xyuanlu commented on a change in pull request #1747: URL: https://github.com/apache/helix/pull/1747#discussion_r644364646
########## File path: helix-rest/src/main/java/org/apache/helix/rest/server/resources/helix/ResourceAssignmentOptimizer.java ########## @@ -0,0 +1,347 @@ +package org.apache.helix.rest.server.resources.helix; + +/* + * 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. + */ + +import java.security.InvalidParameterException; +import java.util.ArrayList; +import java.util.Arrays; +import java.util.HashMap; +import java.util.HashSet; +import java.util.Iterator; +import java.util.List; +import java.util.Map; +import java.util.Set; +import java.util.TreeMap; +import java.util.stream.Collectors; +import javax.ws.rs.POST; +import javax.ws.rs.Path; +import javax.ws.rs.PathParam; +import javax.ws.rs.core.Response; + +import com.codahale.metrics.annotation.ResponseMetered; +import com.codahale.metrics.annotation.Timed; +import com.fasterxml.jackson.core.JsonProcessingException; +import com.fasterxml.jackson.core.type.TypeReference; +import org.apache.helix.ConfigAccessor; +import org.apache.helix.HelixDataAccessor; +import org.apache.helix.controller.rebalancer.strategy.AutoRebalanceStrategy; +import org.apache.helix.controller.rebalancer.strategy.RebalanceStrategy; +import org.apache.helix.controller.rebalancer.waged.WagedRebalancer; +import org.apache.helix.model.ClusterConfig; +import org.apache.helix.model.IdealState; +import org.apache.helix.model.InstanceConfig; +import org.apache.helix.model.ResourceAssignment; +import org.apache.helix.model.ResourceConfig; +import org.apache.helix.model.StateModelDefinition; +import org.apache.helix.rest.common.HttpConstants; +import org.slf4j.Logger; +import org.slf4j.LoggerFactory; + +import static org.apache.helix.util.HelixUtil.computeIdealMapping; +import static org.apache.helix.util.HelixUtil.getIdealAssignmentForFullAuto; +import static org.apache.helix.util.HelixUtil.getTargetAssignmentForWagedFullAuto; + + +@Path("/clusters/{clusterId}/partitionAssignment") +public class ResourceAssignmentOptimizer extends AbstractHelixResource { + private static Logger LOG = LoggerFactory + .getLogger(org.apache.helix.rest.server.resources.helix.ClusterAccessor.class.getName()); + + private static class InputFields { + Set<String> resourceFilter = new HashSet<>(); + Map<String, String> nodeSwap = new HashMap<>(); // old instance -> new instance. + List<String> newInstances = new ArrayList<>(); + List<String> instancesToRemove = new ArrayList<>(); + Set<String> instanceFilter = new HashSet<>(); + } + + private static class ClusterState { + List<InstanceConfig> instanceConfigs = new ArrayList<>(); + ClusterConfig clusterConfig; + List<String> resources = new ArrayList<>(); + List<String> instances; // cluster LiveInstance + addInstances - instancesToRemove. + } + + // Result format: Map of resource -> partition -> instance -> state. + private static class AssignmentResult extends HashMap<String, Map<String, Map<String, String>>> { + public AssignmentResult() { + super(); + } + } + + @ResponseMetered(name = HttpConstants.WRITE_REQUEST) + @Timed(name = HttpConstants.WRITE_REQUEST) + @POST + public Response computePotentialAssignment(@PathParam("clusterId") String clusterId, + String content) { + + InputFields inputFields; + ClusterState clusterState; + AssignmentResult result; + + try { + // 1. Try to parse the content string. If parseable, use it as a KV mapping. Otherwise, return a REASON String + inputFields = readInput(content); + // 2. Read cluster status from ZK. + clusterState = readClusterStateAndValidateInput(clusterId, inputFields); + // 3. Call rebalancer tools for each resource. + result = computeOptimalAssignmentForResources(inputFields, clusterState, clusterId); + // 4. Serialize result to JSON and return. + return JSONRepresentation(result); + } catch (InvalidParameterException ex) { + return badRequest(ex.getMessage()); + } catch (JsonProcessingException e) { + return badRequest("Invalid input: Input can but be parsed into a KV map." + e.getMessage()); + } catch (OutOfMemoryError e) { + LOG.error( + "OutOfMemoryError while calling AssignmentResult" + Arrays.toString(e.getStackTrace())); + return badRequest( + "Response size is too large to serialize. Please query by resources or instance filter"); + } catch (Exception e) { + LOG.error("Failed to compute partition assignment:" + Arrays.toString(e.getStackTrace())); + return badRequest("Failed to compute partition assignment: " + e); + } + } + + private InputFields readInput(String content) + throws InvalidParameterException, JsonProcessingException { + + Map<String, Map<String, Object>> customFieldsMap; + InputFields inputMap = new InputFields(); + customFieldsMap = + OBJECT_MAPPER.readValue(content, new TypeReference<HashMap<String, Map<String, Object>>>() { + }); + + // Content is given as a KV mapping. + for (Map.Entry<String, Map<String, Object>> entry : customFieldsMap.entrySet()) { + String key = entry.getKey(); + switch (key) { + case "InstanceChange": + for (Map.Entry<String, Object> instanceChange : entry.getValue().entrySet()) { + String instanceChangeKey = instanceChange.getKey(); + if (instanceChangeKey.equals("AddInstances") && (instanceChange + .getValue() instanceof List)) { + inputMap.newInstances.addAll((List<String>) instanceChange.getValue()); + } else if (instanceChangeKey.equals("RemoveInstances") && (instanceChange + .getValue() instanceof List)) { + inputMap.instancesToRemove.addAll((List<String>) instanceChange.getValue()); + } else if (instanceChangeKey.equals("SwapInstances") && (instanceChange + .getValue() instanceof Map)) { + for (Map.Entry<String, String> swapPair : ((Map<String, String>) instanceChange + .getValue()).entrySet()) { + inputMap.nodeSwap.put(swapPair.getKey(), swapPair.getValue()); + } + } else { + throw new InvalidParameterException( + "Unsupported command or invalid format for InstanceChange : " + + instanceChangeKey); + } + } + break; + case "Options": + for (Map.Entry<String, Object> option : entry.getValue().entrySet()) { + String optionKey = option.getKey(); + if (optionKey.equals("ResourceFilter") && (option.getValue() instanceof List)) { + inputMap.resourceFilter.addAll((List<String>) option.getValue()); + } else if (optionKey.equals("InstanceFilter") && (option.getValue() instanceof List)) { + inputMap.instanceFilter.addAll((List<String>) option.getValue()); + } else { + throw new InvalidParameterException( + "Unsupported command for invalid format for Options : " + option); + } + } + break; + default: + throw new InvalidParameterException( + "Unsupported command for partitionAssignment : " + key); + } + } + return inputMap; + } + + private ClusterState readClusterStateAndValidateInput(String clusterId, InputFields inputFields) + throws InvalidParameterException { + + ClusterState clusterState = new ClusterState(); + ConfigAccessor cfgAccessor = getConfigAccessor(); + HelixDataAccessor dataAccessor = getDataAccssor(clusterId); + clusterState.resources = dataAccessor.getChildNames(dataAccessor.keyBuilder().idealStates()); + // Add existing live instance and new instance from user input to instances list + clusterState.instances = dataAccessor.getChildNames(dataAccessor.keyBuilder().liveInstances()); + clusterState.instances.addAll(inputFields.newInstances); + + // Check if to be removed instances and old instances in swap node exist in live instance + if (!inputFields.nodeSwap.isEmpty() || !inputFields.instancesToRemove.isEmpty()) { + Set<String> liveInstanceSet = new HashSet<>(clusterState.instances); + for (Map.Entry<String, String> nodeSwapPair : inputFields.nodeSwap.entrySet()) { + if (!liveInstanceSet.contains(nodeSwapPair.getKey())) { + throw new InvalidParameterException("Invalid input: instance [" + nodeSwapPair.getKey() + + "] in SwapInstances does not exist in cluster."); + } + } + for (String instanceToRemove : inputFields.instancesToRemove) { + if (!liveInstanceSet.contains(instanceToRemove)) { + throw new InvalidParameterException("Invalid input: instance [" + instanceToRemove + + "] in RemoveInstances does not exist in cluster."); + } + } + if (!inputFields.instancesToRemove.isEmpty()) { + clusterState.instances.removeIf(inputFields.instancesToRemove::contains); + } + } + + // Read instance and cluster config. + // It will throw exception if there is no instanceConfig for newly added instance. + for (String instance : clusterState.instances) { + InstanceConfig config = cfgAccessor.getInstanceConfig(clusterId, instance); + clusterState.instanceConfigs.add(config); + } + clusterState.clusterConfig = cfgAccessor.getClusterConfig(clusterId); + return clusterState; + } + + private AssignmentResult computeOptimalAssignmentForResources(InputFields inputFields, + ClusterState clusterState, String clusterId) throws Exception { + + AssignmentResult result = new AssignmentResult(); + // Iterate through resources, read resource level info and get potential assignment. + HelixDataAccessor dataAccessor = getDataAccssor(clusterId); + List<IdealState> wagedResourceIdealState = new ArrayList<>(); + + for (String resource : clusterState.resources) { + IdealState idealState = + dataAccessor.getProperty(dataAccessor.keyBuilder().idealStates(resource)); + // Compute all Waged resources in a batch later. + if (idealState.getRebalancerClassName() != null && idealState.getRebalancerClassName() + .equals(WagedRebalancer.class.getName())) { + wagedResourceIdealState.add(idealState); + continue; + } + // For non Waged resources, we don't compute resources not in white list. + if (!inputFields.resourceFilter.isEmpty() && !inputFields.resourceFilter.contains(resource)) { + continue; + } + // Use getIdealAssignmentForFullAuto for FULL_AUTO resource. + if (idealState.getRebalanceMode() == IdealState.RebalanceMode.FULL_AUTO) { + String rebalanceStrategy = idealState.getRebalanceStrategy(); + if (rebalanceStrategy == null || rebalanceStrategy + .equalsIgnoreCase(RebalanceStrategy.DEFAULT_REBALANCE_STRATEGY)) { + rebalanceStrategy = AutoRebalanceStrategy.class.getName(); + } + Map<String, Map<String, String>> partitionAssignments = new TreeMap<>( + getIdealAssignmentForFullAuto(clusterState.clusterConfig, clusterState.instanceConfigs, + clusterState.instances, idealState, new ArrayList<>(idealState.getPartitionSet()), + rebalanceStrategy)); + instanceSwapAndFilter(inputFields, partitionAssignments, resource, result); Review comment: `instanceSwapAndFilter` is called for each resources. For non Waged resources, we compute assignment one by one and filter result one by one. Waged resources need to be computed in a batch and filter result one by one. I am not sure if these three can be merged. -- 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. For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
