Thomas Vachuska | 4f1a60c | 2014-10-28 13:39:07 -0700 | [diff] [blame] | 1 | /* |
Brian O'Connor | a09fe5b | 2017-08-03 21:12:30 -0700 | [diff] [blame] | 2 | * Copyright 2014-present Open Networking Foundation |
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 | |
Jonathan Hart | 4a4d18f | 2015-03-26 12:16:16 -0700 | [diff] [blame] | 18 | import com.google.common.collect.ImmutableSet; |
| 19 | import com.google.common.collect.Maps; |
Madan Jampani | e337506 | 2016-04-19 16:32:10 -0700 | [diff] [blame] | 20 | |
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; |
Deepa Vaddireddy | 6690f59 | 2017-03-12 03:31:48 +0530 | [diff] [blame] | 28 | import org.apache.felix.scr.annotations.ReferencePolicy; |
tom | 2d7c65f | 2014-09-23 01:09:35 -0700 | [diff] [blame] | 29 | import org.apache.felix.scr.annotations.Service; |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 30 | import org.onlab.packet.IpAddress; |
| 31 | import org.onlab.util.KryoNamespace; |
Deepa Vaddireddy | 6690f59 | 2017-03-12 03:31:48 +0530 | [diff] [blame] | 32 | import org.onosproject.cfg.ConfigProperty; |
sangyun-han | 9f0af2d | 2016-08-04 13:04:59 +0900 | [diff] [blame] | 33 | import org.onosproject.cfg.ComponentConfigService; |
Brian O'Connor | abafb50 | 2014-12-02 22:26:20 -0800 | [diff] [blame] | 34 | import org.onosproject.cluster.ClusterEvent; |
Madan Jampani | ec1df02 | 2015-10-13 21:23:03 -0700 | [diff] [blame] | 35 | import org.onosproject.cluster.ClusterMetadataService; |
Brian O'Connor | abafb50 | 2014-12-02 22:26:20 -0800 | [diff] [blame] | 36 | import org.onosproject.cluster.ClusterStore; |
| 37 | import org.onosproject.cluster.ClusterStoreDelegate; |
| 38 | import org.onosproject.cluster.ControllerNode; |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 39 | import org.onosproject.cluster.ControllerNode.State; |
Jonathan Hart | 4a4d18f | 2015-03-26 12:16:16 -0700 | [diff] [blame] | 40 | import org.onosproject.cluster.DefaultControllerNode; |
Brian O'Connor | abafb50 | 2014-12-02 22:26:20 -0800 | [diff] [blame] | 41 | import org.onosproject.cluster.NodeId; |
Jordan Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 42 | import org.onosproject.core.Version; |
| 43 | import org.onosproject.core.VersionService; |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 44 | import org.onosproject.store.AbstractStore; |
Madan Jampani | c26eede | 2015-04-16 11:42:16 -0700 | [diff] [blame] | 45 | import org.onosproject.store.cluster.messaging.Endpoint; |
Madan Jampani | afeebbd | 2015-05-19 15:26:01 -0700 | [diff] [blame] | 46 | import org.onosproject.store.cluster.messaging.MessagingService; |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 47 | import org.onosproject.store.serializers.KryoNamespaces; |
Jordan Halterman | 2c83a10 | 2017-08-20 17:11:41 -0700 | [diff] [blame] | 48 | import org.onosproject.store.service.Serializer; |
sangyun-han | f98df54 | 2016-03-24 20:28:03 +0900 | [diff] [blame] | 49 | import org.osgi.service.component.ComponentContext; |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 50 | import org.slf4j.Logger; |
tom | 2d7c65f | 2014-09-23 01:09:35 -0700 | [diff] [blame] | 51 | |
Yuta HIGUCHI | 0c47d53 | 2017-08-18 23:16:35 -0700 | [diff] [blame] | 52 | import java.time.Instant; |
Jonathan Hart | 4a4d18f | 2015-03-26 12:16:16 -0700 | [diff] [blame] | 53 | import java.util.Map; |
Thomas Vachuska | fba7f3d | 2016-03-23 15:46:25 -0700 | [diff] [blame] | 54 | import java.util.Objects; |
Jonathan Hart | 4a4d18f | 2015-03-26 12:16:16 -0700 | [diff] [blame] | 55 | import java.util.Set; |
| 56 | import java.util.concurrent.ExecutorService; |
| 57 | import java.util.concurrent.Executors; |
| 58 | import java.util.concurrent.ScheduledExecutorService; |
| 59 | import java.util.concurrent.TimeUnit; |
Madan Jampani | d36def0 | 2016-01-13 11:21:56 -0800 | [diff] [blame] | 60 | import java.util.function.BiConsumer; |
Jonathan Hart | 4a4d18f | 2015-03-26 12:16:16 -0700 | [diff] [blame] | 61 | import java.util.stream.Collectors; |
| 62 | |
Jordan Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 63 | import static com.google.common.base.MoreObjects.firstNonNull; |
