github.com/dgraph-io/dgraph@v1.2.8/wiki/content/query-language/index.md

+++
title = "Query Language"
+++

Dgraph's GraphQL+- is based on Facebook's [GraphQL](https://facebook.github.io/graphql/).  GraphQL wasn't developed for Graph databases, but its graph-like query syntax, schema validation and subgraph shaped response make it a great language choice.  We've modified the language to better support graph operations, adding and removing features to get the best fit for graph databases.  We're calling this simplified, feature rich language, ''GraphQL+-''.

GraphQL+- is a work in progress. We're adding more features and we might further simplify existing ones.

## Take a Tour - https://tour.dgraph.io

This document is the Dgraph query reference material.  It is not a tutorial.  It's designed as a reference for users who already know how to write queries in GraphQL+- but need to check syntax, or indices, or functions, etc.

{{% notice "note" %}}If you are new to Dgraph and want to learn how to use Dgraph and GraphQL+-, take the tour - https://tour.dgraph.io{{% /notice %}}


### Running examples

The examples in this reference use a database of 21 million triples about movies and actors.  The example queries run and return results.  The queries are executed by an instance of Dgraph running at https://play.dgraph.io/.  To run the queries locally or experiment a bit more, see the [Getting Started]({{< relref "get-started/index.md" >}}) guide, which also shows how to load the datasets used in the examples here.

## GraphQL+- Fundamentals

A GraphQL+- query finds nodes based on search criteria, matches patterns in a graph and returns a graph as a result.

A query is composed of nested blocks, starting with a query root.  The root finds the initial set of nodes against which the following graph matching and filtering is applied.

{{% notice "note" %}}See more about Queries in [Queries design concept]({{< relref "design-concepts/index.md#queries" >}}) {{% /notice %}}

### Returning Values

Each query has a name, specified at the query root, and the same name identifies the results.

If an edge is of a value type, the value can be returned by giving the edge name.

Query Example: In the example dataset, edges that link movies to directors and actors, movies have a name, release date and identifiers for a number of well known movie databases.  This query, with name `bladerunner`, and root matching a movie name, returns those values for the early 80's sci-fi classic "Blade Runner".

{{< runnable >}}
{
  bladerunner(func: eq(name@en, "Blade Runner")) {
    uid
    name@en
    initial_release_date
    netflix_id
  }
}
{{< /runnable >}}

The query first searches the graph, using indexes to make the search efficient, for all nodes with a `name` edge equaling "Blade Runner".  For the found node the query then returns the listed outgoing edges.

Every node had a unique 64-bit identifier.  The `uid` edge in the query above returns that identifier.  If the required node is already known, then the function `uid` finds the node.

Query Example: "Blade Runner" movie data found by UID.

{{< runnable >}}
{
  bladerunner(func: uid(0x394c)) {
    uid
    name@en
    initial_release_date
    netflix_id
  }
}
{{< /runnable >}}

A query can match many nodes and return the values for each.

Query Example: All nodes that have either "Blade" or "Runner" in the name.

{{< runnable >}}
{
  bladerunner(func: anyofterms(name@en, "Blade Runner")) {
    uid
    name@en
    initial_release_date
    netflix_id
  }
}
{{< /runnable >}}

Multiple IDs can be specified in a list to the `uid` function.

Query Example:
{{< runnable >}}
{
  movies(func: uid(0xb5849, 0x394c)) {
    uid
    name@en
    initial_release_date
    netflix_id
  }
}
{{< /runnable >}}


{{% notice "note" %}} If your predicate has special characters, then you should wrap it with angular
brackets while asking for it in the query. E.g. `<first:name>`{{% /notice %}}

### Expanding Graph Edges

A query expands edges from node to node by nesting query blocks with `{ }`.

Query Example: The actors and characters played in "Blade Runner".  The query first finds the node with name "Blade Runner", then follows  outgoing `starring` edges to nodes representing an actor's performance as a character.  From there the `performance.actor` and `performance.character` edges are expanded to find the actor names and roles for every actor in the movie.
{{< runnable >}}
{
  brCharacters(func: eq(name@en, "Blade Runner")) {
    name@en
    initial_release_date
    starring {
      performance.actor {
        name@en  # actor name
      }
      performance.character {
        name@en  # character name
      }
    }
  }
}
{{< /runnable >}}


### Comments

Anything on a line following a `#` is a comment

### Applying Filters

The query root finds an initial set of nodes and the query proceeds by returning values and following edges to further nodes - any node reached in the query is found by traversal after the search at root.  The nodes found can be filtered by applying `@filter`, either after the root or at any edge.

Query Example: "Blade Runner" director Ridley Scott's movies released before the year 2000.
{{< runnable >}}
{
  scott(func: eq(name@en, "Ridley Scott")) {
    name@en
    initial_release_date
    director.film @filter(le(initial_release_date, "2000")) {
      name@en
      initial_release_date
    }
  }
}
{{< /runnable >}}

Query Example: Movies with either "Blade" or "Runner" in the title and released before the year 2000.

{{< runnable >}}
{
  bladerunner(func: anyofterms(name@en, "Blade Runner")) @filter(le(initial_release_date, "2000")) {
    uid
    name@en
    initial_release_date
    netflix_id
  }
}
{{< /runnable >}}

### Language Support

{{% notice "note" %}}A `@lang` directive must be specified in the schema to query or mutate
predicates with language tags.{{% /notice %}}

Dgraph supports UTF-8 strings.

In a query, for a string valued edge `edge`, the syntax
```
edge@lang1:...:langN
```
specifies the preference order for returned languages, with the following rules.

* At most one result will be returned (except in the case where the language list is set to *).
* The preference list is considered left to right: if a value in given language is not found, the next language from the list is considered.
* If there are no values in any of the specified languages, no value is returned.
* A final `.` means that a value without a specified language is returned or if there is no value without language, a value in ''some'' language is returned.
* Setting the language list value to * will return all the values for that predicate along with their language. Values without a language tag are also returned.

For example:

- `name`   => Look for an untagged string; return nothing if no untagged value exits.
- `name@.` => Look for an untagged string, then any language.
- `name@en` => Look for `en` tagged string; return nothing if no `en` tagged string exists.
- `name@en:.` => Look for `en`, then untagged, then any language.
- `name@en:pl` => Look for `en`, then `pl`, otherwise nothing.
- `name@en:pl:.` => Look for `en`, then `pl`, then untagged, then any language.
- `name@*` => Look for all the values of this predicate and return them along with their language. For example, if there are two values with languages en and hi, this query will return two keys named "name@en" and "name@hi".


{{% notice "note" %}}In functions, language lists (including the `@*` notation) are not allowed. Untagged predicates, Single language tags, and `.` notation work as described above.

---

In [full-text search functions]({{< relref "#full-text-search" >}}) (`alloftext`, `anyoftext`), when no language is specified (untagged or `@.`), the default (English) full-text tokenizer is used.{{% /notice %}}


Query Example: Some of Bollywood director and actor Farhan Akhtar's movies have a name stored in Russian as well as Hindi and English, others do not.

{{< runnable >}}
{
  q(func: allofterms(name@en, "Farhan Akhtar")) {
    name@hi
    name@en

    director.film {
      name@ru:hi:en
      name@en
      name@hi
      name@ru
    }
  }
}
{{< /runnable >}}




## Functions

{{% notice "note" %}}Functions can only be applied to [indexed]({{< relref "#indexing">}}) predicates.{{% /notice %}}

Functions allow filtering based on properties of nodes or variables.  Functions can be applied in the query root or in filters.

For functions on string valued predicates, if no language preference is given, the function is applied to all languages and strings without a language tag; if a language preference is given, the function is applied only to strings of the given language.


### Term matching


#### allofterms

Syntax Example: `allofterms(predicate, "space-separated term list")`

Schema Types: `string`

Index Required: `term`


Matches strings that have all specified terms in any order; case insensitive.

##### Usage at root

Query Example: All nodes that have `name` containing terms `indiana` and `jones`, returning the English name and genre in English.

{{< runnable >}}
{
  me(func: allofterms(name@en, "jones indiana")) {
    name@en
    genre {
      name@en
    }
  }
}
{{< /runnable >}}

##### Usage as Filter

Query Example: All Steven Spielberg films that contain the words `indiana` and `jones`.  The `@filter(has(director.film))` removes nodes with name Steven Spielberg that aren't the director --- the data also contains a character in a film called Steven Spielberg.

{{< runnable >}}
{
  me(func: eq(name@en, "Steven Spielberg")) @filter(has(director.film)) {
    name@en
    director.film @filter(allofterms(name@en, "jones indiana"))  {
      name@en
    }
  }
}
{{< /runnable >}}


#### anyofterms


Syntax Example: `anyofterms(predicate, "space-separated term list")`

Schema Types: `string`

Index Required: `term`


Matches strings that have any of the specified terms in any order; case insensitive.

##### Usage at root

Query Example: All nodes that have a `name` containing either `poison` or `peacock`.  Many of the returned nodes are movies, but people like Joan Peacock also meet the search terms because without a [cascade directive]({{< relref "#cascade-directive">}}) the query doesn't require a genre.

{{< runnable >}}
{
  me(func:anyofterms(name@en, "poison peacock")) {
    name@en
    genre {
      name@en
    }
  }
}
{{< /runnable >}}


##### Usage as filter

Query Example: All Steven Spielberg movies that contain `war` or `spies`.  The `@filter(has(director.film))` removes nodes with name Steven Spielberg that aren't the director --- the data also contains a character in a film called Steven Spielberg.

{{< runnable >}}
{
  me(func: eq(name@en, "Steven Spielberg")) @filter(has(director.film)) {
    name@en
    director.film @filter(anyofterms(name@en, "war spies"))  {
      name@en
    }
  }
}
{{< /runnable >}}


### Regular Expressions


Syntax Examples: `regexp(predicate, /regular-expression/)` or case insensitive `regexp(predicate, /regular-expression/i)`

Schema Types: `string`

Index Required: `trigram`


Matches strings by regular expression.  The regular expression language is that of [go regular expressions](https://golang.org/pkg/regexp/syntax/).

Query Example: At root, match nodes with `Steven Sp` at the start of `name`, followed by any characters.  For each such matched uid, match the films containing `ryan`.  Note the difference with `allofterms`, which would match only `ryan` but regular expression search will also match within terms, such as `bryan`.

{{< runnable >}}
{
  directors(func: regexp(name@en, /^Steven Sp.*$/)) {
    name@en
    director.film @filter(regexp(name@en, /ryan/i)) {
      name@en
    }
  }
}
{{< /runnable >}}


#### Technical details

A Trigram is a substring of three continuous runes. For example, `Dgraph` has trigrams `Dgr`, `gra`, `rap`, `aph`.

To ensure efficiency of regular expression matching, Dgraph uses [trigram indexing](https://swtch.com/~rsc/regexp/regexp4.html).  That is, Dgraph converts the regular expression to a trigram query, uses the trigram index and trigram query to find possible matches and applies the full regular expression search only to the possibles.

#### Writing Efficient Regular Expressions and Limitations

Keep the following in mind when designing regular expression queries.

- At least one trigram must be matched by the regular expression (patterns shorter than 3 runes are not supported).  That is, Dgraph requires regular expressions that can be converted to a trigram query.
- The number of alternative trigrams matched by the regular expression should be as small as possible  (`[a-zA-Z][a-zA-Z][0-9]` is not a good idea).  Many possible matches means the full regular expression is checked against many strings; where as, if the expression enforces more trigrams to match, Dgraph can make better use of the index and check the full regular expression against a smaller set of possible matches.
- Thus, the regular expression should be as precise as possible.  Matching longer strings means more required trigrams, which helps to effectively use the index.
- If repeat specifications (`*`, `+`, `?`, `{n,m}`) are used, the entire regular expression must not match the _empty_ string or _any_ string: for example, `*` may be used like `[Aa]bcd*` but not like `(abcd)*` or `(abcd)|((defg)*)`
- Repeat specifications after bracket expressions (e.g. `[fgh]{7}`, `[0-9]+` or `[a-z]{3,5}`) are often considered as matching any string because they match too many trigrams.
- If the partial result (for subset of trigrams) exceeds 1000000 uids during index scan, the query is stopped to prohibit expensive queries.


### Fuzzy matching


Syntax: `match(predicate, string, distance)`

Schema Types: `string`

Index Required: `trigram`

Matches predicate values by calculating the [Levenshtein distance](https://en.wikipedia.org/wiki/Levenshtein_distance) to the string,
also known as _fuzzy matching_. The distance parameter must be greater than zero (0). Using a greater distance value can yield more but less accurate results.

Query Example: At root, fuzzy match nodes similar to `Stephen`, with a distance value of less than or equal to 8.

{{< runnable >}}
{
  directors(func: match(name@en, Stephen, 8)) {
    name@en
  }
}
{{< /runnable >}}

Same query with a Levenshtein distance of 3.

{{< runnable >}}
{
  directors(func: match(name@en, Stephen, 3)) {
    name@en
  }
}
{{< /runnable >}}


### Full-Text Search

Syntax Examples: `alloftext(predicate, "space-separated text")` and `anyoftext(predicate, "space-separated text")`

Schema Types: `string`

Index Required: `fulltext`


Apply full-text search with stemming and stop words to find strings matching all or any of the given text.

The following steps are applied during index generation and to process full-text search arguments:

1. Tokenization (according to Unicode word boundaries).
1. Conversion to lowercase.
1. Unicode-normalization (to [Normalization Form KC](http://unicode.org/reports/tr15/#Norm_Forms)).
1. Stemming using language-specific stemmer (if supported by language).
1. Stop words removal (if supported by language).

Dgraph uses [bleve](https://github.com/blevesearch/bleve) for its full-text search indexing. See also the bleve language specific [stop word lists](https://github.com/blevesearch/bleve/tree/master/analysis/lang).

