Record IDs | Language Primitives

Note

SurrealDB does not eagerly convert a string into a record. An implicit r prefix or cast is required instead.

SurrealDB record IDs are composed of a table name and a record identifier separated by a : in between, allowing for a simple and consistent way to reference records across the database. Record IDs are used to uniquely identify records within a table, to query, update, and delete records, and serve as links from one record to another.

Record IDs can be constructed from a number of ways, including alphanumeric text, complex Unicode text and symbols, numbers, arrays, objects, built-in ID generation functions, and a function to generate an ID from values.

All of the following are examples of valid record IDs in SurrealQL.

company:surrealdb
company:w6xb3izpgvz4n0gow6q7
reaction:`🤪`
weather:['London', d'2025-02-14T01:52:50.375Z']

As all record IDs are unique, trying to create a new record with an existing record ID will return an error. To create a record or modify it if the ID already exists, use an UPSERT statement or an INSERT statement with an ON DUPLICATE KEY UPDATE clause.

Types of record IDs

Random IDs

When you create a record without specifying the full ID, a random identifier is assigned after the table name. This differs from the traditional default of auto-increment or serial IDs that many developers are used to.

CREATE company;

Output

[
	{
		id: company:ezs644u19mae2p68404j
	}
]

Record IDs can be generated with a number of built-in ID generation functions, which are cryptographically secure and suitable for dispersion across a distributed datastore. These include a 20 digit alphanumeric ID (the default), sequentially incrementing and temporally sortable ULID Record identifiers, and UUID version 7 Record identifiers.

The default ID is 20 characters drawn at random from 36 (a to z and 0 to 9), which gives about 1.3 × 10³¹ possible values. A collision between two IDs in the same table becomes as likely as not only after about 4 × 10¹⁵ IDs have been generated in it. If a collision does happen, the CREATE fails with Database record ... already exists and the existing record is left unchanged. Use ulid() when records should sort by creation time, and uuid() when the ID has to be a standard UUID for another system.

-- Generate a random record ID 20 characters in length
-- Charset: `abcdefghijklmnopqrstuvwxyz0123456789`
CREATE temperature:rand() SET time = time::now(), celsius = 37.5;
-- Identical to the above CREATE statement, because
-- :rand() is the default random ID format
CREATE temperature SET time = time::now(), celsius = 37.5;

-- Generate a ULID-based record ID
CREATE temperature:ulid() SET time = time::now(), celsius = 37.5;
-- Generate a UUIDv7-based record ID
CREATE temperature:uuid() SET time = time::now(), celsius = 37.5;

Text record IDs

Text record IDs can contain letters, numbers and _ characters.

CREATE company:surrealdb SET name = 'SurrealDB';
CREATE user_version_2025 SET name = 'Alucard';

To create a record ID with complex characters, use ` (backticks) around the table name and/or record identifier.

CREATE article:`8424486b-85b3-4448-ac8d-5d51083391c7` SET
    time = time::now(),
    author = person:tobie;

CREATE `Artykuł`:100 SET
    author = person:`Lech_Wałęsa`;

The parts of record IDs with complex characters will display enclosed by ` backticks.

SurrealDB 2.x displayed the identifier part of such a record ID between ⟨ and ⟩ instead, as in person:⟨Lech_Wałęsa⟩. Both forms are still accepted as input and refer to the same record, so person:⟨Lech_Wałęsa⟩ = person:`Lech_Wałęsa` is true.

Output

-------- Query --------

[
	{
		author: person:tobie,
		id: article:`8424486b-85b3-4448-ac8d-5d51083391c7`,
		time: d'2025-02-18T01:48:46.364Z'
	}
]

-------- Query --------

[
	{
		author: person:`Lech_Wałęsa`,
		id: `Artykuł`:100
	}
]

Numeric record IDs

If you create a record ID with a number as a string, it will be stored with ` backticks to differentiate it from a number.

