blob: 084c1982cf256028ed502e5d7ae1145f8769b3ad [file] [log] [blame]
Thomas Vachuska4f1a60c2014-10-28 13:39:07 -07001/*
Brian O'Connora09fe5b2017-08-03 21:12:30 -07002 * Copyright 2014-present Open Networking Foundation
Thomas Vachuska4f1a60c2014-10-28 13:39:07 -07003 *
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'Connorabafb502014-12-02 22:26:20 -080016package org.onosproject.store.cluster.impl;
tom2d7c65f2014-09-23 01:09:35 -070017
Jonathan Hart4a4d18f2015-03-26 12:16:16 -070018import com.google.common.collect.ImmutableSet;
19import com.google.common.collect.Maps;
Madan Jampanie3375062016-04-19 16:32:10 -070020
tom2d7c65f2014-09-23 01:09:35 -070021import org.apache.felix.scr.annotations.Activate;
22import org.apache.felix.scr.annotations.Component;
23import org.apache.felix.scr.annotations.Deactivate;
sangyun-hanf98df542016-03-24 20:28:03 +090024import org.apache.felix.scr.annotations.Modified;
25import org.apache.felix.scr.annotations.Property;
Madan Jampaniafeebbd2015-05-19 15:26:01 -070026import org.apache.felix.scr.annotations.Reference;
27import org.apache.felix.scr.annotations.ReferenceCardinality;
Deepa Vaddireddy6690f592017-03-12 03:31:48 +053028import org.apache.felix.scr.annotations.ReferencePolicy;
tom2d7c65f2014-09-23 01:09:35 -070029import org.apache.felix.scr.annotations.Service;
Ayaka Koshibedd91b842015-03-02 14:48:47 -080030import org.onlab.packet.IpAddress;
31import org.onlab.util.KryoNamespace;
Deepa Vaddireddy6690f592017-03-12 03:31:48 +053032import org.onosproject.cfg.ConfigProperty;
sangyun-han9f0af2d2016-08-04 13:04:59 +090033import org.onosproject.cfg.ComponentConfigService;
Brian O'Connorabafb502014-12-02 22:26:20 -080034import org.onosproject.cluster.ClusterEvent;
Madan Jampaniec1df022015-10-13 21:23:03 -070035import org.onosproject.cluster.ClusterMetadataService;
Brian O'Connorabafb502014-12-02 22:26:20 -080036import org.onosproject.cluster.ClusterStore;
37import org.onosproject.cluster.ClusterStoreDelegate;
38import org.onosproject.cluster.ControllerNode;
Ayaka Koshibedd91b842015-03-02 14:48:47 -080039import org.onosproject.cluster.ControllerNode.State;
Jonathan Hart4a4d18f2015-03-26 12:16:16 -070040import org.onosproject.cluster.DefaultControllerNode;
Jordan Halterman00e92da2018-05-22 23:05:52 -070041import org.onosproject.cluster.Node;
Brian O'Connorabafb502014-12-02 22:26:20 -080042import org.onosproject.cluster.NodeId;
Jordan Haltermanf70bf462017-07-29 13:12:00 -070043import org.onosproject.core.Version;
44import org.onosproject.core.VersionService;
Ayaka Koshibedd91b842015-03-02 14:48:47 -080045import org.onosproject.store.AbstractStore;
Madan Jampanic26eede2015-04-16 11:42:16 -070046import org.onosproject.store.cluster.messaging.Endpoint;
Madan Jampaniafeebbd2015-05-19 15:26:01 -070047import org.onosproject.store.cluster.messaging.MessagingService;
Ayaka Koshibedd91b842015-03-02 14:48:47 -080048import org.onosproject.store.serializers.KryoNamespaces;
Jordan Halterman2c83a102017-08-20 17:11:41 -070049import org.onosproject.store.service.Serializer;
sangyun-hanf98df542016-03-24 20:28:03 +090050import org.osgi.service.component.ComponentContext;
Ayaka Koshibedd91b842015-03-02 14:48:47 -080051import org.slf4j.Logger;
tom2d7c65f2014-09-23 01:09:35 -070052
Yuta HIGUCHI0c47d532017-08-18 23:16:35 -070053import java.time.Instant;
Jonathan Hart4a4d18f2015-03-26 12:16:16 -070054import java.util.Map;
Thomas Vachuskafba7f3d2016-03-23 15:46:25 -070055import java.util.Objects;
