blob: f50ddbb7eae0550b50862550dd896cce246ab7fd [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;
Ayaka Koshibedd91b842015-03-02 14:48:47 -080020import org.onlab.packet.IpAddress;
21import org.onlab.util.KryoNamespace;
sangyun-han9f0af2d2016-08-04 13:04:59 +090022import org.onosproject.cfg.ComponentConfigService;
Ray Milkeyd84f89b2018-08-17 14:54:17 -070023import org.onosproject.cfg.ConfigProperty;
Brian O'Connorabafb502014-12-02 22:26:20 -080024import org.onosproject.cluster.ClusterEvent;
Madan Jampaniec1df022015-10-13 21:23:03 -070025import org.onosproject.cluster.ClusterMetadataService;
Brian O'Connorabafb502014-12-02 22:26:20 -080026import org.onosproject.cluster.ClusterStore;
27import org.onosproject.cluster.ClusterStoreDelegate;
28import org.onosproject.cluster.ControllerNode;
Ayaka Koshibedd91b842015-03-02 14:48:47 -080029import org.onosproject.cluster.ControllerNode.State;
Jonathan Hart4a4d18f2015-03-26 12:16:16 -070030import org.onosproject.cluster.DefaultControllerNode;
Jordan Halterman00e92da2018-05-22 23:05:52 -070031import org.onosproject.cluster.Node;
Brian O'Connorabafb502014-12-02 22:26:20 -080032import org.onosproject.cluster.NodeId;
Jordan Haltermanf70bf462017-07-29 13:12:00 -070033import org.onosproject.core.Version;
34import org.onosproject.core.VersionService;
Ayaka Koshibedd91b842015-03-02 14:48:47 -080035import org.onosproject.store.AbstractStore;
Madan Jampanic26eede2015-04-16 11:42:16 -070036import org.onosproject.store.cluster.messaging.Endpoint;
Madan Jampaniafeebbd2015-05-19 15:26:01 -070037import org.onosproject.store.cluster.messaging.MessagingService;
Ayaka Koshibedd91b842015-03-02 14:48:47 -080038import org.onosproject.store.serializers.KryoNamespaces;
Jordan Halterman2c83a102017-08-20 17:11:41 -070039import org.onosproject.store.service.Serializer;
sangyun-hanf98df542016-03-24 20:28:03 +090040import org.osgi.service.component.ComponentContext;
Ray Milkeyd84f89b2018-08-17 14:54:17 -070041import org.osgi.service.component.annotations.Activate;
42import org.osgi.service.component.annotations.Component;
43import org.osgi.service.component.annotations.Deactivate;
44import org.osgi.service.component.annotations.Modified;
45import org.osgi.service.component.annotations.Reference;
46import org.osgi.service.component.annotations.ReferenceCardinality;
47import org.osgi.service.component.annotations.ReferencePolicy;
Ayaka Koshibedd91b842015-03-02 14:48:47 -080048import org.slf4j.Logger;
tom2d7c65f2014-09-23 01:09:35 -070049
Yuta HIGUCHI0c47d532017-08-18 23:16:35 -070050import java.time.Instant;
Jonathan Hart4a4d18f2015-03-26 12:16:16 -070051import java.util.Map;
Thomas Vachuskafba7f3d2016-03-23 15:46:25 -070052import java.util.Objects;
Jonathan Hart4a4d18f2015-03-26 12:16:16 -070053import java.util.Set;
54import java.util.concurrent.ExecutorService;
55import java.util.concurrent.Executors;
56import java.util.concurrent.ScheduledExecutorService;
57import java.util.concurrent.TimeUnit;
Madan Jampanid36def02016-01-13 11:21:56 -080058import java.util.function.BiConsumer;
Jonathan Hart4a4d18f2015-03-26 12:16:16 -070059import java.util.stream.Collectors;
60
Jordan Haltermanf70bf462017-07-29 13:12:00 -070061import static com.google.common.base.MoreObjects.firstNonNull;
sangyun-hanf98df542016-03-24 20:28:03 +090062import static com.google.common.base.Preconditions.checkArgument;
Jonathan Hart4a4d18f2015-03-26 12:16:16 -070063import static com.google.common.base.Preconditions.checkNotNull;
Deepa Vaddireddy6690f592017-03-12 03:31:48 +053064import static com.google.common.base.Strings.isNullOrEmpty;
Jonathan Hart4a4d18f2015-03-26 12:16:16 -070065import static org.onlab.util.Tools.groupedThreads;
Thomas Vachuskafba7f3d2016-03-23 15:46:25 -070066import static org.onosproject.cluster.ClusterEvent.Type.INSTANCE_ACTIVATED;
67import static org.onosproject.cluster.ClusterEvent.Type.INSTANCE_DEACTIVATED;
68import static org.onosproject.cluster.ClusterEvent.Type.INSTANCE_READY;
Jonathan Hart4a4d18f2015-03-26 12:16:16 -070069import static org.slf4j.LoggerFactory.getLogger;
Ray Milkeyd84f89b2018-08-17 14:54:17 -070070@Component(enabled = false, service = ClusterStore.class)
Ayaka Koshibedd91b842015-03-02 14:48:47 -080071/**
72 * Distributed cluster nodes store that employs an accrual failure
73 * detector to identify cluster member up/down status.