CREATE article SET id = 10;
CREATE article SET id = "10";
CREATE article SET id = "article10";
SELECT VALUE id FROM article;

As the record ID article:10 is different from article:`10` , no errors are returned when creating and both records turn up in the output of the SELECT statement. Meanwhile, the article with the identifier article10 does not use backticks as there is no article10 number to differentiate it from.

Output

[
	article:10,
	article:`10`,
    article:article10
]

If a numeric value is specified without any decimal point suffix and is within the range -9223372036854775808 to 9223372036854775807 then the value will be parsed, stored, and treated as a 64-bit signed integer.

Any numeric numbers outside of the range of a signed 64-bit integer will be stored as a string.

/**[test]

[[test.results]]
value = "[{ celsius: 37.5f, id: temperature:17493, time: d'2025-10-03T01:09:50.155406Z' }]"

[[test.results]]
value = "[{ events: ['Galactic senate convenes', 'Mr. Bean still waits in a field'], id: year:⟨29878977097987987979232⟩ }]"

*/

CREATE temperature:17493 SET time = time::now(), celsius = 37.5;
CREATE year:29878977097987987979232 SET
    events = [
        "Galactic senate convenes",
        "Mr. Bean still waits in a field"
    ];

Output

-------- Query --------

[
	{
		celsius: 37.5f,
		id: temperature:17493,
		time: d'2025-02-17T06:21:08.911Z'
	}
]

-------- Query s--------

[
	{
		events: [
			'Galactic senate convenes',
			'Mr. Bean still waits in a field'
		],
		id: year:`29878977097987987979232`
	}
]

Array-based record IDs

Record IDs can be constructed out of arrays and even objects. This sort of record ID is most used when you have a field or two that will be used to look up records inside a record range, which is extremely performant. This is in contrast to using a WHERE clause to filter, which involves a table scan.

Records in SurrealDB can store arrays of values, including other nested arrays or objects within them. Different types of values can be stored within the same array, unless defined otherwise.

/**[test]

[[test.results]]
value = "[{ conditions: 'cloudy', id: weather:['London', d'2025-02-13T05:00:00Z'], temperature: 5.7f }]"

*/

CREATE weather:['London', d'2025-02-13T05:00:00Z'] SET
    temperature = 5.7,
    conditions = "cloudy";

Output

[
	{
		conditions: 'cloudy',
		id: weather:[
			'London',
			d'2025-02-13T05:00:00Z'
		],
		temperature: 5.7f
	}
]

Why record ranges are performant

The main reason why record ranges are so performant is simply because the database knows ahead of time in which area to look for records in a query, and therefore has a smaller "surface area" to work in.

This can be demonstrated by seeing what happens when a single record range query encompasses all of the records in a database. The example below creates 10,000 player records that have an array-based record ID that begins with 'mage', allowing them to be used in a record range query, as well as a field called class that is also 'mage', which will be used in a WHERE clause to compare performance.

Interestingly, in this case a record range query is only somewhat more performant. This is because both queries end up iterating over 10,000 records, with the only difference being that the query with a WHERE clause also checks to see if the value of the class field is equal to 'mage'.

FOR $_ IN 0..10000 {
    CREATE player:['mage', rand::id()] SET class = 'mage';
};

LET $_ = SELECT * FROM player:['mage', NONE]..['mage', ..];
LET $_ = SELECT * FROM player WHERE class = 'mage';

If the number of player records is extended to a larger number of classes, however, the difference in performance will be much larger. In this case the record range query is still only iterating a relatively small surface area of 10,000 records, while the second one has ten times this number to go through in addition to the WHERE clause on top.