Jonathan Hart4a4d18f2015-03-26 12:16:16 -070056import java.util.Set;
57import java.util.concurrent.ExecutorService;
58import java.util.concurrent.Executors;
59import java.util.concurrent.ScheduledExecutorService;
60import java.util.concurrent.TimeUnit;
Madan Jampanid36def02016-01-13 11:21:56 -080061import java.util.function.BiConsumer;
Jonathan Hart4a4d18f2015-03-26 12:16:16 -070062import java.util.stream.Collectors;
63
Jordan Haltermanf70bf462017-07-29 13:12:00 -070064import static com.google.common.base.MoreObjects.firstNonNull;
sangyun-hanf98df542016-03-24 20:28:03 +090065import static com.google.common.base.Preconditions.checkArgument;
Jonathan Hart4a4d18f2015-03-26 12:16:16 -070066import static com.google.common.base.Preconditions.checkNotNull;
Deepa Vaddireddy6690f592017-03-12 03:31:48 +053067import static com.google.common.base.Strings.isNullOrEmpty;
Jonathan Hart4a4d18f2015-03-26 12:16:16 -070068import static org.onlab.util.Tools.groupedThreads;
Thomas Vachuskafba7f3d2016-03-23 15:46:25 -070069import static org.onosproject.cluster.ClusterEvent.Type.INSTANCE_ACTIVATED;
70import static org.onosproject.cluster.ClusterEvent.Type.INSTANCE_DEACTIVATED;
71import static org.onosproject.cluster.ClusterEvent.Type.INSTANCE_READY;
Jonathan Hart4a4d18f2015-03-26 12:16:16 -070072import static org.slf4j.LoggerFactory.getLogger;
tom2d7c65f2014-09-23 01:09:35 -070073
Jordan Halterman00e92da2018-05-22 23:05:52 -070074@Component(enabled = false)
tom2d7c65f2014-09-23 01:09:35 -070075@Service
Ayaka Koshibedd91b842015-03-02 14:48:47 -080076/**
77 * Distributed cluster nodes store that employs an accrual failure
78 * detector to identify cluster member up/down status.
79 */
tom0755a362014-09-24 11:54:43 -070080public class DistributedClusterStore
Ayaka Koshibedd91b842015-03-02 14:48:47 -080081 extends AbstractStore<ClusterEvent, ClusterStoreDelegate>
tomb41d1ac2014-09-24 01:51:24 -070082 implements ClusterStore {
tom2d7c65f2014-09-23 01:09:35 -070083
Thomas Vachuska8dc1a692015-03-31 01:01:37 -070084 private static final Logger log = getLogger(DistributedClusterStore.class);
tom2d7c65f2014-09-23 01:09:35 -070085
Thomas Vachuskade563cf2015-04-01 00:28:50 -070086 public static final String HEARTBEAT_MESSAGE = "onos-cluster-heartbeat";
87
sangyun-hanf98df542016-03-24 20:28:03 +090088 private static final int DEFAULT_HEARTBEAT_INTERVAL = 100;
89 @Property(name = "heartbeatInterval", intValue = DEFAULT_HEARTBEAT_INTERVAL,
90 label = "Interval time to send heartbeat to other controller nodes (millisecond)")
91 private int heartbeatInterval = DEFAULT_HEARTBEAT_INTERVAL;
92
93 private static final int DEFAULT_PHI_FAILURE_THRESHOLD = 10;
94 @Property(name = "phiFailureThreshold", intValue = DEFAULT_PHI_FAILURE_THRESHOLD,
95 label = "the value of Phi threshold to detect accrual failure")
96 private int phiFailureThreshold = DEFAULT_PHI_FAILURE_THRESHOLD;
tom2d7c65f2014-09-23 01:09:35 -070097
Jordan Halterman4a082ec2018-05-14 15:00:42 -070098 private static final long DEFAULT_MIN_STANDARD_DEVIATION_MILLIS = 50;
99 @Property(name = "minStandardDeviationMillis", longValue = DEFAULT_MIN_STANDARD_DEVIATION_MILLIS,
100 label = "The minimum standard deviation to take into account when computing the Phi value")
101 private long minStandardDeviationMillis = DEFAULT_MIN_STANDARD_DEVIATION_MILLIS;
102
Jordan Halterman2c83a102017-08-20 17:11:41 -0700103 private static final Serializer SERIALIZER = Serializer.using(