74 */
tom0755a362014-09-24 11:54:43 -070075public class DistributedClusterStore
Ayaka Koshibedd91b842015-03-02 14:48:47 -080076 extends AbstractStore<ClusterEvent, ClusterStoreDelegate>
tomb41d1ac2014-09-24 01:51:24 -070077 implements ClusterStore {
tom2d7c65f2014-09-23 01:09:35 -070078
Thomas Vachuska8dc1a692015-03-31 01:01:37 -070079 private static final Logger log = getLogger(DistributedClusterStore.class);
tom2d7c65f2014-09-23 01:09:35 -070080
Thomas Vachuskade563cf2015-04-01 00:28:50 -070081 public static final String HEARTBEAT_MESSAGE = "onos-cluster-heartbeat";
82
sangyun-hanf98df542016-03-24 20:28:03 +090083 private static final int DEFAULT_HEARTBEAT_INTERVAL = 100;
Ray Milkeyd84f89b2018-08-17 14:54:17 -070084 //@Property(name = "heartbeatInterval", intValue = DEFAULT_HEARTBEAT_INTERVAL,
85 // label = "Interval time to send heartbeat to other controller nodes (millisecond)")
sangyun-hanf98df542016-03-24 20:28:03 +090086 private int heartbeatInterval = DEFAULT_HEARTBEAT_INTERVAL;
87
88 private static final int DEFAULT_PHI_FAILURE_THRESHOLD = 10;
Ray Milkeyd84f89b2018-08-17 14:54:17 -070089 //@Property(name = "phiFailureThreshold", intValue = DEFAULT_PHI_FAILURE_THRESHOLD,
90 // label = "the value of Phi threshold to detect accrual failure")
sangyun-hanf98df542016-03-24 20:28:03 +090091 private int phiFailureThreshold = DEFAULT_PHI_FAILURE_THRESHOLD;
tom2d7c65f2014-09-23 01:09:35 -070092
Jordan Halterman4a082ec2018-05-14 15:00:42 -070093 private static final long DEFAULT_MIN_STANDARD_DEVIATION_MILLIS = 50;
Ray Milkeyd84f89b2018-08-17 14:54:17 -070094 //@Property(name = "minStandardDeviationMillis", longValue = DEFAULT_MIN_STANDARD_DEVIATION_MILLIS,
95 // label = "The minimum standard deviation to take into account when computing the Phi value")
Jordan Halterman4a082ec2018-05-14 15:00:42 -070096 private long minStandardDeviationMillis = DEFAULT_MIN_STANDARD_DEVIATION_MILLIS;
97
Jordan Halterman2c83a102017-08-20 17:11:41 -070098 private static final Serializer SERIALIZER = Serializer.using(
Jordan Haltermanf70bf462017-07-29 13:12:00 -070099 KryoNamespace.newBuilder()
Thomas Vachuska8dc1a692015-03-31 01:01:37 -0700100 .register(KryoNamespaces.API)
HIGUCHI Yutae7290652016-05-18 11:29:01 -0700101 .nextId(KryoNamespaces.BEGIN_USER_CUSTOM_ID)
Thomas Vachuska8dc1a692015-03-31 01:01:37 -0700102 .register(HeartbeatMessage.class)
HIGUCHI Yutae7290652016-05-18 11:29:01 -0700103 .build("ClusterStore"));
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800104
105 private static final String INSTANCE_ID_NULL = "Instance ID cannot be null";
106
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800107 private final Map<NodeId, ControllerNode> allNodes = Maps.newConcurrentMap();
108 private final Map<NodeId, State> nodeStates = Maps.newConcurrentMap();
