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; |
Jordan Halterman | 00e92da | 2018-05-22 23:05:52 -0700 | [diff] [blame^] | 41 | import org.onosproject.cluster.Node; |
Brian O'Connor | abafb50 | 2014-12-02 22:26:20 -0800 | [diff] [blame] | 42 | import org.onosproject.cluster.NodeId; |
Jordan Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 43 | import org.onosproject.core.Version; |
| 44 | import org.onosproject.core.VersionService; |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 45 | import org.onosproject.store.AbstractStore; |
Madan Jampani | c26eede | 2015-04-16 11:42:16 -0700 | [diff] [blame] | 46 | import org.onosproject.store.cluster.messaging.Endpoint; |
Madan Jampani | afeebbd | 2015-05-19 15:26:01 -0700 | [diff] [blame] | 47 | import org.onosproject.store.cluster.messaging.MessagingService; |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 48 | import org.onosproject.store.serializers.KryoNamespaces; |
Jordan Halterman | 2c83a10 | 2017-08-20 17:11:41 -0700 | [diff] [blame] | 49 | import org.onosproject.store.service.Serializer; |
sangyun-han | f98df54 | 2016-03-24 20:28:03 +0900 | [diff] [blame] | 50 | import org.osgi.service.component.ComponentContext; |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 51 | import org.slf4j.Logger; |
tom | 2d7c65f | 2014-09-23 01:09:35 -0700 | [diff] [blame] | 52 | |
Yuta HIGUCHI | 0c47d53 | 2017-08-18 23:16:35 -0700 | [diff] [blame] | 53 | import java.time.Instant; |
Jonathan Hart | 4a4d18f | 2015-03-26 12:16:16 -0700 | [diff] [blame] | 54 | import java.util.Map; |
Thomas Vachuska | fba7f3d | 2016-03-23 15:46:25 -0700 | [diff] [blame] | 55 | import java.util.Objects; |
Jonathan Hart | 4a4d18f | 2015-03-26 12:16:16 -0700 | [diff] [blame] | 56 | import java.util.Set; |
| 57 | import java.util.concurrent.ExecutorService; |
| 58 | import java.util.concurrent.Executors; |
| 59 | import java.util.concurrent.ScheduledExecutorService; |
| 60 | import java.util.concurrent.TimeUnit; |
Madan Jampani | d36def0 | 2016-01-13 11:21:56 -0800 | [diff] [blame] | 61 | import java.util.function.BiConsumer; |
Jonathan Hart | 4a4d18f | 2015-03-26 12:16:16 -0700 | [diff] [blame] | 62 | import java.util.stream.Collectors; |
| 63 | |
Jordan Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 64 | import static com.google.common.base.MoreObjects.firstNonNull; |
sangyun-han | f98df54 | 2016-03-24 20:28:03 +0900 | [diff] [blame] | 65 | import static com.google.common.base.Preconditions.checkArgument; |
Jonathan Hart | 4a4d18f | 2015-03-26 12:16:16 -0700 | [diff] [blame] | 66 | import static com.google.common.base.Preconditions.checkNotNull; |
Deepa Vaddireddy | 6690f59 | 2017-03-12 03:31:48 +0530 | [diff] [blame] | 67 | import static com.google.common.base.Strings.isNullOrEmpty; |
Jonathan Hart | 4a4d18f | 2015-03-26 12:16:16 -0700 | [diff] [blame] | 68 | import static org.onlab.util.Tools.groupedThreads; |
Thomas Vachuska | fba7f3d | 2016-03-23 15:46:25 -0700 | [diff] [blame] | 69 | import static org.onosproject.cluster.ClusterEvent.Type.INSTANCE_ACTIVATED; |
| 70 | import static org.onosproject.cluster.ClusterEvent.Type.INSTANCE_DEACTIVATED; |
| 71 | import static org.onosproject.cluster.ClusterEvent.Type.INSTANCE_READY; |
Jonathan Hart | 4a4d18f | 2015-03-26 12:16:16 -0700 | [diff] [blame] | 72 | import static org.slf4j.LoggerFactory.getLogger; |
tom | 2d7c65f | 2014-09-23 01:09:35 -0700 | [diff] [blame] | 73 | |
Jordan Halterman | 00e92da | 2018-05-22 23:05:52 -0700 | [diff] [blame^] | 74 | @Component(enabled = false) |
tom | 2d7c65f | 2014-09-23 01:09:35 -0700 | [diff] [blame] | 75 | @Service |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 76 | /** |