Jordan Haltermanf70bf462017-07-29 13:12:00 -0700104 KryoNamespace.newBuilder()
Thomas Vachuska8dc1a692015-03-31 01:01:37 -0700105 .register(KryoNamespaces.API)
HIGUCHI Yutae7290652016-05-18 11:29:01 -0700106 .nextId(KryoNamespaces.BEGIN_USER_CUSTOM_ID)
Thomas Vachuska8dc1a692015-03-31 01:01:37 -0700107 .register(HeartbeatMessage.class)
HIGUCHI Yutae7290652016-05-18 11:29:01 -0700108 .build("ClusterStore"));
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800109
110 private static final String INSTANCE_ID_NULL = "Instance ID cannot be null";
111
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800112 private final Map<NodeId, ControllerNode> allNodes = Maps.newConcurrentMap();
113 private final Map<NodeId, State> nodeStates = Maps.newConcurrentMap();
Jordan Haltermanf70bf462017-07-29 13:12:00 -0700114 private final Map<NodeId, Version> nodeVersions = Maps.newConcurrentMap();
Yuta HIGUCHI0c47d532017-08-18 23:16:35 -0700115 private final Map<NodeId, Instant> nodeLastUpdatedTimes = Maps.newConcurrentMap();
Thomas Vachuska7a8de842016-03-07 20:56:35 -0800116
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800117 private ScheduledExecutorService heartBeatSender = Executors.newSingleThreadScheduledExecutor(
HIGUCHI Yutad9e01052016-04-14 09:31:42 -0700118 groupedThreads("onos/cluster/membership", "heartbeat-sender", log));
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800119 private ExecutorService heartBeatMessageHandler = Executors.newSingleThreadExecutor(
HIGUCHI Yutad9e01052016-04-14 09:31:42 -0700120 groupedThreads("onos/cluster/membership", "heartbeat-receiver", log));
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800121
122 private PhiAccrualFailureDetector failureDetector;
123
124 private ControllerNode localNode;
Jordan Haltermanf70bf462017-07-29 13:12:00 -0700125 private Version localVersion;
126
127 @Reference(cardinality = ReferenceCardinality.MANDATORY_UNARY)
128 protected VersionService versionService;
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800129
Madan Jampaniafeebbd2015-05-19 15:26:01 -0700130 @Reference(cardinality = ReferenceCardinality.MANDATORY_UNARY)
Madan Jampaniec1df022015-10-13 21:23:03 -0700131 protected ClusterMetadataService clusterMetadataService;
Madan Jampaniafeebbd2015-05-19 15:26:01 -0700132
133 @Reference(cardinality = ReferenceCardinality.MANDATORY_UNARY)
134 protected MessagingService messagingService;
135
Thomas Vachuska1ca7e9f2016-08-05 10:21:41 -0700136 // This must be optional to avoid a cyclic dependency
Deepa Vaddireddy6690f592017-03-12 03:31:48 +0530137 @Reference(cardinality = ReferenceCardinality.OPTIONAL_UNARY,
Jordan Haltermanf70bf462017-07-29 13:12:00 -0700138 bind = "bindComponentConfigService",
139 unbind = "unbindComponentConfigService",
140 policy = ReferencePolicy.DYNAMIC)
sangyun-han9f0af2d2016-08-04 13:04:59 +0900141 protected ComponentConfigService cfgService;
142
Thomas Vachuska1ca7e9f2016-08-05 10:21:41 -0700143 /**
144 * Hook for wiring up optional reference to a service.
145 *
146 * @param service service being announced
147 */
148 protected void bindComponentConfigService(ComponentConfigService service) {
149 if (cfgService == null) {
150 cfgService = service;
151 cfgService.registerProperties(getClass());
Deepa Vaddireddy6690f592017-03-12 03:31:48 +0530152 readComponentConfiguration();
Thomas Vachuska1ca7e9f2016-08-05 10:21:41 -0700153 }
154 }
155
156 /**
157 * Hook for unwiring optional reference to a service.