Jordan Haltermanf70bf462017-07-29 13:12:00 -0700109 private final Map<NodeId, Version> nodeVersions = Maps.newConcurrentMap();
Yuta HIGUCHI0c47d532017-08-18 23:16:35 -0700110 private final Map<NodeId, Instant> nodeLastUpdatedTimes = Maps.newConcurrentMap();
Thomas Vachuska7a8de842016-03-07 20:56:35 -0800111
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800112 private ScheduledExecutorService heartBeatSender = Executors.newSingleThreadScheduledExecutor(
HIGUCHI Yutad9e01052016-04-14 09:31:42 -0700113 groupedThreads("onos/cluster/membership", "heartbeat-sender", log));
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800114 private ExecutorService heartBeatMessageHandler = Executors.newSingleThreadExecutor(
HIGUCHI Yutad9e01052016-04-14 09:31:42 -0700115 groupedThreads("onos/cluster/membership", "heartbeat-receiver", log));
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800116
117 private PhiAccrualFailureDetector failureDetector;
118
119 private ControllerNode localNode;
Jordan Haltermanf70bf462017-07-29 13:12:00 -0700120 private Version localVersion;
121
Ray Milkeyd84f89b2018-08-17 14:54:17 -0700122 @Reference(cardinality = ReferenceCardinality.MANDATORY)
Jordan Haltermanf70bf462017-07-29 13:12:00 -0700123 protected VersionService versionService;
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800124
Ray Milkeyd84f89b2018-08-17 14:54:17 -0700125 @Reference(cardinality = ReferenceCardinality.MANDATORY)
Madan Jampaniec1df022015-10-13 21:23:03 -0700126 protected ClusterMetadataService clusterMetadataService;
Madan Jampaniafeebbd2015-05-19 15:26:01 -0700127
Ray Milkeyd84f89b2018-08-17 14:54:17 -0700128 @Reference(cardinality = ReferenceCardinality.MANDATORY)
Madan Jampaniafeebbd2015-05-19 15:26:01 -0700129 protected MessagingService messagingService;
130
Thomas Vachuska1ca7e9f2016-08-05 10:21:41 -0700131 // This must be optional to avoid a cyclic dependency
Ray Milkeyd84f89b2018-08-17 14:54:17 -0700132 @Reference(cardinality = ReferenceCardinality.OPTIONAL,
Jordan Haltermanf70bf462017-07-29 13:12:00 -0700133 bind = "bindComponentConfigService",
134 unbind = "unbindComponentConfigService",
135 policy = ReferencePolicy.DYNAMIC)
sangyun-han9f0af2d2016-08-04 13:04:59 +0900136 protected ComponentConfigService cfgService;
137
Thomas Vachuska1ca7e9f2016-08-05 10:21:41 -0700138 /**
139 * Hook for wiring up optional reference to a service.
140 *
141 * @param service service being announced
142 */
143 protected void bindComponentConfigService(ComponentConfigService service) {
144 if (cfgService == null) {
145 cfgService = service;
146 cfgService.registerProperties(getClass());
Deepa Vaddireddy6690f592017-03-12 03:31:48 +0530147 readComponentConfiguration();
Thomas Vachuska1ca7e9f2016-08-05 10:21:41 -0700148 }
149 }
150
151 /**
152 * Hook for unwiring optional reference to a service.
