PostgreSQL Partition Manager (pg_partman) is an extension to create and manage both time-based and serial-based table partition sets. pg_partman simplifies managing table partitions based on time or serial IDs, automating their creation and maintenance. While it includes many options, only a few are typically needed, making it user-friendly. Partitioning in YugabyteDB refers to physically dividing large tables into smaller, more manageable tables to improve performance. Typically, tables with columns containing timestamps are subject to partitioning because of the historical and predictable nature of their data. YugabyteDB fully supports pg_partman, with some minor [limitations](#limitations) due to its distributed nature that don't hinder partition management. {{}} For more information on partitioning in YugabyteDB, see [Table partitioning](../../../explore/ysql-language-features/advanced-features/partitions/) {{}} {{}} For a full list of options, see the [pg_partman documentation](https://github.com/yugabyte/yugabyte-db/blob/master/src/postgres/third-party-extensions/pg_partman/doc/pg_partman.md) {{}} ## Enable pg_partman Enable the pg_partman extension separately for each database. To enable pg_partman for a specific database, create the partition maintenance schema (by default the `public` schema is selected), and then create the pg_partman extension. For example: ```sql CREATE SCHEMA partman; CREATE EXTENSION pg_partman WITH SCHEMA partman; ``` ## Use pg_partman To demonstrate its use, create the following sample database table, `orders`, partitioned based on a timestamp column: ```sql CREATE TABLE orders ( order_id SERIAL, order_date DATE NOT NULL, customer_id INT ) PARTITION BY RANGE (order_date); ``` ### create_parent function After you enable pg_partman, use the `create_parent()` function to set up table partitions. This function allows you to configure partitions in the schema designated for partition maintenance. Use the `create_parent()` function as follows: ```sql SELECT partman.create_parent( p_parent_table => 'public.orders', p_control => 'order_date', p_type => 'native', p_interval => 'monthly', p_premake => 5 ); ``` - p_parent_table - Parent partitioned table. - p_control - Column on which table partitioning is done. Must be an integer or time-based. - p_type - Either 'native' or 'partman'. - native - Use the native partitioning methods that are built into PostgreSQL 10+. - partman - Create a trigger-based partition set; this is [not available in YugabyteDB](#supported-partitioning-methods). - p_interval – The time interval or integer range for each partition. For example, daily, hourly, monthly, and so on. - p_premake – The number of partitions to create in advance to support new inserts. {{}} To learn more, see [Creation Functions](https://github.com/yugabyte/yugabyte-db/blob/master/src/postgres/third-party-extensions/pg_partman/doc/pg_partman.md#creation-functions) in the pg_partman documentation. {{}} ### part_config table The `part_config` table stores all configuration data for partition sets managed by pg_partman. A new row for the partition table is added after calling the [create_parent() function](#create-parent-function). The `part_config` table can be used to make decisions about: - What new partitions to create. - Which old partitions to drop or detach from the partition set. - What table updates are required. {{}} For more information, see [Tables](https://github.com/yugabyte/yugabyte-db/blob/master/src/postgres/third-party-extensions/pg_partman/doc/pg_partman.md#tables) in the pg_partman documentation {{}} An example is as follows: ```sql SELECT parent_table, control, partition_interval, premake, retention FROM partman.part_config; ``` ```output parent_table | control | partition_interval | premake | retention ---------------+------------+--------------------+---------+----------- public.orders | order_date | 1 mon | 5 | ``` You can update the `part_config` table as follows: ```sql UPDATE partman.part_config SET partition_interval = 2 months WHERE parent_table = 'public.orders'; ``` ### Maintain partitions using pg_cron [pg_cron](../extension-pgcron) is a cron-based job scheduler for PostgreSQL that runs inside a database as an extension. You can schedule the pg_partman `run_maintenance()` function, which automatically creates child tables and drops old child partitions, using pg_cron. For example: 1. Load the pg_cron extension as follows: ```sql CREATE EXTENSION pg_cron; ``` 1. Update the `part_config` table retention-related fields as follows: ```sql UPDATE partman.part_config SET retention = '3 months', retention_keep_table=true WHERE parent_table = 'public.orders'; ``` - retention - Sets the maximum retention for the table. - retention_keep_table - When set to true, specifies that the table is only detached from the partition set and isn't deleted automatically. 