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; |
Ayaka Koshibe | c19b8b8 | 2015-04-08 15:18:24 -0700 | [diff] [blame] | 5 | import com.google.common.collect.Lists; |
Jonathan Hart | e649c75 | 2015-03-03 18:04:25 -0800 | [diff] [blame] | 6 | import com.google.common.collect.Maps; |
| 7 | import com.google.common.collect.Sets; |
Ayaka Koshibe | c19b8b8 | 2015-04-08 15:18:24 -0700 | [diff] [blame] | 8 | |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 9 | import org.apache.felix.scr.annotations.Activate; |
| 10 | import org.apache.felix.scr.annotations.Component; |
| 11 | import org.apache.felix.scr.annotations.Deactivate; |
| 12 | import org.apache.felix.scr.annotations.Reference; |
| 13 | import org.apache.felix.scr.annotations.ReferenceCardinality; |
| 14 | import org.apache.felix.scr.annotations.Service; |
| 15 | import org.onlab.util.KryoNamespace; |
Madan Jampani | cc58675 | 2015-05-18 16:19:27 -0700 | [diff] [blame] | 16 | import org.onosproject.cluster.ClusterEvent; |
| 17 | import org.onosproject.cluster.ClusterEvent.Type; |
| 18 | import org.onosproject.cluster.ClusterEventListener; |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 19 | import org.onosproject.cluster.ClusterService; |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 20 | import org.onosproject.cluster.Leadership; |
| 21 | import org.onosproject.cluster.LeadershipEvent; |
| 22 | import org.onosproject.cluster.LeadershipEventListener; |
| 23 | import org.onosproject.cluster.LeadershipService; |
| 24 | import org.onosproject.cluster.NodeId; |
Simon Hunt | ff66374 | 2015-05-14 13:33:05 -0700 | [diff] [blame] | 25 | import org.onosproject.event.ListenerRegistry; |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 26 | import org.onosproject.event.EventDeliveryService; |
| 27 | import org.onosproject.store.cluster.messaging.ClusterCommunicationService; |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 28 | import org.onosproject.store.cluster.messaging.MessageSubject; |
| 29 | import org.onosproject.store.serializers.KryoNamespaces; |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 30 | import org.onosproject.store.service.ConsistentMap; |
Ayaka Koshibe | c19b8b8 | 2015-04-08 15:18:24 -0700 | [diff] [blame] | 31 | import org.onosproject.store.service.ConsistentMapException; |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 32 | import org.onosproject.store.service.Serializer; |
| 33 | import org.onosproject.store.service.StorageService; |
| 34 | import org.onosproject.store.service.Versioned; |
| 35 | import org.slf4j.Logger; |
| 36 | |
Madan Jampani | 1af8e13 | 2015-04-30 16:41:18 -0700 | [diff] [blame] | 37 | import java.util.ArrayList; |
Madan Jampani | cc58675 | 2015-05-18 16:19:27 -0700 | [diff] [blame] | 38 | import java.util.Collections; |
Jonathan Hart | e649c75 | 2015-03-03 18:04:25 -0800 | [diff] [blame] | 39 | import java.util.Map; |
| 40 | import java.util.Map.Entry; |
| 41 | import java.util.Objects; |
| 42 | import java.util.Set; |
Ayaka Koshibe | c19b8b8 | 2015-04-08 15:18:24 -0700 | [diff] [blame] | 43 | import java.util.List; |
Madan Jampani | de003d9 | 2015-05-11 17:14:20 -0700 | [diff] [blame] | 44 | import java.util.concurrent.CancellationException; |
| 45 | import java.util.concurrent.CompletableFuture; |
Jonathan Hart | e649c75 | 2015-03-03 18:04:25 -0800 | [diff] [blame] | 46 | import java.util.concurrent.ExecutorService; |
| 47 | import java.util.concurrent.Executors; |
| 48 | import java.util.concurrent.ScheduledExecutorService; |
| 49 | import java.util.concurrent.TimeUnit; |
Madan Jampani | d46e18f | 2015-05-04 23:19:33 -0700 | [diff] [blame] | 50 | import java.util.concurrent.atomic.AtomicBoolean; |
Jonathan Hart | e649c75 | 2015-03-03 18:04:25 -0800 | [diff] [blame] | 51 | import java.util.stream.Collectors; |
| 52 | |
| 53 | import static com.google.common.base.Preconditions.checkArgument; |
| 54 | import static org.onlab.util.Tools.groupedThreads; |
| 55 | import static org.slf4j.LoggerFactory.getLogger; |
Ayaka Koshibe | 4a3c239 | 2015-04-22 18:03:13 -0700 | [diff] [blame] | 56 | import static org.onosproject.cluster.ControllerNode.State.ACTIVE; |
| 57 | import static org.onosproject.cluster.ControllerNode.State.INACTIVE; |
| 58 | |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 59 | /** |
| 60 | * Distributed Lock Manager implemented on top of ConsistentMap. |
| 61 | * <p> |
