Skip to content
Permalink
Branch: master
Find file Copy path
Find file Copy path
@jcoleman jcoleman Fix typo 411b4d1 May 21, 2019
4 contributors

Users who have contributed to this file

@keithf4 @red-led @jcoleman @bclennox
691 lines (543 sloc) 97.4 KB

PostgreSQL Partition Manager Extension (pg_partman)

About

PostgreSQL Partition Manager is an extension to help make managing time or serial id based table partitioning easier. It has many options, but usually only a few are needed, so it's much easier to use than it may first appear (and definitely easier than implementing it yourself). Currenly the trigger functions only handle inserts to the parent table. Updates that would move a value from one partition to another are only supported in PostgreSQL 11 native partitioning. Some features of this extension have been expanded upon in the author's blog - http://www.keithf4.com/tag/pg_partman

As of version 3.0.1, this extension will support the native partitioning methods that are in the PostgreSQL 10 release. A trigger function is no longer necessary in native partitioning, but automatic child table creation is not handled natively, which is where this extension comes into play. Version 4.0.0 adds even more native support for features introduced in PG11 (easier index/fk inheritance, default partition).

For non-native partitioning, if you attempt to insert data into a partition set that contains data for a partition that does not exist, that data will be placed into the set's parent table. This is preferred over automatically creating new partitions to match that data since a mistake that is causing non-partitioned data to be inserted could cause a lot of unwanted child tables to be made as well as contention due to transactional DDL. The check_default() function provides monitoring for any data getting inserted into the parent/default table and the partition_data_* set of functions can easily partition that data for you if it is valid data. That is much easier than having to clean up potentially hundreds or thousands of unwanted partitions. And also better than throwing an error and losing the data! For native partitioning, inserting data with no relevant child causes an error in PostgreSQL 10. A default partition for native is only available in PostgreSQL 11+.

Note that future child table creation is based on the data currently in the partition set. This means that if you put "future" data in, newly created tables will be based off that value. This may cause intervening data to go to the parent/default as stated above if no child table exists. It is recommended that you set the premake value high enough to encompass your expected data range being inserted. And for non-native partitioning, set the optimize_trigger value to efficiently handle your most frequent data range. See below for further explanations on these configuration values.

If you have an existing partition set and you'd like to migrate it to pg_partman, please see the migration.md file in the doc folder. This is for non-native partitioning only at this time. I'm working on a migration plan for getting non-native partition sets moved to native partition sets. If it all works out, will be included in a future version of pg_partman.

Child Table Property Inheritance

For this extension, most of the attributes of the child partitions are all obtained from the original parent. This includes defaults, indexes (primary keys, unique, clustering, etc), foreign keys (optional), tablespace, & constraints. This also includes the OID and UNLOGGED table properties. Note that for PostgreSQL 10, indexes,foreign keys, & tablespaces for native partitioning are done through a template table instead. For PG11, only unique indexes that don't include the parition column require the template table; all other properties are managed by the parent again.

Privileges & ownership are inherited by default for non-native partitioning, but not for native partitioning. Unless you need direct access to the child tables, this should not be needed. You can set the inherit_privileges option if this is needed (see config table information below).

When privileges are being inherited, whenever a new partition is created it will obtain its privilege & ownership information from what the parent has at that time. Previous partition privileges are not changed. If previous partitions require that their privileges be updated, a separate function is available. This is kept as a separate process due to being an expensive operation when the partition set grows larger.

For PG10 and older, the defaults, indexes, tablespace & constraints on the parent (or template table) are only applied to newly created partitions and are not retroactively set on ones that already existed. For P11, only indexes applied to the template table are not automatically applied to new children; all other properties applied to the parent should automatically apply to all child tables.

The new IDENTITY feature introduced in PG10 is only supported in natively partitioned tables and the automatic generation of new sequence values using this feature is only supported when data is inserted through the parent table, not directly into the children.

IMPORTANT NOTE: The template table feature in use for PostgreSQL 10+ to handle certain features is only a temporary solution to help speed up native partitioning adoption. As things are handled better natively (as they were in PG11), the use of the template table will be phased out quickly from pg_partman. So please plan ahead for quick major version upgrades if you use this feature.

Sub-partitioning

Sub-partitioning with multiple levels is supported, but it is of very limited use in PostgreSQL and provides next to NO PERFORMANCE BENEFIT outside of extremely large data in a single partition set (100s of terabytes, petabytes). If you're looking for performance benefits, adjust your partition interval before considering sub-partitioning. It's main use is in data organization and retention management.

You can do time->time, id->id, time->id and id->time. There is no set limit on the level of subpartitioning you can do, but be sensible and keep in mind performance considerations on managing many tables in a single inheritance set. Also, if the number of tables in a single partition set gets very high, you may have to adjust the max_locks_per_transaction postgresql.conf setting above the default of 64. Otherwise you may run into shared memory issues or even crash the cluster. If you have contention issues when run_maintenance() is called for general maintenance of all partition sets, you can set the automatic_maintenance column in the part_config table to false if you do not want that general call to manage your subpartition set. But you must then call run_maintenance(parent_table) directly, and often enough, to have to future partitions made. If you're on PG11+, you can use the new run_maintenance_proc() procedure to cause less contention issues since it automatically commits after each partition set's maintenance.

PUBLICATION/SUBSCRIPTION for logical replication is NOT supported with native sub-partitioning.

See the create_parent_sub() & run_maintenance() functions below for more information.

Retention

If you don't need to keep data in older partitions, a retention system is available to automatically drop unneeded child partitions. By default, they are only uninherited/detached not actually dropped, but that can be configured if desired. There is also a method available to dump the tables out if they don't need to be in the database anymore but still need to be kept. To set the retention policy, enter either an interval or integer value into the retention column of the part_config table. For time-based partitioning, the interval value will set that any partitions containing only data older than that will be dropped. For id-based partitioning, the integer value will set that any partitions with an id value less than the current maximum id value minus the retention value will be dropped. For example, if the current max id is 100 and the retention value is 30, any partitions with id values less than 70 will be dropped. The current maximum id value at the time the drop function is run is always used. Keep in mind that for subpartition sets, when a parent table has a child dropped, if that child table is in turn partitioned, the drop is a CASCADE and ALL child tables down the entire inheritance tree will be dropped. Also note that a partition set managed by pg_partman must always have at least one child, so retention will never drop the last child table in a set.

Constraint Exclusion

One of the big advantages of partitioning is a feature called constraint exclusion (see docs for explanation of functionality and examples http://www.postgresql.org/docs/current/static/ddl-partitioning.html#DDL-PARTITIONING-CONSTRAINT-EXCLUSION). The problem with most partitioning setups however, is that this will only be used on the partitioning control column. If you use a WHERE condition on any other column in the partition set, a scan across all child tables will occur unless there are also constraints on those columns. And predicting what a column's values will be to precreate constraints can be very hard or impossible. pg_partman has a feature to apply constraints on older tables in a partition set that may no longer have any edits done to them ("old" being defined as older than the optimize_constraint config value). It checks the current min/max values in the given columns and then applies a constraint to that child table. This can allow the constraint exclusion feature to potentially eliminate scanning older child tables when other columns are used in WHERE conditions. Be aware that this limits being able to edit those columns, but for the situations where it is applicable it can have a tremendous affect on query performance for very large partition sets. So if you are only inserting new data this can be very useful, but if data is regularly being inserted/updated throughout the entire partition set, this is of limited use. Functions for easily recreating constraints are also available if data does end up having to be edited in those older partitions. Note that constraints managed by PG Partman SHOULD NOT be renamed in order to allow the extension to manage them properly for you. For a better example of how this works, please see this blog post: http://www.keithf4.com/managing-constraint-exclusion-in-table-partitioning

NOTE: This may not work with sub-partitioning. It will work on the first level of partitioning, but is not guarenteed to work properly on further sub-partition sets depending on the interval combinations and the optimize_constraint value. Ex: Weekly -> Daily with a daily optimize_constraint of 7 won't work as expected. Weekly constraints will get created but daily sub-partition ones likely will not.

Custom Time Interval Considerations

The list of time intervals given for create_parent() below are optimized to work as fast as possible with non-native, trigger-based partitioning. Intervals other than those values are possible, but performance will take a non-trivial hit to allow such flexibility. For native partitioning, unlike pg_partman's trigger based method, there's no differing method of partitioning for any given intervals. All possible intervals that use the native method have the same performance characteristics and are better than any trigger-based method. It is HIGHLY recommended to upgrade to PG10 if you need a partitioning interval different than the optimized ones that pg_partman provides.

The smallest time interval supported is 1 second and the upper limit is bounded by the minimum and maximum timestamp values that PostgreSQL supports (http://www.postgresql.org/docs/current/static/datatype-datetime.html). The smallest integer interval supported at this time is 10.

When first running create_parent() to create a partition set, intervals less than a day round down when determining what the first partition to create will be. Intervals less than 24 hours but greater than 1 minute use the nearest hour rounded down. Intervals less than 1 minute use the nearest minute rounded down. However, enough partitions will be made to support up to what the real current time is. This means that when create_parent() is run, more previous partitions may be made than expected and all future partitions may not be made. The first run of run_maintenance() will fix the missing future partitions. This happens due to the nature of being able to support custom time intervals. Any intervals greater than or equal to 24 hours should set things up as would be expected.