158 *
159 * @param service service being withdrawn
160 */
161 protected void unbindComponentConfigService(ComponentConfigService service) {
162 if (cfgService == service) {
163 cfgService.unregisterProperties(getClass(), false);
164 cfgService = null;
165 }
166 }
167
tom2d7c65f2014-09-23 01:09:35 -0700168 @Activate
Thomas Vachuska1ca7e9f2016-08-05 10:21:41 -0700169 public void activate() {
Madan Jampaniec1df022015-10-13 21:23:03 -0700170 localNode = clusterMetadataService.getLocalNode();
Jordan Haltermanf70bf462017-07-29 13:12:00 -0700171 localVersion = versionService.version();
172 nodeVersions.put(localNode.id(), localVersion);
Thomas Vachuska8dc1a692015-03-31 01:01:37 -0700173
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800174 messagingService.registerHandler(HEARTBEAT_MESSAGE,
Jordan Haltermanf70bf462017-07-29 13:12:00 -0700175 new HeartbeatMessageHandler(), heartBeatMessageHandler);
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800176
Jordan Halterman4a082ec2018-05-14 15:00:42 -0700177 failureDetector = new PhiAccrualFailureDetector(minStandardDeviationMillis);
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800178
179 heartBeatSender.scheduleWithFixedDelay(this::heartbeat, 0,
Jordan Haltermanf70bf462017-07-29 13:12:00 -0700180 heartbeatInterval, TimeUnit.MILLISECONDS);
tomb41d1ac2014-09-24 01:51:24 -0700181
182 log.info("Started");
183 }
184
tom2d7c65f2014-09-23 01:09:35 -0700185 @Deactivate
186 public void deactivate() {
Madan Jampaniafeebbd2015-05-19 15:26:01 -0700187 messagingService.unregisterHandler(HEARTBEAT_MESSAGE);
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800188 heartBeatSender.shutdownNow();
189 heartBeatMessageHandler.shutdownNow();
190
tom2d7c65f2014-09-23 01:09:35 -0700191 log.info("Stopped");
192 }
193
sangyun-hanf98df542016-03-24 20:28:03 +0900194 @Modified
195 public void modified(ComponentContext context) {
Deepa Vaddireddy6690f592017-03-12 03:31:48 +0530196 readComponentConfiguration();
sangyun-hanf98df542016-03-24 20:28:03 +0900197 }
198
tom2d7c65f2014-09-23 01:09:35 -0700199 @Override
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800200 public void setDelegate(ClusterStoreDelegate delegate) {
201 checkNotNull(delegate, "Delegate cannot be null");
202 this.delegate = delegate;
203 }
204
205 @Override
206 public void unsetDelegate(ClusterStoreDelegate delegate) {
207 this.delegate = null;
208 }
209
210 @Override
211 public boolean hasDelegate() {
212 return this.delegate != null;
213 }
214
215 @Override
tom2d7c65f2014-09-23 01:09:35 -0700216 public ControllerNode getLocalNode() {
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800217 return localNode;
tom2d7c65f2014-09-23 01:09:35 -0700218 }
219
220 @Override
221 public Set<ControllerNode> getNodes() {
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800222 return ImmutableSet.copyOf(allNodes.values());
tom2d7c65f2014-09-23 01:09:35 -0700223 }
224
225 @Override
Jordan Halterman00e92da2018-05-22 23:05:52 -0700226 public Set<Node> getStorageNodes() {
227 return ImmutableSet.of();
228 }
229
230 @Override
tom2d7c65f2014-09-23 01:09:35 -0700231 public ControllerNode getNode(NodeId nodeId) {
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800232 checkNotNull(nodeId, INSTANCE_ID_NULL);
233 return allNodes.get(nodeId);
tom2d7c65f2014-09-23 01:09:35 -0700234 }
235
236 @Override
tomb41d1ac2014-09-24 01:51:24 -0700237 public State getState(NodeId nodeId) {
