Postgres-如何合并分区？

Question

I have a table that accumulates many millions of rows per day. The most recent data is queried most frequently, as the data gets older it's queried less frequently.

To aid performance I've partitioned the data by month, the table(s) have been created something like this…

CREATE TABLE banana ( recTime TIMESTAMP, blah blah blah) PARTION BY RANGE (recTime);
CREATE TABLE banana_201704 PARTITION OF banana FOR VALUES FROM ('2017-04-01' ) TO (  '2017-05-01' ) ;
CREATE TABLE banana_201705 PARTITION OF banana FOR VALUES FROM ('2017-05-01' ) TO (  '2017-06-01' ) ;

I've noticed that at the start of the month I get very good performance from this table, presumably because at that time of month it has few records in it, as the month progresses (and the tables gets bigger) the performance goes down.

For this reason I'm thinking that I'd get better performance if I partitioned by day, instead of by month. However, I don't want to partition the entire table by day, I'd end up with lots of little partitions to manage.

Is it possible to somehow seamlessly merge partitions together so that after (lets say) 30 days, all the individual 'day' partitions can be merged into one big 'month' partition, without the performance overhead of simply selecting from one table, inserting into another and then deleting the original records?

Or is there a better way to manage this?

Answer 1

After the end of a month you should create a monthly table as select from the base table, drop daily partitions and attach the monthly table as a partition.

Create daily partitions for January 2018:

do $$
declare d date;
begin
    for d in 
        select generate_series('2018-01-01'::date, '2018-01-31', '1d')::date
    loop
    execute format($ex$
        create table banana_%s partition of banana for values from (%L) to (%L)
        $ex$, replace(d::text, '-', ''), d, d+ 1);
    end loop;
end;
$$;

Create a single partition for the month and drop daily partitions:

create table banana_201801 as
select * from banana
where created_at between '2018-01-01' and '2018-01-31';

do $$
declare d date;
begin
    for d in 
        select generate_series('2018-01-01'::date, '2018-01-31', '1d')::date
    loop
    execute format($ex$
        drop table banana_%s
        $ex$, replace(d::text, '-', ''));
    end loop;
end;
$$;

alter table banana 
attach partition banana_201801 for values from ('2018-01-01') to ('2018-02-01');