Keep in mind that for intervals equal to or greater than 100 years, the extension will use the real start of the century or millennium to determine the partition name & constraint rules. For example, the 21st century and 3rd millennium started January 1, 2001 (not 2000). This also means there is no year "0". It's much too difficult to try to work around this and make nice "even" partition names & rules to handle all possible time periods people may need. Blame the Gregorian creators.

Naming Length Limits

PostgreSQL has an object naming length limit of 63 characters. If you try and create an object with a longer name, it truncates off any characters at the end to fit that limit. This can cause obvious issues with partition names that rely on having a specifically named suffix. PG Partman automatically handles this for all child tables, trigger functions and triggers. It will truncate off the existing parent table name to fit the required suffix. Be aware that if you have tables with very long, similar names, you may run into naming conflicts if they are part of separate partition sets. With serial based partitioning, be aware that over time the table name will be truncated more and more to fit a longer partition suffix. So while the extention will try and handle this edge case for you, it is recommended to keep table names that will be partitioned as short as possible.

Unique Constraints & Upsert

Table inheritance in PostgreSQL does not allow a primary key or unique index/constraint on the parent to apply to all child tables. The constraint is applied to each individual table, but not on the entire partition set as a whole. For example, this means a careless application can cause a primary key value to be duplicated in a partition set. In the mean time, a python script is included with pg_partman that can provide monitoring to help ensure the lack of this feature doesn't cause long term harm. See check_unique_constraint.py in the Scripts section.

IMPORTANT NOTE: INSERT ... ON CONFLICT SUPPORT IN PG10 AND OLDER WILL BE REMOVED FROM PG_PARTMAN IN THE NEAR FUTURE. If you require this feature, it is highly recommended you plan on upgrading to PostgreSQL 11 for proper update and upsert support.

For non-native partitioning and PG10 native partitioning, INSERT ... ON CONFLICT (upsert) is supported in the partitioning trigger as well as native partitioning, but is very limited. The major limitations are that the constraint violations that would trigger the ON CONFLICT clause only occur on individual child tables that actually contain data due to reasons explained above. Of a larger concern than data duplication is an ON CONFLICT DO UPDATE clause which may not fire and cause wildly inconsistent data if not accounted for. For situations where only new data is being inserted, upsert can provide significant performance improvements. However, if you're relying on data in older partitions to cause a constraint violation that upsert would normally handle, it likely will not work. Also, if the resulting UPDATE would end up violating the partitioning constraint of that chld table, it will fail. Neither pg_partman & PG10 native partitioning currently support UPDATES that would require moving a row from one child table to another. This is only supported in PG11+.

Upsert is optional, turned off by default and was only included since there was no native support on the core PostgreSQL roadmap at the time it was implemented. At this time, if you have not implemented this feature, it is highly recommended you upgrade to PG11.

Logging/Monitoring

The PG Jobmon extension (https://github.com/omniti-labs/pg_jobmon) is optional and allows auditing and monitoring of partition maintenance. If jobmon is installed and configured properly, it will automatically be used by partman with no additional setup needed. Jobmon can also be turned on or off individually for each partition set by using the jobmon column in the part_config table or with the option to create_parent() during initial setup. Note that if you try to partition pg_jobmon's tables you MUST set the jobmon option in create_parent() to false, otherwise it will be put into a permanent lockwait since pg_jobmon will be trying to write to the table it's trying to partition. By default, any function that fails to run successfully 3 consecutive times will cause jobmon to raise an alert. This is why the default pre-make value is set to 4 so that an alert will be raised in time for intervention with no additional configuration of jobmon needed. You can of course configure jobmon to alert before (or later) than 3 failures if needed. If you're running partman in a production environment it is HIGHLY recommended to have jobmon installed and some sort of 3rd-party monitoring configured with it to alert when partitioning fails (Nagios, Circonus, etc).

Background Worker

With PostgreSQL 9.4, the ability to create custom background workers and dynamically load them during runtime was introduced. pg_partman's BGW is basically just a scheduler that runs the run_maintenance() function for you so that you don't have to use an external scheduler (cron, etc). Right now it doesn't do anything differently than calling run_maintenance() directly, but that may change in the future. See the README.md file for installation instructions. If you need to call run_maintenance() directly on any specific partition sets, you will still need to do so manually using an outside scheduler. This only maintains partition sets that have automatic_maintenance in **part_config** set to true. LOG messages are output to the normal PostgreSQL log file to indicate when the BGW runs. Additional logging messages are available if log_min_messages is set to "DEBUG1".

The following configuration options are available to add into postgresql.conf to control the BGW process:

  • pg_partman_bgw.dbname
    • Required. The database(s) that run_maintenance() will run on. If more than one, use a comma separated list. If not set, BGW will do nothing.
  • pg_partman_bgw.interval
    • Number of seconds between calls to run_maintenance(). Default is 3600 (1 hour).
    • See further documenation below on suggested values for this based on partition types & intervals used.
  • pg_partman_bgw.role
    • The role that run_maintenance() will run as. Default is "postgres". Only a single role name is allowed.
  • pg_partman_bgw.analyze
    • Same purpose as the p_analyze argument to run_maintenance(). See below for more detail. Set to 'on' for TRUE. Set to 'off' for FALSE. Default is 'on'.
  • pg_partman_bgw.jobmon
    • Same purpose as the p_jobmon argument to run_maintenance(). See below for more detail. Set to 'on' for TRUE. Set to 'off' for FALSE. Default is 'on'.

If for some reason the main background worker process crashes, it is set to try and restart every 10 minutes. Check the postgres logs for any issues if the background worker is not starting.

As of version 4.0.0, the background worker still uses the normal run_maintenance() function. An option to use the new procedure is in the works.

Extension Objects

As of 4.0.0, SECURITY DEFINER has been removed from a larger majority of functions in pg_partman. As of version 4.1.0, requiring a superuser to use pg_partman is now completely optional for native partitioning. It is still required for extension installation and non-native partitioning. From now on, the roles that run pg_partman functions and maintenance must have ownership of all partition sets they manage and permissions to create objects in any schema that will contain partition sets that it manages. For ease of use and privilege management, it is recommended to create a role dedicated to partition management. Please see the main README.md file for role & privileges setup instructions.

As a note for people that were not aware, you can name arguments in function calls to make calling them easier and avoid confusion when there are many possible arguments. If a value has a default listed, it is not required to pass a value to that argument. As an example: SELECT create_parent('schema.table', 'col1', 'partman', 'daily', p_start_partition := '2015-10-20');

Creation Functions

create_parent(p_parent_table text, p_control text, p_type text, p_interval text, p_constraint_cols text[] DEFAULT NULL, p_premake int DEFAULT 4, p_automatic_maintenance text DEFAULT 'on', p_start_partition text DEFAULT NULL, p_inherit_fk boolean DEFAULT true, p_epoch text DEFAULT 'none', p_upsert text DEFAULT '', p_publications text[] DEFAULT NULL, p_trigger_return_null boolean DEFAULT true, p_template_table text DEFAULT NULL, p_jobmon boolean DEFAULT true, p_debug boolean DEFAULT false) RETURNS boolean

