Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 1 | package org.onosproject.store.consistent.impl; |
| 2 | |
Ayaka Koshibe | c19b8b8 | 2015-04-08 15:18:24 -0700 | [diff] [blame] | 3 | import com.google.common.collect.ImmutableList; |
Jonathan Hart | e649c75 | 2015-03-03 18:04:25 -0800 | [diff] [blame] | 4 | import com.google.common.collect.ImmutableMap; |
Madan Jampani | 3650fc4 | 2015-05-27 17:01:50 -0700 | [diff] [blame] | 5 | import com.google.common.collect.Iterables; |
Ayaka Koshibe | c19b8b8 | 2015-04-08 15:18:24 -0700 | [diff] [blame] | 6 | import com.google.common.collect.Lists; |
Jonathan Hart | e649c75 | 2015-03-03 18:04:25 -0800 | [diff] [blame] | 7 | import com.google.common.collect.Maps; |
| 8 | import com.google.common.collect.Sets; |
Ayaka Koshibe | c19b8b8 | 2015-04-08 15:18:24 -0700 | [diff] [blame] | 9 | |
Madan Jampani | 8d37efc | 2015-06-01 17:25:48 -0700 | [diff] [blame] | 10 | import org.apache.commons.lang.math.RandomUtils; |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 11 | import org.apache.felix.scr.annotations.Activate; |
| 12 | import org.apache.felix.scr.annotations.Component; |
| 13 | import org.apache.felix.scr.annotations.Deactivate; |
| 14 | import org.apache.felix.scr.annotations.Reference; |
| 15 | import org.apache.felix.scr.annotations.ReferenceCardinality; |
| 16 | import org.apache.felix.scr.annotations.Service; |
| 17 | import org.onlab.util.KryoNamespace; |
Madan Jampani | cc58675 | 2015-05-18 16:19:27 -0700 | [diff] [blame] | 18 | import org.onosproject.cluster.ClusterEvent; |
| 19 | import org.onosproject.cluster.ClusterEvent.Type; |
| 20 | import org.onosproject.cluster.ClusterEventListener; |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 21 | import org.onosproject.cluster.ClusterService; |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 22 | import org.onosproject.cluster.Leadership; |
| 23 | import org.onosproject.cluster.LeadershipEvent; |
| 24 | import org.onosproject.cluster.LeadershipEventListener; |
| 25 | import org.onosproject.cluster.LeadershipService; |
| 26 | import org.onosproject.cluster.NodeId; |
Simon Hunt | ff66374 | 2015-05-14 13:33:05 -0700 | [diff] [blame] | 27 | import org.onosproject.event.ListenerRegistry; |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 28 | import org.onosproject.event.EventDeliveryService; |
| 29 | import org.onosproject.store.cluster.messaging.ClusterCommunicationService; |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 30 | import org.onosproject.store.cluster.messaging.MessageSubject; |
| 31 | import org.onosproject.store.serializers.KryoNamespaces; |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 32 | import org.onosproject.store.service.ConsistentMap; |
Ayaka Koshibe | c19b8b8 | 2015-04-08 15:18:24 -0700 | [diff] [blame] | 33 | import org.onosproject.store.service.ConsistentMapException; |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 34 | import org.onosproject.store.service.Serializer; |
| 35 | import org.onosproject.store.service.StorageService; |
| 36 | import org.onosproject.store.service.Versioned; |
| 37 | import org.slf4j.Logger; |
| 38 | |
Madan Jampani | 1af8e13 | 2015-04-30 16:41:18 -0700 | [diff] [blame] | 39 | import java.util.ArrayList; |
Madan Jampani | cc58675 | 2015-05-18 16:19:27 -0700 | [diff] [blame] | 40 | import java.util.Collections; |
Jonathan Hart | e649c75 | 2015-03-03 18:04:25 -0800 | [diff] [blame] | 41 | import java.util.Map; |
| 42 | import java.util.Map.Entry; |
| 43 | import java.util.Objects; |
| 44 | import java.util.Set; |
Ayaka Koshibe | c19b8b8 | 2015-04-08 15:18:24 -0700 | [diff] [blame] | 45 | import java.util.List; |
Madan Jampani | de003d9 | 2015-05-11 17:14:20 -0700 | [diff] [blame] | 46 | import java.util.concurrent.CancellationException; |
| 47 | import java.util.concurrent.CompletableFuture; |
Jonathan Hart | e649c75 | 2015-03-03 18:04:25 -0800 | [diff] [blame] | 48 | import java.util.concurrent.ExecutorService; |
| 49 | import java.util.concurrent.Executors; |
| 50 | import java.util.concurrent.ScheduledExecutorService; |
| 51 | import java.util.concurrent.TimeUnit; |
Madan Jampani | d46e18f | 2015-05-04 23:19:33 -0700 | [diff] [blame] | 52 | import java.util.concurrent.atomic.AtomicBoolean; |
Jonathan Hart | e649c75 | 2015-03-03 18:04:25 -0800 | [diff] [blame] | 53 | import java.util.stream.Collectors; |
| 54 | |
| 55 | import static com.google.common.base.Preconditions.checkArgument; |
| 56 | import static org.onlab.util.Tools.groupedThreads; |
| 57 | import static org.slf4j.LoggerFactory.getLogger; |
Ayaka Koshibe | 4a3c239 | 2015-04-22 18:03:13 -0700 | [diff] [blame] | 58 | import static org.onosproject.cluster.ControllerNode.State.ACTIVE; |
| 59 | import static org.onosproject.cluster.ControllerNode.State.INACTIVE; |
| 60 | |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 61 | /** |
| 62 | * Distributed Lock Manager implemented on top of ConsistentMap. |
| 63 | * <p> |
