Skip to content

Set of tables and functions wrote to help the management of partitionning tables in PostgreSQL databases.

Notifications You must be signed in to change notification settings

hipay/postgresql-partition-manager

Repository files navigation

Partmgr

Partmgr is a set of tables and functions wrote to help the management of partitionning tables in PostgreSQL databases.

The partition key is a date, matching a table's attribute. It is possible to set up a retention period

Partmgr's tables are catalog which contains informations about partition type, tables and triggers

Functions can manage trigger creation, partition, drop, and monitoring.

Tables ( catalog )

Tables are in "partition" schema :

  • partition.pattern : partition's type
  • partition."table" : partitionning tables
  • partition.trigger : trigger's partitionning tables

Table pattern contiens some partitioning patterns. Until there no new pattern, there is no need to modify it. Different patterns, and therefore partition types are: `` year ``, `` month ``, `` `` week, and `` day ``.

Table table contains user's tables, and must be set by the user.

La table trigger contains partition table's triggers. Filling this table is automatically done when the user set up a new partition table.

Liste des fonctions

  • fonction partition.between() : compute unit number betwenn two date fon a given pattern.
  • fonctions partition.create() : create the partitions
    • partition.create() : generic function, which create partitions for all tables, at the current date.
    • partition.create( date) : create partitions for all tables, at the given date.
    • partition.create( begin_date, end_date ) : create partitions for all tables, at the given period.
    • partition.create( schema, table, begin_date, end_date ) : create partitions for the given table, at the given period.
    • partition.create( schema, table, column, period, pattern, begin_date, end_date ) : low-level function fonction,

called by all create functions.

  • function partition.create_next() : create next functions of all tables. The next period depends of current date,
plus next_part interval from pattern. This fonction could called by a scheduler like cron.
  • function partition.drop() : drop partition, if permit by setup.
  • function partition.check_next_part() : had to be called by Nagios plugin. Can monitoring if next partition exists.
  • function partition.grant_replace( p_acl text, p_grant text, p_ext_grant text ) :
  • function partition.grant( acl text, tablename text ) :
  • function partition.setgrant( p_schemaname text, p_tablename text, p_part text ) : used by partition.create() to apply grants on partitions.
  • function partition.create_part_trigger() : create partitioning triggers, and create the trigger on the partition.
  • function partition.set_trigger_def() : Trigger function which copy trigger definition from mother table to catalog. Triggered on partition."table"

Tutorial

Installation

To install PartMgr, run the script partition.sql. It make the schema, tables and functions, and fill the table

$ cd partmgr
$ psql -U postgres dbname < partition.sql

Setup

There is two opérations needed to setup up partitionning table. One is insertion into partition."table"

INSERT INTO partition."table" ( schemaname, tablename, keycolumn, pattern, actif, cleanable, retention_period)
  values ('test', 'test1mois', 'ev_date', 'M', 't', 'f', null),
         ('test', 'test_mois', 'ev_date', 'M', 't', 't', '1 mon') ;

Triggers on this table are inserted into partition.trigger to be auto-added on partition. These triggers won't be present on the mother table.

Privileges setted up on the mother table are automatically applied on partitions.

The second step is creation and setup of partitionning trigger

SELECT partition.create_part_trigger('schema_name','table_name');

This function make the specific function trigger for the table given. The new trigger function is created in the partition and the trigger _partitionne is created on the table.

Partition Creation

Then, the set of partition should be created with partition.create() functions

part=$ select * from partition.create('2012-09-01','2012-11-01') ;
 o_tables | o_indexes | o_triggers | o_grants
----------+-----------+------------+----------
       74 |        74 |         65 |      126
(1 row)

part=$ select * from partition.create('test','test_mois','2012-11-01','2013-03-01') ;
 o_tables | o_indexes | o_triggers | o_grants
----------+-----------+------------+----------
        4 |         4 |          0 |        4
(1 row)

then dropped by partition.drop() function

part=$ select * from partition.drop() ;
 o_tables
----------
        0
(1 row)

Only partitions cleanable and whose retention period has passed will be deleted.

Schedule Creation

The creation of the next partitions, the next month or the next day, can be created simply with the partition.create_next (). This function is based on the next_part column of the table partition.pattern to determine the date of the partition to create.

Monitoring

partition.check_next_part() function allows monitoring from Nagios

part=$ select * from partition.check_next_part() ;
 nagios_return_code |              message
--------------------+-----------------------------------
                  2 | Missing : test.test1jour_20120628
(1 row)
part=$ select * from partition.create('test','test1jour','2012-06-28','2012-06-29') ;
 o_tables | o_indexes | o_triggers | o_grants
----------+-----------+------------+----------
        2 |         2 |          2 |        4
(1 row)
part=$ select * from partition.check_next_part() ;
 nagios_return_code | message
--------------------+---------
                  0 |
(1 row)

About

Set of tables and functions wrote to help the management of partitionning tables in PostgreSQL databases.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages