blob: 862b3cbecbfd74c1c4ae18deb3e9e280a81d090c [file] [log] [blame]
Madan Jampani84b6b402015-02-25 17:49:54 -08001/*
2 * Copyright 2015 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 */
16package org.onosproject.store.mastership.impl;
17
18import static org.onlab.util.Tools.groupedThreads;
Ayaka Koshibe941f8602015-04-15 14:17:08 -070019import static org.onlab.util.Tools.futureGetOrElse;
Madan Jampani84b6b402015-02-25 17:49:54 -080020import static org.onosproject.mastership.MastershipEvent.Type.BACKUPS_CHANGED;
21import static org.onosproject.mastership.MastershipEvent.Type.MASTER_CHANGED;
22import static org.slf4j.LoggerFactory.getLogger;
23import static com.google.common.base.Preconditions.checkArgument;
24
Madan Jampani84b6b402015-02-25 17:49:54 -080025import java.util.List;
26import java.util.Map;
27import java.util.Set;
Madan Jampanif7536ab2015-05-07 23:23:23 -070028import java.util.concurrent.CompletableFuture;
Madan Jampani84b6b402015-02-25 17:49:54 -080029import java.util.concurrent.ExecutorService;
30import java.util.concurrent.Executors;
Madan Jampanif7536ab2015-05-07 23:23:23 -070031import java.util.concurrent.ScheduledExecutorService;
32import java.util.concurrent.TimeUnit;
Madan Jampani84b6b402015-02-25 17:49:54 -080033import java.util.regex.Matcher;
34import java.util.regex.Pattern;
35import java.util.stream.Collectors;
36
37import org.apache.felix.scr.annotations.Activate;
38import org.apache.felix.scr.annotations.Component;
39import org.apache.felix.scr.annotations.Deactivate;
40import org.apache.felix.scr.annotations.Reference;
41import org.apache.felix.scr.annotations.ReferenceCardinality;
42import org.apache.felix.scr.annotations.Service;
43import org.onlab.util.KryoNamespace;
44import org.onosproject.cluster.ClusterService;
45import org.onosproject.cluster.Leadership;
46import org.onosproject.cluster.LeadershipEvent;
47import org.onosproject.cluster.LeadershipEventListener;
48import org.onosproject.cluster.LeadershipService;
49import org.onosproject.cluster.NodeId;
50import org.onosproject.cluster.RoleInfo;
51import org.onosproject.mastership.MastershipEvent;
52import org.onosproject.mastership.MastershipStore;
53import org.onosproject.mastership.MastershipStoreDelegate;
54import org.onosproject.mastership.MastershipTerm;
55import org.onosproject.net.DeviceId;
56import org.onosproject.net.MastershipRole;
57import org.onosproject.store.AbstractStore;
58import org.onosproject.store.cluster.messaging.ClusterCommunicationService;
Madan Jampani84b6b402015-02-25 17:49:54 -080059import org.onosproject.store.cluster.messaging.MessageSubject;
60import org.onosproject.store.serializers.KryoNamespaces;
61import org.onosproject.store.serializers.KryoSerializer;
62import org.onosproject.store.serializers.StoreSerializer;
63import org.slf4j.Logger;
64
65import com.google.common.base.Objects;
66import com.google.common.collect.Lists;
67import com.google.common.collect.Maps;
68import com.google.common.collect.Sets;
69
70/**
71 * Implementation of the MastershipStore on top of Leadership Service.