Madan Jampani | cc58675 | 2015-05-18 16:19:27 -0700 | [diff] [blame] | 64 | * This implementation makes use of ClusterService's failure |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 65 | * detection capabilities to detect and purge stale locks. |
| 66 | * TODO: Ensure lock safety and liveness. |
| 67 | */ |
Madan Jampani | 7f72c3f | 2015-03-01 17:34:59 -0800 | [diff] [blame] | 68 | @Component(immediate = true, enabled = true) |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 69 | @Service |
| 70 | public class DistributedLeadershipManager implements LeadershipService { |
| 71 | |
| 72 | @Reference(cardinality = ReferenceCardinality.MANDATORY_UNARY) |
| 73 | protected StorageService storageService; |
| 74 | |
| 75 | @Reference(cardinality = ReferenceCardinality.MANDATORY_UNARY) |
| 76 | protected ClusterService clusterService; |
| 77 | |
| 78 | @Reference(cardinality = ReferenceCardinality.MANDATORY_UNARY) |
| 79 | protected ClusterCommunicationService clusterCommunicator; |
| 80 | |
| 81 | @Reference(cardinality = ReferenceCardinality.MANDATORY_UNARY) |
| 82 | protected EventDeliveryService eventDispatcher; |
| 83 | |
| 84 | private static final MessageSubject LEADERSHIP_EVENT_MESSAGE_SUBJECT = |
| 85 | new MessageSubject("distributed-leadership-manager-events"); |
| 86 | |
| 87 | private final Logger log = getLogger(getClass()); |
| 88 | private ExecutorService messageHandlingExecutor; |
Madan Jampani | 7d243db | 2015-05-27 13:16:02 -0700 | [diff] [blame] | 89 | private ScheduledExecutorService electionRunner; |
| 90 | private ScheduledExecutorService lockExecutor; |
Madan Jampani | cc58675 | 2015-05-18 16:19:27 -0700 | [diff] [blame] | 91 | private ScheduledExecutorService staleLeadershipPurgeExecutor; |
Madan Jampani | 08960a4 | 2015-06-04 09:59:29 -0700 | [diff] [blame] | 92 | private ScheduledExecutorService leadershipRefresher; |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 93 | |
Ayaka Koshibe | c19b8b8 | 2015-04-08 15:18:24 -0700 | [diff] [blame] | 94 | private ConsistentMap<String, NodeId> leaderMap; |
| 95 | private ConsistentMap<String, List<NodeId>> candidateMap; |
| 96 | |
Madan Jampani | cc58675 | 2015-05-18 16:19:27 -0700 | [diff] [blame] | 97 | private ListenerRegistry<LeadershipEvent, LeadershipEventListener> listenerRegistry; |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 98 | private final Map<String, Leadership> leaderBoard = Maps.newConcurrentMap(); |
Ayaka Koshibe | fd26a30 | 2015-04-13 13:59:54 -0700 | [diff] [blame] | 99 | private final Map<String, Leadership> candidateBoard = Maps.newConcurrentMap(); |
Madan Jampani | cc58675 | 2015-05-18 16:19:27 -0700 | [diff] [blame] | 100 | private final ClusterEventListener clusterEventListener = new InternalClusterEventListener(); |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 101 | |
Madan Jampani | cc58675 | 2015-05-18 16:19:27 -0700 | [diff] [blame] | 102 | private NodeId localNodeId; |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 103 | private Set<String> activeTopics = Sets.newConcurrentHashSet(); |
Madan Jampani | 7d243db | 2015-05-27 13:16:02 -0700 | [diff] [blame] | 104 | private Map<String, CompletableFuture<Leadership>> pendingFutures = Maps.newConcurrentMap(); |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 105 | |
Madan Jampani | 8d37efc | 2015-06-01 17:25:48 -0700 | [diff] [blame] | 106 | // The actual delay is randomly chosen between the interval [0, WAIT_BEFORE_RETRY_MILLIS) |
| 107 | private static final int WAIT_BEFORE_RETRY_MILLIS = 150; |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 108 | private static final int DELAY_BETWEEN_LEADER_LOCK_ATTEMPTS_SEC = 2; |
Madan Jampani | 08960a4 | 2015-06-04 09:59:29 -0700 | [diff] [blame] | 109 | private static final int LEADERSHIP_REFRESH_INTERVAL_SEC = 2; |
Madan Jampani | cc58675 | 2015-05-18 16:19:27 -0700 | [diff] [blame] | 110 | private static final int DELAY_BETWEEN_STALE_LEADERSHIP_PURGE_ATTEMPTS_SEC = 2; |
Ayaka Koshibe | c19b8b8 | 2015-04-08 15:18:24 -0700 | [diff] [blame] | 111 | |
Madan Jampani | cc58675 | 2015-05-18 16:19:27 -0700 | [diff] [blame] | 112 | private final AtomicBoolean staleLeadershipPurgeScheduled = new AtomicBoolean(false); |
| 113 | |
Ayaka Koshibe | c19b8b8 | 2015-04-08 15:18:24 -0700 | [diff] [blame] | 114 | private static final Serializer SERIALIZER = Serializer.using( |
| 115 | new KryoNamespace.Builder().register(KryoNamespaces.API).build()); |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 116 | |
| 117 | @Activate |
| 118 | public void activate() { |
Ayaka Koshibe | c19b8b8 | 2015-04-08 15:18:24 -0700 | [diff] [blame] | 119 | leaderMap = storageService.<String, NodeId>consistentMapBuilder() |
| 120 | .withName("onos-topic-leaders") |
| 121 | .withSerializer(SERIALIZER) |
| 122 | .withPartitionsDisabled().build(); |
| 123 | candidateMap = storageService.<String, List<NodeId>>consistentMapBuilder() |
| 124 | .withName("onos-topic-candidates") |
| 125 | .withSerializer(SERIALIZER) |
| 126 | .withPartitionsDisabled().build(); |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 127 | |
| 128 | localNodeId = clusterService.getLocalNode().id(); |
