Csaba Ringhofer created IMPALA-15710:
----------------------------------------
Summary: Improve cardinality estimates of deletes in Hive ACID
tables
Key: IMPALA-15710
URL: https://issues.apache.org/jira/browse/IMPALA-15710
Project: IMPALA
Issue Type: Improvement
Components: Frontend
Reporter: Csaba Ringhofer
Currently the estimated number of deleted rows is calculated based on the
number of rows in the whole table, which can massively overestimate the real
number and lead to inflated mem estimates for the anti-join operator that adds
the delete. I don't know of precise stats available for this, but calculating
some cap based on the size of delete files could reduce extreme overestimation.
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)
