问题
I have basically two tables, Orders and Items. As these tables are imported from Google Cloud Datastore backup files, references are not made by a simple ID field, but a <STRUCT> for one-to-one relationship, where its id field represents the actual unique ID I want to match. For one-to-many relationship (REPEATED), the schema uses ARRAY of <STRUCT>.
I can query the one-to-one relationships with a LEFT OUTER JOIN, I also know how to join on a non-repeated struct and a repeated string or int, but I have trouble to achieve a similar join query with a repeated struct.
One Order with one item:
#standardSQL
WITH Orders AS (
SELECT 1 AS __oid__, STRUCT(STRUCT(2 AS id, "default" AS ns) AS key) AS item UNION ALL
SELECT 2 AS __oid__, STRUCT(STRUCT(4 AS id, "default" AS ns) AS key) AS item UNION ALL
SELECT 3 AS __oid__, STRUCT(STRUCT(6 AS id, "default" AS ns) AS key) AS item
),
Items AS (
SELECT STRUCT(1 AS id, "default" AS ns) AS key, "#1.1" AS title UNION ALL
SELECT STRUCT(2 AS id, "default" AS ns) AS key, "#1.2" AS title UNION ALL
SELECT STRUCT(3 AS id, "default" AS ns) AS key, "#1.3" AS title UNION ALL
SELECT STRUCT(4 AS id, "default" AS ns) AS key, "#1.4" AS title UNION ALL
SELECT STRUCT(5 AS id, "default" AS ns) AS key, "#1.5" AS title UNION ALL
SELECT STRUCT(6 AS id, "default" AS ns) AS key, "#1.6" AS title
)
SELECT
__oid__
,Order_item AS item
FROM Orders
LEFT OUTER JOIN(
SELECT
key
,title
FROM Items
) Order_item
ON Order_item.key.id = item.key.id
Result (works as expected):
+-----+---------+--------------+-------------+------------+
| Row | __oid__ | item.key.id | item.key.ns | item.title |
+-----+---------+--------------+-------------+------------+
| 1 | 1 | 2 | default | #1.2 |
+-----+---------+--------------+-------------+------------+
| 2 | 2 | 4 | default | #1.4 |
+-----+---------+--------------+-------------+------------+
| 3 | 3 | 6 | default | #1.6 |
+-----+---------+--------------+-------------+------------+
Similar query, but this time one order with many items:
#standardSQL
WITH Orders AS (
SELECT 1 AS __oid__, ARRAY[STRUCT(STRUCT(1 AS id, "default" AS ns) AS key), STRUCT(STRUCT(2 AS id, "default" AS ns) AS key)] AS items UNION ALL
SELECT 2 AS __oid__, ARRAY[STRUCT(STRUCT(3 AS id, "default" AS ns) AS key), STRUCT(STRUCT(4 AS id, "default" AS ns) AS key)] AS items UNION ALL
SELECT 3 AS __oid__, ARRAY[STRUCT(STRUCT(5 AS id, "default" AS ns) AS key), STRUCT(STRUCT(6 AS id, "default" AS ns) AS key)] AS items
),
Items AS (
SELECT STRUCT(1 AS id, "default" AS ns) AS key, "#1.1" AS title UNION ALL
SELECT STRUCT(2 AS id, "default" AS ns) AS key, "#1.2" AS title UNION ALL
SELECT STRUCT(3 AS id, "default" AS ns) AS key, "#1.3" AS title UNION ALL
SELECT STRUCT(4 AS id, "default" AS ns) AS key, "#1.4" AS title UNION ALL
SELECT STRUCT(5 AS id, "default" AS ns) AS key, "#1.5" AS title UNION ALL
SELECT STRUCT(6 AS id, "default" AS ns) AS key, "#1.6" AS title
)
SELECT
__oid__
,Order_items AS items
FROM Orders
LEFT OUTER JOIN(
SELECT
key
,title
FROM Items
) Order_items
ON Order_items.key.id IN (SELECT item.key.id FROM UNNEST(items) AS item)
Error: IN subquery is not supported inside join predicate.
