Identify PostgreSQL® slow queries

PostgreSQL® allows you to keep track of queries with certain performance metrics and statistics, which comes in handy when identifying slow queries.

Tip

When using Aiven for PostgreSQL®, you can check the Query statistics page for your service in the Aiven Console to identify long running queries.

Under the hood, the Query statistics page uses the pg_stat_statements extension, a module that provides a means for tracking the planning and execution statistics of all SQL statements executed by your PostgreSQL® server, to provide you with the basic information that can be useful for identifying slow queries.

Query statistics

These are the entries provided by Query statistics which are deduced via the pg_stat_statements:

Column Type	Description
Query	Text of a representative statement
Rows	Total number of rows retrieved or affected by the statement
Calls	Number of times the statement was executed
Min (ms)	Minimum time spent executing the statement
Max (ms)	Maximum time spent executing the statement
Mean (ms)	Mean time spent executing the statement
Stddev (ms)	Population standard deviation of time spent executing the statement
Total (ms)	Total time spent executing the statement

You can also create custom queries using the pg_stat_statements view and use all the available columns to investigate your use case.

Pre-requisites

To query the pg_stat_statements view, you’ll need to create the pg_stat_statements extension (included in the list of available extensions) that can be done via the following CREATE EXTENSION command:

CREATE EXTENSION pg_stat_statements;

Discover slow queries

You can run the following command to display the pg_stat_statements view and all the columns contained:

\d pg_stat_statements;

With the result being for PostgreSQL 13:

                     View "public.pg_stat_statements"
       Column        |       Type       | Collation | Nullable | Default
---------------------+------------------+-----------+----------+---------
 userid              | oid              |           |          |
 dbid                | oid              |           |          |
 toplevel            | boolean          |           |          |
 queryid             | bigint           |           |          |
 query               | text             |           |          |
 plans               | bigint           |           |          |
 total_plan_time     | double precision |           |          |
 min_plan_time       | double precision |           |          |
 max_plan_time       | double precision |           |          |
 mean_plan_time      | double precision |           |          |
 stddev_plan_time    | double precision |           |          |
 calls               | bigint           |           |          |
 total_exec_time     | double precision |           |          |
 min_exec_time       | double precision |           |          |
 max_exec_time       | double precision |           |          |
 mean_exec_time      | double precision |           |          |
 stddev_exec_time    | double precision |           |          |
 rows                | bigint           |           |          |
 shared_blks_hit     | bigint           |           |          |
 shared_blks_read    | bigint           |           |          |
 shared_blks_dirtied | bigint           |           |          |
 shared_blks_written | bigint           |           |          |
 local_blks_hit      | bigint           |           |          |
 local_blks_read     | bigint           |           |          |
 local_blks_dirtied  | bigint           |           |          |
 local_blks_written  | bigint           |           |          |
 temp_blks_read      | bigint           |           |          |
 temp_blks_written   | bigint           |           |          |
 blk_read_time       | double precision |           |          |
 blk_write_time      | double precision |           |          |
 wal_records         | bigint           |           |          |
 wal_fpi             | bigint           |           |          |
 wal_bytes           | numeric          |           |          |

On older PostgreSQL versions you might find different column names (e.g. the column previously named max_time is now max_exec_time). Always refer to the PostgreSQL® official documentation with the version you are using for accurate column matching.

Tip

You can write custom queries to pg_stat_statements to help you analyze recently run queries in your database.

Sort database queries based on total_exec_time

The following query, inspired by a GitHub repository, uses the pg_stat_statements view, shows the running queries sorted descending by total_exec_time, re-formats the calls column and deduces the prop_exec_time and sync_io_time:

SELECT interval '1 millisecond' * total_exec_time AS total_exec_time,
    to_char((total_exec_time/sum(total_exec_time) OVER()) * 100, 'FM90D0') || '%'  AS prop_exec_time,
    to_char(calls, 'FM999G999G999G990') AS calls,
    interval '1 millisecond' * (blk_read_time + blk_write_time) AS sync_io_time,
    query AS query
FROM pg_stat_statements
WHERE userid =
    (
        SELECT usesysid
        FROM pg_user
        WHERE usename = current_user
        LIMIT 1
    )
ORDER BY total_exec_time DESC
LIMIT 10;

You can run the above commands on your own PostgreSQL® to gather more information about how the recent queries are performing.

Tip

It is possible to discard the pg_stat_statements previously gathered statistics by using the following command:

SELECT pg_stat_statements_reset()

Find top queries with high I/O activity

The following SQL shows queries with their id and mean time in seconds. The result set is ordered based on the sum of blk_read_time and blk_write_time meaning that queries with the highest read/write are shown at the top.

SELECT userid::regrole,
    dbid,
    query,
    queryid,
    mean_time/1000 as mean_time_seconds
FROM pg_stat_statements
ORDER by (blk_read_time+blk_write_time) DESC
LIMIT 10;

See top time-consuming queries

Aside from the relevant information to the database, the following SQL retrieves the number of calls, consumption time in milliseconds as total_time_seconds, and the minimum, maximum, and mean times such query has ever been executed in milliseconds. The result set is ordered in descending order by mean_time showing the queries with most consumption time first.

SELECT userid::regrole,
    dbid,
    query,
    calls,
    total_time/1000 as total_time_seconds,
    min_time/1000 as min_time_seconds,
    max_time/1000 as max_time_seconds,
    mean_time/1000 as mean_time_seconds
FROM pg_stat_statements
ORDER by mean_time desc
LIMIT 10;

Check queries with high memory usage

The following SQL retrieves the query, its id, and relevant information about the database. The result set in this case is ordered by showing the queries with the highest memory usage at the top, summing the number of shared memory blocks returned from the cache (shared_blks_hit), and the number of shared memory blocks marked as “dirty” during a request needed to be written to disk (shared_blks_dirtied).

SELECT userid::regrole,
    dbid,
    queryid,
    query
FROM pg_stat_statements
ORDER by (shared_blks_hit+shared_blks_dirtied) DESC limit 10;

Tip

Once you have identified slow queries, you can inspect the query plan and execution using EXPLAIN ANALYZE to understand how you can optimise your design to improve the performance.

The how to optimize slow PostgreSQL® queries contains some common suggestion for query optimisation.