ORing a text field with a unique identifier keyword field leading to increased next_doc count and poor performance

[helderdias]

TL;DR: I want to find documents that match a certain query (e.g. "my search") OR match the unique ID of a document.

When I search for the text field alone, the search is super fast. However, when I or the text field with the keyword field search becomes extremely slow. Looking at the profile data, it seems like the shard that contains the document ends up doing a lot more next_doc than the other shards. Why would that be the case?

Below are the mappings and queries I executed for reference.

Mapping:

{
  "mappings": {
    "properties": {
      "searchable_name_1":    { "type": "text" },  
      "searchable_name_2":  { "type": "text"  }, 
      "unique_identifier":   { "type": "keyword"  }     
    }
  }
}

The following query executes super fast:

{
  "profile": true,
  "query": {
    "bool": {
      "should": [
        {
          "multi_match": {
            "query": "my search",
            "fields": [
              "searchable_name_1",
              "searchable_name_2"
            ]
          }
        }
      ]
    }
  }
}

However, when I or the unique identifier with the text query, it becomes super slow:

{
  "profile": true,
  "query": {
    "bool": {
      "should": [
        {
          "multi_match": {
            "query": "my search",
            "fields": [
              "searchable_name_1",
              "searchable_name_2"
            ]
          }
        },
        {
          "term": {
            "unique_identifier": {
              "value": "13590585",
              "boost": 10
            }
          }
        }
      ]
    }
  }
}

[Mark_Harwood1 (Mark Harwood)]

Hi
Sanity check - the mapped field is called “unique_identifier” and the query is “unique_id” in your example

[helderdias]

Thanks for your reply. It was a copy-paste mistake while anonymizing the use case. Fixed in the original post.

[Mark_Harwood1 (Mark Harwood)]

Thanks. How many docs have this unique ID of 13590585?

[helderdias]

Just one. That's why I'm confused why ORing the two queries would result in such latency since the posting list for that ID has only one doc.

[Mark_Harwood1 (Mark Harwood)]

Hmm. Then I expect that the addition of a clause is causing a change to scoring that means an optimisation is not kicking in.

If you set “track_total_hits” to true in both queries then this might help compare speeds with the shortcut optimisation disabled.
(It won’t solve your problem but would help with the diagnostics here)

[helderdias]

I enabled track_total_hits on both queries.

Fast one (just match query): 5323035
Slow one (match query OR keyword term): 5323036

[Mark_Harwood1 (Mark Harwood)]

I presume that’s the number of hits. How do the response times compare?

[helderdias]

That's correct.

Shard response time for match only:

Shard response time for match OR keyword term:

By the way, thanks in advance for your help, Mark!