| 77 | * Distributed cluster nodes store that employs an accrual failure |
| 78 | * detector to identify cluster member up/down status. |
| 79 | */ |
tom | 0755a36 | 2014-09-24 11:54:43 -0700 | [diff] [blame] | 80 | public class DistributedClusterStore |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 81 | extends AbstractStore<ClusterEvent, ClusterStoreDelegate> |
tom | b41d1ac | 2014-09-24 01:51:24 -0700 | [diff] [blame] | 82 | implements ClusterStore { |
tom | 2d7c65f | 2014-09-23 01:09:35 -0700 | [diff] [blame] | 83 | |
Thomas Vachuska | 8dc1a69 | 2015-03-31 01:01:37 -0700 | [diff] [blame] | 84 | private static final Logger log = getLogger(DistributedClusterStore.class); |
tom | 2d7c65f | 2014-09-23 01:09:35 -0700 | [diff] [blame] | 85 | |
Thomas Vachuska | de563cf | 2015-04-01 00:28:50 -0700 | [diff] [blame] | 86 | public static final String HEARTBEAT_MESSAGE = "onos-cluster-heartbeat"; |
| 87 | |
sangyun-han | f98df54 | 2016-03-24 20:28:03 +0900 | [diff] [blame] | 88 | 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; |
tom | 2d7c65f | 2014-09-23 01:09:35 -0700 | [diff] [blame] | 97 | |
Jordan Halterman | 4a082ec | 2018-05-14 15:00:42 -0700 | [diff] [blame] | 98 | 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 Halterman | 2c83a10 | 2017-08-20 17:11:41 -0700 | [diff] [blame] | 103 | private static final Serializer SERIALIZER = Serializer.using( |
Jordan Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 104 | KryoNamespace.newBuilder() |
Thomas Vachuska | 8dc1a69 | 2015-03-31 01:01:37 -0700 | [diff] [blame] | 105 | .register(KryoNamespaces.API) |
HIGUCHI Yuta | e729065 | 2016-05-18 11:29:01 -0700 | [diff] [blame] | 106 | .nextId(KryoNamespaces.BEGIN_USER_CUSTOM_ID) |
Thomas Vachuska | 8dc1a69 | 2015-03-31 01:01:37 -0700 | [diff] [blame] | 107 | .register(HeartbeatMessage.class) |
HIGUCHI Yuta | e729065 | 2016-05-18 11:29:01 -0700 | [diff] [blame] | 108 | .build("ClusterStore")); |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 109 | |
| 110 | private static final String INSTANCE_ID_NULL = "Instance ID cannot be null"; |
| 111 | |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 112 | private final Map<NodeId, ControllerNode> allNodes = Maps.newConcurrentMap(); |
| 113 | private final Map<NodeId, State> nodeStates = Maps.newConcurrentMap(); |
Jordan Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 114 | private final Map<NodeId, Version> nodeVersions = Maps.newConcurrentMap(); |
Yuta HIGUCHI | 0c47d53 | 2017-08-18 23:16:35 -0700 | [diff] [blame] | 115 | private final Map<NodeId, Instant> nodeLastUpdatedTimes = Maps.newConcurrentMap(); |
Thomas Vachuska | 7a8de84 | 2016-03-07 20:56:35 -0800 | [diff] [blame] | 116 | |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 117 | private ScheduledExecutorService heartBeatSender = Executors.newSingleThreadScheduledExecutor( |
HIGUCHI Yuta | d9e0105 | 2016-04-14 09:31:42 -0700 | [diff] [blame] | 118 | groupedThreads("onos/cluster/membership", "heartbeat-sender", log)); |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 119 | private ExecutorService heartBeatMessageHandler = Executors.newSingleThreadExecutor( |
HIGUCHI Yuta | d9e0105 | 2016-04-14 09:31:42 -0700 | [diff] [blame] | 120 | groupedThreads("onos/cluster/membership", "heartbeat-receiver", log)); |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 121 | |
| 122 | private PhiAccrualFailureDetector failureDetector; |
| 123 | |
| 124 | private ControllerNode localNode; |
Jordan Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 125 | private Version localVersion; |
| 126 | |
| 127 | @Reference(cardinality = ReferenceCardinality.MANDATORY_UNARY) |
