Snowflake unsupported subquery when using function - sql

Question

This is the function I created:
CREATE OR REPLACE FUNCTION NS_REPORTS.AP."COUPA_GET_EXCH_RATE"("from_curr_id" NUMBER(38,0), "to_curr_id" NUMBER(38,0), "date" DATE)
RETURNS FLOAT
LANGUAGE SQL
AS '
SELECT
COALESCE((
SELECT
RATE
FROM
(
SELECT
ROW_NUMBER() OVER (PARTITION BY DATE(RATE_DATE) ORDER BY RATE_DATE DESC) ROW_NUM
, RATE
FROM
CONNECTORS.COUPA.EXCHANGE_RATE
WHERE
FROM_CURRENCY_ID = from_curr_id
AND TO_CURRENCY_ID = to_curr_id
AND DATE(RATE_DATE) = date
) R
WHERE
ROW_NUM = 1
), 1)
';
I'm using the ROW_NUMBER function because the RATE_DATE field is actually datetime and so there are multiple records per date.
When I call the function by itself, it works fine. However, when I try to use it in a view, I get the unsupported subquery type error. The view works fine without it. Can anyone think of what I can do to either fix the error or work around it by rewriting the query?
EDIT 1: View code and exact error message
CREATE OR REPLACE VIEW COUPA_REQUISITION
AS
SELECT
RH.ID REQ_NUM
, RL.LINE_NUM REQ_LINE_NUM
, OH.PO_NUMBER
, REPLACE(REPLACE(OH.CUSTOM_FIELDS:"legacy-po-number", '"', ''), '.0', '') LEGACY_PO_NUMBER
, S."NAME" SUPPLIER
, OH.STATUS
, UR.FULLNAME REQUESTED_BY
, UC.FULLNAME CREATED_BY
, OL.RECEIVED
, DATE(RH.SUBMITTED_AT) ORDER_DATE
, DATE(RH.NEED_BY_DATE) NEEDED_BY_DATE
, RL."DESCRIPTION" ITEM
, CAST(NULL AS VARCHAR) CHART_OF_ACCOUNTS
, REPLACE(OH.CUSTOM_FIELDS:"purchase-type", '"', '') PURCHASE_TYPE
, COM."NAME" COMMODITY
, ACT.NS_SUB_NAME SUBSIDIARY
, ACT.NS_ACCT_NAME_FULL "ACCOUNT"
, ACT.NS_DEPT_NAME_FULL DEPARTMENT
, ACT.NS_L3_DEPT_NAME L3_DEPARTMENT
, ACT.NS_LOC_NAME "LOCATION"
, RL.QUANTITY QTY
, OL.LINE_NUM ORDER_LINE_NUM
, RL.TOTAL * NS_REPORTS.AP.COUPA_GET_EXCH_RATE(RL.CURRENCY_ID, 1, DATE(RH.SUBMITTED_AT)) LINE_TOTAL
, RL.TOTAL - OL.INVOICED UNINVOICED_AMOUNT
, OL.INVOICED INVOICED_TOTAL
, RLSUM.TOTAL TOTAL
, REPLACE(IL.CUSTOM_FIELDS:"amortization-schedule"."name", '"', '') AMORTIZATION_SCHEDULE
, CASE WHEN COALESCE(IL.CUSTOM_FIELDS:"amortization-start-date", '') <> '' THEN DATE(REPLACE(IL.CUSTOM_FIELDS:"amortization-start-date", '"', '')) ELSE NULL END AMORTIZATION_START_DATE
, CASE WHEN COALESCE(IL.CUSTOM_FIELDS:"amortization-end-date", '') <> '' THEN DATE(REPLACE(IL.CUSTOM_FIELDS:"amortization-end-date", '"', '')) ELSE NULL END AMORTIZATION_END_DATE
, CASE WHEN COALESCE(OH.CUSTOM_FIELDS:"contract-start-date", '') <> '' THEN DATE(REPLACE(OH.CUSTOM_FIELDS:"contract-start-date", '"', '')) ELSE NULL END CONTRACT_START_DATE
, CASE WHEN COALESCE(OH.CUSTOM_FIELDS:"contract-end-date", '') <> '' THEN DATE(REPLACE(OH.CUSTOM_FIELDS:"contract-end-date", '"', '')) ELSE NULL END CONTRACT_END_DATE
FROM
CONNECTORS.COUPA.REQUISITION_HEADER RH
JOIN CONNECTORS.COUPA.REQUISITION_LINE RL ON RL.REQUISITION_HEADER_ID = RH.ID
JOIN NS_REPORTS.AP.COUPA_ACCOUNT ACT ON ACT.COUPA_ACCT_ID = RL.ACCOUNT_ID
JOIN CONNECTORS.COUPA."USER" UR ON UR.ID = RH.REQUESTED_BY_ID
JOIN CONNECTORS.COUPA."USER" UC ON UC.ID = RH.CREATED_BY_ID
JOIN (
SELECT
REQUISITION_HEADER_ID
, SUM(TOTAL) TOTAL
FROM
CONNECTORS.COUPA.REQUISITION_LINE
GROUP BY
REQUISITION_HEADER_ID
) RLSUM ON RLSUM.REQUISITION_HEADER_ID = RH.ID
LEFT JOIN CONNECTORS.COUPA.ORDER_LINE OL ON OL.ID = RL.ORDER_LINE_ID
LEFT JOIN CONNECTORS.COUPA.ORDER_HEADER OH ON OH.ID = OL.ORDER_HEADER_ID
LEFT JOIN CONNECTORS.COUPA.COMMODITY COM ON COM.ID = OL.COMMODITY_ID
LEFT JOIN CONNECTORS.COUPA.SUPPLIER S ON S.ID = OH.SUPPLIER_ID
LEFT JOIN CONNECTORS.COUPA.INVOICE_LINE IL ON IL.ORDER_LINE_ID = OL.ID
Error message:
SQL Error [2031] [42601]: SQL compilation error:
Unsupported subquery type cannot be evaluated

