Thomas Vachuska | 4f1a60c | 2014-10-28 13:39:07 -0700 | [diff] [blame] | 1 | /* |
Brian O'Connor | 5ab426f | 2016-04-09 01:19:45 -0700 | [diff] [blame] | 2 | * Copyright 2014-present Open Networking Laboratory |
Thomas Vachuska | 4f1a60c | 2014-10-28 13:39:07 -0700 | [diff] [blame] | 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; |
tom | 2d7c65f | 2014-09-23 01:09:35 -0700 | [diff] [blame] | 17 | |
HIGUCHI Yuta | 1979f55 | 2015-12-28 21:24:26 -0800 | [diff] [blame] | 18 | import com.google.common.base.MoreObjects; |
Jonathan Hart | 4a4d18f | 2015-03-26 12:16:16 -0700 | [diff] [blame] | 19 | import com.google.common.collect.ImmutableSet; |
| 20 | import com.google.common.collect.Maps; |
tom | 2d7c65f | 2014-09-23 01:09:35 -0700 | [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; |
sangyun-han | f98df54 | 2016-03-24 20:28:03 +0900 | [diff] [blame] | 24 | import org.apache.felix.scr.annotations.Modified; |
| 25 | import org.apache.felix.scr.annotations.Property; |
Madan Jampani | afeebbd | 2015-05-19 15:26:01 -0700 | [diff] [blame] | 26 | import org.apache.felix.scr.annotations.Reference; |
| 27 | import org.apache.felix.scr.annotations.ReferenceCardinality; |
tom | 2d7c65f | 2014-09-23 01:09:35 -0700 | [diff] [blame] | 28 | import org.apache.felix.scr.annotations.Service; |
Madan Jampani | 7d2fab2 | 2015-03-18 17:21:57 -0700 | [diff] [blame] | 29 | import org.joda.time.DateTime; |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 30 | import org.onlab.packet.IpAddress; |
| 31 | import org.onlab.util.KryoNamespace; |
sangyun-han | f98df54 | 2016-03-24 20:28:03 +0900 | [diff] [blame] | 32 | import org.onlab.util.Tools; |
Brian O'Connor | abafb50 | 2014-12-02 22:26:20 -0800 | [diff] [blame] | 33 | import org.onosproject.cluster.ClusterEvent; |
Madan Jampani | ec1df02 | 2015-10-13 21:23:03 -0700 | [diff] [blame] | 34 | import org.onosproject.cluster.ClusterMetadataService; |
Brian O'Connor | abafb50 | 2014-12-02 22:26:20 -0800 | [diff] [blame] | 35 | import org.onosproject.cluster.ClusterStore; |
| 36 | import org.onosproject.cluster.ClusterStoreDelegate; |
| 37 | import org.onosproject.cluster.ControllerNode; |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 38 | import org.onosproject.cluster.ControllerNode.State; |
Jonathan Hart | 4a4d18f | 2015-03-26 12:16:16 -0700 | [diff] [blame] | 39 | import org.onosproject.cluster.DefaultControllerNode; |
Brian O'Connor | abafb50 | 2014-12-02 22:26:20 -0800 | [diff] [blame] | 40 | import org.onosproject.cluster.NodeId; |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 41 | import org.onosproject.store.AbstractStore; |
Madan Jampani | c26eede | 2015-04-16 11:42:16 -0700 | [diff] [blame] | 42 | import org.onosproject.store.cluster.messaging.Endpoint; |
Madan Jampani | afeebbd | 2015-05-19 15:26:01 -0700 | [diff] [blame] | 43 | import org.onosproject.store.cluster.messaging.MessagingService; |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 44 | import org.onosproject.store.serializers.KryoNamespaces; |
| 45 | import org.onosproject.store.serializers.KryoSerializer; |
sangyun-han | f98df54 | 2016-03-24 20:28:03 +0900 | [diff] [blame] | 46 | import org.osgi.service.component.ComponentContext; |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 47 | import org.slf4j.Logger; |
tom | 2d7c65f | 2014-09-23 01:09:35 -0700 | [diff] [blame] | 48 | |
sangyun-han | f98df54 | 2016-03-24 20:28:03 +0900 | [diff] [blame] | 49 | import java.util.Dictionary; |
Jonathan Hart | 4a4d18f | 2015-03-26 12:16:16 -0700 | [diff] [blame] | 50 | import java.util.Map; |
