How to join tables with REPEATED RECORDS - arrays

Question

I'm struggling with joining tables, when using REPEATED RECORD fields on the ON clause. The error i get is:
No matching signature for operator = for argument types: ARRAY<STRUCT<experiment INT64>>, INT64. Supported signature: ANY = ANY at [6:5]
My REPEATED RECORD is called ab_test and it has 4 fields inside (experiment, group ,name, state)
My Query:
SELECT be.type, be.group, be.user.id, be.uid,
ARRAY(SELECT STRUCT(ab_test.experiment as experiment , ab_test.group as group, ab_test.name as name, ab_test.state, uid_allocation_timestamp) FROM UNNEST(ab_test) AS ab_test) as ab_test
FROM fiverr-bigquery.dwh.bi_events be
JOIN staging_tables.ab_tests_uid_allocation_history uid_alloc
ON be.uid = uid_alloc.uid
AND ***ARRAY(SELECT STRUCT(ab_test.experiment) FROM UNNEST(ab_test) AS ab_test ) = uid_alloc.test_id***
WHERE be._PARTITIONTIME = '2017-04-24 00:00:00'
AND DATE(created_at) = DATE('2017-04-24')
AND ARRAY(SELECT STRUCT(ab_test.experiment) FROM UNNEST(ab_test) AS ab_test ) IS NOT NULL
AND type = 'order.success'
I also tried replacing the second ON clause with:
CAST((SELECT experiment FROM UNNEST(ab_test) as experiment ) AS INT64) = uid_alloc.test_id
But with no luck (the error i get:Invalid cast from STRUCT<experiment INT64,groupINT64, name STRING, ...> to INT64 at [40:10]
Any ideas ?

Answer 1

I also tried replacing ... But with no luck ... Any ideas ?
Below is attempt to mimic your use case - at least that part of it that responsible for the error you see
If you run below (BigQuery Standard SQL) - you will get exactly same error as in your case
#dtandardSQL
WITH data AS (
SELECT 1 AS id, [ STRUCT<experiment INT64, grp INT64, name STRING>
(911, 2, 'a'), (2, 2, 'b'), (3, 2, 'c')] AS ab_test UNION ALL
SELECT 2 AS id, [ STRUCT<experiment INT64, grp INT64, name STRING>
(11, 3, 'a'), (12, 3, 'b'), (13, 3, 'c')] AS ab_test UNION ALL
SELECT 3 AS id, [ STRUCT<experiment INT64, grp INT64, name STRING>
(21, 4, 'a'), (911, 4, 'b'), (23, 4, 'c')] AS ab_test
)
SELECT id
FROM data
WHERE CAST((SELECT experiment FROM UNNEST(ab_test) AS experiment ) AS INT64) = 911
The error will be
Error: Invalid cast from STRUCT<experiment INT64, grp INT64, name STRING> to INT64 at [12:12]
To resolve this - use below approach
#dtandardSQL
WITH data AS (
SELECT 1 AS id, [ STRUCT<experiment INT64, grp INT64, name STRING>
(911, 2, 'a'), (2, 2, 'b'), (3, 2, 'c')] AS ab_test UNION ALL
SELECT 2 AS id, [ STRUCT<experiment INT64, grp INT64, name STRING>
(11, 3, 'a'), (12, 3, 'b'), (13, 3, 'c')] AS ab_test UNION ALL
SELECT 3 AS id, [ STRUCT<experiment INT64, grp INT64, name STRING>
(21, 4, 'a'), (911, 4, 'b'), (23, 4, 'c')] AS ab_test
)
SELECT id
FROM data
WHERE (SELECT COUNT(1)
FROM UNNEST(ab_test) AS ab_test
WHERE ab_test.experiment = 911
) > 0
No errors now and output will be
id
1
3
because those rows have elements of ab_test with experiment = 911
Finally, below is example with test values from JOIN table as in your question
#dtandardSQL
WITH data AS (
SELECT 1 AS id, [ STRUCT<experiment INT64, grp INT64, name STRING>
(911, 2, 'a'), (2, 2, 'b'), (3, 2, 'c')] AS ab_test UNION ALL
SELECT 2 AS id, [ STRUCT<experiment INT64, grp INT64, name STRING>
(11, 3, 'a'), (12, 3, 'b'), (13, 3, 'c')] AS ab_test UNION ALL
SELECT 3 AS id, [ STRUCT<experiment INT64, grp INT64, name STRING>
(21, 4, 'a'), (911, 4, 'b'), (23, 4, 'c')] AS ab_test
),
tests AS (
SELECT 911 AS test_id UNION ALL
SELECT 912 AS test_id
)
SELECT data.id
FROM data
CROSS JOIN tests
WHERE (SELECT COUNT(1)
FROM UNNEST(ab_test) AS ab_test
WHERE ab_test.experiment = tests.test_id
) > 0
Hope you can apply above to your specific case

Answer 2

When joinning two tables, if the column is not supplied with the table name, it will return STRUCT data type.
To solved this, can you try:
select table.column