[
https://issues.apache.org/jira/browse/DRILL-6385?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16569615#comment-16569615
]
ASF GitHub Bot commented on DRILL-6385:
---------------------------------------
sohami commented on a change in pull request #1334: DRILL-6385: Support JPPD
feature
URL: https://github.com/apache/drill/pull/1334#discussion_r207753050
##########
File path:
exec/java-exec/src/main/java/org/apache/drill/exec/work/filter/RuntimeFilterManager.java
##########
@@ -0,0 +1,666 @@
+/*
+ * 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.drill.exec.work.filter;
+
+import org.apache.calcite.plan.volcano.RelSubset;
+import org.apache.calcite.rel.RelNode;
+import org.apache.calcite.rel.core.JoinInfo;
+import org.apache.calcite.rel.core.JoinRelType;
+import org.apache.calcite.rel.metadata.RelMetadataQuery;
+import org.apache.calcite.rel.type.RelDataType;
+import org.apache.calcite.rel.type.RelDataTypeField;
+import org.apache.calcite.util.ImmutableBitSet;
+import org.apache.commons.collections.CollectionUtils;
+import org.apache.drill.exec.ExecConstants;
+import org.apache.drill.exec.ops.AccountingDataTunnel;
+import org.apache.drill.exec.ops.Consumer;
+import org.apache.drill.exec.ops.QueryContext;
+import org.apache.drill.exec.ops.SendingAccountor;
+import org.apache.drill.exec.ops.StatusHandler;
+import org.apache.drill.exec.physical.PhysicalPlan;
+
+import org.apache.drill.exec.physical.base.AbstractPhysicalVisitor;
+import org.apache.drill.exec.physical.base.Exchange;
+import org.apache.drill.exec.physical.base.GroupScan;
+import org.apache.drill.exec.physical.base.PhysicalOperator;
+import org.apache.drill.exec.physical.config.BroadcastExchange;
+import org.apache.drill.exec.physical.config.HashJoinPOP;
+import org.apache.drill.exec.planner.fragment.Fragment;
+import org.apache.drill.exec.planner.fragment.Wrapper;
+import org.apache.drill.exec.planner.physical.HashAggPrel;
+import org.apache.drill.exec.planner.physical.HashJoinPrel;
+import org.apache.drill.exec.planner.physical.Prel;
+import org.apache.drill.exec.planner.physical.ScanPrel;
+import org.apache.drill.exec.planner.physical.StreamAggPrel;
+import org.apache.drill.exec.planner.physical.visitor.BasePrelVisitor;
+import org.apache.drill.exec.proto.BitData;
+import org.apache.drill.exec.proto.CoordinationProtos;
+import org.apache.drill.exec.proto.GeneralRPCProtos;
+import org.apache.drill.exec.proto.UserBitShared;
+import org.apache.drill.exec.proto.helper.QueryIdHelper;
+import org.apache.drill.exec.rpc.RpcException;
+import org.apache.drill.exec.rpc.RpcOutcomeListener;
+import org.apache.drill.exec.rpc.data.DataTunnel;
+import org.apache.drill.exec.server.DrillbitContext;
+import org.apache.drill.exec.util.Pointer;
+import org.apache.drill.exec.work.QueryWorkUnit;
+import org.slf4j.Logger;
+import org.slf4j.LoggerFactory;
+
+import java.util.ArrayList;
+import java.util.HashMap;
+import java.util.List;
+import java.util.Map;
+import java.util.Set;
+import java.util.concurrent.ConcurrentHashMap;
+
+/**
+ * This class traverses the physical operator tree to find the HashJoin
operator
+ * for which is JPPD (join predicate push down) is possible. The prerequisite
to do JPPD
+ * is:
+ * 1. The join condition is equality
+ * 2. The physical join node is a HashJoin one
+ * 3. The probe side children of the HashJoin node should not contain a
