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; |
Madan Jampani | 1d3494e | 2014-11-20 11:24:22 -0800 | [diff] [blame] | 42 | import org.slf4j.Logger; |
| 43 | |
Jonathan Hart | 7d656f4 | 2015-01-27 14:07:23 -0800 | [diff] [blame] | 44 | import java.util.Map; |
| 45 | import java.util.Set; |
| 46 | import java.util.concurrent.Executors; |
| 47 | import java.util.concurrent.ScheduledExecutorService; |
| 48 | import java.util.concurrent.TimeUnit; |
| 49 | |
| 50 | import static com.google.common.base.Preconditions.checkArgument; |
| 51 | import static org.onlab.util.Tools.namedThreads; |
| 52 | import static org.slf4j.LoggerFactory.getLogger; |
Madan Jampani | 1d3494e | 2014-11-20 11:24:22 -0800 | [diff] [blame] | 53 | |
| 54 | /** |
| 55 | * Distributed implementation of LeadershipService that is based on the primitives exposed by |
| 56 | * LockService. |
| 57 | */ |
Jonathan Hart | 949c284 | 2014-11-28 23:44:09 -0800 | [diff] [blame] | 58 | @Component(enabled = false) |
Madan Jampani | 1d3494e | 2014-11-20 11:24:22 -0800 | [diff] [blame] | 59 | @Service |
| 60 | public class LeadershipManager implements LeadershipService { |
| 61 | |
| 62 | private final Logger log = getLogger(getClass()); |
| 63 | |
Madan Jampani | 94c2353 | 2015-02-05 17:40:01 -0800 | [diff] [blame] | 64 | private static final int TERM_DURATION_MS = 2000; |
Madan Jampani | 1d3494e | 2014-11-20 11:24:22 -0800 | [diff] [blame] | 65 | |
Madan Jampani | 1ee9178 | 2014-11-20 20:24:24 -0800 | [diff] [blame] | 66 | // Time to wait before retrying leadership after |
| 67 | // a unexpected error. |
| 68 | private static final int WAIT_BEFORE_RETRY_MS = 2000; |
| 69 | |
Yuta HIGUCHI | 6593489 | 2014-12-04 17:47:44 -0800 | [diff] [blame] | 70 | // TODO: Make Thread pool size configurable. |
Madan Jampani | ddaffd0 | 2014-11-21 13:12:09 -0800 | [diff] [blame] | 71 | private final ScheduledExecutorService threadPool = |
Thomas Vachuska | 9ea3e6f | 2015-01-23 16:34:22 -0800 | [diff] [blame] | 72 | Executors.newScheduledThreadPool(25, namedThreads("onos-leadership-manager-%d")); |
Madan Jampani | 1d3494e | 2014-11-20 11:24:22 -0800 | [diff] [blame] | 73 | |
Madan Jampani | 1ee9178 | 2014-11-20 20:24:24 -0800 | [diff] [blame] | 74 | private static final MessageSubject LEADERSHIP_UPDATES = |
| 75 | new MessageSubject("leadership-contest-updates"); |
| 76 | |
Madan Jampani | 1d3494e | 2014-11-20 11:24:22 -0800 | [diff] [blame] | 77 | @Reference(cardinality = ReferenceCardinality.MANDATORY_UNARY) |
| 78 | private ClusterService clusterService; |
| 79 | |
| 80 | @Reference(cardinality = ReferenceCardinality.MANDATORY_UNARY) |
Madan Jampani | 1ee9178 | 2014-11-20 20:24:24 -0800 | [diff] [blame] | 81 | private ClusterCommunicationService clusterCommunicator; |
| 82 | |
| 83 | @Reference(cardinality = ReferenceCardinality.MANDATORY_UNARY) |
Madan Jampani | 1d3494e | 2014-11-20 11:24:22 -0800 | [diff] [blame] | 84 | private LockService lockService; |
| 85 | |
Madan Jampani | 1ee9178 | 2014-11-20 20:24:24 -0800 | [diff] [blame] | 86 | private final Map<String, Leadership> leaderBoard = Maps.newHashMap(); |
