You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@drill.apache.org by GitBox <gi...@apache.org> on 2019/03/05 19:04:26 UTC
[GitHub] [drill] vdiravka commented on a change in pull request #1640:
DRILL-7038: Queries on partitioned columns scan the entire datasets
vdiravka commented on a change in pull request #1640: DRILL-7038: Queries on partitioned columns scan the entire datasets
URL: https://github.com/apache/drill/pull/1640#discussion_r262637528
##########
File path: exec/java-exec/src/main/java/org/apache/drill/exec/planner/logical/partition/PruneScanRule.java
##########
@@ -550,4 +567,209 @@ private static void setPruneStatus(MetadataContext metaContext, PruneStatus prun
}
}
+ private static class PruneFilesOnScanRule extends PruneScanRule {
+
+ private final Pattern dirPattern;
+
+ private PruneFilesOnScanRule(OptimizerRulesContext optimizerRulesContext) {
+ super(RelOptHelper.some(Aggregate.class, DrillRel.DRILL_LOGICAL, RelOptHelper.any(TableScan.class)),
+ "PruneFilesOnScanRule:Prune_On_Scan", optimizerRulesContext);
+ String partitionColumnLabel = optimizerRulesContext.getPlannerSettings().getFsPartitionColumnLabel();
+ dirPattern = Pattern.compile(partitionColumnLabel + "\\d+");
+ }
+
+ @Override
+ public PartitionDescriptor getPartitionDescriptor(PlannerSettings settings, TableScan scanRel) {
+ return new FileSystemPartitionDescriptor(settings, scanRel);
+ }
+
+ // Checks if query references directory columns only and has DISTINCT or GROUP BY operation
+ @Override
+ public boolean matches(RelOptRuleCall call) {
+ Aggregate aggregate = call.rel(0);
+ TableScan scan = call.rel(1);
+
+ if (!isQualifiedFilePruning(scan)
+ || scan.getRowType().getFieldCount() != aggregate.getRowType().getFieldCount()) {
+ return false;
+ }
+
+ List<String> fieldNames = scan.getRowType().getFieldNames();
+ // Check if select contains partition columns (dir0, dir1, dir2,..., dirN) only
+ for (String field : fieldNames) {
+ if (!dirPattern.matcher(field).matches()) {
+ return false;
+ }
+ }
+
+ return scan.isDistinct() || aggregate.getGroupCount() > 0;
+ }
+
+ /*
+ Transforms Scan node to DrillValuesRel node to avoid unnecessary scanning of selected files.
+ If cache metadata directory file exists, directory columns will be read from it,
+ otherwise directories will be gathered from selection (PartitionLocations).
+ DrillValuesRel will contain gathered constant literals.
+
+ For example, plan for "select dir0, dir1 from `t` group by 1, 2", where table `t` has directory structure year/quarter
+
+ 00-00 Screen
+ 00-01 Project(dir0=[$0], dir1=[$1])
+ 00-02 HashAgg(group=[{0, 1}])
+ 00-03 Scan(table=[[t]], groupscan=[ParquetGroupScan [entries=[ReadEntryWithPath [path=file:/path/t/1996/Q4/orders_96_q4.parquet],
+ ReadEntryWithPath [path=file:/path/t/1996/Q1/file_96_q1.parquet], ReadEntryWithPath [path=file:/path/t/1996/Q3/file_96_q3.parquet],
+ ReadEntryWithPath [path=file:/path/t/1996/Q2/file_96_q2.parquet], ReadEntryWithPath [path=file:/path/t/1994/Q4/file_94_q4.parquet],
+ ReadEntryWithPath [path=file:/path/t/1994/Q1/file_94_q1.parquet], ReadEntryWithPath [path=file:/path/t/1994/Q3/file_94_q3.parquet],
+ ReadEntryWithPath [path=file:/path/t/1994/Q2/file_94_q2.parquet], ReadEntryWithPath [path=file:/path/t/1995/Q4/file_95_q4.parquet],
+ ReadEntryWithPath [path=file:/path/t/1995/Q1/file_95_q1.parquet], ReadEntryWithPath [path=file:/path/t/1995/Q3/file_95_q3.parquet],
+ ReadEntryWithPath [path=file:/path/t/1995/Q2/file_95_q2.parquet]], selectionRoot=file:/path/t, ..., columns=[`dir0`, `dir1`]]])
+
+ will be changed to
+
+ 00-00 Screen
+ 00-01 Project(dir0=[$0], dir1=[$1])
+ 00-02 HashAgg(group=[{0, 1}])
+ 00-03 Values(tuples=[[{ '1995', 'Q1' }, { '1994', 'Q4' }, { '1996', 'Q3' }, { '1996', 'Q2' }, { '1994', 'Q2' },
Review comment:
Good improvement. It will be much faster :)
----------------------------------------------------------------
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:
users@infra.apache.org
With regards,
Apache Git Services