Performance for Filter Context containing Query Context

[Sam_Sudekum]

https://www.elastic.co/guide/en/elasticsearch/reference/2.4/query-filter-context.html

I'm currently running ES 2.4 working on moving to 5.6. One paradigm shift that needs to happen is the use of bool.query/bool.filter.

There are performance incentives to use filters (non-scoring, and query caching). I want to take advantage of that behavior wherever I can.

Consider the following query:

	"query": {
	  	"bool": {
	  		"filter": {
	  			"bool": {
	  				"should": {
			  				"term": {
								"first_name": "bob"
							}
			  		},
                    "must": {
			  			"term": {
							"last_name": "smith"
						}
			  		}
	  			}
	  		}
	  	}
	}

Regardless of whether this is an optimal query, my question is about behavior. Normally, query.bool.must/should are scoring queries. When I run the query above, the score is normalized to 1 because it is wrapped in a filter. However, I'm curious if the underlying bool.should/must clause is still performing the scoring operations or if I've actually avoided that behavior.

[polyfractal (Zachary Tong)]

All clauses that are under the filter section inherit the "filter context", so scoring won't be executed on all those sub-clauses.

That does cause some behavioral changes though. If you embed a bool that has both a should and a must clause inside the filter of another bool, the should clause because useless. E.g. shoulds are optional and only affect scoring, so the inclusion of a must clause means the should is completely useless... it doesn't affect scoring and doesn't affect inclusion/exclusion.

I personally like to convert things to filter/must_not when inside a filter context just to keep semantics clear to a human, but it's not strictly necessary