Jackie-Jiang commented on code in PR #8979:
URL: https://github.com/apache/pinot/pull/8979#discussion_r961851815
##########
pinot-core/src/main/java/org/apache/pinot/core/plan/SelectionPlanNode.java:
##########
@@ -50,67 +52,113 @@ public Operator<IntermediateResultsBlock> run() {
List<ExpressionContext> expressions =
SelectionOperatorUtils.extractExpressions(_queryContext, _indexSegment);
int limit = _queryContext.getLimit();
- if (limit > 0) {
- List<OrderByExpressionContext> orderByExpressions =
_queryContext.getOrderByExpressions();
- if (orderByExpressions == null) {
- // Selection only
- TransformOperator transformOperator = new
TransformPlanNode(_indexSegment, _queryContext, expressions,
- Math.min(limit, DocIdSetPlanNode.MAX_DOC_PER_CALL)).run();
- return new SelectionOnlyOperator(_indexSegment, _queryContext,
expressions, transformOperator);
- } else {
- // Selection order-by
- if (isAllOrderByColumnsSorted(orderByExpressions)) {
- // All order-by columns are sorted, no need to sort the records
- TransformOperator transformOperator = new
TransformPlanNode(_indexSegment, _queryContext, expressions,
- Math.min(limit + _queryContext.getOffset(),
DocIdSetPlanNode.MAX_DOC_PER_CALL)).run();
- return new SelectionOrderByOperator(_indexSegment, _queryContext,
expressions, transformOperator, true);
- } else if (orderByExpressions.size() == expressions.size()) {
- // All output expressions are ordered
- TransformOperator transformOperator =
- new TransformPlanNode(_indexSegment, _queryContext, expressions,
DocIdSetPlanNode.MAX_DOC_PER_CALL).run();
- return new SelectionOrderByOperator(_indexSegment, _queryContext,
expressions, transformOperator, false);
- } else {
- // Not all output expressions are ordered, only fetch the order-by
expressions and docId to avoid the
- // unnecessary data fetch
- List<ExpressionContext> expressionsToTransform = new
ArrayList<>(orderByExpressions.size());
- for (OrderByExpressionContext orderByExpression :
orderByExpressions) {
- expressionsToTransform.add(orderByExpression.getExpression());
- }
- TransformOperator transformOperator =
- new TransformPlanNode(_indexSegment, _queryContext,
expressionsToTransform,
- DocIdSetPlanNode.MAX_DOC_PER_CALL).run();
- return new SelectionOrderByOperator(_indexSegment, _queryContext,
expressions, transformOperator, false);
- }
- }
- } else {
+ if (limit == 0) {
// Empty selection (LIMIT 0)
TransformOperator transformOperator = new
TransformPlanNode(_indexSegment, _queryContext, expressions, 0).run();
return new EmptySelectionOperator(_indexSegment, expressions,
transformOperator);
}
+ List<OrderByExpressionContext> orderByExpressions =
_queryContext.getOrderByExpressions();
+
+ if (orderByExpressions == null) {
+ // Selection only
+ // ie: SELECT ... FROM Table WHERE ... LIMIT 10
+ int actualLimit = Math.min(limit, DocIdSetPlanNode.MAX_DOC_PER_CALL);
+ TransformPlanNode planNode = new TransformPlanNode(_indexSegment,
_queryContext, expressions, actualLimit);
+ TransformOperator transformOperator = planNode.run();
+
+ return new SelectionOnlyOperator(_indexSegment, _queryContext,
expressions, transformOperator);
+ }
+ int numOrderByExpressions = orderByExpressions.size();
+ // Although it is a break of abstraction, some code, specially merging,
assumes that if there is an order by
+ // expression the operator will return a block whose selection result is a
priority queue.
+ int sortedColumnsPrefixSize = getSortedColumnsPrefix(orderByExpressions);
Review Comment:
It might be sorted or not based on the default value of the column. E.g. if
default value for a int column is `0`, then `-1, null, 1` is considered sorted;
but if the default value is `-100` then it is not sorted.
The null handling support feature is added recently, and we treat null as
default value when generating the index (we don't know if null handling is
enabled at ingestion time either). So what we can do here is deciding whether
to use this optimization at query time, and not use it when null handling is
enabled and the column contains null values
--
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.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
