Create materialized views

This document describes how to create materialized views in BigQuery. Before you read this document, familiarize yourself with Introduction to materialized views.

Before you begin

Grant Identity and Access Management (IAM) roles that give users the necessary permissions to perform each task in this document.

Required permissions

To create materialized views, you need the bigquery.tables.create IAM permission.

Each of the following predefined IAM roles includes the permissions that you need in order to create a materialized view:

• bigquery.dataEditor
• bigquery.dataOwner
• bigquery.admin

For more information about BigQuery Identity and Access Management (IAM), see Access control with IAM.

Create materialized views

To create a materialized view, select one of the following options:

CREATE MATERIALIZED VIEW myproject.mydataset.my_mv_table AS (
  SELECT
    product_id,
    SUM(clicks) AS sum_clicks
  FROM
    myproject.mydataset.my_base_table
  GROUP BY
    product_id
);

resource "google_bigquery_dataset" "default" {
  dataset_id                      = "mydataset"
  default_partition_expiration_ms = 2592000000  # 30 days
  default_table_expiration_ms     = 31536000000 # 365 days
  description                     = "dataset description"
  location                        = "US"
  max_time_travel_hours           = 96 # 4 days

  labels = {
    billing_group = "accounting",
    pii           = "sensitive"
  }
}

resource "google_bigquery_table" "default" {
  dataset_id          = google_bigquery_dataset.default.dataset_id
  table_id            = "my_materialized_view"
  deletion_protection = false # set to "true" in production

  materialized_view {
    query                            = "SELECT ID, description, date_created FROM `myproject.orders.items`"
    enable_refresh                   = "true"
    refresh_interval_ms              = 172800000 # 2 days
    allow_non_incremental_definition = "false"
  }

}

{
  "kind": "bigquery#table",
  "tableReference": {
    "projectId": "PROJECT_ID",
    "datasetId": "DATASET",
    "tableId": "MATERIALIZED_VIEW_NAME"
  },
  "materializedView": {
    "query": "QUERY_EXPRESSION"
  }
}

{
  "kind": "bigquery#table",
  "tableReference": {
    "projectId": "myproject",
    "datasetId": "mydataset",
    "tableId": "my_mv"
  },
  "materializedView": {
    "query": "select product_id,sum(clicks) as
                sum_clicks from myproject.mydataset.my_source_table
                group by 1"
  }
}

import com.google.cloud.bigquery.BigQuery;
import com.google.cloud.bigquery.BigQueryException;
import com.google.cloud.bigquery.BigQueryOptions;
import com.google.cloud.bigquery.MaterializedViewDefinition;
import com.google.cloud.bigquery.TableId;
import com.google.cloud.bigquery.TableInfo;

// Sample to create materialized view
public class CreateMaterializedView {

  public static void main(String[] args) {
    // TODO(developer): Replace these variables before running the sample.
    String datasetName = "MY_DATASET_NAME";
    String tableName = "MY_TABLE_NAME";
    String materializedViewName = "MY_MATERIALIZED_VIEW_NAME";
    String query =
        String.format(
            "SELECT MAX(TimestampField) AS TimestampField, StringField, "
                + "MAX(BooleanField) AS BooleanField "
                + "FROM %s.%s GROUP BY StringField",
            datasetName, tableName);
    createMaterializedView(datasetName, materializedViewName, query);
  }

  public static void createMaterializedView(
      String datasetName, String materializedViewName, String query) {
    try {
      // Initialize client that will be used to send requests. This client only needs to be created
      // once, and can be reused for multiple requests.
      BigQuery bigquery = BigQueryOptions.getDefaultInstance().getService();

      TableId tableId = TableId.of(datasetName, materializedViewName);

      MaterializedViewDefinition materializedViewDefinition =
          MaterializedViewDefinition.newBuilder(query).build();

      bigquery.create(TableInfo.of(tableId, materializedViewDefinition));
      System.out.println("Materialized view created successfully");
    } catch (BigQueryException e) {
      System.out.println("Materialized view was not created. \n" + e.toString());
    }
  }
}

After the materialized view is successfully created, it appears in the Explorer panel of BigQuery in the Google Cloud console. The following example shows a materialized view schema:

Unless you disable automatic refresh, BigQuery starts an asynchronous full refresh for the materialized view. The query finishes quickly, but the initial refresh might continue to run.