Thomas Vachuska | fba7f3d | 2016-03-23 15:46:25 -0700 | [diff] [blame] | 51 | import java.util.Objects; |
Jonathan Hart | 4a4d18f | 2015-03-26 12:16:16 -0700 | [diff] [blame] | 52 | import java.util.Set; |
| 53 | import java.util.concurrent.ExecutorService; |
| 54 | import java.util.concurrent.Executors; |
| 55 | import java.util.concurrent.ScheduledExecutorService; |
| 56 | import java.util.concurrent.TimeUnit; |
Madan Jampani | d36def0 | 2016-01-13 11:21:56 -0800 | [diff] [blame] | 57 | import java.util.function.BiConsumer; |
Jonathan Hart | 4a4d18f | 2015-03-26 12:16:16 -0700 | [diff] [blame] | 58 | import java.util.stream.Collectors; |
| 59 | |
sangyun-han | f98df54 | 2016-03-24 20:28:03 +0900 | [diff] [blame] | 60 | import static com.google.common.base.Preconditions.checkArgument; |
Jonathan Hart | 4a4d18f | 2015-03-26 12:16:16 -0700 | [diff] [blame] | 61 | import static com.google.common.base.Preconditions.checkNotNull; |
| 62 | import static org.onlab.util.Tools.groupedThreads; |
Thomas Vachuska | fba7f3d | 2016-03-23 15:46:25 -0700 | [diff] [blame] | 63 | import static org.onosproject.cluster.ClusterEvent.Type.INSTANCE_ACTIVATED; |
| 64 | import static org.onosproject.cluster.ClusterEvent.Type.INSTANCE_DEACTIVATED; |
| 65 | import static org.onosproject.cluster.ClusterEvent.Type.INSTANCE_READY; |
Jonathan Hart | 4a4d18f | 2015-03-26 12:16:16 -0700 | [diff] [blame] | 66 | import static org.slf4j.LoggerFactory.getLogger; |
tom | 2d7c65f | 2014-09-23 01:09:35 -0700 | [diff] [blame] | 67 | |
tom | 2d7c65f | 2014-09-23 01:09:35 -0700 | [diff] [blame] | 68 | @Component(immediate = true) |
| 69 | @Service |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 70 | /** |
| 71 | * Distributed cluster nodes store that employs an accrual failure |
| 72 | * detector to identify cluster member up/down status. |
| 73 | */ |
tom | 0755a36 | 2014-09-24 11:54:43 -0700 | [diff] [blame] | 74 | public class DistributedClusterStore |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 75 | extends AbstractStore<ClusterEvent, ClusterStoreDelegate> |
tom | b41d1ac | 2014-09-24 01:51:24 -0700 | [diff] [blame] | 76 | implements ClusterStore { |
tom | 2d7c65f | 2014-09-23 01:09:35 -0700 | [diff] [blame] | 77 | |
Thomas Vachuska | 8dc1a69 | 2015-03-31 01:01:37 -0700 | [diff] [blame] | 78 | private static final Logger log = getLogger(DistributedClusterStore.class); |
tom | 2d7c65f | 2014-09-23 01:09:35 -0700 | [diff] [blame] | 79 | |
Thomas Vachuska | de563cf | 2015-04-01 00:28:50 -0700 | [diff] [blame] | 80 | public static final String HEARTBEAT_MESSAGE = "onos-cluster-heartbeat"; |
| 81 | |
sangyun-han | f98df54 | 2016-03-24 20:28:03 +0900 | [diff] [blame] | 82 | private static final int DEFAULT_HEARTBEAT_INTERVAL = 100; |
| 83 | @Property(name = "heartbeatInterval", intValue = DEFAULT_HEARTBEAT_INTERVAL, |
| 84 | label = "Interval time to send heartbeat to other controller nodes (millisecond)") |
| 85 | private int heartbeatInterval = DEFAULT_HEARTBEAT_INTERVAL; |
| 86 | |
| 87 | private static final int DEFAULT_PHI_FAILURE_THRESHOLD = 10; |
| 88 | @Property(name = "phiFailureThreshold", intValue = DEFAULT_PHI_FAILURE_THRESHOLD, |
| 89 | label = "the value of Phi threshold to detect accrual failure") |
| 90 | private int phiFailureThreshold = DEFAULT_PHI_FAILURE_THRESHOLD; |
tom | 2d7c65f | 2014-09-23 01:09:35 -0700 | [diff] [blame] | 91 | |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 92 | private static final KryoSerializer SERIALIZER = new KryoSerializer() { |