| 87 | |
Madan Jampani | d1a467f | 2014-11-24 15:11:09 -0800 | [diff] [blame] | 88 | private final Map<String, Lock> openContests = Maps.newConcurrentMap(); |
Madan Jampani | 1ee9178 | 2014-11-20 20:24:24 -0800 | [diff] [blame] | 89 | private final Set<LeadershipEventListener> listeners = Sets.newIdentityHashSet(); |
Madan Jampani | 8d21c79 | 2014-12-01 16:31:07 -0800 | [diff] [blame] | 90 | private NodeId localNodeId; |
Madan Jampani | 1d3494e | 2014-11-20 11:24:22 -0800 | [diff] [blame] | 91 | |
Madan Jampani | 1ee9178 | 2014-11-20 20:24:24 -0800 | [diff] [blame] | 92 | private final LeadershipEventListener peerAdvertiser = new PeerAdvertiser(); |
| 93 | private final LeadershipEventListener leaderBoardUpdater = new LeaderBoardUpdater(); |
| 94 | |
| 95 | public static final KryoSerializer SERIALIZER = new KryoSerializer() { |
| 96 | @Override |
| 97 | protected void setupKryoPool() { |
| 98 | serializerPool = KryoNamespace.newBuilder() |
| 99 | .register(KryoNamespaces.API) |
| 100 | .build() |
| 101 | .populate(1); |
| 102 | } |
| 103 | }; |
| 104 | |
Madan Jampani | 1d3494e | 2014-11-20 11:24:22 -0800 | [diff] [blame] | 105 | @Activate |
| 106 | public void activate() { |
Madan Jampani | 8d21c79 | 2014-12-01 16:31:07 -0800 | [diff] [blame] | 107 | localNodeId = clusterService.getLocalNode().id(); |
Madan Jampani | 1ee9178 | 2014-11-20 20:24:24 -0800 | [diff] [blame] | 108 | |
| 109 | addListener(peerAdvertiser); |
| 110 | addListener(leaderBoardUpdater); |
| 111 | |
| 112 | clusterCommunicator.addSubscriber( |
| 113 | LEADERSHIP_UPDATES, |
| 114 | new PeerAdvertisementHandler()); |
| 115 | |
Madan Jampani | 1d3494e | 2014-11-20 11:24:22 -0800 | [diff] [blame] | 116 | log.info("Started."); |
| 117 | } |
| 118 | |
| 119 | @Deactivate |
| 120 | public void deactivate() { |
Madan Jampani | 1ee9178 | 2014-11-20 20:24:24 -0800 | [diff] [blame] | 121 | removeListener(peerAdvertiser); |
| 122 | removeListener(leaderBoardUpdater); |
| 123 | |
| 124 | clusterCommunicator.removeSubscriber(LEADERSHIP_UPDATES); |
| 125 | |
Madan Jampani | ddaffd0 | 2014-11-21 13:12:09 -0800 | [diff] [blame] | 126 | threadPool.shutdown(); |
Madan Jampani | 1ee9178 | 2014-11-20 20:24:24 -0800 | [diff] [blame] | 127 | |
Madan Jampani | 1d3494e | 2014-11-20 11:24:22 -0800 | [diff] [blame] | 128 | log.info("Stopped."); |
| 129 | } |
| 130 | |
Madan Jampani | 1ee9178 | 2014-11-20 20:24:24 -0800 | [diff] [blame] | 131 | |
| 132 | @Override |
Madan Jampani | 8d21c79 | 2014-12-01 16:31:07 -0800 | [diff] [blame] | 133 | public NodeId getLeader(String path) { |
Madan Jampani | 1ee9178 | 2014-11-20 20:24:24 -0800 | [diff] [blame] | 134 | synchronized (leaderBoard) { |
| 135 | Leadership leadership = leaderBoard.get(path); |
| 136 | if (leadership != null) { |
| 137 | return leadership.leader(); |
| 138 | } |
| 139 | } |
| 140 | return null; |
| 141 | } |
| 142 | |
Madan Jampani | 1d3494e | 2014-11-20 11:24:22 -0800 | [diff] [blame] | 143 | @Override |
| 144 | public void runForLeadership(String path) { |
| 145 | checkArgument(path != null); |
Madan Jampani | d1a467f | 2014-11-24 15:11:09 -0800 | [diff] [blame] | 146 | |