sangyun-han | f98df54 | 2016-03-24 20:28:03 +0900 | [diff] [blame] | 64 | import static com.google.common.base.Preconditions.checkArgument; |
Jonathan Hart | 4a4d18f | 2015-03-26 12:16:16 -0700 | [diff] [blame] | 65 | import static com.google.common.base.Preconditions.checkNotNull; |
Deepa Vaddireddy | 6690f59 | 2017-03-12 03:31:48 +0530 | [diff] [blame] | 66 | import static com.google.common.base.Strings.isNullOrEmpty; |
Jonathan Hart | 4a4d18f | 2015-03-26 12:16:16 -0700 | [diff] [blame] | 67 | import static org.onlab.util.Tools.groupedThreads; |
Thomas Vachuska | fba7f3d | 2016-03-23 15:46:25 -0700 | [diff] [blame] | 68 | import static org.onosproject.cluster.ClusterEvent.Type.INSTANCE_ACTIVATED; |
| 69 | import static org.onosproject.cluster.ClusterEvent.Type.INSTANCE_DEACTIVATED; |
| 70 | import static org.onosproject.cluster.ClusterEvent.Type.INSTANCE_READY; |
Jonathan Hart | 4a4d18f | 2015-03-26 12:16:16 -0700 | [diff] [blame] | 71 | import static org.slf4j.LoggerFactory.getLogger; |
tom | 2d7c65f | 2014-09-23 01:09:35 -0700 | [diff] [blame] | 72 | |
tom | 2d7c65f | 2014-09-23 01:09:35 -0700 | [diff] [blame] | 73 | @Component(immediate = true) |
| 74 | @Service |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 75 | /** |
| 76 | * Distributed cluster nodes store that employs an accrual failure |
| 77 | * detector to identify cluster member up/down status. |
| 78 | */ |
tom | 0755a36 | 2014-09-24 11:54:43 -0700 | [diff] [blame] | 79 | public class DistributedClusterStore |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 80 | extends AbstractStore<ClusterEvent, ClusterStoreDelegate> |
tom | b41d1ac | 2014-09-24 01:51:24 -0700 | [diff] [blame] | 81 | implements ClusterStore { |
tom | 2d7c65f | 2014-09-23 01:09:35 -0700 | [diff] [blame] | 82 | |
Thomas Vachuska | 8dc1a69 | 2015-03-31 01:01:37 -0700 | [diff] [blame] | 83 | private static final Logger log = getLogger(DistributedClusterStore.class); |
tom | 2d7c65f | 2014-09-23 01:09:35 -0700 | [diff] [blame] | 84 | |
Thomas Vachuska | de563cf | 2015-04-01 00:28:50 -0700 | [diff] [blame] | 85 | public static final String HEARTBEAT_MESSAGE = "onos-cluster-heartbeat"; |
| 86 | |
sangyun-han | f98df54 | 2016-03-24 20:28:03 +0900 | [diff] [blame] | 87 | private static final int DEFAULT_HEARTBEAT_INTERVAL = 100; |
| 88 | @Property(name = "heartbeatInterval", intValue = DEFAULT_HEARTBEAT_INTERVAL, |
| 89 | label = "Interval time to send heartbeat to other controller nodes (millisecond)") |
| 90 | private int heartbeatInterval = DEFAULT_HEARTBEAT_INTERVAL; |
| 91 | |
| 92 | private static final int DEFAULT_PHI_FAILURE_THRESHOLD = 10; |
| 93 | @Property(name = "phiFailureThreshold", intValue = DEFAULT_PHI_FAILURE_THRESHOLD, |
| 94 | label = "the value of Phi threshold to detect accrual failure") |
| 95 | private int phiFailureThreshold = DEFAULT_PHI_FAILURE_THRESHOLD; |
tom | 2d7c65f | 2014-09-23 01:09:35 -0700 | [diff] [blame] | 96 | |
Jordan Halterman | b75aa08 | 2018-05-14 15:00:42 -0700 | [diff] [blame] | 97 | private static final long DEFAULT_MIN_STANDARD_DEVIATION_MILLIS = 50; |
| 98 | @Property(name = "minStandardDeviationMillis", longValue = DEFAULT_MIN_STANDARD_DEVIATION_MILLIS, |
| 99 | label = "The minimum standard deviation to take into account when computing the Phi value") |
| 100 | private long minStandardDeviationMillis = DEFAULT_MIN_STANDARD_DEVIATION_MILLIS; |
| 101 | |
Jordan Halterman | 2c83a10 | 2017-08-20 17:11:41 -0700 | [diff] [blame] | 102 | private static final Serializer SERIALIZER = Serializer.using( |
Jordan Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 103 | KryoNamespace.newBuilder() |
Thomas Vachuska | 8dc1a69 | 2015-03-31 01:01:37 -0700 | [diff] [blame] | 104 | .register(KryoNamespaces.API) |
HIGUCHI Yuta | e729065 | 2016-05-18 11:29:01 -0700 | [diff] [blame] | 105 | .nextId(KryoNamespaces.BEGIN_USER_CUSTOM_ID) |
Thomas Vachuska | 8dc1a69 | 2015-03-31 01:01:37 -0700 | [diff] [blame] | 106 | .register(HeartbeatMessage.class) |
HIGUCHI Yuta | e729065 | 2016-05-18 11:29:01 -0700 | [diff] [blame] | 107 | .build("ClusterStore")); |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 108 | |
| 109 | private static final String INSTANCE_ID_NULL = "Instance ID cannot be null"; |