Madan Jampani | cc58675 | 2015-05-18 16:19:27 -0700 | [diff] [blame] | 62 | * This implementation makes use of ClusterService's failure |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 63 | * detection capabilities to detect and purge stale locks. |
| 64 | * TODO: Ensure lock safety and liveness. |
| 65 | */ |
Madan Jampani | 7f72c3f | 2015-03-01 17:34:59 -0800 | [diff] [blame] | 66 | @Component(immediate = true, enabled = true) |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 67 | @Service |
| 68 | public class DistributedLeadershipManager implements LeadershipService { |
| 69 | |
| 70 | @Reference(cardinality = ReferenceCardinality.MANDATORY_UNARY) |
| 71 | protected StorageService storageService; |
| 72 | |
| 73 | @Reference(cardinality = ReferenceCardinality.MANDATORY_UNARY) |
| 74 | protected ClusterService clusterService; |
| 75 | |
| 76 | @Reference(cardinality = ReferenceCardinality.MANDATORY_UNARY) |
| 77 | protected ClusterCommunicationService clusterCommunicator; |
| 78 | |
| 79 | @Reference(cardinality = ReferenceCardinality.MANDATORY_UNARY) |
| 80 | protected EventDeliveryService eventDispatcher; |
| 81 | |
| 82 | private static final MessageSubject LEADERSHIP_EVENT_MESSAGE_SUBJECT = |
| 83 | new MessageSubject("distributed-leadership-manager-events"); |
| 84 | |
| 85 | private final Logger log = getLogger(getClass()); |
| 86 | private ExecutorService messageHandlingExecutor; |
| 87 | private ScheduledExecutorService retryLeaderLockExecutor; |
Madan Jampani | cc58675 | 2015-05-18 16:19:27 -0700 | [diff] [blame] | 88 | private ScheduledExecutorService staleLeadershipPurgeExecutor; |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 89 | private ScheduledExecutorService leadershipStatusBroadcaster; |
| 90 | |
Ayaka Koshibe | c19b8b8 | 2015-04-08 15:18:24 -0700 | [diff] [blame] | 91 | private ConsistentMap<String, NodeId> leaderMap; |
| 92 | private ConsistentMap<String, List<NodeId>> candidateMap; |
| 93 | |
Madan Jampani | cc58675 | 2015-05-18 16:19:27 -0700 | [diff] [blame] | 94 | private ListenerRegistry<LeadershipEvent, LeadershipEventListener> listenerRegistry; |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 95 | private final Map<String, Leadership> leaderBoard = Maps.newConcurrentMap(); |
Ayaka Koshibe | fd26a30 | 2015-04-13 13:59:54 -0700 | [diff] [blame] | 96 | private final Map<String, Leadership> candidateBoard = Maps.newConcurrentMap(); |
Madan Jampani | cc58675 | 2015-05-18 16:19:27 -0700 | [diff] [blame] | 97 | private final ClusterEventListener clusterEventListener = new InternalClusterEventListener(); |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 98 | |
Madan Jampani | cc58675 | 2015-05-18 16:19:27 -0700 | [diff] [blame] | 99 | private NodeId localNodeId; |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 100 | private Set<String> activeTopics = Sets.newConcurrentHashSet(); |
| 101 | |
Ayaka Koshibe | c19b8b8 | 2015-04-08 15:18:24 -0700 | [diff] [blame] | 102 | private static final int ELECTION_JOIN_ATTEMPT_INTERVAL_SEC = 2; |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 103 | private static final int DELAY_BETWEEN_LEADER_LOCK_ATTEMPTS_SEC = 2; |
Madan Jampani | a14047d | 2015-02-25 12:23:02 -0800 | [diff] [blame] | 104 | private static final int LEADERSHIP_STATUS_UPDATE_INTERVAL_SEC = 2; |
Madan Jampani | cc58675 | 2015-05-18 16:19:27 -0700 | [diff] [blame] | 105 | private static final int DELAY_BETWEEN_STALE_LEADERSHIP_PURGE_ATTEMPTS_SEC = 2; |
Ayaka Koshibe | c19b8b8 | 2015-04-08 15:18:24 -0700 | [diff] [blame] | 106 | private static final int LEADER_CANDIDATE_POS = 0; |
| 107 | |
Madan Jampani | cc58675 | 2015-05-18 16:19:27 -0700 | [diff] [blame] | 108 | private final AtomicBoolean staleLeadershipPurgeScheduled = new AtomicBoolean(false); |
| 109 | |
Ayaka Koshibe | c19b8b8 | 2015-04-08 15:18:24 -0700 | [diff] [blame] | 110 | private static final Serializer SERIALIZER = Serializer.using( |
| 111 | new KryoNamespace.Builder().register(KryoNamespaces.API).build()); |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 112 | |
| 113 | @Activate |
| 114 | public void activate() { |
Ayaka Koshibe | c19b8b8 | 2015-04-08 15:18:24 -0700 | [diff] [blame] | 115 | leaderMap = storageService.<String, NodeId>consistentMapBuilder() |
| 116 | .withName("onos-topic-leaders") |
| 117 | .withSerializer(SERIALIZER) |
| 118 | .withPartitionsDisabled().build(); |
| 119 | candidateMap = storageService.<String, List<NodeId>>consistentMapBuilder() |
| 120 | .withName("onos-topic-candidates") |
| 121 | .withSerializer(SERIALIZER) |
