Thomas Vachuska | 4f1a60c | 2014-10-28 13:39:07 -0700 | [diff] [blame] | 1 | /* |
| 2 | * Copyright 2014 Open Networking Laboratory |
| 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 | */ |
Ayaka Koshibe | 1660969 | 2014-09-23 12:46:15 -0700 | [diff] [blame] | 16 | package org.onlab.onos.cluster.impl; |
| 17 | |
alshabib | 339a3d9 | 2014-09-26 17:54:32 -0700 | [diff] [blame] | 18 | import static com.google.common.base.Preconditions.checkNotNull; |
| 19 | import static org.slf4j.LoggerFactory.getLogger; |
| 20 | |
| 21 | import java.util.Set; |
Ayaka Koshibe | ea5b4ce | 2014-10-11 14:17:17 -0700 | [diff] [blame] | 22 | import java.util.concurrent.atomic.AtomicInteger; |
alshabib | 339a3d9 | 2014-09-26 17:54:32 -0700 | [diff] [blame] | 23 | |
Ayaka Koshibe | 1660969 | 2014-09-23 12:46:15 -0700 | [diff] [blame] | 24 | import org.apache.felix.scr.annotations.Activate; |
tom | 4a5d171 | 2014-09-23 17:49:39 -0700 | [diff] [blame] | 25 | import org.apache.felix.scr.annotations.Component; |
Ayaka Koshibe | 1660969 | 2014-09-23 12:46:15 -0700 | [diff] [blame] | 26 | import org.apache.felix.scr.annotations.Deactivate; |
| 27 | import org.apache.felix.scr.annotations.Reference; |
| 28 | import org.apache.felix.scr.annotations.ReferenceCardinality; |
tom | 4a5d171 | 2014-09-23 17:49:39 -0700 | [diff] [blame] | 29 | import org.apache.felix.scr.annotations.Service; |
Ayaka Koshibe | 3de43ca | 2014-09-26 16:40:23 -0700 | [diff] [blame] | 30 | import org.onlab.onos.cluster.ClusterEvent; |
| 31 | import org.onlab.onos.cluster.ClusterEventListener; |
tom | 4a5d171 | 2014-09-23 17:49:39 -0700 | [diff] [blame] | 32 | import org.onlab.onos.cluster.ClusterService; |
Ayaka Koshibe | ea5b4ce | 2014-10-11 14:17:17 -0700 | [diff] [blame] | 33 | import org.onlab.onos.cluster.ControllerNode; |
Ayaka Koshibe | 1660969 | 2014-09-23 12:46:15 -0700 | [diff] [blame] | 34 | import org.onlab.onos.cluster.NodeId; |
Ayaka Koshibe | abedb09 | 2014-10-20 17:01:31 -0700 | [diff] [blame] | 35 | import org.onlab.onos.cluster.RoleInfo; |
Ayaka Koshibe | 1660969 | 2014-09-23 12:46:15 -0700 | [diff] [blame] | 36 | import org.onlab.onos.event.AbstractListenerRegistry; |
| 37 | import org.onlab.onos.event.EventDeliveryService; |
Yuta HIGUCHI | 80912e6 | 2014-10-12 00:15:47 -0700 | [diff] [blame] | 38 | import org.onlab.onos.mastership.MastershipAdminService; |
| 39 | import org.onlab.onos.mastership.MastershipEvent; |
| 40 | import org.onlab.onos.mastership.MastershipListener; |
| 41 | import org.onlab.onos.mastership.MastershipService; |
| 42 | import org.onlab.onos.mastership.MastershipStore; |
| 43 | import org.onlab.onos.mastership.MastershipStoreDelegate; |
| 44 | import org.onlab.onos.mastership.MastershipTerm; |
| 45 | import org.onlab.onos.mastership.MastershipTermService; |
Ayaka Koshibe | 1660969 | 2014-09-23 12:46:15 -0700 | [diff] [blame] | 46 | import org.onlab.onos.net.DeviceId; |
| 47 | import org.onlab.onos.net.MastershipRole; |
Ayaka Koshibe | 1660969 | 2014-09-23 12:46:15 -0700 | [diff] [blame] | 48 | import org.slf4j.Logger; |
| 49 | |