| 128 | protected VersionService versionService; |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 129 | |
Madan Jampani | afeebbd | 2015-05-19 15:26:01 -0700 | [diff] [blame] | 130 | @Reference(cardinality = ReferenceCardinality.MANDATORY_UNARY) |
Madan Jampani | ec1df02 | 2015-10-13 21:23:03 -0700 | [diff] [blame] | 131 | protected ClusterMetadataService clusterMetadataService; |
Madan Jampani | afeebbd | 2015-05-19 15:26:01 -0700 | [diff] [blame] | 132 | |
| 133 | @Reference(cardinality = ReferenceCardinality.MANDATORY_UNARY) |
| 134 | protected MessagingService messagingService; |
| 135 | |
Thomas Vachuska | 1ca7e9f | 2016-08-05 10:21:41 -0700 | [diff] [blame] | 136 | // This must be optional to avoid a cyclic dependency |
Deepa Vaddireddy | 6690f59 | 2017-03-12 03:31:48 +0530 | [diff] [blame] | 137 | @Reference(cardinality = ReferenceCardinality.OPTIONAL_UNARY, |
Jordan Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 138 | bind = "bindComponentConfigService", |
| 139 | unbind = "unbindComponentConfigService", |
| 140 | policy = ReferencePolicy.DYNAMIC) |
sangyun-han | 9f0af2d | 2016-08-04 13:04:59 +0900 | [diff] [blame] | 141 | protected ComponentConfigService cfgService; |
| 142 | |
Thomas Vachuska | 1ca7e9f | 2016-08-05 10:21:41 -0700 | [diff] [blame] | 143 | /** |
| 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 Vaddireddy | 6690f59 | 2017-03-12 03:31:48 +0530 | [diff] [blame] | 152 | readComponentConfiguration(); |
Thomas Vachuska | 1ca7e9f | 2016-08-05 10:21:41 -0700 | [diff] [blame] | 153 | } |
| 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 | |
tom | 2d7c65f | 2014-09-23 01:09:35 -0700 | [diff] [blame] | 168 | @Activate |
Thomas Vachuska | 1ca7e9f | 2016-08-05 10:21:41 -0700 | [diff] [blame] | 169 | public void activate() { |
Madan Jampani | ec1df02 | 2015-10-13 21:23:03 -0700 | [diff] [blame] | 170 | localNode = clusterMetadataService.getLocalNode(); |
Jordan Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 171 | localVersion = versionService.version(); |
| 172 | nodeVersions.put(localNode.id(), localVersion); |
Thomas Vachuska | 8dc1a69 | 2015-03-31 01:01:37 -0700 | [diff] [blame] | 173 | |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 174 | messagingService.registerHandler(HEARTBEAT_MESSAGE, |
Jordan Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 175 | new HeartbeatMessageHandler(), heartBeatMessageHandler); |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 176 | |
Jordan Halterman | 4a082ec | 2018-05-14 15:00:42 -0700 | [diff] [blame] | 177 | failureDetector = new PhiAccrualFailureDetector(minStandardDeviationMillis); |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 178 | |
| 179 | heartBeatSender.scheduleWithFixedDelay(this::heartbeat, 0, |
Jordan Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 180 | heartbeatInterval, TimeUnit.MILLISECONDS); |
tom | b41d1ac | 2014-09-24 01:51:24 -0700 | [diff] [blame] | 181 | |
| 182 | log.info("Started"); |
| 183 | } |
| 184 | |
tom | 2d7c65f | 2014-09-23 01:09:35 -0700 | [diff] [blame] | 185 | @Deactivate |
| 186 | public void deactivate() { |
Madan Jampani | afeebbd | 2015-05-19 15:26:01 -0700 | [diff] [blame] | 187 | messagingService.unregisterHandler(HEARTBEAT_MESSAGE); |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 188 | heartBeatSender.shutdownNow(); |
| 189 | heartBeatMessageHandler.shutdownNow(); |
| 190 | |
tom | 2d7c65f | 2014-09-23 01:09:35 -0700 | [diff] [blame] | 191 | log.info("Stopped"); |
| 192 | } |
| 193 | |
sangyun-han | f98df54 | 2016-03-24 20:28:03 +0900 | [diff] [blame] | 194 | @Modified |
| 195 | public void modified(ComponentContext context) { |
Deepa Vaddireddy | 6690f59 | 2017-03-12 03:31:48 +0530 | [diff] [blame] | 196 | readComponentConfiguration(); |
