alshabib | ab98466 | 2014-12-04 18:56:18 -0800 | [diff] [blame] | 1 | /* |
| 2 | * Copyright 2014 Open Networking Laboratory |
| 3 | * |
| 4 | * Licensed under the Apache License, Version 2.0 (the "License"); |
| 5 | * you may not use this file except in compliance with the License. |
| 6 | * You may obtain a copy of the License at |
| 7 | * |
| 8 | * http://www.apache.org/licenses/LICENSE-2.0 |
| 9 | * |
| 10 | * Unless required by applicable law or agreed to in writing, software |
| 11 | * distributed under the License is distributed on an "AS IS" BASIS, |
| 12 | * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 13 | * See the License for the specific language governing permissions and |
| 14 | * limitations under the License. |
| 15 | */ |
Brian O'Connor | abafb50 | 2014-12-02 22:26:20 -0800 | [diff] [blame] | 16 | package org.onosproject.store.cluster.impl; |
Madan Jampani | 1d3494e | 2014-11-20 11:24:22 -0800 | [diff] [blame] | 17 | |
Jonathan Hart | 7d656f4 | 2015-01-27 14:07:23 -0800 | [diff] [blame^] | 18 | import com.google.common.collect.ImmutableMap; |
| 19 | import com.google.common.collect.Maps; |
| 20 | import com.google.common.collect.Sets; |
Madan Jampani | 1d3494e | 2014-11-20 11:24:22 -0800 | [diff] [blame] | 21 | import org.apache.felix.scr.annotations.Activate; |
| 22 | import org.apache.felix.scr.annotations.Component; |
| 23 | import org.apache.felix.scr.annotations.Deactivate; |
| 24 | import org.apache.felix.scr.annotations.Reference; |
| 25 | import org.apache.felix.scr.annotations.ReferenceCardinality; |
| 26 | import org.apache.felix.scr.annotations.Service; |
Jonathan Hart | 7d656f4 | 2015-01-27 14:07:23 -0800 | [diff] [blame^] | 27 | import org.onlab.util.KryoNamespace; |
Brian O'Connor | abafb50 | 2014-12-02 22:26:20 -0800 | [diff] [blame] | 28 | import org.onosproject.cluster.ClusterService; |
| 29 | import org.onosproject.cluster.Leadership; |
| 30 | import org.onosproject.cluster.LeadershipEvent; |
| 31 | import org.onosproject.cluster.LeadershipEventListener; |
| 32 | import org.onosproject.cluster.LeadershipService; |
| 33 | import org.onosproject.cluster.NodeId; |
| 34 | import org.onosproject.store.cluster.messaging.ClusterCommunicationService; |
| 35 | import org.onosproject.store.cluster.messaging.ClusterMessage; |
| 36 | import org.onosproject.store.cluster.messaging.ClusterMessageHandler; |
| 37 | import org.onosproject.store.cluster.messaging.MessageSubject; |
| 38 | import org.onosproject.store.serializers.KryoNamespaces; |
| 39 | import org.onosproject.store.serializers.KryoSerializer; |
| 40 | import org.onosproject.store.service.Lock; |
| 41 | import org.onosproject.store.service.LockService; |
| 42 | import org.onosproject.store.service.impl.DistributedLockManager; |
Madan Jampani | 1d3494e | 2014-11-20 11:24:22 -0800 | [diff] [blame] | 43 | import org.slf4j.Logger; |
| 44 | |
Jonathan Hart | 7d656f4 | 2015-01-27 14:07:23 -0800 | [diff] [blame^] | 45 | import java.util.Map; |
| 46 | import java.util.Set; |
| 47 | import java.util.concurrent.Executors; |
