Zoltán Borók-Nagy created IMPALA-11797:
------------------------------------------
Summary: LEFT ANTI JOIN's selectivity is always 1.0
Key: IMPALA-11797
URL: https://issues.apache.org/jira/browse/IMPALA-11797
Project: IMPALA
Issue Type: Bug
Components: Frontend
Reporter: Zoltán Borók-Nagy
We use the [following
formula|https://github.com/apache/impala/blob/b88cfadbbde45ccbe5e4b6d644d046a9a275e31c/fe/src/main/java/org/apache/impala/planner/JoinNode.java#L690]
to calculate the LEFT ANTI JOIN's selectivity:
{noformat}
selectivity = (double) Math.max(lhsNdv - rhsNdv, lhsNdv) / (double) lhsNdv;
{noformat}
{{Math.max(lhsNdv - rhsNdv, lhsNdv)}} is always equals to {{lhsNdv}}, hence the
formula becomes:
{noformat}
lhsNdv / lhsNdv
{noformat}
which is always 1.0. We should come up with a better formula.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