  • Main function to create a partition set with one parent table and inherited children. Parent table must already exist. Please apply all defaults, indexes, constraints, privileges & ownership to parent table so they will propagate to children. For native partitioning, the parent table must already be declared as such and config options passed to this function must match that definition. See notes above about handling indexes & foreign keys.
  • An ACCESS EXCLUSIVE lock is taken on the parent table during the running of this function. No data is moved when running this function, so lock should be brief.
  • For PG11+, a default partition is automatically created. A "_default" suffix is added onto the current table name.
  • p_parent_table - the existing parent table. MUST be schema qualified, even if in public schema.
  • p_control - the column that the partitioning will be based on. Must be a time or integer based column.
  • p_type - one of the following values to set the partitioning type that will be used:
    • native
      • Use the native partitioning methods that are built into PostgreSQL 10+.
      • For PG11+, it is highly recommended that native partitioning be used over trigger-based partitioning. PG10 is still lacking significant features for native partitioning, so please see notes above for more info.
      • Provides significantly better write & read performance than "partman" partitioning.
      • Child table creation is kept up to date by running run_maintenance(_proc). There is no trigger maintenance.
    • partman
      • Create a trigger-based partition set using pg_partman's method of partitioning.
      • Whether it is time or serial based is determined by the control column's data type and if the p_epoch flag is set.
      • The number of partitions most efficiently managed behind and ahead of the current one is determined by the optimize_trigger config value in the part_config table (default of 4 means data for 4 previous and 4 future partitions are handled best).
      • Beware setting the optimize_trigger value too high as that will lessen the efficiency boost.
      • Inserts to the parent table outside the optimize_trigger window will go to the proper child table if it exists, but performance will be degraded due to the higher overhead of handling that condition.
      • If the child table does not exist for the value given, the row will go to the parent.
      • Child table creation & trigger function is kept up to date by the run_maintenance() function.
  • p_interval - the time or integer range interval for each partition. No matter the partitioning type, value must be given as text. The generic intervals of "yearly -> quarter-hour" are for time partitioning and giving one of these explicit values when using pg_partman's trigger-based partitioning will allow significantly better performance than using an arbitrary time interval. For native partitioning, any interval value is valid and will have the same performance which is always better than trigger-based.
    • yearly - One partition per year
    • quarterly - One partition per yearly quarter. Partitions are named as YYYYqQ (ex: 2012q4)
    • monthly - One partition per month
    • weekly - One partition per week. Follows ISO week date format (http://en.wikipedia.org/wiki/ISO_week_date). Partitions are named as IYYYwIW (ex: 2012w36)
    • daily - One partition per day
    • hourly - One partition per hour
    • half-hour - One partition per 30 minute interval on the half-hour (1200, 1230)
    • quarter-hour - One partition per 15 minute interval on the quarter-hour (1200, 1215, 1230, 1245)
    • <interval> - Any other interval besides the values above that is valid for the PostgreSQL interval type. Note this will have a significant performance penalty if not using native partitioning. Do not type cast the parameter value, just leave as text.
    • <integer> - For ID based partitions, the integer value range of the ID that should be set per partition. Enter this as an integer in text format ('100' not 100). Must be greater than or equal to 10.
  • p_constraint_cols - an optional array parameter to set the columns that will have additional constraints set. See the About section above for more information on how this works and the apply_constraints() function for how this is used.
  • p_premake - is how many additional partitions to always stay ahead of the current partition. Default value is 4. This will keep at minimum 5 partitions made, including the current one. For example, if today was Sept 6th, and premake was set to 4 for a daily partition, then partitions would be made for the 6th as well as the 7th, 8th, 9th and 10th. Note some intervals may occasionally cause an extra partition to be premade or one to be missed due to leap years, differing month lengths, daylight savings (on non-UTC systems), etc. This won't hurt anything and will self-correct. If partitioning ever falls behind the premake value, normal running of run_maintenance() and data insertion should automatically catch things up.
  • p_automatic_maintenance - parameter to set whether maintenance is managed automatically when run_maintenance() is called without a table parameter or by the background worker process. Current valid values are "on" and "off". Default is "on". When set to off, run_maintenance() can still be called on an individual partition set by passing it as a paramter to the function. See run_mainteanance in Maintenance Functions section below for more info.
  • p_start_partition - allows the first partition of a set to be specified instead of it being automatically determined. Must be a valid timestamp (for time-based) or positive integer (for id-based) value. Be aware, though, the actual parameter data type is text. For time-based partitioning, all partitions starting with the given timestamp up to CURRENT_TIMESTAMP (plus premake) will be created. For id-based partitioning, only the partition starting at the given value (plus premake) will be made.
  • p_inherit_fk - allows pg_partman to automatically manage inheriting any foreign keys that exist on the parent (or template for native) table to all its children. Defaults to TRUE. Note this option is only relevant for PostgreSQL 10 and older. PG11+ automatically inherits any foreign keys placed on the parent and is not optional.
  • p_epoch - tells pg_partman that the control column is an integer type, but actually represents and epoch time value. You can also specify whether the value is seconds or milliseconds. Valid values for this option are: 'seconds', 'milliseconds' & 'none'. The default is 'none'. All triggers, constraints & table names will be time-based. In addition to a normal index on the control column, be sure you create a functional, time-based index on the control column (to_timestamp(controlcolumn)) as well so this works efficiently.
  • p_upsert - adds upsert to insert queries in the partition trigger to allow handeling of conflicts Defaults to '' (empty string) which means it's inactive.
    • IMPORTANT NOTE: As stated above, this feature will be deprecated in the near future once PG11 as been out for a while. Please plan on migrating to PG11 soon if you use this feature.
    • the value entered here is the entire ON CONFLICT clause which will then be appended to the INSERT statement(s) in the trigger
    • Ex: to ignore conflicting rows on a table with primary key "id" set p_upsert to 'ON CONFLICT (id) DO NOTHING'
    • Ex: to update a conflicting row on a table with columns (id(pk), val) set p_upsert to 'ON CONFLICT (id) DO UPDATE SET val=EXCLUDED.val'
    • Requires postgresql 9.5
    • See About section above for more info.
  • p_publications - Option to add child tables to publications for use with logical replication. Value is an array list of publication names, so multiple publications can be added to each child. Currently does not support sub-partitioning for native partition sets since a publication cannot be added to the parent of a natively partitioned table.
  • p_trigger_return_null - Only applies to non-native, trigger-based partitioning. Boolean value that allows controlling the behavior of the partition trigger RETURN. By default this is true and the trigger returns NULL to prevent data going into the parent table as well as the children. However, if you have multiple triggers and are relying on the return to be the NEW column value, this can cause a problem. Setting this config value to false will cause the partition trigger to RETURN NEW. You are then responsible for handling the return value in another trigger appropriately. Otherwise, this will cause new data to go into both the child and parent table of the partition set.
  • p_template_table - For native partitioning in PG10, indexes, foreign keys & tablespaces cannot be set on the parent table. For PG11, only unique indexes that don't include the partition key cannot be created on the parent. Therefore, if you want them to be automatically created on child tables, they must be managed elsewhere. If you do not pass a value here, a template table will automatically be made for you in same schema that pg_partman was installed to. Note that until indexes, foreign keys or tablespaces are made on the template, no child tables will have any. Use the python scripts to reapply the indexes and foreign keys to the partition set when the template table is ready. For tablespaces, you will have to manually move any previously existing child tables. If you pre-create a template table and pass its name here, then the initial child tables will obtain these properties immediately.
  • p_jobmon - allow pg_partman to use the pg_jobmon extension to monitor that partitioning is working correctly. Defaults to TRUE.
  • p_debug - turns on additional debugging information.

create_sub_parent(p_top_parent text, p_control text, p_type text, p_interval text, p_native_check text DEFAULT NULL, p_constraint_cols text[] DEFAULT NULL, p_premake int DEFAULT 4, p_start_partition text DEFAULT NULL, p_inherit_fk boolean DEFAULT true, p_epoch text DEFAULT 'none', p_upsert text DEFAULT '', p_trigger_return_null boolean DEFAULT true, p_jobmon boolean DEFAULT true, p_debug boolean DEFAULT false) RETURNS boolean

  • Create a subpartition set of an already existing partitioned set. See important notes about Subpartitioning above.
  • p_top_parent - This parameter is the parent table of an already existing partition set. It tells pg_partman to turn all child tables of the given partition set into their own parent tables of their own partition sets using the rest of the parameters for this function.
  • p_native_check - Turning an existing native partition set into a sub-partitioned set is a destructive process. A table must be declared natively partitioned at creation time and cannot be altered later. Therefore existing child tables must be dropped and recreated as partitioned parent tables. This flag is here to help ensure this function is not run without prior knowledge that all data in the partition set will be destroyed as part of the creation process. It must be set to "yes" to proceed with sub-partitioning a native partition set. This option can be ignored if you are created a trigger-based pg_partman partition set.
  • All other parameters to this function have the same exact purpose as those of create_parent(), but instead are used to tell pg_partman how each child table shall itself be partitioned.
  • For example if you have an existing partition set done by year and you then want to partition each of the year partitions by day, you would use this function.
  • It is advised that you keep table names short for subpartition sets if you plan on relying on the table names for organization. The suffix added on to the end of a table name is always guarenteed to be there for whatever partition type is active for that set, but if the total length is longer than 63 chars, the original name will get truncated. Longer table names may cause the original parent table names to be truncated and possibly cut off the top level partitioning suffix. I cannot control this and made the requirement that the lowest level partitioning suffix survives.
  • Note that for the first level of subpartitions, the p_parent_table argument you originally gave to create_parent() would be the exact same value you give to create_sub_parent(). If you need further subpartitioning, you would then start giving create_sub_parent() a different value (the child tables of the top level partition set).
  • For native partitioning, the template table that is already set for the given p_top_parent will automatically be used.

partition_data_time(p_parent_table text, p_batch_count int DEFAULT 1, p_batch_interval interval DEFAULT NULL, p_lock_wait numeric DEFAULT 0, p_order text DEFAULT 'ASC', p_analyze boolean DEFAULT true, p_source_table text DEFAULT NULL) RETURNS bigint

  • This function is used to partition data that may have existed prior to setting up the parent table as a time-based partition set. It also fixes data that accidentally gets inserted into the parent table (trigger-based only) or default table (native, PG11+ only).
  • If the needed partition does not exist, it will automatically be created. If the needed partition already exists, the data will be moved there.
  • If you are trying to partition a large amount of data automatically, it is recommended to either use the partiton_data.py script to commit data in smaller batches. Or if you're on PG11+, use the partition_data_proc() procedure to do the same thing. This will greatly reduce issues caused by long running transactions and data contention.
  • For sub-partitioned sets, you must start partitioning data at the highest level and work your way down each level. This means you must first run this function before running create_sub_parent() to create the additional partitioning levels. Then continue running this function again on each new sub-parent once they're created. See the pg_partman_howto.md document for a full example. IMPORTANT NOTE: this may not work as expected for native partitioning since subpartitioning a native set in pg_partman is a destructive operation. See create_sub_parent().
  • p_parent_table - the existing parent table. For non-native partitioning, this is assumed to be where the unpartitioned data is located. MUST be schema qualified, even if in public schema.
  • p_batch_interval - optional argument, only relevant for non-native partitioning. A time interval of how much of the data to move. This can be smaller than the partition interval, allowing for very large sized partitions to be broken up into smaller commit batches. Defaults to the configured partition interval if not given or if you give an interval larger than the partition interval. NOTE: This option CANNOT be used when moving data out of a default partition in PostgreSQL 11+.
  • p_batch_count - optional argument, how many times to run the batch_interval in a single call of this function. Default value is 1. For native partitioning, this sets how many child tables will be processed in a single run.
  • p_lock_wait - optional argument, sets how long in seconds to wait for a row to be unlocked before timing out. Default is to wait forever.
  • p_order - optional argument, by default data is migrated out of the parent in ascending order (ASC). Allows you to change to descending order (DESC).
  • p_analyze - optional argument, by default whenever a new child table is created, an analyze is run on the parent table of the partition set to ensure constraint exclusion works. This analyze can be skipped by setting this to false and help increase the speed of moving large amounts of data. If this is set to false, it is highly recommended that a manual analyze of the partition set be done upon completion to ensure statistics are updated properly.
  • p_source_table - This option can be used when you need to move data into a natively partitioned set. Pass a schema qualified tablename to this parameter and any data in that table will be MOVED to the partition set designated by p_parent_table, creating any child tables as needed.
  • Returns the number of rows that were moved from the parent table to partitions. Returns zero when source/parent table is empty and partitioning is complete.