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800238 checkNotNull(nodeId, INSTANCE_ID_NULL);
Jordan Haltermanf70bf462017-07-29 13:12:00 -0700239 return firstNonNull(nodeStates.get(nodeId), State.INACTIVE);
240 }
241
242 @Override
243 public Version getVersion(NodeId nodeId) {
244 checkNotNull(nodeId, INSTANCE_ID_NULL);
245 return nodeVersions.get(nodeId);
tomb41d1ac2014-09-24 01:51:24 -0700246 }
247
248 @Override
Thomas Vachuska7a8de842016-03-07 20:56:35 -0800249 public void markFullyStarted(boolean started) {
Jordan Haltermanf70bf462017-07-29 13:12:00 -0700250 updateNode(localNode.id(), started ? State.READY : State.ACTIVE, null);
Thomas Vachuska7a8de842016-03-07 20:56:35 -0800251 }
252
253 @Override
Pavlin Radoslavov444b5192014-10-28 10:45:19 -0700254 public ControllerNode addNode(NodeId nodeId, IpAddress ip, int tcpPort) {
sangyun-hanf98df542016-03-24 20:28:03 +0900255 checkNotNull(nodeId, INSTANCE_ID_NULL);
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800256 ControllerNode node = new DefaultControllerNode(nodeId, ip, tcpPort);
Madan Jampaniafeebbd2015-05-19 15:26:01 -0700257 addNode(node);
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800258 return node;
tomee49c372014-09-26 15:14:50 -0700259 }
260
261 @Override
tomb41d1ac2014-09-24 01:51:24 -0700262 public void removeNode(NodeId nodeId) {
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800263 checkNotNull(nodeId, INSTANCE_ID_NULL);
264 ControllerNode node = allNodes.remove(nodeId);
265 if (node != null) {
266 nodeStates.remove(nodeId);
Jordan Haltermanf70bf462017-07-29 13:12:00 -0700267 nodeVersions.remove(nodeId);
Jonathan Hartf1141262015-04-23 11:27:07 -0700268 notifyDelegate(new ClusterEvent(ClusterEvent.Type.INSTANCE_REMOVED, node));
tomb41d1ac2014-09-24 01:51:24 -0700269 }
270 }
271
Madan Jampaniafeebbd2015-05-19 15:26:01 -0700272 private void addNode(ControllerNode node) {
273 allNodes.put(node.id(), node);
Jordan Haltermanf70bf462017-07-29 13:12:00 -0700274 updateNode(node.id(), node.equals(localNode) ? State.ACTIVE : State.INACTIVE, null);
Madan Jampaniafeebbd2015-05-19 15:26:01 -0700275 notifyDelegate(new ClusterEvent(ClusterEvent.Type.INSTANCE_ADDED, node));
Thomas Vachuskade563cf2015-04-01 00:28:50 -0700276 }
277
Jordan Haltermanf70bf462017-07-29 13:12:00 -0700278 private void updateNode(NodeId nodeId, State newState, Version newVersion) {
Thomas Vachuskafba7f3d2016-03-23 15:46:25 -0700279 State currentState = nodeStates.get(nodeId);
Jordan Haltermanf70bf462017-07-29 13:12:00 -0700280 Version currentVersion = nodeVersions.get(nodeId);
Jordan Haltermaneabae962017-08-01 17:29:36 -0700281 if (!Objects.equals(currentState, newState)
282 || (newVersion != null && !Objects.equals(currentVersion, newVersion))) {
Thomas Vachuskafba7f3d2016-03-23 15:46:25 -0700283 nodeStates.put(nodeId, newState);
Jordan Haltermanf70bf462017-07-29 13:12:00 -0700284 if (newVersion != null) {
285 nodeVersions.put(nodeId, newVersion);
286 }
Yuta HIGUCHI0c47d532017-08-18 23:16:35 -0700287 nodeLastUpdatedTimes.put(nodeId, Instant.now());
Jordan Haltermanf70bf462017-07-29 13:12:00 -0700288 notifyChange(nodeId, currentState, newState, currentVersion, newVersion);
Thomas Vachuskafba7f3d2016-03-23 15:46:25 -0700289 }
Madan Jampani7d2fab22015-03-18 17:21:57 -0700290 }
291
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800292 private void heartbeat() {
293 try {
294 Set<ControllerNode> peers = allNodes.values()
295 .stream()
296 .filter(node -> !(node.id().equals(localNode.id())))
297 .collect(Collectors.toSet());