| 129 | |
| 130 | messageHandlingExecutor = Executors.newSingleThreadExecutor( |
| 131 | groupedThreads("onos/store/leadership", "message-handler")); |
Madan Jampani | 7d243db | 2015-05-27 13:16:02 -0700 | [diff] [blame] | 132 | electionRunner = Executors.newSingleThreadScheduledExecutor( |
| 133 | groupedThreads("onos/store/leadership", "election-runner")); |
| 134 | lockExecutor = Executors.newScheduledThreadPool( |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 135 | 4, groupedThreads("onos/store/leadership", "election-thread-%d")); |
Madan Jampani | cc58675 | 2015-05-18 16:19:27 -0700 | [diff] [blame] | 136 | staleLeadershipPurgeExecutor = Executors.newSingleThreadScheduledExecutor( |
| 137 | groupedThreads("onos/store/leadership", "stale-leadership-evictor")); |
Madan Jampani | 08960a4 | 2015-06-04 09:59:29 -0700 | [diff] [blame] | 138 | leadershipRefresher = Executors.newSingleThreadScheduledExecutor( |
| 139 | groupedThreads("onos/store/leadership", "refresh-thread")); |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 140 | clusterCommunicator.addSubscriber( |
| 141 | LEADERSHIP_EVENT_MESSAGE_SUBJECT, |
Madan Jampani | d46e18f | 2015-05-04 23:19:33 -0700 | [diff] [blame] | 142 | SERIALIZER::decode, |
| 143 | this::onLeadershipEvent, |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 144 | messageHandlingExecutor); |
| 145 | |
Madan Jampani | cc58675 | 2015-05-18 16:19:27 -0700 | [diff] [blame] | 146 | clusterService.addListener(clusterEventListener); |
| 147 | |
Madan Jampani | 7d243db | 2015-05-27 13:16:02 -0700 | [diff] [blame] | 148 | electionRunner.scheduleWithFixedDelay( |
| 149 | this::electLeaders, 0, DELAY_BETWEEN_LEADER_LOCK_ATTEMPTS_SEC, TimeUnit.SECONDS); |
| 150 | |
Madan Jampani | 08960a4 | 2015-06-04 09:59:29 -0700 | [diff] [blame] | 151 | leadershipRefresher.scheduleWithFixedDelay( |
| 152 | this::refreshLeaderBoard, 0, LEADERSHIP_REFRESH_INTERVAL_SEC, TimeUnit.SECONDS); |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 153 | |
Simon Hunt | ff66374 | 2015-05-14 13:33:05 -0700 | [diff] [blame] | 154 | listenerRegistry = new ListenerRegistry<>(); |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 155 | eventDispatcher.addSink(LeadershipEvent.class, listenerRegistry); |
| 156 | |
Madan Jampani | d46e18f | 2015-05-04 23:19:33 -0700 | [diff] [blame] | 157 | log.info("Started"); |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 158 | } |
| 159 | |
| 160 | @Deactivate |
| 161 | public void deactivate() { |
| 162 | leaderBoard.forEach((topic, leadership) -> { |
| 163 | if (localNodeId.equals(leadership.leader())) { |
| 164 | withdraw(topic); |
| 165 | } |
| 166 | }); |
| 167 | |
Madan Jampani | cc58675 | 2015-05-18 16:19:27 -0700 | [diff] [blame] | 168 | clusterService.removeListener(clusterEventListener); |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 169 | eventDispatcher.removeSink(LeadershipEvent.class); |
| 170 | clusterCommunicator.removeSubscriber(LEADERSHIP_EVENT_MESSAGE_SUBJECT); |
| 171 | |
Madan Jampani | 7d243db | 2015-05-27 13:16:02 -0700 | [diff] [blame] | 172 | electionRunner.shutdown(); |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 173 | messageHandlingExecutor.shutdown(); |
Madan Jampani | 7d243db | 2015-05-27 13:16:02 -0700 | [diff] [blame] | 174 | lockExecutor.shutdown(); |
Madan Jampani | cc58675 | 2015-05-18 16:19:27 -0700 | [diff] [blame] | 175 | staleLeadershipPurgeExecutor.shutdown(); |
Madan Jampani | 08960a4 | 2015-06-04 09:59:29 -0700 | [diff] [blame] | 176 | leadershipRefresher.shutdown(); |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 177 | |
Madan Jampani | d46e18f | 2015-05-04 23:19:33 -0700 | [diff] [blame] | 178 | log.info("Stopped"); |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 179 | } |
| 180 | |
| 181 | @Override |
| 182 | public Map<String, Leadership> getLeaderBoard() { |
| 183 | return ImmutableMap.copyOf(leaderBoard); |
| 184 | } |
| 185 | |
| 186 | @Override |
Madan Jampani | fd45d5e | 2015-04-20 13:33:21 -0700 | [diff] [blame] | 187 | public Map<String, List<NodeId>> getCandidates() { |
| 188 | return Maps.toMap(candidateBoard.keySet(), this::getCandidates); |
Ayaka Koshibe | c19b8b8 | 2015-04-08 15:18:24 -0700 | [diff] [blame] | 189 | } |
| 190 | |
| 191 | @Override |
| 192 | public List<NodeId> getCandidates(String path) { |
Ayaka Koshibe | fd26a30 | 2015-04-13 13:59:54 -0700 | [diff] [blame] | 193 | Leadership current = candidateBoard.get(path); |
| 194 | return current == null ? ImmutableList.of() : ImmutableList.copyOf(current.candidates()); |
Ayaka Koshibe | c19b8b8 | 2015-04-08 15:18:24 -0700 | [diff] [blame] | 195 | } |
| 196 | |
| 197 | @Override |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 198 | public NodeId getLeader(String path) { |
| 199 | Leadership leadership = leaderBoard.get(path); |
| 200 | return leadership != null ? leadership.leader() : null; |
| 201 | } |
| 202 | |
| 203 | @Override |
Madan Jampani | 5961051 | 2015-02-25 15:25:43 -0800 | [diff] [blame] | 204 | public Leadership getLeadership(String path) { |
