Quanlong Huang created IMPALA-15543:
---------------------------------------
Summary: Use HBO stats in HdfsScanNode.getNumInputRows()
Key: IMPALA-15543
URL: https://issues.apache.org/jira/browse/IMPALA-15543
Project: IMPALA
Issue Type: Improvement
Reporter: Quanlong Huang
If numRows is missing in HMS for any selected partitions,
HdfsScanNode.getNumInputRows() returns -1. This can be improved by using HBO
stats of full scan with the same partition predicates. For instance, if there
are HBO stats of this query
{code:sql}
select max(id) from tbl where part_col = 1;{code}
We can use its scan output cardinality as the number of scan input rows for
this query
{code:sql}
select max(id) from tbl where part_col = 1 and id < 1000{code}
assuming that part_col is a partition column and id is a non-partition column.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
