Develop Reference

"Document interning" in Mongo

I a lot of documents which I know will rarely change and are very similar to each other, specifically I know they have a nested field in the document that is always the same (for some of them)
{
"docid": 1
"nested_field_that_will_always_be_the_same": {
"title": "this will always be the same"
"desc": "this will always be the same, too"
}
}
{
"docid": 2
"nested_field_that_will_always_be_the_same": {
"title": "this will always be the same"
"desc": "this will always be the same, too"
}
}
I don't want to store the same document over and over again, instead I want Mongo to "intern" this field, i.e only store it once and the rest will only store pointers to it.
Something like:
{
"docid": 1
"nested_field_that_will_always_be_the_same": {
"title": "this will always be the same"
"desc": "this will always be the same, too"
}
}
{
"docid": 2
"nested_field_that_will_always_be_the_same": <pointer to doc1.nested_field_that_will_always_be_the_same>
}
Now, of course, I can take out this nested field into a different document and then have Mongo reference its _id field, but I am not looking for app-side solution, because this collection is being accessed via multiple workers and I don't have all the documents that have the same nested_field_that_will_always_be_the_same at any given moment.
Instead, I want a solution provided by Mongo to only store this field once for every instance it is unique.
How can I do that?
I am using Pymongo.

This is quite an interesting challenge - I don't think a "pure" mongo solution is possible - you'll still have to modify your app code at insertion time. Quite interested to see if anyone does come up with a pure solution
What I'd probably do is add a unique index on the nested document with a partialFilterExpression, the index ensures you can quickly find the ID of the matching document, and the unique enforces this strictly.
Something like this (I shortened your field to nested for brevity):
collection.createIndex(
{ nested: 1 },
{ unique: true, partialFilterExpression: { nested: { $type: "object" } } }
);
Then for my inserts, I'd do the following (pseudo code)
try {
found = collection.findOne({ nested }, { projection: { _id: 1 } })
if (found) {
collection.insert({ docId, nested: found._id })
}
else {
collection.insert({ docId, nested })
}
}
catch (e) {
// test for E11000 and retry
}

Elasticsearch DocumentSimilarity dense_vector got multiple values for argument 'body'

I want to store Document Vectors in an Elasticsearch index in order to calculate document similarity. I'm using the Python client for Elasticsearch 7.8.0.
I have a (dummy) Elasticsearch index with the following mapping:
mapping = {
"mappings": {
"properties": {
"title_vector":{
"type": "dense_vector",
"dims": 3
}
}
}
}
es.indices.create(index="test_vector", body=mapping)
And I stored a bunch of vectors in the following way:
vectors = [[1,2,3],[2,2,2],[1,2,2],[2,2,2],[4,5,6],[1,1,1]]
for i, v in enumerate(vectors):
doc = {"title_vector": v}
es.create("test_vector", id=i, body=doc)
According to the documentation, my query to get the most similar documents, should be as follows:
doc = {
"query": {
"script_score": {
"query": {
"match_all": {}
},
"script": {
"source": "cosineSimilarity(params.queryVector, 'title_vector') + 1.0",
"params": {
"queryVector": [1,1,1]
}
}
}
}}
es.search("test_vector", body=doc)
But I'm getting
TypeError: search() got multiple values for argument 'body'
It seems more like a Python error than an Elastic error. But I can't really find the cause of the error and how I should structure my query differently in order to solve it.
Thanks in advance!
Edit: added Elasticsearch version

You are correct, it is a python error. So below is how the es.search is defined according to this link
search(body=None, index=None, params=None, headers=None)
As you see the first parameter is body.
Notice the es.search you have, you haven't specified the key in the first parameter i.e. body, index, params, headers. As a result, python interprets that as value for body according to the above method declaration.
Just add index="test_vector" instead of just "test_vector" in the first parameter and that should do the trick.
es.search(index="test_vector", body=doc)
Hope it helps!

How to index list of object in Elasticsearch?

A document format I ingest into ElasticSearch looks like this:
{
'id':'514d4e9f-09e7-4f13-b6c9-a0aa9b4f37a0'
'created':'2019-09-06 06:09:33.044433',
'meta':{
'userTags':[
{
'intensity':'1',
'sentiment':'0.84',
'keyword':'train'
},
{
'intensity':'1',
'sentiment':'-0.76',
'keyword':'amtrak'
}
]
}
}
...ingested with python:
r = requests.put(itemUrl, auth = authObj, json = document, headers = headers)
The idea here is that ElasticSearch will treat keyword, intensity and sentiment as fields that can be later queried. However, on ElasticSearch side I can observe that this is not happening (I use Kibana for search UI) -- instead, I see field "meta.userTags" with the value that is the whole list of objects.
How can I make ElasticSearch index elements within a list?