Access control

You can grant access to a materialized view at the dataset level, the view level, or the column level. You can also set access at a higher level in the IAM resource hierarchy.

Querying a materialized view requires access to the view as well as its base tables. To share a materialized view, you can grant permissions to the base tables or configure a materialized view as an authorized view. For more information, see Authorized views.

To control access to views in BigQuery, see Authorized views.

Materialized views query support

Materialized views use a restricted SQL syntax. Queries must use the following pattern:

[ WITH cte [, …]]
SELECT  [{ ALL | DISTINCT }]
  expression [ [ AS ] alias ] [, ...]
FROM from_item [, ...]
[ WHERE bool_expression ]
[ GROUP BY expression [, ...] ]

from_item:
    {
      table_name [ as_alias ]
      | { join_operation | ( join_operation ) }
      | field_path
      | unnest_operator
      | cte_name [ as_alias ]
    }

as_alias:
    [ AS ] alias

Query limitations

Materialized views have the following limitations.

Aggregate requirements

Aggregates in the materialized view query must be outputs. Computing, filtering, or joining based on an aggregated value is not supported. For example, creating a view from the following query is not supported because it produces a value computed from an aggregate, COUNT(*) / 10 as cnt.

SELECT TIMESTAMP_TRUNC(ts, HOUR) AS ts_hour, COUNT(*) / 10 AS cnt
FROM mydataset.mytable
GROUP BY ts_hour;

Only the following aggregation functions are currently supported:

• ANY_VALUE (but not over STRUCT)
• APPROX_COUNT_DISTINCT
• ARRAY_AGG (but not over ARRAY or STRUCT)
• AVG
• BIT_AND
• BIT_OR
• BIT_XOR
• COUNT
• COUNTIF
• HLL_COUNT.INIT
• LOGICAL_AND
• LOGICAL_OR
• MAX
• MIN
• MAX_BY (but not over STRUCT)
• MIN_BY (but not over STRUCT)
• SUM

Unsupported SQL features

The following SQL features are not supported in materialized views:

• UNION ALL. (Support in science Preview)
• LEFT OUTER JOIN (Support in science Preview)
• RIGHT/FULL OUTER JOIN.
• Self-joins, also known as using a JOIN on the same table more than once.
• Window functions.
• ARRAY subqueries.
• Non-deterministic functions such as RAND(), CURRENT_DATE(), SESSION_USER(), or CURRENT_TIME().
• User-defined functions (UDFs).
• TABLESAMPLE.
• FOR SYSTEM_TIME AS OF.

LEFT OUTER JOIN and UNION ALL support

To request feedback or support for this feature, send an email to bq-mv-help @google.com.

Incremental materialized views support LEFT OUTER JOIN and UNION ALL. Materialized views with LEFT OUTER JOIN and UNION ALL statements share the limitations of other incremental materialized views. In addition, smart tuning is not supported for materialized views with union all or left outer join.

Examples

The following example creates an aggregate incremental materialized view with a LEFT JOIN. This view is incrementally updated when data appends to the left table.

CREATE MATERIALIZED VIEW dataset.mv
AS (
  SELECT
    s_store_sk,
    s_country,
    s_zip,
    SUM(ss_net_paid) AS sum_sales,
  FROM dataset.store_sales
  LEFT JOIN dataset.store
    ON ss_store_sk = s_store_sk
  GROUP BY 1, 2, 3
);

The following example creates an aggregate incremental materialized view with a UNION ALL. This view is incrementally updated when data appends to either or both tables. For more information about incremental updates, see Incremental Updates.

CREATE MATERIALIZED VIEW dataset.mv PARTITION BY DATE(ts_hour)
AS (
  SELECT
    SELECT TIMESTAMP_TRUNC(ts, HOUR) AS ts_hour, SUM(sales) sum_sales
  FROM
    (SELECT ts, sales from dataset.table1 UNION ALL
     SELECT ts, sales from dataset.table2)
  GROUP BY 1
);

Access control restrictions

• If a user's query of a materialized view includes base table columns that they cannot access due to column-level security, then the query fails with the message Access Denied.
• If a user queries a materialized view but doesn't have full access to all rows in the materialized views' base tables, then BigQuery runs the query against the base tables instead of reading materialized view data. This ensures the query respects all access control constraints. This limitation also applies when querying tables with data-masked columns.

