Elastic Search: aggregation, count by field

I inserted this data into elastic search:

[
  { "name": "Cassandra Irwin",  "location": "Monzon de Campos" ..     },
  { "name": "Gayle Mooney",     "location": "Villarroya del Campo" .. },
  { "name": "Angelita Charles", "location": "Revenga de Campos" ..    }, 
  { "name": "Sheppard Sweet",   "location": "Santiago del Campo" ..   },
  ..
  ..

Sidenote: to reproduce: 1) download: http://wmo.co/20160928_es_query/bulk.json 2) execute: curl -s -XPOST 'http://localhost:9200/testing/external/_bulk?pretty' --data-binary @bulk.json

Question: obtain a count of how many records there are per "location".

Solution 1: bucket aggregation .. doesn't give the desired results

curl -s -XPOST 'localhost:9200/testing/_search?pretty' -d '
{
  "aggs": {  "location_count": { "terms": { "field":"location",   "size":100 }}}
}' | jq  '.aggregations'

Result:

{"location_count":{"doc_count_error_upper_bound":0,"sum_other_doc_count":0,
 "buckets":[
    {"key":"campo",     "doc_count":47},
    {"key":"del",       "doc_count":47},
    {"key":"campos",    "doc_count":29},
    {"key":"de",        "doc_count":29},
    {"key":"villarroya","doc_count":15},
    {"key":"torre",     "doc_count":12},
    {"key":"monzon",    "doc_count":11},
    {"key":"santiago",  "doc_count":11},
    {"key":"pina",      "doc_count":9},
    {"key":"revenga",   "doc_count":9},
    {"key":"uleila",    "doc_count":9}
]}}

Problem: it splits the 'location' fields into words, and returns a doc count per word.

Solution 2: desired results, but performance worries.

I can do it using this query, pulling out ALL locations and doing the aggregation in jq (the every handy JSON cli-tool), but this can turn into a performance nightmare when applied to huge volumes of data :

curl -s -XPOST 'localhost:9200/testing/_search?pretty' -d '
 {
   "query": { "wildcard": { "location": "*" } }, "size":1000,
   "_source": ["location"]
 }' | jq  '[.hits.hits[] |
           {location:._source.location,"count":1}] |
           group_by(.location) |
           map({ key: .[0].location, value: map(.count)|add })'

Result:

[
  { "key": "Monzon de Campos",      "value": 11 },
  { "key": "Pina de Campos",        "value": 9  },
  { "key": "Revenga de Campos",     "value": 9  },
  { "key": "Santiago del Campo",    "value": 11 },
  { "key": "Torre del Campo",       "value": 12 },
  { "key": "Uleila del Campo",      "value": 9  },
  { "key": "Villarroya del Campo",  "value": 15 }
]

This is the exact result that I want.

QUESTION: how can I obtain the same results via elastic search query? (ie. with the aggregation handled by elastic search, and not by jq)

You need to add a not_analyzed sub-field to your location field.

First modify your mapping like this:

curl -XPOST 'http://localhost:9200/testing/_mapping/external' -d '{
   "properties": {
      "location": {
         "type": "string",
         "fields": {
            "raw": {
               "type": "string",
               "index": "not_analyzed"
            }
         }
      }
   }
}'

Then reindex your data again:

curl -s -XPOST 'http://localhost:9200/testing/external/_bulk?pretty' --data-binary @bulk.json

Finally, you'll be able to run your query like this (on the location.raw field) and get the results you expect:

curl -s -XPOST 'localhost:9200/testing/_search?pretty' -d '
{
  "aggs": {  "location_count": { "terms": { "field":"location.raw",   "size":100 }}}
}' | jq  '.aggregations'