| 48 | import java.util.concurrent.ScheduledExecutorService; |
| 49 | import java.util.concurrent.TimeUnit; |
| 50 | |
| 51 | import static com.google.common.base.Preconditions.checkArgument; |
| 52 | import static org.onlab.util.Tools.namedThreads; |
| 53 | import static org.slf4j.LoggerFactory.getLogger; |
Madan Jampani | 1d3494e | 2014-11-20 11:24:22 -0800 | [diff] [blame] | 54 | |
| 55 | /** |
| 56 | * Distributed implementation of LeadershipService that is based on the primitives exposed by |
| 57 | * LockService. |
| 58 | */ |
Jonathan Hart | 949c284 | 2014-11-28 23:44:09 -0800 | [diff] [blame] | 59 | @Component(enabled = false) |
Madan Jampani | 1d3494e | 2014-11-20 11:24:22 -0800 | [diff] [blame] | 60 | @Service |
| 61 | public class LeadershipManager implements LeadershipService { |
| 62 | |
| 63 | private final Logger log = getLogger(getClass()); |
| 64 | |
| 65 | // TODO: Remove this dependency |
| 66 | private static final int TERM_DURATION_MS = |
| 67 | DistributedLockManager.DEAD_LOCK_TIMEOUT_MS; |
| 68 | |
Madan Jampani | 1ee9178 | 2014-11-20 20:24:24 -0800 | [diff] [blame] | 69 | // Time to wait before retrying leadership after |
| 70 | // a unexpected error. |
| 71 | private static final int WAIT_BEFORE_RETRY_MS = 2000; |
| 72 | |
Yuta HIGUCHI | 6593489 | 2014-12-04 17:47:44 -0800 | [diff] [blame] | 73 | // TODO: Make Thread pool size configurable. |
Madan Jampani | ddaffd0 | 2014-11-21 13:12:09 -0800 | [diff] [blame] | 74 | private final ScheduledExecutorService threadPool = |
Thomas Vachuska | 9ea3e6f | 2015-01-23 16:34:22 -0800 | [diff] [blame] | 75 | Executors.newScheduledThreadPool(25, namedThreads("onos-leadership-manager-%d")); |
Madan Jampani | 1d3494e | 2014-11-20 11:24:22 -0800 | [diff] [blame] | 76 | |
Madan Jampani | 1ee9178 | 2014-11-20 20:24:24 -0800 | [diff] [blame] | 77 | private static final MessageSubject LEADERSHIP_UPDATES = |
| 78 | new MessageSubject("leadership-contest-updates"); |
| 79 | |
Madan Jampani | 1d3494e | 2014-11-20 11:24:22 -0800 | [diff] [blame] | 80 | @Reference(cardinality = ReferenceCardinality.MANDATORY_UNARY) |
| 81 | private ClusterService clusterService; |
| 82 | |
| 83 | @Reference(cardinality = ReferenceCardinality.MANDATORY_UNARY) |
Madan Jampani | 1ee9178 | 2014-11-20 20:24:24 -0800 | [diff] [blame] | 84 | private ClusterCommunicationService clusterCommunicator; |
| 85 | |
| 86 | @Reference(cardinality = ReferenceCardinality.MANDATORY_UNARY) |
Madan Jampani | 1d3494e | 2014-11-20 11:24:22 -0800 | [diff] [blame] | 87 | private LockService lockService; |
| 88 | |
Madan Jampani | 1ee9178 | 2014-11-20 20:24:24 -0800 | [diff] [blame] | 89 | private final Map<String, Leadership> leaderBoard = Maps.newHashMap(); |
| 90 | |
Madan Jampani | d1a467f | 2014-11-24 15:11:09 -0800 | [diff] [blame] | 91 | private final Map<String, Lock> openContests = Maps.newConcurrentMap(); |
Madan Jampani | 1ee9178 | 2014-11-20 20:24:24 -0800 | [diff] [blame] | 92 | private final Set<LeadershipEventListener> listeners = Sets.newIdentityHashSet(); |