72 */
Madan Jampani5756c352015-04-29 00:23:58 -070073@Component(immediate = true, enabled = true)
Madan Jampani84b6b402015-02-25 17:49:54 -080074@Service
75public class ConsistentDeviceMastershipStore
76 extends AbstractStore<MastershipEvent, MastershipStoreDelegate>
77 implements MastershipStore {
78
79 private final Logger log = getLogger(getClass());
80
81 @Reference(cardinality = ReferenceCardinality.MANDATORY_UNARY)
82 protected LeadershipService leadershipService;
83
84 @Reference(cardinality = ReferenceCardinality.MANDATORY_UNARY)
85 protected ClusterService clusterService;
86
87 @Reference(cardinality = ReferenceCardinality.MANDATORY_UNARY)
88 protected ClusterCommunicationService clusterCommunicator;
89
90 private NodeId localNodeId;
91 private final Set<DeviceId> connectedDevices = Sets.newHashSet();
92
93 private static final MessageSubject ROLE_QUERY_SUBJECT =
94 new MessageSubject("mastership-store-device-role-query");
95 private static final MessageSubject ROLE_RELINQUISH_SUBJECT =
96 new MessageSubject("mastership-store-device-role-relinquish");
Madan Jampani9bd1f152015-04-30 23:33:35 -070097 private static final MessageSubject TRANSITION_FROM_MASTER_TO_STANDBY_SUBJECT =
Madan Jampani1af8e132015-04-30 16:41:18 -070098 new MessageSubject("mastership-store-device-mastership-relinquish");
Madan Jampani84b6b402015-02-25 17:49:54 -080099
100 private static final Pattern DEVICE_MASTERSHIP_TOPIC_PATTERN =
Madan Jampani5756c352015-04-29 00:23:58 -0700101 Pattern.compile("device:(.*)");
Madan Jampani84b6b402015-02-25 17:49:54 -0800102
Madan Jampani84b6b402015-02-25 17:49:54 -0800103 private ExecutorService messageHandlingExecutor;
Madan Jampanif7536ab2015-05-07 23:23:23 -0700104 private ScheduledExecutorService transferExecutor;
Madan Jampani84b6b402015-02-25 17:49:54 -0800105 private final LeadershipEventListener leadershipEventListener =
106 new InternalDeviceMastershipEventListener();
107
108 private static final String NODE_ID_NULL = "Node ID cannot be null";
Madan Jampanif7536ab2015-05-07 23:23:23 -0700109 private static final String DEVICE_ID_NULL = "Device ID cannot be null";
110 private static final int WAIT_BEFORE_MASTERSHIP_HANDOFF_MILLIS = 3000;
Madan Jampani84b6b402015-02-25 17:49:54 -0800111
112 public static final StoreSerializer SERIALIZER = new KryoSerializer() {
113 @Override
114 protected void setupKryoPool() {
115 serializerPool = KryoNamespace.newBuilder()
116 .register(KryoNamespaces.API)
117 .register(MastershipRole.class)
118 .register(MastershipEvent.class)
Madan Jampani1af8e132015-04-30 16:41:18 -0700119 .register(MastershipEvent.Type.class)
Madan Jampani84b6b402015-02-25 17:49:54 -0800120 .build();
121 }
122 };
123
124 @Activate
125 public void activate() {
126 messageHandlingExecutor =
Madan Jampanif7536ab2015-05-07 23:23:23 -0700127 Executors.newSingleThreadExecutor(
128 groupedThreads("onos/store/device/mastership", "message-handler"));
129 transferExecutor =
130 Executors.newSingleThreadScheduledExecutor(
131 groupedThreads("onos/store/device/mastership", "mastership-transfer-executor"));
Madan Jampanid46e18f2015-05-04 23:19:33 -0700132 clusterCommunicator.<DeviceId, MastershipRole>addSubscriber(ROLE_QUERY_SUBJECT,
133 SERIALIZER::decode,
134 deviceId -> getRole(localNodeId, deviceId),
135 SERIALIZER::encode,