153 *
154 * @param service service being withdrawn
155 */
156 protected void unbindComponentConfigService(ComponentConfigService service) {
157 if (cfgService == service) {
158 cfgService.unregisterProperties(getClass(), false);
159 cfgService = null;
160 }
161 }
162
tom2d7c65f2014-09-23 01:09:35 -0700163 @Activate
Thomas Vachuska1ca7e9f2016-08-05 10:21:41 -0700164 public void activate() {
Madan Jampaniec1df022015-10-13 21:23:03 -0700165 localNode = clusterMetadataService.getLocalNode();
Jordan Haltermanf70bf462017-07-29 13:12:00 -0700166 localVersion = versionService.version();
167 nodeVersions.put(localNode.id(), localVersion);
Thomas Vachuska8dc1a692015-03-31 01:01:37 -0700168
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800169 messagingService.registerHandler(HEARTBEAT_MESSAGE,
Jordan Haltermanf70bf462017-07-29 13:12:00 -0700170 new HeartbeatMessageHandler(), heartBeatMessageHandler);
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800171
Jordan Halterman4a082ec2018-05-14 15:00:42 -0700172 failureDetector = new PhiAccrualFailureDetector(minStandardDeviationMillis);
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800173
174 heartBeatSender.scheduleWithFixedDelay(this::heartbeat, 0,
Jordan Haltermanf70bf462017-07-29 13:12:00 -0700175 heartbeatInterval, TimeUnit.MILLISECONDS);
tomb41d1ac2014-09-24 01:51:24 -0700176
177 log.info("Started");
178 }
179
tom2d7c65f2014-09-23 01:09:35 -0700180 @Deactivate
181 public void deactivate() {
Madan Jampaniafeebbd2015-05-19 15:26:01 -0700182 messagingService.unregisterHandler(HEARTBEAT_MESSAGE);
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800183 heartBeatSender.shutdownNow();
184 heartBeatMessageHandler.shutdownNow();
185
tom2d7c65f2014-09-23 01:09:35 -0700186 log.info("Stopped");
187 }
188
sangyun-hanf98df542016-03-24 20:28:03 +0900189 @Modified
190 public void modified(ComponentContext context) {
Deepa Vaddireddy6690f592017-03-12 03:31:48 +0530191 readComponentConfiguration();
sangyun-hanf98df542016-03-24 20:28:03 +0900192 }
193
tom2d7c65f2014-09-23 01:09:35 -0700194 @Override
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800195 public void setDelegate(ClusterStoreDelegate delegate) {
196 checkNotNull(delegate, "Delegate cannot be null");
197 this.delegate = delegate;
198 }
199
200 @Override
201 public void unsetDelegate(ClusterStoreDelegate delegate) {
202 this.delegate = null;
203 }
204
205 @Override
206 public boolean hasDelegate() {
207 return this.delegate != null;
208 }
209
210 @Override
tom2d7c65f2014-09-23 01:09:35 -0700211 public ControllerNode getLocalNode() {
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800212 return localNode;
tom2d7c65f2014-09-23 01:09:35 -0700213 }
214
215 @Override
216 public Set<ControllerNode> getNodes() {
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800217 return ImmutableSet.copyOf(allNodes.values());
tom2d7c65f2014-09-23 01:09:35 -0700218 }
219
220 @Override
Jordan Halterman00e92da2018-05-22 23:05:52 -0700221 public Set<Node> getStorageNodes() {
222 return ImmutableSet.of();
223 }
224
225 @Override
tom2d7c65f2014-09-23 01:09:35 -0700226 public ControllerNode getNode(NodeId nodeId) {
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800227 checkNotNull(nodeId, INSTANCE_ID_NULL);
228 return allNodes.get(nodeId);
tom2d7c65f2014-09-23 01:09:35 -0700229 }
230
231 @Override
tomb41d1ac2014-09-24 01:51:24 -0700232 public State getState(NodeId nodeId) {
