comnetwork commented on a change in pull request #444: PHOENIX-5148
URL: https://github.com/apache/phoenix/pull/444#discussion_r266027910
##########
File path:
phoenix-core/src/it/java/org/apache/phoenix/end2end/join/SubqueryUsingSortMergeJoinIT.java
##########
@@ -516,8 +516,8 @@ public void testAnyAllComparisonSubquery() throws
Exception {
assertEquals(rs.getString(2), "T6");
assertFalse(rs.next());
-
- query = "SELECT /*+ USE_SORT_MERGE_JOIN*/ \"order_id\", name FROM
" + tableName4 + " o JOIN " + tableName1 + " i ON o.\"item_id\" = i.\"item_id\"
WHERE quantity != ANY(SELECT quantity FROM " + tableName4 + " q WHERE
o.\"item_id\" = q.\"item_id\" GROUP BY quantity)";
+ //add order by to make the query result stable
+ query = "SELECT /*+ USE_SORT_MERGE_JOIN*/ \"order_id\", name FROM
" + tableName4 + " o JOIN " + tableName1 + " i ON o.\"item_id\" = i.\"item_id\"
WHERE quantity != ANY(SELECT quantity FROM " + tableName4 + " q WHERE
o.\"item_id\" = q.\"item_id\" GROUP BY quantity) order by \"order_id\"";
Review comment:
If we don't add the order by , the order of the query result is reverse
after this patch, just as:
assertTrue (rs.next());
assertEquals(rs.getString(1), "000000000000004");
assertEquals(rs.getString(2), "T6");
assertTrue (rs.next());
assertEquals(rs.getString(1), "000000000000002");
assertEquals(rs.getString(2), "T6");
That is because after the `SubqueryRewriter.transform`,the original sql is
rewritten as :
```
SELECT ORDER_ID,NAME FROM ORDERTABLE O
Inner JOIN ITEMTABLE I
ON (O.ITEM_ID = I.ITEM_ID)
Inner JOIN (SELECT COLLECTDISTINCT($3.$4) $5,$3.$6 $2 FROM (SELECT
QUANTITY $4,Q.ITEM_ID $6 FROM ORDERTABLE Q GROUP BY Q.ITEM_ID,QUANTITY) $3
GROUP BY $3.$6) $1
ON (O.ITEM_ID = $1.$2) WHERE QUANTITY != ANY($1.$5)
```
Without this patch, after the following first `SortMergeJoin` is completed,
an additional `CLIENT SORTED BY [O.ITEM_ID]` is added in order to complete
the second `SortMergeJoin`, which unfortunately changes the order of two tuples
with the same `O.ITEM_ID` which is `0000000006`
```
SELECT ORDER_ID,NAME FROM ORDERTABLE O
Inner JOIN ITEMTABLE I
ON (O.ITEM_ID = I.ITEM_ID)
```
With this patch, the unnecessary `CLIENT SORTED BY [O.ITEM_ID]` is avoid,
the order of tuple is untouched after the first `SortMergeJoin`, so the final
SortMergeJoin result is not same as without this patch.
To make the query result stable regardless of any optimization, order by
\"order_id\" is added.
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
With regards,
Apache Git Services