tom | 4a5d171 | 2014-09-23 17:49:39 -0700 | [diff] [blame] | 50 | @Component(immediate = true) |
| 51 | @Service |
Ayaka Koshibe | 3eed2b0 | 2014-09-23 13:28:05 -0700 | [diff] [blame] | 52 | public class MastershipManager |
alshabib | 339a3d9 | 2014-09-26 17:54:32 -0700 | [diff] [blame] | 53 | implements MastershipService, MastershipAdminService { |
Ayaka Koshibe | 1660969 | 2014-09-23 12:46:15 -0700 | [diff] [blame] | 54 | |
| 55 | private static final String NODE_ID_NULL = "Node ID cannot be null"; |
| 56 | private static final String DEVICE_ID_NULL = "Device ID cannot be null"; |
| 57 | private static final String ROLE_NULL = "Mastership role cannot be null"; |
| 58 | |
| 59 | private final Logger log = getLogger(getClass()); |
| 60 | |
| 61 | protected final AbstractListenerRegistry<MastershipEvent, MastershipListener> |
alshabib | 339a3d9 | 2014-09-26 17:54:32 -0700 | [diff] [blame] | 62 | listenerRegistry = new AbstractListenerRegistry<>(); |
| 63 | |
| 64 | private final MastershipStoreDelegate delegate = new InternalDelegate(); |
Ayaka Koshibe | 1660969 | 2014-09-23 12:46:15 -0700 | [diff] [blame] | 65 | |
| 66 | @Reference(cardinality = ReferenceCardinality.MANDATORY_UNARY) |
| 67 | protected MastershipStore store; |
| 68 | |
| 69 | @Reference(cardinality = ReferenceCardinality.MANDATORY_UNARY) |
| 70 | protected EventDeliveryService eventDispatcher; |
| 71 | |
| 72 | @Reference(cardinality = ReferenceCardinality.MANDATORY_UNARY) |
tom | 4a5d171 | 2014-09-23 17:49:39 -0700 | [diff] [blame] | 73 | protected ClusterService clusterService; |
Ayaka Koshibe | 1660969 | 2014-09-23 12:46:15 -0700 | [diff] [blame] | 74 | |
Ayaka Koshibe | 3de43ca | 2014-09-26 16:40:23 -0700 | [diff] [blame] | 75 | private ClusterEventListener clusterListener = new InternalClusterEventListener(); |
| 76 | |
Ayaka Koshibe | 1660969 | 2014-09-23 12:46:15 -0700 | [diff] [blame] | 77 | @Activate |
| 78 | public void activate() { |
| 79 | eventDispatcher.addSink(MastershipEvent.class, listenerRegistry); |
Ayaka Koshibe | 3de43ca | 2014-09-26 16:40:23 -0700 | [diff] [blame] | 80 | clusterService.addListener(clusterListener); |
alshabib | 339a3d9 | 2014-09-26 17:54:32 -0700 | [diff] [blame] | 81 | store.setDelegate(delegate); |
Ayaka Koshibe | 1660969 | 2014-09-23 12:46:15 -0700 | [diff] [blame] | 82 | log.info("Started"); |
| 83 | } |
| 84 | |
| 85 | @Deactivate |
| 86 | public void deactivate() { |
| 87 | eventDispatcher.removeSink(MastershipEvent.class); |
Ayaka Koshibe | 3de43ca | 2014-09-26 16:40:23 -0700 | [diff] [blame] | 88 | clusterService.removeListener(clusterListener); |
alshabib | 339a3d9 | 2014-09-26 17:54:32 -0700 | [diff] [blame] | 89 | store.unsetDelegate(delegate); |
Ayaka Koshibe | 1660969 | 2014-09-23 12:46:15 -0700 | [diff] [blame] | 90 | log.info("Stopped"); |
| 91 | } |
| 92 | |
Ayaka Koshibe | 1660969 | 2014-09-23 12:46:15 -0700 | [diff] [blame] | 93 | @Override |
| 94 | public void setRole(NodeId nodeId, DeviceId deviceId, MastershipRole role) { |
| 95 | checkNotNull(nodeId, NODE_ID_NULL); |
| 96 | checkNotNull(deviceId, DEVICE_ID_NULL); |
