[
https://issues.apache.org/jira/browse/TAJO-1553?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14560291#comment-14560291
]
ASF GitHub Bot commented on TAJO-1553:
--------------------------------------
Github user jihoonson commented on a diff in the pull request:
https://github.com/apache/tajo/pull/583#discussion_r31097759
--- Diff:
tajo-core/src/main/java/org/apache/tajo/engine/planner/global/rewriter/rules/BroadcastJoinRule.java
---
@@ -0,0 +1,341 @@
+/**
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements. See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership. The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License. You may obtain a copy of the License at
+ *
+ * http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.apache.tajo.engine.planner.global.rewriter.rules;
+
+import org.apache.tajo.ExecutionBlockId;
+import org.apache.tajo.OverridableConf;
+import org.apache.tajo.SessionVars;
+import org.apache.tajo.engine.planner.global.ExecutionBlock;
+import org.apache.tajo.engine.planner.global.GlobalPlanner;
+import org.apache.tajo.engine.planner.global.MasterPlan;
+import
org.apache.tajo.engine.planner.global.rewriter.GlobalPlanRewriteRule;
+import org.apache.tajo.plan.LogicalPlan;
+import org.apache.tajo.plan.PlanningException;
+import org.apache.tajo.plan.logical.*;
+import org.apache.tajo.plan.util.PlannerUtil;
+import org.apache.tajo.util.TUtil;
+import org.apache.tajo.util.graph.DirectedGraphVisitor;
+
+import java.util.*;
+
+/**
+ * {@link BroadcastJoinRule} converts repartition join plan into broadcast
join plan.
+ * Broadcast join rules can be defined as follows.
+ *
+ * <h3>Broadcastable relation</h3>
+ * A relation is broadcastable when its size is smaller than a given
threshold.
+ *
+ * <h3>Assumetion</h3>
+ * If every input of an execution block is broadcastable, the output of
the execution block is also broadcastable.
+ *
+ * <h3>Rules to convert repartition join into broadcast join</h3>
+ * <ul>
+ * <li>Given an EB containing a join and its child EBs, those EBs can be
merged into a single EB if at least one child EB's output is broadcastable.</li>
+ * <li>Given a user-defined threshold, the total size of broadcast
relations of an EB cannot exceed such threshold.</li>
+ * <ul>
+ * <li>After merging EBs according to the first rule, the result EB
may not satisfy the second rule. In this case, enforce repartition join for
large relations to satisfy the second rule.</li>
+ * </ul>
+ * <li>Preserved-row relations cannot be broadcasted to avoid duplicated
results. That is, full outer join cannot be executed with broadcast join.</li>
+ * <ul>
+ * <li>Here is brief backgrounds for this rule. Data of preserved-row
relations will be appeared in the join result regardless of join conditions. If
multiple tasks execute outer join with broadcasted preserved-row relations,
they emit duplicates results.</li>
+ * <li>Even though a single task can execute outer join when every
input is broadcastable, broadcast join is not allowed if one of input relation
consists of multiple files.</li>
+ * </ul>
+ * </ul>
+ *
+ */
+public class BroadcastJoinRule implements GlobalPlanRewriteRule {
+
+ private BroadcastJoinPlanBuilder planBuilder;
+ private BroadcastJoinPlanFinalizer planFinalizer;
+
+ @Override
+ public String getName() {
+ return "BroadcastJoinRule";
+ }
+
+ @Override
+ public boolean isEligible(OverridableConf queryContext, MasterPlan plan)
{
+ if (queryContext.getBool(SessionVars.TEST_BROADCAST_JOIN_ENABLED)) {
+ for (LogicalPlan.QueryBlock block :
plan.getLogicalPlan().getQueryBlocks()) {
+ if (block.hasNode(NodeType.JOIN)) {
+ long broadcastSizeThreshold =
queryContext.getLong(SessionVars.BROADCAST_TABLE_SIZE_LIMIT);
+ if (broadcastSizeThreshold > 0) {
+ GlobalPlanRewriteUtil.ParentFinder parentFinder = new
GlobalPlanRewriteUtil.ParentFinder();
+ RelationSizeComparator relSizeComparator = new
RelationSizeComparator();
+ planBuilder = new BroadcastJoinPlanBuilder(plan,
relSizeComparator, parentFinder, broadcastSizeThreshold);
+ planFinalizer = new BroadcastJoinPlanFinalizer(plan,
relSizeComparator);
+ return true;
+ }
+ }
+ }
+ }
+ return false;
+ }
+
+ @Override
+ public MasterPlan rewrite(MasterPlan plan) throws PlanningException{
+ plan.accept(plan.getRoot().getId(), planBuilder);
+ plan.accept(plan.getRoot().getId(), planFinalizer);
+ return plan;
+ }
+
+ private static class RelationSizeComparator implements
Comparator<ScanNode> {
+
+ @Override
+ public int compare(ScanNode o1, ScanNode o2) {
+ return (int) (GlobalPlanRewriteUtil.getTableVolume(o1) -
GlobalPlanRewriteUtil.getTableVolume(o2));
--- End diff --
Improved to consider overflow.
> Improve broadcast join planning
> -------------------------------
>
> Key: TAJO-1553
> URL: https://issues.apache.org/jira/browse/TAJO-1553
> Project: Tajo
> Issue Type: Improvement
> Components: distributed query plan, planner/optimizer
> Reporter: Jihoon Son
> Assignee: Jihoon Son
> Fix For: 0.11.0
>
>
> The global engine generates a logical plan, and then marks some parts of the
> plan as broadcast plan which means that they and their input will be
> broadcasted to all workers.
> Currently, broadcast parts are identified according to some rigid and
> hard-coded rules. This will limit the broadcast opportunities in many cases.
> So, in this issue, I propose refactoring the broadcast planner to be more
> general.
> Broadcast parts can be identified recursively.
> * A leaf node will be broadcasted if its input size does not exceed the
> pre-defined threshold.
> * An intermediate node will be broadcasted if it has at least one broadcast
> child.
> * For outer joins, row-preserved tables must not be broadcasted to avoid
> input data duplication.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
