Search Functions
- reference
Search functions enable you to use full text search (FTS) queries directly within a N1QL query.
Prerequisites
To use any of the search functions, the Search service must be available on the cluster. It is also recommended, but not required, that you should create suitable full text indexes for the searches that you need to perform. For more information, refer to Preparing for Full Text Searches.
The examples in this page all assume that demonstration full text indexes have been created, as described in Demonstration Indexes. |
Authorization
You do not need credentials for the FTS service to be able to use the search functions in a query. The role Data Admin must be assigned to those who intend to create indexes; and the role Data Reader to those who intend to perform searches. For information on creating users and assigning roles, see Authorization.
When to Use Search Functions
The search functions are useful when you need to combine a full text search with the power of a N1QL query; for example, combining joins and natural-language search in the same query.
If you only need to use the capabilities of a full text search without any N1QL features, consider making use of the Search service directly, through the user interface, the REST API, or an SDK.
SEARCH(identifier
, query
[, options
])
Description
This function enables you to use a full text search to filter a result set, or as a join predicate. It is only allowed in the WHERE clause or the ON clause.
If a query contains a SEARCH function, the Query engine analyzes the entire query, including the search specification, to select the best index to use with this search, taking any index hints into account. The Query engine then passes the search specification over to the Search engine to perform the search.
If no suitable full text index can be selected, or no full text index exists, the Query engine falls back on a Primary index or qualified GSI index to produce document keys, and then fetches the documents. The Search service then creates a temporary index in memory to perform the search. This process may be slower than using a suitable full text index. |
Arguments
- identifier
-
[Required] An expression in the form
keyspaceAlias[.path]
, consisting of the keyspace or keyspace alias in which to search, followed by the path to a field in which to search, using dot notation.-
The identifier must contain the keyspace or keyspace alias if there is more than one input source in the FROM clause. If there is only one input source in the FROM clause, and the identifier contains a path, the keyspace or keyspace alias may be omitted. However, if the path is omitted, the keyspace or keyspace alias is mandatory.
-
When the identifier contains a path, it is used as the default field in the query argument, as long as the query argument is a query string. If the path is omitted, the default field is set to
_all
. If the query argument is a query string which specifies a field, this field takes priority, and the path in the identifier is ignored. Similarly, if the query argument is a query object, the path is ignored. -
The path must use Search syntax rather than N1QL syntax; in other words, you cannot specify array locations such as
[*]
or[3]
in the path. -
If the keyspace, keyspace alias, or path contains any characters such as
-
, you must surround that part of the identifier with backticks``
.
The identifier argument cannot be replaced by a N1QL query parameter.
-
- query
-
[Required] The full text search query. This may be one of the following:
Type Description string
A query string. For more details, refer to Query String Query.
object
The query object within a full text search request. For more details, refer to Supported Queries.
object
A complete full text search request, including sort and pagination options, and so on. For more details, refer to Sorting Query Results.
When specifying a complete full text search request with the N1QL SEARCH() function, if the value of the
size
parameter is greater than the maximum number of full text search results, the query ignores thesize
parameter and returns all matching results.This is different to the behavior of a complete full text search request in the Search service, where the query returns an error if the value of the
size
parameter is greater than the maximum number of full text search results.The query argument may be replaced by a N1QL query parameter, as long as the query parameter resolves to a string or an object.
- options
-
[Optional] A JSON object containing options for the search. The object may contain the following fields:
Name Type Description index
string, object
The
index
field may be a string, containing the name of a full text index in the keyspace. (This may be a full text index alias, but only if the full text index is in the same keyspace.) This provides an index hint to the Query engine. If the full text index does not exist, an error occurs.You can also provide an index hint to the Query engine with the USE INDEX clause. This takes precedence over a hint provided by the
index
field.
The
index
field may also be an object, containing an example of a full text index mapping. This is treated as an input to the index mapping. It overrides the default mapping and is used during index selection and filtering.The object must either have a default mapping with no type mapping, or a single type mapping with the default mapping disabled. For more information, refer to Creating Indexes.
indexUUID
string
A string, containing the UUID of a full text index in the keyspace. This provides an index hint to the Query engine. If the full text index cannot be identified, an error occurs.