| 122 | .withPartitionsDisabled().build(); |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 123 | |
| 124 | localNodeId = clusterService.getLocalNode().id(); |
| 125 | |
| 126 | messageHandlingExecutor = Executors.newSingleThreadExecutor( |
| 127 | groupedThreads("onos/store/leadership", "message-handler")); |
| 128 | retryLeaderLockExecutor = Executors.newScheduledThreadPool( |
| 129 | 4, groupedThreads("onos/store/leadership", "election-thread-%d")); |
Madan Jampani | cc58675 | 2015-05-18 16:19:27 -0700 | [diff] [blame] | 130 | staleLeadershipPurgeExecutor = Executors.newSingleThreadScheduledExecutor( |
| 131 | groupedThreads("onos/store/leadership", "stale-leadership-evictor")); |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 132 | leadershipStatusBroadcaster = Executors.newSingleThreadScheduledExecutor( |
| 133 | groupedThreads("onos/store/leadership", "peer-updater")); |
| 134 | clusterCommunicator.addSubscriber( |
| 135 | LEADERSHIP_EVENT_MESSAGE_SUBJECT, |
Madan Jampani | d46e18f | 2015-05-04 23:19:33 -0700 | [diff] [blame] | 136 | SERIALIZER::decode, |
| 137 | this::onLeadershipEvent, |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 138 | messageHandlingExecutor); |
| 139 | |
Madan Jampani | cc58675 | 2015-05-18 16:19:27 -0700 | [diff] [blame] | 140 | clusterService.addListener(clusterEventListener); |
| 141 | |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 142 | leadershipStatusBroadcaster.scheduleWithFixedDelay( |
Madan Jampani | a14047d | 2015-02-25 12:23:02 -0800 | [diff] [blame] | 143 | this::sendLeadershipStatus, 0, LEADERSHIP_STATUS_UPDATE_INTERVAL_SEC, TimeUnit.SECONDS); |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 144 | |
Simon Hunt | ff66374 | 2015-05-14 13:33:05 -0700 | [diff] [blame] | 145 | listenerRegistry = new ListenerRegistry<>(); |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 146 | eventDispatcher.addSink(LeadershipEvent.class, listenerRegistry); |
| 147 | |
Madan Jampani | d46e18f | 2015-05-04 23:19:33 -0700 | [diff] [blame] | 148 | log.info("Started"); |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 149 | } |
| 150 | |
| 151 | @Deactivate |
| 152 | public void deactivate() { |
| 153 | leaderBoard.forEach((topic, leadership) -> { |
| 154 | if (localNodeId.equals(leadership.leader())) { |
| 155 | withdraw(topic); |
| 156 | } |
| 157 | }); |
| 158 | |
Madan Jampani | cc58675 | 2015-05-18 16:19:27 -0700 | [diff] [blame] | 159 | clusterService.removeListener(clusterEventListener); |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 160 | eventDispatcher.removeSink(LeadershipEvent.class); |
| 161 | clusterCommunicator.removeSubscriber(LEADERSHIP_EVENT_MESSAGE_SUBJECT); |
| 162 | |
| 163 | messageHandlingExecutor.shutdown(); |
| 164 | retryLeaderLockExecutor.shutdown(); |
Madan Jampani | cc58675 | 2015-05-18 16:19:27 -0700 | [diff] [blame] | 165 | staleLeadershipPurgeExecutor.shutdown(); |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 166 | leadershipStatusBroadcaster.shutdown(); |
| 167 | |
Madan Jampani | d46e18f | 2015-05-04 23:19:33 -0700 | [diff] [blame] | 168 | log.info("Stopped"); |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 169 | } |
| 170 | |
| 171 | @Override |
| 172 | public Map<String, Leadership> getLeaderBoard() { |
| 173 | return ImmutableMap.copyOf(leaderBoard); |
| 174 | } |
| 175 | |
| 176 | @Override |
Madan Jampani | fd45d5e | 2015-04-20 13:33:21 -0700 | [diff] [blame] | 177 | public Map<String, List<NodeId>> getCandidates() { |
| 178 | return Maps.toMap(candidateBoard.keySet(), this::getCandidates); |
Ayaka Koshibe | c19b8b8 | 2015-04-08 15:18:24 -0700 | [diff] [blame] | 179 | } |
| 180 | |
| 181 | @Override |
| 182 | public List<NodeId> getCandidates(String path) { |
Ayaka Koshibe | fd26a30 | 2015-04-13 13:59:54 -0700 | [diff] [blame] | 183 | Leadership current = candidateBoard.get(path); |
| 184 | return current == null ? ImmutableList.of() : ImmutableList.copyOf(current.candidates()); |
Ayaka Koshibe | c19b8b8 | 2015-04-08 15:18:24 -0700 | [diff] [blame] | 185 | } |
| 186 | |
| 187 | @Override |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 188 | public NodeId getLeader(String path) { |
| 189 | Leadership leadership = leaderBoard.get(path); |
| 190 | return leadership != null ? leadership.leader() : null; |
| 191 | } |
| 192 | |
| 193 | @Override |
Madan Jampani | 5961051 | 2015-02-25 15:25:43 -0800 | [diff] [blame] | 194 | public Leadership getLeadership(String path) { |
| 195 | checkArgument(path != null); |