| 205 | checkArgument(path != null); |
| 206 | return leaderBoard.get(path); |
| 207 | } |
| 208 | |
| 209 | @Override |
| 210 | public Set<String> ownedTopics(NodeId nodeId) { |
| 211 | checkArgument(nodeId != null); |
| 212 | return leaderBoard.entrySet() |
| 213 | .stream() |
| 214 | .filter(entry -> nodeId.equals(entry.getValue().leader())) |
| 215 | .map(Entry::getKey) |
| 216 | .collect(Collectors.toSet()); |
| 217 | } |
| 218 | |
| 219 | @Override |
Madan Jampani | de003d9 | 2015-05-11 17:14:20 -0700 | [diff] [blame] | 220 | public CompletableFuture<Leadership> runForLeadership(String path) { |
Madan Jampani | 52860be | 2015-02-27 12:52:37 -0800 | [diff] [blame] | 221 | log.debug("Running for leadership for topic: {}", path); |
Madan Jampani | de003d9 | 2015-05-11 17:14:20 -0700 | [diff] [blame] | 222 | CompletableFuture<Leadership> resultFuture = new CompletableFuture<>(); |
| 223 | doRunForLeadership(path, resultFuture); |
| 224 | return resultFuture; |
| 225 | } |
| 226 | |
| 227 | private void doRunForLeadership(String path, CompletableFuture<Leadership> future) { |
Ayaka Koshibe | c19b8b8 | 2015-04-08 15:18:24 -0700 | [diff] [blame] | 228 | try { |
Madan Jampani | 346d4f5 | 2015-05-04 11:09:39 -0700 | [diff] [blame] | 229 | Versioned<List<NodeId>> candidates = candidateMap.computeIf(path, |
| 230 | currentList -> currentList == null || !currentList.contains(localNodeId), |
| 231 | (topic, currentList) -> { |
| 232 | if (currentList == null) { |
| 233 | return ImmutableList.of(localNodeId); |
| 234 | } else { |
| 235 | List<NodeId> newList = Lists.newLinkedList(); |
| 236 | newList.addAll(currentList); |
| 237 | newList.add(localNodeId); |
| 238 | return newList; |
| 239 | } |
| 240 | }); |
| 241 | publish(new LeadershipEvent( |
| 242 | LeadershipEvent.Type.CANDIDATES_CHANGED, |
| 243 | new Leadership(path, |
| 244 | candidates.value(), |
| 245 | candidates.version(), |
| 246 | candidates.creationTime()))); |
Ayaka Koshibe | c19b8b8 | 2015-04-08 15:18:24 -0700 | [diff] [blame] | 247 | log.debug("In the leadership race for topic {} with candidates {}", path, candidates); |
| 248 | activeTopics.add(path); |
Madan Jampani | 7d243db | 2015-05-27 13:16:02 -0700 | [diff] [blame] | 249 | Leadership leadership = electLeader(path, candidates.value()); |
| 250 | if (leadership == null) { |
| 251 | pendingFutures.put(path, future); |
| 252 | } else { |
| 253 | future.complete(leadership); |
| 254 | } |
Ayaka Koshibe | c19b8b8 | 2015-04-08 15:18:24 -0700 | [diff] [blame] | 255 | } catch (ConsistentMapException e) { |
| 256 | log.debug("Failed to enter topic leader race for {}. Retrying.", path, e); |
Madan Jampani | de003d9 | 2015-05-11 17:14:20 -0700 | [diff] [blame] | 257 | rerunForLeadership(path, future); |
Ayaka Koshibe | c19b8b8 | 2015-04-08 15:18:24 -0700 | [diff] [blame] | 258 | } |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 259 | } |
| 260 | |
| 261 | @Override |
Madan Jampani | de003d9 | 2015-05-11 17:14:20 -0700 | [diff] [blame] | 262 | public CompletableFuture<Void> withdraw(String path) { |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 263 | activeTopics.remove(path); |
Madan Jampani | de003d9 | 2015-05-11 17:14:20 -0700 | [diff] [blame] | 264 | CompletableFuture<Void> resultFuture = new CompletableFuture<>(); |
| 265 | doWithdraw(path, resultFuture); |
| 266 | return resultFuture; |
| 267 | } |
Ayaka Koshibe | c19b8b8 | 2015-04-08 15:18:24 -0700 | [diff] [blame] | 268 | |
Madan Jampani | de003d9 | 2015-05-11 17:14:20 -0700 | [diff] [blame] | 269 | |
| 270 | private void doWithdraw(String path, CompletableFuture<Void> future) { |
| 271 | if (activeTopics.contains(path)) { |
| 272 | future.completeExceptionally(new CancellationException(String.format("%s is now a active topic", path))); |
| 273 | } |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 274 | try { |
Ayaka Koshibe | c19b8b8 | 2015-04-08 15:18:24 -0700 | [diff] [blame] | 275 | Versioned<NodeId> leader = leaderMap.get(path); |
| 276 | if (leader != null && Objects.equals(leader.value(), localNodeId)) { |
| 277 | if (leaderMap.remove(path, leader.version())) { |
Madan Jampani | eedfe54 | 2015-05-26 09:54:09 -0700 | [diff] [blame] | 278 | log.debug("Gave up leadership for {}", path); |
Madan Jampani | de003d9 | 2015-05-11 17:14:20 -0700 | [diff] [blame] | 279 | future.complete(null); |
Madan Jampani | d46e18f | 2015-05-04 23:19:33 -0700 | [diff] [blame] | 280 | publish(new LeadershipEvent( |
| 281 | LeadershipEvent.Type.LEADER_BOOTED, |
| 282 | new Leadership(path, |
| 283 | localNodeId, |
| 284 | leader.version(), |
| 285 | leader.creationTime()))); |
Jonathan Hart | e649c75 | 2015-03-03 18:04:25 -0800 | [diff] [blame] | 286 | } |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 287 | } |
Ayaka Koshibe | c19b8b8 | 2015-04-08 15:18:24 -0700 | [diff] [blame] | 288 | // else we are not the current leader, can still be a candidate. |
| 289 | Versioned<List<NodeId>> candidates = candidateMap.get(path); |
| 290 | List<NodeId> candidateList = candidates != null |