| 93 | @Override |
| 94 | protected void setupKryoPool() { |
| 95 | serializerPool = KryoNamespace.newBuilder() |
Thomas Vachuska | 8dc1a69 | 2015-03-31 01:01:37 -0700 | [diff] [blame] | 96 | .register(KryoNamespaces.API) |
| 97 | .register(HeartbeatMessage.class) |
| 98 | .build() |
| 99 | .populate(1); |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 100 | } |
| 101 | }; |
| 102 | |
| 103 | private static final String INSTANCE_ID_NULL = "Instance ID cannot be null"; |
| 104 | |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 105 | private final Map<NodeId, ControllerNode> allNodes = Maps.newConcurrentMap(); |
| 106 | private final Map<NodeId, State> nodeStates = Maps.newConcurrentMap(); |
Madan Jampani | 7d2fab2 | 2015-03-18 17:21:57 -0700 | [diff] [blame] | 107 | private final Map<NodeId, DateTime> nodeStateLastUpdatedTimes = Maps.newConcurrentMap(); |
Thomas Vachuska | 7a8de84 | 2016-03-07 20:56:35 -0800 | [diff] [blame] | 108 | |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 109 | private ScheduledExecutorService heartBeatSender = Executors.newSingleThreadScheduledExecutor( |
HIGUCHI Yuta | d9e0105 | 2016-04-14 09:31:42 -0700 | [diff] [blame^] | 110 | groupedThreads("onos/cluster/membership", "heartbeat-sender", log)); |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 111 | private ExecutorService heartBeatMessageHandler = Executors.newSingleThreadExecutor( |
HIGUCHI Yuta | d9e0105 | 2016-04-14 09:31:42 -0700 | [diff] [blame^] | 112 | groupedThreads("onos/cluster/membership", "heartbeat-receiver", log)); |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 113 | |
| 114 | private PhiAccrualFailureDetector failureDetector; |
| 115 | |
| 116 | private ControllerNode localNode; |
| 117 | |
Madan Jampani | afeebbd | 2015-05-19 15:26:01 -0700 | [diff] [blame] | 118 | @Reference(cardinality = ReferenceCardinality.MANDATORY_UNARY) |
Madan Jampani | ec1df02 | 2015-10-13 21:23:03 -0700 | [diff] [blame] | 119 | protected ClusterMetadataService clusterMetadataService; |
Madan Jampani | afeebbd | 2015-05-19 15:26:01 -0700 | [diff] [blame] | 120 | |
| 121 | @Reference(cardinality = ReferenceCardinality.MANDATORY_UNARY) |
| 122 | protected MessagingService messagingService; |
| 123 | |
tom | 2d7c65f | 2014-09-23 01:09:35 -0700 | [diff] [blame] | 124 | @Activate |
| 125 | public void activate() { |
Madan Jampani | ec1df02 | 2015-10-13 21:23:03 -0700 | [diff] [blame] | 126 | localNode = clusterMetadataService.getLocalNode(); |
Thomas Vachuska | 8dc1a69 | 2015-03-31 01:01:37 -0700 | [diff] [blame] | 127 | |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 128 | messagingService.registerHandler(HEARTBEAT_MESSAGE, |
Thomas Vachuska | 8dc1a69 | 2015-03-31 01:01:37 -0700 | [diff] [blame] | 129 | new HeartbeatMessageHandler(), heartBeatMessageHandler); |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 130 | |
| 131 | failureDetector = new PhiAccrualFailureDetector(); |
| 132 | |
| 133 | heartBeatSender.scheduleWithFixedDelay(this::heartbeat, 0, |
sangyun-han | f98df54 | 2016-03-24 20:28:03 +0900 | [diff] [blame] | 134 | heartbeatInterval, TimeUnit.MILLISECONDS); |
tom | b41d1ac | 2014-09-24 01:51:24 -0700 | [diff] [blame] | 135 | |
| 136 | log.info("Started"); |
| 137 | } |
| 138 | |
tom | 2d7c65f | 2014-09-23 01:09:35 -0700 | [diff] [blame] | 139 | @Deactivate |
| 140 | public void deactivate() { |
Madan Jampani | afeebbd | 2015-05-19 15:26:01 -0700 | [diff] [blame] | 141 | messagingService.unregisterHandler(HEARTBEAT_MESSAGE); |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 142 | heartBeatSender.shutdownNow(); |