Following table contains all supported languages, corresponding country-codes, stemming and stop words filtering support.

|  Language  | Country Code | Stemming | Stop words |
| :--------: | :----------: | :------: | :--------: |
|   Arabic   |      ar      | &#10003; |  &#10003;  |
|  Armenian  |      hy      |          |  &#10003;  |
|   Basque   |      eu      |          |  &#10003;  |
| Bulgarian  |      bg      |          |  &#10003;  |
|  Catalan   |      ca      |          |  &#10003;  |
|  Chinese   |      zh      | &#10003; |  &#10003;  |
|   Czech    |      cs      |          |  &#10003;  |
|   Danish   |      da      | &#10003; |  &#10003;  |
|   Dutch    |      nl      | &#10003; |  &#10003;  |
|  English   |      en      | &#10003; |  &#10003;  |
|  Finnish   |      fi      | &#10003; |  &#10003;  |
|   French   |      fr      | &#10003; |  &#10003;  |
|   Gaelic   |      ga      |          |  &#10003;  |
|  Galician  |      gl      |          |  &#10003;  |
|   German   |      de      | &#10003; |  &#10003;  |
|   Greek    |      el      |          |  &#10003;  |
|   Hindi    |      hi      | &#10003; |  &#10003;  |
| Hungarian  |      hu      | &#10003; |  &#10003;  |
| Indonesian |      id      |          |  &#10003;  |
|  Italian   |      it      | &#10003; |  &#10003;  |
|  Japanese  |      ja      | &#10003; |  &#10003;  |
|   Korean   |      ko      | &#10003; |  &#10003;  |
| Norwegian  |      no      | &#10003; |  &#10003;  |
|  Persian   |      fa      |          |  &#10003;  |
| Portuguese |      pt      | &#10003; |  &#10003;  |
|  Romanian  |      ro      | &#10003; |  &#10003;  |
|  Russian   |      ru      | &#10003; |  &#10003;  |
|  Spanish   |      es      | &#10003; |  &#10003;  |
|  Swedish   |      sv      | &#10003; |  &#10003;  |
|  Turkish   |      tr      | &#10003; |  &#10003;  |


Query Example: All names that have `dog`, `dogs`, `bark`, `barks`, `barking`, etc.  Stop word removal eliminates `the` and `which`.

{{< runnable >}}
{
  movie(func:alloftext(name@en, "the dog which barks")) {
    name@en
  }
}
{{< /runnable >}}


### Inequality

#### equal to

Syntax Examples:

* `eq(predicate, value)`
* `eq(val(varName), value)`
* `eq(predicate, val(varName))`
* `eq(count(predicate), value)`
* `eq(predicate, [val1, val2, ..., valN])`
* `eq(predicate, [$var1, "value", ..., $varN])`

Schema Types: `int`, `float`, `bool`, `string`, `dateTime`

Index Required: An index is required for the `eq(predicate, ...)` forms (see table below).  For `count(predicate)` at the query root, the `@count` index is required. For variables the values have been calculated as part of the query, so no index is required.

| Type       | Index Options |
|:-----------|:--------------|
| `int`      | `int`         |
| `float`    | `float`       |
| `bool`     | `bool`        |
| `string`   | `exact`, `hash` |
| `dateTime` | `dateTime`    |

Test for equality of a predicate or variable to a value or find in a list of values.

The boolean constants are `true` and `false`, so with `eq` this becomes, for example, `eq(boolPred, true)`.

Query Example: Movies with exactly thirteen genres.

{{< runnable >}}
{
  me(func: eq(count(genre), 13)) {
    name@en
    genre {
      name@en
    }
  }
}
{{< /runnable >}}


Query Example: Directors called Steven who have directed 1,2 or 3 movies.

{{< runnable >}}
{
  steve as var(func: allofterms(name@en, "Steven")) {
    films as count(director.film)
  }

  stevens(func: uid(steve)) @filter(eq(val(films), [1,2,3])) {
    name@en
    numFilms : val(films)
  }
}
{{< /runnable >}}


#### less than, less than or equal to, greater than and greater than or equal to

Syntax Examples: for inequality `IE`

* `IE(predicate, value)`
* `IE(val(varName), value)`
* `IE(predicate, val(varName))`
* `IE(count(predicate), value)`

With `IE` replaced by

* `le` less than or equal to
* `lt` less than
* `ge` greater than or equal to
* `gt` greather than

Schema Types: `int`, `float`, `string`, `dateTime`

Index required: An index is required for the `IE(predicate, ...)` forms (see table below).  For `count(predicate)` at the query root, the `@count` index is required. For variables the values have been calculated as part of the query, so no index is required.

| Type       | Index Options |
|:-----------|:--------------|
| `int`      | `int`         |
| `float`    | `float`       |
| `string`   | `exact`       |
| `dateTime` | `dateTime`    |


Query Example: Ridley Scott movies released before 1980.

{{< runnable >}}
{
  me(func: eq(name@en, "Ridley Scott")) {
    name@en
    director.film @filter(lt(initial_release_date, "1980-01-01"))  {
      initial_release_date
      name@en
    }
  }
}
{{< /runnable >}}


Query Example: Movies with directors with `Steven` in `name` and have directed more than `100` actors.

{{< runnable >}}
{
  ID as var(func: allofterms(name@en, "Steven")) {
    director.film {
      num_actors as count(starring)
    }
    total as sum(val(num_actors))
  }

  dirs(func: uid(ID)) @filter(gt(val(total), 100)) {
    name@en
    total_actors : val(total)
  }
}
{{< /runnable >}}



Query Example: A movie in each genre that has over 30000 movies.  Because there is no order specified on genres, the order will be by UID.  The [count index]({{< relref "#count-index">}}) records the number of edges out of nodes and makes such queries more .

{{< runnable >}}
{
  genre(func: gt(count(~genre), 30000)){
    name@en
    ~genre (first:1) {
      name@en
    }
  }
}
{{< /runnable >}}

Query Example: Directors called Steven and their movies which have `initial_release_date` greater
than that of the movie Minority Report.

{{< runnable >}}
{
  var(func: eq(name@en,"Minority Report")) {
    d as initial_release_date
  }

  me(func: eq(name@en, "Steven Spielberg")) {
    name@en
    director.film @filter(ge(initial_release_date, val(d))) {
      initial_release_date
      name@en
    }
  }
}
{{< /runnable >}}


### uid

Syntax Examples:

* `q(func: uid(<uid>)) `
* `predicate @filter(uid(<uid1>, ..., <uidn>))`
* `predicate @filter(uid(a))` for variable `a`
* `q(func: uid(a,b))` for variables `a` and `b`


Filters nodes at the current query level to only nodes in the given set of UIDs.

For query variable `a`, `uid(a)` represents the set of UIDs stored in `a`.  For value variable `b`, `uid(b)` represents the UIDs from the UID to value map.  With two or more variables, `uid(a,b,...)` represents the union of all the variables.

`uid(<uid>)`, like an identity function, will return the requested UID even if the node does not have any edges.

Query Example: If the UID of a node is known, values for the node can be read directly.  The films of Priyanka Chopra by known UID

{{< runnable >}}
{
  films(func: uid(0x2c964)) {
    name@hi
    actor.film {
      performance.film {
        name@hi
      }
    }
  }
}
{{< /runnable >}}



Query Example: The films of Taraji Henson by genre.
{{< runnable >}}
{
  var(func: allofterms(name@en, "Taraji Henson")) {
    actor.film {
      F as performance.film {
        G as genre
      }
    }
  }

  Taraji_films_by_genre(func: uid(G)) {
    genre_name : name@en
    films : ~genre @filter(uid(F)) {
      film_name : name@en
    }
  }
}
{{< /runnable >}}



Query Example: Taraji Henson films ordered by number of genres, with genres listed in order of how many films Taraji has made in each genre.
{{< runnable >}}
{
  var(func: allofterms(name@en, "Taraji Henson")) {
    actor.film {
      F as performance.film {
        G as count(genre)
        genre {
          C as count(~genre @filter(uid(F)))
        }
      }
    }
  }

  Taraji_films_by_genre_count(func: uid(G), orderdesc: val(G)) {
    film_name : name@en
    genres : genre (orderdesc: val(C)) {
      genre_name : name@en
    }
  }
}
{{< /runnable >}}


### uid_in


Syntax Examples:

* `q(func: ...) @filter(uid_in(predicate, <uid>))`
* `predicate1 @filter(uid_in(predicate2, <uid>))`

Schema Types: UID

Index Required: none

While the `uid` function filters nodes at the current level based on UID, function `uid_in` allows looking ahead along an edge to check that it leads to a particular UID.  This can often save an extra query block and avoids returning the edge.

`uid_in` cannot be used at root, it accepts one UID constant as its argument (not a variable).


Query Example: The collaborations of Marc Caro and Jean-Pierre Jeunet (UID 0x99706).  If the UID of Jean-Pierre Jeunet is known, querying this way removes the need to have a block extracting his UID into a variable and the extra edge traversal and filter for `~director.film`.
{{< runnable >}}
{
  caro(func: eq(name@en, "Marc Caro")) {
    name@en
    director.film @filter(uid_in(~director.film, 0x99706)) {
      name@en
    }
  }
}
{{< /runnable >}}


### has

Syntax Examples: `has(predicate)`

Schema Types: all

Determines if a node has a particular predicate.

Query Example: First five directors and all their movies that have a release date recorded.  Directors have directed at least one film --- equivalent semantics to `gt(count(director.film), 0)`.
{{< runnable >}}
{
  me(func: has(director.film), first: 5) {
    name@en
    director.film @filter(has(initial_release_date))  {
      initial_release_date
      name@en
    }
  }
}
{{< /runnable >}}

### Geolocation

{{% notice "note" %}} As of now we only support indexing Point, Polygon and MultiPolygon [geometry types](https://github.com/twpayne/go-geom#geometry-types). However, Dgraph can store other types of gelocation data. {{% /notice %}}

Note that for geo queries, any polygon with holes is replace with the outer loop, ignoring holes.  Also, as for version 0.7.7 polygon containment checks are approximate.

#### Mutations

To make use of the geo functions you would need an index on your predicate.
```
loc: geo @index(geo) .
```

Here is how you would add a `Point`.

```
{
  set {
    <_:0xeb1dde9c> <loc> "{'type':'Point','coordinates':[-122.4220186,37.772318]}"^^<geo:geojson> .
    <_:0xeb1dde9c> <name> "Hamon Tower" .
    <_:0xeb1dde9c> <dgraph.type> "Location" .
  }
}
```

Here is how you would associate a `Polygon` with a node. Adding a `MultiPolygon` is also similar.

```
{
  set {
    <_:0xf76c276b> <loc> "{'type':'Polygon','coordinates':[[[-122.409869,37.7785442],[-122.4097444,37.7786443],[-122.4097544,37.7786521],[-122.4096334,37.7787494],[-122.4096233,37.7787416],[-122.4094004,37.7789207],[-122.4095818,37.7790617],[-122.4097883,37.7792189],[-122.4102599,37.7788413],[-122.409869,37.7785442]],[[-122.4097357,37.7787848],[-122.4098499,37.778693],[-122.4099025,37.7787339],[-122.4097882,37.7788257],[-122.4097357,37.7787848]]]}"^^<geo:geojson> .
    <_:0xf76c276b> <name> "Best Western Americana Hotel" .
    <_:0xf76c276b> <dgraph.type> "Location" .
  }
}
```

The above examples have been picked from our [SF Tourism](https://github.com/dgraph-io/benchmarks/blob/master/data/sf.tourism.gz?raw=true) dataset.

#### Query

##### near

Syntax Example: `near(predicate, [long, lat], distance)`

Schema Types: `geo`

Index Required: `geo`

Matches all entities where the location given by `predicate` is within `distance` meters of geojson coordinate `[long, lat]`.

Query Example: Tourist destinations within 1000 meters (1 kilometer) of a point in Golden Gate Park in San Francisco.

{{< runnable >}}
{
  tourist(func: near(loc, [-122.469829, 37.771935], 1000) ) {
    name
  }
}
{{< /runnable >}}


##### within

Syntax Example: `within(predicate, [[[long1, lat1], ..., [longN, latN]]])`

Schema Types: `geo`

Index Required: `geo`

Matches all entities where the location given by `predicate` lies within the polygon specified by the geojson coordinate array.

Query Example: Tourist destinations within the specified area of Golden Gate Park, San Francisco.

{{< runnable >}}
{
  tourist(func: within(loc, [[[-122.47266769409178, 37.769018558337926 ], [ -122.47266769409178, 37.773699921075135 ], [ -122.4651575088501, 37.773699921075135 ], [ -122.4651575088501, 37.769018558337926 ], [ -122.47266769409178, 37.769018558337926]]] )) {
    name
  }
}
{{< /runnable >}}


##### contains

Syntax Examples: `contains(predicate, [long, lat])` or `contains(predicate, [[long1, lat1], ..., [longN, latN]])`

Schema Types: `geo`

Index Required: `geo`

Matches all entities where the polygon describing the location given by `predicate` contains geojson coordinate `[long, lat]` or given geojson polygon.

Query Example : All entities that contain a point in the flamingo enclosure of San Francisco Zoo.
{{< runnable >}}
{
  tourist(func: contains(loc, [ -122.50326097011566, 37.73353615592843 ] )) {
    name
  }
}
{{< /runnable >}}


##### intersects

Syntax Example: `intersects(predicate, [[[long1, lat1], ..., [longN, latN]]])`

Schema Types: `geo`

Index Required: `geo`

Matches all entities where the polygon describing the location given by `predicate` intersects the given geojson polygon.


{{< runnable >}}
{
  tourist(func: intersects(loc, [[[-122.503325343132, 37.73345766902749 ], [ -122.503325343132, 37.733903134117966 ], [ -122.50271648168564, 37.733903134117966 ], [ -122.50271648168564, 37.73345766902749 ], [ -122.503325343132, 37.73345766902749]]] )) {
    name
  }
}
{{< /runnable >}}



## Connecting Filters

Within `@filter` multiple functions can be used with boolean connectives.

