Managing Query Execution Plans for Aurora PostgreSQL

With query plan management for Amazon Aurora with PostgreSQL compatibility, you can control how and when query execution plans change. Query plan management has two main objectives:

• Preventing plan regressions when the database system changes

• Controlling when the query optimizer can use new plans

The quality and consistency of query optimization have a major impact on the performance and stability of any relational database management system (RDBMS). Query optimizers create a query execution plan for a SQL statement at a specific point in time. As conditions change, the optimizer might pick a different plan that makes performance worse. A number of changes can all cause the query optimizer to choose a different plan and lead to performance regression. These changes include changes in statistics, constraints, environment settings, query parameter bindings, and software upgrades. Regression is a major concern for high-performance applications.

With query plan management, you can control execution plans for a set of statements that you want to manage. You can do the following:

• Improve plan stability by forcing the optimizer to choose from a small number of known, good plans.

• Optimize plans centrally and then distribute the best plans globally.

• Identify indexes that aren't used and assess the impact of creating or dropping an index.

• Automatically detect a new minimum-cost plan discovered by the optimizer.

• Try new optimizer features with less risk, because you can choose to approve only the plan changes that improve performance.

Topics

• Enabling Query Plan Management for Aurora PostgreSQL
• Upgrading Query Plan Management
• Basics of Query Plan Management
• Best Practices for Query Plan Management
• Examining Plans in the apg_plan_mgmt.dba_plans view
• Capturing Execution Plans
• Using Managed Plans
• Maintaining Execution Plans
• Parameter Reference for Query Plan Management
• Function Reference for Query Plan Management

Enabling Query Plan Management for Aurora PostgreSQL

Query plan management is available with Amazon Aurora PostgreSQL version 2.1.0 (compatible with PostgreSQL 10.5) and greater versions.

To enable query plan management

1. Open the Amazon RDS console at https://console.aws.amazon.com/rds/.

2. Create a new instance-level parameter group to use for query plan management parameters. For more information, see Creating a DB Parameter Group. Associate the new parameter group with the DB instances in which you want to use query plan management. For more information, see Modify a DB Instance in a DB Cluster.

3. Create a new cluster-level parameter group to use for query plan management parameters. For more information, see Creating a DB Cluster Parameter Group. Associate the new cluster-level parameter group with the DB clusters in which you want to use query plan management. For more information, see Modifying the DB Cluster by Using the Console, CLI, and API.

4. Open your cluster-level parameter group and set the rds.enable_plan_management parameter to 1. For more information, see Modifying Parameters in a DB Cluster Parameter Group.

5. Restart your DB instance to enable this new setting.

6. Connect to your DB instance with a SQL client such as psql.

7. Create the apg_plan_mgmt extension for your DB instance. The following shows an example.

   psql my-database
   my-database=> CREATE EXTENSION apg_plan_mgmt;

   If you create the apg_plan_mgmt extension in the template1 default database, then the query plan management extension is available in each new database that you create.

   Note

   To create the apg_plan_mgmt extension, you need the rds_superuser role. When you create the apg_plan_mgmt extension, the apg_plan_mgmt role is created. Users must be granted the apg_plan_mgmt role to administer the apg_plan_mgmt extension.

Upgrading Query Plan Management

The latest version of query plan management is 2.0. If you installed an earlier version of query plan management, we strongly recommend that you upgrade to version 2.0. For version details, see Extension Versions for Amazon Aurora PostgreSQL.

To upgrade, run the following commands at the cluster or DB instance level.

ALTER EXTENSION apg_plan_mgmt UPDATE TO '2.0';
SELECT apg_plan_mgmt.validate_plans('update_plan_hash');
SELECT apg_plan_mgmt.reload();

Basics of Query Plan Management

You can manage any SELECT, INSERT, UPDATE, or DELETE statement with query plan management, regardless of how complex the statement is. Prepared, dynamic, embedded, and immediate-mode SQL statements are all supported. All PostgreSQL language features can be used, including partitioned tables, inheritance, row-level security, and recursive common table expressions (CTEs).

Performing a Manual Plan Capture

To capture plans for specific statements, use the manual capture mode as in the following example.

/* Turn on manual capture */
SET apg_plan_mgmt.capture_plan_baselines = manual; 
EXPLAIN SELECT COUNT(*) from pg_class;       -- capture the plan baseline
SET apg_plan_mgmt.capture_plan_baselines = off;    -- turn off capture
SET apg_plan_mgmt.use_plan_baselines =     true;   -- turn on plan usage

You can either execute SELECT, INSERT, UPDATE, or DELETE statements, or you can include the EXPLAIN statement as shown above. Use EXPLAIN to capture a plan without the overhead or potential side-effects of executing the statement. For more about manual capture, see Manually Capturing Plans for Specific SQL Statements.

Viewing Captured Plans

When EXPLAIN SELECT runs in the previous example, the optimizer saves the plan. To do so, it inserts a row into the apg_plan_mgmt.dba_plans view and commits the plan in an autonomous transaction. You can see the contents of the apg_plan_mgmt.dba_plans view if you've been granted the apg_plan_mgmt role. The following query displays some important columns of the dba_plans view.