FOR $_ IN 0..10000 {
  CREATE player:['mage', rand::id()] SET class = 'mage';
  CREATE player:['barbarian', rand::id()] SET class = 'barbarian';
  CREATE player:['rogue', rand::id()]     SET class = 'rogue';
  CREATE player:['bard', rand::id()]      SET class = 'bard';
  CREATE player:['sage', rand::id()]      SET class = 'sage';
  CREATE player:['psionic', rand::id()]   SET class = 'psionic';
  CREATE player:['thief', rand::id()]     SET class = 'thief';
  CREATE player:['paladin', rand::id()]   SET class = 'paladin';
  CREATE player:['ranger', rand::id()]    SET class = 'ranger';
  CREATE player:['cleric', rand::id()]    SET class = 'cleric';
};

LET $_ = SELECT * FROM player:['mage', NONE]..['mage', ..];
LET $_ = SELECT * FROM player WHERE class = 'mage';

IDs made with parameters and function calls

Parameters and function calls can be used inside array- and object-based record IDs in the same way as on standalone arrays and objects.

/**[test]

[[test.results]]
value = "NONE"

[[test.results]]
value = "[{ conditions: 'cloudy', id: weather:['Seoul', d'2025-10-03T01:11:38.204589Z'], temperature: -2.3f }]"
skip-datetime = true

[[test.results]]
value = "[{ conditions: 'cloudy', id: weather:['London', d'2025-10-03T01:11:38.204961Z'], temperature: 5.3f }]"
skip-datetime = true

*/

LET $now = time::now();

CREATE weather:['Seoul', $now] SET
    temperature = -2.3,
    conditions = "cloudy";

CREATE weather:['London', time::now()] SET
    temperature = 5.3,
    conditions = "cloudy";

To create a record that uses a parameter or function call as its entire record identifier, the type::record() function can be used. (Note: this function was known as type::thing() before SurrealDB 3.0)

/**[test]

[[test.results]]
value = "NONE"

[[test.results]]
value = "[{ city: 'London', id: weather:⟨2025-10-03T01:13:14.238633Z⟩ }]"
skip-record-id-key = true

*/

LET $now = time::now();

CREATE type::record("weather", $now) SET city = 'London';

Output

[
	{
		city: 'London',
		id: weather:`2025-02-18T02:30:08.563Z`
	}
]

Sort order of record IDs

Records in a table are stored in the order of their IDs, and a query that reads a whole table returns them in that order. ORDER BY id gives the same order. The key of an ID sorts first by its type and then by its value:

Numbers, by numeric value, so 10 comes after 9.
Strings, character by character, so the string `10` comes before `9` .
Arrays, element by element.
Objects.

CREATE k:1, k:2, k:10, k:`10`, k:`9`, k:abc, k:[1, 'a'], k:[0, 'z'], k:{ a: 1 };
SELECT VALUE id FROM k ORDER BY id;

Output

[k:1, k:2, k:10, k:`10`, k:`9`, k:abc, k:[0, 'z'], k:[1, 'a'], k:{ a: 1 }]

To make IDs sort by value, use numbers for numeric keys, rather than strings of digits. A ULID sorts by the time it was created. The same order decides which records a record range returns.

Record ranges

SurrealDB supports the ability to query a range of records, using the record ID. Record ID range queries retrieve records using the natural sorting order of the record IDs, making a table scan unnecessary. These range queries can be used to query a range of records in a timeseries context.

-- Select all person records with IDs between the given range
SELECT * FROM person:1..1000;

-- Select all records for a particular location, inclusive
SELECT * FROM temperature:['London', NONE]..=['London', ..];

-- Select London records before a given time
SELECT * FROM temperature:['London', NONE]..['London', '2022-08-29T08:09:31'];

-- Select London records from a given time onwards
SELECT * FROM temperature:['London', '2022-08-29T08:03:39']..=['London', ..];

-- Select all temperature records with IDs between the specified range
SELECT * FROM temperature:['London', '2022-08-29T08:03:39']..['London', '2022-08-29T08:09:31'];

A range compares whole IDs in their sort order, not each element on its own. A range with an open end therefore continues past the records for one location: temperature:['London', '2022-08-29T08:03:39'].. also returns every record for 'Munich', and temperature:..['London', '2022-08-29T08:09:31'] every record for 'Amsterdam', because those arrays sort after and before the London ones. To stay within one location, close the range with NONE or .. in the second element, as in the examples above. NONE sorts before every other value and .. after every other value.