blocking operator like HashAgg
+ */
+public class RuntimeFilterManager {
+
+ private Wrapper rootWrapper;
+ //HashJoin node's major fragment id to its corresponding probe side nodes's
endpoints
+ private Map<Integer, List<CoordinationProtos.DrillbitEndpoint>>
joinMjId2probdeScanEps = new HashMap<>();
+ //HashJoin node's major fragment id to its corresponding probe side nodes's
number
+ private Map<Integer, Integer> joinMjId2scanSize = new ConcurrentHashMap<>();
+ //HashJoin node's major fragment id to its corresponding probe side scan
node's belonging major fragment id
+ private Map<Integer, Integer> joinMjId2ScanMjId = new HashMap<>();
+
+ private RuntimeFilterWritable aggregatedRuntimeFilter;
+
+ private DrillbitContext drillbitContext;
+
+ private SendingAccountor sendingAccountor = new SendingAccountor();
+
+ private String lineSeparator;
+
+ private static final Logger logger =
LoggerFactory.getLogger(RuntimeFilterManager.class);
+
+ /**
+ * This class maintains context for the runtime join push down's filter
management. It
+ * does a traversal of the physical operators by leveraging the root wrapper
which indirectly
+ * holds the global PhysicalOperator tree and contains the minor fragment
endpoints.
+ * @param workUnit
+ * @param drillbitContext
+ */
+ public RuntimeFilterManager(QueryWorkUnit workUnit, DrillbitContext
drillbitContext) {
+ this.rootWrapper = workUnit.getRootWrapper();
+ this.drillbitContext = drillbitContext;
+ lineSeparator = java.security.AccessController.doPrivileged(new
sun.security.action.GetPropertyAction("line.separator"));
+ }
+
+
+ /**
+ * Step 1 :
+ * Generate a possible RuntimeFilter of a HashJoinPrel, left some BF
parameters of the generated RuntimeFilter
+ * to be set later.
+ * @param hashJoinPrel
+ * @return null or a partial information RuntimeFilterDef
+ */
+ public static RuntimeFilterDef generateRuntimeFilter(HashJoinPrel
hashJoinPrel) {
+ JoinRelType joinRelType = hashJoinPrel.getJoinType();
+ JoinInfo joinInfo = hashJoinPrel.analyzeCondition();
+ boolean allowJoin = (joinInfo.isEqui()) && (joinRelType ==
JoinRelType.INNER || joinRelType == JoinRelType.RIGHT);
+ if (!allowJoin) {
+ return null;
+ }
+ List<BloomFilterDef> bloomFilterDefs = new ArrayList<>();
+ //find the possible left scan node of the left join key
+ GroupScan groupScan = null;
+ RelNode left = hashJoinPrel.getLeft();
+ List<String> leftFields = left.getRowType().getFieldNames();
+ List<Integer> leftKeys = hashJoinPrel.getLeftKeys();
+ RelMetadataQuery metadataQuery = left.getCluster().getMetadataQuery();
+ for (Integer leftKey : leftKeys) {
+ String leftFieldName = leftFields.get(leftKey);
+ //This also avoids the left field of the join condition with a function
call.
+ ScanPrel scanPrel = findLeftScanPrel(leftFieldName, left);
+ if (scanPrel != null) {
+ boolean encounteredBlockNode = containBlockNode((Prel) left, scanPrel);
+ if (encounteredBlockNode) {
+ continue;
+ }
+ //Collect NDV from the Metadata
+ RelDataType scanRowType = scanPrel.getRowType();
+ RelDataTypeField field = scanRowType.getField(leftFieldName, true,
true);
+ int index = field.getIndex();
+ Double ndv = metadataQuery.getDistinctRowCount(scanPrel,
ImmutableBitSet.of(index), null);
+ if (ndv == null) {
+ //If NDV is not supplied, we use the row count to estimate the ndv.