sangyun-han | f98df54 | 2016-03-24 20:28:03 +0900 | [diff] [blame] | 197 | } |
| 198 | |
tom | 2d7c65f | 2014-09-23 01:09:35 -0700 | [diff] [blame] | 199 | @Override |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 200 | 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 |
tom | 2d7c65f | 2014-09-23 01:09:35 -0700 | [diff] [blame] | 216 | public ControllerNode getLocalNode() { |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 217 | return localNode; |
tom | 2d7c65f | 2014-09-23 01:09:35 -0700 | [diff] [blame] | 218 | } |
| 219 | |
| 220 | @Override |
| 221 | public Set<ControllerNode> getNodes() { |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 222 | return ImmutableSet.copyOf(allNodes.values()); |
tom | 2d7c65f | 2014-09-23 01:09:35 -0700 | [diff] [blame] | 223 | } |
| 224 | |
| 225 | @Override |
Jordan Halterman | 00e92da | 2018-05-22 23:05:52 -0700 | [diff] [blame^] | 226 | public Set<Node> getStorageNodes() { |
| 227 | return ImmutableSet.of(); |
| 228 | } |
| 229 | |
| 230 | @Override |
tom | 2d7c65f | 2014-09-23 01:09:35 -0700 | [diff] [blame] | 231 | public ControllerNode getNode(NodeId nodeId) { |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 232 | checkNotNull(nodeId, INSTANCE_ID_NULL); |
| 233 | return allNodes.get(nodeId); |
tom | 2d7c65f | 2014-09-23 01:09:35 -0700 | [diff] [blame] | 234 | } |
| 235 | |
| 236 | @Override |
tom | b41d1ac | 2014-09-24 01:51:24 -0700 | [diff] [blame] | 237 | public State getState(NodeId nodeId) { |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 238 | checkNotNull(nodeId, INSTANCE_ID_NULL); |
Jordan Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 239 | 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); |
tom | b41d1ac | 2014-09-24 01:51:24 -0700 | [diff] [blame] | 246 | } |
| 247 | |
| 248 | @Override |
Thomas Vachuska | 7a8de84 | 2016-03-07 20:56:35 -0800 | [diff] [blame] | 249 | public void markFullyStarted(boolean started) { |
Jordan Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 250 | updateNode(localNode.id(), started ? State.READY : State.ACTIVE, null); |
Thomas Vachuska | 7a8de84 | 2016-03-07 20:56:35 -0800 | [diff] [blame] | 251 | } |
| 252 | |
| 253 | @Override |
Pavlin Radoslavov | 444b519 | 2014-10-28 10:45:19 -0700 | [diff] [blame] | 254 | public ControllerNode addNode(NodeId nodeId, IpAddress ip, int tcpPort) { |
sangyun-han | f98df54 | 2016-03-24 20:28:03 +0900 | [diff] [blame] | 255 | checkNotNull(nodeId, INSTANCE_ID_NULL); |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 256 | ControllerNode node = new DefaultControllerNode(nodeId, ip, tcpPort); |
Madan Jampani | afeebbd | 2015-05-19 15:26:01 -0700 | [diff] [blame] | 257 | addNode(node); |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 258 | return node; |
tom | ee49c37 | 2014-09-26 15:14:50 -0700 | [diff] [blame] | 259 | } |
| 260 | |
| 261 | @Override |
tom | b41d1ac | 2014-09-24 01:51:24 -0700 | [diff] [blame] | 262 | public void removeNode(NodeId nodeId) { |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 263 | checkNotNull(nodeId, INSTANCE_ID_NULL); |
| 264 | ControllerNode node = allNodes.remove(nodeId); |
| 265 | if (node != null) { |
| 266 | nodeStates.remove(nodeId); |
Jordan Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 267 | nodeVersions.remove(nodeId); |
Jonathan Hart | f114126 | 2015-04-23 11:27:07 -0700 | [diff] [blame] | 268 | notifyDelegate(new ClusterEvent(ClusterEvent.Type.INSTANCE_REMOVED, node)); |
tom | b41d1ac | 2014-09-24 01:51:24 -0700 | [diff] [blame] | 269 | } |
| 270 | } |
| 271 | |
Madan Jampani | afeebbd | 2015-05-19 15:26:01 -0700 | [diff] [blame] | 272 | private void addNode(ControllerNode node) { |