partition_data_id(p_parent_table text, p_batch_count int DEFAULT 1, p_batch_interval bigint DEFAULT NULL, p_lock_wait numeric DEFAULT 0, p_order text DEFAULT 'ASC', p_analyze boolean DEFAULT true, p_source_table text DEFAULT NULL) RETURNS bigint

  • This function is used to partition data that may have existed prior to setting up the parent table as a serial id partition set. It also fixes data that accidentally gets inserted into the parent (trigger-based only).
  • If the needed partition does not exist, it will automatically be created. If the needed partition already exists, the data will be moved there.
  • If you are trying to partition a large amount of data automatically, it is recommended to either use the partiton_data.py script to commit data in smaller batches. Or if you're on PG11+, use the partition_data_proc() procedure to do the same thing. This will greatly reduce issues caused by long running transactions and data contention.
  • For sub-partitioned sets, you must start partitioning data at the highest level and work your way down each level. This means you must first run this function before running create_sub_parent() to create the additional partitioning levels. Then continue running this function again on each new sub-parent once they're created. See the pg_partman_howto.md document for a full example. IMPORTANT NOTE: this may not work as expected for native partitioning since subpartitioning a native set in pg_partman is a destructive operation. See create_sub_parent().
  • p_parent_table - the existing parent table. For non-native partitioning, this is assumed to be where the unpartitioned data is located. MUST be schema qualified, even if in public schema.
  • p_batch_interval - optional argument, only relevant for non-native partitioning. A time interval of how much of the data to move. This can be smaller than the partition interval, allowing for very large sized partitions to be broken up into smaller commit batches. Defaults to the configured partition interval if not given or if you give an interval larger than the partition interval. NOTE: This option CANNOT be used when moving data out of a default partition in PostgreSQL 11+.
  • p_batch_count - optional argument, how many times to run the batch_interval in a single call of this function. Default value is 1. For native partitioning, this sets how many child tables will be processed in a single run.
  • p_lock_wait - optional argument, sets how long in seconds to wait for a row to be unlocked before timing out. Default is to wait forever.
  • p_order - optional argument, by default data is migrated out of the parent in ascending order (ASC). Allows you to change to descending order (DESC).
  • p_analyze - optional argument, by default whenever a new child table is created, an analyze is run on the parent table of the partition set to ensure constraint exclusion works. This analyze can be skipped by setting this to false and help increase the speed of moving large amounts of data. If this is set to false, it is highly recommended that a manual analyze of the partition set be done upon completion to ensure statistics are updated properly.
  • p_source_table - This option can be used when you need to move data into a natively partitioned set. Pass a schema qualified tablename to this parameter and any data in that table will be MOVED to the partition set designated by p_parent_table, creating any child tables as needed.
  • Returns the number of rows that were moved from the parent table to partitions. Returns zero when source/parent table is empty and partitioning is complete.

partition_data_proc (p_parent_table text, p_interval text DEFAULT NULL, p_batch int DEFAULT NULL, p_wait int DEFAULT 1, p_source_table text DEFAULT NULL, p_order text DEFAULT 'ASC', p_lock_wait int DEFAULT 0, p_lock_wait_tries int DEFAULT 10, p_quiet boolean DEFAULT false)

  • A procedure that can partition data in distinct commit batches to avoid long running transactions and data contention issues.
  • Only works with PostgreSQL 11+
  • Calls either partition_data_time() or partition_data_id() in a loop depending on partitioning type.
  • p_parent_table - Parent table of an already created partition set.
  • p_interval - Value that is passed on to the partitioning function as p_batch_interval argument. Use this to set an interval smaller than the partition interval to commit data in smaller batches. Defaults to the partition interval if not given. NOTE: This option CANNOT be used when moving data out of a default partition in PostgreSQL 11+.
  • p_batch - How many times to loop through the value given for p_interval. If p_interval not set, will use default partition interval and make at most -b partition(s). Procedure commits at the end of each individual batch. (NOT passed as p_batch_count to partitioning function). If not set, all data in the parent/source table will be partitioned in a single run of the procedure.
  • p_wait - Cause the procedure to pause for a given number of seconds between commits (batches) to reduce write load
  • p_source_table - Same as the p_source_table option in the called partitioning function.
  • p_order - Allows you to specify the order that data is migrated from the parent/default to the children, either ascending (ASC) or descending (DESC). Default is ASC.
  • p_lock_wait - Parameter passed directly through to the underlying partition_data_*() function. Number of seconds to wait on rows that may be locked by another transaction. Default is to wait forever (0).
  • p_lock_wait_tries - Parameter to set how many times the procedure will attempt waiting the amount of time set for p_lock_wait. Default is 10 tries.
  • p_quiet - Procedures cannot return values, so by default it emmits NOTICE's to show progress. Set this option to silence these notices.

create_partition_time(p_parent_table text, p_partition_times timestamptz[], p_analyze boolean DEFAULT true, p_debug boolean DEFAULT false) RETURNS boolean

  • This function is used to create child partitions for the given parent table.
  • Normally this function is never called manually since partition creation is managed by run_maintenance(). But if you need to force the creation of specific child tables outside of normal maintenance, this function can make it easier.
  • For non-native partitioning, you may also need to call create_function_time() to update the partitioning trigger if you created partitions in the "current" optimization window.
  • p_parent_table - parent table to create new child table(s) in.
  • p_partition_times - An array of timestamptz values to create children for. If the child table does not exist, it will be created. If it does exist, it will not be created and the function will still exit cleanly. Be aware that the value given will be used as the lower boundary for the child table and also influence the name given to the child table. So ensure the timestamp value given is consistent with other children or you may encounter a gap in value coverage.
  • p_analyze - If a new child table is created, an analyze is normally kicked off so that the statistics are aware of the constraint boundaries for constraint exclusion. For larger partition sets, this analyze can take a long time. Set this to false to skip this automatic analyze.
  • Returns TRUE if any child tables were created for the given timestamptz values. Returns false if no child tables were created.

create_partition_id(p_parent_table text, p_partition_ids bigint[], p_analyze boolean DEFAULT true, p_debug boolean DEFAULT false) RETURNS boolean

  • This function is used to create child partitions for the given parent table.
  • Normally this function is never called manually since partition creation is managed by run_maintenance(). But if you need to force the creation of specific child tables outside of normal maintenance, this function can make it easier.
  • For non-native partitioning, you may need to also call create_function_id() to update the partitioning trigger if you created partitions in the "current" optimization window.
  • p_parent_table - parent table to create new child table(s) in.
  • p_partition_ids - An array of integer values to create children for. If the child table does not exist, it will be created. If it does exist, it will not be created and the function will still exit cleanly. Be aware that the value given will be used as the lower boundary for the child table and also influence the name given to the child table. So ensure the integer value given is consistent with other children or you may encounter a gap in value coverage.
  • p_analyze - If a new child table is created, an analyze is normally kicked off so that the statistics are aware of the constraint boundaries for constraint exclusion. For larger partition sets, this analyze can take a long time. Set this to false to skip this automatic analyze.
  • Returns TRUE if any child tables were created for the given integer values. Returns false if no child tables were created.

create_function_time(p_parent_table text, p_job_id bigint DEFAULT NULL) RETURNS void

  • This function is used to create the trigger function for non-native time-based partitioning.
  • Normally this function is never called manually since function creation is managed by run_maintenance(). But if you need to force the re-creation of the trigger function, this will let you do that.
  • p_parent_table - parent table to recreate trigger function on.
  • The p_job_id parameter is optional. It's for internal use and allows job logging to be consolidated into the original job that called this function if applicable.

create_function_id(p_parent_table text, p_job_id bigint DEFAULT NULL) RETURNS void

  • This function is used to create the trigger function for non-native serial partitioning.
  • Normally this function is never called manually since function creation is managed by run_maintenance(). But if you need to force the re-creation of the trigger function, this will let you do that.
  • p_parent_table - parent table to recreate trigger function on.
  • The p_job_id parameter is optional. It's for internal use and allows job logging to be consolidated into the original job that called this function if applicable.

