DB partition trigger with PostgreSQL

Database partitioning is about logically splitting one large table into smaller physical pieces, such that improving query performance.  DB partitioning is a good alternate for indexing multiple columns, reducing index size, hence the memory in use. Few common pros of database partitioning:
  • Improved performance – data operations (CRUD) can be performed on a smaller volume of data, for example, in case of collecting data overtime, putting old data in separate partition might help with performance.
  • Bulk create and delete can be efficient by adding or removing separate partitions.
  • Time based partition can be helpful in cleaning old seldom-used data i.e. month based partition we can simply set a cron job for cleaning 12 month old partition, without effecting the table portion heavily in use for ADD, UPDATE, etc.
  • Improved scalability – In case of very large tables, you can partition and have them hosted on a separate server.
There are 2 main approaches to database partitioning:
  • Horizontal partitioning (Sharding) – a table is split horizontally, such that each partition is a subset of the table, having the same schema (i.e. number of fields/columns).
  • Vertical partitioning – a table is split on the fields/columns, such that each subset has separate schema. A common use-case for vertical partitioning is to partition table fields on the basis of pattern of use i.e. frequently accessed fields are to be grouped together, and the less frequently accessed are put in a separate partition.

This blog post is about setting automatic horizontal partitioning (month based) on a table in PostgreSQL.
The high-level steps are:
  1. Create table, or select an existing one.
  2. Execute partitioning function or procedure
  3. Table trigger – to call the partition procedure.
  4. View for parent-child tables (optional)
  5. Verification.

Step 1 – Create Table

CREATE TABLE partition_test(id BIGINT, created_datetime DATE);
Output:
Create table

Step 2 – Trigger function

CREATE OR REPLACE FUNCTION test_partition_function() RETURNS trigger AS
$BODY$
DECLARE
partition_date TEXT;
partition TEXT;
BEGIN
partition_date := to_char(NEW.created_datetime,'YYYY_MM');
partition := TG_RELNAME || '_' || partition_date;
IF NOT EXISTS(SELECT relname FROM pg_class WHERE relname=partition) THEN
RAISE NOTICE 'A partition has been created %',partition;
EXECUTE 'CREATE TABLE ' || partition || ' () INHERITS (' || TG_RELNAME || ');';
END IF;
EXECUTE 'INSERT INTO ' || partition || ' SELECT(' || TG_RELNAME || ' ' || quote_literal(NEW) || ').* RETURNING id;';
RETURN NULL;
END;
$BODY$
LANGUAGE plpgsql VOLATILE
COST 100;
Output:
partition function


Step 3 – Table trigger

CREATE TRIGGER partition_test_trg
AFTER INSERT ON partition_test
FOR EACH ROW EXECUTE PROCEDURE test_partition_function();
Output:
table trigger


Step 4 – Viewing the partition (optional)

CREATE VIEW show_partitions AS
SELECT nmsp_parent.nspname AS parent_schema,
parent.relname AS parent,
nmsp_child.nspname AS child_schema,
child.relname AS child
FROM pg_inherits
JOIN pg_class parent ON pg_inherits.inhparent = parent.oid
JOIN pg_class child ON pg_inherits.inhrelid = child.oid
JOIN pg_namespace nmsp_parent ON nmsp_parent.oid = parent.relnamespace
JOIN pg_namespace nmsp_child ON nmsp_child.oid = child.relnamespace
WHERE parent.relname='partition_test' ;
Output:
db view

Step 5 – Verification (optional)

Let’s test using the show_partitions view, if we have any partitions yet
select * from show_partitions;
Output:
view partition

insert in partition_test

insert into partition_test values (1, '2018-01-19');
Output:
db_part_6

Few more inserts for the same month

insert into partition_test values (2, '2018-01-22');
insert into partition_test values (3, '2018-01-29');

Output:
db_part_7

Insert for next month

insert into partition_test values (2, '2018-02-22');
Output:
db_part_8

View partitions

select * from show_partitions;
Output:
db_part_9


For Django users there’s a very good suite ‘Architect‘  for enhancing the ORM capability for complex database task (i.e. partitioning). Architect is simple to use, updated, and currently supports MySQL and PostgreSQL.

Leave a Reply

Your email address will not be published. Required fields are marked *