### AND, OR and NOT

Connectives `AND`, `OR` and `NOT` join filters and can be built into arbitrarily complex filters, such as `(NOT A OR B) AND (C AND NOT (D OR E))`.  Note that, `NOT` binds more tightly than `AND` which binds more tightly than `OR`.

Query Example : All Steven Spielberg movies that contain either both "indiana" and "jones" OR both "jurassic" and "park".

{{< runnable >}}
{
  me(func: eq(name@en, "Steven Spielberg")) @filter(has(director.film)) {
    name@en
    director.film @filter(allofterms(name@en, "jones indiana") OR allofterms(name@en, "jurassic park"))  {
      uid
      name@en
    }
  }
}
{{< /runnable >}}


## Alias

Syntax Examples:

* `aliasName : predicate`
* `aliasName : predicate { ... }`
* `aliasName : varName as ...`
* `aliasName : count(predicate)`
* `aliasName : max(val(varName))`

An alias provides an alternate name in results.  Predicates, variables and aggregates can be aliased by prefixing with the alias name and `:`.  Aliases do not have to be different to the original predicate name, but, within a block, an alias must be distinct from predicate names and other aliases returned in the same block.  Aliases can be used to return the same predicate multiple times within a block.



Query Example: Directors with `name` matching term `Steven`, their UID, English name, average number of actors per movie, total number of films, and the name of each film in English and French.
{{< runnable >}}
{
  ID as var(func: allofterms(name@en, "Steven")) @filter(has(director.film)) {
    director.film {
      num_actors as count(starring)
    }
    average as avg(val(num_actors))
  }

  films(func: uid(ID)) {
    director_id : uid
    english_name : name@en
    average_actors : val(average)
    num_films : count(director.film)

    films : director.film {
      name : name@en
      english_name : name@en
      french_name : name@fr
    }
  }
}
{{< /runnable >}}


## Pagination

Pagination allows returning only a portion, rather than the whole, result set.  This can be useful for top-k style queries as well as to reduce the size of the result set for client side processing or to allow paged access to results.

Pagination is often used with [sorting]({{< relref "#sorting">}}).

{{% notice "note" %}}Without a sort order specified, the results are sorted by `uid`, which is assigned randomly. So the ordering, while deterministic, might not be what you expected.{{% /notice  %}}

### First

Syntax Examples:

* `q(func: ..., first: N)`
* `predicate (first: N) { ... }`
* `predicate @filter(...) (first: N) { ... }`

For positive `N`, `first: N` retrieves the first `N` results, by sorted or UID order.

For negative `N`, `first: N` retrieves the last `N` results, by sorted or UID order.  Currently, negative is only supported when no order is applied.  To achieve the effect of a negative with a sort, reverse the order of the sort and use a positive `N`.


Query Example: Last two films, by UID order, directed by Steven Spielberg and the first three genres of those movies, sorted alphabetically by English name.

{{< runnable >}}
{
  me(func: allofterms(name@en, "Steven Spielberg")) {
    director.film (first: -2) {
      name@en
      initial_release_date
      genre (orderasc: name@en) (first: 3) {
          name@en
      }
    }
  }
}
{{< /runnable >}}



Query Example: The three directors named Steven who have directed the most actors of all directors named Steven.

{{< runnable >}}
{
  ID as var(func: allofterms(name@en, "Steven")) @filter(has(director.film)) {
    director.film {
      stars as count(starring)
    }
    totalActors as sum(val(stars))
  }

  mostStars(func: uid(ID), orderdesc: val(totalActors), first: 3) {
    name@en
    stars : val(totalActors)

    director.film {
      name@en
    }
  }
}
{{< /runnable >}}

### Offset

Syntax Examples:

* `q(func: ..., offset: N)`
* `predicate (offset: N) { ... }`
* `predicate (first: M, offset: N) { ... }`
* `predicate @filter(...) (offset: N) { ... }`

With `offset: N` the first `N` results are not returned.  Used in combination with first, `first: M, offset: N` skips over `N` results and returns the following `M`.

Query Example: Order Hark Tsui's films by English title, skip over the first 4 and return the following 6.

{{< runnable >}}
{
  me(func: allofterms(name@en, "Hark Tsui")) {
    name@zh
    name@en
    director.film (orderasc: name@en) (first:6, offset:4)  {
      genre {
        name@en
      }
      name@zh
      name@en
      initial_release_date
    }
  }
}
{{< /runnable >}}

### After

Syntax Examples:

* `q(func: ..., after: UID)`
* `predicate (first: N, after: UID) { ... }`
* `predicate @filter(...) (first: N, after: UID) { ... }`

Another way to get results after skipping over some results is to use the default UID ordering and skip directly past a node specified by UID.  For example, a first query could be of the form `predicate (after: 0x0, first: N)`, or just `predicate (first: N)`, with subsequent queries of the form `predicate(after: <uid of last entity in last result>, first: N)`.


Query Example: The first five of Baz Luhrmann's films, sorted by UID order.

{{< runnable >}}
{
  me(func: allofterms(name@en, "Baz Luhrmann")) {
    name@en
    director.film (first:5) {
      uid
      name@en
    }
  }
}
{{< /runnable >}}

The fifth movie is the Australian movie classic Strictly Ballroom.  It has UID `0x99e44`.  The results after Strictly Ballroom can now be obtained with `after`.

{{< runnable >}}
{
  me(func: allofterms(name@en, "Baz Luhrmann")) {
    name@en
    director.film (first:5, after: 0x99e44) {
      uid
      name@en
    }
  }
}
{{< /runnable >}}


## Count

Syntax Examples:

* `count(predicate)`
* `count(uid)`

The form `count(predicate)` counts how many `predicate` edges lead out of a node.

The form `count(uid)` counts the number of UIDs matched in the enclosing block.

Query Example: The number of films acted in by each actor with `Orlando` in their name.

{{< runnable >}}
{
  me(func: allofterms(name@en, "Orlando")) @filter(has(actor.film)) {
    name@en
    count(actor.film)
  }
}
{{< /runnable >}}

Count can be used at root and [aliased]({{< relref "#alias">}}).

Query Example: Count of directors who have directed more than five films.  When used at the query root, the [count index]({{< relref "#count-index">}}) is required.

{{< runnable >}}
{
  directors(func: gt(count(director.film), 5)) {
    totalDirectors : count(uid)
  }
}
{{< /runnable >}}


Count can be assigned to a [value variable]({{< relref "#value-variables">}}).

Query Example: The actors of Ang Lee's "Eat Drink Man Woman" ordered by the number of movies acted in.

{{< runnable >}}
{
  var(func: allofterms(name@en, "eat drink man woman")) {
    starring {
      actors as performance.actor {
        totalRoles as count(actor.film)
      }
    }
  }

  edmw(func: uid(actors), orderdesc: val(totalRoles)) {
    name@en
    name@zh
    totalRoles : val(totalRoles)
  }
}
{{< /runnable >}}


## Sorting

Syntax Examples:

* `q(func: ..., orderasc: predicate)`
* `q(func: ..., orderdesc: val(varName))`
* `predicate (orderdesc: predicate) { ... }`
* `predicate @filter(...) (orderasc: N) { ... }`
* `q(func: ..., orderasc: predicate1, orderdesc: predicate2)`

Sortable Types: `int`, `float`, `String`, `dateTime`, `default`

Results can be sorted in ascending order (`orderasc`) or descending order (`orderdesc`) by a predicate or variable.

For sorting on predicates with [sortable indices]({{< relref "#sortable-indices">}}), Dgraph sorts on the values and with the index in parallel and returns whichever result is computed first.

Sorted queries retrieve up to 1000 results by default. This can be changed with [first]({{< relref "#first">}}).


Query Example: French director Jean-Pierre Jeunet's movies sorted by release date.

{{< runnable >}}
{
  me(func: allofterms(name@en, "Jean-Pierre Jeunet")) {
    name@fr
    director.film(orderasc: initial_release_date) {
      name@fr
      name@en
      initial_release_date
    }
  }
}
{{< /runnable >}}

Sorting can be performed at root and on value variables.

Query Example: All genres sorted alphabetically and the five movies in each genre with the most genres.

{{< runnable >}}
{
  genres as var(func: has(~genre)) {
    ~genre {
      numGenres as count(genre)
    }
  }

  genres(func: uid(genres), orderasc: name@en) {
    name@en
    ~genre (orderdesc: val(numGenres), first: 5) {
      name@en
      genres : val(numGenres)
    }
  }
}
{{< /runnable >}}

Sorting can also be performed by multiple predicates as shown below. If the values are equal for the
first predicate, then they are sorted by the second predicate and so on.

Query Example: Find all nodes which have type Person, sort them by their first_name and among those
that have the same first_name sort them by last_name in descending order.

```
{
  me(func: type("Person"), orderasc: first_name, orderdesc: last_name) {
    first_name
    last_name
  }
}
```

## Multiple Query Blocks

Inside a single query, multiple query blocks are allowed.  The result is all blocks with corresponding block names.

Multiple query blocks are executed in parallel.

The blocks need not be related in any way.

Query Example: All of Angelina Jolie's films, with genres, and Peter Jackson's films since 2008.

{{< runnable >}}
{
 AngelinaInfo(func:allofterms(name@en, "angelina jolie")) {
  name@en
   actor.film {
    performance.film {
      genre {
        name@en
      }
    }
   }
  }

 DirectorInfo(func: eq(name@en, "Peter Jackson")) {
    name@en
    director.film @filter(ge(initial_release_date, "2008"))  {
        Release_date: initial_release_date
        Name: name@en
    }
  }
}
{{< /runnable >}}


If queries contain some overlap in answers, the result sets are still independent.

Query Example: The movies Mackenzie Crook has acted in and the movies Jack Davenport has acted in.  The results sets overlap because both have acted in the Pirates of the Caribbean movies, but the results are independent and both contain the full answers sets.

{{< runnable >}}
{
  Mackenzie(func:allofterms(name@en, "Mackenzie Crook")) {
    name@en
    actor.film {
      performance.film {
        uid
        name@en
      }
      performance.character {
        name@en
      }
    }
  }

  Jack(func:allofterms(name@en, "Jack Davenport")) {
    name@en
    actor.film {
      performance.film {
        uid
        name@en
      }
      performance.character {
        name@en
      }
    }
  }
}
{{< /runnable >}}


### Var Blocks

Var blocks start with the keyword `var` and are not returned in the query results.

Query Example: Angelina Jolie's movies ordered by genre.

{{< runnable >}}
{
  var(func:allofterms(name@en, "angelina jolie")) {
    name@en
    actor.film {
      A AS performance.film {
        B AS genre
      }
    }
  }

  films(func: uid(B), orderasc: name@en) {
    name@en
    ~genre @filter(uid(A)) {
      name@en
    }
  }
}
{{< /runnable >}}


## Query Variables

Syntax Examples:

* `varName as q(func: ...) { ... }`
* `varName as var(func: ...) { ... }`
* `varName as predicate { ... }`
* `varName as predicate @filter(...) { ... }`

Types : `uid`

Nodes (UIDs) matched at one place in a query can be stored in a variable and used elsewhere.  Query variables can be used in other query blocks or in a child node of the defining block.

Query variables do not affect the semantics of the query at the point of definition.  Query variables are evaluated to all nodes matched by the defining block.

In general, query blocks are executed in parallel, but variables impose an evaluation order on some blocks.  Cycles induced by variable dependence are not permitted.

If a variable is defined, it must be used elsewhere in the query.

A query variable is used by extracting the UIDs in it with `uid(var-name)`.

The syntax `func: uid(A,B)` or `@filter(uid(A,B))` means the union of UIDs for variables `A` and `B`.

Query Example: The movies of Angelia Jolie and Brad Pitt where both have acted on movies in the same genre.  Note that `B` and `D` match all genres for all movies, not genres per movie.
{{< runnable >}}
{
 var(func:allofterms(name@en, "angelina jolie")) {
   actor.film {
    A AS performance.film {  # All films acted in by Angelina Jolie
     B As genre  # Genres of all the films acted in by Angelina Jolie
    }
   }
  }

 var(func:allofterms(name@en, "brad pitt")) {
   actor.film {
    C AS performance.film {  # All films acted in by Brad Pitt
     D as genre  # Genres of all the films acted in by Brad Pitt
    }
   }
  }

 films(func: uid(D)) @filter(uid(B)) {   # Genres from both Angelina and Brad
  name@en
   ~genre @filter(uid(A, C)) {  # Movies in either A or C.
     name@en
   }
 }
}
{{< /runnable >}}


## Value Variables

Syntax Examples:

* `varName as scalarPredicate`
* `varName as count(predicate)`
* `varName as avg(...)`
* `varName as math(...)`

Types : `int`, `float`, `String`, `dateTime`, `default`, `geo`, `bool`

Value variables store scalar values.  Value variables are a map from the UIDs of the enclosing block to the corresponding values.

It therefore only makes sense to use the values from a value variable in a context that matches the same UIDs - if used in a block matching different UIDs the value variable is undefined.

It is an error to define a value variable but not use it elsewhere in the query.

Value variables are used by extracting the values with `val(var-name)`, or by extracting the UIDs with `uid(var-name)`.

[Facet]({{< relref "#facets-edge-attributes">}}) values can be stored in value variables.

Query Example: The number of movie roles played by the actors of the 80's classic "The Princess Bride".  Query variable `pbActors` matches the UIDs of all actors from the movie.  Value variable `roles` is thus a map from actor UID to number of roles.  Value variable `roles` can be used in the `totalRoles` query block because that query block also matches the `pbActors` UIDs, so the actor to number of roles map is available.