Madan Jampani | 8d21c79 | 2014-12-01 16:31:07 -0800 | [diff] [blame] | 93 | private NodeId localNodeId; |
Madan Jampani | 1d3494e | 2014-11-20 11:24:22 -0800 | [diff] [blame] | 94 | |
Madan Jampani | 1ee9178 | 2014-11-20 20:24:24 -0800 | [diff] [blame] | 95 | private final LeadershipEventListener peerAdvertiser = new PeerAdvertiser(); |
| 96 | private final LeadershipEventListener leaderBoardUpdater = new LeaderBoardUpdater(); |
| 97 | |
| 98 | public static final KryoSerializer SERIALIZER = new KryoSerializer() { |
| 99 | @Override |
| 100 | protected void setupKryoPool() { |
| 101 | serializerPool = KryoNamespace.newBuilder() |
| 102 | .register(KryoNamespaces.API) |
| 103 | .build() |
| 104 | .populate(1); |
| 105 | } |
| 106 | }; |
| 107 | |
Madan Jampani | 1d3494e | 2014-11-20 11:24:22 -0800 | [diff] [blame] | 108 | @Activate |
| 109 | public void activate() { |
Madan Jampani | 8d21c79 | 2014-12-01 16:31:07 -0800 | [diff] [blame] | 110 | localNodeId = clusterService.getLocalNode().id(); |
Madan Jampani | 1ee9178 | 2014-11-20 20:24:24 -0800 | [diff] [blame] | 111 | |
| 112 | addListener(peerAdvertiser); |
| 113 | addListener(leaderBoardUpdater); |
| 114 | |
| 115 | clusterCommunicator.addSubscriber( |
| 116 | LEADERSHIP_UPDATES, |
| 117 | new PeerAdvertisementHandler()); |
| 118 | |
Madan Jampani | 1d3494e | 2014-11-20 11:24:22 -0800 | [diff] [blame] | 119 | log.info("Started."); |
| 120 | } |
| 121 | |
| 122 | @Deactivate |
| 123 | public void deactivate() { |
Madan Jampani | 1ee9178 | 2014-11-20 20:24:24 -0800 | [diff] [blame] | 124 | removeListener(peerAdvertiser); |
| 125 | removeListener(leaderBoardUpdater); |
| 126 | |
| 127 | clusterCommunicator.removeSubscriber(LEADERSHIP_UPDATES); |
| 128 | |
Madan Jampani | ddaffd0 | 2014-11-21 13:12:09 -0800 | [diff] [blame] | 129 | threadPool.shutdown(); |
Madan Jampani | 1ee9178 | 2014-11-20 20:24:24 -0800 | [diff] [blame] | 130 | |
Madan Jampani | 1d3494e | 2014-11-20 11:24:22 -0800 | [diff] [blame] | 131 | log.info("Stopped."); |
| 132 | } |
| 133 | |
Madan Jampani | 1ee9178 | 2014-11-20 20:24:24 -0800 | [diff] [blame] | 134 | |
| 135 | @Override |
Madan Jampani | 8d21c79 | 2014-12-01 16:31:07 -0800 | [diff] [blame] | 136 | public NodeId getLeader(String path) { |
Madan Jampani | 1ee9178 | 2014-11-20 20:24:24 -0800 | [diff] [blame] | 137 | synchronized (leaderBoard) { |
| 138 | Leadership leadership = leaderBoard.get(path); |
| 139 | if (leadership != null) { |
| 140 | return leadership.leader(); |
| 141 | } |
| 142 | } |
| 143 | return null; |
| 144 | } |
| 145 | |
Madan Jampani | 1d3494e | 2014-11-20 11:24:22 -0800 | [diff] [blame] | 146 | @Override |
| 147 | public void runForLeadership(String path) { |
| 148 | checkArgument(path != null); |
Madan Jampani | d1a467f | 2014-11-24 15:11:09 -0800 | [diff] [blame] | 149 | |
Madan Jampani | 1d3494e | 2014-11-20 11:24:22 -0800 | [diff] [blame] | 150 | if (openContests.containsKey(path)) { |
| 151 | log.info("Already in the leadership contest for {}", path); |
| 152 | return; |