1. Schedule the `run_maintenance()` call using the `cron.schedule` procedure as follows: ```sql SELECT cron.schedule( 'Run maintenance job', '10 seconds', 'SELECT partman.run_maintenance()' ); ``` 1. Monitor the scheduled maintenance job by querying the `cron.job` table as follows: ```sql SELECT jobid, schedule, command, jobname FROM cron.job; ``` ```output jobid | schedule | command | jobname -------+------------+----------------------------------+----------+------------- 1 | 10 seconds | SELECT partman.run_maintenance() | Run maintenance job ``` ## Limitations The pg_partman extension has the following limitations in YugabyteDB. These limitations ensure that pg_partman features align with the capabilities and architecture of YugabyteDB, maintaining system performance and stability. You should adjust your use of partitioning strategies accordingly to adhere to these constraints. ### Supported partitioning methods Only native partitioning is supported in YugabyteDB. Non-native partitioning, which uses table inheritance and triggers to route data to the appropriate child tables, is not supported because YugabyteDB does not support table inheritance. Versions of pg_partman later than v4.7.4 have deprecated non-native partitioning, so migrating partition sets from non-native to native partitioning methods can be challenging. The following example shows an unsupported operation using non-native partitioning: ```sql CREATE SCHEMA partman; CREATE EXTENSION pg_partman WITH SCHEMA partman; CREATE TABLE orders ( order_id SERIAL, order_date DATE NOT NULL, customer_id INT ) PARTITION BY RANGE (order_date); SELECT partman.create_parent( p_parent_table => 'public.orders', p_control => 'order_date', p_type => 'partman', p_interval =>'monthly', p_premake => 5 ); ``` ```output ERROR: partman is not a valid partitioning type for pg_partman ``` ### ACCESS EXCLUSIVE LOCK The pg_partman `create_parent()` function requires an access exclusive lock on the parent table to create new child partitions. Currently, access exclusive locks are not supported in YugabyteDB, and are disabled in this function. ### Advisory locks [Advisory locks](../../../architecture/transactions/concurrency-control/#advisory-locks), used in some pg_partman functions to create or drop/delete partitioned tables, are not currently supported in pg_partman in YugabyteDB. Attempts to acquire these locks are disabled. ### Background worker process The pg_partman background worker process, which automates partition maintenance without the need of an external scheduler, is disabled in favor of using pg_cron to manage such tasks externally. ### undo_partition function The `undo_partition()` function, which moves data from child tables back to a target table, is disabled. As part of `undo_partition()`, all data from child partitions is moved out to the target table and after that the child table is detached from the parent table. This is done in an iterative manner, looping through each child. `undo_partition()` is disabled because YugabyteDB does not currently support transactional DDL operations (the DMLs (deleting data from child tables) and DDLs (detaching child tables) will cause conflicts), which are required for this functionality. For example: ```sql SELECT partman.undo_partition(p_parent_table => 'public.orders'); ``` ```output ERROR: undo_partition not supported yet in YugabyteDB ``` ### Default partition creation Creation and use of default partitions are disabled. If a default partition is created, `partition_data_proc` and `run_maintenance()` operations will exit early without performing any operation. For example: ```sql SELECT partman.create_parent( p_parent_table => 'public.orders', p_control => 'order_date', p_type => 'native', p_interval =>'monthly', p_premake => 5 ); ``` ```output create_parent --------------- t (1 row) ``` ```sql \d ``` ```output List of relations Schema | Name | Type | Owner --------+---------------------+-------------------+---------- public | orders | partitioned table | yugabyte public | orders_order_id_seq | sequence | yugabyte public | orders_p2024_06 | table | yugabyte public | orders_p2024_07 | table | yugabyte public | orders_p2024_08 | table | yugabyte public | orders_p2024_09 | table | yugabyte public | orders_p2024_10 | table | yugabyte public | orders_p2024_11 | table | yugabyte public | orders_p2024_12 | table | yugabyte public | orders_p2025_01 | table | yugabyte public | orders_p2025_02 | table | yugabyte public | orders_p2025_03 | table | yugabyte public | orders_p2025_04 | table | yugabyte (13 rows) No default