Use history-based optimizations

For support during the preview, email bigquery-history-based-optimization-support@google.com.

This guide describes how to enable, disable, and analyze history-based optimizations for queries.

About history-based optimizations

History-based optimizations use information from already completed executions of similar queries to apply additional optimizations and further improve query performance such as slot time consumed and query latency. For example, when you apply history-based optimization, the first query execution might take 60 seconds, but the second query execution might take only 30 seconds if a history-based optimization was identified. This process continues until there are no additional optimizations to add.

The following is an example of how history-based optimizations work with BigQuery:

Execution count	Query slot time consumed	Notes
1	60	Original execution.
2	30	First history based-optimization applied.
3	20	Second history based-optimization applied.
4	21	No additional history based-optimizations to apply.
5	19	No additional history based-optimizations to apply.
6	20	No additional history based-optimizations to apply.

History-based optimizations are only applied when there is high confidence that there will be a beneficial impact to the query performance. In addition, when an optimization does not significantly improve query performance, that optimization is revoked and not used in future executions of that query.

Enable history-based optimizations

To use history-based optimizations in a project, include the following parameter in the ALTER PROJECT or ALTER ORGANIZATION statement: default_query_optimizer_options = 'adaptive=on'

Example:

ALTER PROJECT `user_project`
SET OPTIONS (
  `region-us.default_query_optimizer_options` = 'adaptive=on'
);

Disable history-based optimizations

To disable history-based optimizations in a project, include the default_query_optimizer_options = 'adaptive=off' parameter in the ALTER PROJECT or ALTER ORGANIZATION statement.

Example:

ALTER PROJECT `user_project`
SET OPTIONS (
  `region-us.default_query_optimizer_options` = 'adaptive=off'
);

Review history-based optimizations for a job

To review the history-based optimizations for a job, you can use a SQL query or a REST API method call.

SQL

You can use a query to get the history-based optimizations for a job. The query must include INFORMATION_SCHEMA.JOBS_BY_PROJECT and the query_info.optimization_details column name.

In the following example, the optimization details are returned for a job called sample_job. If no history-based optimizations were applied, NULL is produced for optimization_details:

SELECT
  job_id,
  query_info.optimization_details
FROM `project_name.region-us`.INFORMATION_SCHEMA.JOBS_BY_PROJECT
WHERE job_id = 'sample_job'
LIMIT 1;

The results look similar to the following:

-- The JSON in optimization_details has been formatted for readability.
/*------------+-----------------------------------------------------------------*
 | job_id     | optimization_details                                            |
 +------------+-----------------------------------------------------------------+
 | sample_job | {                                                               |
 |            |   "optimizations": [                                            |
 |            |     {                                                           |
 |            |       "semi_join_reduction": "web_sales.web_date,RIGHT"         |
 |            |     },                                                          |
 |            |     {                                                           |
 |            |       "semi_join_reduction": "catalog_sales.catalog_date,RIGHT" |
 |            |     },                                                          |
 |            |     {                                                           |
 |            |       "semi_join_reduction": "store_sales.store_date,RIGHT"     |
 |            |     },                                                          |
 |            |     {                                                           |
 |            |       "join_commutation": "web_returns.web_item"                |
 |            |     },                                                          |
 |            |     {                                                           |
 |            |       "parallelism_adjustment": "applied"                       |
 |            |     },                                                          |
 |            |   ]                                                             |
 |            | }                                                               |
 *------------+-----------------------------------------------------------------*/

API

To get the optimization details for a job, you can call the jobs.get method.

In the following example, the jobs.get method returns the optimization details (optimizationDetails) in the full response:

{
  "jobReference": {
    "projectId": "myProject",
    "jobId": "sample_job"
  }
}

The results look similar to the following:

-- The unrelated parts in the full response have been removed.
{
  "jobReference": {
    "projectId": "myProject",
    "jobId": "sample_job",
    "location": "US"
  },
  "statistics": {
    "query": {
      "queryInfo": {
        "optimizationDetails": {
          "optimizations": [
            {
              "semi_join_reduction": "web_sales.web_date,RIGHT"
            },
            {
              "semi_join_reduction": "catalog_sales.catalog_date,RIGHT"
            },
            {
              "semi_join_reduction": "store_sales.store_date,RIGHT"
            },
            {
              "join_commutation": "web_returns.web_item"
            },
            {
              "parallelism_adjustment": "applied"
            }
          ]
        }
      }
    }
  }
}

Estimate impact of history-based optimizations

To estimate the impact of history-based optimizations, you can use the following sample SQL query to identify project queries with the greatest estimated improvement to execution time.