Madan Jampani | 1d3494e | 2014-11-20 11:24:22 -0800 | [diff] [blame] | 147 | if (openContests.containsKey(path)) { |
| 148 | log.info("Already in the leadership contest for {}", path); |
| 149 | return; |
| 150 | } else { |
| 151 | Lock lock = lockService.create(path); |
| 152 | openContests.put(path, lock); |
Madan Jampani | d1a467f | 2014-11-24 15:11:09 -0800 | [diff] [blame] | 153 | threadPool.schedule(new TryLeadership(lock), 0, TimeUnit.MILLISECONDS); |
Madan Jampani | 1d3494e | 2014-11-20 11:24:22 -0800 | [diff] [blame] | 154 | } |
| 155 | } |
| 156 | |
| 157 | @Override |
| 158 | public void withdraw(String path) { |
| 159 | checkArgument(path != null); |
Madan Jampani | d1a467f | 2014-11-24 15:11:09 -0800 | [diff] [blame] | 160 | |
Madan Jampani | 1d3494e | 2014-11-20 11:24:22 -0800 | [diff] [blame] | 161 | Lock lock = openContests.remove(path); |
| 162 | |
| 163 | if (lock != null && lock.isLocked()) { |
| 164 | lock.unlock(); |
| 165 | notifyListeners( |
| 166 | new LeadershipEvent( |
| 167 | LeadershipEvent.Type.LEADER_BOOTED, |
Madan Jampani | 8d21c79 | 2014-12-01 16:31:07 -0800 | [diff] [blame] | 168 | new Leadership(lock.path(), localNodeId, lock.epoch()))); |
Madan Jampani | 1d3494e | 2014-11-20 11:24:22 -0800 | [diff] [blame] | 169 | } |
| 170 | } |
| 171 | |
| 172 | @Override |
Yuta HIGUCHI | c2bf3d8 | 2014-11-28 18:50:41 -0800 | [diff] [blame] | 173 | public Map<String, Leadership> getLeaderBoard() { |
| 174 | return ImmutableMap.copyOf(leaderBoard); |
| 175 | } |
| 176 | |
| 177 | @Override |
Madan Jampani | 1d3494e | 2014-11-20 11:24:22 -0800 | [diff] [blame] | 178 | public void addListener(LeadershipEventListener listener) { |
| 179 | checkArgument(listener != null); |
| 180 | listeners.add(listener); |
| 181 | } |
| 182 | |
| 183 | @Override |
| 184 | public void removeListener(LeadershipEventListener listener) { |
| 185 | checkArgument(listener != null); |
| 186 | listeners.remove(listener); |
| 187 | } |
| 188 | |
| 189 | private void notifyListeners(LeadershipEvent event) { |
| 190 | for (LeadershipEventListener listener : listeners) { |
Madan Jampani | d1a467f | 2014-11-24 15:11:09 -0800 | [diff] [blame] | 191 | try { |
| 192 | listener.event(event); |
| 193 | } catch (Exception e) { |
| 194 | log.error("Notifying listener failed with exception.", e); |
| 195 | } |
Madan Jampani | 1d3494e | 2014-11-20 11:24:22 -0800 | [diff] [blame] | 196 | } |
| 197 | } |
| 198 | |
| 199 | private void tryAcquireLeadership(String path) { |
| 200 | Lock lock = openContests.get(path); |
Madan Jampani | d1a467f | 2014-11-24 15:11:09 -0800 | [diff] [blame] | 201 | if (lock == null) { |
| 202 | // withdrew from race. |
| 203 | return; |
| 204 | } |
Madan Jampani | 1d3494e | 2014-11-20 11:24:22 -0800 | [diff] [blame] | 205 | lock.lockAsync(TERM_DURATION_MS).whenComplete((response, error) -> { |
| 206 | if (error == null) { |
Madan Jampani | ddaffd0 | 2014-11-21 13:12:09 -0800 | [diff] [blame] | 207 | threadPool.schedule( |
Madan Jampani | 1ee9178 | 2014-11-20 20:24:24 -0800 | [diff] [blame] | 208 | new ReelectionTask(lock), |
