Warning This document has not been updated for a while now. It may be out of date.

Last updated: 9 Jun 2020

search-api: Schema definitions

The config/schema tree holds the schema definitions for our search indexes. The purpose of the schema is to define:

which fields are allowed to be present in documents passed to the search indexer
how these fields should be processed at indexing time
how these fields should be searched at query time

The schema is used to produce the elasticsearch configuration for each index, but is also used to control how the fields are processed before passing them to elasticsearch, and after retrieving results from elasticsearch.

There are several "layers" of the schema; field types, which are used to define the configuration for a named set of fields, which are then used to define a set of document types, which are then used to make up the configuration for an index.

Field types

The field_types.json file contains a JSON object describing some "high-level" named types which can be applied to fields in documents. Each type is keyed by the name of the type, and has the following properties:

description: a human-readable description of what the type is to be used for (this is essentially a comment field).
es_config: the elasticsearch configuration to be applied to fields using this type.
multivalued: (optional, default of false) - if true, the field is allowed to contain multiple values, and will be represented as an array when documents are being returned.
children: (optional) - if present, the field contains child fields (ie, the field values will be JSON objects containing the child fields). This must be set to one of two values:
- named: The definition of the field (see next section) must contain a children property, defining the allowed child fields.
- dynamic: Arbitrary child fields will be accepted, but how they are handled will be configured dynamically according to the elasticsearch configuration.

Field definitions

To keep things simple we require that for a field of a given name, the same configuration must be used regardless of document type, or even search index that the document is being placed in.

The field_definitions.json file describes what configuration should be used for each named field; this single-point of configuration ensures that differing configuration cannot be used for a single field name.

The file contains a JSON object in which the keys are the field names. Each value has the following properties:

description: a human-readable description of what type of data is expected to exist in the field.
type: one of the type values defined in the field_types.json file.
children: (only for fields for which the type had the children property set to named) the field definitions for the child fields. These are in the same format as the top-level field definitions in the file (and could even be recursive).

Elasticsearch document types

Documents in an elasticsearch index have a type, and each type may have very different configuration. We call this type "elasticsearch type" to differentiate with the document_type used by GOV.UK to describe types of pages.

We define a basic configuration which is used across all document types, and additional configuration for each document type is then merged with this.

The basic configuration is defined in a base_elasticsearch_type.json file. Document types are defined in the elasticsearch_types directory, with the additional configuration for each type being defined by a JSON file in that directory.

The files contain JSON object with the following keys:

fields: An array of field names which are allowed in documents of this type. The field names must be defined in the field_definitions.json file.
expanded_search_result_fields: DEPRECATED. Sets up field "expansion" for a field. This is used by the search result presenter to replace a raw value.

For example:
```
"expanded_search_result_fields": {
  "foo": [
    {
      "label": "Bar the bar",
      "value": "bar"
    },
    {
      "label": "Baz the baz",
      "value": "baz"
    }
  ],
```
Will return make the presenter replace the foo with value bar with the hash { "label": "Bar the bar", "value": "bar" }. This can be used when displaying the search results.

Even though we have different schemas for different "elasticsearch document types", in practice elasticsearch only knows about one "type": which is the union of all the schemas. This is because Elasticsearch 6 does not allow multiple types in the same index.

Indexes

Indexes in elasticsearch are defined by files in the indexes directory. These files contain a JSON object with the following keys:

elasticsearch_types: An array of the names of the document types allowed in this index.

Synonyms

Synonyms are defined in the synonyms.yml file.

Synonyms are specified in "lucene" syntax. Each synonym group is represented as comma separated lists of synonyms, and optionally a "=>" symbol.

If the => is provided, the left hand side contains a list of words which are combined at search time into a group, and the right hand side contains the list of words which are combined at index time into that same group. ie, the left hand side is the mapping applied to searches, and the right hand side is the mapping applied to documents.

If there is no =>, the same group of words is used at index time as at search time.

Some examples:

foo, bar => baz

means "a search for 'foo' or 'bar' should return documents with 'baz' in them"

foo => bar, baz

means "a search for 'foo' should return documents with 'bar' or 'baz' in them"

foo, bar, baz

is the same as foo, bar, baz => foo, bar, baz and means "a search for foo, bar or baz should return documents with any of them".

Additional configuration

Additional configuration is defined in the elasticsearch_schema.yml and stems.yml files. This configuration is merged with the JSON configuration, and then passed to elasticsearch directly.