| 196 | return leaderBoard.get(path); |
| 197 | } |
| 198 | |
| 199 | @Override |
| 200 | public Set<String> ownedTopics(NodeId nodeId) { |
| 201 | checkArgument(nodeId != null); |
| 202 | return leaderBoard.entrySet() |
| 203 | .stream() |
| 204 | .filter(entry -> nodeId.equals(entry.getValue().leader())) |
| 205 | .map(Entry::getKey) |
| 206 | .collect(Collectors.toSet()); |
| 207 | } |
| 208 | |
| 209 | @Override |
Madan Jampani | de003d9 | 2015-05-11 17:14:20 -0700 | [diff] [blame] | 210 | public CompletableFuture<Leadership> runForLeadership(String path) { |
Madan Jampani | 52860be | 2015-02-27 12:52:37 -0800 | [diff] [blame] | 211 | log.debug("Running for leadership for topic: {}", path); |
Madan Jampani | de003d9 | 2015-05-11 17:14:20 -0700 | [diff] [blame] | 212 | CompletableFuture<Leadership> resultFuture = new CompletableFuture<>(); |
| 213 | doRunForLeadership(path, resultFuture); |
| 214 | return resultFuture; |
| 215 | } |
| 216 | |
| 217 | private void doRunForLeadership(String path, CompletableFuture<Leadership> future) { |
Ayaka Koshibe | c19b8b8 | 2015-04-08 15:18:24 -0700 | [diff] [blame] | 218 | try { |
| 219 | Versioned<List<NodeId>> candidates = candidateMap.get(path); |
| 220 | if (candidates != null) { |
| 221 | List<NodeId> candidateList = Lists.newArrayList(candidates.value()); |
| 222 | if (!candidateList.contains(localNodeId)) { |
| 223 | candidateList.add(localNodeId); |
Ayaka Koshibe | fd26a30 | 2015-04-13 13:59:54 -0700 | [diff] [blame] | 224 | if (candidateMap.replace(path, candidates.version(), candidateList)) { |
| 225 | Versioned<List<NodeId>> newCandidates = candidateMap.get(path); |
Madan Jampani | d46e18f | 2015-05-04 23:19:33 -0700 | [diff] [blame] | 226 | publish(new LeadershipEvent( |
| 227 | LeadershipEvent.Type.CANDIDATES_CHANGED, |
| 228 | new Leadership(path, |
| 229 | newCandidates.value(), |
| 230 | newCandidates.version(), |
| 231 | newCandidates.creationTime()))); |
Ayaka Koshibe | fd26a30 | 2015-04-13 13:59:54 -0700 | [diff] [blame] | 232 | } else { |
Madan Jampani | de003d9 | 2015-05-11 17:14:20 -0700 | [diff] [blame] | 233 | rerunForLeadership(path, future); |
Ayaka Koshibe | c19b8b8 | 2015-04-08 15:18:24 -0700 | [diff] [blame] | 234 | return; |
| 235 | } |
| 236 | } |
| 237 | } else { |
Ayaka Koshibe | fd26a30 | 2015-04-13 13:59:54 -0700 | [diff] [blame] | 238 | List<NodeId> candidateList = ImmutableList.of(localNodeId); |
| 239 | if ((candidateMap.putIfAbsent(path, candidateList) == null)) { |
| 240 | Versioned<List<NodeId>> newCandidates = candidateMap.get(path); |
Madan Jampani | d46e18f | 2015-05-04 23:19:33 -0700 | [diff] [blame] | 241 | publish(new LeadershipEvent( |
| 242 | LeadershipEvent.Type.CANDIDATES_CHANGED, |
| 243 | new Leadership(path, |
| 244 | newCandidates.value(), |
| 245 | newCandidates.version(), |
| 246 | newCandidates.creationTime()))); |
Ayaka Koshibe | fd26a30 | 2015-04-13 13:59:54 -0700 | [diff] [blame] | 247 | } else { |
Madan Jampani | de003d9 | 2015-05-11 17:14:20 -0700 | [diff] [blame] | 248 | rerunForLeadership(path, future); |
Ayaka Koshibe | c19b8b8 | 2015-04-08 15:18:24 -0700 | [diff] [blame] | 249 | return; |
| 250 | } |
| 251 | } |
| 252 | log.debug("In the leadership race for topic {} with candidates {}", path, candidates); |
| 253 | activeTopics.add(path); |
Madan Jampani | de003d9 | 2015-05-11 17:14:20 -0700 | [diff] [blame] | 254 | tryLeaderLock(path, future); |
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())) { |
Jonathan Hart | e649c75 | 2015-03-03 18:04:25 -0800 | [diff] [blame] | 278 | log.info("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 { |
Ayaka Koshibe | c19b8b8 | 2015-04-08 15:18:24 -0700 | [diff] [blame] | 307 | log.warn("Failed to withdraw from candidates list. Will retry"); |
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 | d46e18f | 2015-05-04 23:19:33 -0700 | [diff] [blame] | 326 | log.info("Stepped down from leadership for {}", path); |
| 327 | publish(new LeadershipEvent( |
| 328 | LeadershipEvent.Type.LEADER_BOOTED, |
| 329 | new Leadership(path, |
| 330 | localNodeId, |
| 331 | leader.version(), |
| 332 | leader.creationTime()))); |
Madan Jampani | de003d9 | 2015-05-11 17:14:20 -0700 | [diff] [blame] | 333 | retryLock(path, new CompletableFuture<>()); |
Madan Jampani | 1af8e13 | 2015-04-30 16:41:18 -0700 | [diff] [blame] | 334 | return true; |
| 335 | } |
| 336 | } |
| 337 | } catch (Exception e) { |
| 338 | log.warn("Error executing stepdown for {}", path, e); |
| 339 | } |
| 340 | return false; |
