CREATE INDEX

CREATE INDEX creates an in-memory index on a source, view, or materialized view.

In Materialize, indexes store query results in memory within a cluster, and keep these results incrementally updated as new data arrives. By making up-to-date results available in memory, indexes can help optimize query performance, both when serving results and maintaining resource-heavy operations like joins.

Usage patterns

Indexes on views vs. materialized views

In Materialize, both indexes on views and materialized views incrementally update the view results when Materialize ingests new data. Whereas materialized views persist the view results in durable storage and can be accessed across clusters, indexes on views compute and store view results in memory within a single cluster.

Maintaining a materialized view in durable storage has resource and latency costs that should be carefully considered depending on the main usage of the view, while maintaining an index has memory costs.

Some general guidelines for usage patterns include:

Indexes and query optimizations

You might want to create indexes when…

• You want to use non-primary keys (e.g. foreign keys) as a join condition. In this case, you could create an index on the columns in the join condition.
• You want to speed up searches filtering by literal values or expressions.

Specific instances where indexes can be useful to improve performance include:

• When used in ad-hoc queries.

• When used by multiple queries within the same cluster.

• When used to enable delta joins.

For more information, see Optimization.

Best practices

Before creating an index, consider the following:

• If you create stacked views (i.e., views that depend on other views) to reduce SQL complexity, we recommend that you create an index only on the view that will serve results, taking into account the expected data access patterns.

• Materialize can reuse indexes across queries that concurrently access the same data in memory, which reduces redundancy and resource utilization per query. In particular, this means that joins do not need to store data in memory multiple times.

• For queries that have no supporting indexes, Materialize uses the same mechanics used by indexes to optimize computations. However, since this underlying work is discarded after each query run, take into account the expected data access patterns to determine if you need to index or not.

Syntax

with_options

Details

Restrictions

• You can only reference the columns available in the SELECT list of the query that defines the view. For example, if your view was defined as SELECT a, b FROM src, you can only reference columns a and b, even if src contains additional columns.

• You cannot exclude any columns from being in the index’s “value” set. For example, if your view is defined as SELECT a, b FROM ..., all indexes will contain {a, b} as their values.

  If you want to create an index that only stores a subset of these columns, consider creating another materialized view that uses SELECT some_subset FROM this_view....

Structure

Indexes in Materialize have the following structure for each unique row:

((tuple of indexed expressions), (tuple of the row, i.e. stored columns))

Indexed expressions vs. stored columns

Automatically created indexes will use all columns as key expressions for the index, unless Materialize is provided or can infer a unique key for the source or view.

For instance, unique keys can be…

• Provided by the schema provided for the source, e.g. through the Confluent Schema Registry.
• Inferred when the query…
  • Concludes with a GROUP BY.
  • Uses sources or views that have a unique key without damaging this property. For example, joining a view with unique keys against a second, where the join constraint uses foreign keys.

When creating your own indexes, you can choose the indexed expressions.

Memory footprint

The in-memory sizes of indexes are proportional to the current size of the source or view they represent. The actual amount of memory required depends on several details related to the rate of compaction and the representation of the types of data in the source or view.

Creating an index may also force the first materialization of a view, which may cause Materialize to install a dataflow to determine and maintain the results of the view. This dataflow may have a memory footprint itself, in addition to that of the index.

Examples

Optimizing joins with indexes

You can optimize the performance of JOIN on two relations by ensuring their join keys are the key columns in an index.

CREATE MATERIALIZED VIEW active_customers AS
    SELECT guid, geo_id, last_active_on
    FROM customer_source
    WHERE last_active_on > now() - INTERVAL '30' DAYS;

CREATE INDEX active_customers_geo_idx ON active_customers (geo_id);

CREATE MATERIALIZED VIEW active_customer_per_geo AS
    SELECT geo.name, count(*)
    FROM geo_regions AS geo
    JOIN active_customers ON active_customers.geo_id = geo.id
    GROUP BY geo.name;

In the above example, the index active_customers_geo_idx…

• Helps us because it contains a key that the view active_customer_per_geo can use to look up values for the join condition (active_customers.geo_id).

  Because this index is exactly what the query requires, the Materialize optimizer will choose to use active_customers_geo_idx rather than build and maintain a private copy of the index just for this query.

• Obeys our restrictions by containing only a subset of columns in the result set.

Speed up filtering with indexes

If you commonly filter by a certain column being equal to a literal value, you can set up an index over that column to speed up your queries:

CREATE MATERIALIZED VIEW active_customers AS
    SELECT guid, geo_id, last_active_on
    FROM customer_source
    GROUP BY geo_id;

CREATE INDEX active_customers_idx ON active_customers (guid);

-- This should now be very fast!
SELECT * FROM active_customers WHERE guid = 'd868a5bf-2430-461d-a665-40418b1125e7';

-- Using indexed expressions:
CREATE INDEX active_customers_exp_idx ON active_customers (upper(guid));
SELECT * FROM active_customers WHERE upper(guid) = 'D868A5BF-2430-461D-A665-40418B1125E7';

-- Filter using an expression in one field and a literal in another field:
CREATE INDEX active_customers_exp_field_idx ON active_customers (upper(guid), geo_id);
SELECT * FROM active_customers WHERE upper(guid) = 'D868A5BF-2430-461D-A665-40418B1125E7' and geo_id = 'ID_8482';

Create an index with an expression to improve query performance over a frequently used expression, and avoid building downstream views to apply the function like the one used in the example: upper(). Take into account that aggregations like count() cannot be used as indexed expressions.

For more details on using indexes to optimize queries, see Optimization.

Privileges

The privileges required to execute this statement are:

• Ownership of obj_name.
• CREATE privileges on the containing schema.
• CREATE privileges on the containing cluster.
• USAGE privileges on all types used in the index definition.
• USAGE privileges on the schemas that all types in the statement are contained in.

Related pages

• SHOW INDEXES
• DROP INDEX