| 291 | ? Lists.newArrayList(candidates.value()) |
| 292 | : Lists.newArrayList(); |
| 293 | if (!candidateList.remove(localNodeId)) { |
Madan Jampani | de003d9 | 2015-05-11 17:14:20 -0700 | [diff] [blame] | 294 | future.complete(null); |
Ayaka Koshibe | c19b8b8 | 2015-04-08 15:18:24 -0700 | [diff] [blame] | 295 | return; |
| 296 | } |
Ayaka Koshibe | 93451b4 | 2015-04-21 15:12:03 -0700 | [diff] [blame] | 297 | if (candidateMap.replace(path, candidates.version(), candidateList)) { |
Ayaka Koshibe | 67cf7de | 2015-04-17 11:47:27 -0700 | [diff] [blame] | 298 | Versioned<List<NodeId>> newCandidates = candidateMap.get(path); |
Madan Jampani | de003d9 | 2015-05-11 17:14:20 -0700 | [diff] [blame] | 299 | future.complete(null); |
Madan Jampani | d46e18f | 2015-05-04 23:19:33 -0700 | [diff] [blame] | 300 | publish(new LeadershipEvent( |
| 301 | LeadershipEvent.Type.CANDIDATES_CHANGED, |
| 302 | new Leadership(path, |
| 303 | newCandidates.value(), |
| 304 | newCandidates.version(), |
| 305 | newCandidates.creationTime()))); |
Ayaka Koshibe | fd26a30 | 2015-04-13 13:59:54 -0700 | [diff] [blame] | 306 | } else { |
Madan Jampani | 8d37efc | 2015-06-01 17:25:48 -0700 | [diff] [blame] | 307 | log.debug("Failed to withdraw from candidates list for {}. Will retry", path); |
Madan Jampani | de003d9 | 2015-05-11 17:14:20 -0700 | [diff] [blame] | 308 | retryWithdraw(path, future); |
Ayaka Koshibe | c19b8b8 | 2015-04-08 15:18:24 -0700 | [diff] [blame] | 309 | } |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 310 | } catch (Exception e) { |
Madan Jampani | a14047d | 2015-02-25 12:23:02 -0800 | [diff] [blame] | 311 | log.debug("Failed to verify (and clear) any lock this node might be holding for {}", path, e); |
Madan Jampani | de003d9 | 2015-05-11 17:14:20 -0700 | [diff] [blame] | 312 | retryWithdraw(path, future); |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 313 | } |
| 314 | } |
| 315 | |
| 316 | @Override |
Madan Jampani | 1af8e13 | 2015-04-30 16:41:18 -0700 | [diff] [blame] | 317 | public boolean stepdown(String path) { |
Madan Jampani | 9bd1f15 | 2015-04-30 23:33:35 -0700 | [diff] [blame] | 318 | if (!activeTopics.contains(path) || !Objects.equals(localNodeId, getLeader(path))) { |
Madan Jampani | 1af8e13 | 2015-04-30 16:41:18 -0700 | [diff] [blame] | 319 | return false; |
| 320 | } |
| 321 | |
| 322 | try { |
| 323 | Versioned<NodeId> leader = leaderMap.get(path); |
| 324 | if (leader != null && Objects.equals(leader.value(), localNodeId)) { |
| 325 | if (leaderMap.remove(path, leader.version())) { |
Madan Jampani | eedfe54 | 2015-05-26 09:54:09 -0700 | [diff] [blame] | 326 | log.debug("Stepped down from leadership for {}", path); |
Madan Jampani | d46e18f | 2015-05-04 23:19:33 -0700 | [diff] [blame] | 327 | publish(new LeadershipEvent( |
| 328 | LeadershipEvent.Type.LEADER_BOOTED, |
| 329 | new Leadership(path, |
| 330 | localNodeId, |
| 331 | leader.version(), |
| 332 | leader.creationTime()))); |
Madan Jampani | 1af8e13 | 2015-04-30 16:41:18 -0700 | [diff] [blame] | 333 | return true; |
| 334 | } |
| 335 | } |
| 336 | } catch (Exception e) { |
| 337 | log.warn("Error executing stepdown for {}", path, e); |
| 338 | } |
| 339 | return false; |
| 340 | } |
| 341 | |
| 342 | @Override |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 343 | public void addListener(LeadershipEventListener listener) { |
| 344 | listenerRegistry.addListener(listener); |
| 345 | } |
| 346 | |
| 347 | @Override |
| 348 | public void removeListener(LeadershipEventListener listener) { |
| 349 | listenerRegistry.removeListener(listener); |
| 350 | } |
| 351 | |
Madan Jampani | 1af8e13 | 2015-04-30 16:41:18 -0700 | [diff] [blame] | 352 | @Override |
| 353 | public boolean makeTopCandidate(String path, NodeId nodeId) { |
Madan Jampani | 346d4f5 | 2015-05-04 11:09:39 -0700 | [diff] [blame] | 354 | Versioned<List<NodeId>> newCandidates = candidateMap.computeIf(path, |
Madan Jampani | 3650fc4 | 2015-05-27 17:01:50 -0700 | [diff] [blame] | 355 | candidates -> candidates != null && |
| 356 | candidates.contains(nodeId) && |
| 357 | !nodeId.equals(Iterables.getFirst(candidates, null)), |
Madan Jampani | 346d4f5 | 2015-05-04 11:09:39 -0700 | [diff] [blame] | 358 | (topic, candidates) -> { |
| 359 | List<NodeId> updatedCandidates = new ArrayList<>(candidates.size()); |
| 360 | updatedCandidates.add(nodeId); |
| 361 | candidates.stream().filter(id -> !nodeId.equals(id)).forEach(updatedCandidates::add); |
| 362 | return updatedCandidates; |
| 363 | }); |
| 364 | publish(new LeadershipEvent( |
Madan Jampani | d46e18f | 2015-05-04 23:19:33 -0700 | [diff] [blame] | 365 | LeadershipEvent.Type.CANDIDATES_CHANGED, |
| 366 | new Leadership(path, |
| 367 | newCandidates.value(), |
| 368 | newCandidates.version(), |
| 369 | newCandidates.creationTime()))); |
Madan Jampani | 346d4f5 | 2015-05-04 11:09:39 -0700 | [diff] [blame] | 370 | return true; |
Madan Jampani | 1af8e13 | 2015-04-30 16:41:18 -0700 | [diff] [blame] | 371 | } |
| 372 | |
