is it possible to nest an array_agg inside another array_agg - sql

Question

I have a SQL table message(application, type, action, date, ...) and I would like to get all the actions for a type and all the types for an application in a single query if possible.
So far I have managed to get the result in two separate queries like so:
select application, array_agg(distinct type) as types from message group by application;
application | types
--------------+----------------------------------------------------------------------------------------------------------------------------
app1 | {company,user}
app2 | {document,template}
app3 | {organization,user}
and the second query:
select type, array_agg(distinct action) as actions from message group by type;
type | actions
--------------------------------------+-----------------------------------------
company | {created,updated}
document | {created,tested,approved}
organization | {updated}
template | {deleted}
user | {created,logged,updated}
The most obvious single query I could come up with so far is just:
select application, type, array_agg(distinct action) from message group by application, type;
Which would require some programmatic processing to build the type array.
What I wanted to do was something theoretically like:
select application, array_agg(type, array_agg(action)) from message group by application, type which isn't possible as is but I feel there is a way to do it. I have also thought about nesting the second query into the first one but haven't found how to make it work yet.

Answer 1

demo:db<>fiddle
You can create tuples (records): (col1, col2). So if col2 is of type array, you created (text, text[]). These tuples can be aggregated as well into array of tuples:
SELECT
app,
array_agg((type, actions)) -- here is the magic
FROM (
SELECT
app,
type,
array_agg(actions) actions
FROM
message
GROUP BY app, type
) s
GROUP BY app
To get access, you have to explicitely define the record type at unnesting:
SELECT
*
FROM (
-- your query with tuples
)s,
unnest(types) AS t(type text, actions text[]) -- unnesting the tuple array
Nevertheless, as stated in the comments, maybe JSON may be a better approach for you:
demo:db<>fiddle
SELECT
app,
json_agg(json_build_object('type', type, 'actions', actions))
FROM (
SELECT
app,
type,
json_agg(actions) actions
FROM
message
GROUP BY app, type
) s
GROUP BY app
Result:
[{
"type": "company",
"actions": ["created","updated"]
},
{
"type": "user",
"actions": ["logged","updated"]
}]
Another possible JSON output:
demo:db<>fiddle
SELECT
json_agg(data)
FROM (
SELECT
json_build_object(app, json_agg(types)) as data
FROM (
SELECT
app,
json_build_object(type, json_agg(actions)) AS types
FROM
message
GROUP BY app, type
) s
GROUP BY app
) s
Result:
[{
"app1": [{
"company": ["created","updated"]
},
{
"user": ["logged","updated"]
}]
},
{
"app2": [{
"company": ["created"]
}]
}]