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800233 checkNotNull(nodeId, INSTANCE_ID_NULL);
Jordan Haltermanf70bf462017-07-29 13:12:00 -0700234 return firstNonNull(nodeStates.get(nodeId), State.INACTIVE);
235 }
236
237 @Override
238 public Version getVersion(NodeId nodeId) {
239 checkNotNull(nodeId, INSTANCE_ID_NULL);
240 return nodeVersions.get(nodeId);
tomb41d1ac2014-09-24 01:51:24 -0700241 }
242
243 @Override
Thomas Vachuska7a8de842016-03-07 20:56:35 -0800244 public void markFullyStarted(boolean started) {
Jordan Haltermanf70bf462017-07-29 13:12:00 -0700245 updateNode(localNode.id(), started ? State.READY : State.ACTIVE, null);
Thomas Vachuska7a8de842016-03-07 20:56:35 -0800246 }
247
248 @Override
Pavlin Radoslavov444b5192014-10-28 10:45:19 -0700249 public ControllerNode addNode(NodeId nodeId, IpAddress ip, int tcpPort) {
sangyun-hanf98df542016-03-24 20:28:03 +0900250 checkNotNull(nodeId, INSTANCE_ID_NULL);
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800251 ControllerNode node = new DefaultControllerNode(nodeId, ip, tcpPort);
Madan Jampaniafeebbd2015-05-19 15:26:01 -0700252 addNode(node);
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800253 return node;
tomee49c372014-09-26 15:14:50 -0700254 }
255
256 @Override
tomb41d1ac2014-09-24 01:51:24 -0700257 public void removeNode(NodeId nodeId) {
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800258 checkNotNull(nodeId, INSTANCE_ID_NULL);
259 ControllerNode node = allNodes.remove(nodeId);
260 if (node != null) {
261 nodeStates.remove(nodeId);
Jordan Haltermanf70bf462017-07-29 13:12:00 -0700262 nodeVersions.remove(nodeId);
Jonathan Hartf1141262015-04-23 11:27:07 -0700263 notifyDelegate(new ClusterEvent(ClusterEvent.Type.INSTANCE_REMOVED, node));
tomb41d1ac2014-09-24 01:51:24 -0700264 }
265 }
266
Madan Jampaniafeebbd2015-05-19 15:26:01 -0700267 private void addNode(ControllerNode node) {
268 allNodes.put(node.id(), node);
Jordan Haltermanf70bf462017-07-29 13:12:00 -0700269 updateNode(node.id(), node.equals(localNode) ? State.ACTIVE : State.INACTIVE, null);
Madan Jampaniafeebbd2015-05-19 15:26:01 -0700270 notifyDelegate(new ClusterEvent(ClusterEvent.Type.INSTANCE_ADDED, node));
Thomas Vachuskade563cf2015-04-01 00:28:50 -0700271 }
272
Jordan Haltermanf70bf462017-07-29 13:12:00 -0700273 private void updateNode(NodeId nodeId, State newState, Version newVersion) {
Thomas Vachuskafba7f3d2016-03-23 15:46:25 -0700274 State currentState = nodeStates.get(nodeId);
Jordan Haltermanf70bf462017-07-29 13:12:00 -0700275 Version currentVersion = nodeVersions.get(nodeId);
Jordan Haltermaneabae962017-08-01 17:29:36 -0700276 if (!Objects.equals(currentState, newState)
277 || (newVersion != null && !Objects.equals(currentVersion, newVersion))) {
Thomas Vachuskafba7f3d2016-03-23 15:46:25 -0700278 nodeStates.put(nodeId, newState);
Jordan Haltermanf70bf462017-07-29 13:12:00 -0700279 if (newVersion != null) {
280 nodeVersions.put(nodeId, newVersion);
281 }
Yuta HIGUCHI0c47d532017-08-18 23:16:35 -0700282 nodeLastUpdatedTimes.put(nodeId, Instant.now());
Jordan Haltermanf70bf462017-07-29 13:12:00 -0700283 notifyChange(nodeId, currentState, newState, currentVersion, newVersion);
Thomas Vachuskafba7f3d2016-03-23 15:46:25 -0700284 }
Madan Jampani7d2fab22015-03-18 17:21:57 -0700285 }
286
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800287 private void heartbeat() {
288 try {
289 Set<ControllerNode> peers = allNodes.values()
290 .stream()
291 .filter(node -> !(node.id().equals(localNode.id())))
292 .collect(Collectors.toSet());