| 153 | } else { |
| 154 | Lock lock = lockService.create(path); |
| 155 | openContests.put(path, lock); |
Madan Jampani | d1a467f | 2014-11-24 15:11:09 -0800 | [diff] [blame] | 156 | threadPool.schedule(new TryLeadership(lock), 0, TimeUnit.MILLISECONDS); |
Madan Jampani | 1d3494e | 2014-11-20 11:24:22 -0800 | [diff] [blame] | 157 | } |
| 158 | } |
| 159 | |
| 160 | @Override |
| 161 | public void withdraw(String path) { |
| 162 | checkArgument(path != null); |
Madan Jampani | d1a467f | 2014-11-24 15:11:09 -0800 | [diff] [blame] | 163 | |
Madan Jampani | 1d3494e | 2014-11-20 11:24:22 -0800 | [diff] [blame] | 164 | Lock lock = openContests.remove(path); |
| 165 | |
| 166 | if (lock != null && lock.isLocked()) { |
| 167 | lock.unlock(); |
| 168 | notifyListeners( |
| 169 | new LeadershipEvent( |
| 170 | LeadershipEvent.Type.LEADER_BOOTED, |
Madan Jampani | 8d21c79 | 2014-12-01 16:31:07 -0800 | [diff] [blame] | 171 | new Leadership(lock.path(), localNodeId, lock.epoch()))); |
Madan Jampani | 1d3494e | 2014-11-20 11:24:22 -0800 | [diff] [blame] | 172 | } |
| 173 | } |
| 174 | |
| 175 | @Override |
Yuta HIGUCHI | c2bf3d8 | 2014-11-28 18:50:41 -0800 | [diff] [blame] | 176 | public Map<String, Leadership> getLeaderBoard() { |
| 177 | return ImmutableMap.copyOf(leaderBoard); |
| 178 | } |
| 179 | |
| 180 | @Override |
Madan Jampani | 1d3494e | 2014-11-20 11:24:22 -0800 | [diff] [blame] | 181 | public void addListener(LeadershipEventListener listener) { |
| 182 | checkArgument(listener != null); |
| 183 | listeners.add(listener); |
| 184 | } |
| 185 | |
| 186 | @Override |
| 187 | public void removeListener(LeadershipEventListener listener) { |
| 188 | checkArgument(listener != null); |
| 189 | listeners.remove(listener); |
| 190 | } |
| 191 | |
| 192 | private void notifyListeners(LeadershipEvent event) { |
| 193 | for (LeadershipEventListener listener : listeners) { |
Madan Jampani | d1a467f | 2014-11-24 15:11:09 -0800 | [diff] [blame] | 194 | try { |
| 195 | listener.event(event); |
| 196 | } catch (Exception e) { |
| 197 | log.error("Notifying listener failed with exception.", e); |
| 198 | } |
Madan Jampani | 1d3494e | 2014-11-20 11:24:22 -0800 | [diff] [blame] | 199 | } |
| 200 | } |
| 201 | |
| 202 | private void tryAcquireLeadership(String path) { |
| 203 | Lock lock = openContests.get(path); |
Madan Jampani | d1a467f | 2014-11-24 15:11:09 -0800 | [diff] [blame] | 204 | if (lock == null) { |
| 205 | // withdrew from race. |
| 206 | return; |
| 207 | } |
Madan Jampani | 1d3494e | 2014-11-20 11:24:22 -0800 | [diff] [blame] | 208 | lock.lockAsync(TERM_DURATION_MS).whenComplete((response, error) -> { |
| 209 | if (error == null) { |
Madan Jampani | ddaffd0 | 2014-11-21 13:12:09 -0800 | [diff] [blame] | 210 | threadPool.schedule( |
Madan Jampani | 1ee9178 | 2014-11-20 20:24:24 -0800 | [diff] [blame] | 211 | new ReelectionTask(lock), |
Madan Jampani | 1d3494e | 2014-11-20 11:24:22 -0800 | [diff] [blame] | 212 | TERM_DURATION_MS / 2, |
| 213 | TimeUnit.MILLISECONDS); |
| 214 | notifyListeners( |
