paul-rogers commented on a change in pull request #1955: DRILL-7491: Incorrect
count() returned for complex types in parquet
URL: https://github.com/apache/drill/pull/1955#discussion_r366676515
##########
File path:
exec/java-exec/src/main/java/org/apache/drill/exec/physical/base/AbstractGroupScanWithMetadata.java
##########
@@ -167,29 +167,43 @@ public boolean isMatchAllMetadata() {
*/
@Override
public long getColumnValueCount(SchemaPath column) {
- long tableRowCount, colNulls;
- Long nulls;
ColumnStatistics<?> columnStats =
getTableMetadata().getColumnStatistics(column);
- ColumnStatistics<?> nonInterestingColStats = null;
- if (columnStats == null) {
- nonInterestingColStats =
getNonInterestingColumnsMetadata().getColumnStatistics(column);
- }
+ ColumnStatistics<?> nonInterestingColStats = (columnStats == null)
+ ? getNonInterestingColumnsMetadata().getColumnStatistics(column) :
null;
+ long tableRowCount;
if (columnStats != null) {
tableRowCount =
TableStatisticsKind.ROW_COUNT.getValue(getTableMetadata());
} else if (nonInterestingColStats != null) {
tableRowCount =
TableStatisticsKind.ROW_COUNT.getValue(getNonInterestingColumnsMetadata());
+ columnStats = nonInterestingColStats;
+ } else if (existsNestedStatsForColumn(column, getTableMetadata())
+ || existsNestedStatsForColumn(column,
getNonInterestingColumnsMetadata())) {
+ return Statistic.NO_COLUMN_STATS;
} else {
return 0; // returns 0 if the column doesn't exist in the table.
}
- columnStats = columnStats != null ? columnStats : nonInterestingColStats;
- nulls = ColumnStatisticsKind.NULLS_COUNT.getFrom(columnStats);
- colNulls = nulls != null ? nulls : Statistic.NO_COLUMN_STATS;
+ Long nulls = ColumnStatisticsKind.NULLS_COUNT.getFrom(columnStats);
+ if (nulls == null || Statistic.NO_COLUMN_STATS == nulls ||
Statistic.NO_COLUMN_STATS == tableRowCount) {
+ return Statistic.NO_COLUMN_STATS;
+ } else {
+ return tableRowCount - nulls;
+ }
+ }
- return Statistic.NO_COLUMN_STATS == tableRowCount
- || Statistic.NO_COLUMN_STATS == colNulls
- ? Statistic.NO_COLUMN_STATS : tableRowCount - colNulls;
+ /**
+ * For complex columns, stats may be present only for nested fields. For
example, a column path is `a`,
+ * but stats present for `a`.`b`. So before making a decision that column is
absent, the case needs
+ * to be tested.
+ *
+ * @param column column path
+ * @param metadata metadata with column statistics
+ * @return whether stats exists for nested fields
+ */
+ private boolean existsNestedStatsForColumn(SchemaPath column, Metadata
metadata) {
Review comment:
I wonder about the premise of this function. The Java doc suggests we have
the path `a.b`. We want to know if `a` is an uninteresting column. Are we
suggesting that we have metadata that says that `a.b` is uninteresting, but we
have no metadata for `a` itself?
Why would this be? Do we store columns in a non-structured way? That is, do
we have a list of columns like 'a', 'b', 'c.d', 'e.f.g', 'c.h', and so on?
Rather than `(a, b, c(d, g), e((g)))`?
Further we seem to be assuming that we will gather stats for, say, `a.c`,
but not `a.b`.
The only place I can see that such complexity would make sense is to
estimate cardinality for Flatten or Lateral Join. In such a case, might it make
more sense to treat this recursively as a nested table? That is 'a` is a table
and `a.b` and `a.c` are just `b` and `c` columns in the nested table.
Given this, we *must* have a metadata entry for `a` (the map) in order to
have metadata for any of its nested columns.
So, again, I wonder if we've defined our metadata semantics as clearly as we
might.
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to 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
With regards,
Apache Git Services
