Hasura "Running Total" Computed Column - sql

Question

I'm looking to create a computed column in Hasura which returns a set of another table and includes a running total column of the set. For example:
table_a
id product_id
----------- ----------
1 "1"
2 "2"
2 "3"
table_b
id product_id qty created_at
----------- ---------- --- ----------
1 "1" 6 01/01/20
2 "2" 4 01/02/20
3 "3" 2 01/02/20
4 "3" 2 01/02/20
5 "1" 4 01/03/20
6 "2" 6 01/03/20
Desired GQL Response:
{
"data": {
"table_a": [
{
"id": 1,
"product_id": "1",
"computed_table_b_rows": [
{
id: 1,
product_id: "1",
qty: 6,
created_at: 01/01/20,
running_total: 6,
},
{
id: 5,
product_id: "1",
qty: 4,
created_at: 01/03/20,
running_total: 10,
},
]
}
]
}
}
Here's what I have so far which does not work:
CREATE FUNCTION filter_table_b(table_a_row table_a)
RETURNS SETOF table_b AS $$
SELECT *,
SUM (qty) OVER (PARTITION BY product_id ORDER BY created_at) as running_total
FROM table_b
WHERE product_id = table_a_row.product_id
$$ LANGUAGE sql STABLE;

Answer 1

It seems like SETOF is supposed to be a subset of the table, and cannot have new columns. E.g. I tried:
CREATE FUNCTION getfoo(int) RETURNS SETOF table_a AS $$
SELECT *, 'asdf' as extra FROM table_a WHERE id = $1;
$$ LANGUAGE SQL;
resulting in the Hasura-reported error:
SQL Execution Failed
postgres-error: return type mismatch in function declared to return table_a
{
"path": "$.args[0].args",
"error": "query execution failed",
"internal": {
"arguments": [],
"error": {
"status_code": "42P13",
"exec_status": "FatalError",
"message": "return type mismatch in function declared to return table_a",
"description": "Final statement returns too many columns.",
"hint": ""
},
"prepared": false,
"statement": "CREATE FUNCTION getfoo(int) RETURNS SETOF table_a AS $$\n SELECT *, 'asdf' as extra FROM table_a WHERE id = $1;\n$$ LANGUAGE SQL;"
},
"code": "postgres-error"
}
There are a few other options, however:
Add the "running total" column to table_b itself (by changing the function definition. Detailed below.
If this is not performant enough, another option is to create a view over table_b that has the running_total column using sum and partition. And then create a relationship from table_a to table_b_view.
Create a trigger on table_b, that when a new row is added it computes the running_total column and stores it, as that data seems to be static as it's based on historical data.
Option 1 is detailed below as it most closely resembles original implementation (use of a function):
CREATE OR REPLACE FUNCTION public.table_b_running_total(table_b_row table_b)
RETURNS bigint
LANGUAGE sql
STABLE
AS $function$
SELECT SUM (qty)
FROM table_b
WHERE product_id = table_b_row.product_id
AND created_at <= table_b_row.created_at
LIMIT 1
$function$
With that, I was able to get the desired result -- tho the desired graphql response does not exactly match.
query:
query MyQuery {
table_a(where:{ id: { _eq: 1 }}) {
id
product_id
table_bs {
id
product_id
qty
created_at
table_b_running_total
}
}
}
response:
{
"data": {
"table_a": [
{
"id": 1,
"product_id": "1",
"table_bs": [
{
"id": 1,
"product_id": "1",
"qty": 6,
"created_at": "2020-01-01T00:00:00+00:00",
"table_b_running_total": 6
},
{
"id": 5,
"product_id": "1",
"qty": 4,
"created_at": "2020-01-03T00:00:00+00:00",
"table_b_running_total": 10
}
]
}
]
}
}