| 341 | } |
| 342 | |
| 343 | @Override |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 344 | public void addListener(LeadershipEventListener listener) { |
| 345 | listenerRegistry.addListener(listener); |
| 346 | } |
| 347 | |
| 348 | @Override |
| 349 | public void removeListener(LeadershipEventListener listener) { |
| 350 | listenerRegistry.removeListener(listener); |
| 351 | } |
| 352 | |
Madan Jampani | 1af8e13 | 2015-04-30 16:41:18 -0700 | [diff] [blame] | 353 | @Override |
| 354 | public boolean makeTopCandidate(String path, NodeId nodeId) { |
| 355 | Versioned<List<NodeId>> candidates = candidateMap.get(path); |
| 356 | if (candidates == null || !candidates.value().contains(nodeId)) { |
| 357 | return false; |
| 358 | } |
Madan Jampani | d46e18f | 2015-05-04 23:19:33 -0700 | [diff] [blame] | 359 | List<NodeId> currentRoster = candidates.value(); |
| 360 | if (nodeId.equals(currentRoster.get(LEADER_CANDIDATE_POS))) { |
Madan Jampani | 1af8e13 | 2015-04-30 16:41:18 -0700 | [diff] [blame] | 361 | return true; |
| 362 | } |
Madan Jampani | 1af8e13 | 2015-04-30 16:41:18 -0700 | [diff] [blame] | 363 | List<NodeId> newRoster = new ArrayList<>(currentRoster.size()); |
| 364 | newRoster.add(nodeId); |
| 365 | currentRoster.stream().filter(id -> !nodeId.equals(id)).forEach(newRoster::add); |
| 366 | boolean updated = candidateMap.replace(path, candidates.version(), newRoster); |
| 367 | if (updated) { |
| 368 | Versioned<List<NodeId>> newCandidates = candidateMap.get(path); |
Madan Jampani | d46e18f | 2015-05-04 23:19:33 -0700 | [diff] [blame] | 369 | publish(new LeadershipEvent( |
| 370 | LeadershipEvent.Type.CANDIDATES_CHANGED, |
| 371 | new Leadership(path, |
| 372 | newCandidates.value(), |
| 373 | newCandidates.version(), |
| 374 | newCandidates.creationTime()))); |
Madan Jampani | 1af8e13 | 2015-04-30 16:41:18 -0700 | [diff] [blame] | 375 | } |
| 376 | return updated; |
| 377 | } |
| 378 | |
Madan Jampani | de003d9 | 2015-05-11 17:14:20 -0700 | [diff] [blame] | 379 | private void tryLeaderLock(String path, CompletableFuture<Leadership> future) { |
Madan Jampani | d46e18f | 2015-05-04 23:19:33 -0700 | [diff] [blame] | 380 | if (!activeTopics.contains(path) || Objects.equals(localNodeId, getLeader(path))) { |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 381 | return; |
| 382 | } |
Ayaka Koshibe | 0d886fc | 2015-04-23 11:53:41 -0700 | [diff] [blame] | 383 | try { |
| 384 | Versioned<List<NodeId>> candidates = candidateMap.get(path); |
| 385 | if (candidates != null) { |
Madan Jampani | d46e18f | 2015-05-04 23:19:33 -0700 | [diff] [blame] | 386 | List<NodeId> activeNodes = candidates.value() |
| 387 | .stream() |
Ayaka Koshibe | 0d886fc | 2015-04-23 11:53:41 -0700 | [diff] [blame] | 388 | .filter(n -> clusterService.getState(n) == ACTIVE) |
| 389 | .collect(Collectors.toList()); |
| 390 | if (localNodeId.equals(activeNodes.get(LEADER_CANDIDATE_POS))) { |
Madan Jampani | de003d9 | 2015-05-11 17:14:20 -0700 | [diff] [blame] | 391 | leaderLockAttempt(path, candidates.value(), future); |
Ayaka Koshibe | 0d886fc | 2015-04-23 11:53:41 -0700 | [diff] [blame] | 392 | } else { |
Madan Jampani | de003d9 | 2015-05-11 17:14:20 -0700 | [diff] [blame] | 393 | retryLock(path, future); |
Ayaka Koshibe | 0d886fc | 2015-04-23 11:53:41 -0700 | [diff] [blame] | 394 | } |
Ayaka Koshibe | c19b8b8 | 2015-04-08 15:18:24 -0700 | [diff] [blame] | 395 | } else { |
Ayaka Koshibe | 0d886fc | 2015-04-23 11:53:41 -0700 | [diff] [blame] | 396 | throw new IllegalStateException("should not be here"); |
Ayaka Koshibe | c19b8b8 | 2015-04-08 15:18:24 -0700 | [diff] [blame] | 397 | } |
Ayaka Koshibe | 0d886fc | 2015-04-23 11:53:41 -0700 | [diff] [blame] | 398 | } catch (Exception e) { |
| 399 | log.debug("Failed to fetch candidate information for {}", path, e); |
Madan Jampani | de003d9 | 2015-05-11 17:14:20 -0700 | [diff] [blame] | 400 | retryLock(path, future); |
Ayaka Koshibe | c19b8b8 | 2015-04-08 15:18:24 -0700 | [diff] [blame] | 401 | } |
| 402 | } |
| 403 | |
Madan Jampani | de003d9 | 2015-05-11 17:14:20 -0700 | [diff] [blame] | 404 | private void leaderLockAttempt(String path, List<NodeId> candidates, CompletableFuture<Leadership> future) { |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 405 | try { |
Ayaka Koshibe | c19b8b8 | 2015-04-08 15:18:24 -0700 | [diff] [blame] | 406 | Versioned<NodeId> currentLeader = leaderMap.get(path); |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 407 | if (currentLeader != null) { |