Thomas Vachuska7a8de842016-03-07 20:56:35 -0800298 State state = nodeStates.get(localNode.id());
Jordan Haltermanf70bf462017-07-29 13:12:00 -0700299 byte[] hbMessagePayload = SERIALIZER.encode(new HeartbeatMessage(localNode, state, localVersion));
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800300 peers.forEach((node) -> {
301 heartbeatToPeer(hbMessagePayload, node);
302 State currentState = nodeStates.get(node.id());
303 double phi = failureDetector.phi(node.id());
sangyun-hanf98df542016-03-24 20:28:03 +0900304 if (phi >= phiFailureThreshold) {
Thomas Vachuska7a8de842016-03-07 20:56:35 -0800305 if (currentState.isActive()) {
Jordan Haltermanf70bf462017-07-29 13:12:00 -0700306 updateNode(node.id(), State.INACTIVE, null);
Jordan Haltermane7062532018-01-22 11:22:46 -0800307 failureDetector.reset(node.id());
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800308 }
309 } else {
310 if (currentState == State.INACTIVE) {
Jordan Haltermanf70bf462017-07-29 13:12:00 -0700311 updateNode(node.id(), State.ACTIVE, null);
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800312 }
313 }
314 });
315 } catch (Exception e) {
316 log.debug("Failed to send heartbeat", e);
317 }
tomb41d1ac2014-09-24 01:51:24 -0700318 }
319
Jordan Haltermanf70bf462017-07-29 13:12:00 -0700320 private void notifyChange(NodeId nodeId, State oldState, State newState, Version oldVersion, Version newVersion) {
321 if (oldState != newState || !Objects.equals(oldVersion, newVersion)) {
Thomas Vachuskafba7f3d2016-03-23 15:46:25 -0700322 ControllerNode node = allNodes.get(nodeId);
Jon Hall1f13d642016-05-13 17:53:01 -0700323 // Either this node or that node is no longer part of the same cluster
324 if (node == null) {
Jon Hall66870baa2016-06-20 12:12:10 -0700325 log.debug("Could not find node {} in the cluster, ignoring state change", nodeId);
Jon Hall1f13d642016-05-13 17:53:01 -0700326 return;
327 }
Thomas Vachuskafba7f3d2016-03-23 15:46:25 -0700328 ClusterEvent.Type type = newState == State.READY ? INSTANCE_READY :
329 newState == State.ACTIVE ? INSTANCE_ACTIVATED :
330 INSTANCE_DEACTIVATED;
331 notifyDelegate(new ClusterEvent(type, node));
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800332 }
tomb41d1ac2014-09-24 01:51:24 -0700333 }
334
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800335 private void heartbeatToPeer(byte[] messagePayload, ControllerNode peer) {
Madan Jampaniafeebbd2015-05-19 15:26:01 -0700336 Endpoint remoteEp = new Endpoint(peer.ip(), peer.tcpPort());
Madan Jampani175e8fd2015-05-20 14:10:45 -0700337 messagingService.sendAsync(remoteEp, HEARTBEAT_MESSAGE, messagePayload).whenComplete((result, error) -> {
338 if (error != null) {
339 log.trace("Sending heartbeat to {} failed", remoteEp, error);
340 }
341 });
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800342 }
343
Madan Jampanid36def02016-01-13 11:21:56 -0800344 private class HeartbeatMessageHandler implements BiConsumer<Endpoint, byte[]> {
tomb41d1ac2014-09-24 01:51:24 -0700345 @Override
Madan Jampanid36def02016-01-13 11:21:56 -0800346 public void accept(Endpoint sender, byte[] message) {
Madan Jampanic26eede2015-04-16 11:42:16 -0700347 HeartbeatMessage hb = SERIALIZER.decode(message);
Madan Jampanie3375062016-04-19 16:32:10 -0700348 if (clusterMetadataService.getClusterMetadata().getNodes().contains(hb.source())) {
Jordan Haltermane7062532018-01-22 11:22:46 -0800349 // Avoid reporting heartbeats that have been enqueued by setting a minimum interval.
