Sort the basic of Number of documents in ElasticSearch - elasticsearch

Question

I am saving user relations in ES Index
i.e
{'id' => 1, 'User_id_1' => '2001', 'relation' => 'friend', 'User_id_2' => '1002'}
{'id' => 2, 'User_id_1' => '2002', 'relation' => 'friend', 'User_id_2' => '1002'}
{'id' => 3, 'User_id_1' => '2002', 'relation' => 'friend', 'User_id_2' => '1001'}
{'id' => 4, 'User_id_1' => '2003', 'relation' => 'friend', 'User_id_2' => '1003'}
no suppose i want to get the user_id_2 who has most friends,
in above case its 1002 as 2001, and 2002 are its friends. (Count = 2)
I just can't figure out the query
Thanks.
EDIT:
Well as suggested by #imotov, term facets is very good choice, but
The problem I have is 2 Indexes
1st index is for saving the main docs and 2nd index for saving the relations
now problem is
Suppose I have 100 USER Docs in my main index, only 50 of them has made relations, so I'll have only 50 USER Docs in my relationship index
So when i implement the "term facet", it sorts the results and gives the correct output i want, but I am missing those left 50 users who don't have any relations yet, i need them in my final output after the 50 sorted users.

Answer 1

First of all, we need to ensure that relationships saved in ES are unique. It can be done by replacing arbitrary ids with ids constructed from user_id_1, relation and user_id_2. We also need to make sure that analyzer for user_ids doesn't produce multiple tokens. If ids are strings, they have to be indexed not_analyzed. With these two conditions satisfied, we can simply use terms facet query for the field user_id_2 on the result list limited by relation:friend. This query will retrieve top user_id_2 ids sorted by number of occurrences in the index. All together it could look something like this:
curl -XPUT http://localhost:9200/relationships -d '{
"mappings" : {
"relation" : {
"_source" : {"enabled" : false },
"properties" : {
"user_id_1": { "type": "string", "index" : "not_analyzed"},
"relation": { "type": "string", "index" : "not_analyzed"},
"user_id_2": { "type": "string", "index" : "not_analyzed"}
}
}
}
}'
curl -XPUT http://localhost:9200/relationships/relation/2001-friend-1002 -d '{"user_id_1": "2001", "relation":"friend", "user_id_2": "1002"}'
curl -XPUT http://localhost:9200/relationships/relation/2002-friend-1002 -d '{"user_id_1": "2002", "relation":"friend", "user_id_2": "1002"}'
curl -XPUT http://localhost:9200/relationships/relation/2002-friend-1001 -d '{"user_id_1": "2002", "relation":"friend", "user_id_2": "1001"}'
curl -XPUT http://localhost:9200/relationships/relation/2003-friend-1003 -d '{"user_id_1": "2003", "relation":"friend", "user_id_2": "1003"}'
curl -XPOST http://localhost:9200/relationships/_refresh
echo
curl -XGET 'http://localhost:9200/relationships/relation/_search?pretty=true&search_type=count' -d '{
"query": {
"term" : {
"relation" : "friend"
}
},
"facets" : {
"popular" : {
"terms" : {
"field" : "user_id_2"
}
}
}
}'
Please, note that due to distributed nature of facets calculation, counts reported by the facet query might be lower than the actual number of records if multiple shards are used. See elasticsearch issue 1832
EDIT:
There are two solutions for the edited question. One solution is to use facet on two fields:
curl -XPUT http://localhost:9200/relationships -d '{
"mappings" : {
"relation" : {
"_source" : {"enabled" : false },
"properties" : {
"user_id_1": { "type": "string", "index" : "not_analyzed"},
"relation": { "type": "string", "index" : "not_analyzed"},
"user_id_2": { "type": "string", "index" : "not_analyzed"}
}
}
}
}'
curl -XPUT http://localhost:9200/users -d '{
"mappings" : {
"user" : {
"_source" : {"enabled" : false },
"properties" : {
"user_id": { "type": "string", "index" : "not_analyzed"}
}
}
}
}'
curl -XPUT http://localhost:9200/users/user/1001 -d '{"user_id": 1001}'
curl -XPUT http://localhost:9200/users/user/1002 -d '{"user_id": 1002}'
curl -XPUT http://localhost:9200/users/user/1003 -d '{"user_id": 1003}'
curl -XPUT http://localhost:9200/users/user/1004 -d '{"user_id": 1004}'
curl -XPUT http://localhost:9200/users/user/1005 -d '{"user_id": 1005}'
curl -XPUT http://localhost:9200/relationships/relation/2001-friend-1002 -d '{"user_id_1": "2001", "relation":"friend", "user_id_2": "1002"}'
curl -XPUT http://localhost:9200/relationships/relation/2002-friend-1002 -d '{"user_id_1": "2002", "relation":"friend", "user_id_2": "1002"}'
curl -XPUT http://localhost:9200/relationships/relation/2002-friend-1001 -d '{"user_id_1": "2002", "relation":"friend", "user_id_2": "1001"}'
curl -XPUT http://localhost:9200/relationships/relation/2003-friend-1003 -d '{"user_id_1": "2003", "relation":"friend", "user_id_2": "1003"}'
curl -XPOST http://localhost:9200/relationships/_refresh
curl -XPOST http://localhost:9200/users/_refresh
echo
curl -XGET 'http://localhost:9200/relationships,users/_search?pretty=true&search_type=count' -d '{
"query": {
"indices" : {
"indices" : ["relationships"],
"query" : {
"filtered" : {
"query" : {
"term" : {
"relation" : "friend"
}
},
"filter" : {
"type" : {
"value" : "relation"
}
}
}
},
"no_match_query" : {
"filtered" : {
"query" : {
"match_all" : { }
},
"filter" : {
"type" : {
"value" : "user"
}
}
}
}
}
},
"facets" : {
"popular" : {
"terms" : {
"fields" : ["user_id", "user_id_2"]
}
}
}
}'
Another solution is to add "self" relation to the relationships index for every user when user is created. I would prefer the second solution since it seems to be less complicated.