| 110 | |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 111 | private final Map<NodeId, ControllerNode> allNodes = Maps.newConcurrentMap(); |
| 112 | private final Map<NodeId, State> nodeStates = Maps.newConcurrentMap(); |
Jordan Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 113 | private final Map<NodeId, Version> nodeVersions = Maps.newConcurrentMap(); |
Yuta HIGUCHI | 0c47d53 | 2017-08-18 23:16:35 -0700 | [diff] [blame] | 114 | private final Map<NodeId, Instant> nodeLastUpdatedTimes = Maps.newConcurrentMap(); |
Thomas Vachuska | 7a8de84 | 2016-03-07 20:56:35 -0800 | [diff] [blame] | 115 | |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 116 | private ScheduledExecutorService heartBeatSender = Executors.newSingleThreadScheduledExecutor( |
HIGUCHI Yuta | d9e0105 | 2016-04-14 09:31:42 -0700 | [diff] [blame] | 117 | groupedThreads("onos/cluster/membership", "heartbeat-sender", log)); |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 118 | private ExecutorService heartBeatMessageHandler = Executors.newSingleThreadExecutor( |
HIGUCHI Yuta | d9e0105 | 2016-04-14 09:31:42 -0700 | [diff] [blame] | 119 | groupedThreads("onos/cluster/membership", "heartbeat-receiver", log)); |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 120 | |
| 121 | private PhiAccrualFailureDetector failureDetector; |
| 122 | |
| 123 | private ControllerNode localNode; |
Jordan Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 124 | private Version localVersion; |
| 125 | |
| 126 | @Reference(cardinality = ReferenceCardinality.MANDATORY_UNARY) |
| 127 | protected VersionService versionService; |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 128 | |
Madan Jampani | afeebbd | 2015-05-19 15:26:01 -0700 | [diff] [blame] | 129 | @Reference(cardinality = ReferenceCardinality.MANDATORY_UNARY) |
Madan Jampani | ec1df02 | 2015-10-13 21:23:03 -0700 | [diff] [blame] | 130 | protected ClusterMetadataService clusterMetadataService; |
Madan Jampani | afeebbd | 2015-05-19 15:26:01 -0700 | [diff] [blame] | 131 | |
| 132 | @Reference(cardinality = ReferenceCardinality.MANDATORY_UNARY) |
| 133 | protected MessagingService messagingService; |
| 134 | |
Thomas Vachuska | 1ca7e9f | 2016-08-05 10:21:41 -0700 | [diff] [blame] | 135 | // This must be optional to avoid a cyclic dependency |
Deepa Vaddireddy | 6690f59 | 2017-03-12 03:31:48 +0530 | [diff] [blame] | 136 | @Reference(cardinality = ReferenceCardinality.OPTIONAL_UNARY, |
Jordan Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 137 | bind = "bindComponentConfigService", |
| 138 | unbind = "unbindComponentConfigService", |
| 139 | policy = ReferencePolicy.DYNAMIC) |
sangyun-han | 9f0af2d | 2016-08-04 13:04:59 +0900 | [diff] [blame] | 140 | protected ComponentConfigService cfgService; |
| 141 | |
Thomas Vachuska | 1ca7e9f | 2016-08-05 10:21:41 -0700 | [diff] [blame] | 142 | /** |
| 143 | * Hook for wiring up optional reference to a service. |
| 144 | * |
| 145 | * @param service service being announced |
| 146 | */ |
| 147 | protected void bindComponentConfigService(ComponentConfigService service) { |
| 148 | if (cfgService == null) { |
| 149 | cfgService = service; |
| 150 | cfgService.registerProperties(getClass()); |
Deepa Vaddireddy | 6690f59 | 2017-03-12 03:31:48 +0530 | [diff] [blame] | 151 | readComponentConfiguration(); |
Thomas Vachuska | 1ca7e9f | 2016-08-05 10:21:41 -0700 | [diff] [blame] | 152 | } |
| 153 | } |
| 154 | |
| 155 | /** |
| 156 | * Hook for unwiring optional reference to a service. |
| 157 | * |
| 158 | * @param service service being withdrawn |
| 159 | */ |
| 160 | protected void unbindComponentConfigService(ComponentConfigService service) { |
| 161 | if (cfgService == service) { |
| 162 | cfgService.unregisterProperties(getClass(), false); |
| 163 | cfgService = null; |
| 164 | } |
| 165 | } |
| 166 | |
tom | 2d7c65f | 2014-09-23 01:09:35 -0700 | [diff] [blame] | 167 | @Activate |
Thomas Vachuska | 1ca7e9f | 2016-08-05 10:21:41 -0700 | [diff] [blame] | 168 | public void activate() { |
Madan Jampani | ec1df02 | 2015-10-13 21:23:03 -0700 | [diff] [blame] | 169 | localNode = clusterMetadataService.getLocalNode(); |