You can use the
indexUUID
field alongside theindex
field to help identify a full text index. TheindexUUID
field and theindex
field must both identify the same full text index. If they identify different full text indexes, or if either of them does not identify a full text index, an error occurs.You can find the UUID of a full text index by viewing the index definition. You can do this using the Index Definition Preview in the Query Workbench, or the Index Definition endpoints provided by the Full Text Search REST API.
out
string
A name given to this full text search operation in this keyspace. You can use this name to refer to this operation using the SEARCH_META() and SEARCH_SCORE() functions. If this field is omitted, the name of this full text search operation defaults to
"out"
.(other)
(any)
Other fields are ignored by the Query engine and are passed on to the Search engine as options. The values of these options may be replaced with N1QL query parameters, such as
"analyzer": $analyzer
.The options argument cannot be replaced by a N1QL query parameter, but it may contain N1QL query parameters.
Return Value
A boolean, representing whether the search query is found within the input path.
This returns true
if the search query is found within the input path, or false
otherwise.
Limitations
The Query service can select a full text index for efficient search in the following cases:
-
If the SEARCH() function is used in a WHERE clause or in an ANSI JOIN. The SEARCH() function must be on the leftmost (first) JOIN. It may be on the outer side of a nested-loop JOIN, or either side of a hash JOIN. RIGHT OUTER JOINs are rewritten as LEFT OUTER JOINs.
-
If the SEARCH() function is evaluated on the
true
condition in positive cases: for example,SEARCH(field, query, options)
,SEARCH(field, query, options) = true
,SEARCH(field, query, options) IN [true, true, true]
, or a condition including one of these withAND
orOR
.
The Query service cannot select a full text index for efficient search in the following cases:
-
If a USE KEYS hint is present; or if the SEARCH() function is used on the inner side of a nested-loop JOIN, a lookup JOIN or lookup NEST, an index JOIN or index NEST, an UNNEST clause, a subquery expression, a subquery result, or a correlated query.
-
If the SEARCH() function is evaluated on the
false
condition, or in negative cases: for example,NOT SEARCH(field, query, options)
,SEARCH(field, query, options) = false
,SEARCH(field, query, options) != false
,SEARCH(field, query, options) IN [false, true, 1, "a"]
, or in a condition using the relation operators<
,<=
,>
,>=
,BETWEEN
,NOT
,LIKE
, orNOT LIKE
.
In these cases, the Query service must fetch the documents, and the Search service creates a temporary index in memory to perform the search. This may affect performance.
If the SEARCH() function is present for a keyspace, no GSI covering scan is possible on that keyspace. If more than one FTS or GSI index are used in the plan, IntersectScan or Ordered IntersectScan is performed. To avoid this, use a USE INDEX hint.
Order pushdown is possible only if query ORDER BY has only SEARCH_SCORE() on the leftmost keyspace. Offset and Limit pushdown is possible if the query only has a SEARCH() predicate, using a single search index — no IntersectScan or OrderIntersectScan. Group aggregates and projection are not pushed.
Examples
The following queries are equivalent:
SELECT META(t1).id
FROM `travel-sample`.inventory.airline AS t1
WHERE SEARCH(t1.country, "+United +States");
SELECT META(t1).id
FROM `travel-sample`.inventory.airline AS t1
WHERE SEARCH(t1, "country:\"United States\"");
[
{
"id": "airline_10"
},
{
"id": "airline_10123"
},
{
"id": "airline_10226"
},
{
"id": "airline_10748"
},
...
]
The results are unordered, so they may be returned in a different order each time.
SELECT t1.name
FROM `travel-sample`.inventory.hotel AS t1
WHERE SEARCH(t1, {
"match": "bathrobes",
"field": "reviews.content",
"analyzer": "standard"
});
[
{
"name": "Typoeth Cottage"
},
{
"name": "Great Orme Lighthouse"
},
{
"name": "New Road Guest House (B&B)"
},
...
]
The results are unordered, so they may be returned in a different order each time.
SELECT t1.name
FROM `travel-sample`.inventory.hotel AS t1
WHERE SEARCH(t1, {
"explain": false,
"fields": [
"*"
],
"highlight": {},
"query": {
"match": "bathrobes",
"field": "reviews.content",
"analyzer": "standard"
},
"size" : 5,
"sort": [
{
"by" : "field",
"field" : "reviews.ratings.Overall",
"mode" : "max",
"missing" : "last"
}
]
});
[
{
"name": "Waunifor"
},
{
"name": "Bistro Prego With Rooms"
},
{
"name": "Thornehill Broome Beach Campground"
},
...
]
SELECT META(t1).id
FROM `travel-sample`.inventory.hotel AS t1
WHERE t1.type = "hotel" AND SEARCH(t1.description, "amazing");
[
{
"id": "hotel_20422"
},
{
"id": "hotel_22096"
},
{
"id": "hotel_25243"
},
{
"id": "hotel_27741"
}
]
If the full text search index being queried has its default mapping disabled and has a custom type mapping defined, the query needs to specify the type explicitly. The above query uses the demonstration index travel-sample-index-hotel-description, which has the custom type mapping "hotel".
