Work with multi-statement queries

This document describes how to use multi-statement queries in BigQuery, such as how to write multi-statement queries, use temporary tables in multi-statement queries, reference variables in multi-statement queries, and debug multi-statement queries.

A multi-statement query is a collection of SQL statements that you can execute in one request. With multi-statement queries you can run multiple statements in a sequence, with shared state. Multi-statement queries can have side effects such as adding or modifying table data.

Multi-statement queries are often used in stored procedures and support procedural language statements, which let you do things like define variables and implement control flow.

Write, run, and save multi-statement queries

A multi-statement query consists of one or more SQL statements separated by semicolons. Any valid SQL statement can be used in a multi-statement query. Multi-statement queries can also include procedural language statements, which let you use variables or implement control flow with your SQL statements.

Write a multi-statement query

You can write a multi-statement query in BigQuery. The following multi-query statement query declares a variable and uses the variable inside an IF statement:

DECLARE day INT64;
SET day = (SELECT EXTRACT(DAYOFWEEK from CURRENT_DATE));
if day = 1 or day = 7 THEN
  SELECT 'Weekend';
ELSE
  SELECT 'Weekday';
END IF

BigQuery interprets any request with multiple statements as a multi-statement query, unless the statements consist entirely of CREATE TEMP FUNCTION statements followed by a single SELECT statement. For example, the following is not considered a multi-statement query:

CREATE TEMP FUNCTION Add(x INT64, y INT64) AS (x + y);
SELECT Add(3, 4);

Run a multi-statement query

You can run a multi-statement query in the same way as any other query, for example, in the Google Cloud console or using the bq command-line tool.

Dry-run a multi-statement query

To estimate the number of bytes read by a multi-statement query, consider a dry run. A dry run of a multi-statement query is most accurate for queries that only contain SELECT statements.

Dry runs have special handling for the following query and statement types:

CALL statements: the dry run validates that the called procedure exists and has a signature matching the arguments provided. The content of the called procedure and all statements after the CALL statement are not validated.
DDL statements: the dry run validates the first DDL statement and then stops. All subsequent statements are skipped.
DML statements: the dry run validates the DML statement and then continues to validate subsequent statements. In this case, byte estimates are based on original table sizes, and don't take into account the outcome of the DML statement.
EXECUTE IMMEDIATE statements: the dry run validates the query expression, but does not evaluate the dynamic query itself. All statements following the EXECUTE IMMEDIATE statement are skipped.
Queries that use variables in a partition filter: the dry run validates the initial query and subsequent statements. However, the dry run is unable to calculate the runtime value of variables in a partition filter. This affects the bytes read estimate.
Queries that use variables in the timestamp expression of a FOR SYSTEM TIME AS OF clause: the dry run uses the table's current content and ignores the FOR SYSTEM TIME AS OF clause. This affects the bytes read estimate if there are size differences between the current table and the prior iteration of the table.
FOR, IF and WHILE control statements: the dry run stops immediately. Condition expressions, bodies of the control statement, and all subsequent statements are not validated.

Dry runs operate on a best-effort basis, and the underlying process is subject to change. Dry runs are subject to the following stipulations:

A query that successfully completes a dry run might not execute successfully. For example, queries might fail at runtime due to reasons that are not detected by dry runs.
A query that successfully executes might not complete a dry run successfully. For example, queries might fail dry runs due to reasons caught at execution.
Dry runs that successfully run today are not guaranteed to always run in the future. For example, changes to the dry run implementation might detect errors in a query that were previously undetected.

Save a multi-statement query

To save a multi-statement query, see Work with saved queries.

Use variables in a multi-statement query

A multi-statement query can contain user-created variables and system variables.

You can declare user-created variables, assign values to them, and reference them throughout the query.
You can reference system variables in a query and assign values to some of them, but unlike user-defined variables, you don't declare them. System variables are built into BigQuery.

Declare a user-created variable

You must declare user-created variables either at the start of the multi-statement query or at the start of a BEGIN block. Variables declared at the start of the multi-statement query are in scope for the entire query. Variables declared inside a BEGIN block have scope for the block. They go out of scope after the corresponding END statement. The maximum size of a variable is 1 MB, and the maximum size of all variables used in a multi-statement query is 10 MB.

