Top metrics aggregation

edit

The top_metrics aggregation selects metrics from the document with the largest or smallest "sort" value. For example, this gets the value of the m field on the document with the largest value of s:

response = client.bulk(
  index: 'test',
  refresh: true,
  body: [
    {
      index: {}
    },
    {
      s: 1,
      m: 3.1415
    },
    {
      index: {}
    },
    {
      s: 2,
      m: 1
    },
    {
      index: {}
    },
    {
      s: 3,
      m: 2.71828
    }
  ]
)
puts response

response = client.search(
  index: 'test',
  filter_path: 'aggregations',
  body: {
    aggregations: {
      tm: {
        top_metrics: {
          metrics: {
            field: 'm'
          },
          sort: {
            s: 'desc'
          }
        }
      }
    }
  }
)
puts response
POST /test/_bulk?refresh
{"index": {}}
{"s": 1, "m": 3.1415}
{"index": {}}
{"s": 2, "m": 1.0}
{"index": {}}
{"s": 3, "m": 2.71828}
POST /test/_search?filter_path=aggregations
{
  "aggs": {
    "tm": {
      "top_metrics": {
        "metrics": {"field": "m"},
        "sort": {"s": "desc"}
      }
    }
  }
}

Which returns:

{
  "aggregations": {
    "tm": {
      "top": [ {"sort": [3], "metrics": {"m": 2.718280076980591 } } ]
    }
  }
}

top_metrics is fairly similar to top_hits in spirit but because it is more limited it is able to do its job using less memory and is often faster.

sort

edit

The sort field in the metric request functions exactly the same as the sort field in the search request except:

  • It can’t be used on binary, flattened, ip, keyword, or text fields.
  • It only supports a single sort value so which document wins ties is not specified.

The metrics that the aggregation returns is the first hit that would be returned by the search request. So,

"sort": {"s": "desc"}
gets metrics from the document with the highest s
"sort": {"s": "asc"}
gets the metrics from the document with the lowest s
"sort": {"_geo_distance": {"location": "POINT (-78.6382 35.7796)"}}
gets metrics from the documents with location closest to 35.7796, -78.6382
"sort": "_score"
gets metrics from the document with the highest score

metrics

edit

metrics selects the fields of the "top" document to return. You can request a single metric with something like "metrics": {"field": "m"} or multiple metrics by requesting a list of metrics like "metrics": [{"field": "m"}, {"field": "i"}.

metrics.field supports the following field types:

Except for keywords, runtime fields for corresponding types are also supported. metrics.field doesn’t support fields with array values. A top_metric aggregation on array values may return inconsistent results.

The following example runs a top_metrics aggregation on several field types.

response = client.indices.create(
  index: 'test',
  body: {
    mappings: {
      properties: {
        d: {
          type: 'date'
        }
      }
    }
  }
)
puts response

response = client.bulk(
  index: 'test',
  refresh: true,
  body: [
    {
      index: {}
    },
    {
      s: 1,
      m: 3.1415,
      i: 1,
      d: '2020-01-01T00:12:12Z',
      t: 'cat'
    },
    {
      index: {}
    },
    {
      s: 2,
      m: 1,
      i: 6,
      d: '2020-01-02T00:12:12Z',
      t: 'dog'
    },
    {
      index: {}
    },
    {
      s: 3,
      m: 2.71828,
      i: -12,
      d: '2019-12-31T00:12:12Z',
      t: 'chicken'
    }
  ]
)
puts response