Thomas Vachuska7a8de842016-03-07 20:56:35 -0800293 State state = nodeStates.get(localNode.id());
Jordan Haltermanf70bf462017-07-29 13:12:00 -0700294 byte[] hbMessagePayload = SERIALIZER.encode(new HeartbeatMessage(localNode, state, localVersion));
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800295 peers.forEach((node) -> {
296 heartbeatToPeer(hbMessagePayload, node);
297 State currentState = nodeStates.get(node.id());
298 double phi = failureDetector.phi(node.id());
sangyun-hanf98df542016-03-24 20:28:03 +0900299 if (phi >= phiFailureThreshold) {
Thomas Vachuska7a8de842016-03-07 20:56:35 -0800300 if (currentState.isActive()) {
Jordan Haltermanf70bf462017-07-29 13:12:00 -0700301 updateNode(node.id(), State.INACTIVE, null);
Jordan Haltermane7062532018-01-22 11:22:46 -0800302 failureDetector.reset(node.id());
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800303 }
304 } else {
305 if (currentState == State.INACTIVE) {
Jordan Haltermanf70bf462017-07-29 13:12:00 -0700306 updateNode(node.id(), State.ACTIVE, null);
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800307 }
308 }
309 });
310 } catch (Exception e) {
311 log.debug("Failed to send heartbeat", e);
312 }
tomb41d1ac2014-09-24 01:51:24 -0700313 }
314
Jordan Haltermanf70bf462017-07-29 13:12:00 -0700315 private void notifyChange(NodeId nodeId, State oldState, State newState, Version oldVersion, Version newVersion) {
316 if (oldState != newState || !Objects.equals(oldVersion, newVersion)) {
Thomas Vachuskafba7f3d2016-03-23 15:46:25 -0700317 ControllerNode node = allNodes.get(nodeId);
Jon Hall1f13d642016-05-13 17:53:01 -0700318 // Either this node or that node is no longer part of the same cluster
319 if (node == null) {
Jon Hall66870baa2016-06-20 12:12:10 -0700320 log.debug("Could not find node {} in the cluster, ignoring state change", nodeId);
Jon Hall1f13d642016-05-13 17:53:01 -0700321 return;
322 }
Thomas Vachuskafba7f3d2016-03-23 15:46:25 -0700323 ClusterEvent.Type type = newState == State.READY ? INSTANCE_READY :
324 newState == State.ACTIVE ? INSTANCE_ACTIVATED :
325 INSTANCE_DEACTIVATED;
326 notifyDelegate(new ClusterEvent(type, node));
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800327 }
tomb41d1ac2014-09-24 01:51:24 -0700328 }
329
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800330 private void heartbeatToPeer(byte[] messagePayload, ControllerNode peer) {
Madan Jampaniafeebbd2015-05-19 15:26:01 -0700331 Endpoint remoteEp = new Endpoint(peer.ip(), peer.tcpPort());
Madan Jampani175e8fd2015-05-20 14:10:45 -0700332 messagingService.sendAsync(remoteEp, HEARTBEAT_MESSAGE, messagePayload).whenComplete((result, error) -> {
333 if (error != null) {
334 log.trace("Sending heartbeat to {} failed", remoteEp, error);
335 }
336 });
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800337 }
338
Madan Jampanid36def02016-01-13 11:21:56 -0800339 private class HeartbeatMessageHandler implements BiConsumer<Endpoint, byte[]> {
tomb41d1ac2014-09-24 01:51:24 -0700340 @Override
Madan Jampanid36def02016-01-13 11:21:56 -0800341 public void accept(Endpoint sender, byte[] message) {
Madan Jampanic26eede2015-04-16 11:42:16 -0700342 HeartbeatMessage hb = SERIALIZER.decode(message);
Madan Jampanie3375062016-04-19 16:32:10 -0700343 if (clusterMetadataService.getClusterMetadata().getNodes().contains(hb.source())) {
Jordan Haltermane7062532018-01-22 11:22:46 -0800344 // Avoid reporting heartbeats that have been enqueued by setting a minimum interval.
