tillrohrmann commented on a change in pull request #13644: URL: https://github.com/apache/flink/pull/13644#discussion_r509942532
########## File path: flink-kubernetes/src/main/java/org/apache/flink/kubernetes/highavailability/KubernetesLeaderElectionService.java ########## @@ -0,0 +1,219 @@ +/* + * 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.flink.kubernetes.highavailability; + +import org.apache.flink.kubernetes.kubeclient.FlinkKubeClient; +import org.apache.flink.kubernetes.kubeclient.KubernetesLeaderElectionConfiguration; +import org.apache.flink.kubernetes.kubeclient.resources.KubernetesConfigMap; +import org.apache.flink.kubernetes.kubeclient.resources.KubernetesLeaderElector; +import org.apache.flink.kubernetes.kubeclient.resources.KubernetesWatch; +import org.apache.flink.kubernetes.utils.KubernetesUtils; +import org.apache.flink.runtime.leaderelection.AbstractLeaderElectionService; +import org.apache.flink.runtime.leaderelection.LeaderContender; +import org.apache.flink.util.function.FunctionUtils; + +import java.util.List; +import java.util.Map; +import java.util.Objects; +import java.util.Optional; +import java.util.concurrent.CompletableFuture; +import java.util.concurrent.Executor; + +import static org.apache.flink.kubernetes.utils.Constants.LABEL_CONFIGMAP_TYPE_HIGH_AVAILABILITY; +import static org.apache.flink.kubernetes.utils.Constants.LEADER_ADDRESS_KEY; +import static org.apache.flink.kubernetes.utils.Constants.LEADER_SESSION_ID_KEY; +import static org.apache.flink.util.Preconditions.checkNotNull; + +/** + * Leader election service for multiple JobManagers. The active JobManager is elected using Kubernetes. + * The current leader's address as well as its leader session ID is published via Kubernetes ConfigMap. + * Note that the contending lock and leader storage are using the same ConfigMap. And every component(e.g. + * ResourceManager, Dispatcher, RestEndpoint, JobManager for each job) will have a separate ConfigMap. + */ +public class KubernetesLeaderElectionService extends AbstractLeaderElectionService { + + private final FlinkKubeClient kubeClient; + + private final Executor executor; + + private final String configMapName; + + private final KubernetesLeaderElector leaderElector; + + private KubernetesWatch kubernetesWatch; + + // Labels will be used to clean up the ha related ConfigMaps. + private Map<String, String> configMapLabels; + + KubernetesLeaderElectionService( + FlinkKubeClient kubeClient, + Executor executor, + KubernetesLeaderElectionConfiguration leaderConfig) { + + this.kubeClient = checkNotNull(kubeClient, "Kubernetes client should not be null."); + this.executor = checkNotNull(executor, "Executor should not be null."); + this.configMapName = leaderConfig.getConfigMapName(); + this.leaderElector = kubeClient.createLeaderElector(leaderConfig, new LeaderCallbackHandlerImpl()); + this.leaderContender = null; + this.configMapLabels = KubernetesUtils.getConfigMapLabels( + leaderConfig.getClusterId(), LABEL_CONFIGMAP_TYPE_HIGH_AVAILABILITY); + } + + @Override + public void internalStart(LeaderContender contender) { + CompletableFuture.runAsync(leaderElector::run, executor); + kubernetesWatch = kubeClient.watchConfigMaps(configMapName, new ConfigMapCallbackHandlerImpl()); + } + + @Override + public void internalStop() { + if (kubernetesWatch != null) { + kubernetesWatch.close(); + } + } + + @Override + protected void writeLeaderInformation() { + try { + kubeClient.checkAndUpdateConfigMap( + configMapName, + configMap -> { + if (leaderElector.hasLeadership(configMap)) { + // Get the updated ConfigMap with new leader information + if (confirmedLeaderAddress != null && confirmedLeaderSessionID != null) { + configMap.getData().put(LEADER_ADDRESS_KEY, confirmedLeaderAddress); + configMap.getData().put(LEADER_SESSION_ID_KEY, confirmedLeaderSessionID.toString()); + } + configMap.getLabels().putAll(configMapLabels); + return Optional.of(configMap); + } + return Optional.empty(); + }).get(); + } catch (Exception e) { + leaderContender.handleError(new Exception("Could not update ConfigMap " + configMapName, e)); + } + } + + @Override + protected boolean hasLeadership() { + return kubeClient.getConfigMap(configMapName) + .map(leaderElector::hasLeadership) + .orElse(false); + } + + @Override + public String toString() { + return "KubernetesLeaderElectionService{configMapName='" + configMapName + "'}"; + } + + private class LeaderCallbackHandlerImpl extends KubernetesLeaderElector.LeaderCallbackHandler { + + @Override + public void isLeader() { + onGrantLeadership(); + } + + @Override + public void notLeader() { + // Clear the leader information in ConfigMap + try { + kubeClient.checkAndUpdateConfigMap( + configMapName, + configMap -> { + // Do not need to check the leader here + configMap.getData().remove(LEADER_ADDRESS_KEY); + configMap.getData().remove(LEADER_SESSION_ID_KEY); + return Optional.of(configMap); + } + ).get(); + } catch (Exception e) { + leaderContender.handleError( + new Exception("Could not remove leader information from ConfigMap " + configMapName, e)); + } + onRevokeLeadership(); + // Continue to contend the leader + CompletableFuture.runAsync(leaderElector::run, executor); + } + } + + private class ConfigMapCallbackHandlerImpl implements FlinkKubeClient.WatchCallbackHandler<KubernetesConfigMap> { + + @Override + public void onAdded(List<KubernetesConfigMap> configMaps) { + // noop + } + + @Override + public void onModified(List<KubernetesConfigMap> configMaps) { + if (hasLeadership()) { + configMaps.forEach(configMap -> { + if (isLeaderUpdatedExternally(configMap)) { + if (configMap.getName().equals(configMapName)) { + // the data field does not correspond to the expected leader information + if (logger.isDebugEnabled()) { + logger.debug("Correcting leader information in {} by {}.", + configMapName, leaderContender.getDescription()); + } + writeLeaderInformation(); + } else { + logger.warn("Ignoring the modified event since it does not belong to {}.", this); + } + } + }); + } + } + + @Override + public void onDeleted(List<KubernetesConfigMap> configMaps) { Review comment: Wouldn't this be covered by the `onModified` callback which the leader will receive? The leader should then try to re-create the lost information. This actually brings me to another question: What happens if the configMap is being deleted by the user and at the same time the current leader dies for some reason? Who would then re-create the configMap? And even if we are able to recreate the configMap, how would we be able to recover the old state (e.g. all the submitted jobs)? ---------------------------------------------------------------- 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: us...@infra.apache.org