Madan Jampani84b6b402015-02-25 17:49:54 -0800136 messageHandlingExecutor);
Madan Jampanid46e18f2015-05-04 23:19:33 -0700137 clusterCommunicator.<DeviceId, MastershipEvent>addSubscriber(ROLE_RELINQUISH_SUBJECT,
138 SERIALIZER::decode,
Madan Jampanif7536ab2015-05-07 23:23:23 -0700139 this::relinquishLocalRole,
Madan Jampanid46e18f2015-05-04 23:19:33 -0700140 SERIALIZER::encode,
141 messageHandlingExecutor);
Madan Jampani9bd1f152015-04-30 23:33:35 -0700142 clusterCommunicator.addSubscriber(TRANSITION_FROM_MASTER_TO_STANDBY_SUBJECT,
Madan Jampani1af8e132015-04-30 16:41:18 -0700143 SERIALIZER::decode,
Madan Jampani9bd1f152015-04-30 23:33:35 -0700144 this::transitionFromMasterToStandby,
Madan Jampani1af8e132015-04-30 16:41:18 -0700145 SERIALIZER::encode,
146 messageHandlingExecutor);
Madan Jampani84b6b402015-02-25 17:49:54 -0800147 localNodeId = clusterService.getLocalNode().id();
148 leadershipService.addListener(leadershipEventListener);
149
150 log.info("Started.");
151 }
152
153 @Deactivate
154 public void deactivate() {
155 clusterCommunicator.removeSubscriber(ROLE_QUERY_SUBJECT);
156 clusterCommunicator.removeSubscriber(ROLE_RELINQUISH_SUBJECT);
Madan Jampani9bd1f152015-04-30 23:33:35 -0700157 clusterCommunicator.removeSubscriber(TRANSITION_FROM_MASTER_TO_STANDBY_SUBJECT);
Madan Jampani84b6b402015-02-25 17:49:54 -0800158 messageHandlingExecutor.shutdown();
Madan Jampanif7536ab2015-05-07 23:23:23 -0700159 transferExecutor.shutdown();
Madan Jampani84b6b402015-02-25 17:49:54 -0800160 leadershipService.removeListener(leadershipEventListener);
161
162 log.info("Stoppped.");
163 }
164
165 @Override
166 public MastershipRole requestRole(DeviceId deviceId) {
167 checkArgument(deviceId != null, DEVICE_ID_NULL);
168
169 String leadershipTopic = createDeviceMastershipTopic(deviceId);
170 if (connectedDevices.add(deviceId)) {
171 leadershipService.runForLeadership(leadershipTopic);
172 return MastershipRole.STANDBY;
173 } else {
174 Leadership leadership = leadershipService.getLeadership(leadershipTopic);
175 if (leadership != null && leadership.leader().equals(localNodeId)) {
176 return MastershipRole.MASTER;
177 } else {
178 return MastershipRole.STANDBY;
179 }
180 }
181 }
182
183 @Override
184 public MastershipRole getRole(NodeId nodeId, DeviceId deviceId) {
185 checkArgument(nodeId != null, NODE_ID_NULL);
186 checkArgument(deviceId != null, DEVICE_ID_NULL);
187
188 String leadershipTopic = createDeviceMastershipTopic(deviceId);
189 Leadership leadership = leadershipService.getLeadership(leadershipTopic);
190 if (leadership != null && nodeId.equals(leadership.leader())) {
191 return MastershipRole.MASTER;
192 }
193
194 if (localNodeId.equals(nodeId)) {
195 if (connectedDevices.contains(deviceId)) {
196 return MastershipRole.STANDBY;
197 } else {
198 return MastershipRole.NONE;
199 }
Madan Jampani84b6b402015-02-25 17:49:54 -0800200 }
Ayaka Koshibe941f8602015-04-15 14:17:08 -0700201 MastershipRole role = futureGetOrElse(clusterCommunicator.sendAndReceive(
Madan Jampani2bfa94c2015-04-11 05:03:49 -0700202 deviceId,
203 ROLE_QUERY_SUBJECT,
204 SERIALIZER::encode,
205 SERIALIZER::decode,
Ayaka Koshibe941f8602015-04-15 14:17:08 -0700206 nodeId), null);
Madan Jampani2bfa94c2015-04-11 05:03:49 -0700207 return role == null ? MastershipRole.NONE : role;
Madan Jampani84b6b402015-02-25 17:49:54 -0800208 }
