Field data formats

The field data format controls how field data should be stored.

Depending on the field type, there might be several field data types available. In particular, string and numeric types support the doc_values format which allows for computing the field data data-structures at indexing time and storing them on disk. Although it will make the index larger and may be slightly slower, this implementation will be more near-realtime-friendly and will require much less memory from the JVM than other implementations.

Here is an example of how to configure the tag field to use the fst field data format.

{
    "tag": {
        "type":      "string",
        "fielddata": {
            "format": "fst"
        }
    }
}

It is possible to change the field data format (and the field data settings in general) on a live index by using the update mapping API. When doing so, field data which had already been loaded for existing segments will remain alive while new segments will use the new field data configuration. Thanks to the background merging process, all segments will eventually use the new field data format.

Global ordinals is a data-structure on top of field data, that maintains an incremental numbering for all the terms in field data in a lexicographic order. Each term has a unique number and the number of term A is lower than the number of term B. Global ordinals are only supported on string fields.

Field data on string also has ordinals, which is a unique numbering for all terms in a particular segment and field. Global ordinals just build on top of this, by providing a mapping between the segment ordinals and the global ordinals. The latter being unique across the entire shard.

Global ordinals can be beneficial in search features that use segment ordinals already such as the terms aggregator to improve the execution time. Often these search features need to merge the segment ordinal results to a cross segment terms result. With global ordinals this mapping happens during field data load time instead of during each query execution. With global ordinals search features only need to resolve the actual term when building the (shard) response, but during the execution there is no need at all to use the actual terms and the unique numbering global ordinals provided is sufficient and improves the execution time.

Global ordinals for a specified field are tied to all the segments of a shard (Lucene index), which is different than for field data for a specific field which is tied to a single segment. For this reason global ordinals need to be rebuilt in its entirety once new segments become visible. This one time cost would happen anyway without global ordinals, but then it would happen for each search execution instead!

The loading time of global ordinals depends on the number of terms in a field, but in general it is low, since it source field data has already been loaded. The memory overhead of global ordinals is a small because it is very efficiently compressed. Eager loading of global ordinals can move the loading time from the first search request, to the refresh itself.

By default, field data is loaded lazily, ie. the first time that a query that requires them is executed. However, this can make the first requests that follow a merge operation quite slow since fielddata loading is a heavy operation.

It is possible to force field data to be loaded and cached eagerly through the loading setting of fielddata:

{
    "category": {
        "type":      "string",
        "fielddata": {
            "loading": "eager"
        }
    }
}

Global ordinals can also be eagerly loaded:

{
    "category": {
        "type":      "string",
        "fielddata": {
            "loading": "eager_global_ordinals"
        }
    }
}

With the above setting both field data and global ordinals for a specific field are eagerly loaded.

Field data can take a lot of RAM so it makes sense to disable field data loading on the fields that don’t need field data, for example those that are used for full-text search only. In order to disable field data loading, just change the field data format to disabled. When disabled, all requests that will try to load field data, e.g. when they include aggregations and/or sorting, will return an error.

{
    "text": {
        "type":      "string",
        "fielddata": {
            "format": "disabled"
        }
    }
}

The disabled format is supported by all field types.

It is possible to control which field values are loaded into memory, which is particularly useful for string fields. When specifying the mapping for a field, you can also specify a fielddata filter.

Fielddata filters can be changed using the PUT mapping API. After changing the filters, use the Clear Cache API to reload the fielddata using the new filters.

The frequency filter allows you to only load terms whose frequency falls between a min and max value, which can be expressed an absolute number or as a percentage (eg 0.01 is 1%). Frequency is calculated per segment. Percentages are based on the number of docs which have a value for the field, as opposed to all docs in the segment.

Small segments can be excluded completely by specifying the minimum number of docs that the segment should contain with min_segment_size:

{
    "tag": {
        "type":      "string",
        "fielddata": {
            "filter": {
                "frequency": {
                    "min":              0.001,
                    "max":              0.1,
                    "min_segment_size": 500
                }
            }
        }
    }
}

Terms can also be filtered by regular expression - only values which match the regular expression are loaded. Note: the regular expression is applied to each term in the field, not to the whole field value. For instance, to only load hashtags from a tweet, we can use a regular expression which matches terms beginning with #:

{
    "tweet": {
        "type":      "string",
        "analyzer":  "whitespace"
        "fielddata": {
            "filter": {
                "regex": {
                    "pattern": "^#.*"
                }
            }
        }
    }
}

The frequency and regex filters can be combined:

{
    "tweet": {
        "type":      "string",
        "analyzer":  "whitespace"
        "fielddata": {
            "filter": {
                "regex": {
                    "pattern":          "^#.*",
                },
                "frequency": {
                    "min":              0.001,
                    "max":              0.1,
                    "min_segment_size": 500
                }
            }
        }
    }
}