| 273 | allNodes.put(node.id(), node); |
Jordan Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 274 | updateNode(node.id(), node.equals(localNode) ? State.ACTIVE : State.INACTIVE, null); |
Madan Jampani | afeebbd | 2015-05-19 15:26:01 -0700 | [diff] [blame] | 275 | notifyDelegate(new ClusterEvent(ClusterEvent.Type.INSTANCE_ADDED, node)); |
Thomas Vachuska | de563cf | 2015-04-01 00:28:50 -0700 | [diff] [blame] | 276 | } |
| 277 | |
Jordan Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 278 | private void updateNode(NodeId nodeId, State newState, Version newVersion) { |
Thomas Vachuska | fba7f3d | 2016-03-23 15:46:25 -0700 | [diff] [blame] | 279 | State currentState = nodeStates.get(nodeId); |
Jordan Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 280 | Version currentVersion = nodeVersions.get(nodeId); |
Jordan Halterman | eabae96 | 2017-08-01 17:29:36 -0700 | [diff] [blame] | 281 | if (!Objects.equals(currentState, newState) |
| 282 | || (newVersion != null && !Objects.equals(currentVersion, newVersion))) { |
Thomas Vachuska | fba7f3d | 2016-03-23 15:46:25 -0700 | [diff] [blame] | 283 | nodeStates.put(nodeId, newState); |
Jordan Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 284 | if (newVersion != null) { |
| 285 | nodeVersions.put(nodeId, newVersion); |
| 286 | } |
Yuta HIGUCHI | 0c47d53 | 2017-08-18 23:16:35 -0700 | [diff] [blame] | 287 | nodeLastUpdatedTimes.put(nodeId, Instant.now()); |
Jordan Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 288 | notifyChange(nodeId, currentState, newState, currentVersion, newVersion); |
Thomas Vachuska | fba7f3d | 2016-03-23 15:46:25 -0700 | [diff] [blame] | 289 | } |
Madan Jampani | 7d2fab2 | 2015-03-18 17:21:57 -0700 | [diff] [blame] | 290 | } |
| 291 | |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 292 | 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 Vachuska | 7a8de84 | 2016-03-07 20:56:35 -0800 | [diff] [blame] | 298 | State state = nodeStates.get(localNode.id()); |
Jordan Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 299 | byte[] hbMessagePayload = SERIALIZER.encode(new HeartbeatMessage(localNode, state, localVersion)); |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 300 | peers.forEach((node) -> { |
| 301 | heartbeatToPeer(hbMessagePayload, node); |
| 302 | State currentState = nodeStates.get(node.id()); |
| 303 | double phi = failureDetector.phi(node.id()); |
sangyun-han | f98df54 | 2016-03-24 20:28:03 +0900 | [diff] [blame] | 304 | if (phi >= phiFailureThreshold) { |
Thomas Vachuska | 7a8de84 | 2016-03-07 20:56:35 -0800 | [diff] [blame] | 305 | if (currentState.isActive()) { |
Jordan Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 306 | updateNode(node.id(), State.INACTIVE, null); |
Jordan Halterman | e706253 | 2018-01-22 11:22:46 -0800 | [diff] [blame] | 307 | failureDetector.reset(node.id()); |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 308 | } |
| 309 | } else { |
| 310 | if (currentState == State.INACTIVE) { |
Jordan Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 311 | updateNode(node.id(), State.ACTIVE, null); |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 312 | } |
| 313 | } |
| 314 | }); |
| 315 | } catch (Exception e) { |
| 316 | log.debug("Failed to send heartbeat", e); |
| 317 | } |
tom | b41d1ac | 2014-09-24 01:51:24 -0700 | [diff] [blame] | 318 | } |
| 319 | |
Jordan Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 320 | private void notifyChange(NodeId nodeId, State oldState, State newState, Version oldVersion, Version newVersion) { |
| 321 | if (oldState != newState || !Objects.equals(oldVersion, newVersion)) { |
