Github user amansinha100 commented on a diff in the pull request: https://github.com/apache/drill/pull/368#discussion_r58631107 --- Diff: exec/java-exec/src/main/java/org/apache/drill/exec/planner/fragment/SimpleParallelizerMultiPlans.java --- @@ -0,0 +1,222 @@ +/** + * 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.planner.fragment; + +import java.util.Collection; +import java.util.Iterator; +import java.util.List; + +import org.apache.drill.common.exceptions.ExecutionSetupException; +import org.apache.drill.common.util.DrillStringUtils; +import org.apache.drill.exec.ops.QueryContext; +import org.apache.drill.exec.physical.base.Exchange; +import org.apache.drill.exec.physical.base.FragmentRoot; +import org.apache.drill.exec.physical.base.PhysicalOperator; +import org.apache.drill.exec.planner.PhysicalPlanReader; +import org.apache.drill.exec.planner.fragment.Materializer.IndexedFragmentNode; +import org.apache.drill.exec.proto.BitControl.PlanFragment; +import org.apache.drill.exec.proto.BitControl.QueryContextInformation; +import org.apache.drill.exec.proto.CoordinationProtos.DrillbitEndpoint; +import org.apache.drill.exec.proto.ExecProtos.FragmentHandle; +import org.apache.drill.exec.proto.UserBitShared.QueryId; +import org.apache.drill.exec.rpc.user.UserSession; +import org.apache.drill.exec.server.options.OptionList; +import org.apache.drill.exec.work.QueryWorkUnit; +import org.apache.drill.exec.work.foreman.ForemanSetupException; + +import com.fasterxml.jackson.core.JsonProcessingException; +import com.google.common.base.Preconditions; +import com.google.common.collect.Lists; + +/** + * SimpleParallelizerMultiPlans class is an extension to SimpleParallelizer + * to help with getting PlanFragments for split plan. + * Split plan is essentially ability to create multiple physical plans from a single logical plan + * to be able to run them separately. + * Moving functionality specific to splitting the plan to this class + * allows not to pollute parent class with non-authentic functionality + * + */ +public class SimpleParallelizerMultiPlans extends SimpleParallelizer { + + public SimpleParallelizerMultiPlans(QueryContext context) { + super(context); + } + + /** + * Create multiple physical plans from original query planning, it will allow execute them eventually independently + * @param options + * @param foremanNode + * @param queryId + * @param activeEndpoints + * @param reader + * @param rootFragment + * @param session + * @param queryContextInfo + * @return + * @throws ExecutionSetupException + */ + public List<QueryWorkUnit> getSplitFragments(OptionList options, DrillbitEndpoint foremanNode, QueryId queryId, + Collection<DrillbitEndpoint> activeEndpoints, PhysicalPlanReader reader, Fragment rootFragment, + UserSession session, QueryContextInformation queryContextInfo) throws ExecutionSetupException { + + final PlanningSet planningSet = getFragmentsHelper(activeEndpoints, rootFragment); + + return generateWorkUnits( + options, foremanNode, queryId, reader, rootFragment, planningSet, session, queryContextInfo); + } + + /** + * Split plan into multiple plans based on parallelization + * Ideally it is applicable only to plans with two major fragments: Screen and UnionExchange + * But there could be cases where we can remove even multiple exchanges like in case of "order by" + * End goal is to get single major fragment: Screen with chain that ends up with a single minor fragment + * from Leaf Exchange. This way each plan can run independently without any exchange involvement + * @param options + * @param foremanNode - not really applicable + * @param queryId + * @param reader + * @param rootNode + * @param planningSet + * @param session + * @param queryContextInfo + * @return + * @throws ExecutionSetupException + */ + private List<QueryWorkUnit> generateWorkUnits(OptionList options, DrillbitEndpoint foremanNode, QueryId queryId, + PhysicalPlanReader reader, Fragment rootNode, PlanningSet planningSet, + UserSession session, QueryContextInformation queryContextInfo) throws ExecutionSetupException { + + // now we generate all the individual plan fragments and associated assignments. Note, we need all endpoints + // assigned before we can materialize, so we start a new loop here rather than utilizing the previous one. + + List<QueryWorkUnit> workUnits = Lists.newArrayList(); + int plansCount = 0; + DrillbitEndpoint[] endPoints = null; + long initialAllocation = 0; + long maxAllocation = 0; + + final Iterator<Wrapper> iter = planningSet.iterator(); + while (iter.hasNext()) { + Wrapper wrapper = iter.next(); + Fragment node = wrapper.getNode(); + boolean isLeafFragment = node.getReceivingExchangePairs().size() == 0; + final PhysicalOperator physicalOperatorRoot = node.getRoot(); + // get all the needed info from leaf fragment + if ( (physicalOperatorRoot instanceof Exchange) && isLeafFragment) { + // need to get info about + // number of minor fragments + // assignedEndPoints + // allocation + plansCount = wrapper.getWidth(); + initialAllocation = (wrapper.getInitialAllocation() != 0 ) ? wrapper.getInitialAllocation()/plansCount : 0; + maxAllocation = (wrapper.getMaxAllocation() != 0 ) ? wrapper.getMaxAllocation()/plansCount : 0; + endPoints = new DrillbitEndpoint[plansCount]; + for (int mfId = 0; mfId < plansCount; mfId++) { + endPoints[mfId] = wrapper.getAssignedEndpoint(mfId); --- End diff -- Since the end-point assignments are saved with the plan fragment which will be submitted later (maybe after several minutes or even hours depending on the client), it is possible that due to data movement (e.g rebalancing) the data locality could have changed in the meantime. Is there additional check done at the time the split plan is run ? Is that a concern ?
--- If your project is set up for it, you can reply to this email and have your reply appear on GitHub as well. If your project does not have this feature enabled and wishes so, or if the feature is enabled but not working, please contact infrastructure at infrastruct...@apache.org or file a JIRA ticket with INFRA. ---