| 215 | new LeadershipEvent( |
| 216 | LeadershipEvent.Type.LEADER_ELECTED, |
Madan Jampani | 8d21c79 | 2014-12-01 16:31:07 -0800 | [diff] [blame] | 217 | new Leadership(lock.path(), localNodeId, lock.epoch()))); |
Madan Jampani | 1ee9178 | 2014-11-20 20:24:24 -0800 | [diff] [blame] | 218 | return; |
Madan Jampani | 1d3494e | 2014-11-20 11:24:22 -0800 | [diff] [blame] | 219 | } else { |
Madan Jampani | d1a467f | 2014-11-24 15:11:09 -0800 | [diff] [blame] | 220 | log.warn("Failed to acquire lock for {}. Will retry in {} ms", path, WAIT_BEFORE_RETRY_MS, error); |
| 221 | threadPool.schedule(new TryLeadership(lock), WAIT_BEFORE_RETRY_MS, TimeUnit.MILLISECONDS); |
Madan Jampani | 1d3494e | 2014-11-20 11:24:22 -0800 | [diff] [blame] | 222 | } |
| 223 | }); |
| 224 | } |
| 225 | |
Madan Jampani | 1ee9178 | 2014-11-20 20:24:24 -0800 | [diff] [blame] | 226 | private class ReelectionTask implements Runnable { |
Madan Jampani | 1d3494e | 2014-11-20 11:24:22 -0800 | [diff] [blame] | 227 | |
| 228 | private final Lock lock; |
| 229 | |
Madan Jampani | 1ee9178 | 2014-11-20 20:24:24 -0800 | [diff] [blame] | 230 | public ReelectionTask(Lock lock) { |
Madan Jampani | 1d3494e | 2014-11-20 11:24:22 -0800 | [diff] [blame] | 231 | this.lock = lock; |
| 232 | } |
| 233 | |
| 234 | @Override |
| 235 | public void run() { |
Madan Jampani | d1a467f | 2014-11-24 15:11:09 -0800 | [diff] [blame] | 236 | if (!openContests.containsKey(lock.path())) { |
| 237 | log.debug("Node withdrew from leadership race for {}. Cancelling reelection task.", lock.path()); |
| 238 | return; |
| 239 | } |
| 240 | |
| 241 | boolean lockExtended = false; |
| 242 | try { |
| 243 | lockExtended = lock.extendExpiration(TERM_DURATION_MS); |
| 244 | } catch (Exception e) { |
| 245 | log.warn("Attempt to extend lock failed with an exception.", e); |
| 246 | } |
| 247 | |
| 248 | if (lockExtended) { |
Madan Jampani | 1d3494e | 2014-11-20 11:24:22 -0800 | [diff] [blame] | 249 | notifyListeners( |
| 250 | new LeadershipEvent( |
| 251 | LeadershipEvent.Type.LEADER_REELECTED, |
Madan Jampani | 8d21c79 | 2014-12-01 16:31:07 -0800 | [diff] [blame] | 252 | new Leadership(lock.path(), localNodeId, lock.epoch()))); |
Madan Jampani | ddaffd0 | 2014-11-21 13:12:09 -0800 | [diff] [blame] | 253 | threadPool.schedule(this, TERM_DURATION_MS / 2, TimeUnit.MILLISECONDS); |
Madan Jampani | 1d3494e | 2014-11-20 11:24:22 -0800 | [diff] [blame] | 254 | } else { |
Madan Jampani | d1a467f | 2014-11-24 15:11:09 -0800 | [diff] [blame] | 255 | // Check if this node already withdrew from the contest, in which case |
| 256 | // we don't need to notify here. |
Madan Jampani | 1d3494e | 2014-11-20 11:24:22 -0800 | [diff] [blame] | 257 | if (openContests.containsKey(lock.path())) { |
| 258 | notifyListeners( |
| 259 | new LeadershipEvent( |
| 260 | LeadershipEvent.Type.LEADER_BOOTED, |
Madan Jampani | 8d21c79 | 2014-12-01 16:31:07 -0800 | [diff] [blame] | 261 | new Leadership(lock.path(), localNodeId, lock.epoch()))); |
