Snowflake: SQL compilation error: not a valid group by expression - sql

Question

I'm trying to call a window function inside of a case statement, as such:
SELECT
DISTINCT properties.property_id
COALESCE(MAX(CASE
WHEN units_count.unit_type = 'NORMAL' THEN units_count.unit_count
END) OVER (PARTITION BY properties.property_id),
0)::INT AS normal_units_count
FROM units_count
JOIN properties ON units_count.property_id = properties.property_id
I'm receiving the following error:
SQL compilation error: [IFF(UNITS_COUNT.UNIT_TYPE = 'NORMAL', UNITS_COUNT.UNIT_COUNT, SYSTEM$NULL_TO_FIXED(null))] is not a valid group by expression
I've tried adding a qualify clause to remove the MAX() function:
SELECT
DISTINCT properties.property_id
COALESCE(CASE
WHEN units_count.unit_type = 'NORMAL' THEN units_count.unit_count
END OVER (PARTITION BY properties.property_id),
0)::INT AS normal_units_count
FROM units_count
JOIN properties ON units_count.property_id = properties.property_id
QUALIFY units_count.unit_count = MAX(units_count.unit_count) OVER (PARTITION BY properties.property_id)
The code executes, but the qualify clause results in unwanted filtering for other fields. Can I keep the existing logic (with MAX()) or do I need to include a qualify clause?

Answer 1

The DISTINCT is in effect a grouping operation, but your MAX is a for every row with the OVER clause.
which implies as "basic SQL" this should work:
SELECT
p.property_id,
MAX(IFF(uc.unit_type = 'NORMAL', uc.unit_count, 0)) as v1 max_units_count
FROM units_count AS uc
JOIN properties AS p
ON uc.property_id = p.property_id
GROUP BY 1
but as your "other fields" implies you are selecting other things and without knowing what/how you are doing that is hard to see what you are wanting todo.
also your max(unit_count) does not so much feel like it is a normal_unit_count.
But you example needs to pull in a second+ column of what you are wanting to do to see how they should be co-handled. But I would be inclined with zero information to suggest you use a CTE to find the per property_id the MAX unit_count and then join that result to a second read from your data. Because with zero insights to the other operations and how they could re-use that read, I have experienced it is better to GROUP/JOIN then WINDOW(over partition)/ANYVALUE. But that is another option.
Thus (without testing) this might work:
SELECT
DISTINCT p.property_id
ANYVALUE(COALESCE(MAX(CASE
WHEN uc.unit_type = 'NORMAL' THEN uc.unit_count
END) OVER (PARTITION BY p.property_id),
0)::INT) AS normal_units_count
FROM units_count AS uc
JOIN properties AS p
ON uc.property_id = p.property_id