Madan Jampani | 15b8ef5 | 2016-02-02 17:35:05 -0800 | [diff] [blame] | 1 | /* |
Brian O'Connor | a09fe5b | 2017-08-03 21:12:30 -0700 | [diff] [blame] | 2 | * Copyright 2016-present Open Networking Foundation |
Madan Jampani | 15b8ef5 | 2016-02-02 17:35:05 -0800 | [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 | */ |
| 16 | |
| 17 | package org.onosproject.store.primitives.impl; |
| 18 | |
Madan Jampani | 15b8ef5 | 2016-02-02 17:35:05 -0800 | [diff] [blame] | 19 | import java.io.File; |
Jordan Halterman | 07f052b | 2017-10-08 14:22:41 -0700 | [diff] [blame] | 20 | import java.io.IOException; |
| 21 | import java.nio.file.Files; |
Madan Jampani | e14a09c | 2016-02-11 10:43:21 -0800 | [diff] [blame] | 22 | import java.util.List; |
Madan Jampani | 15b8ef5 | 2016-02-02 17:35:05 -0800 | [diff] [blame] | 23 | import java.util.Map; |
Jordan Halterman | 07f052b | 2017-10-08 14:22:41 -0700 | [diff] [blame] | 24 | import java.util.Optional; |
Madan Jampani | 15b8ef5 | 2016-02-02 17:35:05 -0800 | [diff] [blame] | 25 | import java.util.Set; |
| 26 | import java.util.concurrent.CompletableFuture; |
Madan Jampani | 3354745 | 2016-02-29 16:45:04 -0800 | [diff] [blame] | 27 | import java.util.concurrent.atomic.AtomicReference; |
Madan Jampani | e14a09c | 2016-02-11 10:43:21 -0800 | [diff] [blame] | 28 | import java.util.stream.Collectors; |
Madan Jampani | 15b8ef5 | 2016-02-02 17:35:05 -0800 | [diff] [blame] | 29 | |
Jordan Halterman | 2bf177c | 2017-06-29 01:49:08 -0700 | [diff] [blame] | 30 | import com.google.common.collect.ImmutableSet; |
| 31 | import com.google.common.collect.Maps; |
Madan Jampani | 15b8ef5 | 2016-02-02 17:35:05 -0800 | [diff] [blame] | 32 | import org.apache.felix.scr.annotations.Activate; |
| 33 | import org.apache.felix.scr.annotations.Component; |
Madan Jampani | 86cb243 | 2016-02-17 11:07:56 -0800 | [diff] [blame] | 34 | import org.apache.felix.scr.annotations.Deactivate; |
Madan Jampani | 15b8ef5 | 2016-02-02 17:35:05 -0800 | [diff] [blame] | 35 | import org.apache.felix.scr.annotations.Reference; |
| 36 | import org.apache.felix.scr.annotations.ReferenceCardinality; |
| 37 | import org.apache.felix.scr.annotations.Service; |
| 38 | import org.onlab.util.Tools; |
Jordan Halterman | 07f052b | 2017-10-08 14:22:41 -0700 | [diff] [blame] | 39 | import org.onosproject.cluster.ClusterEvent; |
| 40 | import org.onosproject.cluster.ClusterEventListener; |
Madan Jampani | 3354745 | 2016-02-29 16:45:04 -0800 | [diff] [blame] | 41 | import org.onosproject.cluster.ClusterMetadata; |
| 42 | import org.onosproject.cluster.ClusterMetadataDiff; |
| 43 | import org.onosproject.cluster.ClusterMetadataEvent; |
| 44 | import org.onosproject.cluster.ClusterMetadataEventListener; |
Madan Jampani | 15b8ef5 | 2016-02-02 17:35:05 -0800 | [diff] [blame] | 45 | import org.onosproject.cluster.ClusterMetadataService; |
| 46 | import org.onosproject.cluster.ClusterService; |
Jordan Halterman | 07f052b | 2017-10-08 14:22:41 -0700 | [diff] [blame] | 47 | import org.onosproject.cluster.DefaultPartition; |
| 48 | import org.onosproject.cluster.Member; |
| 49 | import org.onosproject.cluster.MembershipService; |
Madan Jampani | 15b8ef5 | 2016-02-02 17:35:05 -0800 | [diff] [blame] | 50 | import org.onosproject.cluster.NodeId; |
Jordan Halterman | 07f052b | 2017-10-08 14:22:41 -0700 | [diff] [blame] | 51 | import org.onosproject.cluster.Partition; |