Thomas Vachuska | fba7f3d | 2016-03-23 15:46:25 -0700 | [diff] [blame] | 322 | ControllerNode node = allNodes.get(nodeId); |
Jon Hall | 1f13d64 | 2016-05-13 17:53:01 -0700 | [diff] [blame] | 323 | // Either this node or that node is no longer part of the same cluster |
| 324 | if (node == null) { |
Jon Hall | 66870baa | 2016-06-20 12:12:10 -0700 | [diff] [blame] | 325 | 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] | 326 | return; |
| 327 | } |
Thomas Vachuska | fba7f3d | 2016-03-23 15:46:25 -0700 | [diff] [blame] | 328 | ClusterEvent.Type type = newState == State.READY ? INSTANCE_READY : |
| 329 | newState == State.ACTIVE ? INSTANCE_ACTIVATED : |
| 330 | INSTANCE_DEACTIVATED; |
| 331 | notifyDelegate(new ClusterEvent(type, node)); |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 332 | } |
tom | b41d1ac | 2014-09-24 01:51:24 -0700 | [diff] [blame] | 333 | } |
| 334 | |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 335 | private void heartbeatToPeer(byte[] messagePayload, ControllerNode peer) { |
Madan Jampani | afeebbd | 2015-05-19 15:26:01 -0700 | [diff] [blame] | 336 | Endpoint remoteEp = new Endpoint(peer.ip(), peer.tcpPort()); |
Madan Jampani | 175e8fd | 2015-05-20 14:10:45 -0700 | [diff] [blame] | 337 | 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 Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 342 | } |
| 343 | |
Madan Jampani | d36def0 | 2016-01-13 11:21:56 -0800 | [diff] [blame] | 344 | private class HeartbeatMessageHandler implements BiConsumer<Endpoint, byte[]> { |
tom | b41d1ac | 2014-09-24 01:51:24 -0700 | [diff] [blame] | 345 | @Override |
Madan Jampani | d36def0 | 2016-01-13 11:21:56 -0800 | [diff] [blame] | 346 | public void accept(Endpoint sender, byte[] message) { |
Madan Jampani | c26eede | 2015-04-16 11:42:16 -0700 | [diff] [blame] | 347 | HeartbeatMessage hb = SERIALIZER.decode(message); |
Madan Jampani | e337506 | 2016-04-19 16:32:10 -0700 | [diff] [blame] | 348 | if (clusterMetadataService.getClusterMetadata().getNodes().contains(hb.source())) { |
Jordan Halterman | e706253 | 2018-01-22 11:22:46 -0800 | [diff] [blame] | 349 | // 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 Halterman | 1315d3e | 2018-01-16 19:41:31 -0800 | [diff] [blame] | 354 | } |
Jordan Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 355 | updateNode(hb.source().id(), hb.state, hb.version); |
Madan Jampani | e337506 | 2016-04-19 16:32:10 -0700 | [diff] [blame] | 356 | } |
tom | b41d1ac | 2014-09-24 01:51:24 -0700 | [diff] [blame] | 357 | } |
tom | 2d7c65f | 2014-09-23 01:09:35 -0700 | [diff] [blame] | 358 | } |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 359 | |
| 360 | private static class HeartbeatMessage { |
| 361 | private ControllerNode source; |
Thomas Vachuska | 7a8de84 | 2016-03-07 20:56:35 -0800 | [diff] [blame] | 362 | private State state; |
Jordan Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 363 | private Version version; |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 364 | |
Jordan Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 365 | public HeartbeatMessage(ControllerNode source, State state, Version version) { |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 366 | this.source = source; |
Thomas Vachuska | 7a8de84 | 2016-03-07 20:56:35 -0800 | [diff] [blame] | 367 | this.state = state != null ? state : State.ACTIVE; |
Jordan Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 368 | this.version = version; |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 369 | } |
| 370 | |
| 371 | public ControllerNode source() { |
| 372 | return source; |
| 373 | } |
Ayaka Koshibe | dd91b84 | 2015-03-02 14:48:47 -0800 | [diff] [blame] | 374 | } |
| 375 | |
