[ https://issues.apache.org/jira/browse/DRILL-5796?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16550948#comment-16550948 ]
ASF GitHub Bot commented on DRILL-5796: --------------------------------------- vrozov commented on a change in pull request #1298: DRILL-5796: Filter pruning for multi rowgroup parquet file URL: https://github.com/apache/drill/pull/1298#discussion_r204099135 ########## File path: exec/java-exec/src/main/java/org/apache/drill/exec/expr/stat/ParquetIsPredicate.java ########## @@ -124,8 +124,7 @@ private static LogicalExpression createIsTruePredicate(LogicalExpression expr) { */ private static LogicalExpression createIsFalsePredicate(LogicalExpression expr) { return new ParquetIsPredicate<Boolean>(expr, (exprStat, evaluator) -> - //if min value is not false or if there are all nulls -> canDrop - isAllNulls(exprStat, evaluator.getRowCount()) || exprStat.hasNonNullValue() && ((BooleanStatistics) exprStat).getMin() + isAllNulls(exprStat, evaluator.getRowCount()) || exprStat.hasNonNullValue() && ((BooleanStatistics) exprStat).getMin() ? RowsMatch.NONE : checkNull(exprStat) Review comment: I don't understand how it works for the use cases you provided, but I would expect the following unit test to pass while it actually fails: ``` public void testSOME() { LogicalExpression le = mock(LogicalExpression.class); BooleanStatistics booleanStatistics = mock(BooleanStatistics.class); doReturn(booleanStatistics).when(le).accept(ArgumentMatchers.any(), ArgumentMatchers.any()); RangeExprEvaluator<Boolean> re = mock(RangeExprEvaluator.class); when(re.getRowCount()).thenReturn(Long.valueOf(2)); // 2 rows when(booleanStatistics.isEmpty()).thenReturn(FALSE); // stat is not empty when(booleanStatistics.isNumNullsSet()).thenReturn(TRUE); // num_nulls set when(booleanStatistics.getNumNulls()).thenReturn(Long.valueOf(0)); // no nulls when(booleanStatistics.hasNonNullValue()).thenReturn(TRUE); // min/max set when(booleanStatistics.getMin()).thenReturn(Boolean.FALSE); // min false when(booleanStatistics.getMax()).thenReturn(TRUE); // max true assertEquals(SOME, ((ParquetIsPredicate<Boolean>) createIsPredicate(IS_TRUE, le)).matches(re)); assertEquals(SOME, ((ParquetIsPredicate<Boolean>) createIsPredicate(IS_FALSE, le)).matches(re)); } ``` ``` java.lang.AssertionError: Expected :SOME Actual :ALL ``` I would prefer to see similar unit tests and not the integration tests that you point me to and that may succeed for a different reason (I'd like to see an explanation why they succeed as well). ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org > Filter pruning for multi rowgroup parquet file > ---------------------------------------------- > > Key: DRILL-5796 > URL: https://issues.apache.org/jira/browse/DRILL-5796 > Project: Apache Drill > Issue Type: Improvement > Components: Storage - Parquet > Reporter: Damien Profeta > Assignee: Jean-Blas IMBERT > Priority: Major > Fix For: 1.14.0 > > > Today, filter pruning use the file name as the partitioning key. This means > you can remove a partition only if the whole file is for the same partition. > With parquet, you can prune the filter if the rowgroup make a partition of > your dataset as the unit of work if the rowgroup not the file. -- This message was sent by Atlassian JIRA (v7.6.3#76005)