You can declare a variable with the DECLARE procedural statement like this:

DECLARE x INT64;

BEGIN
DECLARE y INT64;
-- Here you can reference x and y
END;

-- Here you can reference x, but not y

Set a user-created variable

After you declare a user-created variable, you can assign a value to it with the SET procedural statement like this:

DECLARE x INT64 DEFAULT 0;
SET x = 10;

Set a system variable

You don't create system variables, but you can override the default value for some of them like this:

SET @@dataset_project_id = 'MyProject';

You can also set and implicitly use a system variable in a multi-statement query. For example, in the following query you must include the project each time you wish to create a new table:

BEGIN
  CREATE TABLE MyProject.MyDataset.MyTempTableA (id STRING);
  CREATE TABLE MyProject.MyDataset.MyTempTableB (id STRING);
END;

If you don't want to add the project to table paths multiple times, you can assign the dataset project ID MyProject to the @@dataset_project_id system variable in the multi-statement query. This assignment makes MyProject the default project for the rest of the query.

SET @@dataset_project_id = 'MyProject';

BEGIN
  CREATE TABLE MyDataset.MyTempTableA (id STRING);
  CREATE TABLE MyDataset.MyTempTableB (id STRING);
END;

Similarly, you can set the @@dataset_id system variable to assign a default dataset for the query. For example:

SET @@dataset_project_id = 'MyProject';
SET @@dataset_id = 'MyDataset';

BEGIN
  CREATE TABLE MyTempTableA (id STRING);
  CREATE TABLE MyTempTableB (id STRING);
END;

You can also explicitly reference system variables like @@dataset_id in many parts of a multi-statement query. To learn more, see Reference a system variable.

Reference a user-created variable

After you have declared and set a user-created variable, you can reference it in a multi-statement query. If a variable and column share the same name, the column takes precedence.

This returns column x + column x:

DECLARE x INT64 DEFAULT 0;
SET x = 10;

WITH Numbers AS (SELECT 50 AS x)
SELECT (x+x) AS result FROM Numbers;

+--------+
| result |
+--------+
| 100    |
+--------+

This returns column y + variable x:

DECLARE x INT64 DEFAULT 0;
SET x = 10;

WITH Numbers AS (SELECT 50 AS y)
SELECT (y+x) AS result FROM Numbers;

+--------+
| result |
+--------+
| 60     |
+--------+

Reference a system variable

You can reference a system variable in a multi-statement query.

The following query returns the default time zone:

BEGIN
  SELECT @@time_zone AS default_time_zone;
END;

+-------------------+
| default_time_zone |
+-------------------+
| UTC               |
+-------------------+

You can use system variables with DDL and DML queries. For example, here are a few ways to use the system variable @@time_zone when creating and updating a table:

BEGIN
  CREATE TEMP TABLE MyTempTable
  AS SELECT @@time_zone AS default_time_zone;
END;

BEGIN
  CREATE OR REPLACE TABLE MyDataset.MyTable(default_time_zone STRING)
  OPTIONS (description = @@time_zone);
END;

BEGIN
  UPDATE MyDataset.MyTable
  SET default_time_zone = @@time_zone
  WHERE TRUE;
END;

There are some places where system variables cannot be used in DDL and DML queries. For example, you can't use a system variable as a project name, dataset, or table name. This produces an error when you attempt to include the @@dataset_id system variable in a table path:

BEGIN
  CREATE TEMP TABLE @@dataset_id.MyTempTable (id STRING);
END;

Use temporary tables in a multi-statement query

Temporary tables let you save intermediate results to a table. Temporary tables are managed by BigQuery, so you don't need to save or maintain them in a dataset. You are charged for storage of temporary tables.

You can create and reference a temporary table in a multi-statement query. When you are finished with the temporary table, you can delete it manually to minimize storage costs, or wait for BigQuery to delete it after 24 hours.

Create a temporary table

You can create a temporary table for a multi-statement query with the CREATE TABLE statement. The following example creates a temporary table to store the results of a query and uses the temporary table in a subquery:

-- Find the top 100 names from the year 2017.
CREATE TEMP TABLE top_names(name STRING)
AS
 SELECT name
 FROM `bigquery-public-data`.usa_names.usa_1910_current
 WHERE year = 2017
 ORDER BY number DESC LIMIT 100
;
-- Which names appear as words in Shakespeare's plays?
SELECT
 name AS shakespeare_name
FROM top_names
WHERE name IN (
 SELECT word
 FROM `bigquery-public-data`.samples.shakespeare
);

Other than the use of TEMP or TEMPORARY, the syntax is identical to the CREATE TABLE syntax.

When you create a temporary table, don't use a project or dataset qualifier in the table name. The table is automatically created in a special dataset.

Reference a temporary table

You can refer to a temporary table by name for the duration of the current multi-statement query. This includes temporary tables created by a procedure within the multi-statement query. You cannot share temporary tables. Temporary tables reside in hidden _script% datasets with randomly generated names. Listing datasets article describes how to list hidden datasets.

Delete temporary tables

You can delete a temporary table explicitly before the multi-statement query completes by using the DROP TABLE statement:

CREATE TEMP TABLE table1(x INT64);
SELECT * FROM table1;  -- Succeeds
DROP TABLE table1;
SELECT * FROM table1;  -- Results in an error

After a multi-statement query finishes, the temporary table exists for up to 24 hours.

