postgresql unnest and pivot int array column - sql

Question

I have below table
create table test(id serial, key int,type text,words text[],numbers int[] );
insert into test(key,type,words) select 1,'Name',array['Table'];
insert into test(key,type,numbers) select 1,'product_id',array[2];
insert into test(key,type,numbers) select 1,'price',array[40];
insert into test(key,type,numbers) select 1,'Region',array[23,59];
insert into test(key,type,words) select 2,'Name',array['Table1'];
insert into test(key,type,numbers) select 2,'product_id',array[1];
insert into test(key,type,numbers) select 2,'price',array[34];
insert into test(key,type,numbers) select 2,'Region',array[23,59,61];
insert into test(key,type,words) select 3,'Name',array['Chair'];
insert into test(key,type,numbers) select 3,'product_id',array[5];
I was using below query to pivot table for users.
select key,
max(array_to_string(words,',')) filter(where type='Name') as "Name",
cast(max(array_to_string(numbers,',')) filter(where type='product_id') as int) as "product_id",
cast(max(array_to_string(numbers,',')) filter(where type='price') as int) as "price" ,
max(array_to_string(numbers,',')) filter(where type='Region') as "Region"
from test group by key
But I couldn't unnest the Region column during Pivot in-order to use Region column to join with another table .
My expected output is below

Answer 1

Since we are using unnest("Region") to do to pivot. There must be a row with region data for each product.
Or below code will do the trick by creating an array of null.
unnest(CASE WHEN array_length("Region", 1) >= 1
THEN "Region"
ELSE '{null}'::int[] END)
Schema:
create table test(id serial, key int,type text,words text[],numbers int[] );
insert into test(key,type,words) select 1,'Name',array['Table'];
insert into test(key,type,numbers) select 1,'product_id',array[2];
insert into test(key,type,numbers) select 1,'price',array[40];
insert into test(key,type,numbers) select 1,'Region',array[23,59];
insert into test(key,type,words) select 2,'Name',array['Table1'];
insert into test(key,type,numbers) select 2,'product_id',array[1];
insert into test(key,type,numbers) select 2,'price',array[34];
insert into test(key,type,numbers) select 2,'Region',array[23,59,61];
insert into test(key,type,words) select 3,'Name',array['Chair'];
insert into test(key,type,numbers) select 3,'product_id',array[5];
select key,"Name",product_id,price,unnest(CASE WHEN array_length("Region", 1) >= 1
THEN "Region"
ELSE '{null}'::int[] END) from
(
select key,
max(array_to_string(words,',')) filter(where type='Name') as "Name",
cast(max(array_to_string(numbers,',')) filter(where type='product_id') as int) as "product_id",
cast(max(array_to_string(numbers,',')) filter(where type='price') as int) as "price" ,
max(numbers) filter(where type='Region') as "Region"
from test group by key
)t order by key
key
Name
product_id
price
unnest
1
Table
2
40
23
1
Table
2
40
59
2
Table1
1
34
23
2
Table1
1
34
59
2
Table1
1
34
61
3
Chair
5
null
null
db<>fiddle here

Answer 2

Very strange database design... I'm assuming you inherited it?
If none of the other array values will ever have a cardinality > 1 then, you can simply unnest:
select
key,
(max (words) filter (where type = 'Name'))[1] as name,
(max (numbers) filter (where type = 'product_id'))[1] as product_id,
(max (numbers) filter (where type = 'price'))[1] as price,
unnest (max (numbers) filter (where type = 'Region')) as region
from test
group by key
If they can have multiple values, that can also be handled.
-- EDIT 3/15/2021 --
Short version: an unnest against a null won't product a row, so if you coalesce the null value into an array of a single null element, that should take care of this part:
select
key,
(max (words) filter (where type = 'Name'))[1] as name,
(max (numbers) filter (where type = 'product_id'))[1] as product_id,
(max (numbers) filter (where type = 'price'))[1] as price,
unnest (coalesce (max (numbers) filter (where type = 'Region'), array[null]::integer[])) as region
from test
group by key
order by key
Now for the part you didn't ask... I and at least one other have been gently nudging you that your database design is going to cause multiple problems at every turn. The fact that it's in production doesn't mean you shouldn't fix it as soon as you can.
This design is what's known as EAV - Entity - Attribute - Value. It has its use cases, but like most good things it can also be applied when it shouldn't. The use case that comes to mind is if you want users to be able to dynamically add attributes to certain objects. Even then, there might be better/easier ways.
And as one example, if you have one million objects, five attributes means you have to store that as five million rows, and the majority of that space will be occupied with repeating the key and attribute names.
Just food for thought. We can continue to triage this with every new scenario you find, but it would be better to redo the design.