response = client.search(
  index: 'test',
  filter_path: 'aggregations',
  body: {
    aggregations: {
      tm: {
        top_metrics: {
          metrics: [
            {
              field: 'm'
            },
            {
              field: 'i'
            },
            {
              field: 'd'
            },
            {
              field: 't.keyword'
            }
          ],
          sort: {
            s: 'desc'
          }
        }
      }
    }
  }
)
puts response
PUT /test
{
  "mappings": {
    "properties": {
      "d": {"type": "date"}
    }
  }
}
POST /test/_bulk?refresh
{"index": {}}
{"s": 1, "m": 3.1415, "i": 1, "d": "2020-01-01T00:12:12Z", "t": "cat"}
{"index": {}}
{"s": 2, "m": 1.0, "i": 6, "d": "2020-01-02T00:12:12Z", "t": "dog"}
{"index": {}}
{"s": 3, "m": 2.71828, "i": -12, "d": "2019-12-31T00:12:12Z", "t": "chicken"}
POST /test/_search?filter_path=aggregations
{
  "aggs": {
    "tm": {
      "top_metrics": {
        "metrics": [
          {"field": "m"},
          {"field": "i"},
          {"field": "d"},
          {"field": "t.keyword"}
        ],
        "sort": {"s": "desc"}
      }
    }
  }
}

Which returns:

{
  "aggregations": {
    "tm": {
      "top": [ {
        "sort": [3],
        "metrics": {
          "m": 2.718280076980591,
          "i": -12,
          "d": "2019-12-31T00:12:12.000Z",
          "t.keyword": "chicken"
        }
      } ]
    }
  }
}

missing

edit

The missing parameter defines how documents with a missing value are treated. By default, if any of the key components are missing, the entire document is ignored. It is possible to treat the missing components as if they had a value by using the missing parameter.

PUT /my-index
{
  "mappings": {
    "properties": {
      "nr":    { "type": "integer" },
      "state":  { "type": "keyword"  } 
    }
  }
}
POST /my-index/_bulk?refresh
{"index": {}}
{"nr": 1, "state": "started"}
{"index": {}}
{"nr": 2, "state": "stopped"}
{"index": {}}
{"nr": 3, "state": "N/A"}
{"index": {}}
{"nr": 4} 
POST /my-index/_search?filter_path=aggregations
{
  "aggs": {
    "my_top_metrics": {
      "top_metrics": {
        "metrics": {
          "field": "state",
          "missing": "N/A"}, 
        "sort": {"nr": "desc"}
      }
    }
  }
}

If you want to use an aggregation on textual content, it must be a keyword type field or you must enable fielddata on that field.

This document has a missing state field value.

The missing parameter defines that if state field has a missing value, it should be treated as if it had the N/A value.

The request results in the following response:

{
  "aggregations": {
    "my_top_metrics": {
      "top": [
        {
          "sort": [
            4
          ],
          "metrics": {
            "state": "N/A"
          }
        }
      ]
    }
  }
}

size

edit

top_metrics can return the top few document’s worth of metrics using the size parameter:

response = client.bulk(
  index: 'test',
  refresh: true,
  body: [
    {
      index: {}
    },
    {
      s: 1,
      m: 3.1415
    },
    {
      index: {}
    },
    {
      s: 2,
      m: 1
    },
    {
      index: {}
    },
    {
      s: 3,
      m: 2.71828
    }
  ]
)
puts response

response = client.search(
  index: 'test',
  filter_path: 'aggregations',
  body: {
    aggregations: {
      tm: {
        top_metrics: {
          metrics: {
            field: 'm'
          },
          sort: {
            s: 'desc'
          },
          size: 3
        }
      }
    }
  }
)
puts response
POST /test/_bulk?refresh
{"index": {}}
{"s": 1, "m": 3.1415}
{"index": {}}
{"s": 2, "m": 1.0}
{"index": {}}
{"s": 3, "m": 2.71828}
POST /test/_search?filter_path=aggregations
{
  "aggs": {
    "tm": {
      "top_metrics": {
        "metrics": {"field": "m"},
        "sort": {"s": "desc"},
        "size": 3
      }
    }
  }
}

Which returns:

{
  "aggregations": {
    "tm": {
      "top": [
        {"sort": [3], "metrics": {"m": 2.718280076980591 } },
        {"sort": [2], "metrics": {"m": 1.0 } },
        {"sort": [1], "metrics": {"m": 3.1414999961853027 } }
      ]
    }
  }
}

The default size is 1. The maximum default size is 10 because the aggregation’s working storage is "dense", meaning we allocate size slots for every bucket. 10 is a very conservative default maximum and you can raise it if you need to by changing the top_metrics_max_size index setting. But know that large sizes can take a fair bit of memory, especially if they are inside of an aggregation which makes many buckes like a large terms aggregation. If you till want to raise it, use something like:

response = client.indices.put_settings(
  index: 'test',
  body: {
    top_metrics_max_size: 100
  }
)
puts response
PUT /test/_settings
{
  "top_metrics_max_size": 100
}

If size is more than 1 the top_metrics aggregation can’t be the target of a sort.

