Replicating MongoDB $bucket with conditional sum in Postgres - mongodb

Question

I have a database with hundreds of thousands of rows with this schema:
+----+----------+---------+
| id | duration | type |
+----+----------+---------+
| 1 | 41 | cycling |
+----+----------+---------+
| 2 | 15 | walking |
+----+----------+---------+
| 3 | 6 | walking |
+----+----------+---------+
| 4 | 26 | running |
+----+----------+---------+
| 5 | 30 | cycling |
+----+----------+---------+
| 6 | 13 | running |
+----+----------+---------+
| 7 | 10 | running |
+----+----------+---------+
I was previously using a MongoDB aggregation to do this and get a distribution of activities by type and total count:
{
$bucket: {
groupBy: '$duration',
boundaries: [0, 16, 31, 61, 91, 121],
default: 121,
output: {
total: { $sum: 1 },
walking: {
$sum: { $cond: [{ $eq: ['$type', 'walking'] }, 1, 0] },
},
running: {
$sum: { $cond: [{ $eq: ['$type', 'running'] }, 1, 0] },
},
cycling: {
$sum: { $cond: [{ $eq: ['$type', 'cycling'] }, 1, 0] },
},
},
},
}
I have just transitioned to using Postgres and can't figure out how to do the conditional sums there. What would the query be to get a result table like this?
+---------------+---------+---------+---------+-------+
| duration_band | walking | running | cycling | total |
+---------------+---------+---------+---------+-------+
| 0-15 | 41 | 21 | 12 | 74 |
+---------------+---------+---------+---------+-------+
| 15-30 | 15 | 1 | 44 | 60 |
+---------------+---------+---------+---------+-------+
| 30-60 | 6 | 56 | 7 | 69 |
+---------------+---------+---------+---------+-------+
| 60-90 | 26 | 89 | 32 | 150 |
+---------------+---------+---------+---------+-------+
| 90-120 | 30 | 0 | 6 | 36 |
+---------------+---------+---------+---------+-------+
| 120+ | 13 | 90 | 0 | 103 |
+---------------+---------+---------+---------+-------+
| Total | 131 | 257 | 101 | 492 |
+---------------+---------+---------+---------+-------+

Answer 1

SQL is very good at retrieving and making calculations on data, and delivering it so getting the values you want is an easy task. It is not so good at formatting results, that why that task is typically left to the presentation layer. That said, however, does not mean it cannot be done - it can and in a single query. The difficulty is the pivot process - transforming rows into columns. But first some setup. You should put the duration data on a its own table (if not already). With the addition of a identifier which then allows multiple criteria sets (more on that later). I will proceed that way.
create table bands( name text, period int4range, title text );
insert into bands(name, period, title)
values ('Standard', '[ 0, 15)'::int4range , '0 - 15')
, ('Standard', '[ 15, 30)'::int4range , '15 - 30')
, ('Standard', '[ 30, 60)'::int4range , '30 - 60')
, ('Standard', '[ 60, 90)'::int4range , '60 - 00')
, ('Standard', '[ 90,120)'::int4range , '90 - 120')
, ('Standard', '[120,)'::int4range , '120+');
This sets up the your current criteria. The name column is the prior mentioned identifier where the title column becomes the duration band on the output. The interesting column is the period; defined as an integer range. In this case a [closed,open) range that includes the 1st number but not the 2nd - yea the brackets have meaning. That definition becomes the heart of resulting query. The query builds as follows:
Retrieve the desired interval set ( [0-5) ... ) set and append to it
a "totals" entry.
Define the list of activities (cycling, ...).
Combine these sets to create a list of interval set with each
activity.
The above gives the activity intervals which becomes the matrix generated when pivoted.
Combine the "test" table values into the above list calculating the
total time for each activity within each interval. This is the work
horse of the query. It does ALL of the calculations.
The above now contains intervals plus total activity for each cell in the matrix. However it still exists in row orientation.
With the results calculated pivot them from row orientation to
column orientation.
Finally compress the pivoted results into a single row for each interval and set the final interval ordering.
And the result is:
with buckets ( period , title, ord) as
( select period , title, row_number() over (order by lower(b.period)) ord ---- 1
from bands b
where name = 'Standard'
union all
select '[0,)','Total',count(*) + 1
from bands b
where name = 'Standard'
)
, activities (activity) as ( values ('running'),('walking'),('cycling'), ('Total')) ---- 2
, activity_buckets (period, title, ord, activity) as
(select * from buckets cross join activities) ---- 3
select s2.title "Duration Band" ---- 6
, max(cycling) "Cycling"
, max(running) "Running"
, max(walking) "Walking"
, max(Total) "Total "
from ( select s1.title, s1.ord
, case when s1.activity = 'cycling' then duration else null end cycling ---- 5
, case when s1.activity = 'running' then duration else null end running
, case when s1.activity = 'walking' then duration else null end walking
, case when s1.activity = 'Total' then duration else null end total
from ( select ab.ord, ab.title, ab.activity
, sum(coalesce(t.duration,0)) duration ---- 4
from activity_buckets ab
left join test t
on ( (t.type = ab.activity or ab.activity = 'Total')
and t.duration <# ab.period --** determines which time interval(s) the value belongs
)
group by ab.ord, ab.title, ab.activity
) s1
) s2
group by s2.ord,s2.title
order by s2.ord;
See demo. It contains each of the major steps along the way. Additionally it shows how creating a table for the intervals can be put to use. Since I dislike long queries I generally hide them behind a SQL function and then just use the function. Demo also contains this.