Jordan Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 170 | localVersion = versionService.version(); |
| 171 | nodeVersions.put(localNode.id(), localVersion); |
Thomas Vachuska | 8dc1a69 | 2015-03-31 01:01:37 -0700 | [diff] [blame] | 172 | |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 173 | messagingService.registerHandler(HEARTBEAT_MESSAGE, |
Jordan Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 174 | new HeartbeatMessageHandler(), heartBeatMessageHandler); |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 175 | |
Jordan Halterman | b75aa08 | 2018-05-14 15:00:42 -0700 | [diff] [blame] | 176 | failureDetector = new PhiAccrualFailureDetector(minStandardDeviationMillis); |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 177 | |
| 178 | heartBeatSender.scheduleWithFixedDelay(this::heartbeat, 0, |
Jordan Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 179 | heartbeatInterval, TimeUnit.MILLISECONDS); |
tom | b41d1ac | 2014-09-24 01:51:24 -0700 | [diff] [blame] | 180 | |
| 181 | log.info("Started"); |
| 182 | } |
| 183 | |
tom | 2d7c65f | 2014-09-23 01:09:35 -0700 | [diff] [blame] | 184 | @Deactivate |
| 185 | public void deactivate() { |
Madan Jampani | afeebbd | 2015-05-19 15:26:01 -0700 | [diff] [blame] | 186 | messagingService.unregisterHandler(HEARTBEAT_MESSAGE); |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 187 | heartBeatSender.shutdownNow(); |
| 188 | heartBeatMessageHandler.shutdownNow(); |
| 189 | |
tom | 2d7c65f | 2014-09-23 01:09:35 -0700 | [diff] [blame] | 190 | log.info("Stopped"); |
| 191 | } |
| 192 | |
sangyun-han | f98df54 | 2016-03-24 20:28:03 +0900 | [diff] [blame] | 193 | @Modified |
| 194 | public void modified(ComponentContext context) { |
Deepa Vaddireddy | 6690f59 | 2017-03-12 03:31:48 +0530 | [diff] [blame] | 195 | readComponentConfiguration(); |
sangyun-han | f98df54 | 2016-03-24 20:28:03 +0900 | [diff] [blame] | 196 | } |
| 197 | |
tom | 2d7c65f | 2014-09-23 01:09:35 -0700 | [diff] [blame] | 198 | @Override |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 199 | public void setDelegate(ClusterStoreDelegate delegate) { |
| 200 | checkNotNull(delegate, "Delegate cannot be null"); |
| 201 | this.delegate = delegate; |
| 202 | } |
| 203 | |
| 204 | @Override |
| 205 | public void unsetDelegate(ClusterStoreDelegate delegate) { |
| 206 | this.delegate = null; |
| 207 | } |
| 208 | |
| 209 | @Override |
| 210 | public boolean hasDelegate() { |
| 211 | return this.delegate != null; |
| 212 | } |
| 213 | |
| 214 | @Override |
tom | 2d7c65f | 2014-09-23 01:09:35 -0700 | [diff] [blame] | 215 | public ControllerNode getLocalNode() { |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 216 | return localNode; |
tom | 2d7c65f | 2014-09-23 01:09:35 -0700 | [diff] [blame] | 217 | } |
| 218 | |
| 219 | @Override |
| 220 | public Set<ControllerNode> getNodes() { |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 221 | return ImmutableSet.copyOf(allNodes.values()); |
tom | 2d7c65f | 2014-09-23 01:09:35 -0700 | [diff] [blame] | 222 | } |
| 223 | |
| 224 | @Override |
| 225 | public ControllerNode getNode(NodeId nodeId) { |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 226 | checkNotNull(nodeId, INSTANCE_ID_NULL); |
| 227 | return allNodes.get(nodeId); |
tom | 2d7c65f | 2014-09-23 01:09:35 -0700 | [diff] [blame] | 228 | } |
| 229 | |
| 230 | @Override |
tom | b41d1ac | 2014-09-24 01:51:24 -0700 | [diff] [blame] | 231 | public State getState(NodeId nodeId) { |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 232 | checkNotNull(nodeId, INSTANCE_ID_NULL); |
Jordan Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 233 | return firstNonNull(nodeStates.get(nodeId), State.INACTIVE); |
| 234 | } |
| 235 | |
| 236 | @Override |
| 237 | public Version getVersion(NodeId nodeId) { |
| 238 | checkNotNull(nodeId, INSTANCE_ID_NULL); |
| 239 | return nodeVersions.get(nodeId); |
tom | b41d1ac | 2014-09-24 01:51:24 -0700 | [diff] [blame] | 240 | } |
| 241 | |
| 242 | @Override |
Thomas Vachuska | 7a8de84 | 2016-03-07 20:56:35 -0800 | [diff] [blame] | 243 | public void markFullyStarted(boolean started) { |
Jordan Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 244 | updateNode(localNode.id(), started ? State.READY : State.ACTIVE, null); |