+ ndv = left.estimateRowCount(metadataQuery) * 0.1;
+ }
+ //only the probe side field name and hash seed is definite, other
information left to pad later
+ BloomFilterDef bloomFilterDef = new BloomFilterDef(0, 0,false,
leftFieldName);
+ bloomFilterDef.setLeftNDV(ndv);
+ bloomFilterDefs.add(bloomFilterDef);
+ groupScan = scanPrel.getGroupScan();
+ }
+ }
+ if (bloomFilterDefs.size() > 0) {
+ //only bloom filters parameter & probe side GroupScan is definitely set
here
+ RuntimeFilterDef runtimeFilterDef = new RuntimeFilterDef(true, false,
bloomFilterDefs, false);
+ runtimeFilterDef.setProbeSideGroupScan(groupScan);
+ return runtimeFilterDef;
+ }
+ return null;
+ }
+
+ /**
+ * Step 2:
+ * Complement the RuntimeFilter information
+ * @param plan
+ * @param queryContext
+ */
+ public static void complementRuntimeFilterInfo(PhysicalPlan plan,
QueryContext queryContext) {
+ final PhysicalOperator rootOperator =
plan.getSortedOperators(false).iterator().next();
+ RuntimeFilterInfoPaddingHelper runtimeFilterInfoPaddingHelper = new
RuntimeFilterInfoPaddingHelper(queryContext);
+ rootOperator.accept(runtimeFilterInfoPaddingHelper, null);
+ }
+
+
+ /**
+ * Step3 :
+ * This method is to collect the parallel information of the
RuntimetimeFilters. Then it res a RuntimeFilterRouting to
+ * record the relationship between the RuntimeFilter producers and consumers.
+ */
+ public void collectRuntimeFilterParallelAndControlInfo(Pointer<String>
textPlan) {
+ Map<String, String> mjOpIdPair2runtimeFilter = new HashMap<>();
+ RuntimeFilterParallelismCollector runtimeFilterParallelismCollector = new
RuntimeFilterParallelismCollector();
+ rootWrapper.getNode().getRoot().accept(runtimeFilterParallelismCollector,
null);
+ List<RFHelperHolder> holders =
runtimeFilterParallelismCollector.getHolders();
+
+ for (RFHelperHolder holder : holders) {
+ List<CoordinationProtos.DrillbitEndpoint> probeSideEndpoints =
holder.getProbeSideScanEndpoints();
+ int probeSideScanMajorId = holder.getProbeSideScanMajorId();
+ int joinNodeMajorId = holder.getJoinMajorId();
+ RuntimeFilterDef runtimeFilterDef = holder.getRuntimeFilterDef();
+ boolean sendToForeman = runtimeFilterDef.isSendToForeman();
+ //mark the runtime filter info to the profile
+ int probeSideScanOpId = holder.getProbeSideScanOpId();
+ List<BloomFilterDef> bloomFilterDefs =
runtimeFilterDef.getBloomFilterDefs();
+ String mjOpIdPair = String.format("%02d-%02d", probeSideScanMajorId,
probeSideScanOpId);
+ StringBuilder stringBuilder = new StringBuilder();
+ stringBuilder.append("RuntimeFilter[");
+ for (BloomFilterDef bloomFilterDef : bloomFilterDefs) {
+ stringBuilder.append(bloomFilterDef.toString()).append(",");
+ }
+ stringBuilder.append("]");
+ String runtimeFiltersJson = stringBuilder.toString();
+ mjOpIdPair2runtimeFilter.put(mjOpIdPair, runtimeFiltersJson);
+ if (sendToForeman) {
+ //send RuntimeFilter to Foreman
+ joinMjId2probdeScanEps.put(joinNodeMajorId, probeSideEndpoints);
+ joinMjId2scanSize.put(joinNodeMajorId, probeSideEndpoints.size());
+ joinMjId2ScanMjId.put(joinNodeMajorId, probeSideScanMajorId);
+ }
+ }
+ reconstructTextPlan(textPlan, mjOpIdPair2runtimeFilter);
+ }
+
+
+ public void waitForComplete() {
+ sendingAccountor.waitForSendComplete();
+ }
+
+ /**
+ * This method is passively invoked by receiving a runtime filter from the
network
+ * @param runtimeFilterWritable
+ */
+ public void registerRuntimeFilter(RuntimeFilterWritable
runtimeFilterWritable) {
+ BitData.RuntimeFilterBDef runtimeFilterB =
runtimeFilterWritable.getRuntimeFilterBDef();
+ int majorId = runtimeFilterB.getMajorFragmentId();
+ UserBitShared.QueryId queryId = runtimeFilterB.getQueryId();
+ List<String> probeFields = runtimeFilterB.getProbeFieldsList();
+ logger.info("RuntimeFilterManager receives a runtime filter , majorId:{},
queryId:{}", majorId, QueryIdHelper.getQueryId(queryId));
+ int size;
+ synchronized (this) {
+ size = joinMjId2scanSize.get(majorId);
Review comment:
Check for `Precondition.checkState(size > 0)` in first place.