Maintenance Functions

run_maintenance(p_parent_table text DEFAULT NULL, p_analyze boolean DEFAULT NULL, p_jobmon boolean DEFAULT true, p_debug boolean DEFAULT false) RETURNS void

  • Run this function as a scheduled job (cron, etc) to automatically create child tables for partition sets configured to use it.
  • You can also use the included background worker (BGW) to have this automatically run for you by PostgreSQL itself. Note that the p_parent_table parameter is not available with this method, so if you need to run it for a specific partition set, you must do that manually or scheduled as noted above. The other parameters have postgresql.conf values that can be set. See BGW section above.
  • This function also maintains the partition retention system for any partitions sets that have it turned on (see About and part_config table below).
  • Every run checks for all tables listed in the part_config table with automatic_maintenance set to true and either creates new partitions for them or runs their retention policy.
  • By default, all partition sets have automatic_maintenance set to true.
  • New partitions are only created if the number of child tables ahead of the current one is less than the premake value, so you can run this more often than needed without fear of needlessly creating more partitions.
  • Will automatically update the trigger function for non-native partition sets to keep the parent table pointing at the correct partitions. When using time, run this function more often than the partitioning interval to keep the trigger function running its most efficient. For example, if using quarter-hour, run every 5 minutes; if using daily, run at least twice a day, etc.
  • p_parent_table - an optional parameter that if passed will cause run_maintenance() to be run for ONLY that given table, no matter what automatic_maintenance is set to. High transcation rate tables can cause contention when maintenance is being run for many tables at the same time, so this allows finer control of when partition maintenance is run for specific tables. Note that this will also cause the retention system to only be run for the given table as well.
  • p_analyze - For non-native partitioning and native partitioning in PG10, when a new child table is created, an analyze is run on the parent to ensure statistics are updated. For PG11+, this is no longer done, so it is not run by default then. For large partition sets, this analyze can take a while and if run_maintenance() is managing several partitions in a single run, this can cause contention while the analyze finishes. Set this to false (or just leave NULL for PG11+) to disable the analyze run and avoid this contention. For PG10 and older, please note that you must then schedule an analyze of the parent table at some point.
  • p_jobmon - an optional paramter to control whether run_maintenance() itself uses the pg_jobmon extension to log what it does. Whether the maintenance of a particular table uses pg_jobmon is controlled by the setting in the part_config table and this setting will have no affect on that. Defaults to true if not set.
  • p_debug - Output additional notices to help with debugging problems or to more closely examine what is being done during the run.

run_maintenance_proc(p_wait int DEFAULT 0, p_analyze boolean DEFAULT NULL, p_jobmon boolean DEFAULT true, p_debug boolean DEFAULT false)

  • For PG11+, this is the preferred method to run partition maintenance vs directly calling the run_maintenance() function.
  • This procedure can be called instead of the run_maintenance() function to cause PostgreSQL to commit after each partition set's maintenance has finished. This greatly reduces contention issues with long running transactions when there are many partition sets to maintain.
  • NOTE: The BGW does not yet use this procedure and still uses the standard run_maintenance() function.
  • p_wait - How many seconds to wait between each partition set's maintenance run. Defaults to 0.
  • p_analyze - See p_analyze option in run_maintenance.

show_partitions (p_parent_table text, p_order text DEFAULT 'ASC', p_include_default boolean DEFAULT false) RETURNS TABLE (partition_schemaname text, partition_tablename text)

  • List all child tables of a given partition set managed by pg_partman. Each child table returned as a single row.
  • Tables are returned in the order that makes sense for the partition interval, not by the locale ordering of their names.
  • For PG11+, the default partition can be returned in this result set as well if p_include_default is set to true. It is false by default since that is far more common with internal code.
  • p_order - optional parameter to set the order the child tables are returned in. Defaults to ASCending. Set to 'DESC' to return in descending order. If the default is included, it is always listed first.

show_partition_name(p_parent_table text, p_value text, OUT partition_table text, OUT suffix_timestamp timestamp, OUT suffix_id bigint, OUT table_exists boolean)

  • Given a parent table managed by pg_partman (p_parent_table) and an appropriate value (time or id but given in text form for p_value), return the name of the child partition that that value would exist in.
  • If using epoch time partitioning, give the timestamp value, NOT the integer epoch value (use to_timestamp() to convert an epoch value).
  • Returns a child table name whether the child table actually exists or not
  • Also returns a raw value (suffix_timestamp or suffix_id) for the partition suffix for the given child table
  • Also returns a boolean value (table_exists) to say whether that child table actually exists

check_default(p_exact_count boolean DEFAULT true)

  • Run this function to monitor that the parent tables (non-native) or default tables (native PG11+) of the partition sets that pg_partman manages do not get rows inserted to them.
  • Returns a row for each parent/default table along with the number of rows it contains. Returns zero rows if none found.
  • partition_data_time() & partition_data_id() can be used to move data from these parent/default tables into the proper children.
  • p_exact_count will tell the function to give back an exact count of how many rows are in each parent if any is found. This is the default if the parameter is left out. If you don't care about an exact count, you can set this to false and it will return if it finds even just a single row in any parent. This can significantly speed up the check if a lot of data ends up in a parent or there are many partitions being managed.

apply_constraints(p_parent_table text, p_child_table text DEFAULT NULL, p_job_id bigint DEFAULT NULL, p_debug BOOLEAN DEFAULT FALSE)

  • Apply constraints to child tables in a given partition set for the columns that are configured (constraint names are all prefixed with "partmanconstr_").
  • Note that this does not need to be called manually to maintain custom constraints. The creation of new partitions automatically manages adding constraints to old child tables.
  • Columns that are to have constraints are set in the part_config table constraint_cols array column or during creation with the parameter to create_parent().
  • If the pg_partman constraints already exists on the child table, the function will cleanly skip over the ones that exist and not create duplicates.
  • If the column(s) given contain all NULL values, no constraint will be made.
  • If the child table parameter is given, only that child table will have constraints applied.
  • If the p_child_table parameter is not given, constraints are placed on the last child table older than the optimize_constraint value. For example, if the optimize_constraint value is 30, then constraints will be placed on the child table that is 31 back from the current partition (as long as partition pre-creation has been kept up to date).
  • If you need to apply constraints to all older child tables, use the included python script (reapply_constraint.py). Or if you're on PG11+, use the reapply_constraints_proc procedure. Both these methods have options to make constraint application easier with as little impact on performance as possible.
  • The p_job_id parameter is optional. It's for internal use and allows job logging to be consolidated into the original job that called this function if applicable.
  • The p_debug parameter will show you the constraint creation statement that was used.

drop_constraints(p_parent_table text, p_child_table text, p_debug boolean DEFAULT false)

  • Drop constraints that have been created by pg_partman for the columns that are configured in part_config. This makes it easy to clean up constraints if old data needs to be edited and the constraints aren't allowing it.
  • Will only drop constraints that begin with partmanconstr_* for the given child table and configured columns.
  • If you need to drop constraints on all child tables, use the included python script (reapply_constraint.py). Or if you're on PG11+, use reapply_constraints_proc(). These both have options to make constraint removal easier with as little impact on performance as possible.
  • The debug parameter will show you the constraint drop statement that was used.

reapply_constraints_proc(p_parent_table text, p_drop_constraints boolean DEFAULT false, p_apply_constraints boolean DEFAULT false, p_wait int DEFAULT 0, p_dryrun boolean DEFAULT false)

  • Procedures for PG11+ to reapply the extra constraint managed by pg_partman (see Constraint Exclusion section in About section above).
  • Calls drop_constraints() and/or apply_constraint() in a loop, committing after each object is either dropped or added. This helps to avoid long running transaction and contention when doing this on large partition sets.
  • Typical useage would be to drop constraints first, edit the data as needed, then apply constraints again.
  • p_parent_table - Parent table of an already created partition set.
  • p_drop_constraints - Drop all constraints managed by pg_partman. Drops constraints on all child tables including current & future.
  • p_apply_constraints - Apply constraints on configured columns to all child tables older than the premake value.
  • p_wait - Wait the given number of seconds after a table has had its constraints dropped or applied before moving on to the next.

reapply_privileges(p_parent_table text)

  • This function is used to reapply ownership & grants on all child tables based on what the parent table has set.
  • Privileges that the parent table has will be granted to all child tables and privilges that the parent does not have will be revoked (with CASCADE).
  • Privilges that are checked for are SELECT, INSERT, UPDATE, DELETE, TRUNCATE, REFERENCES, & TRIGGER.
  • Be aware that for large partition sets, this can be a very long running operation and is why it was made into a separate function to run independently. Only privileges that are different between the parent & child are applied, but it still has to do system catalog lookups and comparisons for every single child partition and all individual privileges on each.
  • p_parent_table - parent table of the partition set. Must be schema qualified and match a parent table name already configured in pg_partman.

apply_foreign_keys(p_parent_table text, p_child_table text DEFAULT NULL, p_job_id bigint DEFAULT NULL, p_debug boolean DEFAULT false)

  • IMPORTANT: This function is no longer necessary for PG11+ since FK inheritance is automatically managed.
  • Applies any foreign keys that exist on a parent table in a partition set to all the child tables for PG10 and older.
  • This function is automatically called whenever a new child table is created, so there is no need to manually run it unless you need to fix an existing child table.
  • If you need to apply this to an entire partition set, see the reapply_foreign_keys.py python script. This will commit after every FK creation to avoid contention.
  • This function can be used on any table inheritance set, not just ones managed by pg_partman.
  • The p_job_id parameter is optional. It's for internal use and allows job logging to be consolidated into the original job that called this function if applicable.
  • The p_debug parameter will show you the constraint creation statement that was used.

stop_sub_partition(p_parent_table text, p_jobmon boolean DEFAULT true) RETURNS boolean

  • By default, if you undo a child table that is also partitioned, it will not stop additional sibling children of the parent partition set from being subpartitioned unless that parent is also undone. To handle this situation where you may not be removing the parent but don't want any additional subpartitioned children, this function can be used.
  • This function simply deletes the parent_table entry from the part_config_sub table. But this gives a predictable, programatic way to do so and also provides jobmon logging for the operation.