Thomas Vachuska | 7a8de84 | 2016-03-07 20:56:35 -0800 | [diff] [blame] | 245 | } |
| 246 | |
| 247 | @Override |
Pavlin Radoslavov | 444b519 | 2014-10-28 10:45:19 -0700 | [diff] [blame] | 248 | public ControllerNode addNode(NodeId nodeId, IpAddress ip, int tcpPort) { |
sangyun-han | f98df54 | 2016-03-24 20:28:03 +0900 | [diff] [blame] | 249 | checkNotNull(nodeId, INSTANCE_ID_NULL); |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 250 | ControllerNode node = new DefaultControllerNode(nodeId, ip, tcpPort); |
Madan Jampani | afeebbd | 2015-05-19 15:26:01 -0700 | [diff] [blame] | 251 | addNode(node); |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 252 | return node; |
tom | ee49c37 | 2014-09-26 15:14:50 -0700 | [diff] [blame] | 253 | } |
| 254 | |
| 255 | @Override |
tom | b41d1ac | 2014-09-24 01:51:24 -0700 | [diff] [blame] | 256 | public void removeNode(NodeId nodeId) { |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 257 | checkNotNull(nodeId, INSTANCE_ID_NULL); |
| 258 | ControllerNode node = allNodes.remove(nodeId); |
| 259 | if (node != null) { |
| 260 | nodeStates.remove(nodeId); |
Jordan Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 261 | nodeVersions.remove(nodeId); |
Jonathan Hart | f114126 | 2015-04-23 11:27:07 -0700 | [diff] [blame] | 262 | notifyDelegate(new ClusterEvent(ClusterEvent.Type.INSTANCE_REMOVED, node)); |
tom | b41d1ac | 2014-09-24 01:51:24 -0700 | [diff] [blame] | 263 | } |
| 264 | } |
| 265 | |
Madan Jampani | afeebbd | 2015-05-19 15:26:01 -0700 | [diff] [blame] | 266 | private void addNode(ControllerNode node) { |
| 267 | allNodes.put(node.id(), node); |
Jordan Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 268 | updateNode(node.id(), node.equals(localNode) ? State.ACTIVE : State.INACTIVE, null); |
Madan Jampani | afeebbd | 2015-05-19 15:26:01 -0700 | [diff] [blame] | 269 | notifyDelegate(new ClusterEvent(ClusterEvent.Type.INSTANCE_ADDED, node)); |
Thomas Vachuska | de563cf | 2015-04-01 00:28:50 -0700 | [diff] [blame] | 270 | } |
| 271 | |
Jordan Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 272 | private void updateNode(NodeId nodeId, State newState, Version newVersion) { |
Thomas Vachuska | fba7f3d | 2016-03-23 15:46:25 -0700 | [diff] [blame] | 273 | State currentState = nodeStates.get(nodeId); |
Jordan Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 274 | Version currentVersion = nodeVersions.get(nodeId); |
Jordan Halterman | eabae96 | 2017-08-01 17:29:36 -0700 | [diff] [blame] | 275 | if (!Objects.equals(currentState, newState) |
| 276 | || (newVersion != null && !Objects.equals(currentVersion, newVersion))) { |
Thomas Vachuska | fba7f3d | 2016-03-23 15:46:25 -0700 | [diff] [blame] | 277 | nodeStates.put(nodeId, newState); |
Jordan Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 278 | if (newVersion != null) { |
| 279 | nodeVersions.put(nodeId, newVersion); |
| 280 | } |
Yuta HIGUCHI | 0c47d53 | 2017-08-18 23:16:35 -0700 | [diff] [blame] | 281 | nodeLastUpdatedTimes.put(nodeId, Instant.now()); |
Jordan Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 282 | notifyChange(nodeId, currentState, newState, currentVersion, newVersion); |
Thomas Vachuska | fba7f3d | 2016-03-23 15:46:25 -0700 | [diff] [blame] | 283 | } |
Madan Jampani | 7d2fab2 | 2015-03-18 17:21:57 -0700 | [diff] [blame] | 284 | } |
| 285 | |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 286 | private void heartbeat() { |
| 287 | try { |
| 288 | Set<ControllerNode> peers = allNodes.values() |
| 289 | .stream() |
| 290 | .filter(node -> !(node.id().equals(localNode.id()))) |
| 291 | .collect(Collectors.toSet()); |
Thomas Vachuska | 7a8de84 | 2016-03-07 20:56:35 -0800 | [diff] [blame] | 292 | State state = nodeStates.get(localNode.id()); |
Jordan Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 293 | byte[] hbMessagePayload = SERIALIZER.encode(new HeartbeatMessage(localNode, state, localVersion)); |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 294 | peers.forEach((node) -> { |
| 295 | heartbeatToPeer(hbMessagePayload, node); |
| 296 | State currentState = nodeStates.get(node.id()); |