Madan Jampani | 3354745 | 2016-02-29 16:45:04 -0800 | [diff] [blame] | 52 | import org.onosproject.cluster.PartitionDiff; |
Madan Jampani | 15b8ef5 | 2016-02-02 17:35:05 -0800 | [diff] [blame] | 53 | import org.onosproject.cluster.PartitionId; |
Jordan Halterman | 980a8c1 | 2017-09-22 18:01:19 -0700 | [diff] [blame] | 54 | import org.onosproject.core.Version; |
Madan Jampani | 15b8ef5 | 2016-02-02 17:35:05 -0800 | [diff] [blame] | 55 | import org.onosproject.event.AbstractListenerManager; |
Jordan Halterman | 28183ee | 2017-10-17 17:29:10 -0700 | [diff] [blame] | 56 | import org.onosproject.store.cluster.messaging.ClusterCommunicationService; |
Madan Jampani | 15b8ef5 | 2016-02-02 17:35:05 -0800 | [diff] [blame] | 57 | import org.onosproject.store.primitives.DistributedPrimitiveCreator; |
| 58 | import org.onosproject.store.primitives.PartitionAdminService; |
| 59 | import org.onosproject.store.primitives.PartitionEvent; |
| 60 | import org.onosproject.store.primitives.PartitionEventListener; |
| 61 | import org.onosproject.store.primitives.PartitionService; |
Madan Jampani | ccdf9da | 2016-05-05 14:37:27 -0700 | [diff] [blame] | 62 | import org.onosproject.store.service.PartitionClientInfo; |
Madan Jampani | e14a09c | 2016-02-11 10:43:21 -0800 | [diff] [blame] | 63 | import org.onosproject.store.service.PartitionInfo; |
Jordan Halterman | 07f052b | 2017-10-08 14:22:41 -0700 | [diff] [blame] | 64 | import org.onosproject.upgrade.Upgrade; |
| 65 | import org.onosproject.upgrade.UpgradeEvent; |
| 66 | import org.onosproject.upgrade.UpgradeEventListener; |
Jordan Halterman | 980a8c1 | 2017-09-22 18:01:19 -0700 | [diff] [blame] | 67 | import org.onosproject.upgrade.UpgradeService; |
Madan Jampani | 15b8ef5 | 2016-02-02 17:35:05 -0800 | [diff] [blame] | 68 | import org.slf4j.Logger; |
| 69 | |
Heedo Kang | 4a47a30 | 2016-02-29 17:40:23 +0900 | [diff] [blame] | 70 | import static org.onosproject.security.AppGuard.checkPermission; |
| 71 | import static org.onosproject.security.AppPermission.Type.PARTITION_READ; |
Jordan Halterman | 2bf177c | 2017-06-29 01:49:08 -0700 | [diff] [blame] | 72 | import static org.slf4j.LoggerFactory.getLogger; |
Heedo Kang | 4a47a30 | 2016-02-29 17:40:23 +0900 | [diff] [blame] | 73 | |
Madan Jampani | 15b8ef5 | 2016-02-02 17:35:05 -0800 | [diff] [blame] | 74 | /** |
| 75 | * Implementation of {@code PartitionService} and {@code PartitionAdminService}. |
| 76 | */ |
| 77 | @Component |
| 78 | @Service |
| 79 | public class PartitionManager extends AbstractListenerManager<PartitionEvent, PartitionEventListener> |
| 80 | implements PartitionService, PartitionAdminService { |
| 81 | |
| 82 | private final Logger log = getLogger(getClass()); |
| 83 | |
| 84 | @Reference(cardinality = ReferenceCardinality.MANDATORY_UNARY) |
Jordan Halterman | 28183ee | 2017-10-17 17:29:10 -0700 | [diff] [blame] | 85 | protected ClusterCommunicationService clusterCommunicator; |
Madan Jampani | 15b8ef5 | 2016-02-02 17:35:05 -0800 | [diff] [blame] | 86 | |
| 87 | @Reference(cardinality = ReferenceCardinality.MANDATORY_UNARY) |
| 88 | protected ClusterMetadataService metadataService; |
| 89 | |
| 90 | @Reference(cardinality = ReferenceCardinality.MANDATORY_UNARY) |
| 91 | protected ClusterService clusterService; |
| 92 | |