Madan Jampani | 7d243db | 2015-05-27 13:16:02 -0700 | [diff] [blame] | 373 | private Leadership electLeader(String path, List<NodeId> candidates) { |
| 374 | Leadership currentLeadership = getLeadership(path); |
| 375 | if (currentLeadership != null) { |
| 376 | return currentLeadership; |
| 377 | } else { |
| 378 | NodeId topCandidate = candidates |
| 379 | .stream() |
| 380 | .filter(n -> clusterService.getState(n) == ACTIVE) |
| 381 | .findFirst() |
| 382 | .orElse(null); |
| 383 | try { |
| 384 | Versioned<NodeId> leader = localNodeId.equals(topCandidate) |
| 385 | ? leaderMap.computeIfAbsent(path, p -> localNodeId) : leaderMap.get(path); |
| 386 | if (leader != null) { |
| 387 | Leadership newLeadership = new Leadership(path, |
| 388 | leader.value(), |
| 389 | leader.version(), |
| 390 | leader.creationTime()); |
| 391 | publish(new LeadershipEvent( |
| 392 | LeadershipEvent.Type.LEADER_ELECTED, |
| 393 | newLeadership)); |
| 394 | return newLeadership; |
Ayaka Koshibe | 0d886fc | 2015-04-23 11:53:41 -0700 | [diff] [blame] | 395 | } |
Madan Jampani | 7d243db | 2015-05-27 13:16:02 -0700 | [diff] [blame] | 396 | } catch (Exception e) { |
| 397 | log.debug("Failed to elect leader for {}", path, e); |
Ayaka Koshibe | c19b8b8 | 2015-04-08 15:18:24 -0700 | [diff] [blame] | 398 | } |
Ayaka Koshibe | c19b8b8 | 2015-04-08 15:18:24 -0700 | [diff] [blame] | 399 | } |
Madan Jampani | 7d243db | 2015-05-27 13:16:02 -0700 | [diff] [blame] | 400 | return null; |
Ayaka Koshibe | c19b8b8 | 2015-04-08 15:18:24 -0700 | [diff] [blame] | 401 | } |
| 402 | |
Madan Jampani | 7d243db | 2015-05-27 13:16:02 -0700 | [diff] [blame] | 403 | private void electLeaders() { |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 404 | try { |
Madan Jampani | 7d243db | 2015-05-27 13:16:02 -0700 | [diff] [blame] | 405 | candidateMap.entrySet().forEach(entry -> { |
| 406 | String path = entry.getKey(); |
Madan Jampani | cd40e5b | 2015-05-27 17:19:48 -0700 | [diff] [blame] | 407 | Versioned<List<NodeId>> candidates = entry.getValue(); |
| 408 | // for active topics, check if this node can become a leader (if it isn't already) |
Madan Jampani | 7d243db | 2015-05-27 13:16:02 -0700 | [diff] [blame] | 409 | if (activeTopics.contains(path)) { |
| 410 | lockExecutor.submit(() -> { |
Madan Jampani | cd40e5b | 2015-05-27 17:19:48 -0700 | [diff] [blame] | 411 | Leadership leadership = electLeader(path, candidates.value()); |
Madan Jampani | 7d243db | 2015-05-27 13:16:02 -0700 | [diff] [blame] | 412 | if (leadership != null) { |
| 413 | CompletableFuture<Leadership> future = pendingFutures.remove(path); |
| 414 | if (future != null) { |
| 415 | future.complete(leadership); |
| 416 | } |
| 417 | } |
| 418 | }); |
| 419 | } |
Madan Jampani | cd40e5b | 2015-05-27 17:19:48 -0700 | [diff] [blame] | 420 | // Raise a CANDIDATES_CHANGED event to force refresh local candidate board |
| 421 | // and also to update local listeners. |
| 422 | // Don't worry about duplicate events as they will be suppressed. |
| 423 | onLeadershipEvent(new LeadershipEvent(LeadershipEvent.Type.CANDIDATES_CHANGED, |
| 424 | new Leadership(path, |
| 425 | candidates.value(), |
| 426 | candidates.version(), |
| 427 | candidates.creationTime()))); |
Madan Jampani | 7d243db | 2015-05-27 13:16:02 -0700 | [diff] [blame] | 428 | }); |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 429 | } catch (Exception e) { |
Madan Jampani | 7d243db | 2015-05-27 13:16:02 -0700 | [diff] [blame] | 430 | log.debug("Failure electing leaders", e); |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 431 | } |
| 432 | } |
| 433 | |
Madan Jampani | d46e18f | 2015-05-04 23:19:33 -0700 | [diff] [blame] | 434 | private void publish(LeadershipEvent event) { |
| 435 | onLeadershipEvent(event); |
| 436 | clusterCommunicator.broadcast(event, LEADERSHIP_EVENT_MESSAGE_SUBJECT, SERIALIZER::encode); |
| 437 | } |
| 438 | |
| 439 | private void onLeadershipEvent(LeadershipEvent leadershipEvent) { |
| 440 | log.trace("Leadership Event: time = {} type = {} event = {}", |
| 441 | leadershipEvent.time(), leadershipEvent.type(), |
| 442 | leadershipEvent); |
| 443 | |
| 444 | Leadership leadershipUpdate = leadershipEvent.subject(); |
| 445 | LeadershipEvent.Type eventType = leadershipEvent.type(); |
| 446 | String topic = leadershipUpdate.topic(); |
| 447 | |
| 448 | AtomicBoolean updateAccepted = new AtomicBoolean(false); |
| 449 | if (eventType.equals(LeadershipEvent.Type.LEADER_ELECTED)) { |
| 450 | leaderBoard.compute(topic, (k, currentLeadership) -> { |
| 451 | if (currentLeadership == null || currentLeadership.epoch() < leadershipUpdate.epoch()) { |
| 452 | updateAccepted.set(true); |
| 453 | return leadershipUpdate; |
| 454 | } |
| 455 | return currentLeadership; |
| 456 | }); |
| 457 | } else if (eventType.equals(LeadershipEvent.Type.LEADER_BOOTED)) { |
| 458 | leaderBoard.compute(topic, (k, currentLeadership) -> { |
| 459 | if (currentLeadership == null || currentLeadership.epoch() <= leadershipUpdate.epoch()) { |
| 460 | updateAccepted.set(true); |