350 long heartbeatTime = System.currentTimeMillis();
351 long lastHeartbeatTime = failureDetector.getLastHeartbeatTime(hb.source().id());
352 if (heartbeatTime - lastHeartbeatTime > heartbeatInterval / 2) {
353 failureDetector.report(hb.source().id(), heartbeatTime);
Jordan Halterman1315d3e2018-01-16 19:41:31 -0800354 }
Jordan Haltermanf70bf462017-07-29 13:12:00 -0700355 updateNode(hb.source().id(), hb.state, hb.version);
Madan Jampanie3375062016-04-19 16:32:10 -0700356 }
tomb41d1ac2014-09-24 01:51:24 -0700357 }
tom2d7c65f2014-09-23 01:09:35 -0700358 }
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800359
360 private static class HeartbeatMessage {
361 private ControllerNode source;
Thomas Vachuska7a8de842016-03-07 20:56:35 -0800362 private State state;
Jordan Haltermanf70bf462017-07-29 13:12:00 -0700363 private Version version;
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800364
Jordan Haltermanf70bf462017-07-29 13:12:00 -0700365 public HeartbeatMessage(ControllerNode source, State state, Version version) {
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800366 this.source = source;
Thomas Vachuska7a8de842016-03-07 20:56:35 -0800367 this.state = state != null ? state : State.ACTIVE;
Jordan Haltermanf70bf462017-07-29 13:12:00 -0700368 this.version = version;
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800369 }
370
371 public ControllerNode source() {
372 return source;
373 }
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800374 }
375
Madan Jampani7d2fab22015-03-18 17:21:57 -0700376 @Override
Yuta HIGUCHI0c47d532017-08-18 23:16:35 -0700377 public Instant getLastUpdatedInstant(NodeId nodeId) {
Jordan Haltermanf70bf462017-07-29 13:12:00 -0700378 return nodeLastUpdatedTimes.get(nodeId);
Madan Jampani7d2fab22015-03-18 17:21:57 -0700379 }
Thomas Vachuskade563cf2015-04-01 00:28:50 -0700380
sangyun-hanf98df542016-03-24 20:28:03 +0900381 /**
Deepa Vaddireddy6690f592017-03-12 03:31:48 +0530382 * Extracts properties from the component configuration.
sangyun-hanf98df542016-03-24 20:28:03 +0900383 *
sangyun-hanf98df542016-03-24 20:28:03 +0900384 */
Deepa Vaddireddy6690f592017-03-12 03:31:48 +0530385 private void readComponentConfiguration() {
386 Set<ConfigProperty> configProperties = cfgService.getProperties(getClass().getName());
387 for (ConfigProperty property : configProperties) {
Jon Halla3fcf672017-03-28 16:53:22 -0700388 if ("heartbeatInterval".equals(property.name())) {
Deepa Vaddireddy6690f592017-03-12 03:31:48 +0530389 String s = property.value();
390 if (s == null) {
391 setHeartbeatInterval(DEFAULT_HEARTBEAT_INTERVAL);
392 log.info("Heartbeat interval time is not configured, default value is {}",
393 DEFAULT_HEARTBEAT_INTERVAL);
394 } else {
395 int newHeartbeatInterval = isNullOrEmpty(s) ? DEFAULT_HEARTBEAT_INTERVAL
Jordan Haltermanf70bf462017-07-29 13:12:00 -0700396 : Integer.parseInt(s.trim());
Deepa Vaddireddy6690f592017-03-12 03:31:48 +0530397 if (newHeartbeatInterval > 0 && heartbeatInterval != newHeartbeatInterval) {
398 heartbeatInterval = newHeartbeatInterval;
399 restartHeartbeatSender();
400 }
401 log.info("Configured. Heartbeat interval time is configured to {}",
402 heartbeatInterval);
403 }
404 }
Jon Halla3fcf672017-03-28 16:53:22 -0700405 if ("phiFailureThreshold".equals(property.name())) {
Deepa Vaddireddy6690f592017-03-12 03:31:48 +0530406 String s = property.value();
407 if (s == null) {
408 setPhiFailureThreshold(DEFAULT_PHI_FAILURE_THRESHOLD);
409 log.info("Phi failure threshold is not configured, default value is {}",