Jordan Halterman | 980a8c1 | 2017-09-22 18:01:19 -0700 | [diff] [blame] | 93 | @Reference(cardinality = ReferenceCardinality.MANDATORY_UNARY) |
Jordan Halterman | 07f052b | 2017-10-08 14:22:41 -0700 | [diff] [blame] | 94 | protected MembershipService membershipService; |
Jordan Halterman | 980a8c1 | 2017-09-22 18:01:19 -0700 | [diff] [blame] | 95 | |
| 96 | @Reference(cardinality = ReferenceCardinality.MANDATORY_UNARY) |
Jordan Halterman | 07f052b | 2017-10-08 14:22:41 -0700 | [diff] [blame] | 97 | protected UpgradeService upgradeService; |
Jordan Halterman | 980a8c1 | 2017-09-22 18:01:19 -0700 | [diff] [blame] | 98 | |
| 99 | private final Map<PartitionId, StoragePartition> inactivePartitions = Maps.newConcurrentMap(); |
| 100 | private final Map<PartitionId, StoragePartition> activePartitions = Maps.newConcurrentMap(); |
Madan Jampani | 3354745 | 2016-02-29 16:45:04 -0800 | [diff] [blame] | 101 | private final AtomicReference<ClusterMetadata> currentClusterMetadata = new AtomicReference<>(); |
Jordan Halterman | 07f052b | 2017-10-08 14:22:41 -0700 | [diff] [blame] | 102 | |
| 103 | private final ClusterEventListener clusterListener = new InternalClusterEventListener(); |
| 104 | private final UpgradeEventListener upgradeListener = new InternalUpgradeEventListener(); |
| 105 | private final ClusterMetadataEventListener metadataListener = new InternalClusterMetadataListener(); |
Madan Jampani | 15b8ef5 | 2016-02-02 17:35:05 -0800 | [diff] [blame] | 106 | |
| 107 | @Activate |
| 108 | public void activate() { |
| 109 | eventDispatcher.addSink(PartitionEvent.class, listenerRegistry); |
Madan Jampani | 3354745 | 2016-02-29 16:45:04 -0800 | [diff] [blame] | 110 | currentClusterMetadata.set(metadataService.getClusterMetadata()); |
Jordan Halterman | 07f052b | 2017-10-08 14:22:41 -0700 | [diff] [blame] | 111 | |
| 112 | clusterService.addListener(clusterListener); |
| 113 | upgradeService.addListener(upgradeListener); |
Madan Jampani | 3354745 | 2016-02-29 16:45:04 -0800 | [diff] [blame] | 114 | metadataService.addListener(metadataListener); |
Madan Jampani | 15b8ef5 | 2016-02-02 17:35:05 -0800 | [diff] [blame] | 115 | |
Jordan Halterman | 980a8c1 | 2017-09-22 18:01:19 -0700 | [diff] [blame] | 116 | // If an upgrade is currently in progress and this node is an upgraded node, initialize upgrade partitions. |
| 117 | CompletableFuture<Void> openFuture; |
| 118 | if (upgradeService.isUpgrading() && upgradeService.isLocalUpgraded()) { |
Jordan Halterman | 980a8c1 | 2017-09-22 18:01:19 -0700 | [diff] [blame] | 119 | currentClusterMetadata.get() |
| 120 | .getPartitions() |
Jordan Halterman | 28183ee | 2017-10-17 17:29:10 -0700 | [diff] [blame] | 121 | .forEach(partition -> { |
Jordan Halterman | 07f052b | 2017-10-08 14:22:41 -0700 | [diff] [blame] | 122 | // Create a default partition and assign it to inactive partitions. This node will join |
| 123 | // inactive partitions to participate in consensus for fault tolerance, but the partitions |
| 124 | // won't be accessible via client proxies. |
| 125 | inactivePartitions.put(partition.getId(), new InactiveStoragePartition( |
Jordan Halterman | 28183ee | 2017-10-17 17:29:10 -0700 | [diff] [blame] | 126 | partition, |
Jordan Halterman | 28183ee | 2017-10-17 17:29:10 -0700 | [diff] [blame] | 127 | clusterCommunicator, |
