f226f1"),
"uid" : 99999,
"pid" : 444,
"comment" : "qwer",
"userss" : [
{
"_id" : ObjectId("5511238129709afa03f226f3"),
"uid" : 99999,
"name" : "mia"
}
]
}`
Hope so this will help.
This is sample of one document in my mongodb collection page_link_titles:
{
"_id" : ObjectId("553b11f30b81511d64152416"),
"id" : 36470831,
"linkTitles" : [
"Syrian civil war",
"Damascus",
"Geographic coordinate system",
"Bashar al-Assad",
"Al Jazeera English",
"Free Syrian Army",
...
"February 2012 Aleppo bombings",
"2012 Deir ez-Zor bombing",
"Aleppo University bombings"
]
}
I want to find all the documents that the text in their linkTitles contains a phrase like '%term1%' or '%term2%' or (so on). term1 and term2 must have a line break in both sides. For example looking into "Syrian civil war". If term1 = "war" I want this document to be returned as the result of query, however if term1 = "yria" which is a part of a word in this document, it shouldn't be returned.
This is my java code:
for (String term : segment.terms) {
DBObject clause1 = new BasicDBObject("linkTitles",
java.util.regex.Pattern.compile("\\b"
+ stprocess.singularize(term) + "\\b"));
or.add(clause1);
}
DBObject mongoQuery = new BasicDBObject("$or", or);
DBCursor cursor = pageLinks.find(mongoQuery);
In line: java.util.regex.Pattern.compile("\\b"+ stprocess.singularize(term) + "\\b")); I only assumed line break. I don't know how I should write the regex to consider all my conditions : line break, case insensitive, like.
Any ideas?
It is possible to do a regular expression that achieves what you want. You can also use a single regular expression rather using $or.
I'm using the shell for a quick example and wanting to search for boxer or cat. First insert the test data:
db.test.drop()
db.test.insert([
{ "a" : "Boxer One" },
{ "a" : "A boxer dog" },
{ "a" : "A box shouldn't match" },
{ "a" : "should match BOXER" },
{ "a" : "wont match as this it the plural BOXERs" },
{ "a" : "also match on cat" }])
Using the following regular expression we can search for all our terms:
/(^|\b)(boxer|cat)(\b|$)/i
+---+ +-------+ +---+
| | |
| | |
Start or space | Space or end
|
Search terms
And do a find like so:
db.test.find({a: /(^|\b)(boxer|cat)(\b|$)/i})
That query will return the following results:
{ "_id" : ObjectId("555f18eee7b6d1b7e622de36"), "a" : "Boxer One" }
{ "_id" : ObjectId("555f18eee7b6d1b7e622de37"), "a" : "A boxer dog" }
{ "_id" : ObjectId("555f18eee7b6d1b7e622de39"), "a" : "should match BOXER" }
{ "_id" : ObjectId("555f18eee7b6d1b7e622de3b"), "a" : "also match on cat" }
In Java you might build this query up like so:
StringBuilder singularizedTerms = new StringBuilder();
for (String term : terms) {
singularizedTerms.append("|").append(stprocess.singularize(term));
}
String regexPattern = format("(^|\\b)(%s)(\\b|$)", singularizedTerms.substring(1));
Pattern regex = Pattern.compile(regexPattern, Pattern.CASE_INSENSITIVE);
Theres two problems with this approach.
It will be slow
It can't use an index so will do a full scan of the collection, if you have 10 million documents it will check each one!
It won't match plurals
For example it won't match the document containing "BOXERs" because our regular expression explicitly doesn't allow for partial matches!
Text indexes support this. Using an index will make the operation faster as well as matching plural or single values, for example:
db.test.createIndex( { a: "text" } )
db.test.find({ $text: { $search: "boxer cat"}})
{ "_id" : ObjectId("555f18eee7b6d1b7e622de3b"), "a" : "also match on cat" }
{ "_id" : ObjectId("555f18eee7b6d1b7e622de3a"), "a" : "wont match as this it the plural BOXERs" }
{ "_id" : ObjectId("555f18eee7b6d1b7e622de36"), "a" : "Boxer One" }
{ "_id" : ObjectId("555f18eee7b6d1b7e622de37"), "a" : "A boxer dog" }
{ "_id" : ObjectId("555f18eee7b6d1b7e622de39"), "a" : "should match BOXER" }