SELECT sql_hash, plan_hash, status, enabled, plan_outline, sql_text::varchar(40)
FROM apg_plan_mgmt.dba_plans
ORDER BY sql_text, plan_created;

Each row displayed represents a managed plan. The preceding example displays the following information.

• sql_hash – The ID of the managed statement that the plan is for.

• plan_hash – The ID of the managed plan.

• status – The status of the plan. The optimizer can run an approved plan.

• enabled – A value that indicates whether the plan is enabled for use or disabled and not for use.

• plan_outline – Details of the managed plan.

For more about the apg_plan_mgmt.dba_plans view, see Examining Plans in the apg_plan_mgmt.dba_plans view.

Working with Managed Statements and the SQL Hash

A managed statement is a SQL statement captured by the optimizer under query plan management. You specify which SQL statements to capture as managed statements using either manual or automatic capture:

• For manual capture, you provide the specific statements to the optimizer as shown in the previous example.

• For automatic capture, the optimizer captures plans for statements that run multiple times. Automatic capture is shown in a later example.

In the apg_plan_mgmt.dba_plans view, you can identify a managed statement with a SQL hash value. The SQL hash is calculated on a normalized representation of the SQL statement that removes some differences such as the literal values. Using normalization means that when multiple SQL statements differ only in their literal or parameter values, they are represented by the same SQL hash in the apg_plan_mgmt.dba_plans view. Therefore, there can be multiple plans for the same SQL hash where each plan is optimal under different conditions.

When the optimizer processes any SQL statement, it uses the following rules to create the normalized SQL statement:

• Removes any leading block comment

• Removes the EXPLAIN keyword and EXPLAIN options, if present

• Removes trailing spaces

• Removes all literals

• Preserves whitespace and case for readability

For example, take the following statement.

/*Leading comment*/ EXPLAIN SELECT /* Query 1 */ * FROM t WHERE x > 7 AND y = 1; 

The optimizer normalizes this statement as the following.

SELECT /* Query 1 */ * FROM t WHERE x > CONST AND y = CONST; 

Working with Automatic Plan Capture

Use automatic plan capture if you want to capture plans for all SQL statements in your application, or if you can’t use manual capture. With automatic plan capture, by default, the optimizer captures plans for statements that run at least two times. Do the following to use automatic plan capture.

1. Turn on automatic plan capture by setting apg_plan_mgmt.capture_plan_baselines to automatic in the parameter group for the DB instance. For more information, see Modifying Parameters in a DB Parameter Group.

2. Restart your DB instance.

As your application runs, the optimizer captures plans for any statement that runs more than once. The optimizer always sets the status of a managed statement's first captured plan to approved. A managed statement's set of approved plans is known as its plan baseline.

As your application continues to run, the optimizer might find additional plans for the managed statements. The optimizer sets additional captured plans to a status of Unapproved.

The set of all captured plans for a managed statement is known as the plan history. Later, you can decide if the Unapproved plans perform well and change them to Approved, Rejected, or Preferred by using the apg_plan_mgmt.evolve_plan_baselines function or the apg_plan_mgmt.set_plan_status function.

To turn off automatic plan capture, set apg_plan_mgmt.capture_plan_baselines to off in the parameter group for the DB instance. Then restart the database for the setting to take effect.

For more about plan capture, see Capturing Execution Plans.

Validating Plans

Managed plans can become invalid ("stale") when objects that they depend on are removed, such as an index. To find and delete all plans that are stale, use the apg_plan_mgmt.validate_plans function.

SELECT apg_plan_mgmt.validate_plans('delete');

For more information, see Validating Plans.

Approving New Plans That Improve Performance

While using your managed plans, you can verify whether newer, lower-cost plans discovered by the optimizer are faster than the minimum-cost plan already in the plan baseline. To do the performance comparison and optionally approve the faster plans, call the apg_plan_mgmt.evolve_plan_baselines function.

The following example automatically approves any unapproved plan that is enabled and faster by at least 10 percent than the minimum-cost plan in the plan baseline.

SELECT apg_plan_mgmt.evolve_plan_baselines(
   sql_hash,
   plan_hash, 
   1.1,
   'approve'
)
FROM apg_plan_mgmt.dba_plans 
WHERE status = 'Unapproved' AND enabled = true;

When the apg_plan_mgmt.evolve_plan_baselines function runs, it collects performance statistics and saves them in the apg_plan_mgmt.dba_plans view in the columns planning_time_ms, execution_time_ms, cardinality_error, total_time_benefit_ms, and execution_time_benefit_ms. The apg_plan_mgmt.evolve_plan_baselines function also updates the columns last_verified or last_validated timestamps, in which you can see the most recent time the performance statistics were collected.

SELECT sql_hash, plan_hash, status, last_verified, sql_text::varchar(40)
FROM apg_plan_mgmt.dba_plans
ORDER BY last_verified DESC;  -- value updated by evolve_plan_baselines()  

For more information about verifying plans, see Evaluating Plan Performance.

Deleting Plans

The optimizer deletes plans automatically if they have not been executed or chosen as the minimum-cost plan for the plan retention period. By default, the plan retention period is 32 days. To change the plan retention period, set the apg_plan_mgmt.plan_retention_period parameter.

You can also review the contents of the apg_plan_mgmt.dba_plans view and delete any plans you don't want by using the apg_plan_mgmt.delete_plan function. For more information, see Deleting Plans.