Jordan Halterman | 07f052b | 2017-10-08 14:22:41 -0700 | [diff] [blame] | 128 | clusterService)); |
| 129 | |
| 130 | // Create a forked partition and assign it to active partitions. These partitions will be |
| 131 | // forked from commit logs for previous version partitions. |
| 132 | Partition forkedPartition = computeInitialPartition( |
Jordan Halterman | 28183ee | 2017-10-17 17:29:10 -0700 | [diff] [blame] | 133 | partition, |
Jordan Halterman | 07f052b | 2017-10-08 14:22:41 -0700 | [diff] [blame] | 134 | upgradeService.getState().target(), |
| 135 | getLocalNodes()); |
| 136 | activePartitions.put(partition.getId(), new ForkedStoragePartition( |
| 137 | forkedPartition, |
| 138 | partition, |
Jordan Halterman | 28183ee | 2017-10-17 17:29:10 -0700 | [diff] [blame] | 139 | clusterCommunicator, |
Jordan Halterman | 07f052b | 2017-10-08 14:22:41 -0700 | [diff] [blame] | 140 | clusterService)); |
Jordan Halterman | 28183ee | 2017-10-17 17:29:10 -0700 | [diff] [blame] | 141 | }); |
Jordan Halterman | 980a8c1 | 2017-09-22 18:01:19 -0700 | [diff] [blame] | 142 | |
| 143 | // We have to fork existing partitions before we can start inactive partition servers to |
| 144 | // avoid duplicate message handlers when both servers are running. |
| 145 | openFuture = CompletableFuture.allOf(activePartitions.values().stream() |
| 146 | .map(StoragePartition::open) |
| 147 | .toArray(CompletableFuture[]::new)) |
| 148 | .thenCompose(v -> CompletableFuture.allOf(inactivePartitions.values().stream() |
| 149 | .map(StoragePartition::open) |
| 150 | .toArray(CompletableFuture[]::new))); |
| 151 | } else { |
Jordan Halterman | 980a8c1 | 2017-09-22 18:01:19 -0700 | [diff] [blame] | 152 | currentClusterMetadata.get() |
| 153 | .getPartitions() |
Jordan Halterman | 07f052b | 2017-10-08 14:22:41 -0700 | [diff] [blame] | 154 | .forEach(partition -> activePartitions.put(partition.getId(), new ActiveStoragePartition( |
Jordan Halterman | 980a8c1 | 2017-09-22 18:01:19 -0700 | [diff] [blame] | 155 | partition, |
Jordan Halterman | 980a8c1 | 2017-09-22 18:01:19 -0700 | [diff] [blame] | 156 | clusterCommunicator, |
Jordan Halterman | 07f052b | 2017-10-08 14:22:41 -0700 | [diff] [blame] | 157 | clusterService))); |
Jordan Halterman | 980a8c1 | 2017-09-22 18:01:19 -0700 | [diff] [blame] | 158 | openFuture = CompletableFuture.allOf(activePartitions.values().stream() |
| 159 | .map(StoragePartition::open) |
| 160 | .toArray(CompletableFuture[]::new)); |
| 161 | } |
| 162 | |
Madan Jampani | 15b8ef5 | 2016-02-02 17:35:05 -0800 | [diff] [blame] | 163 | openFuture.join(); |
| 164 | log.info("Started"); |
| 165 | } |
| 166 | |
Madan Jampani | 86cb243 | 2016-02-17 11:07:56 -0800 | [diff] [blame] | 167 | @Deactivate |
Madan Jampani | 15b8ef5 | 2016-02-02 17:35:05 -0800 | [diff] [blame] | 168 | public void deactivate() { |
Jordan Halterman | 07f052b | 2017-10-08 14:22:41 -0700 | [diff] [blame] | 169 | clusterService.removeListener(clusterListener); |
| 170 | upgradeService.removeListener(upgradeListener); |
Madan Jampani | 3354745 | 2016-02-29 16:45:04 -0800 | [diff] [blame] | 171 | metadataService.removeListener(metadataListener); |
Madan Jampani | 15b8ef5 | 2016-02-02 17:35:05 -0800 | [diff] [blame] | 172 | eventDispatcher.removeSink(PartitionEvent.class); |
| 173 | |