Madan Jampani | 08960a4 | 2015-06-04 09:59:29 -0700 | [diff] [blame] | 461 | // FIXME: Removing entries from leaderboard is not safe and should be visited. |
Madan Jampani | d46e18f | 2015-05-04 23:19:33 -0700 | [diff] [blame] | 462 | return null; |
| 463 | } |
| 464 | return currentLeadership; |
| 465 | }); |
| 466 | } else if (eventType.equals(LeadershipEvent.Type.CANDIDATES_CHANGED)) { |
| 467 | candidateBoard.compute(topic, (k, currentInfo) -> { |
| 468 | if (currentInfo == null || currentInfo.epoch() < leadershipUpdate.epoch()) { |
| 469 | updateAccepted.set(true); |
| 470 | return leadershipUpdate; |
| 471 | } |
| 472 | return currentInfo; |
| 473 | }); |
| 474 | } else { |
| 475 | throw new IllegalStateException("Unknown event type."); |
Ayaka Koshibe | fd26a30 | 2015-04-13 13:59:54 -0700 | [diff] [blame] | 476 | } |
Ayaka Koshibe | fd26a30 | 2015-04-13 13:59:54 -0700 | [diff] [blame] | 477 | |
Madan Jampani | d46e18f | 2015-05-04 23:19:33 -0700 | [diff] [blame] | 478 | if (updateAccepted.get()) { |
| 479 | eventDispatcher.post(leadershipEvent); |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 480 | } |
| 481 | } |
| 482 | |
Madan Jampani | de003d9 | 2015-05-11 17:14:20 -0700 | [diff] [blame] | 483 | private void rerunForLeadership(String path, CompletableFuture<Leadership> future) { |
Madan Jampani | 7d243db | 2015-05-27 13:16:02 -0700 | [diff] [blame] | 484 | lockExecutor.schedule( |
Madan Jampani | de003d9 | 2015-05-11 17:14:20 -0700 | [diff] [blame] | 485 | () -> doRunForLeadership(path, future), |
Madan Jampani | 8d37efc | 2015-06-01 17:25:48 -0700 | [diff] [blame] | 486 | RandomUtils.nextInt(WAIT_BEFORE_RETRY_MILLIS), |
| 487 | TimeUnit.MILLISECONDS); |
Ayaka Koshibe | c19b8b8 | 2015-04-08 15:18:24 -0700 | [diff] [blame] | 488 | } |
| 489 | |
Madan Jampani | de003d9 | 2015-05-11 17:14:20 -0700 | [diff] [blame] | 490 | private void retryWithdraw(String path, CompletableFuture<Void> future) { |
Madan Jampani | 7d243db | 2015-05-27 13:16:02 -0700 | [diff] [blame] | 491 | lockExecutor.schedule( |
Madan Jampani | de003d9 | 2015-05-11 17:14:20 -0700 | [diff] [blame] | 492 | () -> doWithdraw(path, future), |
Madan Jampani | 8d37efc | 2015-06-01 17:25:48 -0700 | [diff] [blame] | 493 | RandomUtils.nextInt(WAIT_BEFORE_RETRY_MILLIS), |
| 494 | TimeUnit.MILLISECONDS); |
Ayaka Koshibe | c19b8b8 | 2015-04-08 15:18:24 -0700 | [diff] [blame] | 495 | } |
| 496 | |
Madan Jampani | cc58675 | 2015-05-18 16:19:27 -0700 | [diff] [blame] | 497 | private void scheduleStaleLeadershipPurge(int afterDelaySec) { |
| 498 | if (staleLeadershipPurgeScheduled.compareAndSet(false, true)) { |
| 499 | staleLeadershipPurgeExecutor.schedule( |
| 500 | this::purgeStaleLeadership, |
| 501 | afterDelaySec, |
| 502 | TimeUnit.SECONDS); |
| 503 | } |
| 504 | } |
| 505 | |
| 506 | /** |
| 507 | * Purges locks held by inactive nodes and evicts inactive nodes from candidacy. |
| 508 | */ |
| 509 | private void purgeStaleLeadership() { |
| 510 | AtomicBoolean rerunPurge = new AtomicBoolean(false); |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 511 | try { |
Madan Jampani | cc58675 | 2015-05-18 16:19:27 -0700 | [diff] [blame] | 512 | staleLeadershipPurgeScheduled.set(false); |
Ayaka Koshibe | c19b8b8 | 2015-04-08 15:18:24 -0700 | [diff] [blame] | 513 | leaderMap.entrySet() |
| 514 | .stream() |
Ayaka Koshibe | 4a3c239 | 2015-04-22 18:03:13 -0700 | [diff] [blame] | 515 | .filter(e -> clusterService.getState(e.getValue().value()) == INACTIVE) |
Ayaka Koshibe | c19b8b8 | 2015-04-08 15:18:24 -0700 | [diff] [blame] | 516 | .forEach(entry -> { |
| 517 | String path = entry.getKey(); |
| 518 | NodeId nodeId = entry.getValue().value(); |
| 519 | long epoch = entry.getValue().version(); |
| 520 | long creationTime = entry.getValue().creationTime(); |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 521 | try { |
Ayaka Koshibe | c19b8b8 | 2015-04-08 15:18:24 -0700 | [diff] [blame] | 522 | if (leaderMap.remove(path, epoch)) { |
Madan Jampani | eedfe54 | 2015-05-26 09:54:09 -0700 | [diff] [blame] | 523 | log.debug("Purged stale lock held by {} for {}", nodeId, path); |
Madan Jampani | d46e18f | 2015-05-04 23:19:33 -0700 | [diff] [blame] | 524 | publish(new LeadershipEvent( |
| 525 | LeadershipEvent.Type.LEADER_BOOTED, |
| 526 | new Leadership(path, nodeId, epoch, creationTime))); |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 527 | } |
| 528 | } catch (Exception e) { |
Madan Jampani | ce8392b | 2015-05-20 11:13:47 -0700 | [diff] [blame] | 529 | log.debug("Failed to purge stale lock held by {} for {}", nodeId, path, e); |
Madan Jampani | cc58675 | 2015-05-18 16:19:27 -0700 | [diff] [blame] | 530 | rerunPurge.set(true); |
| 531 | } |
| 532 | }); |
| 533 | |
| 534 | candidateMap.entrySet() |
| 535 | .forEach(entry -> { |
| 536 | String path = entry.getKey(); |
| 537 | Versioned<List<NodeId>> candidates = entry.getValue(); |
| 538 | List<NodeId> candidatesList = candidates != null |
| 539 | ? candidates.value() : Collections.emptyList(); |
| 540 | List<NodeId> activeCandidatesList = |
| 541 | candidatesList.stream() |