| 97 | checkNotNull(role, ROLE_NULL); |
Ayaka Koshibe | d9f693e | 2014-09-29 18:04:54 -0700 | [diff] [blame] | 98 | |
Ayaka Koshibe | 971a38a | 2014-09-30 11:56:23 -0700 | [diff] [blame] | 99 | MastershipEvent event = null; |
Ayaka Koshibe | e60d452 | 2014-10-28 15:07:00 -0700 | [diff] [blame] | 100 | |
| 101 | switch (role) { |
| 102 | case MASTER: |
| 103 | event = store.setMaster(nodeId, deviceId); |
| 104 | break; |
| 105 | case STANDBY: |
| 106 | event = store.setStandby(nodeId, deviceId); |
| 107 | break; |
| 108 | case NONE: |
| 109 | event = store.relinquishRole(nodeId, deviceId); |
| 110 | break; |
| 111 | default: |
| 112 | log.info("Unknown role; ignoring"); |
| 113 | return; |
Ayaka Koshibe | 971a38a | 2014-09-30 11:56:23 -0700 | [diff] [blame] | 114 | } |
Ayaka Koshibe | d9f693e | 2014-09-29 18:04:54 -0700 | [diff] [blame] | 115 | |
Ayaka Koshibe | 971a38a | 2014-09-30 11:56:23 -0700 | [diff] [blame] | 116 | if (event != null) { |
| 117 | post(event); |
Ayaka Koshibe | a7f044e | 2014-09-23 16:56:20 -0700 | [diff] [blame] | 118 | } |
Ayaka Koshibe | 1660969 | 2014-09-23 12:46:15 -0700 | [diff] [blame] | 119 | } |
| 120 | |
| 121 | @Override |
tom | b41d1ac | 2014-09-24 01:51:24 -0700 | [diff] [blame] | 122 | public MastershipRole getLocalRole(DeviceId deviceId) { |
| 123 | checkNotNull(deviceId, DEVICE_ID_NULL); |
| 124 | return store.getRole(clusterService.getLocalNode().id(), deviceId); |
| 125 | } |
| 126 | |
| 127 | @Override |
| 128 | public void relinquishMastership(DeviceId deviceId) { |
Ayaka Koshibe | c404770 | 2014-10-07 14:43:52 -0700 | [diff] [blame] | 129 | MastershipEvent event = null; |
Ayaka Koshibe | 1c292d7 | 2014-10-08 17:46:07 -0700 | [diff] [blame] | 130 | event = store.relinquishRole( |
| 131 | clusterService.getLocalNode().id(), deviceId); |
Ayaka Koshibe | d9f693e | 2014-09-29 18:04:54 -0700 | [diff] [blame] | 132 | if (event != null) { |
| 133 | post(event); |
| 134 | } |
tom | b41d1ac | 2014-09-24 01:51:24 -0700 | [diff] [blame] | 135 | } |
| 136 | |
| 137 | @Override |
| 138 | public MastershipRole requestRoleFor(DeviceId deviceId) { |
| 139 | checkNotNull(deviceId, DEVICE_ID_NULL); |
| 140 | return store.requestRole(deviceId); |
| 141 | } |
| 142 | |
| 143 | @Override |
Ayaka Koshibe | 1660969 | 2014-09-23 12:46:15 -0700 | [diff] [blame] | 144 | public NodeId getMasterFor(DeviceId deviceId) { |
| 145 | checkNotNull(deviceId, DEVICE_ID_NULL); |
| 146 | return store.getMaster(deviceId); |
| 147 | } |
| 148 | |
| 149 | @Override |
| 150 | public Set<DeviceId> getDevicesOf(NodeId nodeId) { |
| 151 | checkNotNull(nodeId, NODE_ID_NULL); |
| 152 | return store.getDevices(nodeId); |
| 153 | } |
| 154 | |
Ayaka Koshibe | 45503ce | 2014-10-14 11:26:45 -0700 | [diff] [blame] | 155 | @Override |
Ayaka Koshibe | abedb09 | 2014-10-20 17:01:31 -0700 | [diff] [blame] | 156 | public RoleInfo getNodesFor(DeviceId deviceId) { |
Ayaka Koshibe | 45503ce | 2014-10-14 11:26:45 -0700 | [diff] [blame] | 157 | checkNotNull(deviceId, DEVICE_ID_NULL); |
| 158 | return store.getNodes(deviceId); |
| 159 | } |
