_exists_ doesn't work on objects

[bobrik]

At least in 1.0.0

# create index and object
curl -X PUT 'http://127.0.0.1:9200/wtf/?pretty'
curl -X PUT 'http://127.0.0.1:9200/wtf/test/1?pretty' -d '{"complex": { "object": "it is" }, "simple": "value"}'

# no results
curl -X POST 'http://127.0.0.1:9200/wtf/test/_search?pretty&q=_exists_:complex'
# has results
curl -X POST 'http://127.0.0.1:9200/wtf/test/_search?pretty&q=_exists_:simple'

# has results
curl -X POST 'http://127.0.0.1:9200/wtf/test/_search?pretty' -d '{ "query": { "filtered": { "filter": { "not": { "missing": { "field": "complex" } } } } } }'

# has results
curl -X POST 'http://127.0.0.1:9200/wtf/test/_search?pretty' -d '{ "query": { "filtered": { "filter": { "not": { "missing": { "field": "simple" } } } } } }'

This behaviour should be either fixed or noted in docs.

[jpountz]

The exists filter works by looking at matches in the inverted index for any value of the given field. The reason why it doesn't work on object is that they are not indexed. In the example that you gave, the inverted index only has entries for complex.object and simple.

I will a a note to the documentation.

[bobrik]

So missing filter will use field data and not inverted index, right?

[jpountz]

The missing filter works exactly as if you wrapped the exists filter under a not filter, so the same limitations apply.

[bobrik]

Since this works:

curl -X POST 'http://127.0.0.1:9200/wtf/test/_search?pretty' -d '{ "query": { "filtered": { "filter": { "exists": { "field": "complex" } } } } }'

and this doesn't:

curl -X POST 'http://127.0.0.1:9200/wtf/test/_search?pretty&q=_exists_:complex'

I understand that _exists_ in query only uses inverted index, but exists filter knows that complex is object and cannot be found in index so field data should be used. Do I get this right?

[jpountz]

I had to check the code to understand what happens, the difference is that exists tries to be smarter than _exists_ when matching an object by using sub-fields. That is, exists is able to translate "exists": { "field": "complex"} to "exists": { "field": "complex.object"}. Let me see if I can fix _exists_ to behave consistently with exists.