| 297 | double phi = failureDetector.phi(node.id()); |
sangyun-han | f98df54 | 2016-03-24 20:28:03 +0900 | [diff] [blame] | 298 | if (phi >= phiFailureThreshold) { |
Thomas Vachuska | 7a8de84 | 2016-03-07 20:56:35 -0800 | [diff] [blame] | 299 | if (currentState.isActive()) { |
Jordan Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 300 | updateNode(node.id(), State.INACTIVE, null); |
Jordan Halterman | e706253 | 2018-01-22 11:22:46 -0800 | [diff] [blame] | 301 | failureDetector.reset(node.id()); |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 302 | } |
| 303 | } else { |
| 304 | if (currentState == State.INACTIVE) { |
Jordan Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 305 | updateNode(node.id(), State.ACTIVE, null); |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 306 | } |
| 307 | } |
| 308 | }); |
| 309 | } catch (Exception e) { |
| 310 | log.debug("Failed to send heartbeat", e); |
| 311 | } |
tom | b41d1ac | 2014-09-24 01:51:24 -0700 | [diff] [blame] | 312 | } |
| 313 | |
Jordan Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 314 | private void notifyChange(NodeId nodeId, State oldState, State newState, Version oldVersion, Version newVersion) { |
| 315 | if (oldState != newState || !Objects.equals(oldVersion, newVersion)) { |
Thomas Vachuska | fba7f3d | 2016-03-23 15:46:25 -0700 | [diff] [blame] | 316 | ControllerNode node = allNodes.get(nodeId); |
Jon Hall | 1f13d64 | 2016-05-13 17:53:01 -0700 | [diff] [blame] | 317 | // Either this node or that node is no longer part of the same cluster |
| 318 | if (node == null) { |
Jon Hall | 66870baa | 2016-06-20 12:12:10 -0700 | [diff] [blame] | 319 | log.debug("Could not find node {} in the cluster, ignoring state change", nodeId); |
Jon Hall | 1f13d64 | 2016-05-13 17:53:01 -0700 | [diff] [blame] | 320 | return; |
| 321 | } |
Thomas Vachuska | fba7f3d | 2016-03-23 15:46:25 -0700 | [diff] [blame] | 322 | ClusterEvent.Type type = newState == State.READY ? INSTANCE_READY : |
| 323 | newState == State.ACTIVE ? INSTANCE_ACTIVATED : |
| 324 | INSTANCE_DEACTIVATED; |
| 325 | notifyDelegate(new ClusterEvent(type, node)); |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 326 | } |
tom | b41d1ac | 2014-09-24 01:51:24 -0700 | [diff] [blame] | 327 | } |
| 328 | |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 329 | private void heartbeatToPeer(byte[] messagePayload, ControllerNode peer) { |
Madan Jampani | afeebbd | 2015-05-19 15:26:01 -0700 | [diff] [blame] | 330 | Endpoint remoteEp = new Endpoint(peer.ip(), peer.tcpPort()); |
Madan Jampani | 175e8fd | 2015-05-20 14:10:45 -0700 | [diff] [blame] | 331 | messagingService.sendAsync(remoteEp, HEARTBEAT_MESSAGE, messagePayload).whenComplete((result, error) -> { |
| 332 | if (error != null) { |
| 333 | log.trace("Sending heartbeat to {} failed", remoteEp, error); |
| 334 | } |
| 335 | }); |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 336 | } |
| 337 | |
Madan Jampani | d36def0 | 2016-01-13 11:21:56 -0800 | [diff] [blame] | 338 | private class HeartbeatMessageHandler implements BiConsumer<Endpoint, byte[]> { |
tom | b41d1ac | 2014-09-24 01:51:24 -0700 | [diff] [blame] | 339 | @Override |
Madan Jampani | d36def0 | 2016-01-13 11:21:56 -0800 | [diff] [blame] | 340 | public void accept(Endpoint sender, byte[] message) { |
Madan Jampani | c26eede | 2015-04-16 11:42:16 -0700 | [diff] [blame] | 341 | HeartbeatMessage hb = SERIALIZER.decode(message); |
Madan Jampani | e337506 | 2016-04-19 16:32:10 -0700 | [diff] [blame] | 342 | if (clusterMetadataService.getClusterMetadata().getNodes().contains(hb.source())) { |
Jordan Halterman | e706253 | 2018-01-22 11:22:46 -0800 | [diff] [blame] | 343 | // Avoid reporting heartbeats that have been enqueued by setting a minimum interval. |
| 344 | long heartbeatTime = System.currentTimeMillis(); |
Jordan Halterman | a899362 | 2018-07-29 00:22:37 -0700 | [diff] [blame^] | 345 | Long lastHeartbeatTime = failureDetector.getLastHeartbeatTime(hb.source().id()); |
| 346 | if (lastHeartbeatTime == null || heartbeatTime - lastHeartbeatTime > heartbeatInterval / 2) { |
Jordan Halterman | e706253 | 2018-01-22 11:22:46 -0800 | [diff] [blame] | 347 | failureDetector.report(hb.source().id(), heartbeatTime); |
