Multi term search

[Mattchewone]

Steps to reproduce

Trying to get results by using multi-$phrase search using this query:

?tags.name[$phrase]=rock&tags.name[$phrase]=pop

I am wanting to find all items which have both rock and pop as tags

Here is the data structure in Elasticsearch:

{
  "title": "Track One",
  "description": "description",
  "slug": "track-one",
  "updatedAt": "2017-04-24T20:29:38.178Z",
  "createdAt": "2017-04-24T20:29:38.178Z",
  "tags": [
	{
	  "_id": "58fe600db684ab640b68eb83",
	  "name": "pop",
	  "parent": "58fe5fbfdf4b3863eae06cd1",
	  "slug": "pop",
	  "updatedAt": "2017-04-24T20:29:01.251Z",
	  "createdAt": "2017-04-24T20:29:01.251Z"
	},
	{
	  "_id": "58fe5fc9df4b3863eae06cd2",
	  "name": "rock",
	  "parent": "58fe5fbfdf4b3863eae06cd1",
          "slug": "rock",
	  "updatedAt": "2017-04-24T20:27:53.873Z",
	  "createdAt": "2017-04-24T20:27:53.873Z"
	},
	{
	  "_id": "58fe5fc9df4b3863eae06cd2",
	  "name": "chill",
	  "parent": "58fe5fbfdf4b3863eae06cd1",
          "slug": "chill",
	  "updatedAt": "2017-04-24T20:27:53.873Z",
	  "createdAt": "2017-04-24T20:27:53.873Z"
	}
  ]
}

Expected behavior

I would expect it to build query and find for both the $phrased terms.

Actual behavior

I get this error [illegal_state_exception] Can't get text on a START_ARRAY at 1:49

System configuration

feathers@2.1.1
feathers-elasticsearch@0.2.2

[daffl]

@jciolek Any thoughts on this?

[jciolek]

Apologies for the delay. Ok, so here's what seems to be happening. First of all the query string:

?tags.name[$phrase]=rock&tags.name[$phrase]=pop

gets parsed by qs to { 'tags.name': { '$phrase': [ 'rock', 'pop' ] } } and rightly so.

Now, $phrase does not deal with arrays in any special way currently - it passes the value it got from the query as is to Elasticsearch. The latter does not expect an array in the context of the match_phrase query, and so it throws an error.

I think it is a very good use case, and I am going to make some subtle changes to support it. I am thinking of adding array support to $prefix, $match, $phrase and $phrase_prefix, as all of them would benefit.

On a separate note, regarding your particular data structure @Mattchewone, I am wondering if you wouldn't benefit from having keyword instead of text mapping for the tag names. You could then use filter terms query $in which is also faster.

[Mattchewone]

@jciolek Thanks very much for looking in to this.

I am new to Elasticsearch and understanding the mappings I am still trying to get to grips with.

Would having keyword mapping work for an array of objects? Would you be kind enough to show a small example to give me something to start with?

Many thanks

[jciolek]

@Mattchewone Not a problem.

So in your case, the query string would be:

?tags.name[$in]=rock&tags.name[$in]=pop

Which translates to an Elasticsearch terms query in filter context:

{
  "query": {
    "bool": {
      "filter": [
        { "terms": { "tags.name": ["rock", "pop"] } }
      ]
    }
  }
}

You can also have a look at the queries fired by https://hacker-search.net/, which uses feathers-elasticsearch. Specifically meta.type is using the $in operator.

I hope that helps.

[jciolek]

Right, so I have read your original issue again and I gather what I just said above is probably not exactly what you are after. I understand you want ALL of the values to be present, not just some. I am working on it.

[daffl]

Closed via #16