| 143 | heartBeatMessageHandler.shutdownNow(); |
| 144 | |
tom | 2d7c65f | 2014-09-23 01:09:35 -0700 | [diff] [blame] | 145 | log.info("Stopped"); |
| 146 | } |
| 147 | |
sangyun-han | f98df54 | 2016-03-24 20:28:03 +0900 | [diff] [blame] | 148 | @Modified |
| 149 | public void modified(ComponentContext context) { |
| 150 | readComponentConfiguration(context); |
| 151 | restartHeartbeatSender(); |
| 152 | } |
| 153 | |
tom | 2d7c65f | 2014-09-23 01:09:35 -0700 | [diff] [blame] | 154 | @Override |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 155 | public void setDelegate(ClusterStoreDelegate delegate) { |
| 156 | checkNotNull(delegate, "Delegate cannot be null"); |
| 157 | this.delegate = delegate; |
| 158 | } |
| 159 | |
| 160 | @Override |
| 161 | public void unsetDelegate(ClusterStoreDelegate delegate) { |
| 162 | this.delegate = null; |
| 163 | } |
| 164 | |
| 165 | @Override |
| 166 | public boolean hasDelegate() { |
| 167 | return this.delegate != null; |
| 168 | } |
| 169 | |
| 170 | @Override |
tom | 2d7c65f | 2014-09-23 01:09:35 -0700 | [diff] [blame] | 171 | public ControllerNode getLocalNode() { |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 172 | return localNode; |
tom | 2d7c65f | 2014-09-23 01:09:35 -0700 | [diff] [blame] | 173 | } |
| 174 | |
| 175 | @Override |
| 176 | public Set<ControllerNode> getNodes() { |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 177 | return ImmutableSet.copyOf(allNodes.values()); |
tom | 2d7c65f | 2014-09-23 01:09:35 -0700 | [diff] [blame] | 178 | } |
| 179 | |
| 180 | @Override |
| 181 | public ControllerNode getNode(NodeId nodeId) { |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 182 | checkNotNull(nodeId, INSTANCE_ID_NULL); |
| 183 | return allNodes.get(nodeId); |
tom | 2d7c65f | 2014-09-23 01:09:35 -0700 | [diff] [blame] | 184 | } |
| 185 | |
| 186 | @Override |
tom | b41d1ac | 2014-09-24 01:51:24 -0700 | [diff] [blame] | 187 | public State getState(NodeId nodeId) { |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 188 | checkNotNull(nodeId, INSTANCE_ID_NULL); |
HIGUCHI Yuta | 1979f55 | 2015-12-28 21:24:26 -0800 | [diff] [blame] | 189 | return MoreObjects.firstNonNull(nodeStates.get(nodeId), State.INACTIVE); |
tom | b41d1ac | 2014-09-24 01:51:24 -0700 | [diff] [blame] | 190 | } |
| 191 | |
| 192 | @Override |
Thomas Vachuska | 7a8de84 | 2016-03-07 20:56:35 -0800 | [diff] [blame] | 193 | public void markFullyStarted(boolean started) { |
| 194 | updateState(localNode.id(), started ? State.READY : State.ACTIVE); |
| 195 | } |
| 196 | |
| 197 | @Override |
Pavlin Radoslavov | 444b519 | 2014-10-28 10:45:19 -0700 | [diff] [blame] | 198 | public ControllerNode addNode(NodeId nodeId, IpAddress ip, int tcpPort) { |
sangyun-han | f98df54 | 2016-03-24 20:28:03 +0900 | [diff] [blame] | 199 | checkNotNull(nodeId, INSTANCE_ID_NULL); |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 200 | ControllerNode node = new DefaultControllerNode(nodeId, ip, tcpPort); |
Madan Jampani | afeebbd | 2015-05-19 15:26:01 -0700 | [diff] [blame] | 201 | addNode(node); |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 202 | return node; |
tom | ee49c37 | 2014-09-26 15:14:50 -0700 | [diff] [blame] | 203 | } |
| 204 | |
| 205 | @Override |
tom | b41d1ac | 2014-09-24 01:51:24 -0700 | [diff] [blame] | 206 | public void removeNode(NodeId nodeId) { |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 207 | checkNotNull(nodeId, INSTANCE_ID_NULL); |