Destruction Functions

undo_partition(p_parent_table text, p_batch_count int DEFAULT 1, p_batch_interval text DEFAULT NULL, p_keep_table boolean DEFAULT true, p_lock_wait numeric DEFAULT 0, p_target_table text DEFAULT NULL, OUT partitions_undone int, OUT rows_undone bigint) RETURNS record

  • Undo a partition set created by pg_partman. This function MOVES the data from the child tables to either the parent table (non-native) or the given target table (native).
  • If you are trying to un-partition a large amount of data automatically, it is recommended to either use the undo_partiton.py script to commit data in smaller batches. Or if you're on PG11+, use the undo_partition_data() procedure to do the same thing. This will greatly reduce issues caused by long running transactions and data contention.
  • When this function is run, the undo_in_progress column in the configuration table is set to true. This causes all partition creation and retention management to stop.
  • By default, partitions are not DROPPED, they are UNINHERITED/UNATTACHED. This leave previous child tables as empty, independent tables.
  • For non-native, when this function is run, the trigger on the parent table & the trigger function are immediately dropped (if they still exist). This means any further writes are done to the parent.
  • Without setting either batch argument manually, each run of the function will move all the data from a single partition into the parent/target.
  • Once all child tables have been uninherited/dropped, the configuration data is removed from pg_partman automatically.
  • For subpartitioned tables, you must start at the lowest level parent table and undo from there then work your way up. If you attempt to undo partitioning on a subpartition set, the function will stop with a warning to let you know.
  • p_parent_table - parent table of the partition set. Must be schema qualified and match a parent table name already configured in pg_partman.
  • p_batch_count - an optional argument, this sets how many times to move the amount of data equal to the p_batch_interval argument (or default partition interval if not set) in a single run of the function. Defaults to 1.
  • p_batch_interval - optional argument. A time or id interval of how much of the data to move. This can be smaller than the partition interval, allowing for very large sized partitions to be broken up into smaller commit batches. Defaults to the configured partition interval if not given or if you give an interval larger than the partition interval. Note that the value must be given as text to this parameter.
  • p_keep_table - an optional argument, setting this to false will cause the old child table to be dropped instead of uninherited/unattached after all of its data has been moved. Note that it takes at least two batches to actually drop a table from the set.
  • p_lock_wait - optional argument, sets how long in seconds to wait for either the table or a row to be unlocked before timing out. Default is to wait forever.
  • p_target_table - A schema-qualified table to move the old partitioned table's data to. Required for undoing a native partition set since data cannot be moved to the parent. Schema can be different from original table.
  • Returns the number of partitions undone and the number of rows moved to the parent table. The partitions undone value returns -1 if a problem is encountered.

@extschema@.undo_partition_proc(p_parent_table text, p_interval text DEFAULT NULL, p_batch int DEFAULT NULL, p_wait int DEFAULT 1, p_target_table text DEFAULT NULL, p_keep_table boolean DEFAULT true, p_lock_wait int DEFAULT 0, p_lock_wait_tries int DEFAULT 10, p_quiet boolean DEFAULT false)

  • A procedure that can un-partition data in distinct commit batches to avoid long running transactions and data contention issues.
  • Only works with PostgreSQL 11+
  • Calls either undo_partition() function in a loop committing as needed.
  • p_parent_table - Parent table of an already created partition set.
  • p_interval - Value that is passed on to the undo_partition function as p_batch_interval argument. Use this to set an interval smaller than the partition interval to commit data in smaller batches. Defaults to the partition interval if not given.
  • p_batch - How many times to loop through the value given for --interval. If --interval not set, will use default partition interval and undo at most -b partition(s). Procedure commits at the end of each individual batch. (NOT passed as p_batch_count to undo_partition function). If not set, all data in the entire partition set will be moved in a single run of the procedure.
  • p_wait - Cause the procedure to pause for a given number of seconds between commits (batches) to reduce write load
  • p_target_table - Same as the p_target_table option in the undo_partition() function.
  • p_keep_table - Same as the p_keep_table option in the undo_partition() function.
  • p_lock_wait - Parameter passed directly through to the underlying partition_data_*() function. Number of seconds to wait on rows that may be locked by another transaction. Default is to wait forever (0).
  • p_lock_wait_tries - Parameter to set how many times the procedure will attempt waiting the amount of time set for p_lock_wait. Default is 10 tries.
  • p_quiet - Procedures cannot return values, so by default it emmits NOTICE's to show progress. Set this option to silence these notices.

drop_partition_time(p_parent_table text, p_retention interval DEFAULT NULL, p_keep_table boolean DEFAULT NULL, p_keep_index boolean DEFAULT NULL, p_retention_schema text DEFAULT NULL) RETURNS int

  • This function is used to drop child tables from a time-based partition set. By default, the table is just uninherited and not actually dropped. For automatically dropping old tables, it is recommended to use the run_maintenance() function with retention configured instead of calling this directly.
  • p_parent_table - the existing parent table of a time-based partition set. MUST be schema qualified, even if in public schema.
  • p_retention - optional parameter to give a retention time interval and immediately drop tables containing only data older than the given interval. If you have a retention value set in the config table already, the function will use that, otherwise this will override it. If not, this parameter is required. See the About section above for more information on retention settings.
  • p_keep_table - optional parameter to tell partman whether to keep or drop the table in addition to uninheriting it. TRUE means the table will not actually be dropped; FALSE means the table will be dropped. This function will just use the value configured in part_config if not explicitly set. This option is ignored if retention_schema is set.
  • p_keep_index - optional parameter to tell partman whether to keep or drop the indexes of the child table when it is uninherited. TRUE means the indexes will be kept; FALSE means all indexes will be dropped. This function will just use the value configured in part_config if not explicitly set. This option is ignored if p_keep_table is set to FALSE or if retention_schema is set.
  • p_retention_schema - optional parameter to tell partman to move a table to another schema instead of dropping it. Set this to the schema you want the table moved to. This function will just use the value configured in part_config if not explicitly set. If this option is set, the retention p_keep_table & p_keep_index parameters are ignored.
  • Returns the number of partitions affected.

drop_partition_id(p_parent_table text, p_retention bigint DEFAULT NULL, p_keep_table boolean DEFAULT NULL, p_keep_index boolean DEFAULT NULL, p_retention_schema text DEFAULT NULL) RETURNS int

  • This function is used to drop child tables from an id-based partition set. By default, the table just uninherited and not actually dropped. For automatically dropping old tables, it is recommended to use the run_maintenance() function with retention configured instead of calling this directly.
  • p_parent_table - the existing parent table of a time-based partition set. MUST be schema qualified, even if in public schema.
  • p_retention - optional parameter to give a retention integer interval and immediately drop tables containing only data less than the current maximum id value minus the given retention value. If you have a retention value set in the config table already, the function will use that, otherwise this will override it. If not, this parameter is required. See the About section above for more information on retention settings.
  • p_keep_table - optional parameter to tell partman whether to keep or drop the table in addition to uninheriting it. TRUE means the table will not actually be dropped; FALSE means the table will be dropped. This function will just use the value configured in part_config if not explicitly set. This option is ignored if retention_schema is set.
  • p_keep_index - optional parameter to tell partman whether to keep or drop the indexes of the child table when it is uninherited. TRUE means the indexes will be kept; FALSE means all indexes will be dropped. This function will just use the value configured in part_config if not explicitly set. This option is ignored if p_keep_table is set to FALSE or if retention_schema is set.
  • p_retention_schema - optional parameter to tell partman to move a table to another schema instead of dropping it. Set this to the schema you want the table moved to. This function will just use the value configured in part_config if not explicitly set. If this option is set, the retention p_keep_table & p_keep_index parameters are ignored.
  • Returns the number of partitions affected.

drop_partition_column(p_parent_table text, p_column text) RETURNS void

  • Depending on when a column was added (before or after partitioning was set up), dropping it on the parent may or may not drop it from all children. This function is used to ensure a column is always dropped from the parent and all children in a partition set.
  • This should only be relevant for non-native partition sets.
  • Uses the IF EXISTS clause in all drop statements, so it may spit out notices/warnings that a column was not found. You can safely ignore these warnings. It should not spit out any errors.

Tables

part_config

Stores all configuration data for partition sets mananged by the extension. The only columns in this table that should ever need to be manually changed are:

  1. retention, retention_schema, retention_keep_table & retention_keep_index to configure the partition set's retention policy
  2. constraint_cols to have partman manage additional constraints & optimize_constraint to control when they're added
  3. premake, optimize_trigger, inherit_fk, automatic_maintenance, template_table, inherit_privileges & jobmon to change the default behavior.

The rest are managed by the extension itself and should not be changed unless absolutely necessary.

