Accuracy

Numeric, date and geospatial fields index extra terms at coarser granularities, called accuracies. A range query can then match a handful of coarse “bucket” terms instead of scanning every distinct value, which makes ranges and range based aggregations much faster. The trade-off is a larger index: each accuracy level adds terms.

Accuracies are set per field with the _accuracy option. If it is omitted a sensible default is used, so you only need to tune it when you know your query patterns.

Numeric Accuracy

For numeric fields, _accuracy is an array of step sizes. The field below buckets its values at multiples of 100, 1000 and 10000:

PUT /test_accuracy/

{
  "_schema": {
    "balance": {
      "_type": "float",
      "_accuracy": [100, 1000, 10000]
    }
  }
}

PUT /test_accuracy/1

{
  "balance": 1234.56
}

Ranges over the field use those buckets transparently, the results are exactly the same as without accuracies, only faster:

SEARCH /test_accuracy/

{
  "_query": {
    "balance": {
      "_in": {
        "_range": { "_from": 1000, "_to": 2000 }
      }
    }
  }
}

When _accuracy is omitted, numeric fields default to [100, 1000, 10000, 100000, 1000000, 100000000].

Date Accuracy

For date fields, _accuracy is an array of named levels instead of step sizes: "second", "minute", "hour", "day", "month", "year", "decade", "century" and "millennium".

See Date Type and Numeric Type for the per-type details.