I actually expected this result:
+-----+---------+--------------+-------------+------------+
| Row | __oid__ | item.key.id | item.key.ns | item.title |
+-----+---------+--------------+-------------+------------+
| 1 | 1 | 1 | default | #1.1 |
| | | 2 | default | #1.2 |
+-----+---------+--------------+-------------+------------+
| 2 | 2 | 3 | default | #1.3 |
| | | 4 | default | #1.4 |
+-----+---------+--------------+-------------+------------+
| 3 | 3 | 5 | default | #1.5 |
| | | 6 | default | #1.6 |
+-----+---------+--------------+-------------+------------+
How do I change the second query to get the expected result?
回答1:
Alternative option is to do CROSS JOIN instead of LEFT JOIN
#standardSQL
WITH Orders AS (
SELECT 1 AS __oid__, ARRAY[STRUCT(STRUCT(1 AS id, "default" AS ns) AS key), STRUCT(STRUCT(2 AS id, "default" AS ns) AS key)] AS items UNION ALL
SELECT 2 AS __oid__, ARRAY[STRUCT(STRUCT(3 AS id, "default" AS ns) AS key), STRUCT(STRUCT(4 AS id, "default" AS ns) AS key)] AS items UNION ALL
SELECT 3 AS __oid__, ARRAY[STRUCT(STRUCT(5 AS id, "default" AS ns) AS key), STRUCT(STRUCT(6 AS id, "default" AS ns) AS key)] AS items
),
Items AS (
SELECT STRUCT(1 AS id, "default" AS ns) AS key, "#1.1" AS title UNION ALL
SELECT STRUCT(2 AS id, "default" AS ns) AS key, "#1.2" AS title UNION ALL
SELECT STRUCT(3 AS id, "default" AS ns) AS key, "#1.3" AS title UNION ALL
SELECT STRUCT(4 AS id, "default" AS ns) AS key, "#1.4" AS title UNION ALL
SELECT STRUCT(5 AS id, "default" AS ns) AS key, "#1.5" AS title UNION ALL
SELECT STRUCT(6 AS id, "default" AS ns) AS key, "#1.6" AS title
)
SELECT
__oid__
,ARRAY_AGG(Order_items) AS items
FROM Orders
CROSS JOIN(
SELECT
key
,title
FROM Items
) Order_items
WHERE Order_items.key.id IN (SELECT item.key.id FROM UNNEST(items) AS item)
GROUP BY __oid__
回答2:
The problem is that BigQuery can't hash-partition the join keys from the two sides (since the join is expressed as an IN condition). You can make this work by flattening the array on the left-hand side and then aggregating the items from the right:
#standardSQL
WITH Orders AS (
SELECT 1 AS __oid__, ARRAY[STRUCT(STRUCT(1 AS id, "default" AS ns) AS key), STRUCT(STRUCT(2 AS id, "default" AS ns) AS key)] AS items UNION ALL
SELECT 2 AS __oid__, ARRAY[STRUCT(STRUCT(3 AS id, "default" AS ns) AS key), STRUCT(STRUCT(4 AS id, "default" AS ns) AS key)] AS items UNION ALL
SELECT 3 AS __oid__, ARRAY[STRUCT(STRUCT(5 AS id, "default" AS ns) AS key), STRUCT(STRUCT(6 AS id, "default" AS ns) AS key)] AS items
),
Items AS (
SELECT STRUCT(1 AS id, "default" AS ns) AS key, "#1.1" AS title UNION ALL
SELECT STRUCT(2 AS id, "default" AS ns) AS key, "#1.2" AS title UNION ALL
SELECT STRUCT(3 AS id, "default" AS ns) AS key, "#1.3" AS title UNION ALL
SELECT STRUCT(4 AS id, "default" AS ns) AS key, "#1.4" AS title UNION ALL
SELECT STRUCT(5 AS id, "default" AS ns) AS key, "#1.5" AS title UNION ALL
SELECT STRUCT(6 AS id, "default" AS ns) AS key, "#1.6" AS title
)
SELECT
__oid__
,ARRAY_AGG(Order_items) AS items
FROM Orders,
UNNEST(items) AS item
LEFT OUTER JOIN(
SELECT
key
,title
FROM Items
) Order_items
ON Order_items.key.id = item.key.id
GROUP BY __oid__
This looks like what you wanted in any case, since your original query would have had items just as a struct rather than as an array of structs.
来源：https://stackoverflow.com/questions/51136595/bigquery-join-on-with-repeated-array-struct-field-in-standard-sql