| 208 | ControllerNode node = allNodes.remove(nodeId); |
| 209 | if (node != null) { |
| 210 | nodeStates.remove(nodeId); |
Jonathan Hart | f114126 | 2015-04-23 11:27:07 -0700 | [diff] [blame] | 211 | notifyDelegate(new ClusterEvent(ClusterEvent.Type.INSTANCE_REMOVED, node)); |
tom | b41d1ac | 2014-09-24 01:51:24 -0700 | [diff] [blame] | 212 | } |
| 213 | } |
| 214 | |
Madan Jampani | afeebbd | 2015-05-19 15:26:01 -0700 | [diff] [blame] | 215 | private void addNode(ControllerNode node) { |
| 216 | allNodes.put(node.id(), node); |
Madan Jampani | ec1df02 | 2015-10-13 21:23:03 -0700 | [diff] [blame] | 217 | updateState(node.id(), node.equals(localNode) ? State.ACTIVE : State.INACTIVE); |
Madan Jampani | afeebbd | 2015-05-19 15:26:01 -0700 | [diff] [blame] | 218 | notifyDelegate(new ClusterEvent(ClusterEvent.Type.INSTANCE_ADDED, node)); |
Thomas Vachuska | de563cf | 2015-04-01 00:28:50 -0700 | [diff] [blame] | 219 | } |
| 220 | |
Madan Jampani | 7d2fab2 | 2015-03-18 17:21:57 -0700 | [diff] [blame] | 221 | private void updateState(NodeId nodeId, State newState) { |
Thomas Vachuska | fba7f3d | 2016-03-23 15:46:25 -0700 | [diff] [blame] | 222 | State currentState = nodeStates.get(nodeId); |
| 223 | if (!Objects.equals(currentState, newState)) { |
| 224 | nodeStates.put(nodeId, newState); |
| 225 | nodeStateLastUpdatedTimes.put(nodeId, DateTime.now()); |
| 226 | notifyStateChange(nodeId, currentState, newState); |
| 227 | } |
Madan Jampani | 7d2fab2 | 2015-03-18 17:21:57 -0700 | [diff] [blame] | 228 | } |
| 229 | |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 230 | private void heartbeat() { |
| 231 | try { |
| 232 | Set<ControllerNode> peers = allNodes.values() |
| 233 | .stream() |
| 234 | .filter(node -> !(node.id().equals(localNode.id()))) |
| 235 | .collect(Collectors.toSet()); |
Thomas Vachuska | 7a8de84 | 2016-03-07 20:56:35 -0800 | [diff] [blame] | 236 | State state = nodeStates.get(localNode.id()); |
| 237 | byte[] hbMessagePayload = SERIALIZER.encode(new HeartbeatMessage(localNode, state, peers)); |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 238 | peers.forEach((node) -> { |
| 239 | heartbeatToPeer(hbMessagePayload, node); |
| 240 | State currentState = nodeStates.get(node.id()); |
| 241 | double phi = failureDetector.phi(node.id()); |
sangyun-han | f98df54 | 2016-03-24 20:28:03 +0900 | [diff] [blame] | 242 | if (phi >= phiFailureThreshold) { |
Thomas Vachuska | 7a8de84 | 2016-03-07 20:56:35 -0800 | [diff] [blame] | 243 | if (currentState.isActive()) { |
Madan Jampani | 7d2fab2 | 2015-03-18 17:21:57 -0700 | [diff] [blame] | 244 | updateState(node.id(), State.INACTIVE); |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 245 | } |
| 246 | } else { |
| 247 | if (currentState == State.INACTIVE) { |
Madan Jampani | 7d2fab2 | 2015-03-18 17:21:57 -0700 | [diff] [blame] | 248 | updateState(node.id(), State.ACTIVE); |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 249 | } |
| 250 | } |
| 251 | }); |
| 252 | } catch (Exception e) { |
| 253 | log.debug("Failed to send heartbeat", e); |
| 254 | } |
tom | b41d1ac | 2014-09-24 01:51:24 -0700 | [diff] [blame] | 255 | } |
| 256 | |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 257 | private void notifyStateChange(NodeId nodeId, State oldState, State newState) { |
Thomas Vachuska | fba7f3d | 2016-03-23 15:46:25 -0700 | [diff] [blame] | 258 | if (oldState != newState) { |
| 259 | ControllerNode node = allNodes.get(nodeId); |
| 260 | ClusterEvent.Type type = newState == State.READY ? INSTANCE_READY : |