{{< runnable >}}
{
  var(func:allofterms(name@en, "The Princess Bride")) {
    starring {
      pbActors as performance.actor {
        roles as count(actor.film)
      }
    }
  }
  totalRoles(func: uid(pbActors), orderasc: val(roles)) {
    name@en
    numRoles : val(roles)
  }
}
{{< /runnable >}}


Value variables can be used in place of UID variables by extracting the UID list from the map.

Query Example: The same query as the previous example, but using value variable `roles` for matching UIDs in the `totalRoles` query block.

{{< runnable >}}
{
  var(func:allofterms(name@en, "The Princess Bride")) {
    starring {
      performance.actor {
        roles as count(actor.film)
      }
    }
  }
  totalRoles(func: uid(roles), orderasc: val(roles)) {
    name@en
    numRoles : val(roles)
  }
}
{{< /runnable >}}


### Variable Propagation

Like query variables, value variables can be used in other query blocks and in blocks nested within the defining block.  When used in a block nested within the block that defines the variable, the value is computed as a sum of the variable for parent nodes along all paths to the point of use.  This is called variable propagation.

For example:
```
{
  q(func: uid(0x01)) {
    myscore as math(1)          # A
    friends {                   # B
      friends {                 # C
        ...myscore...
      }
    }
  }
}
```
At line A, a value variable `myscore` is defined as mapping node with UID `0x01` to value 1.  At B, the value for each friend is still 1: there is only one path to each friend.  Traversing the friend edge twice reaches the friends of friends. The variable `myscore` gets propagated such that each friend of friend will receive the sum of its parents values:  if a friend of a friend is reachable from only one friend, the value is still 1, if they are reachable from two friends, the value is two and so on.  That is, the value of `myscore` for each friend of friends inside the block marked C will be the number of paths to them.

**The value that a node receives for a propagated variable is the sum of the values of all its parent nodes.**

This propagation is useful, for example, in normalizing a sum across users, finding the number of paths between nodes and accumulating a sum through a graph.



Query Example: For each Harry Potter movie, the number of roles played by actor Warwick Davis.
{{< runnable >}}
{
    num_roles(func: eq(name@en, "Warwick Davis")) @cascade @normalize {

    paths as math(1)  # records number of paths to each character

    actor : name@en

    actor.film {
      performance.film @filter(allofterms(name@en, "Harry Potter")) {
        film_name : name@en
        characters : math(paths)  # how many paths (i.e. characters) reach this film
      }
    }
  }
}
{{< /runnable >}}


Query Example: Each actor who has been in a Peter Jackson movie and the fraction of Peter Jackson movies they have appeared in.
{{< runnable >}}
{
    movie_fraction(func:eq(name@en, "Peter Jackson")) @normalize {

    paths as math(1)
    total_films : num_films as count(director.film)
    director : name@en

    director.film {
      starring {
        performance.actor {
          fraction : math(paths / (num_films/paths))
          actor : name@en
        }
      }
    }
  }
}
{{< /runnable >}}

More examples can be found in two Dgraph blog posts about using variable propagation for recommendation engines ([post 1](https://open.dgraph.io/post/recommendation/), [post 2](https://open.dgraph.io/post/recommendation2/)).

## Aggregation

Syntax Example: `AG(val(varName))`

For `AG` replaced with

* `min` : select the minimum value in the value variable `varName`
* `max` : select the maximum value
* `sum` : sum all values in value variable `varName`
* `avg` : calculate the average of values in `varName`

Schema Types:

| Aggregation       | Schema Types |
|:-----------|:--------------|
| `min` / `max`     | `int`, `float`, `string`, `dateTime`, `default`         |
| `sum` / `avg`    | `int`, `float`       |

Aggregation can only be applied to [value variables]({{< relref "#value-variables">}}).  An index is not required (the values have already been found and stored in the value variable mapping).

An aggregation is applied at the query block enclosing the variable definition.  As opposed to query variables and value variables, which are global, aggregation is computed locally.  For example:
```
A as predicateA {
  ...
  B as predicateB {
    x as ...some value...
  }
  min(val(x))
}
```
Here, `A` and `B` are the lists of all UIDs that match these blocks.  Value variable `x` is a mapping from UIDs in `B` to values.  The aggregation `min(val(x))`, however, is computed for each UID in `A`.  That is, it has a semantics of: for each UID in `A`, take the slice of `x` that corresponds to `A`'s outgoing `predicateB` edges and compute the aggregation for those values.

Aggregations can themselves be assigned to value variables, making a UID to aggregation map.


### Min

#### Usage at Root

Query Example: Get the min initial release date for any Harry Potter movie.

The release date is assigned to a variable, then it is aggregated and fetched in an empty block.
{{< runnable >}}
{
  var(func: allofterms(name@en, "Harry Potter")) {
    d as initial_release_date
  }
  me() {
    min(val(d))
  }
}
{{< /runnable >}}

#### Usage at other levels

Query Example:  Directors called Steven and the date of release of their first movie, in ascending order of first movie.

{{< runnable >}}
{
  stevens as var(func: allofterms(name@en, "steven")) {
    director.film {
      ird as initial_release_date
      # ird is a value variable mapping a film UID to its release date
    }
    minIRD as min(val(ird))
    # minIRD is a value variable mapping a director UID to their first release date
  }

  byIRD(func: uid(stevens), orderasc: val(minIRD)) {
    name@en
    firstRelease: val(minIRD)
  }
}
{{< /runnable >}}

### Max

#### Usage at Root

Query Example: Get the max initial release date for any Harry Potter movie.

The release date is assigned to a variable, then it is aggregated and fetched in an empty block.
{{< runnable >}}
{
  var(func: allofterms(name@en, "Harry Potter")) {
    d as initial_release_date
  }
  me() {
    max(val(d))
  }
}
{{< /runnable >}}

#### Usage at other levels

Query Example: Quentin Tarantino's movies and date of release of the most recent movie.

{{< runnable >}}
{
  director(func: allofterms(name@en, "Quentin Tarantino")) {
    director.film {
      name@en
      x as initial_release_date
    }
    max(val(x))
  }
}
{{< /runnable >}}

### Sum and Avg

#### Usage at Root

Query Example: Get the sum and average of number of count of movies directed by people who have
Steven or Tom in their name.

{{< runnable >}}
{
  var(func: anyofterms(name@en, "Steven Tom")) {
    a as count(director.film)
  }

  me() {
    avg(val(a))
    sum(val(a))
  }
}
{{< /runnable >}}

#### Usage at other levels

Query Example: Steven Spielberg's movies, with the number of recorded genres per movie, and the total number of genres and average genres per movie.

{{< runnable >}}
{
  director(func: eq(name@en, "Steven Spielberg")) {
    name@en
    director.film {
      name@en
      numGenres : g as count(genre)
    }
    totalGenres : sum(val(g))
    genresPerMovie : avg(val(g))
  }
}
{{< /runnable >}}


### Aggregating Aggregates

Aggregations can be assigned to value variables, and so these variables can in turn be aggregated.

Query Example: For each actor in a Peter Jackson film, find the number of roles played in any movie.  Sum these to find the total number of roles ever played by all actors in the movie.  Then sum the lot to find the total number of roles ever played by actors who have appeared in Peter Jackson movies.  Note that this demonstrates how to aggregate aggregates; the answer in this case isn't quite precise though, because actors that have appeared in multiple Peter Jackson movies are counted more than once.

{{< runnable >}}
{
  PJ as var(func:allofterms(name@en, "Peter Jackson")) {
    director.film {
      starring {  # starring an actor
        performance.actor {
          movies as count(actor.film)
          # number of roles for this actor
        }
        perf_total as sum(val(movies))
      }
      movie_total as sum(val(perf_total))
      # total roles for all actors in this movie
    }
    gt as sum(val(movie_total))
  }

  PJmovies(func: uid(PJ)) {
    name@en
    director.film (orderdesc: val(movie_total), first: 5) {
      name@en
      totalRoles : val(movie_total)
    }
    grandTotal : val(gt)
  }
}
{{< /runnable >}}


## Math on value variables

Value variables can be combined using mathematical functions.  For example, this could be used to associate a score which is then used to order or perform other operations, such as might be used in building news feeds, simple recommendation systems, and so on.

Math statements must be enclosed within `math( <exp> )` and must be stored to a value variable.

The supported operators are as follows:

| Operators                       | Types accepted                                 | What it does                                                   |
| :------------:                  | :--------------:                               | :------------------------:                                     |
| `+` `-` `*` `/` `%`             | `int`, `float`                                     | performs the corresponding operation                           |
| `min` `max`                     | All types except `geo`, `bool`  (binary functions) | selects the min/max value among the two                        |
| `<` `>` `<=` `>=` `==` `!=`     | All types except `geo`, `bool`                     | Returns true or false based on the values                      |
| `floor` `ceil` `ln` `exp` `sqrt` | `int`, `float` (unary function)                    | performs the corresponding operation                           |
| `since`                         | `dateTime`                                 | Returns the number of seconds in float from the time specified |
| `pow(a, b)`                     | `int`, `float`                                     | Returns `a to the power b`                                     |
| `logbase(a,b)`                  | `int`, `float`                                     | Returns `log(a)` to the base `b`                               |
| `cond(a, b, c)`                 | first operand must be a boolean                | selects `b` if `a` is true else `c`                            |


Query Example:  Form a score for each of Steven Spielberg's movies as the sum of number of actors, number of genres and number of countries.  List the top five such movies in order of decreasing score.

{{< runnable >}}
{
	var(func:allofterms(name@en, "steven spielberg")) {
		films as director.film {
			p as count(starring)
			q as count(genre)
			r as count(country)
			score as math(p + q + r)
		}
	}

	TopMovies(func: uid(films), orderdesc: val(score), first: 5){
		name@en
		val(score)
	}
}
{{< /runnable >}}

Value variables and aggregations of them can be used in filters.

Query Example: Calculate a score for each Steven Spielberg movie with a condition on release date to penalize movies that are more than 10 years old, filtering on the resulting score.

{{< runnable >}}
{
  var(func:allofterms(name@en, "steven spielberg")) {
    films as director.film {
      p as count(starring)
      q as count(genre)
      date as initial_release_date
      years as math(since(date)/(365*24*60*60))
      score as math(cond(years > 10, 0, ln(p)+q-ln(years)))
    }
  }

  TopMovies(func: uid(films), orderdesc: val(score)) @filter(gt(val(score), 2)){
    name@en
    val(score)
    val(date)
  }
}
{{< /runnable >}}


Values calculated with math operations are stored to value variables and so can be aggregated.

Query Example: Compute a score for each Steven Spielberg movie and then aggregate the score.

{{< runnable >}}
{
	steven as var(func:eq(name@en, "Steven Spielberg")) @filter(has(director.film)) {
		director.film {
			p as count(starring)
			q as count(genre)
			r as count(country)
			score as math(p + q + r)
		}
		directorScore as sum(val(score))
	}

	score(func: uid(steven)){
		name@en
		val(directorScore)
	}
}
{{< /runnable >}}


## GroupBy

Syntax Examples:

* `q(func: ...) @groupby(predicate) { min(...) }`
* `predicate @groupby(pred) { count(uid) }``


A `groupby` query aggregates query results given a set of properties on which to group elements.  For example, a query containing the block `friend @groupby(age) { count(uid) }`, finds all nodes reachable along the friend edge, partitions these into groups based on age, then counts how many nodes are in each group.  The returned result is the grouped edges and the aggregations.

Inside a `groupby` block, only aggregations are allowed and `count` may only be applied to `uid`.

If the `groupby` is applied to a `uid` predicate, the resulting aggregations can be saved in a variable (mapping the grouped UIDs to aggregate values) and used elsewhere in the query to extract information other than the grouped or aggregated edges.

Query Example: For Steven Spielberg movies, count the number of movies in each genre and for each of those genres return the genre name and the count.  The name can't be extracted in the `groupby` because it is not an aggregate, but `uid(a)` can be used to extract the UIDs from the UID to value map and thus organize the `byGenre` query by genre UID.


{{< runnable >}}
{
  var(func:allofterms(name@en, "steven spielberg")) {
    director.film @groupby(genre) {
      a as count(uid)
      # a is a genre UID to count value variable
    }
  }

  byGenre(func: uid(a), orderdesc: val(a)) {
    name@en
    total_movies : val(a)
  }
}
{{< /runnable >}}

Query Example: Actors from Tim Burton movies and how many roles they have played in Tim Burton movies.
{{< runnable >}}
{
  var(func:allofterms(name@en, "Tim Burton")) {
    director.film {
      starring @groupby(performance.actor) {
        a as count(uid)
        # a is an actor UID to count value variable
      }
    }
  }

  byActor(func: uid(a), orderdesc: val(a)) {
    name@en
    val(a)
  }
}
{{< /runnable >}}



## Expand Predicates

The `expand()` function can be used to expand the predicates out of a node. To
 use `expand()`, the [type system]({{< relref "#type-system" >}}) is required.
Refer to the section on the type system to check how to set the types
nodes. The rest of this section assumes familiarity with that section.

There are four ways to use the `expand` function.

* Predicates can be stored in a variable and passed to `expand()` to expand all
  the predicates in the variable.
* If `_all_` is passed as an argument to `expand()`, the predicates to be
expanded will be the union of fields in the types assigned to a given node.

The `_all_` keyword requires that the nodes have types. Dgraph will look for all
the types that have been assigned to a node, query the types to check which
attributes they have, and use those to compute the list of predicates to expand.

For example, consider a node that has types `Animal` and `Pet`, which have 
the following definitions:

```
type Animal {
    name
    species
    dob
}

type Pet {
    owner
    veterinarian
}
```

When `expand(_all_)` is called on this node, Dgraph will first check which types
the node has (`Animal` and `Pet`). Then it will get the definitions of `Animal`
and `Pet` and build a list of predicates from their type definitions.

```
name
species
dob
owner
veterinarian
```

For `string` predicates, `expand` only returns values not tagged with a language
(see [language preference]({{< relref "#language-support" >}})).  So it's often 
required to add `name@fr` or `name@.` as well to an expand query.

### Filtering during expand.

Expand queries support filters on the type of the outgoing edge. For example,
`expand(_all_) @filter(type(Person))` will expand on all the predicates but will
only include edges whose destination node is of type Person. Since only nodes of
type `uid` can have a type, this query will filter out any scalar values.

Please note that other type of filters and directives are not currently supported
with the expand function. The filter needs to use the `type` function for the
filter to be allowed. Logical `AND` and `OR` operations are allowed. For
example, `expand(_all_) @filter(type(Person) OR type(Animal))` will only expand
the edges that point to nodes of either type.

## Cascade Directive

With the `@cascade` directive, nodes that don't have all predicates specified in the query are removed. This can be useful in cases where some filter was applied or if nodes might not have all listed predicates.


Query Example: Harry Potter movies, with each actor and characters played.  With `@cascade`, any character not played by an actor called Warwick is removed, as is any Harry Potter movie without any actors called Warwick.  Without `@cascade`, every character is returned, but only those played by actors called Warwick also have the actor name.
{{< runnable >}}
{
  HP(func: allofterms(name@en, "Harry Potter")) @cascade {
    name@en
    starring{
        performance.character {
          name@en
        }
        performance.actor @filter(allofterms(name@en, "Warwick")){
            name@en
         }
    }
  }
}
{{< /runnable >}}

You can apply `@cascade` on inner query blocks as well.
{{< runnable >}}
{
  HP(func: allofterms(name@en, "Harry Potter")) {
    name@en
    genre {
      name@en
    }
    starring @cascade {
        performance.character {
          name@en
        }
        performance.actor @filter(allofterms(name@en, "Warwick")){
            name@en
         }
    }
  }
}
{{< /runnable >}}

## Normalize directive

With the `@normalize` directive, only aliased predicates are returned and the result is flattened to remove nesting.

Query Example: Film name, country and first two actors (by UID order) of every Steven Spielberg movie, without `initial_release_date` because no alias is given and flattened by `@normalize`
{{< runnable >}}
{
  director(func:allofterms(name@en, "steven spielberg")) @normalize {
    director: name@en
    director.film {
      film: name@en
      initial_release_date
      starring(first: 2) {
        performance.actor {
          actor: name@en
        }
        performance.character {
          character: name@en
        }
      }
      country {
        country: name@en
      }
    }
  }
}
{{< /runnable >}}

You can also apply `@normalize` on nested query blocks. It will work similarly but only flatten the result of the nested query block where `@normalize` has been applied. `@normalize` will return a list irrespective of the type of attribute on which it is applied.
{{< runnable >}}
{
  director(func:allofterms(name@en, "steven spielberg")) {
    director: name@en
    director.film {
      film: name@en
      initial_release_date
      starring(first: 2) @normalize {
        performance.actor {
          actor: name@en
        }
        performance.character {
          character: name@en
        }
      }
      country {
        country: name@en
      }
    }
  }
}
{{< /runnable >}}


## Ignorereflex directive

The `@ignorereflex` directive forces the removal of child nodes that are reachable from themselves as a parent, through any path in the query result

Query Example: All the co-actors of Rutger Hauer.  Without `@ignorereflex`, the result would also include Rutger Hauer for every movie.

{{< runnable >}}
{
  coactors(func: eq(name@en, "Rutger Hauer")) @ignorereflex {
    actor.film {
      performance.film {
        starring {
          performance.actor {
            name@en
          }
        }
      }
    }
  }
}
{{< /runnable >}}

## Debug

For the purposes of debugging, you can attach a query parameter `debug=true` to a query. Attaching this parameter lets you retrieve the `uid` attribute for all the entities along with the `server_latency` and `start_ts` information under the `extensions` key of the response.

- `parsing_ns`: Latency in nanoseconds to parse the query.
- `processing_ns`: Latency in nanoseconds to process the query.
- `encoding_ns`: Latency in nanoseconds to encode the JSON response.
- `start_ts`: The logical start timestamp of the transaction.

Query with debug as a query parameter
```sh
curl -H "Content-Type: application/graphql+-" http://localhost:8080/query?debug=true -XPOST -d $'{
  tbl(func: allofterms(name@en, "The Big Lebowski")) {
    name@en
  }
}' | python -m json.tool | less
```

Returns `uid` and `server_latency`
```
{
  "data": {
    "tbl": [
      {
        "uid": "0x41434",
        "name@en": "The Big Lebowski"
      },
      {
        "uid": "0x145834",
        "name@en": "The Big Lebowski 2"
      },
      {
        "uid": "0x2c8a40",
        "name@en": "Jeffrey \"The Big\" Lebowski"
      },
      {
        "uid": "0x3454c4",
        "name@en": "The Big Lebowski"
      }
    ],
    "extensions": {
      "server_latency": {
        "parsing_ns": 18559,
        "processing_ns": 802990982,
        "encoding_ns": 1177565
      },
      "txn": {
        "start_ts": 40010
      }
    }
  }
}
```


## Schema

For each predicate, the schema specifies the target's type.  If a predicate `p` has type `T`, then for all subject-predicate-object triples `s p o` the object `o` is of schema type `T`.

* On mutations, scalar types are checked and an error thrown if the value cannot be converted to the schema type.

* On query, value results are returned according to the schema type of the predicate.

If a schema type isn't specified before a mutation adds triples for a predicate, then the type is inferred from the first mutation.  This type is either:

* type `uid`, if the first mutation for the predicate has nodes for the subject and object, or

* derived from the [RDF type]({{< relref "#rdf-types" >}}), if the object is a literal and an RDF type is present in the first mutation, or

* `default` type, otherwise.


### Schema Types

Dgraph supports scalar types and the UID type.

#### Scalar Types

For all triples with a predicate of scalar types the object is a literal.

| Dgraph Type | Go type |
| ------------|:--------|
|  `default`  | string  |
|  `int`      | int64   |
|  `float`    | float   |
|  `string`   | string  |
|  `bool`     | bool    |
|  `dateTime` | time.Time (RFC3339 format [Optional timezone] eg: 2006-01-02T15:04:05.999999999+10:00 or 2006-01-02T15:04:05.999999999)    |
|  `geo`      | [go-geom](https://github.com/twpayne/go-geom)    |
|  `password` | string (encrypted) |


{{% notice "note" %}}Dgraph supports date and time formats for `dateTime` scalar type only if they
are RFC 3339 compatible which is different from ISO 8601(as defined in the RDF spec). You should
convert your values to RFC 3339 format before sending them to Dgraph.{{% /notice  %}}

#### UID Type

The `uid` type denotes a node-node edge; internally each node is represented as a `uint64` id.

| Dgraph Type | Go type |
| ------------|:--------|
|  `uid`      | uint64  |


### Adding or Modifying Schema

Schema mutations add or modify schema.

Multiple scalar values can also be added for a `S P` by specifying the schema to be of
list type. Occupations in the example below can store a list of strings for each `S P`.

An index is specified with `@index`, with arguments to specify the tokenizer. When specifying an
index for a predicate it is mandatory to specify the type of the index. For example:

```
name: string @index(exact, fulltext) @count .
multiname: string @lang .
age: int @index(int) .
friend: [uid] @count .
dob: dateTime .
location: geo @index(geo) .
occupations: [string] @index(term) .
```

If no data has been stored for the predicates, a schema mutation sets up an empty schema ready to receive triples.

If data is already stored before the mutation, existing values are not checked to conform to the new schema.  On query, Dgraph tries to convert existing values to the new schema types, ignoring any that fail conversion.

If data exists and new indices are specified in a schema mutation, any index not in the updated list is dropped and a new index is created for every new tokenizer specified.

Reverse edges are also computed if specified by a schema mutation.


### Predicate name rules

Any alphanumeric combination of a predicate name is permitted.
Dgraph also supports [Internationalized Resource Identifiers](https://en.wikipedia.org/wiki/Internationalized_Resource_Identifier) (IRIs).
You can read more in [Predicates i18n](#predicates-i18n).

#### Allowed special characters

Single special characters are not accepted, which includes the special characters from IRIs.
They have to be prefixed/suffixed with alphanumeric characters.

```
][&*()_-+=!#$%
```

*Note: You are not restricted to use @ suffix, but the suffix character gets ignored.*

#### Forbidden special characters

The special characters below are not accepted.

```
^}|{`\~
```


### Predicates i18n

If your predicate is a URI or has language-specific characters, then enclose
it with angle brackets `<>` when executing the schema mutation.

{{% notice "note" %}}Dgraph supports [Internationalized Resource Identifiers](https://en.wikipedia.org/wiki/Internationalized_Resource_Identifier) (IRIs) for predicate names and values.{{% /notice  %}}

Schema syntax:
```
<职业>: string @index(exact) .
<年龄>: int @index(int) .
<地点>: geo @index(geo) .
<公司>: string .
```

This syntax allows for internationalized predicate names, but full-text indexing still defaults to English.
To use the right tokenizer for your language, you need to use the `@lang` directive and enter values using your
language tag.

Schema:
```
<公司>: string @index(fulltext) @lang .
```
Mutation:
```
{
  set {
    _:a <公司> "Dgraph Labs Inc"@en .
    _:b <公司> "夏新科技有限责任公司"@zh .
    _:a <dgraph.type> "Company" .
  }
}
```
Query:
```
{
  q(func: alloftext(<公司>@zh, "夏新科技有限责任公司")) {
    uid
    <公司>@.
  }
}
```


### Upsert directive

To use [upsert operations]({{< relref "howto/index.md#upserts">}}) on a
predicate, specify the `@upsert` directive in the schema. When committing
transactions involving predicates with the `@upsert` directive, Dgraph checks
index keys for conflicts, helping to enforce uniqueness constraints when running
concurrent upserts.

This is how you specify the upsert directive for a predicate.
```
email: string @index(exact) @upsert .
```

### RDF Types

Dgraph supports a number of [RDF types in mutations]({{< relref "mutations/index.md#language-and-rdf-types" >}}).

As well as implying a schema type for a [first mutation]({{< relref "#schema" >}}), an RDF type can override a schema type for storage.

If a predicate has a schema type and a mutation has an RDF type with a different underlying Dgraph type, the convertibility to schema type is checked, and an error is thrown if they are incompatible, but the value is stored in the RDF type's corresponding Dgraph type.  Query results are always returned in schema type.

For example, if no schema is set for the `age` predicate.  Given the mutation
```
{
 set {
  _:a <age> "15"^^<xs:int> .
  _:b <age> "13" .
  _:c <age> "14"^^<xs:string> .
  _:d <age> "14.5"^^<xs:string> .
  _:e <age> "14.5" .
 }
}
```
Dgraph:

* sets the schema type to `int`, as implied by the first triple,
* converts `"13"` to `int` on storage,
* checks `"14"` can be converted to `int`, but stores as `string`,
* throws an error for the remaining two triples, because `"14.5"` can't be converted to `int`.

### Extended Types

The following types are also accepted.

#### Password type

A password for an entity is set with setting the schema for the attribute to be of type `password`.  Passwords cannot be queried directly, only checked for a match using the `checkpwd` function.
The passwords are encrypted using [bcrypt](https://en.wikipedia.org/wiki/Bcrypt).

For example: to set a password, first set schema, then the password:
```
pass: password .
```

```
{
  set {
    <0x123> <name> "Password Example" .
    <0x123> <pass> "ThePassword" .
  }
}
```

to check a password:
```
{
  check(func: uid(0x123)) {
    name
    checkpwd(pass, "ThePassword")
  }
}
```

output:
```
{
  "data": {
    "check": [
      {
        "name": "Password Example",
        "checkpwd(pass)": true
      }
    ]
  }
}
```

You can also use alias with password type.

```
{
  check(func: uid(0x123)) {
    name
    secret: checkpwd(pass, "ThePassword")
  }
}
```

output:
```
{
  "data": {
    "check": [
      {
        "name": "Password Example",
        "secret": true
      }
    ]
  }
}
```

### Indexing

{{% notice "note" %}}Filtering on a predicate by applying a [function]({{< relref "#functions" >}}) requires an index.{{% /notice %}}

When filtering by applying a function, Dgraph uses the index to make the search through a potentially large dataset efficient.

All scalar types can be indexed.

Types `int`, `float`, `bool` and `geo` have only a default index each: with tokenizers named `int`, `float`, `bool` and `geo`.

Types `string` and `dateTime` have a number of indices.

#### String Indices
The indices available for strings are as follows.

| Dgraph function            | Required index / tokenizer             | Notes |
| :-----------------------   | :------------                          | :---  |
| `eq`                       | `hash`, `exact`, `term`, or `fulltext` | The most performant index for `eq` is `hash`. Only use `term` or `fulltext` if you also require term or full-text search. If you're already using `term`, there is no need to use `hash` or `exact` as well. |
| `le`, `ge`, `lt`, `gt`     | `exact`                                | Allows faster sorting.                                   |
| `allofterms`, `anyofterms` | `term`                                 | Allows searching by a term in a sentence.                |
| `alloftext`, `anyoftext`   | `fulltext`                             | Matching with language specific stemming and stopwords.  |
| `regexp`                   | `trigram`                              | Regular expression matching. Can also be used for equality checking. |

{{% notice "warning" %}}
Incorrect index choice can impose performance penalties and an increased
transaction conflict rate. Use only the minimum number of and simplest indexes
that your application needs.
{{% /notice %}}


#### DateTime Indices

The indices available for `dateTime` are as follows.

| Index name / Tokenizer   | Part of date indexed                                      |
| :----------- | :------------------------------------------------------------------ |
| `year`      | index on year (default)                                        |
| `month`       | index on year and month                                         |
| `day`       | index on year, month and day                                      |
| `hour`       | index on year, month, day and hour                               |

The choices of `dateTime` index allow selecting the precision of the index.  Applications, such as the movies examples in these docs, that require searching over dates but have relatively few nodes per year may prefer the `year` tokenizer; applications that are dependent on fine grained date searches, such as real-time sensor readings, may prefer the `hour` index.


