PostgreSQL Table Partition by Date - postgresql

Question

I am creating a table that will have over million records. The data will append on a daily basis (once a day) appending previous whole day data. I want to partition the table using daily_updated field. The table will maintain rolling 1 month of data and I am thinking to partition on a 1 day basis. Any suggestions here. How can I maintain a rolling partition. I looked over some example where it states that you write a statement using partition by range and the table will auto-partition based on the data it holds. (statement PARTITION BY RANGE (daily_updated)).
Suggestion please.
thanks,

Answer 1

As note by others, PostgreSQL doesn't have any built in functionality for automatically doing this. I've been using the (slightly edited) following and running it on a daily cron:
CREATE OR REPLACE FUNCTION manage_partitions ()
RETURNS VOID
LANGUAGE plpgsql
SECURITY DEFINER
AS $$
/**
Procedure manage_partitions creates (and disposes of) table partitions
for the SCHEMA_NAME.TABLE_NAME table.
*/
DECLARE
l_schema_name varchar := 'SCHEMA_NAME' ; -- The name of the schema for the partitioned table
l_table_name varchar := 'TABLE_NAME' ; -- The name of the table to manage partitions for
l_retention_days integer := 30 ; -- The number of past days to retain partitions for.
l_pre_days integer := 10 ; -- The number of future days to pre-create partitions for.
-- The intent is to maintain a buffer so that, in the
-- event that this function is not run for a few days,
-- the logging functionality can continue to work.
dt record ;
l_cmd text ;
BEGIN
-- ASSERTION: the schema name and table name of the partioned table do not require quoting
-- ASSERTION: there are no other partitioned tables in the schema that have a similar name
-- ASSERTION: the table partitions reside in the same schema as the parent table
-- NB: in the event that there are ever any partitions that are desired to be preserved
-- beyond the retention schedule then they either need to be renamed in such fashion that
-- breaks the naming pattern or, better yet, manually detached from the parent table
FOR dt IN (
WITH args AS (
SELECT l_retention_days AS retention_days,
l_pre_days AS pre_days,
l_retention_days + l_pre_days AS total_days,
l_schema_name AS schema_name,
l_table_name AS table_name,
l_schema_name || '.' || l_table_name AS parent_table
),
dates AS (
SELECT ( current_date + ( ( s.idx - args.retention_days )::text || ' days'::text )::interval )::date AS partition_date
FROM args
CROSS JOIN (
SELECT idx
FROM generate_series ( 1, ( SELECT total_days FROM args ), 1 ) AS gs ( idx )
) s
),
new_parts AS (
SELECT args.parent_table || '_' || to_char ( dates.partition_date, 'yyyymmdd' ) AS partition_name,
to_char ( dates.partition_date, 'yyyy-mm-dd' ) AS partition_date
FROM dates
CROSS JOIN args
),
cur_parts AS (
SELECT n.nspname || '.' || c.relname AS partition_name
FROM pg_catalog.pg_class c
JOIN pg_catalog.pg_namespace n
ON ( n.oid = c.relnamespace )
JOIN pg_catalog.pg_inherits i
ON ( c.oid = i.inhrelid )
CROSS JOIN args
WHERE n.nspname = args.schema_name
AND c.relname::text ~ ( args.table_name || '_.+' )::text
)
SELECT args.parent_table,
cur_parts.partition_name AS current_partition,
new_parts.partition_name AS new_partition,
new_parts.partition_date
FROM cur_parts
FULL JOIN new_parts
ON ( cur_parts.partition_name = new_parts.partition_name )
CROSS JOIN args ) LOOP
IF dt.current_partition IS NULL THEN
l_cmd := 'CREATE TABLE '
|| dt.new_partition
|| ' PARTITION OF '
|| dt.parent_table
|| ' FOR VALUES FROM ( '''
|| dt.partition_date
|| '''::date ) TO ( ( '''
|| dt.partition_date
|| '''::date + ''1 day''::interval )::date )' ;
--RAISE NOTICE E' % ;', l_cmd ;
EXECUTE l_cmd ;
ELSIF dt.new_partition IS NULL THEN
l_cmd := 'ALTER TABLE '
|| dt.parent_table
|| ' DETACH PARTITION '
|| dt.current_partition ;
--RAISE NOTICE E' % ;', l_cmd ;
EXECUTE l_cmd ;
l_cmd := 'DROP TABLE ' || dt.current_partition ;
--RAISE NOTICE E' % ;', l_cmd ;
EXECUTE l_cmd ;
END IF ;
END LOOP ;
END ;
$$ ;