training_duration with the "distinct on" paradigm

This improves performance against the previous "partition by" approach
This commit is contained in:
Riccardo Berto 2021-09-12 11:25:30 +02:00
parent 64c4938c88
commit 2d802b67d5

View File

@ -24,30 +24,28 @@ create index if not exists idx_training_data_training_id on training_data(traini
create index if not exists idx_training_data_geog on training_data using GIST(geog);
create or replace view training_duration as
(select td_start.id as id,
td_start.geog as start_location,
td_end.geog as end_location,
td_start.t as start_time,
td_end.t as end_time,
date_trunc('second', (td_end.t - td_start.t)) as duration
from (select td_row_n.t, td_row_n.id, td_row_n.geog
from (select *,
row_number() over (partition by td.training_id order by td.t asc) as row_n
from training_data td join training t on (t.id = td.training_id)
order by td.t asc) as td_row_n
where td_row_n.row_n = 1
order by td_row_n.t asc) td_start
join
(select td_row_n.t, td_row_n.id, td_row_n.geog
from (select *,
row_number() over (partition by td.training_id order by td.t desc) as row_n
from training_data td join training t on (t.id = td.training_id)
order by td.t asc) as td_row_n
where td_row_n.row_n = 1) td_end
on (td_start.id = td_end.id)
(
select
td_start.training_id as training_id,
td_start.t as start_time,
td_end.t as end_time,
resolve_geo_location(td_start.geog) AS start_location,
resolve_geo_location(td_end.geog) AS end_location,
td_start.geog as start_location_point,
td_end.geog as end_location_point,
date_trunc('second'::text, td_end.t - td_start.t) AS duration
from
(
select distinct on (td2.training_id) *
from training_data td2 order by td2.training_id, t asc
) td_start
join
(
select distinct on (td2.training_id) *
from training_data td2 order by td2.training_id, t desc
) td_end
on (td_start.training_id = td_end.training_id)
order by td_start.t asc
);
create or replace view training_info as