Examples

edit

Use with terms

edit

This aggregation should be quite useful inside of terms aggregation, to, say, find the last value reported by each server.

response = client.indices.create(
  index: 'node',
  body: {
    mappings: {
      properties: {
        ip: {
          type: 'ip'
        },
        date: {
          type: 'date'
        }
      }
    }
  }
)
puts response

response = client.bulk(
  index: 'node',
  refresh: true,
  body: [
    {
      index: {}
    },
    {
      ip: '192.168.0.1',
      date: '2020-01-01T01:01:01',
      m: 1
    },
    {
      index: {}
    },
    {
      ip: '192.168.0.1',
      date: '2020-01-01T02:01:01',
      m: 2
    },
    {
      index: {}
    },
    {
      ip: '192.168.0.2',
      date: '2020-01-01T02:01:01',
      m: 3
    }
  ]
)
puts response

response = client.search(
  index: 'node',
  filter_path: 'aggregations',
  body: {
    aggregations: {
      ip: {
        terms: {
          field: 'ip'
        },
        aggregations: {
          tm: {
            top_metrics: {
              metrics: {
                field: 'm'
              },
              sort: {
                date: 'desc'
              }
            }
          }
        }
      }
    }
  }
)
puts response
PUT /node
{
  "mappings": {
    "properties": {
      "ip": {"type": "ip"},
      "date": {"type": "date"}
    }
  }
}
POST /node/_bulk?refresh
{"index": {}}
{"ip": "192.168.0.1", "date": "2020-01-01T01:01:01", "m": 1}
{"index": {}}
{"ip": "192.168.0.1", "date": "2020-01-01T02:01:01", "m": 2}
{"index": {}}
{"ip": "192.168.0.2", "date": "2020-01-01T02:01:01", "m": 3}
POST /node/_search?filter_path=aggregations
{
  "aggs": {
    "ip": {
      "terms": {
        "field": "ip"
      },
      "aggs": {
        "tm": {
          "top_metrics": {
            "metrics": {"field": "m"},
            "sort": {"date": "desc"}
          }
        }
      }
    }
  }
}

Which returns:

{
  "aggregations": {
    "ip": {
      "buckets": [
        {
          "key": "192.168.0.1",
          "doc_count": 2,
          "tm": {
            "top": [ {"sort": ["2020-01-01T02:01:01.000Z"], "metrics": {"m": 2 } } ]
          }
        },
        {
          "key": "192.168.0.2",
          "doc_count": 1,
          "tm": {
            "top": [ {"sort": ["2020-01-01T02:01:01.000Z"], "metrics": {"m": 3 } } ]
          }
        }
      ],
      "doc_count_error_upper_bound": 0,
      "sum_other_doc_count": 0
    }
  }
}

Unlike top_hits, you can sort buckets by the results of this metric:

response = client.search(
  index: 'node',
  filter_path: 'aggregations',
  body: {
    aggregations: {
      ip: {
        terms: {
          field: 'ip',
          order: {
            "tm.m": 'desc'
          }
        },
        aggregations: {
          tm: {
            top_metrics: {
              metrics: {
                field: 'm'
              },
              sort: {
                date: 'desc'
              }
            }
          }
        }
      }
    }
  }
)
puts response
POST /node/_search?filter_path=aggregations
{
  "aggs": {
    "ip": {
      "terms": {
        "field": "ip",
        "order": {"tm.m": "desc"}
      },
      "aggs": {
        "tm": {
          "top_metrics": {
            "metrics": {"field": "m"},
            "sort": {"date": "desc"}
          }
        }
      }
    }
  }
}