209
210 @Override
211 public NodeId getMaster(DeviceId deviceId) {
212 checkArgument(deviceId != null, DEVICE_ID_NULL);
213
214 String leadershipTopic = createDeviceMastershipTopic(deviceId);
215 Leadership leadership = leadershipService.getLeadership(leadershipTopic);
216 return leadership != null ? leadership.leader() : null;
217 }
218
219 @Override
220 public RoleInfo getNodes(DeviceId deviceId) {
221 checkArgument(deviceId != null, DEVICE_ID_NULL);
222
223 Map<NodeId, MastershipRole> roles = Maps.newHashMap();
224 clusterService
225 .getNodes()
Madan Jampani84b6b402015-02-25 17:49:54 -0800226 .forEach((node) -> roles.put(node.id(), getRole(node.id(), deviceId)));
227
228 NodeId master = null;
229 final List<NodeId> standbys = Lists.newLinkedList();
230
Madan Jampani86940d92015-05-06 11:47:57 -0700231 List<NodeId> candidates = leadershipService.getCandidates(createDeviceMastershipTopic(deviceId));
232
Madan Jampani84b6b402015-02-25 17:49:54 -0800233 for (Map.Entry<NodeId, MastershipRole> entry : roles.entrySet()) {
234 if (entry.getValue() == MastershipRole.MASTER) {
235 master = entry.getKey();
236 } else if (entry.getValue() == MastershipRole.STANDBY) {
237 standbys.add(entry.getKey());
238 }
239 }
240
Madan Jampani86940d92015-05-06 11:47:57 -0700241 List<NodeId> sortedStandbyList = candidates.stream().filter(standbys::contains).collect(Collectors.toList());
242
243 return new RoleInfo(master, sortedStandbyList);
Madan Jampani84b6b402015-02-25 17:49:54 -0800244 }
245
246 @Override
247 public Set<DeviceId> getDevices(NodeId nodeId) {
248 checkArgument(nodeId != null, NODE_ID_NULL);
249
250 return leadershipService
251 .ownedTopics(nodeId)
252 .stream()
253 .filter(this::isDeviceMastershipTopic)
254 .map(this::extractDeviceIdFromTopic)
255 .collect(Collectors.toSet());
256 }
257
258 @Override
Madan Jampanif7536ab2015-05-07 23:23:23 -0700259 public CompletableFuture<MastershipEvent> setMaster(NodeId nodeId, DeviceId deviceId) {
Madan Jampani84b6b402015-02-25 17:49:54 -0800260 checkArgument(nodeId != null, NODE_ID_NULL);
261 checkArgument(deviceId != null, DEVICE_ID_NULL);
262
Madan Jampani1af8e132015-04-30 16:41:18 -0700263 NodeId currentMaster = getMaster(deviceId);
264 if (nodeId.equals(currentMaster)) {
Madan Jampanif7536ab2015-05-07 23:23:23 -0700265 return CompletableFuture.completedFuture(null);
Madan Jampani1af8e132015-04-30 16:41:18 -0700266 } else {
267 String leadershipTopic = createDeviceMastershipTopic(deviceId);
268 List<NodeId> candidates = leadershipService.getCandidates(leadershipTopic);
269 if (candidates.isEmpty()) {
Madan Jampanif7536ab2015-05-07 23:23:23 -0700270 return CompletableFuture.completedFuture(null);
Madan Jampani1af8e132015-04-30 16:41:18 -0700271 }
272 if (leadershipService.makeTopCandidate(leadershipTopic, nodeId)) {
Madan Jampanif7536ab2015-05-07 23:23:23 -0700273 CompletableFuture<MastershipEvent> result = new CompletableFuture<>();
274 // There is brief wait before we step down from mastership.
275 // This is to ensure any work that happens when standby preference
276 // order changes can complete. For example: flow entries need to be backed
277 // to the new top standby (ONOS-1883)
278 // FIXME: This potentially introduces a race-condition.
279 // Right now role changes are only forced via CLI.
