Query Acceleration

Index

Overview

Indexing, or property indexing, is a technique used in Ultipa to accelerate the retrieval of nodes and edges with specific properties. By avoiding full graph scans, indexes enable the database to quickly locate relevant data. This is especially advantageous when working with large graphs.

Index Types

Ultipa supports single index on one property and composite index which involve multiple properties from a schema.

Showing Indexes

To retrieve node indexes in the current graph:

GQL
SHOW NODE INDEX

To retrieve edge indexes in the current graph:

GQL
SHOW EDGE INDEX

The information about indexes is organized into the _nodeIndex or _edgeIndex table with the following fields:

Creating an Index

You can create an index using the CREATE INDEX statement. Note that each property can only have one single index. The index creation runs as a job, you may run SHOW JOB <id?> afterward to verify the success of the creation.

System properties in Ultipa are inherently optimized for query performance and have built-in efficiencies. They do not support indexing.

Syntax
<create index statement> ::=
  "CREATE INDEX" <index name> "ON" < "NODE" | "EDGE" > <schema name>
  "(" <property index item> [ { "," <property index item> }... ] ")"

<property index item> ::=
  <property name> [ "(" <bytes> ")" ]

Details

• The <index name> must be unique among nodes and among edges, but a node index and an edge index may share the same name.
• For a single index, specifies one <property index item>; for a composite index, lists multiple <property index item>.
• If a specified property is of type string or text, you can specify the maximum number of bytes ^[1] (count from left) to be indexed for each value. If omitted, the default indexing length is 1024 bytes for string and 2048 bytes for text. Learn more about how this byte-length limitation affects queries.

^[1] In standard English text, most encodings (such as ASCII or UTF-8) use 1 byte per character. However, for non-English characters, the byte size may vary—for example, one Chinese character typically occupies 3 bytes.

To create single index named cBalance for the property balance of card nodes:

GQL
CREATE INDEX cBalance ON NODE card (balance)

To create single index named name for the property name (string type) of card nodes, restricting the indexed byte-length as 10:

GQL
CREATE INDEX name ON NODE card (name(10))

To create composite index named transAmountNotes for properties amount and notes (text type, restricting the indexed byte-length as 10) for transfer edges:

GQL
CREATE INDEX transAmountNotes ON EDGE transfer (amount, notes(10))

Dropping an Index

You can drop an index using the DROP NODE INDEX or DROP EDGE INDEX statement. Dropping an index does not affect the actual property values stored in shards.

NOTE

A property with an index cannot be dropped until the index is deleted.

To drop the node index cBalance:

GQL
DROP NODE INDEX cBalance

To drop the edge index transAmountNotes:

GQL
DROP EDGE INDEX transAmountNotes

Using Indexes

Applicable Queries

Indexes are automatically applied when the corresponding properties are used in the following types of queries. They are not effective in other types of queries.

1. Node retrieval using a single node pattern. For example,

GQL
CREATE INDEX user_age_index ON NODE user (age)

The user_age_index is effective in the following queries:

GQL
MATCH (n:user {age: 45}) RETURN n

GQL
MATCH (n) WHERE n.age > 45 RETURN n

In the second query, the node label is not specified, so user_age_index is only partially used during the search for user nodes.

2. Edge retrieval using a one-step path pattern. For example,

GQL
CREATE INDEX links_weight_index ON EDGE links (weight)

The links_weight_index is effective in the following query:

GQL
MATCH ()-[e:links WHERE e.weight = 2]->() RETURN e

The edge direction can be left (<-[]-), right (-[]->), or any (-[]-).

The following query does not use links_weight_index because it retrieves paths, not edges:

GQL
MATCH p = ()-[e:links WHERE e.weight = 2]->() RETURN p

3. Start node filtering in path patterns.

The above user_age_index is effective in the following query:

GQL
MATCH p = (n:user WHERE n.age > 45)-[]-()-[]-() RETURN p

It does not apply to the following query:

GQL
MATCH p = ()-[]-(n:user WHERE n.age > 45) RETURN p

Leftmost Prefix Rule

The order of properties in a composite index matters — queries that match the leftmost properties of the index (i.e., the first property or the first few properties in the defined order) will benefit from the index.

For example:

GQL
CREATE INDEX name_age ON NODE user (name(10),age)

• MATCH (u:user WHERE u.name = "Kavi" AND u.age > 20) uses the index.
• MATCH (u:user WHERE u.name = "Kavi") uses the index.
• MATCH (u:user WHERE u.age > 20) doesn't use the index.
• MATCH (u:user WHERE u.name = "Kavi" AND u.age > 20 AND u.grade = 7) uses the index, meanwhile it contains the filtering for the grade property which lacks an index.

String Byte-Length Limitation

When using indexes with string or text properties, ensure the byte-length of the string used in the filter does not exceed the defined limit when creating the index.

For example, an index Username is created for the name property of the user nodes with a 8-byte limitation:

GQL
CREATE INDEX Username ON NODE user (name(8))

The query below won't utilize the Username index as the specified string Aventurine exceeds the 8-byte limit:

GQL
MATCH (n:user {name: "Aventurine"})
RETURN n