[missinglink]

I noticed a bug today where which causes admin_multi_match queries to score results incorrectly for multi-term admin queries.

I'll attach a simple test-case below, prior to this PR we were using the default best_fields type which would essentially ignore the "Eureka" part of the query, meaning that the results were returned in the wrong order:

1) Target, California, MD, USA
2) Target, Eureka, CA, USA

By changing this to cross_fields we are able to use the combined score across all fields, which in-turn scores the query correctly, returning the "Eureka" result first.

{
    "query": {
        "bool": {
            "must": [
                {
                    "match": {
                        "name.default": {
                            "analyzer": "peliasQuery",
                            "query": "Target"
                        }
                    }
                },
                {
                    "multi_match": {
                        "fields": [
                            "parent.country^1",
                            "parent.region^1",
                            "parent.county^1",
                            "parent.localadmin^1",
                            "parent.locality^1",
                            "parent.borough^1",
                            "parent.neighbourhood^1",
                            "parent.region_a^1"
                        ],
                        "type": "best_fields",
                        "query": "Eureka California",
                        "analyzer": "peliasAdmin",
                        "cutoff_frequency": 0.01
                    }
                }
            ]
        }
    },
    "size": 20
}

[missinglink]

This issue is covered by a failing test case which indicates that the text length possibly has something to do with how the best_fields calculation was being made:

search_poi
  ✔ [searchpoi-1] "/v1/search?text=Target Eureka CA"
  ✘ regression [searchpoi-2] "/v1/search?text=Target Eureka California": score 4 out of 5
  diff:
    priorityThresh is 1 but found at position 2

[missinglink]

I added a second commit to disable cutoff_frequency, I would expect this setting to have no effect on the sorting of results, but in fact it does!

Below is a minimal test-case which shows that removing cutoff_frequency changes the order of results.

@orangejulius any idea what's going on here? I read the docs for multi_match cross_fields and cutoff_frequency several times and still can't figure out what's going on..

I'm assuming it's something to do with mixing multi_match and cutoff_frequency.
If we decide that it's not safe to use the two together then we might have to put a check in pelias/query to prevent them from being used together? or maybe it's only triggered (or noticeable) when cross_fields is used?

{
    "query": {
        "bool": {
            "must": [
                {
                    "match": {
                        "name.default": {
                            "analyzer": "peliasQuery",
                            "query": "Target",
                            "cutoff_frequency": 0.01
                        }
                    }
                },
                {
                    "multi_match": {
                        "fields": [
                            "parent.country^1",
                            "parent.region^1",
                            "parent.county^1",
                            "parent.localadmin^1",
                            "parent.locality^1",
                            "parent.borough^1",
                            "parent.neighbourhood^1",
                            "parent.region_a^1"
                        ],
                        "type": "cross_fields",
                        "query": "Eureka California",
                        "analyzer": "peliasAdmin",
                        "cutoff_frequency": 0.01
                    }
                }
            ]
        }
    },
    "size": 20
}

[missinglink]

I tested this PR on our staging environment and it caused 0 regressions and 1 improvement:

✔ improvement [searchpoi-2] "/v1/search?text=Target Eureka California"