[IMPALA-7895] Incorrect expected results for spillable-buffer-sizing.test

• Type: Bug
• Status: Resolved
• Priority: Minor
• Resolution: Fixed
• Affects Version/s: Impala 3.0
• Fix Version/s: Impala 3.2.0
• Component/s: Frontend
• Epic Color: ghx-label-3
• Assignee: Paul Rogers
• Reporter: Paul Rogers
• Created: 27/Nov/18 00:44
• Updated: 01/Feb/19 14:22
• Resolved: 01/Feb/19 14:22

Description

A recent change appears to have caused a test to expect the wrong rewritten SQL in spillable-buffer-sizing.test.

# Mid NDV aggregation - should scale down buffers to intermediate size.
select straight_join l_orderkey, o_orderstatus, count(*)
from tpch_parquet.lineitem
    join tpch_parquet.orders on o_orderkey = l_orderkey
group by 1, 2
having count(*) = 1
---- DISTRIBUTEDPLAN
Max Per-Host Resource Reservation: Memory=82.00MB Threads=7
Per-Host Resource Estimates: Memory=244MB
Analyzed query: SELECT 
-- +straight_join
l_orderkey, o_orderstatus, count(*) FROM tpch_parquet.lineitem INNER JOIN
tpch_parquet.orders ON o_orderkey = l_orderkey GROUP BY CAST(1 AS INVALID_TYPE),
CAST(2 AS INVALID_TYPE) HAVING count(*) = CAST(1 AS BIGINT)

Correct rewritten SQL:

Analyzed query: SELECT 
-- +straight_join
l_orderkey, o_orderstatus, count(*) FROM tpch_parquet.lineitem INNER JOIN
tpch_parquet.orders ON o_orderkey = l_orderkey GROUP BY l_orderkey,
o_orderstatus HAVING count(*) = CAST(1 AS BIGINT)

The same problem occurs in max-rows-test.test.

The problem is due to the existence of two copies of the grouping expressions. The toSql() function used the original, unanalyzed copy, not the rewritten copy with ordinal replacements.