345 long heartbeatTime = System.currentTimeMillis();
346 long lastHeartbeatTime = failureDetector.getLastHeartbeatTime(hb.source().id());
347 if (heartbeatTime - lastHeartbeatTime > heartbeatInterval / 2) {
348 failureDetector.report(hb.source().id(), heartbeatTime);
Jordan Halterman1315d3e2018-01-16 19:41:31 -0800349 }
Jordan Haltermanf70bf462017-07-29 13:12:00 -0700350 updateNode(hb.source().id(), hb.state, hb.version);
Madan Jampanie3375062016-04-19 16:32:10 -0700351 }
tomb41d1ac2014-09-24 01:51:24 -0700352 }
tom2d7c65f2014-09-23 01:09:35 -0700353 }
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800354
355 private static class HeartbeatMessage {
356 private ControllerNode source;
Thomas Vachuska7a8de842016-03-07 20:56:35 -0800357 private State state;
Jordan Haltermanf70bf462017-07-29 13:12:00 -0700358 private Version version;
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800359
Jordan Haltermanf70bf462017-07-29 13:12:00 -0700360 public HeartbeatMessage(ControllerNode source, State state, Version version) {
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800361 this.source = source;
Thomas Vachuska7a8de842016-03-07 20:56:35 -0800362 this.state = state != null ? state : State.ACTIVE;
Jordan Haltermanf70bf462017-07-29 13:12:00 -0700363 this.version = version;
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800364 }
365
366 public ControllerNode source() {
367 return source;
368 }
Ayaka Koshibedd91b842015-03-02 14:48:47 -0800369 }
370
Madan Jampani7d2fab22015-03-18 17:21:57 -0700371 @Override
Yuta HIGUCHI0c47d532017-08-18 23:16:35 -0700372 public Instant getLastUpdatedInstant(NodeId nodeId) {
Jordan Haltermanf70bf462017-07-29 13:12:00 -0700373 return nodeLastUpdatedTimes.get(nodeId);
Madan Jampani7d2fab22015-03-18 17:21:57 -0700374 }
Thomas Vachuskade563cf2015-04-01 00:28:50 -0700375
sangyun-hanf98df542016-03-24 20:28:03 +0900376 /**
Deepa Vaddireddy6690f592017-03-12 03:31:48 +0530377 * Extracts properties from the component configuration.
sangyun-hanf98df542016-03-24 20:28:03 +0900378 *
sangyun-hanf98df542016-03-24 20:28:03 +0900379 */
Deepa Vaddireddy6690f592017-03-12 03:31:48 +0530380 private void readComponentConfiguration() {
381 Set<ConfigProperty> configProperties = cfgService.getProperties(getClass().getName());
382 for (ConfigProperty property : configProperties) {
Jon Halla3fcf672017-03-28 16:53:22 -0700383 if ("heartbeatInterval".equals(property.name())) {
Deepa Vaddireddy6690f592017-03-12 03:31:48 +0530384 String s = property.value();
385 if (s == null) {
386 setHeartbeatInterval(DEFAULT_HEARTBEAT_INTERVAL);
387 log.info("Heartbeat interval time is not configured, default value is {}",
388 DEFAULT_HEARTBEAT_INTERVAL);
389 } else {
390 int newHeartbeatInterval = isNullOrEmpty(s) ? DEFAULT_HEARTBEAT_INTERVAL
Jordan Haltermanf70bf462017-07-29 13:12:00 -0700391 : Integer.parseInt(s.trim());
Deepa Vaddireddy6690f592017-03-12 03:31:48 +0530392 if (newHeartbeatInterval > 0 && heartbeatInterval != newHeartbeatInterval) {
393 heartbeatInterval = newHeartbeatInterval;
394 restartHeartbeatSender();
395 }
396 log.info("Configured. Heartbeat interval time is configured to {}",
397 heartbeatInterval);
398 }
399 }
Jon Halla3fcf672017-03-28 16:53:22 -0700400 if ("phiFailureThreshold".equals(property.name())) {
Deepa Vaddireddy6690f592017-03-12 03:31:48 +0530401 String s = property.value();
402 if (s == null) {
403 setPhiFailureThreshold(DEFAULT_PHI_FAILURE_THRESHOLD);
404 log.info("Phi failure threshold is not configured, default value is {}",
