Transform using geotile_grid pivot with missing_bucket=true triggers NullPointerException

[elemask]

Elasticsearch Version

8.4.3

Installed Plugins

No response

Java Version

bundled

OS Version

Ubuntu 22.04

Problem Description

Using a transform which pivots on a geotile_grid will trigger a NullPointerException when setting the missing_buckets option to true.

The same composite aggregation works fine and correctly generates a bucket whose key contains a null value (as the value to aggregate on is missing). I believe it is a bug in the transform code when manipulating the composite bucket containing a key with a null value.

Steps to Reproduce

1. Create index with 3 documents:

PUT /test
{
  "mappings": {
    "properties": {
      "username": {
        "type": "keyword"
      },
      "date": {
        "type": "date"
      },
      "location": {
        "type": "geo_point"
      }
    }
  }
}

PUT /test/_doc/1
{
  "username": "bob",
  "date": "2025-12-25",
  "location": "50.62096405029297,5.5580315589904785"
}

PUT /test/_doc/2
{
  "username": "bob",
  "date": "2024-12-25",
  "location": "50.620975494384766,5.558044910430908"
}

PUT /test/_doc/3
{
  "username": "bob",
  "date": "2023-12-25"
}

2. Test composite aggregation on username + location fields:

GET /test/_search?filter_path=aggregations.pivot.buckets
{
  "query": {
    "match_all": {}
  },
  "size": 0, 
  "aggs": {
    "pivot": {
      "composite": {
        "sources": [
          {
            "user": {
              "terms": {
                "field": "username"
              }
            }
          },
          {
            "pings": {
              "geotile_grid": {
                "field": "location",
                "precision": 1,
                "missing_bucket": true
              }
            }
          }
        ]
      },
      "aggs": {
        "seen_first": {
          "min": {
            "field": "date"
          }
        }
      }
    }
  }
}

Result is as excepted. 2 buckets, one matching the 2 documents with geo_points and one matching the document with no geo_points.

{
  "aggregations": {
    "pivot": {
      "buckets": [
        {
          "key": {
            "user": "bob",
            "pings": null
          },
          "doc_count": 1,
          "seen_first": {
            "value": 1703462400000,
            "value_as_string": "2023-12-25T00:00:00.000Z"
          }
        },
        {
          "key": {
            "user": "bob",
            "pings": "1/1/0"
          },
          "doc_count": 2,
          "seen_first": {
            "value": 1735084800000,
            "value_as_string": "2024-12-25T00:00:00.000Z"
          }
        }
      ]
    }
  }
}

3. Run a transform preview with the equivalent aggregations:

GET /_transform/_preview
{
  "source": {
    "index": "test"
  },
  "pivot": {
    "group_by": {
      "user": {
        "terms": {
          "field": "username"
        }
      },
      "pings": {
        "geotile_grid": {
          "field": "location",
          "precision": 1,
          "missing_bucket": true
        }
      }
    },
    "aggregations": {
      "seen_first": {
        "min": {
          "field": "date"
        }
      }
    }
  }
}

Result in HTTP 500 error

{
  "error": {
    "root_cause": [
      {
        "type": "null_pointer_exception",
        "reason": """Cannot invoke "Object.toString()" because "key" is null"""
      }
    ],
    "type": "null_pointer_exception",
    "reason": """Cannot invoke "Object.toString()" because "key" is null"""
  },
  "status": 500
}

Expected results: 2 buckets

Logs (if relevant)

Server logs contain the following backtrace:

java.lang.NullPointerException: Cannot invoke "Object.toString()" because "key" is null
    at [email protected]/org.elasticsearch.xpack.transform.transforms.pivot.AggregationResultUtils$GeoTileBucketKeyExtractor.value(AggregationResultUtils.java:486)
    at [email protected]/org.elasticsearch.xpack.transform.transforms.pivot.AggregationResultUtils.lambda$extractCompositeAggregationResults$0(AggregationResultUtils.java:123)
    at java.base/java.util.LinkedHashMap.forEach(LinkedHashMap.java:986)
    at [email protected]/org.elasticsearch.xpack.transform.transforms.pivot.AggregationResultUtils.lambda$extractCompositeAggregationResults$2(AggregationResultUtils.java:116)
[...]

[elasticsearchmachine]

Pinging @elastic/ml-core (Team:ML)