Madan Jampani | 7d2fab2 | 2015-03-18 17:21:57 -0700 | [diff] [blame] | 376 | @Override |
Yuta HIGUCHI | 0c47d53 | 2017-08-18 23:16:35 -0700 | [diff] [blame] | 377 | public Instant getLastUpdatedInstant(NodeId nodeId) { |
Jordan Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 378 | return nodeLastUpdatedTimes.get(nodeId); |
Madan Jampani | 7d2fab2 | 2015-03-18 17:21:57 -0700 | [diff] [blame] | 379 | } |
Thomas Vachuska | de563cf | 2015-04-01 00:28:50 -0700 | [diff] [blame] | 380 | |
sangyun-han | f98df54 | 2016-03-24 20:28:03 +0900 | [diff] [blame] | 381 | /** |
Deepa Vaddireddy | 6690f59 | 2017-03-12 03:31:48 +0530 | [diff] [blame] | 382 | * Extracts properties from the component configuration. |
sangyun-han | f98df54 | 2016-03-24 20:28:03 +0900 | [diff] [blame] | 383 | * |
sangyun-han | f98df54 | 2016-03-24 20:28:03 +0900 | [diff] [blame] | 384 | */ |
Deepa Vaddireddy | 6690f59 | 2017-03-12 03:31:48 +0530 | [diff] [blame] | 385 | private void readComponentConfiguration() { |
| 386 | Set<ConfigProperty> configProperties = cfgService.getProperties(getClass().getName()); |
| 387 | for (ConfigProperty property : configProperties) { |
Jon Hall | a3fcf67 | 2017-03-28 16:53:22 -0700 | [diff] [blame] | 388 | if ("heartbeatInterval".equals(property.name())) { |
Deepa Vaddireddy | 6690f59 | 2017-03-12 03:31:48 +0530 | [diff] [blame] | 389 | 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 Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 396 | : Integer.parseInt(s.trim()); |
Deepa Vaddireddy | 6690f59 | 2017-03-12 03:31:48 +0530 | [diff] [blame] | 397 | 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 Hall | a3fcf67 | 2017-03-28 16:53:22 -0700 | [diff] [blame] | 405 | if ("phiFailureThreshold".equals(property.name())) { |
Deepa Vaddireddy | 6690f59 | 2017-03-12 03:31:48 +0530 | [diff] [blame] | 406 | 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 Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 413 | : Integer.parseInt(s.trim()); |
Deepa Vaddireddy | 6690f59 | 2017-03-12 03:31:48 +0530 | [diff] [blame] | 414 | setPhiFailureThreshold(newPhiFailureThreshold); |
| 415 | log.info("Configured. Phi failure threshold is configured to {}", |
| 416 | phiFailureThreshold); |
| 417 | } |
| 418 | } |
Jordan Halterman | 4a082ec | 2018-05-14 15:00:42 -0700 | [diff] [blame] | 419 | 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-han | f98df54 | 2016-03-24 20:28:03 +0900 | [diff] [blame] | 434 | } |
| 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 Halterman | 4a082ec | 2018-05-14 15:00:42 -0700 | [diff] [blame] | 464 | * 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-han | f98df54 | 2016-03-24 20:28:03 +0900 | [diff] [blame] | 480 | * Restarts heartbeatSender executor. |
sangyun-han | f98df54 | 2016-03-24 20:28:03 +0900 | [diff] [blame] | 481 | */ |
| 482 | private void restartHeartbeatSender() { |
| 483 | try { |
| 484 | ScheduledExecutorService prevSender = heartBeatSender; |
| 485 | heartBeatSender = Executors.newSingleThreadScheduledExecutor( |
HIGUCHI Yuta | d9e0105 | 2016-04-14 09:31:42 -0700 | [diff] [blame] | 486 | groupedThreads("onos/cluster/membership", "heartbeat-sender-%d", log)); |
sangyun-han | f98df54 | 2016-03-24 20:28:03 +0900 | [diff] [blame] | 487 | heartBeatSender.scheduleWithFixedDelay(this::heartbeat, 0, |
Jordan Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 488 | heartbeatInterval, TimeUnit.MILLISECONDS); |
sangyun-han | f98df54 | 2016-03-24 20:28:03 +0900 | [diff] [blame] | 489 | prevSender.shutdown(); |
| 490 | } catch (Exception e) { |
| 491 | log.warn(e.getMessage()); |
| 492 | } |
| 493 | } |
Jordan Halterman | f70bf46 | 2017-07-29 13:12:00 -0700 | [diff] [blame] | 494 | } |