The following example shows the difference in performance between a regular query that uses a WHERE clause and a record range scan.

FOR $num IN 0..=100000 {
  CREATE person SET id = $num, num = $num  
};

-- Assign the output to an unused parameter
-- to avoid excessive output
LET $_ = SELECT * FROM person WHERE num IN 0..=1000;
LET $_ = SELECT * FROM person:0..=1000;

Defining record IDs in a schema

The type name of a record ID is record, which by default allows any sort of record. This type can be set inside a DEFINE FIELD statement.

/**[test]

[[test.results]]
value = "NONE"

[[test.results]]
value = "NONE"

[[test.results]]
value = "[{ friends: [person:one, person:two], id: person:663uogu8gnw31irybeer, possessions: [book:one, house:one] }]"
skip-record-id-key = true

*/

DEFINE FIELD possessions ON TABLE person TYPE option<array<record>>;
DEFINE FIELD friends ON TABLE person TYPE option<array<record<person>>>;

CREATE person SET
    possessions = [ book:one, house:one],
    friends = [ person:one, person:two ];

Be sure to use just record instead of record<any>, as <any> here would imply actual records of a table called any.

/**[test]

[[test.results]]
value = "NONE"

[[test.results]]
match = "$error = "Couldn't coerce value for field `possessions` of `person:*`: Expected `none | array<record<any>>` but found `[book:one, house:one]`""
error = true

[[test.results]]
value = "[{ id: person:u6qd2t4ij2h45bkf2gk4, possessions: [any:one, any:two] }]"
skip-record-id-key = true

*/

DEFINE FIELD possessions ON TABLE person TYPE option<array<record<any>>>;

-- Won't work, 'book' and 'house' are not of table 'any'
CREATE person SET
    possessions = [ book:one, house:one ];

-- Actually expects this, which is probably
-- not what the DEFINE FIELD intended
CREATE person SET
    possessions = [ any:one, any:two ];

Limitations

A DEFINE FIELD statement for the id field cannot use the VALUE clause.

/**[test]

[[test.results]]
error = "Cannot use the `VALUE` keyword on the `id` field."

*/

DEFINE FIELD id ON user VALUE rand::int(1, 1000000000) READONLY;

Error output

'Cannot use the `VALUE` keyword on the `id` field.'

To achieve the desired behaviour, the id field can be set inside the statement to create the record.

/**[test]

[[test.results]]
value = "[{ id: user:639167349 }]"
skip-record-id-key = true

*/


CREATE user SET id = rand::int(1, 1000000000);

Changing a record's ID

A record's ID cannot be changed after the record is created. Setting id in an UPDATE, or in the CONTENT of an UPSERT, fails with Found person:new for the `id` field, but a specific record has been specified.

To move a record to a new ID, create a copy under the new ID and delete the original. Links to the record are not updated, so update the fields that point to it as well:

CREATE person:old SET name = 'Ann';
CREATE post:1 SET author = person:old;

CREATE person:new CONTENT (SELECT * OMIT id FROM ONLY person:old);
DELETE person:old;
UPDATE post SET author = person:new WHERE author = person:old;

SELECT author.name AS author_name FROM post:1;

Output

[{ author_name: 'Ann' }]

Such statements are best run inside a manual transaction so that no query sees the record half moved. Graph edges that point to the record have to be recreated in the same way, because an edge's in and out cannot be changed either: an UPDATE that sets them leaves them as they were, without an error. To see all the incoming and outgoing edges for a record, the ? operator can be used.

-- Destructuring operator can be used...
person:one.{
    out: ->?,
    in: <-?
}

-- Or a regular SELECT statement
SELECT 
	->? AS out, 
	<-? AS in
FROM ONLY person:one;

Learn more

Learn more about record IDs in this blogpost and on this youtube video.