From f62cdecc08bdc6cd3ec691f885034adcaa0706fd Mon Sep 17 00:00:00 2001 From: minghong Date: Fri, 22 Mar 2024 20:53:30 +0800 Subject: [PATCH] [fix](Nereids) do not push down topn-filter through right/full outer join if the first orderkey is nulls first (#32633) * do not push down topn-filter through right/full outer join if the first order key is nulls first --- .../nereids/processor/post/TopNScanOpt.java | 16 +++++++++++++--- .../nereids_tpch_p0/tpch/topn-filter.groovy | 15 +++++++++++++++ 2 files changed, 28 insertions(+), 3 deletions(-) diff --git a/fe/fe-core/src/main/java/org/apache/doris/nereids/processor/post/TopNScanOpt.java b/fe/fe-core/src/main/java/org/apache/doris/nereids/processor/post/TopNScanOpt.java index a9425cb715..63754afe4d 100644 --- a/fe/fe-core/src/main/java/org/apache/doris/nereids/processor/post/TopNScanOpt.java +++ b/fe/fe-core/src/main/java/org/apache/doris/nereids/processor/post/TopNScanOpt.java @@ -22,6 +22,7 @@ import org.apache.doris.nereids.trees.expressions.Expression; import org.apache.doris.nereids.trees.expressions.SlotReference; import org.apache.doris.nereids.trees.plans.Plan; import org.apache.doris.nereids.trees.plans.SortPhase; +import org.apache.doris.nereids.trees.plans.algebra.Join; import org.apache.doris.nereids.trees.plans.algebra.OlapScan; import org.apache.doris.nereids.trees.plans.physical.PhysicalCatalogRelation; import org.apache.doris.nereids.trees.plans.physical.PhysicalDeferMaterializeTopN; @@ -89,7 +90,9 @@ public class TopNScanOpt extends PlanPostProcessor { if (! (firstKey instanceof SlotReference)) { return Optional.empty(); } - OlapScan olapScan = findScanNodeBySlotReference(topN, (SlotReference) firstKey); + + boolean nullsFirst = topN.getOrderKeys().get(0).isNullFirst(); + OlapScan olapScan = findScanNodeBySlotReference(topN, (SlotReference) firstKey, nullsFirst); if (olapScan != null && olapScan.getTable().isDupKeysOrMergeOnWrite() && olapScan instanceof PhysicalCatalogRelation) { @@ -99,7 +102,14 @@ public class TopNScanOpt extends PlanPostProcessor { return Optional.empty(); } - private OlapScan findScanNodeBySlotReference(Plan root, SlotReference slot) { + private OlapScan findScanNodeBySlotReference(Plan root, SlotReference slot, boolean nullsFirst) { + // topn-filter cannot be pushed through right/full outer join if the first orderKey is nulls first + if (nullsFirst && root instanceof Join) { + Join join = (Join) root; + if (join.getJoinType().isRightOuterJoin() || join.getJoinType().isFullOuterJoin()) { + return null; + } + } OlapScan target = null; if (root instanceof OlapScan && root.getOutputSet().contains(slot)) { return (OlapScan) root; @@ -109,7 +119,7 @@ public class TopNScanOpt extends PlanPostProcessor { // TODO for union, topn-filter can be pushed down to all of its children. Plan child = root.child(0); if (!(child instanceof PhysicalWindow) && child.getOutputSet().contains(slot)) { - target = findScanNodeBySlotReference(child, slot); + target = findScanNodeBySlotReference(child, slot, nullsFirst); if (target != null) { return target; } diff --git a/regression-test/suites/nereids_tpch_p0/tpch/topn-filter.groovy b/regression-test/suites/nereids_tpch_p0/tpch/topn-filter.groovy index 14733d1303..3a29274a2f 100644 --- a/regression-test/suites/nereids_tpch_p0/tpch/topn-filter.groovy +++ b/regression-test/suites/nereids_tpch_p0/tpch/topn-filter.groovy @@ -116,6 +116,21 @@ suite("topn-filter") { } qt_groupingsets "select n_regionkey, sum(n_nationkey) from nation group by grouping sets((n_regionkey)) order by n_regionkey limit 2;" + + explain { + sql "select * from nation right outer join region on r_regionkey = n_regionkey order by n_regionkey nulls first limit 1; " + notContains "TOPN OPT:" + } + + explain { + sql "select * from nation right outer join region on r_regionkey = n_regionkey order by n_regionkey nulls last limit 1; " + contains "TOPN OPT:" + } + + explain { + sql "select * from nation right outer join region on r_regionkey = n_regionkey order by n_regionkey nulls last, n_name nulls first limit 1; " + contains "TOPN OPT:" + } sql "set enable_pipeline_engine=false;" sql "set enable_pipeline_x_engine=false;"