| 408 | if (localNodeId.equals(currentLeader.value())) { |
| 409 | log.info("Already has leadership for {}", path); |
Ayaka Koshibe | c19b8b8 | 2015-04-08 15:18:24 -0700 | [diff] [blame] | 410 | // FIXME: candidates can get out of sync. |
Madan Jampani | de003d9 | 2015-05-11 17:14:20 -0700 | [diff] [blame] | 411 | Leadership leadership = new Leadership(path, |
| 412 | localNodeId, |
| 413 | currentLeader.version(), |
| 414 | currentLeader.creationTime()); |
| 415 | future.complete(leadership); |
Madan Jampani | d46e18f | 2015-05-04 23:19:33 -0700 | [diff] [blame] | 416 | publish(new LeadershipEvent( |
| 417 | LeadershipEvent.Type.LEADER_ELECTED, |
Madan Jampani | de003d9 | 2015-05-11 17:14:20 -0700 | [diff] [blame] | 418 | leadership)); |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 419 | } else { |
| 420 | // someone else has leadership. will retry after sometime. |
Madan Jampani | de003d9 | 2015-05-11 17:14:20 -0700 | [diff] [blame] | 421 | retryLock(path, future); |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 422 | } |
| 423 | } else { |
Ayaka Koshibe | c19b8b8 | 2015-04-08 15:18:24 -0700 | [diff] [blame] | 424 | if (leaderMap.putIfAbsent(path, localNodeId) == null) { |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 425 | log.info("Assumed leadership for {}", path); |
| 426 | // do a get again to get the version (epoch) |
Ayaka Koshibe | c19b8b8 | 2015-04-08 15:18:24 -0700 | [diff] [blame] | 427 | Versioned<NodeId> newLeader = leaderMap.get(path); |
| 428 | // FIXME: candidates can get out of sync |
Madan Jampani | de003d9 | 2015-05-11 17:14:20 -0700 | [diff] [blame] | 429 | Leadership leadership = new Leadership(path, |
| 430 | newLeader.value(), |
| 431 | newLeader.version(), |
| 432 | newLeader.creationTime()); |
| 433 | future.complete(leadership); |
Madan Jampani | d46e18f | 2015-05-04 23:19:33 -0700 | [diff] [blame] | 434 | publish(new LeadershipEvent( |
| 435 | LeadershipEvent.Type.LEADER_ELECTED, |
Madan Jampani | de003d9 | 2015-05-11 17:14:20 -0700 | [diff] [blame] | 436 | leadership)); |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 437 | } else { |
| 438 | // someone beat us to it. |
Madan Jampani | de003d9 | 2015-05-11 17:14:20 -0700 | [diff] [blame] | 439 | retryLock(path, future); |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 440 | } |
| 441 | } |
| 442 | } catch (Exception e) { |
Madan Jampani | a14047d | 2015-02-25 12:23:02 -0800 | [diff] [blame] | 443 | log.debug("Attempt to acquire leadership lock for topic {} failed", path, e); |
Madan Jampani | de003d9 | 2015-05-11 17:14:20 -0700 | [diff] [blame] | 444 | retryLock(path, future); |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 445 | } |
| 446 | } |
| 447 | |
Madan Jampani | d46e18f | 2015-05-04 23:19:33 -0700 | [diff] [blame] | 448 | private void publish(LeadershipEvent event) { |
| 449 | onLeadershipEvent(event); |
| 450 | clusterCommunicator.broadcast(event, LEADERSHIP_EVENT_MESSAGE_SUBJECT, SERIALIZER::encode); |
| 451 | } |
| 452 | |
| 453 | private void onLeadershipEvent(LeadershipEvent leadershipEvent) { |
| 454 | log.trace("Leadership Event: time = {} type = {} event = {}", |
| 455 | leadershipEvent.time(), leadershipEvent.type(), |
| 456 | leadershipEvent); |
| 457 | |
| 458 | Leadership leadershipUpdate = leadershipEvent.subject(); |
| 459 | LeadershipEvent.Type eventType = leadershipEvent.type(); |
| 460 | String topic = leadershipUpdate.topic(); |
| 461 | |
| 462 | AtomicBoolean updateAccepted = new AtomicBoolean(false); |
| 463 | if (eventType.equals(LeadershipEvent.Type.LEADER_ELECTED)) { |
| 464 | leaderBoard.compute(topic, (k, currentLeadership) -> { |
| 465 | if (currentLeadership == null || currentLeadership.epoch() < leadershipUpdate.epoch()) { |
| 466 | updateAccepted.set(true); |
| 467 | return leadershipUpdate; |
| 468 | } |
| 469 | return currentLeadership; |
| 470 | }); |
| 471 | } else if (eventType.equals(LeadershipEvent.Type.LEADER_BOOTED)) { |
| 472 | leaderBoard.compute(topic, (k, currentLeadership) -> { |
| 473 | if (currentLeadership == null || currentLeadership.epoch() <= leadershipUpdate.epoch()) { |
| 474 | updateAccepted.set(true); |
| 475 | return null; |
| 476 | } |
| 477 | return currentLeadership; |
| 478 | }); |
| 479 | } else if (eventType.equals(LeadershipEvent.Type.CANDIDATES_CHANGED)) { |
| 480 | candidateBoard.compute(topic, (k, currentInfo) -> { |
| 481 | if (currentInfo == null || currentInfo.epoch() < leadershipUpdate.epoch()) { |
| 482 | updateAccepted.set(true); |