  • parent_table
    • Parent table of the partition set
  • control
    • Column used as the control for partition constraints. Must be a time or integer based column.
  • partition_type
    • Type of partitioning. Must be one of the types mentioned above in the create_parent() info.
  • partition_interval
    • Text type value that determines the interval for each partition.
    • Must be a value that can either be cast to the interval or bigint data types.
  • constraint_cols
    • Array column that lists columns to have additional constraints applied. See About section for more information on how this feature works.
  • premake
    • How many partitions to keep pre-made ahead of the current partition. Default is 4.
  • optimize_trigger
    • Manages number of partitions which are handled most efficiently by trigger. See create_parent() function for more info. Default 4.
    • This option is ignored for native partitioning.
  • optimize_constraint
    • Manages which old tables get additional constraints set if configured to do so. See About section for more info. Default 30.
  • epoch
    • Flag the table to be partitioned by time by an integer epoch value instead of a timestamp. See create_parent() function for more info. Default 'none'.
  • inherit_fk
    • Set whether pg_partman manages inheriting foreign keys from the parent table to all children.
    • Defaults to TRUE. Can be set with the create_parent() function at creation time as well.
    • This option is currently ignored for native partitioning.
  • retention
    • Text type value that determines how old the data in a child partition can be before it is dropped.
    • Must be a value that can either be cast to the interval (for time-based partitioning) or bigint (for serial partitioning) data types.
    • Leave this column NULL (the default) to always keep all child partitions. See About section for more info.
  • retention_schema
    • Schema to move tables to as part of the retentions system instead of dropping them. Overrides retention_keep_* options.
  • retention_keep_table
    • Boolean value to determine whether dropped child tables are kept or actually dropped.
    • Default is TRUE to keep the table and only uninherit it. Set to FALSE to have the child tables removed from the database completely.
  • retention_keep_index
    • NOTE: This setting has no affect on native partitioning in PG11+. You cannot drop natively inherited child indexes.
    • Boolean value to determine whether indexes are dropped for child tables that are uninherited.
    • Default is TRUE. Set to FALSE to have the child table's indexes dropped when it is uninherited.
  • infinite_time_partitions
    • By default, new partitions in a time-based set will not be created if new data is not inserted to keep an infinte amount of empty tables from being created.
    • If you'd still like new partitions to be made despite there being no new data, set this to TRUE.
    • Defaults to FALSE.
  • datetime_string
    • For time-based partitioning, this is the datetime format string used when naming child partitions.
  • automatic_maintenance
    • Flag to set whether maintenance is managed automatically when run_maintenance() is called without a table parameter or by the background worker process.
    • Current valid values are "on" and "off". Default is "on".
    • When set to off, run_maintenance() can still be called on in individual partition set by passing it as a paramter to the function.
  • jobmon
    • Boolean value to determine whether the pg_jobmon extension is used to log/monitor partition maintenance. Defaults to true.
  • sub_partition_set_full
    • Boolean value to denote that the final partition for a sub-partition set has been created. Allows run_maintenance() to run more efficiently when there are large numbers of subpartition sets.
  • undo_in_progress
    • Set by the undo_partition functions whenever they are run. If true, this causes all partition creation and retention management by the run_maintenance() function to stop. Default is false.
  • trigger_exception_handling
    • This option is ignored for native partitioning.
    • Boolean value that can be set to allow the partitioning trigger function to handle any exceptions encountered while writing to this table. Handling it in this case means putting the data into the parent table to try and ensure no data loss in case of errors. Be aware that catching the exception here will override any other exception handling that may be done when writing to this partitioned set (Ex. handling a unique constraint violation to ignore it). Just the existence of this exception block will also increase xid consumption since every row inserted will increment the global xid value. If this is table has a high insert rate, you can quickly reach xid wraparound, so use this carefully. This option is set to false by default to avoid causing unexpected behavior in other exception handling situations.
  • p_upsert
    • Please note this option will be going away in the near future once PG11 has been out for a while.
    • text value of the ON CONFLICT clause to include in the partition trigger Defaults to '' (empty string) which means it's inactive. See create_parent() function definition & About section for more info.
    • This option is currently ignored for native partitioning since there is no trigger, but upsert is still able to work in a limited fashion.
  • trigger_return_null
    • Boolean value that allows controlling the behavior of the partition trigger RETURN. By default this is true and the trigger returns NULL to prevent data going into the parent table as well as the children. However, if you have multiple triggers and are relying on the return to be the NEW column value, this can cause a problem. Setting this config value to false will cause the partition trigger to RETURN NEW. You are then responsible for handling the return value in another trigger appropriately. Otherwise, this will cause new data to go into both the child and parent table of the partition set.
    • This option is ignored for native partitioning
  • template_table
    • The schema-qualified name of the table used as a template for applying any inheritance options not handled by the native partitioning options in PG.
  • inherit_privileges
    • Sets whether to inherit the ownership/privileges of the parent table to all child tables. Defaults to true for non-native & PG10. Defaults to false for native PG11+ and should only be necessary if you need direct access to child tables, by-passing the parent table. If true, requires that a superuser either manages partition maintenance or that a superuser at least ownes all pg_partman functions to take advantage of SECURITY DEFINER. If false, it means a superuser is not necessary for any pg_partman operations.

part_config_sub

  • Stores all configuration data for sub-partitioned sets managed by pg_partman.
  • The sub_parent column is the parent table of the subpartition set and all other columns govern how that parent's children are subpartitioned.
  • All columns except sub_parent work the same exact way as their counterparts in the part_config table.

Scripts

If the extension was installed using make, the below script files should have been installed to the PostgreSQL binary directory.

partition_data.py

  • NOTE: This script is only installed for PostgreSQL 10 and lower. It has been replaced by partition_data_proc().
  • A python script to make partitioning in committed batches easier.
  • Script currently does not work with native partitioning.
  • Calls either partition_data_time() or partition_data_id() depending on the value given for --type.
  • A commit is done at the end of each --interval and/or fully created partition.
  • Returns the total number of rows moved to partitions. Automatically stops when parent is empty.
  • To help avoid heavy load and contention during partitioning, autovacuum is turned off for the parent table and all child tables when this script is run. When partitioning is complete, autovacuum is set back to its default value and the parent table is vacuumed when it is emptied.
  • --parent (-p): Parent table of an already created partition set. Required.
  • --type (-t): Type of partitioning. Valid values are "time" and "id". Required.
  • --connection (-c): Connection string for use by psycopg. Defaults to "host=" (local socket).
  • --interval (-i): Value that is passed on to the partitioning function as p_batch_interval argument. Use this to set an interval smaller than the partition interval to commit data in smaller batches. Defaults to the partition interval if not given.
  • --batch (-b): How many times to loop through the value given for --interval. If --interval not set, will use default partition interval and make at most -b partition(s). Script commits at the end of each individual batch. (NOT passed as p_batch_count to partitioning function). If not set, all data in the parent table will be partitioned in a single run of the script.
  • --wait (-w): Cause the script to pause for a given number of seconds between commits (batches).
  • --order (-o): Allows you to specify the order that data is migrated from the parent to the children, either ascending (ASC) or descending (DESC). Default is ASC.
  • --lockwait (-l): Have a lock timeout of this many seconds on the data move. If a lock is not obtained, that batch will be tried again.
  • --lockwait_tries: Number of times to allow a lockwait to time out before giving up on the partitioning. Defaults to 10.
  • --autovacuum_on: Turning autovacuum off requires a brief lock to ALTER the table property. Set this option to leave autovacuum on and avoid the lock attempt.
  • --quiet (-q): Switch setting to stop all output during and after partitioning.
  • --version: Print out the minimum version of pg_partman this script is meant to work with. The version of pg_partman installed may be greater than this.
  • --debug Show additional debugging output
  • Examples:
Partition all data in a parent table. Commit after each partition is made.
      python partition_data.py -c "host=localhost dbname=mydb" -p schema.parent_table -t time
Partition by id in smaller intervals and pause between them for 5 seconds (assume >100 partition interval)
      python partition_data.py -p schema.parent_table -t id -i 100 -w 5
Partition by time in smaller intervals for at most 10 partitions in a single run (assume monthly partition interval)
      python partition_data.py -p schema.parent_table -t time -i "1 week" -b 10

undo_partition.py

  • NOTE: This script is only installed for PostgreSQL 10 and lower. It has been replaced by undo_partition_proc().
  • A python script to make undoing partitions in committed batches easier.
  • Can also work on any non-native parent/child partition set not managed by pg_partman if --type option is not set.
  • This script calls either undo_partition(), undo_partition_time() or undo_partition_id depending on the value given for --type.
  • A commit is done at the end of each --interval and/or emptied partition.
  • Returns the total number of child tables undone. Automatically stops when last child table is undone.
  • --parent (-p): Parent table of the partition set. Required.
  • --type (-t): Type of partitioning. Valid values are "time", "id", & "native". Not setting this argument will use undo_partition() and work on any non-native parent/child table set.
  • --connection (-c): Connection string for use by psycopg. Defaults to "host=" (local socket).
  • --interval (-i): Value that is passed on to the partitioning function as p_batch_interval. Use this to set an interval smaller than the partition interval to commit data in smaller batches. Defaults to the partition interval if not given.
  • --batch (-b): How many times to loop through the value given for --interval. If --interval not set, will use default partition interval and undo at most -b partition(s). Script commits at the end of each individual batch. (NOT passed as p_batch_count to undo function). If not set, all data will be moved to the parent table in a single run of the script.
  • --wait (-w): Cause the script to pause for a given number of seconds between commits (batches).
  • --droptable (-d): Switch setting for whether to drop child tables when they are empty. Leave off option to just uninherit.
  • --quiet (-q): Switch setting to stop all output during and after partitioning undo.
  • --version: Print out the minimum version of pg_partman this script is meant to work with. The version of pg_partman installed may be greater than this.
  • --debug: Show additional debugging output

dump_partition.py

  • A python script to dump out tables contained in the given schema. Uses pg_dump, creates a SHA-512 hash file of the dump file, and then drops the table.
  • When combined with the retention_schema configuration option, provides a way to reliably dump out tables that would normally just be dropped by the retention system.
  • Tables are not dropped if pg_dump does not return successfully.
  • The connection options for psycopg and pg_dump were separated out due to distinct differences in their requirements depending on your database connection configuration.
  • All dump_* option defaults are the same as they would be for pg_dump if they are not given.
  • Will work on any given schema, not just the one used to manage pg_partman retention.
  • --schema (-n): The schema that contains the tables that will be dumped. (Required).
  • --connection (-c): Connection string for use by psycopg. Role used must be able to select from pg_catalog.pg_tables in the relevant database and drop all tables in the given schema. Defaults to "host=" (local socket). Note this is distinct from the parameters sent to pg_dump.
  • --output (-o): Path to dump file output location. Default is where the script is run from.
  • --dump_database (-d): Used for pg_dump, same as its --dbname option or final database name parameter.
  • --dump_host: Used for pg_dump, same as its --host option.
  • --dump_username: Used for pg_dump, same as its --username option.
  • --dump_port: Used for pg_dump, same as its --port option.
  • --pg_dump_path: Path to pg_dump binary location. Must set if not in current PATH.
  • --Fp: Dump using pg_dump plain text format. Default is binary custom (-Fc).
  • --nohashfile: Do NOT create a separate file with the SHA-512 hash of the dump. If dump files are very large, hash generation can possibly take a long time.
  • --nodrop: Do NOT drop the tables from the given schema after dumping/hashing.
  • --verbose (-v): Provide more verbose output.
  • --version: Print out the minimum version of pg_partman this script is meant to work with. The version of pg_partman installed may be greater than this.