Madan Jampani | 1d3494e | 2014-11-20 11:24:22 -0800 | [diff] [blame] | 209 | TERM_DURATION_MS / 2, |
| 210 | TimeUnit.MILLISECONDS); |
| 211 | notifyListeners( |
| 212 | new LeadershipEvent( |
| 213 | LeadershipEvent.Type.LEADER_ELECTED, |
Madan Jampani | 8d21c79 | 2014-12-01 16:31:07 -0800 | [diff] [blame] | 214 | new Leadership(lock.path(), localNodeId, lock.epoch()))); |
Madan Jampani | 1ee9178 | 2014-11-20 20:24:24 -0800 | [diff] [blame] | 215 | return; |
Madan Jampani | 1d3494e | 2014-11-20 11:24:22 -0800 | [diff] [blame] | 216 | } else { |
Madan Jampani | d1a467f | 2014-11-24 15:11:09 -0800 | [diff] [blame] | 217 | log.warn("Failed to acquire lock for {}. Will retry in {} ms", path, WAIT_BEFORE_RETRY_MS, error); |
| 218 | threadPool.schedule(new TryLeadership(lock), WAIT_BEFORE_RETRY_MS, TimeUnit.MILLISECONDS); |
Madan Jampani | 1d3494e | 2014-11-20 11:24:22 -0800 | [diff] [blame] | 219 | } |
| 220 | }); |
| 221 | } |
| 222 | |
Madan Jampani | 1ee9178 | 2014-11-20 20:24:24 -0800 | [diff] [blame] | 223 | private class ReelectionTask implements Runnable { |
Madan Jampani | 1d3494e | 2014-11-20 11:24:22 -0800 | [diff] [blame] | 224 | |
| 225 | private final Lock lock; |
| 226 | |
Madan Jampani | 1ee9178 | 2014-11-20 20:24:24 -0800 | [diff] [blame] | 227 | public ReelectionTask(Lock lock) { |
Madan Jampani | 1d3494e | 2014-11-20 11:24:22 -0800 | [diff] [blame] | 228 | this.lock = lock; |
| 229 | } |
| 230 | |
| 231 | @Override |
| 232 | public void run() { |
Madan Jampani | d1a467f | 2014-11-24 15:11:09 -0800 | [diff] [blame] | 233 | if (!openContests.containsKey(lock.path())) { |
| 234 | log.debug("Node withdrew from leadership race for {}. Cancelling reelection task.", lock.path()); |
| 235 | return; |
| 236 | } |
| 237 | |
| 238 | boolean lockExtended = false; |
| 239 | try { |
| 240 | lockExtended = lock.extendExpiration(TERM_DURATION_MS); |
| 241 | } catch (Exception e) { |
| 242 | log.warn("Attempt to extend lock failed with an exception.", e); |
| 243 | } |
| 244 | |
| 245 | if (lockExtended) { |
Madan Jampani | 1d3494e | 2014-11-20 11:24:22 -0800 | [diff] [blame] | 246 | notifyListeners( |
| 247 | new LeadershipEvent( |
| 248 | LeadershipEvent.Type.LEADER_REELECTED, |
Madan Jampani | 8d21c79 | 2014-12-01 16:31:07 -0800 | [diff] [blame] | 249 | new Leadership(lock.path(), localNodeId, lock.epoch()))); |
Madan Jampani | ddaffd0 | 2014-11-21 13:12:09 -0800 | [diff] [blame] | 250 | threadPool.schedule(this, TERM_DURATION_MS / 2, TimeUnit.MILLISECONDS); |
Madan Jampani | 1d3494e | 2014-11-20 11:24:22 -0800 | [diff] [blame] | 251 | } else { |
Madan Jampani | d1a467f | 2014-11-24 15:11:09 -0800 | [diff] [blame] | 252 | // Check if this node already withdrew from the contest, in which case |
| 253 | // we don't need to notify here. |
Madan Jampani | 1d3494e | 2014-11-20 11:24:22 -0800 | [diff] [blame] | 254 | if (openContests.containsKey(lock.path())) { |
| 255 | notifyListeners( |
| 256 | new LeadershipEvent( |
| 257 | LeadershipEvent.Type.LEADER_BOOTED, |
Madan Jampani | 8d21c79 | 2014-12-01 16:31:07 -0800 | [diff] [blame] | 258 | new Leadership(lock.path(), localNodeId, lock.epoch()))); |