280 transferExecutor.schedule(() -> {
281 result.complete(transitionFromMasterToStandby(deviceId));
282 }, WAIT_BEFORE_MASTERSHIP_HANDOFF_MILLIS, TimeUnit.MILLISECONDS);
283 return result;
Madan Jampani1af8e132015-04-30 16:41:18 -0700284 } else {
285 log.warn("Failed to promote {} to mastership for {}", nodeId, deviceId);
286 }
287 }
Madan Jampanif7536ab2015-05-07 23:23:23 -0700288 return CompletableFuture.completedFuture(null);
Madan Jampani84b6b402015-02-25 17:49:54 -0800289 }
290
291 @Override
292 public MastershipTerm getTermFor(DeviceId deviceId) {
293 checkArgument(deviceId != null, DEVICE_ID_NULL);
294
295 String leadershipTopic = createDeviceMastershipTopic(deviceId);
296 Leadership leadership = leadershipService.getLeadership(leadershipTopic);
297 return leadership != null ? MastershipTerm.of(leadership.leader(), leadership.epoch()) : null;
298 }
299
300 @Override
Madan Jampanif7536ab2015-05-07 23:23:23 -0700301 public CompletableFuture<MastershipEvent> setStandby(NodeId nodeId, DeviceId deviceId) {
Madan Jampani84b6b402015-02-25 17:49:54 -0800302 checkArgument(nodeId != null, NODE_ID_NULL);
303 checkArgument(deviceId != null, DEVICE_ID_NULL);
304
Madan Jampani1af8e132015-04-30 16:41:18 -0700305 NodeId currentMaster = getMaster(deviceId);
306 if (!nodeId.equals(currentMaster)) {
Madan Jampanif7536ab2015-05-07 23:23:23 -0700307 return CompletableFuture.completedFuture(null);
Madan Jampani1af8e132015-04-30 16:41:18 -0700308 }
Madan Jampanid46e18f2015-05-04 23:19:33 -0700309
310 String leadershipTopic = createDeviceMastershipTopic(deviceId);
311 List<NodeId> candidates = leadershipService.getCandidates(leadershipTopic);
312
313 NodeId newMaster = candidates.stream()
314 .filter(candidate -> !Objects.equal(nodeId, candidate))
315 .findFirst()
316 .orElse(null);
317 log.info("Transitioning to role {} for {}. Next master: {}",
318 newMaster != null ? MastershipRole.STANDBY : MastershipRole.NONE, deviceId, newMaster);
319
320 if (newMaster != null) {
321 return setMaster(newMaster, deviceId);
322 }
323 return relinquishRole(nodeId, deviceId);
Madan Jampani84b6b402015-02-25 17:49:54 -0800324 }
325
326 @Override
Madan Jampanif7536ab2015-05-07 23:23:23 -0700327 public CompletableFuture<MastershipEvent> relinquishRole(NodeId nodeId, DeviceId deviceId) {
Madan Jampani84b6b402015-02-25 17:49:54 -0800328 checkArgument(nodeId != null, NODE_ID_NULL);
329 checkArgument(deviceId != null, DEVICE_ID_NULL);
330
331 if (!nodeId.equals(localNodeId)) {
332 log.debug("Forwarding request to relinquish "
333 + "role for device {} to {}", deviceId, nodeId);
Madan Jampanif7536ab2015-05-07 23:23:23 -0700334 return clusterCommunicator.sendAndReceive(
Madan Jampani2bfa94c2015-04-11 05:03:49 -0700335 deviceId,
336 ROLE_RELINQUISH_SUBJECT,
337 SERIALIZER::encode,
338 SERIALIZER::decode,
Madan Jampanif7536ab2015-05-07 23:23:23 -0700339 nodeId);
Madan Jampani84b6b402015-02-25 17:49:54 -0800340 }
Madan Jampanif7536ab2015-05-07 23:23:23 -0700341 return CompletableFuture.completedFuture(relinquishLocalRole(deviceId));
342 }
343
344 private MastershipEvent relinquishLocalRole(DeviceId deviceId) {
345 checkArgument(deviceId != null, DEVICE_ID_NULL);
Madan Jampani84b6b402015-02-25 17:49:54 -0800346
347 // Check if this node is can be managed by this node.