| 483 | return leadershipUpdate; |
| 484 | } |
| 485 | return currentInfo; |
| 486 | }); |
| 487 | } else { |
| 488 | throw new IllegalStateException("Unknown event type."); |
Ayaka Koshibe | fd26a30 | 2015-04-13 13:59:54 -0700 | [diff] [blame] | 489 | } |
Ayaka Koshibe | fd26a30 | 2015-04-13 13:59:54 -0700 | [diff] [blame] | 490 | |
Madan Jampani | d46e18f | 2015-05-04 23:19:33 -0700 | [diff] [blame] | 491 | if (updateAccepted.get()) { |
| 492 | eventDispatcher.post(leadershipEvent); |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 493 | } |
| 494 | } |
| 495 | |
Madan Jampani | de003d9 | 2015-05-11 17:14:20 -0700 | [diff] [blame] | 496 | private void rerunForLeadership(String path, CompletableFuture<Leadership> future) { |
Ayaka Koshibe | c19b8b8 | 2015-04-08 15:18:24 -0700 | [diff] [blame] | 497 | retryLeaderLockExecutor.schedule( |
Madan Jampani | de003d9 | 2015-05-11 17:14:20 -0700 | [diff] [blame] | 498 | () -> doRunForLeadership(path, future), |
Ayaka Koshibe | c19b8b8 | 2015-04-08 15:18:24 -0700 | [diff] [blame] | 499 | ELECTION_JOIN_ATTEMPT_INTERVAL_SEC, |
| 500 | TimeUnit.SECONDS); |
| 501 | } |
| 502 | |
Madan Jampani | de003d9 | 2015-05-11 17:14:20 -0700 | [diff] [blame] | 503 | private void retryLock(String path, CompletableFuture<Leadership> future) { |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 504 | retryLeaderLockExecutor.schedule( |
Madan Jampani | de003d9 | 2015-05-11 17:14:20 -0700 | [diff] [blame] | 505 | () -> tryLeaderLock(path, future), |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 506 | DELAY_BETWEEN_LEADER_LOCK_ATTEMPTS_SEC, |
| 507 | TimeUnit.SECONDS); |
| 508 | } |
| 509 | |
Madan Jampani | de003d9 | 2015-05-11 17:14:20 -0700 | [diff] [blame] | 510 | private void retryWithdraw(String path, CompletableFuture<Void> future) { |
Ayaka Koshibe | c19b8b8 | 2015-04-08 15:18:24 -0700 | [diff] [blame] | 511 | retryLeaderLockExecutor.schedule( |
Madan Jampani | de003d9 | 2015-05-11 17:14:20 -0700 | [diff] [blame] | 512 | () -> doWithdraw(path, future), |
Ayaka Koshibe | c19b8b8 | 2015-04-08 15:18:24 -0700 | [diff] [blame] | 513 | DELAY_BETWEEN_LEADER_LOCK_ATTEMPTS_SEC, |
| 514 | TimeUnit.SECONDS); |
| 515 | } |
| 516 | |
Madan Jampani | cc58675 | 2015-05-18 16:19:27 -0700 | [diff] [blame] | 517 | private void scheduleStaleLeadershipPurge(int afterDelaySec) { |
| 518 | if (staleLeadershipPurgeScheduled.compareAndSet(false, true)) { |
| 519 | staleLeadershipPurgeExecutor.schedule( |
| 520 | this::purgeStaleLeadership, |
| 521 | afterDelaySec, |
| 522 | TimeUnit.SECONDS); |
| 523 | } |
| 524 | } |
| 525 | |
| 526 | /** |
| 527 | * Purges locks held by inactive nodes and evicts inactive nodes from candidacy. |
| 528 | */ |
| 529 | private void purgeStaleLeadership() { |
| 530 | AtomicBoolean rerunPurge = new AtomicBoolean(false); |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 531 | try { |
Madan Jampani | cc58675 | 2015-05-18 16:19:27 -0700 | [diff] [blame] | 532 | staleLeadershipPurgeScheduled.set(false); |
Ayaka Koshibe | c19b8b8 | 2015-04-08 15:18:24 -0700 | [diff] [blame] | 533 | leaderMap.entrySet() |
| 534 | .stream() |
Ayaka Koshibe | 4a3c239 | 2015-04-22 18:03:13 -0700 | [diff] [blame] | 535 | .filter(e -> clusterService.getState(e.getValue().value()) == INACTIVE) |
Ayaka Koshibe | c19b8b8 | 2015-04-08 15:18:24 -0700 | [diff] [blame] | 536 | .forEach(entry -> { |
| 537 | String path = entry.getKey(); |
| 538 | NodeId nodeId = entry.getValue().value(); |
| 539 | long epoch = entry.getValue().version(); |
| 540 | long creationTime = entry.getValue().creationTime(); |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 541 | try { |
Ayaka Koshibe | c19b8b8 | 2015-04-08 15:18:24 -0700 | [diff] [blame] | 542 | if (leaderMap.remove(path, epoch)) { |
Madan Jampani | a14047d | 2015-02-25 12:23:02 -0800 | [diff] [blame] | 543 | log.info("Purged stale lock held by {} for {}", nodeId, path); |
Madan Jampani | d46e18f | 2015-05-04 23:19:33 -0700 | [diff] [blame] | 544 | publish(new LeadershipEvent( |
| 545 | LeadershipEvent.Type.LEADER_BOOTED, |
| 546 | new Leadership(path, nodeId, epoch, creationTime))); |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 547 | } |
| 548 | } catch (Exception e) { |
Madan Jampani | ce8392b | 2015-05-20 11:13:47 -0700 | [diff] [blame] | 549 | 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] | 550 | rerunPurge.set(true); |
| 551 | } |
| 552 | }); |
| 553 | |
| 554 | candidateMap.entrySet() |
| 555 | .forEach(entry -> { |