All the `dateTime` indices are sortable.


#### Sortable Indices

Not all the indices establish a total order among the values that they index. Sortable indices allow inequality functions and sorting.

* Indexes `int` and `float` are sortable.
* `string` index `exact` is sortable.
* All `dateTime` indices are sortable.

For example, given an edge `name` of `string` type, to sort by `name` or perform inequality filtering on names, the `exact` index must have been specified.  In which case a schema query would return at least the following tokenizers.

```
{
  "predicate": "name",
  "type": "string",
  "index": true,
  "tokenizer": [
    "exact"
  ]
}
```

#### Count index

For predicates with the `@count` Dgraph indexes the number of edges out of each node.  This enables fast queries of the form:
```
{
  q(func: gt(count(pred), threshold)) {
    ...
  }
}
```

### List Type

Predicate with scalar types can also store a list of values if specified in the schema. The scalar
type needs to be enclosed within `[]` to indicate that its a list type. These lists are like an
unordered set.

```
occupations: [string] .
score: [int] .
```

* A set operation adds to the list of values. The order of the stored values is non-deterministic.
* A delete operation deletes the value from the list.
* Querying for these predicates would return the list in an array.
* Indexes can be applied on predicates which have a list type and you can use [Functions]({{<ref
  "#functions">}}) on them.
* Sorting is not allowed using these predicates.


### Reverse Edges

A graph edge is unidirectional. For node-node edges, sometimes modeling requires reverse edges.  If only some subject-predicate-object triples have a reverse, these must be manually added.  But if a predicate always has a reverse, Dgraph computes the reverse edges if `@reverse` is specified in the schema.

The reverse edge of `anEdge` is `~anEdge`.

For existing data, Dgraph computes all reverse edges.  For data added after the schema mutation, Dgraph computes and stores the reverse edge for each added triple.

### Querying Schema

A schema query queries for the whole schema:

```
schema {}
```

{{% notice "note" %}} Unlike regular queries, the schema query is not surrounded
by curly braces. Also, schema queries and regular queries cannot be combined.
{{% /notice %}}

You can query for particular schema fields in the query body.

```
schema {
  type
  index
  reverse
  tokenizer
  list
  count
  upsert
  lang
}
```

You can also query for particular predicates:

```
schema(pred: [name, friend]) {
  type
  index
  reverse
  tokenizer
  list
  count
  upsert
  lang
}
```

Types can also be queried. Below are some example queries.

```
schema(type: Movie) {}
schema(type: [Person, Animal]) {}
```

Note that type queries do not contain anything between the curly braces. The
output will be the entire definition of the requested types.

## Type System

Dgraph supports a type system that can be used to categorize nodes and query
them based on their type. The type system is also used during expand queries.

### Type definition

Types are defined using a GraphQL-like syntax. For example:

```
type Student {
  name
  dob
  home_address
  year
  friends
}
```

Types are declared along with the schema using the Alter endpoint. In order to
properly support the above type, a predicate for each of the attributes
in the type is also needed, such as:

```
name: string @index(term) .
dob: datetime .
home_address: string .
year: int .
friends: [uid] .
```

Reverse predicates can also be included inside a type definition. For example, the type above
could be expanded to include the parent of the student if there's a predicate `children` with
a reverse edge (the brackets around the predicate name are needed to properly understand the
special character `~`).

```
children: [uid] @reverse .

type Student {
  name
  dob
  home_address
  year
  friends
  <~children>
}
```

Edges can be used in multiple types: for example, `name` might be used for both
a person and a pet. Sometimes, however, it's required to use a different
predicate for each type to represent a similar concept. For example, if student
names and book names required different indexes, then the predicates must be
different.

```
type Student {
  student_name
}

type Textbook {
  textbook_name
}

student_name: string @index(exact) .
textbook_name: string @lang @index(fulltext) .
```

Altering the schema for a type that already exists, overwrites the existing
definition.

### Setting the type of a node

Scalar nodes cannot have types since they only have one attribute and its type
is the type of the node. UID nodes can have a type. The type is set by setting
the value of the `dgraph.type` predicate for that node. A node can have multiple
types. Here's an example of how to set the types of a node:

```
{
  set {
    _:a <name> "Garfield" .
    _:a <dgraph.type> "Pet" .
    _:a <dgraph.type> "Animal" .
  }
}
```

`dgraph.type` is a reserved predicate and cannot be removed or modified.

### Using types during queries

Types can be used as a top level function in the query language. For example:

```
{
  q(func: type(Animal)) {
    uid
    name
  }
}
```

This query will only return nodes whose type is set to `Animal`.

Types can also be used to filter results inside a query. For example:

```
{
  q(func: has(parent)) {
    uid
    parent @filter(type(Person)) {
      uid
      name
    }
  }
}
```

This query will return the nodes that have a parent predicate and only the
`parent`'s of type `Person`.

### Deleting a type

Type definitions can be deleted using the Alter endpoint. All that is needed is
to send an operation object with the field `DropOp` (or `drop_op` depending on
the client) to the enum value `TYPE` and the field 'DropValue' (or `drop_value`)
to the type that is meant to be deleted.

Below is an example deleting the type `Person` using the Go client:
```go
err := c.Alter(context.Background(), &api.Operation{
                DropOp: api.Operation_TYPE,
                DropValue: "Person"})
```

### Expand queries and types

Queries using [expand]({{< relref "#expand-predicates" >}}) (i.e.:
`expand(_all_)`) require that the nodes to be expanded have types.

## Facets : Edge attributes

Dgraph supports facets --- **key value pairs on edges** --- as an extension to RDF triples. That is, facets add properties to edges, rather than to nodes.
For example, a `friend` edge between two nodes may have a boolean property of `close` friendship.
Facets can also be used as `weights` for edges.

Though you may find yourself leaning towards facets many times, they should not be misused.  It wouldn't be correct modeling to give the `friend` edge a facet `date_of_birth`. That should be an edge for the friend.  However, a facet like `start_of_friendship` might be appropriate.  Facets are however not first class citizen in Dgraph like predicates.

Facet keys are strings and values can be `string`, `bool`, `int`, `float` and `dateTime`.
For `int` and `float`, only 32-bit signed integers and 64-bit floats are accepted.

The following mutation is used throughout this section on facets.  The mutation adds data for some peoples and, for example, records a `since` facet in `mobile` and `car` to record when Alice bought the car and started using the mobile number.

First we add some schema.
```sh
curl localhost:8080/alter -XPOST -d $'
    name: string @index(exact, term) .
    rated: [uid] @reverse @count .
' | python -m json.tool | less

```

```sh
curl -H "Content-Type: application/rdf" localhost:8080/mutate?commitNow=true -XPOST -d $'
{
  set {

    # -- Facets on scalar predicates
    _:alice <name> "Alice" .
    _:alice <dgraph.type> "Person" .
    _:alice <mobile> "040123456" (since=2006-01-02T15:04:05) .
    _:alice <car> "MA0123" (since=2006-02-02T13:01:09, first=true) .

    _:bob <name> "Bob" .
    _:bob <dgraph.type> "Person" .
    _:bob <car> "MA0134" (since=2006-02-02T13:01:09) .

    _:charlie <name> "Charlie" .
    _:charlie <dgraph.type> "Person" .
    _:dave <name> "Dave" .
    _:dave <dgraph.type> "Person" .


    # -- Facets on UID predicates
    _:alice <friend> _:bob (close=true, relative=false) .
    _:alice <friend> _:charlie (close=false, relative=true) .
    _:alice <friend> _:dave (close=true, relative=true) .


    # -- Facets for variable propagation
    _:movie1 <name> "Movie 1" .
    _:movie1 <dgraph.type> "Movie" .
    _:movie2 <name> "Movie 2" .
    _:movie2 <dgraph.type> "Movie" .
    _:movie3 <name> "Movie 3" .
    _:movie3 <dgraph.type> "Movie" .

    _:alice <rated> _:movie1 (rating=3) .
    _:alice <rated> _:movie2 (rating=2) .
    _:alice <rated> _:movie3 (rating=5) .

    _:bob <rated> _:movie1 (rating=5) .
    _:bob <rated> _:movie2 (rating=5) .
    _:bob <rated> _:movie3 (rating=5) .

    _:charlie <rated> _:movie1 (rating=2) .
    _:charlie <rated> _:movie2 (rating=5) .
    _:charlie <rated> _:movie3 (rating=1) .
  }
}' | python -m json.tool | less
```

### Facets on scalar predicates


Querying `name`, `mobile` and `car` of Alice gives the same result as without facets.

{{< runnable >}}
{
  data(func: eq(name, "Alice")) {
     name
     mobile
     car
  }
}
{{</ runnable >}}


The syntax `@facets(facet-name)` is used to query facet data. For Alice the `since` facet for `mobile` and `car` are queried as follows.

{{< runnable >}}
{
  data(func: eq(name, "Alice")) {
     name
     mobile @facets(since)
     car @facets(since)
  }
}
{{</ runnable >}}


Facets are returned at the same level as the corresponding edge and have keys like edge|facet.

All facets on an edge are queried with `@facets`.

{{< runnable >}}
{
  data(func: eq(name, "Alice")) {
     name
     mobile @facets
     car @facets
  }
}
{{</ runnable >}}

### Facets i18n

Facets keys and values can use language-specific characters directly when mutating. But facet keys need to be enclosed in angle brackets `<>` when querying. This is similar to predicates. See [Predicates i18n](#predicates-i18n) for more info.

{{% notice "note" %}}Dgraph supports [Internationalized Resource Identifiers](https://en.wikipedia.org/wiki/Internationalized_Resource_Identifier) (IRIs) for facet keys when querying.{{% /notice  %}}

Example:
```
{
  set {
    _:person1 <name> "Daniel" (वंश="स्पेनी", ancestry="Español") .
    _:person1 <dgraph.type> "Person" .
    _:person2 <name> "Raj" (वंश="हिंदी", ancestry="हिंदी") .
    _:person2 <dgraph.type> "Person" .
    _:person3 <name> "Zhang Wei" (वंश="चीनी", ancestry="中文") .
    _:person3 <dgraph.type> "Person" .
  }
}
```
Query, notice the `<>`'s:
```
{
  q(func: has(name)) {
    name @facets(<वंश>)
  }
}
```

### Alias with facets

Alias can be specified while requesting specific predicates. Syntax is similar to how would request
alias for other predicates. `orderasc` and `orderdesc` are not allowed as alias as they have special
meaning. Apart from that anything else can be set as alias.

Here we set `car_since`, `close_friend` alias for `since`, `close` facets respectively.
{{< runnable >}}
{
   data(func: eq(name, "Alice")) {
     name
     mobile
     car @facets(car_since: since)
     friend @facets(close_friend: close) {
       name
     }
   }
}
{{</ runnable >}}



### Facets on UID predicates

Facets on UID edges work similarly to facets on value edges.

For example, `friend` is an edge with facet `close`.
It was set to true for friendship between Alice and Bob
and false for friendship between Alice and Charlie.

A query for friends of Alice.

{{< runnable >}}
{
  data(func: eq(name, "Alice")) {
    name
    friend {
      name
    }
  }
}
{{</ runnable >}}

A query for friends and the facet `close` with `@facets(close)`.

{{< runnable >}}
{
   data(func: eq(name, "Alice")) {
     name
     friend @facets(close) {
       name
     }
   }
}
{{</ runnable >}}


For uid edges like `friend`, facets go to the corresponding child under the key edge|facet. In the above
example you can see that the `close` facet on the edge between Alice and Bob appears with the key `friend|close`
along with Bob's results.

{{< runnable >}}
{
  data(func: eq(name, "Alice")) {
    name
    friend @facets {
      name
      car @facets
    }
  }
}
{{</ runnable >}}

Bob has a `car` and it has a facet `since`, which, in the results, is part of the same object as Bob
under the key car|since.
Also, the `close` relationship between Bob and Alice is part of Bob's output object.
Charlie does not have `car` edge and thus only UID facets.

### Filtering on facets

Dgraph supports filtering edges based on facets.
Filtering works similarly to how it works on edges without facets and has the same available functions.


Find Alice's close friends
{{< runnable >}}
{
  data(func: eq(name, "Alice")) {
    friend @facets(eq(close, true)) {
      name
    }
  }
}
{{</ runnable >}}


To return facets as well as filter, add another `@facets(<facetname>)` to the query.

{{< runnable >}}
{
  data(func: eq(name, "Alice")) {
    friend @facets(eq(close, true)) @facets(relative) { # filter close friends and give relative status
      name
    }
  }
}
{{</ runnable >}}


Facet queries can be composed with `AND`, `OR` and `NOT`.

{{< runnable >}}
{
  data(func: eq(name, "Alice")) {
    friend @facets(eq(close, true) AND eq(relative, true)) @facets(relative) { # filter close friends in my relation
      name
    }
  }
}
{{</ runnable >}}


### Sorting using facets

Sorting is possible for a facet on a uid edge. Here we sort the movies rated by Alice, Bob and
Charlie by their `rating` which is a facet.

{{< runnable >}}
{
  me(func: anyofterms(name, "Alice Bob Charlie")) {
    name
    rated @facets(orderdesc: rating) {
      name
    }
  }
}
{{</ runnable >}}



### Assigning Facet values to a variable

Facets on UID edges can be stored in [value variables]({{< relref "#value-variables" >}}).  The variable is a map from the edge target to the facet value.

Alice's friends reported by variables for `close` and `relative`.
{{< runnable >}}
{
  var(func: eq(name, "Alice")) {
    friend @facets(a as close, b as relative)
  }

  friend(func: uid(a)) {
    name
    val(a)
  }

  relative(func: uid(b)) {
    name
    val(b)
  }
}
{{</ runnable >}}


### Facets and Variable Propagation

Facet values of `int` and `float` can be assigned to variables and thus the [values propagate]({{< relref "#variable-propagation" >}}).