Answer 1

The error is a correlated subquery and the are not supported (beyond some tiny toy examples)
But the basic form is
SELECT a.a
(select b.b from b where b.a = a.a order by b.y limit 1)
FROM a;
in effect for each row, a sub-query is run on another table to get a value. There are many tricks done in other DB's to make this "work" but in effect there is work done on each row. Snowflake does not types of for each row operations.
The good news is there are other patterns that are effectively the same, that snowflake does support, the two patterns are really the same use a CTE or join to a sub-select which is the same thing.
so the above becomes:
WITH subquery AS (
SELECT b.a, b.b FROM b
QUALIFY row_number() over (partition by b.a order by b.y) = 1
)
SELECT a.a
sq.b
FROM a
JOIN subquery AS sq
ON sq.a = a.a
So we first process/shape "all records" from the other/sub table, and we only keep the rows that have the count/shape we want, and then join to that result. The is very parallelizable, so performs well. The reason Snowflake does not auto translate a sub-query for you, is it rather easy to get it wrong, and they presently are spending there development efforts working on features that do not exist at all, etc etc, and it can be rewritten by you, given you understand your model.

Answer 2

What if you move this to the FROM clause? I would phrase this as:
SELECT COALESCE(MAX(er.rate), 1)
FROM (SELECT er.*
FROM CONNECTORS.COUPA.EXCHANGE_RATE er
WHERE er.FROM_CURRENCY_ID = in_from_curr_id AND
er.TO_CURRENCY_ID = in_to_curr_id AND
DATE(er.RATE_DATE) = in_date
ORDER BY RATE_DATE DESC
LIMIT 1
) er;
Notice that I changed the names of the parameters so they are more obviously input parameters.