View temporary table data

After you create a temporary table, you can view the structure of the table and any data in it. To view the table structure and data, follow these steps:

In the Google Cloud console, go to the BigQuery Explorer page.

Go to Explorer
Click Query history.
Choose the query that created the temporary table.
In the Destination table row, click Temporary table.

Note: The name of the temporary table used in the multi-statement query is not preserved. In the Google Cloud console, it has a random name.

Qualify temporary tables with `_SESSION`

When temporary tables are used together with a default dataset, unqualified table names refer to a temporary table if one exists, or a table in the default dataset. The exception is for CREATE TABLE statements, where the target table is considered a temporary table if and only if the TEMP or TEMPORARY keyword is present.

For example, consider the following multi-statement query:

-- Create table t1 in the default dataset
CREATE TABLE t1 (x INT64);

-- Create temporary table t1.
CREATE TEMP TABLE t1 (x INT64);

-- This statement selects from the temporary table.
SELECT * FROM t1;

-- Drop the temporary table
DROP TABLE t1;

-- Now that the temporary table is dropped, this statement selects from the
-- table in the default dataset.
SELECT * FROM t1;

You can explicitly indicate that you are referring to a temporary table by qualifying the table name with _SESSION:

-- Create a temp table
CREATE TEMP TABLE t1 (x INT64);

-- Create a temp table using the `_SESSION` qualifier
CREATE TEMP TABLE _SESSION.t2 (x INT64);

-- Select from a temporary table using the `_SESSION` qualifier
SELECT * FROM _SESSION.t1;

If you use the _SESSION qualifier for a query of a temporary table that does not exist, the multi-statement query throws an error indicating that the table does not exist. For example, if there is no temporary table named t3, the multi-statement query throws an error even if a table named t3 exists in the default dataset.

You cannot use _SESSION to create a non-temporary table:

CREATE TABLE _SESSION.t4 (x INT64);  -- Fails

Collect information about a multi-statement query job

A multi-statement query job contains information about a multi-statement query that has been executed. Some common tasks that you can perform with job data include returning the last statement executed with the multi-statement query or returning all statements executed with the multi-statement query.

Return the last executed statement

The jobs.getQueryResults method returns the query results for the last statement to execute in the multi-statement query. If no statement was executed, no results are returned.

Return all executed statements

To get the results of all statements in a multi-statement query, enumerate the child jobs and call jobs.getQueryResults on each of them.

Enumerate child jobs

Multi-statement queries are executed in BigQuery using jobs.insert, similar to any other query, with the multi-statement queries specified as the query text. When a multi-statement query runs, additional jobs, known as child jobs, are created for each statement in the multi-statement query. You can enumerate the child jobs of a multi-statement query by calling jobs.list, passing in the multi-statement query job ID as the parentJobId parameter.

Debug a multi-statement-query

Here are some tips for debugging multi-statement queries:

Use the ASSERT statement to assert that a Boolean condition is true.
Use BEGIN...EXCEPTION...END to catch errors and display the error message and stack trace.
Use SELECT FORMAT("....") to show intermediate results.
When you run a multi-statement query in the Google Cloud console, you can view the output of each statement in the multi-statement query. The bq command-line tool's 'bq query` command also shows the results of each step when you run a multi-statement query.
In the Google Cloud console, you can select an individual statement inside the query editor and run it.

Permissions

Permission to access a table, model, or other resource is checked at the time of execution. If a statement is not executed, or an expression is not evaluated, BigQuery does not check whether the user executing the multi-statement query has access to any resources referenced by it.

Within a multi-statement query, the permissions for each expression or statement are validated separately. For example:

SELECT * FROM dataset_with_access.table1;
SELECT * FROM dataset_without_access.table2;

If the user executing the query has access to table1 but does not have access to table2, the first query succeeds and the second query fails. The multi-statement query job itself also fails.

Security constraints

In multi-statement queries, you can use dynamic SQL to build SQL statements at runtime. This is convenient, but can offer new opportunities for misuse. For example, executing the following query poses a potential SQL injection security threat since the table parameter could be improperly filtered, allow access to, and be executed on unintended tables.

-- Risky query vulnerable to SQL injection attack.
EXECUTE IMMEDIATE CONCAT('SELECT * FROM SensitiveTable WHERE id = ', @id);

To avoid exposing or leaking sensitive data in a table or running commands like DROP TABLE to delete data in a table, BigQuery's dynamic procedural statements support multiple security measures to reduce exposure to SQL injection attacks, including:

An EXECUTE IMMEDIATE statement does not allow its query, expanded with query parameters and variables, to embed multiple SQL statements.
The following commands are restricted from being executed dynamically: BEGIN/END, CALL, CASE, IF, LOOP, WHILE, and EXECUTE IMMEDIATE.

