Steve Carlin created IMPALA-14324:
-------------------------------------
Summary: Calcite planner: Wrong cardinality shown with is_not_null
predicate in scan node
Key: IMPALA-14324
URL: https://issues.apache.org/jira/browse/IMPALA-14324
Project: IMPALA
Issue Type: Sub-task
Reporter: Steve Carlin
Taken from review: [https://gerrit.cloudera.org/#/c/22870]
When running TpcdsCpuCostPlannerTest under the java/calcite-planner directory,
the cardinality for store_returns is showing up as: (filtered from 86.40M)
It should be 864M. The reason it is only .1 times the size is because the
is_not_null pred filter selectivity is being treated as filtering out 90% of
the rows. This is due to the Planner::computeSelectivities method using the
following lines:
{code:java}
if (selectivities.size() != conjunctSize) {
// Some conjuncts have no estimated selectivity. Use a single default
// representative selectivity for all those conjuncts.
selectivities.add(Expr.DEFAULT_SELECTIVITY);
}{code}
We need the getSelectivity for the "is not null" predicate expression to return
the right result.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