Madan Jampani | d1a467f | 2014-11-24 15:11:09 -0800 | [diff] [blame] | 262 | // Retry leadership after a brief wait. |
| 263 | threadPool.schedule(new TryLeadership(lock), WAIT_BEFORE_RETRY_MS, TimeUnit.MILLISECONDS); |
Madan Jampani | 1d3494e | 2014-11-20 11:24:22 -0800 | [diff] [blame] | 264 | } |
| 265 | } |
| 266 | } |
| 267 | } |
Madan Jampani | 1ee9178 | 2014-11-20 20:24:24 -0800 | [diff] [blame] | 268 | |
Madan Jampani | d1a467f | 2014-11-24 15:11:09 -0800 | [diff] [blame] | 269 | private class TryLeadership implements Runnable { |
| 270 | private final Lock lock; |
| 271 | |
| 272 | public TryLeadership(Lock lock) { |
| 273 | this.lock = lock; |
| 274 | } |
| 275 | |
| 276 | @Override |
| 277 | public void run() { |
| 278 | tryAcquireLeadership(lock.path()); |
| 279 | } |
| 280 | } |
| 281 | |
Madan Jampani | 1ee9178 | 2014-11-20 20:24:24 -0800 | [diff] [blame] | 282 | private class PeerAdvertiser implements LeadershipEventListener { |
| 283 | @Override |
| 284 | public void event(LeadershipEvent event) { |
| 285 | // publish events originating on this host. |
Madan Jampani | 8d21c79 | 2014-12-01 16:31:07 -0800 | [diff] [blame] | 286 | if (event.subject().leader().equals(localNodeId)) { |
Jonathan Hart | 7d656f4 | 2015-01-27 14:07:23 -0800 | [diff] [blame^] | 287 | clusterCommunicator.broadcast( |
| 288 | new ClusterMessage( |
| 289 | localNodeId, |
| 290 | LEADERSHIP_UPDATES, |
| 291 | SERIALIZER.encode(event))); |
Madan Jampani | 1ee9178 | 2014-11-20 20:24:24 -0800 | [diff] [blame] | 292 | } |
| 293 | } |
| 294 | } |
| 295 | |
| 296 | private class PeerAdvertisementHandler implements ClusterMessageHandler { |
| 297 | @Override |
| 298 | public void handle(ClusterMessage message) { |
| 299 | LeadershipEvent event = SERIALIZER.decode(message.payload()); |
Yuta HIGUCHI | 47b2f55 | 2014-11-28 20:13:15 -0800 | [diff] [blame] | 300 | log.trace("Received {} from {}", event, message.sender()); |
Madan Jampani | 1ee9178 | 2014-11-20 20:24:24 -0800 | [diff] [blame] | 301 | notifyListeners(event); |
| 302 | } |
| 303 | } |
| 304 | |
| 305 | private class LeaderBoardUpdater implements LeadershipEventListener { |
| 306 | @Override |
| 307 | public void event(LeadershipEvent event) { |
| 308 | Leadership leadershipUpdate = event.subject(); |
| 309 | synchronized (leaderBoard) { |
| 310 | Leadership currentLeadership = leaderBoard.get(leadershipUpdate.topic()); |
| 311 | switch (event.type()) { |
| 312 | case LEADER_ELECTED: |
| 313 | case LEADER_REELECTED: |
| 314 | if (currentLeadership == null || currentLeadership.epoch() < leadershipUpdate.epoch()) { |
| 315 | leaderBoard.put(leadershipUpdate.topic(), leadershipUpdate); |
| 316 | } |
| 317 | break; |
| 318 | case LEADER_BOOTED: |
| 319 | if (currentLeadership != null && currentLeadership.epoch() <= leadershipUpdate.epoch()) { |
| 320 | leaderBoard.remove(leadershipUpdate.topic()); |
| 321 | } |
| 322 | break; |
| 323 | default: |
| 324 | break; |
| 325 | } |
| 326 | } |
| 327 | } |
| 328 | } |
Brian O'Connor | abafb50 | 2014-12-02 22:26:20 -0800 | [diff] [blame] | 329 | } |