Ayaka Koshibe | b70d34b | 2014-09-25 15:43:01 -0700 | [diff] [blame] | 160 | |
| 161 | @Override |
| 162 | public MastershipTermService requestTermService() { |
| 163 | return new InternalMastershipTermService(); |
| 164 | } |
| 165 | |
Ayaka Koshibe | 1660969 | 2014-09-23 12:46:15 -0700 | [diff] [blame] | 166 | @Override |
Ayaka Koshibe | 1660969 | 2014-09-23 12:46:15 -0700 | [diff] [blame] | 167 | public void addListener(MastershipListener listener) { |
| 168 | checkNotNull(listener); |
| 169 | listenerRegistry.addListener(listener); |
| 170 | } |
| 171 | |
| 172 | @Override |
| 173 | public void removeListener(MastershipListener listener) { |
| 174 | checkNotNull(listener); |
| 175 | listenerRegistry.removeListener(listener); |
| 176 | } |
| 177 | |
tom | b41d1ac | 2014-09-24 01:51:24 -0700 | [diff] [blame] | 178 | // FIXME: provide wiring to allow events to be triggered by changes within the store |
Ayaka Koshibe | 1660969 | 2014-09-23 12:46:15 -0700 | [diff] [blame] | 179 | |
| 180 | // Posts the specified event to the local event dispatcher. |
| 181 | private void post(MastershipEvent event) { |
| 182 | if (event != null && eventDispatcher != null) { |
| 183 | eventDispatcher.post(event); |
| 184 | } |
| 185 | } |
Ayaka Koshibe | 3eed2b0 | 2014-09-23 13:28:05 -0700 | [diff] [blame] | 186 | |
Ayaka Koshibe | b70d34b | 2014-09-25 15:43:01 -0700 | [diff] [blame] | 187 | private class InternalMastershipTermService implements MastershipTermService { |
| 188 | |
| 189 | @Override |
| 190 | public MastershipTerm getMastershipTerm(DeviceId deviceId) { |
| 191 | return store.getTermFor(deviceId); |
| 192 | } |
| 193 | |
| 194 | } |
| 195 | |
Ayaka Koshibe | 3de43ca | 2014-09-26 16:40:23 -0700 | [diff] [blame] | 196 | //callback for reacting to cluster events |
| 197 | private class InternalClusterEventListener implements ClusterEventListener { |
| 198 | |
Ayaka Koshibe | ea5b4ce | 2014-10-11 14:17:17 -0700 | [diff] [blame] | 199 | // A notion of a local maximum cluster size, used to tie-break. |
| 200 | // Think of a better way to do this. |
| 201 | private AtomicInteger clusterSize; |
| 202 | |
| 203 | InternalClusterEventListener() { |
| 204 | clusterSize = new AtomicInteger(0); |
| 205 | } |
| 206 | |
Ayaka Koshibe | 3de43ca | 2014-09-26 16:40:23 -0700 | [diff] [blame] | 207 | @Override |
| 208 | public void event(ClusterEvent event) { |
| 209 | switch (event.type()) { |
| 210 | //FIXME: worry about addition when the time comes |
| 211 | case INSTANCE_ADDED: |
| 212 | case INSTANCE_ACTIVATED: |
Ayaka Koshibe | ea5b4ce | 2014-10-11 14:17:17 -0700 | [diff] [blame] | 213 | clusterSize.incrementAndGet(); |
| 214 | log.info("instance {} added/activated", event.subject()); |
| 215 | break; |
Ayaka Koshibe | 3de43ca | 2014-09-26 16:40:23 -0700 | [diff] [blame] | 216 | case INSTANCE_REMOVED: |
| 217 | case INSTANCE_DEACTIVATED: |
Ayaka Koshibe | ea5b4ce | 2014-10-11 14:17:17 -0700 | [diff] [blame] | 218 | ControllerNode node = event.subject(); |
| 219 | |
| 220 | if (node.equals(clusterService.getLocalNode())) { |