| 542 | .filter(n -> clusterService.getState(n) == ACTIVE) |
| 543 | .filter(n -> !localNodeId.equals(n) || activeTopics.contains(path)) |
| 544 | .collect(Collectors.toList()); |
| 545 | if (activeCandidatesList.size() < candidatesList.size()) { |
| 546 | Set<NodeId> removedCandidates = |
| 547 | Sets.difference(Sets.newHashSet(candidatesList), |
| 548 | Sets.newHashSet(activeCandidatesList)); |
| 549 | try { |
| 550 | if (candidateMap.replace(path, entry.getValue().version(), activeCandidatesList)) { |
| 551 | log.info("Evicted inactive candidates {} from " |
| 552 | + "candidate list for {}", removedCandidates, path); |
| 553 | Versioned<List<NodeId>> updatedCandidates = candidateMap.get(path); |
| 554 | publish(new LeadershipEvent( |
| 555 | LeadershipEvent.Type.CANDIDATES_CHANGED, |
| 556 | new Leadership(path, |
| 557 | updatedCandidates.value(), |
| 558 | updatedCandidates.version(), |
| 559 | updatedCandidates.creationTime()))); |
Madan Jampani | 3d1727c | 2015-05-22 11:46:29 -0700 | [diff] [blame] | 560 | } else { |
| 561 | // Conflicting update detected. Rerun purge to make sure |
| 562 | // inactive candidates are evicted. |
| 563 | rerunPurge.set(true); |
Madan Jampani | cc58675 | 2015-05-18 16:19:27 -0700 | [diff] [blame] | 564 | } |
| 565 | } catch (Exception e) { |
Madan Jampani | ce8392b | 2015-05-20 11:13:47 -0700 | [diff] [blame] | 566 | log.debug("Failed to evict inactive candidates {} from " |
Madan Jampani | cc58675 | 2015-05-18 16:19:27 -0700 | [diff] [blame] | 567 | + "candidate list for {}", removedCandidates, path, e); |
| 568 | rerunPurge.set(true); |
| 569 | } |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 570 | } |
Ayaka Koshibe | c19b8b8 | 2015-04-08 15:18:24 -0700 | [diff] [blame] | 571 | }); |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 572 | } catch (Exception e) { |
Madan Jampani | ce8392b | 2015-05-20 11:13:47 -0700 | [diff] [blame] | 573 | log.debug("Failure purging state leadership.", e); |
Madan Jampani | cc58675 | 2015-05-18 16:19:27 -0700 | [diff] [blame] | 574 | rerunPurge.set(true); |
| 575 | } |
| 576 | |
| 577 | if (rerunPurge.get()) { |
Madan Jampani | ce8392b | 2015-05-20 11:13:47 -0700 | [diff] [blame] | 578 | log.debug("Rescheduling stale leadership purge due to errors encountered in previous run"); |
Madan Jampani | cc58675 | 2015-05-18 16:19:27 -0700 | [diff] [blame] | 579 | scheduleStaleLeadershipPurge(DELAY_BETWEEN_STALE_LEADERSHIP_PURGE_ATTEMPTS_SEC); |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 580 | } |
| 581 | } |
| 582 | |
Madan Jampani | 08960a4 | 2015-06-04 09:59:29 -0700 | [diff] [blame] | 583 | private void refreshLeaderBoard() { |
Madan Jampani | a14047d | 2015-02-25 12:23:02 -0800 | [diff] [blame] | 584 | try { |
Madan Jampani | 08960a4 | 2015-06-04 09:59:29 -0700 | [diff] [blame] | 585 | Map<String, Leadership> newLeaderBoard = Maps.newHashMap(); |
| 586 | leaderMap.entrySet().forEach(entry -> { |
| 587 | String path = entry.getKey(); |
| 588 | Versioned<NodeId> leader = entry.getValue(); |
| 589 | Leadership leadership = new Leadership(path, |
| 590 | leader.value(), |
| 591 | leader.version(), |
| 592 | leader.creationTime()); |
| 593 | newLeaderBoard.put(path, leadership); |
| 594 | }); |
| 595 | |
| 596 | // first take snapshot of current leader board. |
| 597 | Map<String, Leadership> currentLeaderBoard = ImmutableMap.copyOf(leaderBoard); |
| 598 | |
| 599 | // evict stale leaders |
| 600 | Maps.difference(currentLeaderBoard, newLeaderBoard).entriesOnlyOnLeft().forEach((path, leadership) -> { |
| 601 | log.debug("Evicting {} from leaderboard. It is no longer active leader.", leadership); |
| 602 | onLeadershipEvent(new LeadershipEvent(LeadershipEvent.Type.LEADER_BOOTED, leadership)); |
| 603 | }); |
| 604 | |
| 605 | // add missing leaders |
| 606 | Maps.difference(currentLeaderBoard, newLeaderBoard).entriesDiffering().forEach((path, difference) -> { |
| 607 | Leadership leadership = difference.rightValue(); |
| 608 | log.debug("Adding {} to leaderboard. It is now the active leader.", leadership); |
| 609 | onLeadershipEvent(new LeadershipEvent(LeadershipEvent.Type.LEADER_ELECTED, leadership)); |
Madan Jampani | a14047d | 2015-02-25 12:23:02 -0800 | [diff] [blame] | 610 | }); |
| 611 | } catch (Exception e) { |
Madan Jampani | 08960a4 | 2015-06-04 09:59:29 -0700 | [diff] [blame] | 612 | log.debug("Failed to refresh leader board", e); |
Madan Jampani | a14047d | 2015-02-25 12:23:02 -0800 | [diff] [blame] | 613 | } |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 614 | } |
Madan Jampani | cc58675 | 2015-05-18 16:19:27 -0700 | [diff] [blame] | 615 | |
| 616 | private class InternalClusterEventListener implements ClusterEventListener { |
| 617 | |
| 618 | @Override |
| 619 | public void event(ClusterEvent event) { |
| 620 | if (event.type() == Type.INSTANCE_DEACTIVATED || event.type() == Type.INSTANCE_REMOVED) { |
| 621 | scheduleStaleLeadershipPurge(0); |
| 622 | } |
| 623 | } |
| 624 | } |
Madan Jampani | a14047d | 2015-02-25 12:23:02 -0800 | [diff] [blame] | 625 | } |