I used the document body you provided to create a new index 'testind' and type 'testTyp' using the Postman REST client.:
POST http://localhost:9200/testind/testTyp
{
"id":"514d4e9f-09e7-4f13-b6c9-a0aa9b4f37a0",
"created":"2019-09-06 06:09:33.044433",
"meta":{
"userTags":[
{
"intensity":"1",
"sentiment":"0.84",
"keyword":"train"
},
{
"intensity":"1",
"sentiment":"-0.76",
"keyword":"amtrak"
}
]
}
}
When I queried for the index's mapping this is what i get :
GET http://localhost:9200/testind/testTyp/_mapping
{
"testind":{
"mappings":{
"testTyp":{
"properties":{
"created":{
"type":"text",
"fields":{
"keyword":{
"type":"keyword",
"ignore_above":256
}
}
},
"id":{
"type":"text",
"fields":{
"keyword":{
"type":"keyword",
"ignore_above":256
}
}
},
"meta":{
"properties":{
"userTags":{
"properties":{
"intensity":{
"type":"text",
"fields":{
"keyword":{
"type":"keyword",
"ignore_above":256
}
}
},
"keyword":{
"type":"text",
"fields":{
"keyword":{
"type":"keyword",
"ignore_above":256
}
}
},
"sentiment":{
"type":"text",
"fields":{
"keyword":{
"type":"keyword",
"ignore_above":256
}
}
}
}
}
}
}
}
}
}
}
}
As you can see in the mapping the fields are part of the mapping and can be queried as per need in future, so I don't see the problem here as long as the field names are not one of these - https://www.elastic.co/guide/en/elasticsearch/reference/6.4/sql-syntax-reserved.html ( you might want to avoid the term 'keyword' as it might be confusing later when writing search queries as the fieldname and type are both same - 'keyword') . Also, note one thing, the mapping gets created via dynamic mapping (https://www.elastic.co/guide/en/elasticsearch/reference/6.3/dynamic-field-mapping.html#dynamic-field-mapping ) in Elasticsearch and so the data types are determined by elasticsearch based on the values you have provided.However, this may not be always accurate , so to prevent that you can use the PUT _mapping API to define your own mapping for the index, and then prevent new fields within a type from being added to mappings.

You don't need a special mapping to index a list - every field can contain one or more values of the same type. See array datatype.
In the case of a list of objects, they can be indexed as object or nested datatype. Per default elastic uses object datatype. In this case you can query meta.userTags.keyword or/and meta.userTags.sentiment. The result will allways contains whole documents with values matched independently, ie. searching keyword=train and sentiment=-0.76 you WILL find document with id=514d4e9f-09e7-4f13-b6c9-a0aa9b4f37a0.
If this is not what you want, you need to define nested datatype mapping for field userTags and use a nested query.

Upsert in mongoengine is not generating ObjectId

I am trying to execute an upsert function in mongoengine. That is, if a document is present, I want to update it with new values, and if it isn't present, I want to create and insert.
I have list of objects. These objects can or cannot have ObjectIds. Example is:
[
{
"id" : ObjectId("5c1791b7397df4a9c8518342"),
"type": "Line"
},
{
"type": "Line"
}
]
As you can see the second object does not have an Id.
I have written my query as:
updates = Collection.objects(
id=obj.get('id', None)).modify(
new=True,
upsert= True,
**update_dict
)
obj is each object when I iterate through the list.
Note: update_dict is another dict that gets its value from a function that returns the attributes to set. (For example: set__type: "Line")
Problem
The first object is getting modified just fine. However there is an error:
"'None' is not a valid ObjectId, it must be a 12-byte input or a
24-character hex string"
Clearly it's because of the obj.get('id', None) part.
So, is there a way that an id can be generated if it is passed as None?