WITH clause and common table expressions (CTEs)

Materialized views support WITH clauses and common table expressions. Materialized views with WITH clauses must still follow the pattern and limitations of materialized views without WITH clauses.

Examples

The following example shows a materialized view using a WITH clause:

WITH tmp AS (
  SELECT TIMESTAMP_TRUNC(ts, HOUR) AS ts_hour, *
  FROM mydataset.mytable
)
SELECT ts_hour, COUNT(*) AS cnt
FROM tmp
GROUP BY ts_hour;

The following example shows a materialized view using a WITH clause that is not supported because it contains two GROUP BY clauses:

WITH tmp AS (
  SELECT city, COUNT(*) AS population
  FROM mydataset.mytable
  GROUP BY city
)
SELECT population, COUNT(*) AS cnt
GROUP BY population;

Materialized views over BigLake tables

To create materialized views over BigLake tables, the BigLake table must have metadata caching enabled over Cloud Storage data and the materialized view must have a max_staleness option value greater than the base table. Materialized views over BigLake tables support the same set of queries as other materialized views.

Example

Creation of a simple aggregate view using a BigLake base table:

CREATE MATERIALIZED VIEW sample_dataset.sample_mv
    OPTIONS (max_staleness=INTERVAL "0:30:0" HOUR TO SECOND)
AS SELECT COUNT(*) cnt
FROM dataset.biglake_base_table;

For details about the limitations of materialized views over BigLake tables, see materialized views over BigLake tables.

Materialized views over Apache Iceberg tables

To request feedback or support for this feature, send an email to [email protected].

You can reference large Iceberg tables in materialized views instead of migrating that data to BigQuery-managed storage.

Create a materialized view over an Iceberg table

To create a materialized view over an Iceberg, follow these steps:

1. Obtain an Iceberg table using one of the following methods:

   • Create an Iceberg table with the JSON metadata file.
   • Create an Iceberg table using the BigLake Metastore.
   • Discover in AWS Glue federated datasets.

   Example

   CREATE EXTERNAL TABLE mydataset.myicebergtable
     WITH CONNECTION `myproject.us.myconnection`
     OPTIONS (
           format = 'ICEBERG',
           uris = ["gs://mybucket/mydata/mytable/metadata/iceberg.metadata.json"]
     )
   

2. Reference your Iceberg table with the following partition-specifications:

   "partition-specs" : [ {
      "spec-id" : 0,
      "fields" : [ {
       "name" : "birth_month",
       "transform" : "month",
       "source-id" : 3,
       "field-id" : 1000
   } ]
   

3. Create a partition-aligned materialized view:

   CREATE MATERIALIZED VIEW mydataset.myicebergmv
     PARTITION BY DATE_TRUNC(birth_month, MONTH)
   AS
     SELECT * FROM mydataset.myicebergtable;
   

Limitations

In addition to the limitations of standard Iceberg tables, materialized views over Iceberg tables have the following limitations:

• You can create a materialized view that is partition aligned with the base table. However, the materialized view only supports time-based partition transformation, for example, YEAR, MONTH, DAY, and HOUR.
• The granularity of the materialized view's partition cannot be finer than the granularity of the base table's partition. For example, if you partition the base table yearly using the birth_date column, creating a materialized view with PARTITION BY DATE_TRUNC(birth_date, MONTH) doesn't work.
• Any schema change invalidates the materialized view.
• Partition evolutions is supported. However, changing the partitioning columns of a base table without recreating the materialized view might result in full invalidation that cannot be fixed by refresh.
• There must be at least one snapshot in the base table.
• The Iceberg table must be a BigLake table, for example, an authorized external table.
• If VPC Service Controls is enabled, service accounts of the authorized external table must be added to your ingress rules, otherwise, VPC Service Controls blocks automatic background refresh for the materialized view.

The metadata.json file of your Iceberg table must have the following specifications. Without these specifications, your queries scan the base table, failing to use the materialized result.