| 221 | //If we are in smaller cluster, relinquish and return |
| 222 | for (DeviceId device : getDevicesOf(node.id())) { |
| 223 | if (!isInMajority()) { |
| 224 | //own DeviceManager should catch event and tell switch |
| 225 | store.relinquishRole(node.id(), device); |
| 226 | } |
| 227 | } |
| 228 | log.info("broke off from cluster, relinquished devices"); |
| 229 | break; |
| 230 | } |
| 231 | |
| 232 | // if we are the larger one and the removed node(s) are brain dead, |
| 233 | // force relinquish on behalf of disabled node. |
| 234 | // check network channel to do this? |
| 235 | for (DeviceId device : getDevicesOf(node.id())) { |
| 236 | //some things to check: |
| 237 | // 1. we didn't break off as well while we're at it |
| 238 | // 2. others don't pile in and try too - maybe a lock |
| 239 | if (isInMajority()) { |
| 240 | store.relinquishRole(node.id(), device); |
| 241 | } |
| 242 | } |
| 243 | clusterSize.decrementAndGet(); |
| 244 | log.info("instance {} removed/deactivated", event.subject()); |
Ayaka Koshibe | 3de43ca | 2014-09-26 16:40:23 -0700 | [diff] [blame] | 245 | break; |
| 246 | default: |
| 247 | log.warn("unknown cluster event {}", event); |
| 248 | } |
| 249 | } |
| 250 | |
Ayaka Koshibe | ea5b4ce | 2014-10-11 14:17:17 -0700 | [diff] [blame] | 251 | private boolean isInMajority() { |
| 252 | if (clusterService.getNodes().size() > (clusterSize.intValue() / 2)) { |
| 253 | return true; |
| 254 | } |
Ayaka Koshibe | a7384a8 | 2014-10-22 18:59:06 -0700 | [diff] [blame] | 255 | // else { |
Ayaka Koshibe | 67af1f4 | 2014-10-20 15:26:37 -0700 | [diff] [blame] | 256 | //FIXME: break tie for equal-sized clusters, by number of |
| 257 | // connected switches, then masters, then nodeId hash |
Ayaka Koshibe | a7384a8 | 2014-10-22 18:59:06 -0700 | [diff] [blame] | 258 | // problem is, how do we get at channel info cleanly here? |
| 259 | // Also, what's the time hit for a distributed store look-up |
| 260 | // versus channel re-negotiation? bet on the latter being worse. |
| 261 | |
| 262 | // } |
Ayaka Koshibe | ea5b4ce | 2014-10-11 14:17:17 -0700 | [diff] [blame] | 263 | return false; |
| 264 | } |
| 265 | |
Ayaka Koshibe | 3de43ca | 2014-09-26 16:40:23 -0700 | [diff] [blame] | 266 | } |
Ayaka Koshibe | 65efaef | 2014-09-29 18:21:56 -0700 | [diff] [blame] | 267 | |
alshabib | 339a3d9 | 2014-09-26 17:54:32 -0700 | [diff] [blame] | 268 | public class InternalDelegate implements MastershipStoreDelegate { |
| 269 | |
| 270 | @Override |
| 271 | public void notify(MastershipEvent event) { |
Ayaka Koshibe | e60d452 | 2014-10-28 15:07:00 -0700 | [diff] [blame] | 272 | if (clusterService.getLocalNode().id().equals(event.roleInfo().master())) { |
| 273 | log.info("ignoring locally-generated event {}", event); |
| 274 | // return; |
| 275 | } |
alshabib | 339a3d9 | 2014-09-26 17:54:32 -0700 | [diff] [blame] | 276 | log.info("dispatching mastership event {}", event); |
| 277 | eventDispatcher.post(event); |
| 278 | } |
| 279 | |
| 280 | } |
| 281 | |
Ayaka Koshibe | 1660969 | 2014-09-23 12:46:15 -0700 | [diff] [blame] | 282 | } |