partition created ``` If you try to create a default partition explicitly and call `run_maintenance()`, the run maintenance job is skipped as follows: ```sql CREATE TABLE orders_default PARTITION OF orders DEFAULT; SELECT partman.run_maintenance('public.orders'); ``` ```output NOTICE: Skip run maintenace job for table public.orders as it has default table attached run_maintenance ----------------- (1 row) ``` ### Colocated databases Registration of a partitioned table for maintenance by pg_partman is not allowed if the table belongs to a colocated database. For example: ```sql yugabyte=# CREATE DATABASE mo WITH colocation=true; yugabyte=# \c mo mo=# SELECT create_parent( p_parent_table => 'public.orders', p_control => 'order_date', p_type => 'native', p_interval =>'monthly', p_premake => 5); mo=# CREATE extension pg_partman; mo=# SELECT create_parent( p_parent_table => 'public.orders', p_control => 'order_date', p_type => 'native', p_interval =>'monthly', p_premake => 5); ``` ```output ERROR: Partition table public.orders is a colocated table hence registering it to pg_partman maintenance is not supported ``` ### xCluster Depending on the type of xCluster deployment, pg_partman may be used. Note that if you are [using pg_cron to manage tasks](#maintain-partitions-using-pg-cron): - In v2025.1.0 or earlier, or if you are using semi-automatic or manual mode, you must install pg_cron on a separate database that is not part of the xCluster configuration. - In v2025.1.1 or later _in automatic mode_, you can install pg_cron on the same database. For more information on xCluster limitations, refer to [Limitations](../../../architecture/docdb-replication/async-replication/#limitations). #### Transactional xCluster and xCluster DR **Automatic mode** The pg_partman maintenance cron job can only be enabled on the xCluster primary. Disable it on the xCluster standby. During disaster recovery switchover (or failover and repair): 1. Disable the pg_partman cron job on the original xCluster primary before initiating switchover. 1. After switchover or failover, enable the pg_partman cron job on the new primary. In v2025.1.1 and later, pg_cron will only execute on the Primary universe. On switchover, pg_cron will automatically switch over to the new Primary universe as well. **Semi-automatic mode** At setup, do the following: 1. On the primary universe, enable the pg_partman extension and the pg_audit extension. 1. On the standby universe, enable the pg_partman extension, and disable the pg_partman maintenance cron job. During normal operations, as DDLs occur on the primary universe, do the following: 1. Monitor the pg_audit log on the primary universe to detect partition-related DDLs. 1. When any partition-related DDLs are detected, manually run them also on the standby universe; this must be done within the WAL log retention time (which is 4 hours by default). **Manual DDL change handling mode** pg_partman is not supported. #### Non-transactional xCluster (including bi-directional) pg_partman is not recommended. Because pg_partman creates and drops tables, if you were to try to run pg_partman on both universes, the partitions might not be created at the same time on both sides. Even if they were identical, manual work is needed to add the new tables and remove dropped tables from the xCluster configuration. If you were to try to run pg_partman on just one side, then any tables created or dropped by pg_partman would have to be created or dropped on the target, which would require difficult manual administration. Although not recommended, it is possible to use pg_partman using the following steps: **Setup** 1. On one universe (call this universe A): - Enable the pg_partman extension. - Enable the pg_audit extension. Note: in a uni-directional replication configuration, choose universe A to be your source universe. In a bi-directional replication configuration, both universes act as source and target (usually for distinctly different tables, however). 1. On the other universe (call this universe B): - Enable the pg_partman extension. - Disable the pg_partman maintenance cron job. **Operation** During normal operations, as DDLs occur on the source universe, monitor the pg_audit log (on the source universe) to detect partition-related DDLs. When you detect partition-related DDLs, follow the instructions in [Handling DDL changes](/stable/deploy/multi-dc/async-replication/async-deployment/#handling-ddl-changes) to issue the same DDL command on the replica universe and include the table in xCluster replication. Note that the sequence of operations can vary by DDL command (for example, CREATE partition and DROP partition require different follow-up actions in different orders).