I tried same thing with mongoose and nodejs and it works for me if i am using like below:
Here is i my array Object:
var arr = [
{
_id: "5c13de7d47zfe91e3484362f",
email: 'test1#gmail.com',
},
{
_id: "5c13de7d47zfe91e3484362f",
email: 'test2#gmail.com',
},
{
// _id: "5c66aa87751fz5368759f9bc", // Commented
email: 'test3#gmail.com',
}
]
Now i am iterating through the array as below with nodejs.
arr.forEach(async element => {
await Driver.findOneAndUpdate(
{
_id: Types.ObjectId(element._id)
},
{
email: element.email
},{ upsert: true, new: true }
).lean().exec();
});
And it works for me. It's updating documents in first two cases and inserting new doc for last case.
The main thing is to use Types.ObjectId which is used to specify a type of ObjectId. If i am doing it without specifying Schema.Types.ObjectId then it does not working.

get filtered embeded elements from all class on mongoengine

I got two class on Mongoengine:
class UserPoints(EmbeddedDocument):
user = ReferenceField(User, verbose_name='user')
points = IntField(verbose_name='points', required=True)
def __unicode__(self):
return self.points
And
class Local(Document):
token = StringField(max_length=250,verbose_name='token_identifier',unique=True)
points = ListField(EmbeddedDocumentField(UserPoints),required=False)
def __unicode__(self):
return self.name
If i do something like: "LP = Local.objects.filter(points__user=user)" I got all the locals with userpoints from my user. But i Want all the UserPoints from a User. How can i?
I try also: "lUs = UserPoints.objects.filter(user=user)" but i got an empty Array.
PD: I do something like this to solve the problem, but it's not efficient.
LDPoints = []
LP = Local.objects.filter(points__user=user)
print 'List P: '+str(len(LP))
for local in LP:
for points in local.points:
if points.user == user:
dPoints = parsePoints(points)
lDPoints.append(dPoints)

Adding to the original and getting venerable answer is that the aggregation framework has $filter now for some time, which is a lot cleaner that the $map and $setDifference method used in the original answer.
Local._get_collection().aggregate([
{ "$match": { "points.user": user } },
{ "$project": {
"token": 1,
"points": {
"$filter": {
"input": "$points",
"as": "el",
"cond": { "$eq": [ "$$el.user", user ] }
}
}
}}
])
The same principles apply though for obtaining "multiple" matches from an array in the collection you use the aggregate() method of the underlying driver, as called from _get_collection().
Original
The answer to avoid "filtering" your embedded documents for the selected "user" only is to use the aggregation framework. This allows you to manipulate the "array content" on the database server rather than filtering the results in your client code.
Aggregation is done with the raw pymongo driver methods, but since Mongoengine is built on top of this driver you access the raw collection object from your class with the ._get_collection() method:
Local._get_collection().aggregate([
# Match the documents that have the required user
{ "$match": {
"points.user": user
}},
# unwind the embedded array to de-normalize
{ "$unwind": "$points" },
# Matching now filters the elements
{ "$match": {
"points.user": user
}},
# Group back as an array
{ "$group": {
"_id": "$_id",
"token": { "$first": "$token" },
"points": { "$push": "$points" }
}}
])
If you have MongoDB 2.6 or greater on your server and your "user/points" combination is always unique you can alternately filter without the $unwind|$match|$group cycle using the $map and $setDifference operators available there:
Local._get_collection().aggregate([
# Match the documents that have the required user
{ "$match": {
"points.user": user
}},
# Filter the array in place
{ "$project": {
"token": 1,
"points": {
"$setDifference": [
{
"$map": {
"input": "$points",
"as": "el",
"in": {
"$cond": [
{ "$eq": [ "$$el.user", user ] },
"$$el",
false
]
}
}
},
[false]
]
}
}}
])
In the second case there the $cond is a ternary operator which takes a logical expression as it's first argument and the values to return when that expression is either true or false as it's other arguments. Inside the $map, each element is tested to see if the condition is true, in this case "is the user field equal to the selected user".
Either the content of that array position is returned or otherwise false. The $setDifference takes the resulting array and "filters" the false values out, so only the matching elements are returned.
In the legacy approach, the $unwind pipeline operator is used to effectively turn each array element into it's own document with all other parent properties. This allows you to apply the same $match condition, which unlike the initial query actually removes the documents which now as single elements no longer match your condition. You always want the first stage as there is no point processing this $unwind|$match combination on all of the documents that might not contain your matching condition.
The $group stage brings everything back into line per document. Using the $first option to return all other fields that were essentially duplicated by the $unwind and the $push operator to rebuild the array with the matching elements.
So while there no "built-in" methods to MongoEngine to do this sort of query, you can do this the MongoDB way by accessing the raw driver.
Also note that if you only expected one element to match in any array for your given "user" or other query, then you could alternately use the field projection form available to the raw driver as well. But the aggregation method is required for any more than one matching element of the array.