410 DEFAULT_PHI_FAILURE_THRESHOLD);
411 } else {
412 int newPhiFailureThreshold = isNullOrEmpty(s) ? DEFAULT_HEARTBEAT_INTERVAL
Jordan Haltermanf70bf462017-07-29 13:12:00 -0700413 : Integer.parseInt(s.trim());
Deepa Vaddireddy6690f592017-03-12 03:31:48 +0530414 setPhiFailureThreshold(newPhiFailureThreshold);
415 log.info("Configured. Phi failure threshold is configured to {}",
416 phiFailureThreshold);
417 }
418 }
Jordan Halterman4a082ec2018-05-14 15:00:42 -0700419 if ("minStandardDeviationMillis".equals(property.name())) {
420 String s = property.value();
421 if (s == null) {
422 setMinStandardDeviationMillis(DEFAULT_MIN_STANDARD_DEVIATION_MILLIS);
423 log.info("Minimum standard deviation is not configured, default value is {}",
424 DEFAULT_MIN_STANDARD_DEVIATION_MILLIS);
425 } else {
426 long newMinStandardDeviationMillis = isNullOrEmpty(s)
427 ? DEFAULT_MIN_STANDARD_DEVIATION_MILLIS
428 : Long.parseLong(s.trim());
429 setMinStandardDeviationMillis(newMinStandardDeviationMillis);
430 log.info("Configured. Minimum standard deviation is configured to {}",
431 newMinStandardDeviationMillis);
432 }
433 }
sangyun-hanf98df542016-03-24 20:28:03 +0900434 }
435 }
436
437 /**
438 * Sets heartbeat interval between the termination of one execution of heartbeat
439 * and the commencement of the next.
440 *
441 * @param interval term between each heartbeat
442 */
443 private void setHeartbeatInterval(int interval) {
444 try {
445 checkArgument(interval > 0, "Interval must be greater than zero");
446 heartbeatInterval = interval;
447 } catch (IllegalArgumentException e) {
448 log.warn(e.getMessage());
449 heartbeatInterval = DEFAULT_HEARTBEAT_INTERVAL;
450 }
451 }
452
453 /**
454 * Sets Phi failure threshold.
455 * Phi is based on a paper titled: "The φ Accrual Failure Detector" by Hayashibara, et al.
456 *
457 * @param threshold
458 */
459 private void setPhiFailureThreshold(int threshold) {
460 phiFailureThreshold = threshold;
461 }
462
463 /**
Jordan Halterman4a082ec2018-05-14 15:00:42 -0700464 * Sets the minimum standard deviation milliseconds.
465 *
466 * @param minStandardDeviationMillis the updated minimum standard deviation
467 */
468 private void setMinStandardDeviationMillis(long minStandardDeviationMillis) {
469 this.minStandardDeviationMillis = minStandardDeviationMillis;
470 try {
471 failureDetector = new PhiAccrualFailureDetector(minStandardDeviationMillis);
472 } catch (IllegalArgumentException e) {
473 log.warn(e.getMessage());
474 this.minStandardDeviationMillis = DEFAULT_MIN_STANDARD_DEVIATION_MILLIS;
475 failureDetector = new PhiAccrualFailureDetector(this.minStandardDeviationMillis);
476 }
477 }
478
479 /**
sangyun-hanf98df542016-03-24 20:28:03 +0900480 * Restarts heartbeatSender executor.
sangyun-hanf98df542016-03-24 20:28:03 +0900481 */
482 private void restartHeartbeatSender() {
483 try {
484 ScheduledExecutorService prevSender = heartBeatSender;
485 heartBeatSender = Executors.newSingleThreadScheduledExecutor(
HIGUCHI Yutad9e01052016-04-14 09:31:42 -0700486 groupedThreads("onos/cluster/membership", "heartbeat-sender-%d", log));
sangyun-hanf98df542016-03-24 20:28:03 +0900487 heartBeatSender.scheduleWithFixedDelay(this::heartbeat, 0,
Jordan Haltermanf70bf462017-07-29 13:12:00 -0700488 heartbeatInterval, TimeUnit.MILLISECONDS);
sangyun-hanf98df542016-03-24 20:28:03 +0900489 prevSender.shutdown();
490 } catch (Exception e) {
491 log.warn(e.getMessage());
492 }
493 }
Jordan Haltermanf70bf462017-07-29 13:12:00 -0700494}