  WITH
    jobs AS (
      SELECT
        *,
        query_info.query_hashes.normalized_literals AS query_hash,
        TIMESTAMP_DIFF(end_time, start_time, MILLISECOND) AS elapsed_ms,
        IFNULL(
          ARRAY_LENGTH(JSON_QUERY_ARRAY(query_info.optimization_details.optimizations)) > 0,
          FALSE)
          AS has_history_based_optimization,
      FROM region-us.INFORMATION_SCHEMA.JOBS_BY_PROJECT
      WHERE EXTRACT(DATE FROM creation_time) > DATE_SUB(CURRENT_DATE(), INTERVAL 30 DAY)
    ),
    most_recent_jobs_without_history_based_optimizations AS (
      SELECT *
      FROM jobs
      WHERE NOT has_history_based_optimization
      QUALIFY ROW_NUMBER() OVER (PARTITION BY query_hash ORDER BY end_time DESC) = 1
    )
  SELECT
    job.job_id,
    SAFE_DIVIDE(
      original_job.elapsed_ms - job.elapsed_ms,
      original_job.elapsed_ms) AS percent_execution_time_saved,
    job.elapsed_ms AS new_elapsed_ms,
    original_job.elapsed_ms AS original_elapsed_ms,
  FROM jobs AS job
  INNER JOIN most_recent_jobs_without_history_based_optimizations AS original_job
    USING (query_hash)
  WHERE
    job.has_history_based_optimization
    AND original_job.end_time < job.start_time
  ORDER BY percent_execution_time_saved DESC
  LIMIT 10;

The result of the preceding query is similar to the following if history-based optimizations were applied:

  /*--------------+------------------------------+------------------+-----------------------*
   |    job_id    | percent_execution_time_saved | new_execution_ms | original_execution_ms |
   +--------------+------------------------------+------------------+-----------------------+
   | sample_job1  |           0.6780685018624512 |             7087 |                 22014 |
   | sample_job2  |           0.6648580041250198 |            10562 |                 31515 |
   | sample_job3  |          0.63285605271764256 |            97668 |                266021 |
   | sample_job4  |            0.611341417268879 |           923384 |               2375823 |
   | sample_job5  |           0.5538127208971375 |          1060062 |               2375823 |
   | sample_job6  |           0.4539694316803648 |          2324071 |               4256302 |
   | sample_job7  |          0.38227031526376026 |            17811 |                 28833 |
   | sample_job8  |          0.33826608962725113 |            66360 |                100282 |
   | sample_job9  |          0.32087813758311606 |            44020 |                 64819 |
   | sample_job10 |           0.2835641631948354 |            19088 |                 26643 |
   *--------------+------------------------------+------------------+-----------------------*/

Details

This is only an estimation of history-based optimization impact. Many factors can influence query performance, including but not limited to slot availability, change in data over time, differences in query parameter values, and view or UDF definitions.
This query can be applied to other query performance metrics such as total_slot_ms and total_bytes_billed. For more information, see the schema for INFORMATION_SCHEMA.JOBS.
If the result of this sample query is empty, then no jobs used history-based optimizations or all queries were optimized more than 30 days ago.

Roles and permissions

To opt in to history-based optimizations, you must have the required permissions to create BigQuery default configurations, and then you must use the ALTER PROJECT statement to enable history-based optimizations. Once you've enabled history-based optimizations, all jobs in that project use history-based optimizations, regardless of which user created the job. To learn more about required permissions for default configurations, see Required permissions for default configurations. To enable history-based optimizations, see Enable history-based optimizations.
To review the history-based optimizations for a job using the INFORMATION_SCHEMA.JOBS view, you must have the required role. For more information, see Required role for INFORMATION_SCHEMA.JOBS view.