Alice, Bob and Charlie each rated every movie.  A value variable on facet `rating` maps movies to ratings.  A query that reaches a movie through multiple paths sums the ratings on each path.  The following sums Alice, Bob and Charlie's ratings for the three movies.

{{<runnable >}}
{
  var(func: anyofterms(name, "Alice Bob Charlie")) {
    num_raters as math(1)
    rated @facets(r as rating) {
      total_rating as math(r) # sum of the 3 ratings
      average_rating as math(total_rating / num_raters)
    }
  }
  data(func: uid(total_rating)) {
    name
    val(total_rating)
    val(average_rating)
  }

}
{{</ runnable >}}



### Facets and Aggregation

Facet values assigned to value variables can be aggregated.

{{< runnable >}}
{
  data(func: eq(name, "Alice")) {
    name
    rated @facets(r as rating) {
      name
    }
    avg(val(r))
  }
}
{{</ runnable >}}


Note though that `r` is a map from movies to the sum of ratings on edges in the query reaching the movie.  Hence, the following does not correctly calculate the average ratings for Alice and Bob individually --- it calculates 2 times the average of both Alice and Bob's ratings.

{{< runnable >}}

{
  data(func: anyofterms(name, "Alice Bob")) {
    name
    rated @facets(r as rating) {
      name
    }
    avg(val(r))
  }
}
{{</ runnable >}}

Calculating the average ratings of users requires a variable that maps users to the sum of their ratings.

{{< runnable >}}

{
  var(func: has(rated)) {
    num_rated as math(1)
    rated @facets(r as rating) {
      avg_rating as math(r / num_rated)
    }
  }

  data(func: uid(avg_rating)) {
    name
    val(avg_rating)
  }
}
{{</ runnable >}}

## Shortest Path Queries

The shortest path between a source (`from`) node and destination (`to`) node can be found using the keyword `shortest` for the query block name. It requires the source node UID, destination node UID and the predicates (at least one) that have to be considered for traversal. A `shortest` query block returns the shortest path under `_path_` in the query response. The path can also be stored in a variable which is used in other query blocks.

**K-Shortest Path queries:** By default the shortest path is returned. With `numpaths: k`, and `k > 1`, the k-shortest paths are returned. Cyclical paths are pruned out from the result of k-shortest path query. With `depth: n`, the paths up to `n` depth away are returned.

{{% notice "note" %}}
- If no predicates are specified in the `shortest` block, no path can be fetched as no edge is traversed.
- If you're seeing queries take a long time, you can set a [gRPC deadline](https://grpc.io/blog/deadlines) to stop the query after a certain amount of time.
{{% /notice %}}

For example:

```sh
curl localhost:8080/alter -XPOST -d $'
    name: string @index(exact) .
' | python -m json.tool | less
```

```sh
curl -H "Content-Type: application/rdf" localhost:8080/mutate?commitNow=true -XPOST -d $'
{
  set {
    _:a <friend> _:b (weight=0.1) .
    _:b <friend> _:c (weight=0.2) .
    _:c <friend> _:d (weight=0.3) .
    _:a <friend> _:d (weight=1) .
    _:a <name> "Alice" .
    _:a <dgraph.type> "Person" .
    _:b <name> "Bob" .
    _:b <dgraph.type> "Person" .
    _:c <name> "Tom" .
    _:c <dgraph.type> "Person" .
    _:d <name> "Mallory" .
    _:d <dgraph.type> "Person" .
  }
}' | python -m json.tool | less
```

The shortest path between Alice and Mallory (assuming UIDs 0x2 and 0x5 respectively) can be found with query:

```sh
curl -H "Content-Type: application/graphql+-" localhost:8080/query -XPOST -d $'{
 path as shortest(from: 0x2, to: 0x5) {
  friend
 }
 path(func: uid(path)) {
   name
 }
}' | python -m json.tool | less
```

Which returns the following results. (Note, without considering the `weight` facet, each edges' weight is considered as 1)

```
{
  "data": {
    "path": [
      {
        "name": "Alice"
      },
      {
        "name": "Mallory"
      }
    ],
    "_path_": [
      {
        "uid": "0x2",
        "friend": [
          {
            "uid": "0x5"
          }
        ]
      }
    ]
  }
}
```

We can return more paths by specifying `numpaths`. Setting `numpaths: 2` returns the shortest two paths:

```sh
curl -H "Content-Type: application/graphql+-" localhost:8080/query -XPOST -d $'{

 A as var(func: eq(name, "Alice"))
 M as var(func: eq(name, "Mallory"))

 path as shortest(from: uid(A), to: uid(M), numpaths: 2) {
  friend
 }
 path(func: uid(path)) {
   name
 }
}' | python -m json.tool | less
```

{{% notice "note" %}}In the query above, instead of using UID literals, we query both people using var blocks and the `uid()` function. You can also combine it with [GraphQL Variables]({{< relref "#graphql-variables" >}}).{{% /notice %}}

Edges weights are included by using facets on the edges as follows.

{{% notice "note" %}}Only one facet per predicate is allowed in the shortest query block.{{% /notice %}}

```sh
curl -H "Content-Type: application/graphql+-" localhost:8080/query -XPOST -d $'{
 path as shortest(from: 0x2, to: 0x5) {
  friend @facets(weight)
 }

 path(func: uid(path)) {
  name
 }
}' | python -m json.tool | less
```

```
{
  "data": {
    "path": [
      {
        "name": "Alice"
      },
      {
        "name": "Bob"
      },
      {
        "name": "Tom"
      },
      {
        "name": "Mallory"
      }
    ],
    "_path_": [
      {
        "uid": "0x2",
        "friend": [
          {
            "uid": "0x3",
            "friend|weight": 0.1,
            "friend": [
              {
                "uid": "0x4",
                "friend|weight": 0.2,
                "friend": [
                  {
                    "uid": "0x5",
                    "friend|weight": 0.3
                  }
                ]
              }
            ]
          }
        ]
      }
    ]
  }
}
```

Constraints can be applied to the intermediate nodes as follows.

```sh
curl -H "Content-Type: application/graphql+-" localhost:8080/query -XPOST -d $'{
  path as shortest(from: 0x2, to: 0x5) {
    friend @filter(not eq(name, "Bob")) @facets(weight)
    relative @facets(liking)
  }

  relationship(func: uid(path)) {
    name
  }
}' | python -m json.tool | less
```

The k-shortest path algorithm (used when `numpaths` > 1) also accepts the arguments `minweight` and `maxweight`, which take a float as their value. When they are passed, only paths within the weight range `[minweight, maxweight]` will be considered as valid paths. This can be used, for example, to query the shortest paths that traverse between 2 and 4 nodes.

```sh
curl -H "Content-Type: application/graphql+-" localhost:8080/query -XPOST -d $'{
 path as shortest(from: 0x2, to: 0x5, numpaths: 2, minweight: 2, maxweight: 4) {
  friend
 }
 path(func: uid(path)) {
   name
 }
}' | python -m json.tool | less
```

Some points to keep in mind for shortest path queries:

- Weights must be non-negative. Dijkstra's algorithm is used to calculate the shortest paths.
- Only one facet per predicate in the shortest query block is allowed.
- Only one `shortest` path block is allowed per query. Only one `_path_` is returned in the result. For queries with `numpaths` > 1, `_path_` contains all the paths.
- Cyclical paths are not included in the result of k-shortest path query.
- For k-shortest paths (when `numpaths` > 1), the result of the shortest path query variable will only return a single path which will be the shortest path among the k paths. All k paths are returned in `_path_`.

## Recurse Query

`Recurse` queries let you traverse a set of predicates (with filter, facets, etc.) until we reach all leaf nodes or we reach the maximum depth which is specified by the `depth` parameter.

To get 10 movies from a genre that has more than 30000 films and then get two actors for those movies we'd do something as follows:
{{< runnable >}}
{
	me(func: gt(count(~genre), 30000), first: 1) @recurse(depth: 5, loop: true) {
		name@en
		~genre (first:10) @filter(gt(count(starring), 2))
		starring (first: 2)
		performance.actor
	}
}
{{< /runnable >}}
Some points to keep in mind while using recurse queries are:

- You can specify only one level of predicates after root. These would be traversed recursively. Both scalar and entity-nodes are treated similarly.
- Only one recurse block is advised per query.
- Be careful as the result size could explode quickly and an error would be returned if the result set gets too large. In such cases use more filters, limit results using pagination, or provide a depth parameter at root as shown in the example above.
- The `loop` parameter can be set to false, in which case paths which lead to a loop would be ignored
  while traversing.
- If not specified, the value of the `loop` parameter defaults to false.
- If the value of the `loop` parameter is false and depth is not specified, `depth` will default to `math.MaxUint64`, which means that the entire graph might be traversed until all the leaf nodes are reached.


## Fragments

`fragment` keyword allows you to define new fragments that can be referenced in a query, as per [GraphQL specification](https://facebook.github.io/graphql/#sec-Language.Fragments). The point is that if there are multiple parts which query the same set of fields, you can define a fragment and refer to it multiple times instead. Fragments can be nested inside fragments, but no cycles are allowed. Here is one contrived example.

```sh
curl -H "Content-Type: application/graphql+-" localhost:8080/query -XPOST -d $'
query {
  debug(func: uid(1)) {
    name@en
    ...TestFrag
  }
}
fragment TestFrag {
  initial_release_date
  ...TestFragB
}
fragment TestFragB {
  country
}' | python -m json.tool | less
```

## GraphQL Variables

`Variables` can be defined and used in queries which helps in query reuse and avoids costly string building in clients at runtime by passing a separate variable map. A variable starts with a `$` symbol.
For **HTTP requests** with GraphQL Variables, we must use `Content-Type: application/json` header and pass data with a JSON object containing `query` and `variables`.

```sh
curl -H "Content-Type: application/json" localhost:8080/query -XPOST -d $'{
  "query": "query test($a: string) { test(func: eq(name, $a)) { \n uid \n name \n } }",
  "variables": { "$a": "Alice" }
}' | python -m json.tool | less
```

{{< runnable vars="{\"$a\": \"5\", \"$b\": \"10\", \"$name\": \"Steven Spielberg\"}" >}}
query test($a: int, $b: int, $name: string) {
  me(func: allofterms(name@en, $name)) {
    name@en
    director.film (first: $a, offset: $b) {
      name @en
      genre(first: $a) {
        name@en
      }
    }
  }
}
{{< /runnable >}}

* Variables can have default values. In the example below, `$a` has a default value of `2`. Since the value for `$a` isn't provided in the variable map, `$a` takes on the default value.
* Variables whose type is suffixed with a `!` can't have a default value but must have a value as part of the variables map.
* The value of the variable must be parsable to the given type, if not, an error is thrown.
* The variable types that are supported as of now are: `int`, `float`, `bool` and `string`.
* Any variable that is being used must be declared in the named query clause in the beginning.

{{< runnable vars="{\"$b\": \"10\", \"$name\": \"Steven Spielberg\"}" >}}
query test($a: int = 2, $b: int!, $name: string) {
  me(func: allofterms(name@en, $name)) {
    director.film (first: $a, offset: $b) {
      genre(first: $a) {
        name@en
      }
    }
  }
}
{{< /runnable >}}

You can also use array with GraphQL Variables.

{{< runnable vars="{\"$b\": \"10\", \"$aName\": \"Steven Spielberg\", \"$bName\": \"Quentin Tarantino\"}" >}}
query test($a: int = 2, $b: int!, $aName: string, $bName: string) {
  me(func: eq(name@en, [$aName, $bName])) {
    director.film (first: $a, offset: $b) {
      genre(first: $a) {
        name@en
      }
    }
  }
}
{{< /runnable >}}

We also support variable substituion in facets now.
{{< runnable vars="{\"$name\": \"Alice\"}" >}}
query test($name: string = "Alice") {
  data(func: eq(name, $name)) {
    friend @facets(eq(close, true)) {
      name
    }
  }
}
{{</ runnable >}}

{{% notice "note" %}}
If you want to input a list of uids as a GraphQL variable value, you can have the variable as string type and
have the value surrounded by square brackets like `["13", "14"]`.
{{% /notice %}}

## Indexing with Custom Tokenizers

Dgraph comes with a large toolkit of builtin indexes, but sometimes for niche
use cases they're not always enough.

Dgraph allows you to implement custom tokenizers via a plugin system in order
to fill the gaps.

### Caveats

The plugin system uses Go's [`pkg/plugin`](https://golang.org/pkg/plugin/).
This brings some restrictions to how plugins can be used.

- Plugins must be written in Go.

- As of Go 1.9, `pkg/plugin` only works on Linux. Therefore, plugins will only
  work on Dgraph instances deployed in a Linux environment.

- The version of Go used to compile the plugin should be the same as the version
  of Go used to compile Dgraph itself. Dgraph always uses the latest version of
Go (and so should you!).

### Implementing a plugin

{{% notice "note" %}}
You should consider Go's [plugin](https://golang.org/pkg/plugin/) documentation
to be supplementary to the documentation provided here.
{{% /notice %}}

Plugins are implemented as their own main package. They must export a
particular symbol that allows Dgraph to hook into the custom logic the plugin
provides.