Jordan Halterman | 1315d3e | 2018-01-16 19:41:31 -0800 | [diff] [blame] | 348 | } |
Jordan Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 349 | updateNode(hb.source().id(), hb.state, hb.version); |
Madan Jampani | e337506 | 2016-04-19 16:32:10 -0700 | [diff] [blame] | 350 | } |
tom | b41d1ac | 2014-09-24 01:51:24 -0700 | [diff] [blame] | 351 | } |
tom | 2d7c65f | 2014-09-23 01:09:35 -0700 | [diff] [blame] | 352 | } |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 353 | |
| 354 | private static class HeartbeatMessage { |
| 355 | private ControllerNode source; |
Thomas Vachuska | 7a8de84 | 2016-03-07 20:56:35 -0800 | [diff] [blame] | 356 | private State state; |
Jordan Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 357 | private Version version; |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 358 | |
Jordan Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 359 | public HeartbeatMessage(ControllerNode source, State state, Version version) { |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 360 | this.source = source; |
Thomas Vachuska | 7a8de84 | 2016-03-07 20:56:35 -0800 | [diff] [blame] | 361 | this.state = state != null ? state : State.ACTIVE; |
Jordan Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 362 | this.version = version; |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 363 | } |
| 364 | |
| 365 | public ControllerNode source() { |
| 366 | return source; |
| 367 | } |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 368 | } |
| 369 | |
Madan Jampani | 7d2fab2 | 2015-03-18 17:21:57 -0700 | [diff] [blame] | 370 | @Override |
Yuta HIGUCHI | 0c47d53 | 2017-08-18 23:16:35 -0700 | [diff] [blame] | 371 | public Instant getLastUpdatedInstant(NodeId nodeId) { |
Jordan Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 372 | return nodeLastUpdatedTimes.get(nodeId); |
Madan Jampani | 7d2fab2 | 2015-03-18 17:21:57 -0700 | [diff] [blame] | 373 | } |
Thomas Vachuska | de563cf | 2015-04-01 00:28:50 -0700 | [diff] [blame] | 374 | |
sangyun-han | f98df54 | 2016-03-24 20:28:03 +0900 | [diff] [blame] | 375 | /** |
Deepa Vaddireddy | 6690f59 | 2017-03-12 03:31:48 +0530 | [diff] [blame] | 376 | * Extracts properties from the component configuration. |
sangyun-han | f98df54 | 2016-03-24 20:28:03 +0900 | [diff] [blame] | 377 | * |
sangyun-han | f98df54 | 2016-03-24 20:28:03 +0900 | [diff] [blame] | 378 | */ |
Deepa Vaddireddy | 6690f59 | 2017-03-12 03:31:48 +0530 | [diff] [blame] | 379 | private void readComponentConfiguration() { |
| 380 | Set<ConfigProperty> configProperties = cfgService.getProperties(getClass().getName()); |
| 381 | for (ConfigProperty property : configProperties) { |
Jon Hall | a3fcf67 | 2017-03-28 16:53:22 -0700 | [diff] [blame] | 382 | if ("heartbeatInterval".equals(property.name())) { |
Deepa Vaddireddy | 6690f59 | 2017-03-12 03:31:48 +0530 | [diff] [blame] | 383 | String s = property.value(); |
| 384 | if (s == null) { |
| 385 | setHeartbeatInterval(DEFAULT_HEARTBEAT_INTERVAL); |
| 386 | log.info("Heartbeat interval time is not configured, default value is {}", |
| 387 | DEFAULT_HEARTBEAT_INTERVAL); |
| 388 | } else { |
| 389 | int newHeartbeatInterval = isNullOrEmpty(s) ? DEFAULT_HEARTBEAT_INTERVAL |
Jordan Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 390 | : Integer.parseInt(s.trim()); |
Deepa Vaddireddy | 6690f59 | 2017-03-12 03:31:48 +0530 | [diff] [blame] | 391 | if (newHeartbeatInterval > 0 && heartbeatInterval != newHeartbeatInterval) { |
| 392 | heartbeatInterval = newHeartbeatInterval; |
| 393 | restartHeartbeatSender(); |
| 394 | } |
| 395 | log.info("Configured. Heartbeat interval time is configured to {}", |
| 396 | heartbeatInterval); |
| 397 | } |
| 398 | } |
Jon Hall | a3fcf67 | 2017-03-28 16:53:22 -0700 | [diff] [blame] | 399 | if ("phiFailureThreshold".equals(property.name())) { |
Deepa Vaddireddy | 6690f59 | 2017-03-12 03:31:48 +0530 | [diff] [blame] | 400 | String s = property.value(); |
| 401 | if (s == null) { |
| 402 | setPhiFailureThreshold(DEFAULT_PHI_FAILURE_THRESHOLD); |
| 403 | log.info("Phi failure threshold is not configured, default value is {}", |
| 404 | DEFAULT_PHI_FAILURE_THRESHOLD); |
| 405 | } else { |
