CODEWITHSUNDEEP
elasticsearchelasticsearch-query
RPM1984
RPM1984

Reputation: 73123

How to do a case-insensitive search on a keyword field in Elasticsearch without reindexing?

I have a keyword field like this:

"address": {
   "type": "keyword"
}

The value is written in camel-case, as it's intended for display. e.g "1/10 Somewhere Rd, Somewhere AAA 3333"

Now, i want to do a case-insensitive search on that field, ideally without reindexing the entire index.

What i've found so far is that match can be used for case-insensitive search, but only works on text fields.

Is my only option reindexing, e.g via a text field, or using a lowercase/custom normalizer/analyser? Any way i can do it without reindexing?

Upvotes: 11

Views: 18018

Answers (2)

kym
kym

Reputation: 1203

As of version 7.10, the Term query supports a case_insensitive parameter

GET /_search
{
  "query": {
    "term": {
      "user.id": {
        "value": "kimchy",
        "case_insensitive": true
      }
    }
  }
}

However, as of v8.10 the Terms query does not support case_insensitive.

The best approach I've found is this that uses a normalizer which lowercases the string at index time. That way you can use any query that supports normalization and get case insensitive matches.

Upvotes: 2

Amit
Amit

Reputation: 32386

You are right that match queries can be used for case-insensitive search as it applied the same analyzer which was used at index time but works only for text field.

Problem here is that while indexing, as you have used keyword field, so tokens in elasticsearch inverted index, which is used to match the tokens of search query are not lowercased so it is not possible at all to provide case-insensitive search.

Let's understand above statement using an example:

Let's suppose you have Foo BAR in your document, which you indexed using keyword field. Please note the case of each character, so inverted index will have below token.

Foo BAR, now at query time by some hook or crook, you can convert search term to either all uppercase or lowercase but in this case still it won't match the tokens, so you will still have a lot of issues in search results.

I would suggest, add a new field which uses text and using reindex API create a fresh index and implement it in clean manner, also using reindex API, you can build a new index from old index, and its much faster to build a new index, rather than build from source of truth(SQL in most cases).

Upvotes: 14

Related Questions