vacuum_maintenance.py

  • A python script to perform additional VACUUM maintenance on a given partition set. The main purpose of this is to provide an easier means of freezing tuples in older partitions that are no longer written to. This allows autovacuum to skip over them safely without causing transaction id wraparound issues. See the PostgreSQL documentation for more information on this maintenance isssue: http://www.postgresql.org/docs/current/static/routine-vacuuming.html#VACUUM-FOR-WRAPAROUND.
  • Vacuums all child tables in a given partition set who's age(relfrozenxid) is greater than vacuum_freeze_min_age, including the parent table.
  • Highly recommend scheduled runs of this script with the --freeze option if you have child tables that never have writes after a certain period of time.
  • --parent (-p): Parent table of an already created partition set. (Required)
  • --connection (-c): Connection string for use by psycopg. Defaults to "host=" (local socket).
  • --freeze (-z): Sets the FREEZE option to the VACUUM command.
  • --full (-f): Sets the FULL option to the VACUUM command. Note that --freeze is not necessary if you set this. Recommend reviewing --dryrun before running this since it will lock all tables it runs against, possibly including the parent.
  • --vacuum_freeze_min_age (-a): By default the script obtains this value from the system catalogs. By setting this, you can override the value obtained from the database. Note this does not change the value in the database, only the value this script uses.
  • --noparent: Normally the parent table is included in the list of tables to vacuum if its age(relfrozenxid) is higher than vacuum_freeze_min_age. Set this to force exclusion of the parent table, even if it meets that criteria.
  • --dryrun: Show what the script will do without actually running it against the database. Highly recommend reviewing this before running for the first time.
  • --quiet (-q): Turn off all output.
  • --debug: Show additional debugging output.

reapply_indexes.py

  • A python script for reapplying indexes on child tables in a partition set after they are changed on the parent table.
  • NOTE: This script only works with non-native partitioning and native partitioning in PG10. It does not worth with PG11+ and likely won't be needed since most indexes are now managed automatically in those versions. The only exception may be primary & unique indexes that don't include the partition column. There's unfortunately no easy way to manage index inheritance in PG11+ via this script and will have to be handled manually.
  • Any indexes that currently exist on the children and match the definition on the parent will be left as is. There is an option to recreate matching as well indexes if desired, as well as the primary key.
  • Indexes that do not exist on the parent will be dropped from all children.
  • Commits are done after each index is dropped/created to help prevent long running transactions & locks.
  • NOTE: New index names are made based off the child table name & columns used, so their naming may differ from the name given on the parent. This is done to allow the tool to account for long or duplicate index names. If an index name would be duplicated, an incremental counter is added on to the end of the index name to allow it to be created. Use the --dryrun option first to see what it will do and which names may cause dupes to be handled like this.
  • --parent (-p): Parent table of an already created partition set. Required.
  • --connection (-c): Connection string for use by psycopg. Defaults to "host=" (local socket).
  • --concurrent: Create indexes with the CONCURRENTLY option. Note this does not work on primary keys when --primary is given.
  • --drop_concurrent: Drop indexes concurrently when recreating them (PostgreSQL >= v9.2). Note this does not work on primary keys when --primary is given.
  • --recreate_all (-R): By default, if an index exists on a child and matches the parent, it will not be touched. Setting this option will force all child indexes to be dropped & recreated. Will obey the --concurrent & --drop_concurrent options if given. Will not recreate primary keys unless --primary option is also given.
  • --primary: By default the primary key is not recreated. Set this option if that is needed. Note this will cause an exclusive lock on the child table for the duration of the recreation.
  • --jobs (-j): Use the python multiprocessing library to recreate indexes in parallel. Note that this is per table, not per index. Be very careful setting this option if load is a concern on your systems.
  • --wait (-w): Wait the given number of seconds after indexes have finished being created on a table before moving on to the next. When used with -j, this will set the pause between the batches of parallel jobs instead.
  • --dryrun: Show what the script will do without actually running it against the database. Highly recommend reviewing this before running. Note that if multiple indexes would get the same default name, the duplicated names will show in the dryrun (because the index doesn't exist in the catalog to check for it). When the real thing is run, the duplicated names will be handled as stated in the NOTE above.
  • --quiet: Turn off all output.
  • --nonpartman If the partition set you are running this on is not managed by pg_partman, set this flag otherwise this script may not work. Note that the pg_partman extension is still required to be installed for this to work since it uses certain internal functions. When this is set the order that the tables are reindexed is alphabetical instead of logical.
  • --version: Print out the minimum version of pg_partman this script is meant to work with. The version of pg_partman installed may be greater than this.

reapply_constraints.py

  • NOTE: This script is only installed for PostgreSQL 10 and lower. It has been replaced by reapply_constraints_proc().
  • A python script for redoing constraints on child tables in a given partition set for the columns that are configured in part_config table.
  • Typical useage would be -d mode to drop constraints, edit the data as needed, then -a mode to reapply constraints.
  • --parent (-p): Parent table of an already created partition set. (Required)
  • --connection (-c): Connection string for use by psycopg. Defaults to "host=" (local socket).
  • --drop_constraints (-d): Drop all constraints managed by pg_partman. Drops constraints on ALL child tables in the partition set.
  • --add_constraints (-a): Apply constraints on configured columns to all child tables older than the premake value.
  • --jobs (-j): Use the python multiprocessing library to recreate indexes in parallel. Value for -j is number of simultaneous jobs to run. Note that this is per table, not per index. Be very careful setting this option if load is a concern on your systems.
  • --wait (-w): Wait the given number of seconds after a table has had its constraints dropped or applied before moving on to the next. When used with -j, this will set the pause between the batches of parallel jobs instead.
  • --dryrun: Show what the script will do without actually running it against the database. Highly recommend reviewing this before running.
  • --quiet (-q): Turn off all output.
  • --version: Print out the minimum version of pg_partman this script is meant to work with. The version of pg_partman installed may be greater than this.

reapply_foreign_keys.py

  • NOTE: This script is only installed for PostgreSQL 10 and lower.
  • A python script for redoing the inherited foreign keys for an entire partition set.
  • Script currently does not work with native partitioning and FKs are handled natively as of PG11+.
  • All existing foreign keys on all child tables are dropped and the foreign keys that exist on the parent at the time this is run will be applied to all children.
  • Commits after each foreign key is created to avoid long periods of contention.
  • --parent (-p): Parent table of an already created partition set. (Required)
  • --connection (-c): Connection string for use by psycopg. Defaults to "host=" (local socket).
  • --quiet (-q): Switch setting to stop all output during and after partitioning undo.
  • --dryrun: Show what the script will do without actually running it against the database. Highly recommend reviewing this before running.
  • --nonpartman If the partition set you are running this on is not managed by pg_partman, set this flag. Otherwise internal pg_partman functions are used and this script may not work. When this is set the order that the tables are rekeyed is alphabetical instead of logical.
  • --version: Print out the minimum version of pg_partman this script is meant to work with. The version of pg_partman installed may be greater than this.
  • --debug: Show additional debugging output

check_unique_constraints.py

  • Partitioning using inheritance (native still does this internally) has the shortcoming of not allowing a unique constraint to apply to all tables in the entire partition set without causing large performance issues once the partition set begins to grow very large. This script is used to check that all rows in a partition set are unique for the given columns.
  • Note that on very large partition sets this can be an expensive operation to run that can consume a large chunk of storage space. The amount of storage space required is enough to dump out the entire index's column data as a plaintext file.
  • If there is a column value that violates the unique constraint, this script will return those column values along with a count of how many of each value there are. Output can also be simplified to a single, total integer value to make it easier to use with monitoring applications.
  • --parent (-p): Parent table of the partition set to be checked. (Required)
  • --column_list (-l): Comma separated list of columns that make up the unique constraint to be checked. (Required)
  • --connection (-c): Connection string for use by psycopg. Defaults to "host=" (local socket).
  • --temp (-t): Path to a writable folder that can be used for temp working files. Defaults system temp folder.
  • --psql: Full path to psql binary if not in current PATH.
  • --simple: Output a single integer value with the total duplicate count. Use this for monitoring software that requires a simple value to be checked for.
  • --quiet (-q): Suppress all output unless there is a constraint violation found.
  • --version: Print out the minimum version of pg_partman this script is meant to work with. The version of pg_partman installed may be greater than this.
You can’t perform that action at this time.