Zoltán Borók-Nagy created IMPALA-15526:
------------------------------------------
Summary: tpcds-unmodified q78 joins cs_item_sk to itself instead
of ss_item_sk
Key: IMPALA-15526
URL: https://issues.apache.org/jira/browse/IMPALA-15526
Project: IMPALA
Issue Type: Bug
Components: Infrastructure, Test
Reporter: Zoltán Borók-Nagy
testdata/workloads/tpcds-unmodified/queries/tpcds-q78.test:49 (master
1f34ea915e) has:
{code:sql}
left join cs on (cs_sold_year=ss_sold_year and cs_item_sk=cs_item_sk and
cs_customer_sk=ss_customer_sk)
{code}
The TPC-DS specification, and the tpcds and tpcds_partitioned copies of the
query, use cs_item_sk=ss_item_sk.
With the typo, the cs join matches on (year, customer) only, so it becomes
many-to-many, and the planner gives the self-equality the default 0.1
selectivity. Results differ from the specification query, and performance
numbers from this workload are not comparable with other engines or published
TPC-DS results. Since tpcds-unmodified is the workload used for performance
tests, the fix also changes baseline numbers.
Fix: replace cs_item_sk=cs_item_sk with cs_item_sk=ss_item_sk.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
