Skip to content

Latest commit

 

History

History
43 lines (32 loc) · 1.53 KB

README.md

File metadata and controls

43 lines (32 loc) · 1.53 KB

Postgres SQL: Simple timeseries in postgres.

You have a massive table with records going back for years. It has become unmanageable. Even deleting old data is slow and inconvenient. What do you do?

Postgres inheritance provides a transparent way of splitting a massive table into many smaller tables, one for each day. Queries don't have to be changed - postgres will transparently run a select query on all the small tables and glue the results together:

-- Create a parent table:
create table ht
( my_daynum	int
, score		int
, size		int
, domain	text
, country	text
, browser	text
, hack_type	int
);

-- Create small tables for UNIX daynums 16791 and 16792:
create table ht_16791 () inherits (ht);
create table ht_16792 () inherits (ht);

-- Insert some data fro those days:
copy ht_16791 from '/tmp/ht_16791.csv' csv;
copy ht_16792 from '/tmp/ht_16792.csv' csv;

-- Querying the parent table gives you data in the small tables:
select * from ht limit 20;

-- Data too old to be interesting?  Delete that day:
drop table ht_16791;

However postgres does NOT manage insertions transparently. Not out of the box. That is what these functions are for:

CREATE TRIGGER ht_insert_trigger
    BEFORE INSERT ON ht
    FOR EACH ROW EXECUTE PROCEDURE daytable_insert_trigger('my_daynum');
insert into ht values (19,10,3,'loopy','NZ','Panda',96758769);
\dt
select * from ht;

This will automagically create the small table, if necessary, and insert the data into the appropriate small table.