| 261 | newState == State.ACTIVE ? INSTANCE_ACTIVATED : |
| 262 | INSTANCE_DEACTIVATED; |
| 263 | notifyDelegate(new ClusterEvent(type, node)); |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 264 | } |
tom | b41d1ac | 2014-09-24 01:51:24 -0700 | [diff] [blame] | 265 | } |
| 266 | |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 267 | private void heartbeatToPeer(byte[] messagePayload, ControllerNode peer) { |
Madan Jampani | afeebbd | 2015-05-19 15:26:01 -0700 | [diff] [blame] | 268 | Endpoint remoteEp = new Endpoint(peer.ip(), peer.tcpPort()); |
Madan Jampani | 175e8fd | 2015-05-20 14:10:45 -0700 | [diff] [blame] | 269 | messagingService.sendAsync(remoteEp, HEARTBEAT_MESSAGE, messagePayload).whenComplete((result, error) -> { |
| 270 | if (error != null) { |
| 271 | log.trace("Sending heartbeat to {} failed", remoteEp, error); |
| 272 | } |
| 273 | }); |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 274 | } |
| 275 | |
Madan Jampani | d36def0 | 2016-01-13 11:21:56 -0800 | [diff] [blame] | 276 | private class HeartbeatMessageHandler implements BiConsumer<Endpoint, byte[]> { |
tom | b41d1ac | 2014-09-24 01:51:24 -0700 | [diff] [blame] | 277 | @Override |
Madan Jampani | d36def0 | 2016-01-13 11:21:56 -0800 | [diff] [blame] | 278 | public void accept(Endpoint sender, byte[] message) { |
Madan Jampani | c26eede | 2015-04-16 11:42:16 -0700 | [diff] [blame] | 279 | HeartbeatMessage hb = SERIALIZER.decode(message); |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 280 | failureDetector.report(hb.source().id()); |
Thomas Vachuska | 7a8de84 | 2016-03-07 20:56:35 -0800 | [diff] [blame] | 281 | updateState(hb.source().id(), hb.state); |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 282 | hb.knownPeers().forEach(node -> { |
| 283 | allNodes.put(node.id(), node); |
| 284 | }); |
tom | b41d1ac | 2014-09-24 01:51:24 -0700 | [diff] [blame] | 285 | } |
tom | 2d7c65f | 2014-09-23 01:09:35 -0700 | [diff] [blame] | 286 | } |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 287 | |
| 288 | private static class HeartbeatMessage { |
| 289 | private ControllerNode source; |
Thomas Vachuska | 7a8de84 | 2016-03-07 20:56:35 -0800 | [diff] [blame] | 290 | private State state; |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 291 | private Set<ControllerNode> knownPeers; |
| 292 | |
Thomas Vachuska | 7a8de84 | 2016-03-07 20:56:35 -0800 | [diff] [blame] | 293 | public HeartbeatMessage(ControllerNode source, State state, Set<ControllerNode> members) { |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 294 | this.source = source; |
Thomas Vachuska | 7a8de84 | 2016-03-07 20:56:35 -0800 | [diff] [blame] | 295 | this.state = state != null ? state : State.ACTIVE; |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 296 | this.knownPeers = ImmutableSet.copyOf(members); |
| 297 | } |
| 298 | |
| 299 | public ControllerNode source() { |
| 300 | return source; |
| 301 | } |
| 302 | |
| 303 | public Set<ControllerNode> knownPeers() { |
| 304 | return knownPeers; |
| 305 | } |
| 306 | } |
| 307 | |
Madan Jampani | 7d2fab2 | 2015-03-18 17:21:57 -0700 | [diff] [blame] | 308 | @Override |
| 309 | public DateTime getLastUpdated(NodeId nodeId) { |
| 310 | return nodeStateLastUpdatedTimes.get(nodeId); |
| 311 | } |
Thomas Vachuska | de563cf | 2015-04-01 00:28:50 -0700 | [diff] [blame] | 312 | |
sangyun-han | f98df54 | 2016-03-24 20:28:03 +0900 | [diff] [blame] | 313 | /** |
| 314 | * Extracts properties from the component configuration context. |
| 315 | * |
