Multi-Field Statistical Facets in ElasticSearch

TL;DR - When running multiple statistical facets with the same name, ElasticSearch will catenate the data from all fields in running the results. Using Ruby's Tire wrapper on ElasticSearch, only the last named facet will be sent to ElasticSearch.

The data in these examples are trivial, but do a clear job illustrating this unintended 'feature'.

Example 1

Sales.search do 
  query { all }
  facet 'stats' do
    statistical :amount
  end

  puts to_curl
end

curl -X GET 'http://localhost:9200/sales/sale/_search?size=10&pretty' -d '{"query":{"match_all":{}},"facets":{"stats":{"statistical":{"field":"amount"}}},"size":10}'

@facets={"stats"=>{"_type"=>"statistical",
  "count"=>1476,
  "total"=>865389.1399999999, 
  "min"=>1.0, "max"=>9289.0, 
  "mean"=>586.3070054200541, 
  "sum_of_squares"=>1252317429.8326, 
  "variance"=>504697.63864238764, 
  "std_deviation"=>710.4207476153745}}

Example 2

Sales.search do 
  query { all }
  facet 'stats' do
    statistical :id
  end

  puts to_curl
end

curl -X GET 'http://localhost:9200/sales/sale/_search?size=10&pretty' -d '{"query":{"match_all":{}},"facets":{"stats":{"statistical":{"field":"id"}}},"size":10}'

@facets={"stats"=>{"_type"=>"statistical", 
  "count"=>20, 
  "total"=>34.0, 
  "min"=>1.0, 
  "max"=>4.0, 
  "mean"=>1.7, 
  "sum_of_squares"=>78.0, 
  "variance"=>1.0100000000000002, 
  "std_deviation"=>1.0049875621120892}}

Example 3

Sales.search do 
  query { all }
  facet 'stats' do
    statistical :id
  end
  facet 'stats' do
    statistical :amount
  end

  puts to_curl
end

curl -X GET 'http://localhost:9200/sales/sale/_search?size=10&pretty' -d '{"query":{"match_all":{}},"facets":{"stats":{"statistical":{"field":"amount"}}},"size":10}'

@facets={"stats"=>{"_type"=>"statistical",
  "count"=>1476, 
  "total"=>865389.1399999999, 
  "min"=>1.0,
  "max"=>9289.0,
  "mean"=>586.3070054200541,
  "sum_of_squares"=>1252317429.8326,
  "variance"=>504697.63864238764,
  "std_deviation"=>710.4207476153745}}

Notice above that Tire ignored the first facet and only sent the second.

Example 4

Handwritten curl request, no Tire wrapper

curl -X GET 'http://localhost:9200/sales/sale/_search?size=10&pretty' -d '{"query":{"match_all":{}},"facets":{"stats":{"statistical":{"field":"amount"}},"stats":{"statistical":{"field":"id"}}},"size":10}'

"facets" : {
  "stats" : {
    "_type" : "statistical",
    "count" : 1496,
    "total" : 865423.1399999999,
    "min" : 1.0,
    "max" : 9289.0,
    "mean" : 578.4914037433155,
    "sum_of_squares" : 1.2523175078326E9,
    "variance" : 502458.32937302964,
    "std_deviation" : 708.8429511344735
  },
  "stats" : {
    "_type" : "statistical",
    "count" : 1496,
    "total" : 865423.1399999999,
    "min" : 1.0,
    "max" : 9289.0,
    "mean" : 578.4914037433155,
    "sum_of_squares" : 1.2523175078326E9,
    "variance" : 502458.32937302964,
    "std_deviation" : 708.8429511344735
  }
}

Here, both facets are registered, and both contain information from both fields.

see: http://elasticsearch-users.115913.n3.nabble.com/Statistical-facet-on-multiple-fields-td1676993.html for a discussion on the issue.
also see: http://www.elasticsearch.org/guide/reference/api/search/facets/statistical-facet/ for the proper way to intentionally run a multi-field facet.