Madan Jampani | d1a467f | 2014-11-24 15:11:09 -0800 | [diff] [blame] | 259 | // Retry leadership after a brief wait. |
| 260 | threadPool.schedule(new TryLeadership(lock), WAIT_BEFORE_RETRY_MS, TimeUnit.MILLISECONDS); |
Madan Jampani | 1d3494e | 2014-11-20 11:24:22 -0800 | [diff] [blame] | 261 | } |
| 262 | } |
| 263 | } |
| 264 | } |
Madan Jampani | 1ee9178 | 2014-11-20 20:24:24 -0800 | [diff] [blame] | 265 | |
Madan Jampani | d1a467f | 2014-11-24 15:11:09 -0800 | [diff] [blame] | 266 | private class TryLeadership implements Runnable { |
| 267 | private final Lock lock; |
| 268 | |
| 269 | public TryLeadership(Lock lock) { |
| 270 | this.lock = lock; |
| 271 | } |
| 272 | |
| 273 | @Override |
| 274 | public void run() { |
| 275 | tryAcquireLeadership(lock.path()); |
| 276 | } |
| 277 | } |
| 278 | |
Madan Jampani | 1ee9178 | 2014-11-20 20:24:24 -0800 | [diff] [blame] | 279 | private class PeerAdvertiser implements LeadershipEventListener { |
| 280 | @Override |
| 281 | public void event(LeadershipEvent event) { |
| 282 | // publish events originating on this host. |
Madan Jampani | 8d21c79 | 2014-12-01 16:31:07 -0800 | [diff] [blame] | 283 | if (event.subject().leader().equals(localNodeId)) { |
Jonathan Hart | 7d656f4 | 2015-01-27 14:07:23 -0800 | [diff] [blame] | 284 | clusterCommunicator.broadcast( |
| 285 | new ClusterMessage( |
| 286 | localNodeId, |
| 287 | LEADERSHIP_UPDATES, |
| 288 | SERIALIZER.encode(event))); |
Madan Jampani | 1ee9178 | 2014-11-20 20:24:24 -0800 | [diff] [blame] | 289 | } |
| 290 | } |
| 291 | } |
| 292 | |
| 293 | private class PeerAdvertisementHandler implements ClusterMessageHandler { |
| 294 | @Override |
| 295 | public void handle(ClusterMessage message) { |
| 296 | LeadershipEvent event = SERIALIZER.decode(message.payload()); |
Yuta HIGUCHI | 47b2f55 | 2014-11-28 20:13:15 -0800 | [diff] [blame] | 297 | log.trace("Received {} from {}", event, message.sender()); |
Madan Jampani | 1ee9178 | 2014-11-20 20:24:24 -0800 | [diff] [blame] | 298 | notifyListeners(event); |
| 299 | } |
| 300 | } |
| 301 | |
| 302 | private class LeaderBoardUpdater implements LeadershipEventListener { |
| 303 | @Override |
| 304 | public void event(LeadershipEvent event) { |
| 305 | Leadership leadershipUpdate = event.subject(); |
| 306 | synchronized (leaderBoard) { |
| 307 | Leadership currentLeadership = leaderBoard.get(leadershipUpdate.topic()); |
| 308 | switch (event.type()) { |
| 309 | case LEADER_ELECTED: |
| 310 | case LEADER_REELECTED: |
| 311 | if (currentLeadership == null || currentLeadership.epoch() < leadershipUpdate.epoch()) { |
| 312 | leaderBoard.put(leadershipUpdate.topic(), leadershipUpdate); |
| 313 | } |
| 314 | break; |
| 315 | case LEADER_BOOTED: |
| 316 | if (currentLeadership != null && currentLeadership.epoch() <= leadershipUpdate.epoch()) { |
| 317 | leaderBoard.remove(leadershipUpdate.topic()); |
| 318 | } |
| 319 | break; |
| 320 | default: |
| 321 | break; |
| 322 | } |
| 323 | } |
| 324 | } |
| 325 | } |
Brian O'Connor | abafb50 | 2014-12-02 22:26:20 -0800 | [diff] [blame] | 326 | } |