| 406 | int newPhiFailureThreshold = isNullOrEmpty(s) ? DEFAULT_HEARTBEAT_INTERVAL |
Jordan Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 407 | : Integer.parseInt(s.trim()); |
Deepa Vaddireddy | 6690f59 | 2017-03-12 03:31:48 +0530 | [diff] [blame] | 408 | setPhiFailureThreshold(newPhiFailureThreshold); |
| 409 | log.info("Configured. Phi failure threshold is configured to {}", |
| 410 | phiFailureThreshold); |
| 411 | } |
| 412 | } |
Jordan Halterman | b75aa08 | 2018-05-14 15:00:42 -0700 | [diff] [blame] | 413 | if ("minStandardDeviationMillis".equals(property.name())) { |
| 414 | String s = property.value(); |
| 415 | if (s == null) { |
| 416 | setMinStandardDeviationMillis(DEFAULT_MIN_STANDARD_DEVIATION_MILLIS); |
| 417 | log.info("Minimum standard deviation is not configured, default value is {}", |
| 418 | DEFAULT_MIN_STANDARD_DEVIATION_MILLIS); |
| 419 | } else { |
| 420 | long newMinStandardDeviationMillis = isNullOrEmpty(s) |
| 421 | ? DEFAULT_MIN_STANDARD_DEVIATION_MILLIS |
| 422 | : Long.parseLong(s.trim()); |
| 423 | setMinStandardDeviationMillis(newMinStandardDeviationMillis); |
| 424 | log.info("Configured. Minimum standard deviation is configured to {}", |
| 425 | newMinStandardDeviationMillis); |
| 426 | } |
| 427 | } |
sangyun-han | f98df54 | 2016-03-24 20:28:03 +0900 | [diff] [blame] | 428 | } |
| 429 | } |
| 430 | |
| 431 | /** |
| 432 | * Sets heartbeat interval between the termination of one execution of heartbeat |
| 433 | * and the commencement of the next. |
| 434 | * |
| 435 | * @param interval term between each heartbeat |
| 436 | */ |
| 437 | private void setHeartbeatInterval(int interval) { |
| 438 | try { |
| 439 | checkArgument(interval > 0, "Interval must be greater than zero"); |
| 440 | heartbeatInterval = interval; |
| 441 | } catch (IllegalArgumentException e) { |
| 442 | log.warn(e.getMessage()); |
| 443 | heartbeatInterval = DEFAULT_HEARTBEAT_INTERVAL; |
| 444 | } |
| 445 | } |
| 446 | |
| 447 | /** |
| 448 | * Sets Phi failure threshold. |
| 449 | * Phi is based on a paper titled: "The φ Accrual Failure Detector" by Hayashibara, et al. |
| 450 | * |
| 451 | * @param threshold |
| 452 | */ |
| 453 | private void setPhiFailureThreshold(int threshold) { |
| 454 | phiFailureThreshold = threshold; |
| 455 | } |
| 456 | |
| 457 | /** |
Jordan Halterman | b75aa08 | 2018-05-14 15:00:42 -0700 | [diff] [blame] | 458 | * Sets the minimum standard deviation milliseconds. |
| 459 | * |
| 460 | * @param minStandardDeviationMillis the updated minimum standard deviation |
| 461 | */ |
| 462 | private void setMinStandardDeviationMillis(long minStandardDeviationMillis) { |
| 463 | this.minStandardDeviationMillis = minStandardDeviationMillis; |
| 464 | try { |
| 465 | failureDetector = new PhiAccrualFailureDetector(minStandardDeviationMillis); |
| 466 | } catch (IllegalArgumentException e) { |
| 467 | log.warn(e.getMessage()); |
| 468 | this.minStandardDeviationMillis = DEFAULT_MIN_STANDARD_DEVIATION_MILLIS; |
| 469 | failureDetector = new PhiAccrualFailureDetector(this.minStandardDeviationMillis); |
| 470 | } |
| 471 | } |
| 472 | |
| 473 | /** |
sangyun-han | f98df54 | 2016-03-24 20:28:03 +0900 | [diff] [blame] | 474 | * Restarts heartbeatSender executor. |
sangyun-han | f98df54 | 2016-03-24 20:28:03 +0900 | [diff] [blame] | 475 | */ |
| 476 | private void restartHeartbeatSender() { |
| 477 | try { |
| 478 | ScheduledExecutorService prevSender = heartBeatSender; |
| 479 | heartBeatSender = Executors.newSingleThreadScheduledExecutor( |
HIGUCHI Yuta | d9e0105 | 2016-04-14 09:31:42 -0700 | [diff] [blame] | 480 | groupedThreads("onos/cluster/membership", "heartbeat-sender-%d", log)); |
sangyun-han | f98df54 | 2016-03-24 20:28:03 +0900 | [diff] [blame] | 481 | heartBeatSender.scheduleWithFixedDelay(this::heartbeat, 0, |
Jordan Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 482 | heartbeatInterval, TimeUnit.MILLISECONDS); |
sangyun-han | f98df54 | 2016-03-24 20:28:03 +0900 | [diff] [blame] | 483 | prevSender.shutdown(); |
| 484 | } catch (Exception e) { |
| 485 | log.warn(e.getMessage()); |
| 486 | } |
| 487 | } |
Jordan Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 488 | } |