348 if (!connectedDevices.contains(deviceId)) {
349 return null;
350 }
351
352 String leadershipTopic = createDeviceMastershipTopic(deviceId);
Madan Jampani9bd1f152015-04-30 23:33:35 -0700353 NodeId currentLeader = leadershipService.getLeader(leadershipTopic);
Madan Jampani84b6b402015-02-25 17:49:54 -0800354
Madan Jampani9bd1f152015-04-30 23:33:35 -0700355 MastershipEvent.Type eventType = Objects.equal(currentLeader, localNodeId)
356 ? MastershipEvent.Type.MASTER_CHANGED
357 : MastershipEvent.Type.BACKUPS_CHANGED;
Madan Jampani84b6b402015-02-25 17:49:54 -0800358
359 connectedDevices.remove(deviceId);
360 leadershipService.withdraw(leadershipTopic);
361
362 return new MastershipEvent(eventType, deviceId, getNodes(deviceId));
363 }
364
Madan Jampani9bd1f152015-04-30 23:33:35 -0700365 private MastershipEvent transitionFromMasterToStandby(DeviceId deviceId) {
Madan Jampani1af8e132015-04-30 16:41:18 -0700366 checkArgument(deviceId != null, DEVICE_ID_NULL);
367
368 NodeId currentMaster = getMaster(deviceId);
369 if (currentMaster == null) {
370 return null;
371 }
372
373 if (!currentMaster.equals(localNodeId)) {
374 log.info("Forwarding request to relinquish "
375 + "mastership for device {} to {}", deviceId, currentMaster);
376 return futureGetOrElse(clusterCommunicator.sendAndReceive(
377 deviceId,
Madan Jampani9bd1f152015-04-30 23:33:35 -0700378 TRANSITION_FROM_MASTER_TO_STANDBY_SUBJECT,
Madan Jampani1af8e132015-04-30 16:41:18 -0700379 SERIALIZER::encode,
380 SERIALIZER::decode,
381 currentMaster), null);
382 }
383
Madan Jampani9bd1f152015-04-30 23:33:35 -0700384 return leadershipService.stepdown(createDeviceMastershipTopic(deviceId))
385 ? new MastershipEvent(MastershipEvent.Type.MASTER_CHANGED, deviceId, getNodes(deviceId)) : null;
Madan Jampani1af8e132015-04-30 16:41:18 -0700386 }
387
Madan Jampani84b6b402015-02-25 17:49:54 -0800388 @Override
389 public void relinquishAllRole(NodeId nodeId) {
390 // Noop. LeadershipService already takes care of detecting and purging deadlocks.
391 }
392
Madan Jampani84b6b402015-02-25 17:49:54 -0800393 private class InternalDeviceMastershipEventListener implements LeadershipEventListener {
394 @Override
395 public void event(LeadershipEvent event) {
396 Leadership leadership = event.subject();
397 if (!isDeviceMastershipTopic(leadership.topic())) {
398 return;
399 }
400 NodeId nodeId = leadership.leader();
401 DeviceId deviceId = extractDeviceIdFromTopic(leadership.topic());
402 if (Objects.equal(nodeId, localNodeId) && connectedDevices.contains(deviceId)) {
403 switch (event.type()) {
404 case LEADER_ELECTED:
405 notifyDelegate(new MastershipEvent(MASTER_CHANGED, deviceId, getNodes(deviceId)));
406 break;
407 case LEADER_REELECTED:
408 // There is no concept of leader re-election in the new distributed leadership manager.
409 throw new IllegalStateException("Unexpected event type");
410 case LEADER_BOOTED:
411 notifyDelegate(new MastershipEvent(BACKUPS_CHANGED, deviceId, getNodes(deviceId)));
412 break;
413 default:
414 return;
415 }
416 }
417 }
418 }
419
420 private String createDeviceMastershipTopic(DeviceId deviceId) {
Madan Jampani5756c352015-04-29 00:23:58 -0700421 return String.format("device:%s", deviceId.toString());
Madan Jampani84b6b402015-02-25 17:49:54 -0800422 }
423
424 private DeviceId extractDeviceIdFromTopic(String topic) {
425 Matcher m = DEVICE_MASTERSHIP_TOPIC_PATTERN.matcher(topic);
426 if (m.matches()) {
427 return DeviceId.deviceId(m.group(1));
428 } else {
429 throw new IllegalArgumentException("Invalid device mastership topic: " + topic);
430 }
431 }
432
433 private boolean isDeviceMastershipTopic(String topic) {
434 Matcher m = DEVICE_MASTERSHIP_TOPIC_PATTERN.matcher(topic);
435 return m.matches();
436 }
437
Madan Jampanic26eede2015-04-16 11:42:16 -0700438}