• In table metadata:

  • current-snapshot-id
  • current-schema-id
  • snapshots
  • snapshot-log

• In snapshots:

  • parent-snapshot-id (if available)
  • schema-id
  • operation (in the summary field)

• Partitioning (for the partitioned materialized view)

Partitioned materialized views

Materialized views on partitioned tables can be partitioned. Partitioning a materialized view is similar to partitioning a normal table, in that it provides benefit when queries often access a subset of the partitions. In addition, partitioning a materialized view can improve the view's behavior when data in the base table or tables is modified or deleted. For more information, see Partition alignment.

If the base table is partitioned, then you can partition a materialized view on the same partitioning column. For time-based partitions, the granularity must match (hourly, daily, monthly, or yearly). For integer-range partitions, the range specification must exactly match. You cannot partition a materialized view over a non-partitioned base table.

If the base table is partitioned by ingestion time, then a materialized view can group by the _PARTITIONDATE column of the base table, and also partition by it. If you don't explicitly specify partitioning when you create the materialized view, then the materialized view is unpartitioned.

If the base table is partitioned, consider partitioning your materialized view as well to reduce refresh job maintenance cost and query cost.

Partition expiration

Partition expiration can't be set on materialized views. A materialized view implicitly inherits the partition expiration time from the base table. Materialized view partitions are aligned with the base table partitions, so they expire synchronously.

Example 1

In this example, the base table is partitioned on the transaction_time column with daily partitions. The materialized view is partitioned on the same column and clustered on the employee_id column.

CREATE TABLE my_project.my_dataset.my_base_table(
  employee_id INT64,
  transaction_time TIMESTAMP)
  PARTITION BY DATE(transaction_time)
  OPTIONS (partition_expiration_days = 2);

CREATE MATERIALIZED VIEW my_project.my_dataset.my_mv_table
  PARTITION BY DATE(transaction_time)
  CLUSTER BY employee_id
AS (
  SELECT
    employee_id,
    transaction_time,
    COUNT(employee_id) AS cnt
  FROM
    my_dataset.my_base_table
  GROUP BY
    employee_id, transaction_time
);

Example 2

In this example, the base table is partitioned by ingestion time with daily partitions. The materialized view selects the ingestion time as a column named date. The materialized view is grouped by the date column and partitioned by the same column.

CREATE MATERIALIZED VIEW my_project.my_dataset.my_mv_table
  PARTITION BY date
  CLUSTER BY employee_id
AS (
  SELECT
    employee_id,
    _PARTITIONDATE AS date,
    COUNT(1) AS count
  FROM
    my_dataset.my_base_table
  GROUP BY
    employee_id,
    date
);

Example 3

In this example, the base table is partitioned on a TIMESTAMP column named transaction_time, with daily partitions. The materialized view defines a column named transaction_hour, using the TIMESTAMP_TRUNC function to truncate the value to the nearest hour. The materialized view is grouped by transaction_hour and also partitioned by it.

Note the following:

• The truncation function that is applied to the partitioning column must be at least as granular as the partitioning of the base table. For example, if the base table uses daily partitions, the truncation function cannot use MONTH or YEAR granularity.

• In the materialized view's partition specification, the granularity has to match the base table.

CREATE TABLE my_project.my_dataset.my_base_table (
  employee_id INT64,
  transaction_time TIMESTAMP)
  PARTITION BY DATE(transaction_time);

CREATE MATERIALIZED VIEW my_project.my_dataset.my_mv_table
  PARTITION BY DATE(transaction_hour)
AS (
  SELECT
    employee_id,
    TIMESTAMP_TRUNC(transaction_time, HOUR) AS transaction_hour,
    COUNT(employee_id) AS cnt
  FROM
    my_dataset.my_base_table
  GROUP BY
    employee_id,
    transaction_hour
);

Cluster materialized views

You can cluster materialized views by their output columns, subject to the BigQuery clustered table limitations. Aggregate output columns cannot be used as clustering columns. Adding clustering columns to materialized views can improve the performance of queries that include filters on those columns.

Reference logical views

To request feedback or support for this feature, send email to [email protected].

Materialized view queries can reference logical views but are subject to the following limitations:

• Materialized view limitations apply.
• If the logical view changes, then the materialized view becomes invalid and must be fully refreshed.
• Smart tuning is not supported.

Considerations when creating materialized views

Which materialized views to create

When creating a materialized view, ensure your materialized view definition reflects query patterns against the base tables. Materialized views are more effective when they serve a broad set of queries rather than just one specific query pattern.