| 556 | String path = entry.getKey(); |
| 557 | Versioned<List<NodeId>> candidates = entry.getValue(); |
| 558 | List<NodeId> candidatesList = candidates != null |
| 559 | ? candidates.value() : Collections.emptyList(); |
| 560 | List<NodeId> activeCandidatesList = |
| 561 | candidatesList.stream() |
| 562 | .filter(n -> clusterService.getState(n) == ACTIVE) |
| 563 | .filter(n -> !localNodeId.equals(n) || activeTopics.contains(path)) |
| 564 | .collect(Collectors.toList()); |
| 565 | if (activeCandidatesList.size() < candidatesList.size()) { |
| 566 | Set<NodeId> removedCandidates = |
| 567 | Sets.difference(Sets.newHashSet(candidatesList), |
| 568 | Sets.newHashSet(activeCandidatesList)); |
| 569 | try { |
| 570 | if (candidateMap.replace(path, entry.getValue().version(), activeCandidatesList)) { |
| 571 | log.info("Evicted inactive candidates {} from " |
| 572 | + "candidate list for {}", removedCandidates, path); |
| 573 | Versioned<List<NodeId>> updatedCandidates = candidateMap.get(path); |
| 574 | publish(new LeadershipEvent( |
| 575 | LeadershipEvent.Type.CANDIDATES_CHANGED, |
| 576 | new Leadership(path, |
| 577 | updatedCandidates.value(), |
| 578 | updatedCandidates.version(), |
| 579 | updatedCandidates.creationTime()))); |
Madan Jampani | 3d1727c | 2015-05-22 11:46:29 -0700 | [diff] [blame^] | 580 | } else { |
| 581 | // Conflicting update detected. Rerun purge to make sure |
| 582 | // inactive candidates are evicted. |
| 583 | rerunPurge.set(true); |
Madan Jampani | cc58675 | 2015-05-18 16:19:27 -0700 | [diff] [blame] | 584 | } |
| 585 | } catch (Exception e) { |
Madan Jampani | ce8392b | 2015-05-20 11:13:47 -0700 | [diff] [blame] | 586 | log.debug("Failed to evict inactive candidates {} from " |
Madan Jampani | cc58675 | 2015-05-18 16:19:27 -0700 | [diff] [blame] | 587 | + "candidate list for {}", removedCandidates, path, e); |
| 588 | rerunPurge.set(true); |
| 589 | } |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 590 | } |
Ayaka Koshibe | c19b8b8 | 2015-04-08 15:18:24 -0700 | [diff] [blame] | 591 | }); |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 592 | } catch (Exception e) { |
Madan Jampani | ce8392b | 2015-05-20 11:13:47 -0700 | [diff] [blame] | 593 | log.debug("Failure purging state leadership.", e); |
Madan Jampani | cc58675 | 2015-05-18 16:19:27 -0700 | [diff] [blame] | 594 | rerunPurge.set(true); |
| 595 | } |
| 596 | |
| 597 | if (rerunPurge.get()) { |
Madan Jampani | ce8392b | 2015-05-20 11:13:47 -0700 | [diff] [blame] | 598 | 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] | 599 | scheduleStaleLeadershipPurge(DELAY_BETWEEN_STALE_LEADERSHIP_PURGE_ATTEMPTS_SEC); |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 600 | } |
| 601 | } |
| 602 | |
| 603 | private void sendLeadershipStatus() { |
Madan Jampani | a14047d | 2015-02-25 12:23:02 -0800 | [diff] [blame] | 604 | try { |
| 605 | leaderBoard.forEach((path, leadership) -> { |
| 606 | if (leadership.leader().equals(localNodeId)) { |
| 607 | LeadershipEvent event = new LeadershipEvent(LeadershipEvent.Type.LEADER_ELECTED, leadership); |
Madan Jampani | 2bfa94c | 2015-04-11 05:03:49 -0700 | [diff] [blame] | 608 | clusterCommunicator.broadcast(event, |
| 609 | LEADERSHIP_EVENT_MESSAGE_SUBJECT, |
| 610 | SERIALIZER::encode); |
Madan Jampani | a14047d | 2015-02-25 12:23:02 -0800 | [diff] [blame] | 611 | } |
| 612 | }); |
Ayaka Koshibe | fd26a30 | 2015-04-13 13:59:54 -0700 | [diff] [blame] | 613 | candidateBoard.forEach((path, leadership) -> { |
| 614 | LeadershipEvent event = new LeadershipEvent(LeadershipEvent.Type.CANDIDATES_CHANGED, leadership); |
| 615 | clusterCommunicator.broadcast(event, |
| 616 | LEADERSHIP_EVENT_MESSAGE_SUBJECT, |
| 617 | SERIALIZER::encode); |
| 618 | }); |
Madan Jampani | a14047d | 2015-02-25 12:23:02 -0800 | [diff] [blame] | 619 | } catch (Exception e) { |
| 620 | log.debug("Failed to send leadership updates", e); |
| 621 | } |
Madan Jampani | d14166a | 2015-02-24 17:37:51 -0800 | [diff] [blame] | 622 | } |
Madan Jampani | cc58675 | 2015-05-18 16:19:27 -0700 | [diff] [blame] | 623 | |
| 624 | private class InternalClusterEventListener implements ClusterEventListener { |
| 625 | |
| 626 | @Override |
| 627 | public void event(ClusterEvent event) { |
| 628 | if (event.type() == Type.INSTANCE_DEACTIVATED || event.type() == Type.INSTANCE_REMOVED) { |
| 629 | scheduleStaleLeadershipPurge(0); |
| 630 | } |
| 631 | } |
| 632 | } |
Madan Jampani | a14047d | 2015-02-25 12:23:02 -0800 | [diff] [blame] | 633 | } |