Details
-
Sub-task
-
Status: Resolved
-
Major
-
Resolution: Fixed
-
Impala 2.2, Impala 2.3.0
Description
The Parquet scanner attempts to collect statistics about how many rows are filtered out using the bitmap filters. However, it incorrectly counts the number of filtered rows so that it thinks that every row is rejected by the filter, i.e. that the filter is awesome. This means the bitmap filters are never disabled.
Fix this obvious bug and do some performance validation that the heuristic is good.