Configuration field limitations

The following job configuration query fields cannot be set for a multi-statement query:

clustering
create_disposition
destination_table
destination_encryption_configuration
range_partitioning
schema_update_options
time_partitioning
user_defined_function_resources
write_disposition

Pricing

Pricing for multi-statement queries includes charges for queries (when using the on-demand billing model) and storage for temporary tables. When you are using reservations, query usage is covered by your reservation charges.

On-demand query size calculation

If you use on-demand billing, BigQuery charges for multi-statement queries based on the number of bytes processed during execution of the multi-statement queries.

To get an estimate of how many bytes a multi-statement query might process, you can run a dry run.

The following pricing applies for these multi-statement queries:

DECLARE: the sum of bytes scanned for any tables referenced in the DEFAULT expression. DECLARE statements with no table references don't incur a cost.
SET: the sum of bytes scanned for any tables referenced in the expression. SET statements with no table references don't incur a cost.
IF: the sum of bytes scanned for any tables referenced in the condition expression. IF condition expressions with no table reference don't incur a cost. Any statements within the IF block that are not executed don't incur a cost.
WHILE: the sum of bytes scanned for any tables referenced in the condition expression. WHILE statements with no table references in the condition expression don't incur a cost. Any statements within the WHILE block that are not executed don't incur a cost.
CONTINUE or ITERATE: No associated cost.
BREAK or LEAVE: No associated cost.
BEGIN or END: No associated cost.

If a multi-statement query fails, the cost of any statements up until the failure still applies. The statement that failed does not incur any costs.

For example, the following sample code contains comments preceding every statement that explain what cost, if any, is incurred by each statement:

-- No cost, since no tables are referenced.
DECLARE x DATE DEFAULT CURRENT_DATE();
-- Incurs the cost of scanning string_col from dataset.table.
DECLARE y STRING DEFAULT (SELECT MAX(string_col) FROM dataset.table);
-- Incurs the cost of copying the data from dataset.big_table.  Once the
-- table is created, you are not charged for storage while the rest of the
-- multi-statement query runs.
CREATE TEMP TABLE t AS SELECT * FROM dataset.big_table;
-- Incurs the cost of scanning column1 from temporary table t.
SELECT column1 FROM t;
-- No cost, since y = 'foo' doesn't reference a table.
IF y = 'foo' THEN
  -- Incurs the cost of scanning all columns from dataset.other_table, if
  -- y was equal to 'foo', or otherwise no cost since it is not executed.
  SELECT * FROM dataset.other_table;
ELSE
  -- Incurs the cost of scanning all columns from dataset.different_table, if
  -- y was not equal to 'foo', or otherwise no cost since it is not executed.
  UPDATE dataset.different_table
  SET col = 10
  WHERE true;
END IF;
-- Incurs the cost of scanning date_col from dataset.table for each
-- iteration of the loop.
WHILE x < (SELECT MIN(date_col) FROM dataset.table) DO
  -- No cost, since the expression does not reference any tables.
  SET x = DATE_ADD(x, INTERVAL 1 DAY);
  -- No cost, since the expression does not reference any tables.
  IF true THEN
    -- LEAVE has no associated cost.
    LEAVE;
  END IF;
  -- Never executed, since the IF branch is always taken, so does not incur
  -- a cost.
  SELECT * FROM dataset.big_table;
END WHILE;

For more information, see Query size calculation.

Storage pricing

You are charged for temporary tables created by multi-statement queries. You can use the TABLE_STORAGE or TABLE_STORAGE_USAGE_TIMELINE views to see the storage used by these temporary tables. Temporary tables reside in hidden _script% datasets with randomly generated names.

Quotas

For information about multi-statement query quotas, see Quotas and limits.

View the number of multi-statement queries

You can view the number of active multi-statement queries using the INFORMATION_SCHEMA.JOBS_BY_PROJECT view. The following example uses the INFORMATION_SCHEMA.JOBS_BY_PROJECT view to show the number of multi-statement queries from the previous day:

SELECT
  COUNT(*)
FROM
  `region-us`.INFORMATION_SCHEMA.JOBS_BY_PROJECT
WHERE
  creation_time BETWEEN TIMESTAMP_SUB(CURRENT_TIMESTAMP(), INTERVAL 1 DAY) AND CURRENT_TIMESTAMP()
AND job_type = "QUERY"
AND state = 'RUNNING'
AND statement_type = 'SCRIPT'

For more information about querying INFORMATION_SCHEMA.JOBS for multi-statement queries, see Multi-statement query job.