| 316 | * @param context the component context |
| 317 | */ |
| 318 | private void readComponentConfiguration(ComponentContext context) { |
| 319 | Dictionary<?, ?> properties = context.getProperties(); |
| 320 | |
| 321 | Integer newHeartbeatInterval = Tools.getIntegerProperty(properties, |
| 322 | "heartbeatInterval"); |
| 323 | if (newHeartbeatInterval == null) { |
| 324 | setHeartbeatInterval(DEFAULT_HEARTBEAT_INTERVAL); |
| 325 | log.info("Heartbeat interval time is not configured, default value is {}", |
| 326 | DEFAULT_HEARTBEAT_INTERVAL); |
| 327 | } else { |
| 328 | setHeartbeatInterval(newHeartbeatInterval); |
| 329 | log.info("Configured. Heartbeat interval time is configured to {}", |
| 330 | heartbeatInterval); |
| 331 | } |
| 332 | |
| 333 | Integer newPhiFailureThreshold = Tools.getIntegerProperty(properties, |
| 334 | "phiFailureThreshold"); |
| 335 | if (newPhiFailureThreshold == null) { |
| 336 | setPhiFailureThreshold(DEFAULT_PHI_FAILURE_THRESHOLD); |
| 337 | log.info("Phi failure threshold is not configured, default value is {}", |
| 338 | DEFAULT_PHI_FAILURE_THRESHOLD); |
| 339 | } else { |
| 340 | setPhiFailureThreshold(newPhiFailureThreshold); |
| 341 | log.info("Configured. Phi failure threshold is configured to {}", |
| 342 | phiFailureThreshold); |
| 343 | } |
| 344 | } |
| 345 | |
| 346 | /** |
| 347 | * Sets heartbeat interval between the termination of one execution of heartbeat |
| 348 | * and the commencement of the next. |
| 349 | * |
| 350 | * @param interval term between each heartbeat |
| 351 | */ |
| 352 | private void setHeartbeatInterval(int interval) { |
| 353 | try { |
| 354 | checkArgument(interval > 0, "Interval must be greater than zero"); |
| 355 | heartbeatInterval = interval; |
| 356 | } catch (IllegalArgumentException e) { |
| 357 | log.warn(e.getMessage()); |
| 358 | heartbeatInterval = DEFAULT_HEARTBEAT_INTERVAL; |
| 359 | } |
| 360 | } |
| 361 | |
| 362 | /** |
| 363 | * Sets Phi failure threshold. |
| 364 | * Phi is based on a paper titled: "The φ Accrual Failure Detector" by Hayashibara, et al. |
| 365 | * |
| 366 | * @param threshold |
| 367 | */ |
| 368 | private void setPhiFailureThreshold(int threshold) { |
| 369 | phiFailureThreshold = threshold; |
| 370 | } |
| 371 | |
| 372 | /** |
| 373 | * Restarts heartbeatSender executor. |
| 374 | * |
| 375 | */ |
| 376 | private void restartHeartbeatSender() { |
| 377 | try { |
| 378 | ScheduledExecutorService prevSender = heartBeatSender; |
| 379 | heartBeatSender = Executors.newSingleThreadScheduledExecutor( |
HIGUCHI Yuta | d9e0105 | 2016-04-14 09:31:42 -0700 | [diff] [blame^] | 380 | groupedThreads("onos/cluster/membership", "heartbeat-sender-%d", log)); |
sangyun-han | f98df54 | 2016-03-24 20:28:03 +0900 | [diff] [blame] | 381 | heartBeatSender.scheduleWithFixedDelay(this::heartbeat, 0, |
| 382 | heartbeatInterval, TimeUnit.MILLISECONDS); |
| 383 | prevSender.shutdown(); |
| 384 | } catch (Exception e) { |
| 385 | log.warn(e.getMessage()); |
| 386 | } |
| 387 | } |
| 388 | |
| 389 | /** |
| 390 | * Gets current heartbeat interval. |
| 391 | * |
| 392 | * @return heartbeatInterval |
| 393 | */ |
| 394 | private int getHeartbeatInterval() { |
| 395 | return heartbeatInterval; |
| 396 | } |
| 397 | |
| 398 | /** |
| 399 | * Gets current Phi failure threshold for Accrual Failure Detector. |
| 400 | * |
| 401 | * @return phiFailureThreshold |
| 402 | */ |
| 403 | private int getPhiFailureThreshold() { |
| 404 | return phiFailureThreshold; |
| 405 | } |
| 406 | |
| 407 | } |