For example, consider a query on a table where users often filter by the columns user_id or department. You can group by these columns and optionally cluster by them, instead of adding filters like user_id = 123 into the materialized view.

As another example, users often use date filters, either by specific date, such as WHERE order_date = CURRENT_DATE(), or date range, such as WHERE order_date BETWEEN '2019-10-01' AND '2019-10-31'. Add a date range filter in the materialized view that covers expected date ranges in the query:

CREATE MATERIALIZED VIEW ...
  ...
  WHERE date > '2019-01-01'
  GROUP BY date

Joins

The following recommendations apply to materialized views with JOINs.

Put the most frequently changing table first

Ensure that the largest or most frequently changing table is the first/leftmost table referenced in the view query. Materialized views with joins support incremental queries and refresh when the first or left-most table in the query is appended, but changes to other tables fully invalidate the view cache. In star or snowflake schemas the first or leftmost table should generally be the fact table.

Avoid joining on clustering keys

Materialized views with joins work best in cases where the data is heavily aggregated or the original join query is expensive. For selective queries, BigQuery is often already able to perform the join efficiently and no materialized view is needed. For example consider the following materialized view definitions.

CREATE MATERIALIZED VIEW dataset.mv
  CLUSTER BY s_market_id
AS (
  SELECT
    s_market_id,
    s_country,
    SUM(ss_net_paid) AS sum_sales,
    COUNT(*) AS cnt_sales
  FROM dataset.store_sales
  INNER JOIN dataset.store
    ON ss_store_sk = s_store_sk
  GROUP BY s_market_id, s_country
);

Suppose store_sales is clustered on ss_store_sk and you often run queries like the following:

SELECT
  SUM(ss_net_paid)
FROM dataset.store_sales
INNER JOIN dataset.store
ON ss_store_sk = s_store_sk
WHERE s_country = 'Germany';

The materialized view might not be as efficient as the original query. For best results, experiment with a representative set of queries, with and without the materialized view.

Use materialized views with max_staleness option

The max_staleness materialized view option helps you achieve consistently high query performance with controlled costs when processing large, frequently changing datasets. With the max_staleness parameter, you can reduce cost and latency on your queries by setting an interval of time where data staleness of query results is acceptable. This behavior can be useful for dashboards and reports for which fully up-to-date query results aren't essential.

Data staleness

When you query a materialized view with the max_staleness option set, BigQuery returns the result based on the max_staleness value and the time at which the last refresh occurred.

If the last refresh occurred within the max_staleness interval, then BigQuery returns data directly from the materialized view without reading the base tables. For example, this applies if your max_staleness interval is 4 hours, and the last refresh occurred 2 hours ago.

If the last refresh occurred outside the max_staleness interval, then BigQuery reads the data from the materialized view, combines it with changes to the base table since the last refresh, and returns the combined result. This combined result might still be stale, up to your max_staleness interval. For example, this applies if your max_staleness interval is 4 hours, and the last refresh occurred 7 hours ago.

Create with max_staleness option

Select one of the following options:

{
  "kind": "bigquery#table",
  "tableReference": {
    "projectId": "project-id",
    "datasetId": "my_dataset",
    "tableId": "my_mv_table"
  },
  "materializedView": {
    "query": "select product_id,sum(clicks) as
                sum_clicks from project-id.my_dataset.my_base_table
                group by 1"
  }
  "maxStaleness": "4:0:0"
}

Apply max_staleness option

You can apply this parameter to existing materialized views by using the ALTER MATERIALIZED VIEW statement. For example:

ALTER MATERIALIZED VIEW project-id.my_dataset.my_mv_table
SET OPTIONS (enable_refresh = true, refresh_interval_minutes = 120,
  max_staleness = INTERVAL "8:0:0" HOUR TO SECOND);

Query with max_staleness

You can query materialized views with the max_staleness option as you would query any other materialized view, logical view, or table.

For example:

SELECT * FROM  project-id.my_dataset.my_mv_table

This query returns data from the last refresh if the data is not older than the max_staleness parameter. If the materialized view has not been refreshed within the max_staleness interval, BigQuery merges the results of the latest available refresh with the base table changes to return results within the max_staleness interval.