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on 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
> Support JPPD (Join Predicate Push Down)
> ---------------------------------------
>
> Key: DRILL-6385
> URL: https://issues.apache.org/jira/browse/DRILL-6385
> Project: Apache Drill
> Issue Type: New Feature
> Components: Server, Execution - Flow
> Affects Versions: 1.14.0
> Reporter: weijie.tong
> Assignee: weijie.tong
> Priority: Major
>
> This feature is to support the JPPD (Join Predicate Push Down). It will
> benefit the HashJoin ,Broadcast HashJoin performance by reducing the number
> of rows to send across the network ,the memory consumed. This feature is
> already supported by Impala which calls it RuntimeFilter
> ([https://www.cloudera.com/documentation/enterprise/5-9-x/topics/impala_runtime_filtering.html]).
> The first PR will try to push down a bloom filter of HashJoin node to
> Parquet’s scan node. The propose basic procedure is described as follow:
> # The HashJoin build side accumulate the equal join condition rows to
> construct a bloom filter. Then it sends out the bloom filter to the foreman
> node.
> # The foreman node accept the bloom filters passively from all the fragments
> that has the HashJoin operator. It then aggregates the bloom filters to form
> a global bloom filter.
> # The foreman node broadcasts the global bloom filter to all the probe side
> scan nodes which maybe already have send out partial data to the hash join
> nodes(currently the hash join node will prefetch one batch from both sides ).
> 4. The scan node accepts a global bloom filter from the foreman node.
> It will filter the rest rows satisfying the bloom filter.
>
> To implement above execution flow, some main new notion described as below:
> 1. RuntimeFilter
> It’s a filter container which may contain BloomFilter or MinMaxFilter.
> 2. RuntimeFilterReporter
> It wraps the logic to send hash join’s bloom filter to the foreman.The
> serialized bloom filter will be sent out through the data tunnel.This object
> will be instanced by the FragmentExecutor and passed to the
> FragmentContext.So the HashJoin operator can obtain it through the
> FragmentContext.
> 3. RuntimeFilterRequestHandler
> It is responsible to accept a SendRuntimeFilterRequest RPC to strip the
> actual BloomFilter from the network. It then translates this filter to the
> WorkerBee’s new interface registerRuntimeFilter.
> Another RPC type is BroadcastRuntimeFilterRequest. It will register the
> accepted global bloom filter to the WorkerBee by the registerRuntimeFilter
> method and then propagate to the FragmentContext through which the probe side
> scan node can fetch the aggregated bloom filter.
> 4.RuntimeFilterManager
> The foreman will instance a RuntimeFilterManager .It will indirectly get
> every RuntimeFilter by the WorkerBee. Once all the BloomFilters have been
> accepted and aggregated . It will broadcast the aggregated bloom filter to
> all the probe side scan nodes through the data tunnel by a
> BroadcastRuntimeFilterRequest RPC.
> 5. RuntimeFilterEnableOption
> A global option will be added to decide whether to enable this new feature.
>
> Welcome suggestion and advice from you.The related PR will be presented as
> soon as possible.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