405 DEFAULT_PHI_FAILURE_THRESHOLD);
406 } else {
407 int newPhiFailureThreshold = isNullOrEmpty(s) ? DEFAULT_HEARTBEAT_INTERVAL
Jordan Haltermanf70bf462017-07-29 13:12:00 -0700408 : Integer.parseInt(s.trim());
Deepa Vaddireddy6690f592017-03-12 03:31:48 +0530409 setPhiFailureThreshold(newPhiFailureThreshold);
410 log.info("Configured. Phi failure threshold is configured to {}",
411 phiFailureThreshold);
412 }
413 }
Jordan Halterman4a082ec2018-05-14 15:00:42 -0700414 if ("minStandardDeviationMillis".equals(property.name())) {
415 String s = property.value();
416 if (s == null) {
417 setMinStandardDeviationMillis(DEFAULT_MIN_STANDARD_DEVIATION_MILLIS);
418 log.info("Minimum standard deviation is not configured, default value is {}",
419 DEFAULT_MIN_STANDARD_DEVIATION_MILLIS);
420 } else {
421 long newMinStandardDeviationMillis = isNullOrEmpty(s)
422 ? DEFAULT_MIN_STANDARD_DEVIATION_MILLIS
423 : Long.parseLong(s.trim());
424 setMinStandardDeviationMillis(newMinStandardDeviationMillis);
425 log.info("Configured. Minimum standard deviation is configured to {}",
426 newMinStandardDeviationMillis);
427 }
428 }
sangyun-hanf98df542016-03-24 20:28:03 +0900429 }
430 }
431
432 /**
433 * Sets heartbeat interval between the termination of one execution of heartbeat
434 * and the commencement of the next.
435 *
436 * @param interval term between each heartbeat
437 */
438 private void setHeartbeatInterval(int interval) {
439 try {
440 checkArgument(interval > 0, "Interval must be greater than zero");
441 heartbeatInterval = interval;
442 } catch (IllegalArgumentException e) {
443 log.warn(e.getMessage());
444 heartbeatInterval = DEFAULT_HEARTBEAT_INTERVAL;
445 }
446 }
447
448 /**
449 * Sets Phi failure threshold.
450 * Phi is based on a paper titled: "The φ Accrual Failure Detector" by Hayashibara, et al.
451 *
452 * @param threshold
453 */
454 private void setPhiFailureThreshold(int threshold) {
455 phiFailureThreshold = threshold;
456 }
457
458 /**
Jordan Halterman4a082ec2018-05-14 15:00:42 -0700459 * Sets the minimum standard deviation milliseconds.
460 *
461 * @param minStandardDeviationMillis the updated minimum standard deviation
462 */
463 private void setMinStandardDeviationMillis(long minStandardDeviationMillis) {
464 this.minStandardDeviationMillis = minStandardDeviationMillis;
465 try {
466 failureDetector = new PhiAccrualFailureDetector(minStandardDeviationMillis);
467 } catch (IllegalArgumentException e) {
468 log.warn(e.getMessage());
469 this.minStandardDeviationMillis = DEFAULT_MIN_STANDARD_DEVIATION_MILLIS;
470 failureDetector = new PhiAccrualFailureDetector(this.minStandardDeviationMillis);
471 }
472 }
473
474 /**
sangyun-hanf98df542016-03-24 20:28:03 +0900475 * Restarts heartbeatSender executor.
sangyun-hanf98df542016-03-24 20:28:03 +0900476 */
477 private void restartHeartbeatSender() {
478 try {
479 ScheduledExecutorService prevSender = heartBeatSender;
480 heartBeatSender = Executors.newSingleThreadScheduledExecutor(
HIGUCHI Yutad9e01052016-04-14 09:31:42 -0700481 groupedThreads("onos/cluster/membership", "heartbeat-sender-%d", log));
sangyun-hanf98df542016-03-24 20:28:03 +0900482 heartBeatSender.scheduleWithFixedDelay(this::heartbeat, 0,
Jordan Haltermanf70bf462017-07-29 13:12:00 -0700483 heartbeatInterval, TimeUnit.MILLISECONDS);
sangyun-hanf98df542016-03-24 20:28:03 +0900484 prevSender.shutdown();
485 } catch (Exception e) {
486 log.warn(e.getMessage());
487 }
488 }
Jordan Haltermanf70bf462017-07-29 13:12:00 -0700489}