For more information on defining custom type mappings within the full text search index, refer to Specifying Type Mappings. Note that for N1QL queries, only full text search indexes with one type mapping are searchable. Also the supported type identifiers at the moment are "type_field" and "docid_prefix"; "docid_regexp" isn’t supported yet for SEARCH queries via N1QL.
SEARCH_META([identifier
])
Description
This function is intended to be used in a query which contains a SEARCH() function. It returns the metadata given by the Search engine for each document found by the SEARCH() function. If there is no SEARCH() function in the query, or if a full text index was not used to evaluate the search, the function returns MISSING.
Arguments
- identifier
-
[Optional] An expression in the form
[keyspaceAlias.]outname
, consisting of the keyspace or keyspace alias in which the full text search operation was performed, followed by the outname of the full text search operation, using dot notation.
|
Return Value
A JSON object containing the metadata returned by the Search engine. By default, the metadata includes the score and ID of the search result. It may also include other metadata requested by advanced search options, such as the location of the search terms or an explanation of the search results.
Examples
SELECT SEARCH_META() AS meta (1)
FROM `travel-sample`.inventory.hotel AS t1
WHERE SEARCH(t1, {
"query": {
"match": "bathrobes",
"field": "reviews.content",
"analyzer": "standard"
},
"includeLocations": true (2)
})
LIMIT 3;
[
{
"meta": {
"id": "hotel_12068", (3)
"locations": { (4)
"reviews.content": {
"bathrobes": [
{
"array_positions": [
8
],
"end": 664,
"pos": 122,
"start": 655
}
]
}
},
"score": 0.3471730605306995 (5)
}
},
{
"meta": {
"id": "hotel_18819",
"locations": {
"reviews.content": {
"bathrobes": [
{
"array_positions": [
6
],
"end": 110,
"pos": 19,
"start": 101
}
]
}
},
"score": 0.3778486940124847
}
},
{
"meta": {
"id": "hotel_5841",
"locations": {
"reviews.content": {
"bathrobes": [
{
"array_positions": [
0
],
"end": 1248,
"pos": 242,
"start": 1239
}
]
}
},
"score": 0.3696905918027607
}
}
]
1 | There is only one SEARCH() function in this query, so the SEARCH_META() function does not need to specify the outname. |
2 | The full text search specifies that locations should be included in the search result metadata. |
3 | The id is included in the search result metadata by default. |
4 | The location of the search term is included in the search result metadata as requested. |
5 | The score is included in the search result metadata by default. |
SELECT t1.name, SEARCH_META(s1) AS meta (1)
FROM `travel-sample`.inventory.hotel AS t1
WHERE SEARCH(t1.description, "mountain", {"out": "s1"}) (2)
AND SEARCH(t1, {
"query": {
"match": "bathrobes",
"field": "reviews.content",
"analyzer": "standard"
}
});
[
{
"name": "Marina del Rey Marriott"
}
]
SEARCH_SCORE([identifier
])
Description
This function is intended to be used in a query which contains a SEARCH() function. It returns the score given by the Search engine for each document found by the SEARCH() function. If there is no SEARCH() function in the query, or if a full text index was not used to evaluate the search, the function returns MISSING.
This function is the same as SEARCH_META().score.
Arguments
- identifier
-
[Optional] An expression in the form
[keyspaceAlias.]outname
, consisting of the keyspace or keyspace alias in which the full text search operation was performed, followed by the outname of the full text search operation, using dot notation.
|
Examples
SELECT name, description, SEARCH_SCORE() AS score (1)
FROM `travel-sample`.inventory.hotel AS t1
WHERE SEARCH(t1.description, "mountain")
ORDER BY score DESC
LIMIT 3;
[
{
"description": "3 Star Hotel next to the Mountain Railway terminus and set in 30 acres of grounds which include Dolbadarn Castle",
"name": "The Royal Victoria Hotel"
},
{
"description": "370 guest rooms offering both water and mountain view.",
"name": "Marina del Rey Marriott"
},
{
"description": "This small family run hotel captures the spirit of Mull and is a perfect rural holiday retreat. The mountain and sea blend together to give fantastic, panoramic views from the hotel which is in an elevated position on the shoreline. Panoramic views are also available from the bar and restaurant which serves local produce 7 days a week.",
"name": "The Glenforsa Hotel"
}
]
1 | There is only one SEARCH() function in this query, so the SEARCH_SCORE() function does not need to specify the outname. |