The plugin must export a symbol named `Tokenizer`. The type of the symbol must
be `func() interface{}`. When the function is called the result returned should
be a value that implements the following interface:

```
type PluginTokenizer interface {
    // Name is the name of the tokenizer. It should be unique among all
    // builtin tokenizers and other custom tokenizers. It identifies the
    // tokenizer when an index is set in the schema and when search/filter
    // is used in queries.
    Name() string

    // Identifier is a byte that uniquely identifiers the tokenizer.
    // Bytes in the range 0x80 to 0xff (inclusive) are reserved for
    // custom tokenizers.
    Identifier() byte

    // Type is a string representing the type of data that is to be
    // tokenized. This must match the schema type of the predicate
    // being indexed. Allowable values are shown in the table below.
    Type() string

    // Tokens should implement the tokenization logic. The input is
    // the value to be tokenized, and will always have a concrete type
    // corresponding to Type(). The return value should be a list of
    // the tokens generated.
    Tokens(interface{}) ([]string, error)
}
```

The return value of `Type()` corresponds to the concrete input type of
`Tokens(interface{})` in the following way:

 `Type()` return value | `Tokens(interface{})` input type
-----------------------|----------------------------------
 `"int"`               | `int64`
 `"float"`             | `float64`
 `"string"`            | `string`
 `"bool"`              | `bool`
 `"datetime"`          | `time.Time`

### Building the plugin

The plugin has to be built using the `plugin` build mode so that an `.so` file
is produced instead of a regular executable. For example:

```sh
go build -buildmode=plugin -o myplugin.so ~/go/src/myplugin/main.go
```

### Running Dgraph with plugins

When starting Dgraph, use the `--custom_tokenizers` flag to tell Dgraph which
tokenizers to load. It accepts a comma separated list of plugins. E.g.

```sh
dgraph ...other-args... --custom_tokenizers=plugin1.so,plugin2.so
```

{{% notice "note" %}}
Plugin validation is performed on startup. If a problem is detected, Dgraph
will refuse to initialise.
{{% /notice %}}

### Adding the index to the schema

To use a tokenization plugin, an index has to be created in the schema.

The syntax is the same as adding any built-in index. To add an custom index
using a tokenizer plugin named `foo` to a `string` predicate named
`my_predicate`, use the following in the schema:

```sh
my_predicate: string @index(foo) .
```

### Using the index in queries

There are two functions that can use custom indexes:

 Mode | Behaviour
--------|-------
 `anyof` | Returns nodes that match on *any* of the tokens generated
 `allof` | Returns nodes that match on *all* of the tokens generated

The functions can be used either at the query root or in filters.

There behaviour here an analogous to `anyofterms`/`allofterms` and
`anyoftext`/`alloftext`.

### Examples

The following examples should make the process of writing a tokenization plugin
more concrete.

#### Unicode Characters

This example shows the type of tokenization that is similar to term
tokenization of full-text search. Instead of being broken down into terms or
stem words, the text is instead broken down into its constituent unicode
codepoints (in Go terminology these are called *runes*).

{{% notice "note" %}}
This tokenizer would create a very large index that would be expensive to
manage and store. That's one of the reasons that text indexing usually occurs
at a higher level; stem words for full-text search or terms for term search.
{{% /notice %}}

The implementation of the plugin looks like this:

```go
package main

import "encoding/binary"

func Tokenizer() interface{} { return RuneTokenizer{} }

type RuneTokenizer struct{}

func (RuneTokenizer) Name() string     { return "rune" }
func (RuneTokenizer) Type() string     { return "string" }
func (RuneTokenizer) Identifier() byte { return 0xfd }

func (t RuneTokenizer) Tokens(value interface{}) ([]string, error) {
	var toks []string
	for _, r := range value.(string) {
		var buf [binary.MaxVarintLen32]byte
		n := binary.PutVarint(buf[:], int64(r))
		tok := string(buf[:n])
		toks = append(toks, tok)
	}
	return toks, nil
}
```

**Hints and tips:**

- Inside `Tokens`, you can assume that `value` will have concrete type
  corresponding to that specified by `Type()`. It's safe to do a type
assertion.

- Even though the return value is `[]string`, you can always store non-unicode
  data inside the string. See [this blogpost](https://blog.golang.org/strings)
for some interesting background how string are implemented in Go and why they
can be used to store non-textual data. By storing arbitrary data in the string,
you can make the index more compact. In this case, varints are stored in the
return values.

Setting up the indexing and adding data:
```
name: string @index(rune) .
```


```
{
  set{
    _:ad <name> "Adam" .
    _:ad <dgraph.type> "Person" .
    _:aa <name> "Aaron" .
    _:aa <dgraph.type> "Person" .
    _:am <name> "Amy" .
    _:am <dgraph.type> "Person" .
    _:ro <name> "Ronald" .
    _:ro <dgraph.type> "Person" .
  }
}
```
Now queries can be performed.

The only person that has all of the runes `A` and `n` in their `name` is Aaron:
```
{
  q(func: allof(name, rune, "An")) {
    name
  }
}
=>
{
  "data": {
    "q": [
      { "name": "Aaron" }
    ]
  }
}
```
But there are multiple people who have both of the runes `A` and `m`:
```
{
  q(func: allof(name, rune, "Am")) {
    name
  }
}
=>
{
  "data": {
    "q": [
      { "name": "Amy" },
      { "name": "Adam" }
    ]
  }
}
```
Case is taken into account, so if you search for all names containing `"ron"`,
you would find `"Aaron"`, but not `"Ronald"`. But if you were to search for
`"no"`, you would match both `"Aaron"` and `"Ronald"`. The order of the runes in
the strings doesn't matter.

It's possible to search for people that have *any* of the supplied runes in
their names (rather than *all* of the supplied runes). To do this, use `anyof`
instead of `allof`:
```
{
  q(func: anyof(name, rune, "mr")) {
    name
  }
}
=>
{
  "data": {
    "q": [
      { "name": "Adam" },
      { "name": "Aaron" },
      { "name": "Amy" }
    ]
  }
}
```
`"Ronald"` doesn't contain `m` or `r`, so isn't found by the search.

{{% notice "note" %}}
Understanding what's going on under the hood can help you intuitively
understand how `Tokens` method should be implemented.

When Dgraph sees new edges that are to be indexed by your tokenizer, it
will tokenize the value. The resultant tokens are used as keys for posting
lists. The edge subject is then added to the posting list for each token.

When a query root search occurs, the search value is tokenized. The result of
the search is all of the nodes in the union or intersection of the corresponding
posting lists (depending on whether `anyof` or `allof` was used).
{{% /notice %}}

#### CIDR Range

Tokenizers don't always have to be about splitting text up into its constituent
parts. This example indexes [IP addresses into their CIDR
ranges](https://en.wikipedia.org/wiki/Classless_Inter-Domain_Routing). This
allows you to search for all IP addresses that fall into a particular CIDR
range.

The plugin code is more complicated than the rune example. The input is an IP
address stored as a string, e.g. `"100.55.22.11/32"`. The output are the CIDR
ranges that the IP address could possibly fall into. There could be up to 32
different outputs (`"100.55.22.11/32"` does indeed have 32 possible ranges, one
for each mask size).

```go
package main

import "net"

func Tokenizer() interface{} { return CIDRTokenizer{} }

type CIDRTokenizer struct{}

func (CIDRTokenizer) Name() string     { return "cidr" }
func (CIDRTokenizer) Type() string     { return "string" }
func (CIDRTokenizer) Identifier() byte { return 0xff }

func (t CIDRTokenizer) Tokens(value interface{}) ([]string, error) {
	_, ipnet, err := net.ParseCIDR(value.(string))
	if err != nil {
		return nil, err
	}
	ones, bits := ipnet.Mask.Size()
	var toks []string
	for i := ones; i >= 1; i-- {
		m := net.CIDRMask(i, bits)
		tok := net.IPNet{
			IP:   ipnet.IP.Mask(m),
			Mask: m,
		}
		toks = append(toks, tok.String())
	}
	return toks, nil
}
```
An example of using the tokenizer:

Setting up the indexing and adding data:
```
ip: string @index(cidr) .

```

```
{
  set{
    _:a <ip> "100.55.22.11/32" .
    _:b <ip> "100.33.81.19/32" .
    _:c <ip> "100.49.21.25/32" .
    _:d <ip> "101.0.0.5/32" .
    _:e <ip> "100.176.2.1/32" .
  }
}
```
```
{
  q(func: allof(ip, cidr, "100.48.0.0/12")) {
    ip
  }
}
=>
{
  "data": {
    "q": [
      { "ip": "100.55.22.11/32" },
      { "ip": "100.49.21.25/32" }
    ]
  }
}
```
The CIDR ranges of `100.55.22.11/32` and `100.49.21.25/32` are both
`100.48.0.0/12`.  The other IP addresses in the database aren't included in the
search result, since they have different CIDR ranges for 12 bit masks
(`100.32.0.0/12`, `101.0.0.0/12`, `100.154.0.0/12` for `100.33.81.19/32`,
`101.0.0.5/32`, and `100.176.2.1/32` respectively).

Note that we're using `allof` instead of `anyof`. Only `allof` will work
correctly with this index. Remember that the tokenizer generates all possible
CIDR ranges for an IP address. If we were to use `anyof` then the search result
would include all IP addresses under the 1 bit mask (in this case, `0.0.0.0/1`,
which would match all IPs in this dataset).

#### Anagram

Tokenizers don't always have to return multiple tokens. If you just want to
index data into groups, have the tokenizer just return an identifying member of
that group.

In this example, we want to find groups of words that are
[anagrams](https://en.wikipedia.org/wiki/Anagram) of each
other.

A token to correspond to a group of anagrams could just be the letters in the
anagram in sorted order, as implemented below:

```go
package main

import "sort"

func Tokenizer() interface{} { return AnagramTokenizer{} }

type AnagramTokenizer struct{}

func (AnagramTokenizer) Name() string     { return "anagram" }
func (AnagramTokenizer) Type() string     { return "string" }
func (AnagramTokenizer) Identifier() byte { return 0xfc }

func (t AnagramTokenizer) Tokens(value interface{}) ([]string, error) {
	b := []byte(value.(string))
	sort.Slice(b, func(i, j int) bool { return b[i] < b[j] })
	return []string{string(b)}, nil
}
```
In action:

Setting up the indexing and adding data:
```
word: string @index(anagram) .
```

```
{
  set{
    _:1 <word> "airmen" .
    _:2 <word> "marine" .
    _:3 <word> "beat" .
    _:4 <word> "beta" .
    _:5 <word> "race" .
    _:6 <word> "care" .
  }
}
```
```
{
  q(func: allof(word, anagram, "remain")) {
    word
  }
}
=>
{
  "data": {
    "q": [
      { "word": "airmen" },
      { "word": "marine" }
    ]
  }
}
```

Since a single token is only ever generated, it doesn't matter if `anyof` or
`allof` is used. The result will always be the same.

#### Integer prime factors

All of the custom tokenizers shown previously have worked with strings.
However, other data types can be used as well. This example is contrived, but
nonetheless shows some advanced usages of custom tokenizers.

The tokenizer creates a token for each prime factor in the input.

```
package main

import (
    "encoding/binary"
    "fmt"
)

func Tokenizer() interface{} { return FactorTokenizer{} }

type FactorTokenizer struct{}

func (FactorTokenizer) Name() string     { return "factor" }
func (FactorTokenizer) Type() string     { return "int" }
func (FactorTokenizer) Identifier() byte { return 0xfe }

func (FactorTokenizer) Tokens(value interface{}) ([]string, error) {
    x := value.(int64)
    if x <= 1 {
        return nil, fmt.Errorf("Cannot factor int <= 1: %d", x)
    }
    var toks []string
    for p := int64(2); x > 1; p++ {
        if x%p == 0 {
            toks = append(toks, encodeInt(p))
            for x%p == 0 {
                x /= p
            }
        }
    }
    return toks, nil

}

func encodeInt(x int64) string {
    var buf [binary.MaxVarintLen64]byte
    n := binary.PutVarint(buf[:], x)
    return string(buf[:n])
}
```
{{% notice "note" %}}
Notice that the return of `Type()` is `"int"`, corresponding to the concrete
type of the input to `Tokens` (which is `int64`).
{{% /notice %}}

This allows you do things like search for all numbers that share prime
factors with a particular number.

In particular, we search for numbers that contain any of the prime factors of
15, i.e. any numbers that are divisible by either 3 or 5.

Setting up the indexing and adding data:
```
num: int @index(factor) .
```

```
{
  set{
    _:2 <num> "2"^^<xs:int> .
    _:3 <num> "3"^^<xs:int> .
    _:4 <num> "4"^^<xs:int> .
    _:5 <num> "5"^^<xs:int> .
    _:6 <num> "6"^^<xs:int> .
    _:7 <num> "7"^^<xs:int> .
    _:8 <num> "8"^^<xs:int> .
    _:9 <num> "9"^^<xs:int> .
    _:10 <num> "10"^^<xs:int> .
    _:11 <num> "11"^^<xs:int> .
    _:12 <num> "12"^^<xs:int> .
    _:13 <num> "13"^^<xs:int> .
    _:14 <num> "14"^^<xs:int> .
    _:15 <num> "15"^^<xs:int> .
    _:16 <num> "16"^^<xs:int> .
    _:17 <num> "17"^^<xs:int> .
    _:18 <num> "18"^^<xs:int> .
    _:19 <num> "19"^^<xs:int> .
    _:20 <num> "20"^^<xs:int> .
    _:21 <num> "21"^^<xs:int> .
    _:22 <num> "22"^^<xs:int> .
    _:23 <num> "23"^^<xs:int> .
    _:24 <num> "24"^^<xs:int> .
    _:25 <num> "25"^^<xs:int> .
    _:26 <num> "26"^^<xs:int> .
    _:27 <num> "27"^^<xs:int> .
    _:28 <num> "28"^^<xs:int> .
    _:29 <num> "29"^^<xs:int> .
    _:30 <num> "30"^^<xs:int> .
  }
}
```
```
{
  q(func: anyof(num, factor, 15)) {
    num
  }
}
=>
{
  "data": {
    "q": [
      { "num": 3 },
      { "num": 5 },
      { "num": 6 },
      { "num": 9 },
      { "num": 10 },
      { "num": 12 },
      { "num": 15 },
      { "num": 18 }
      { "num": 20 },
      { "num": 21 },
      { "num": 25 },
      { "num": 24 },
      { "num": 27 },
      { "num": 30 },
    ]
  }
}
```