Which returns:

{
  "aggregations": {
    "ip": {
      "buckets": [
        {
          "key": "192.168.0.2",
          "doc_count": 1,
          "tm": {
            "top": [ {"sort": ["2020-01-01T02:01:01.000Z"], "metrics": {"m": 3 } } ]
          }
        },
        {
          "key": "192.168.0.1",
          "doc_count": 2,
          "tm": {
            "top": [ {"sort": ["2020-01-01T02:01:01.000Z"], "metrics": {"m": 2 } } ]
          }
        }
      ],
      "doc_count_error_upper_bound": 0,
      "sum_other_doc_count": 0
    }
  }
}

Mixed sort types

edit

Sorting top_metrics by a field that has different types across different indices producs somewhat surprising results: floating point fields are always sorted independently of whole numbered fields.

response = client.bulk(
  index: 'test',
  refresh: true,
  body: [
    {
      index: {
        _index: 'test1'
      }
    },
    {
      s: 1,
      m: 3.1415
    },
    {
      index: {
        _index: 'test1'
      }
    },
    {
      s: 2,
      m: 1
    },
    {
      index: {
        _index: 'test2'
      }
    },
    {
      s: 3.1,
      m: 2.71828
    }
  ]
)
puts response

response = client.search(
  index: 'test*',
  filter_path: 'aggregations',
  body: {
    aggregations: {
      tm: {
        top_metrics: {
          metrics: {
            field: 'm'
          },
          sort: {
            s: 'asc'
          }
        }
      }
    }
  }
)
puts response
POST /test/_bulk?refresh
{"index": {"_index": "test1"}}
{"s": 1, "m": 3.1415}
{"index": {"_index": "test1"}}
{"s": 2, "m": 1}
{"index": {"_index": "test2"}}
{"s": 3.1, "m": 2.71828}
POST /test*/_search?filter_path=aggregations
{
  "aggs": {
    "tm": {
      "top_metrics": {
        "metrics": {"field": "m"},
        "sort": {"s": "asc"}
      }
    }
  }
}

Which returns:

{
  "aggregations": {
    "tm": {
      "top": [ {"sort": [3.0999999046325684], "metrics": {"m": 2.718280076980591 } } ]
    }
  }
}

While this is better than an error it probably isn’t what you were going for. While it does lose some precision, you can explicitly cast the whole number fields to floating points with something like:

response = client.search(
  index: 'test*',
  filter_path: 'aggregations',
  body: {
    aggregations: {
      tm: {
        top_metrics: {
          metrics: {
            field: 'm'
          },
          sort: {
            s: {
              order: 'asc',
              numeric_type: 'double'
            }
          }
        }
      }
    }
  }
)
puts response
POST /test*/_search?filter_path=aggregations
{
  "aggs": {
    "tm": {
      "top_metrics": {
        "metrics": {"field": "m"},
        "sort": {"s": {"order": "asc", "numeric_type": "double"}}
      }
    }
  }
}

Which returns the much more expected:

{
  "aggregations": {
    "tm": {
      "top": [ {"sort": [1.0], "metrics": {"m": 3.1414999961853027 } } ]
    }
  }
}

Use in pipeline aggregations

edit

top_metrics can be used in pipeline aggregations that consume a single value per bucket, such as bucket_selector that applies per bucket filtering, similar to using a HAVING clause in SQL. This requires setting size to 1, and specifying the right path for the (single) metric to be passed to the wrapping aggregator. For example:

POST /test*/_search?filter_path=aggregations
{
  "aggs": {
    "ip": {
      "terms": {
        "field": "ip"
      },
      "aggs": {
        "tm": {
          "top_metrics": {
            "metrics": {"field": "m"},
            "sort": {"s": "desc"},
            "size": 1
          }
        },
        "having_tm": {
          "bucket_selector": {
            "buckets_path": {
              "top_m": "tm[m]"
            },
            "script": "params.top_m < 1000"
          }
        }
      }
    }
  }
}

The bucket_path uses the top_metrics name tm and a keyword for the metric providing the aggregate value, namely m.