Data streaming and max_staleness results

If you stream data into the base tables of a materialized view with the max_staleness option, then the query of the materialized view might exclude records that were streamed into its tables before the beginning of the staleness interval. As a result, a materialized view that includes data from multiple tables and max_staleness option might not represent a point-in-time snapshot of those tables.

Smart tuning and the max_staleness option

Smart tuning automatically rewrites queries to use materialized views whenever possible regardless of the max_staleness option, even if the query does not reference a materialized view. The max_staleness option on a materialized view does not affect the results of the rewritten query. The max_staleness option only affects queries that directly query the materialized view.

Manage staleness and refresh frequency

You should set max_staleness based on your requirements. To avoid reading data from base tables, configure the refresh interval so that the refresh takes place within the staleness interval. You can account for the average refresh runtime plus a margin for growth.

For example, if one hour is required to refresh your materialized view and you want a one-hour buffer for growth, then you should set the refresh interval to two hours. This configuration ensures that the refresh occurs within your report's four-hour maximum for staleness.

CREATE MATERIALIZED VIEW project-id.my_dataset.my_mv_table
OPTIONS (enable_refresh = true, refresh_interval_minutes = 120, max_staleness =
INTERVAL "4:0:0" HOUR TO SECOND)
AS SELECT
  employee_id,
  DATE(transaction_time),
  COUNT(1) AS cnt
FROM my_dataset.my_base_table
GROUP BY 1, 2;

Non-incremental materialized views

Non-incremental materialized views support most SQL queries, including OUTER JOIN, UNION, and HAVING clauses, and analytic functions. To determine whether a materialized view was used in your query, check the cost estimates by using a dry run. In scenarios where data staleness is acceptable, for example for batch data processing or reporting, non-incremental materialized views can improve query performance and reduce cost. By using the max_staleness option, you can build arbitrary, complex materialized views that are automatically maintained and have built-in staleness guarantees.

Use non-incremental materialized views

You can create non-incremental materialized views by using the allow_non_incremental_definition option. This option must be accompanied by the max_staleness option. To ensure a periodic refresh of the materialized view, you should also configure a refresh policy. Without a refresh policy, you must manually refresh the materialized view.

The materialized view always represents the state of the base tables within the max_staleness interval. If the last refresh is too stale and doesn't represent the base tables within the max_staleness interval, then the query reads the base tables. To learn more about possible performance implications, see Data staleness.

Create with allow_non_incremental_definition

To create a materialized view with the allow_non_incremental_definition option, follow these steps. After you create the materialized view, you cannot modify the allow_non_incremental_definition option. For example, you cannot change the value true to false, or remove the allow_non_incremental_definition option from the materialized view.

{
  "kind": "bigquery#table",
  "tableReference": {
    "projectId": "my_project",
    "datasetId": "my_dataset",
    "tableId": "my_mv_table"
  },
  "materializedView": {
    "query": "`SELECT`
        s_store_sk,
        SUM(ss_net_paid) AS sum_sales,
        APPROX_QUANTILES(ss_net_paid, 2)[safe_offset(1)] median
      FROM my_project.my_dataset.store
      LEFT OUTER JOIN my_project.my_dataset.store_sales
        ON ss_store_sk = s_store_sk
      GROUP BY s_store_sk
      HAVING median < 40 OR median is NULL`",
    "allowNonIncrementalDefinition": true
  }
  "maxStaleness": "4:0:0"
}

Query with allow_non_incremental_definition

You can query non-incremental materialized views as you would query any other materialized view, logical view, or table.

For example:

SELECT * FROM  my_project.my_dataset.my_mv_table

If the data is not older than the max_staleness parameter, then this query returns data from the last refresh. For details about the staleness and freshness of data, see data staleness.

Limitations specific to non-incremental materialized views

The following limitations only apply to materialized views with the allow_non_incremental_definition option. With the exception of limitations on supported query syntax, all materialized view limitations still apply.

• Smart-tuning is not applied to the materialized views that include the allow_non_incremental_definition option. The only way to benefit from materialized views with the allow_non_incremental_definition option is to query them directly.
• Materialized views without the allow_non_incremental_definition option can incrementally refresh a subset of their data. Materialized views with the allow_non_incremental_definition option must be refreshed in their entirety.
• Materialized views with max_staleness option validates presence of the column-level security constraints during query execution. See more details about this in column-level access control

What's next

• Manage materialized views.
• Use materialized views.