Jordan Halterman | 980a8c1 | 2017-09-22 18:01:19 -0700 | [diff] [blame] | 174 | CompletableFuture<Void> closeFuture = CompletableFuture.allOf( |
| 175 | CompletableFuture.allOf(inactivePartitions.values().stream() |
| 176 | .map(StoragePartition::close) |
| 177 | .toArray(CompletableFuture[]::new)), |
| 178 | CompletableFuture.allOf(activePartitions.values().stream() |
| 179 | .map(StoragePartition::close) |
| 180 | .toArray(CompletableFuture[]::new))); |
Madan Jampani | 15b8ef5 | 2016-02-02 17:35:05 -0800 | [diff] [blame] | 181 | closeFuture.join(); |
| 182 | log.info("Stopped"); |
| 183 | } |
| 184 | |
| 185 | @Override |
Madan Jampani | 15b8ef5 | 2016-02-02 17:35:05 -0800 | [diff] [blame] | 186 | public int getNumberOfPartitions() { |
Heedo Kang | 4a47a30 | 2016-02-29 17:40:23 +0900 | [diff] [blame] | 187 | checkPermission(PARTITION_READ); |
Jordan Halterman | 980a8c1 | 2017-09-22 18:01:19 -0700 | [diff] [blame] | 188 | return activePartitions.size(); |
Madan Jampani | 15b8ef5 | 2016-02-02 17:35:05 -0800 | [diff] [blame] | 189 | } |
| 190 | |
| 191 | @Override |
| 192 | public Set<PartitionId> getAllPartitionIds() { |
Heedo Kang | 4a47a30 | 2016-02-29 17:40:23 +0900 | [diff] [blame] | 193 | checkPermission(PARTITION_READ); |
Jordan Halterman | 980a8c1 | 2017-09-22 18:01:19 -0700 | [diff] [blame] | 194 | return activePartitions.keySet(); |
Madan Jampani | 15b8ef5 | 2016-02-02 17:35:05 -0800 | [diff] [blame] | 195 | } |
| 196 | |
| 197 | @Override |
| 198 | public DistributedPrimitiveCreator getDistributedPrimitiveCreator(PartitionId partitionId) { |
Heedo Kang | 4a47a30 | 2016-02-29 17:40:23 +0900 | [diff] [blame] | 199 | checkPermission(PARTITION_READ); |
Jordan Halterman | 980a8c1 | 2017-09-22 18:01:19 -0700 | [diff] [blame] | 200 | return activePartitions.get(partitionId).client(); |
Madan Jampani | 15b8ef5 | 2016-02-02 17:35:05 -0800 | [diff] [blame] | 201 | } |
| 202 | |
| 203 | @Override |
| 204 | public Set<NodeId> getConfiguredMembers(PartitionId partitionId) { |
Heedo Kang | 4a47a30 | 2016-02-29 17:40:23 +0900 | [diff] [blame] | 205 | checkPermission(PARTITION_READ); |
Jordan Halterman | 980a8c1 | 2017-09-22 18:01:19 -0700 | [diff] [blame] | 206 | StoragePartition partition = activePartitions.get(partitionId); |
Madan Jampani | 15b8ef5 | 2016-02-02 17:35:05 -0800 | [diff] [blame] | 207 | return ImmutableSet.copyOf(partition.getMembers()); |
| 208 | } |
| 209 | |
| 210 | @Override |
| 211 | public Set<NodeId> getActiveMembersMembers(PartitionId partitionId) { |
Heedo Kang | 4a47a30 | 2016-02-29 17:40:23 +0900 | [diff] [blame] | 212 | checkPermission(PARTITION_READ); |
Madan Jampani | 15b8ef5 | 2016-02-02 17:35:05 -0800 | [diff] [blame] | 213 | // TODO: This needs to query metadata to determine currently active |
| 214 | // members of partition |
| 215 | return getConfiguredMembers(partitionId); |
| 216 | } |
Madan Jampani | e14a09c | 2016-02-11 10:43:21 -0800 | [diff] [blame] | 217 | |
| 218 | @Override |
| 219 | public List<PartitionInfo> partitionInfo() { |
Jordan Halterman | 980a8c1 | 2017-09-22 18:01:19 -0700 | [diff] [blame] | 220 | return activePartitions.values() |
Madan Jampani | e14a09c | 2016-02-11 10:43:21 -0800 | [diff] [blame] | 221 | .stream() |
Sho SHIMIZU | 5fab6e5 | 2016-02-15 11:54:15 -0800 | [diff] [blame] | 222 | .flatMap(x -> Tools.stream(x.info())) |
Madan Jampani | e14a09c | 2016-02-11 10:43:21 -0800 | [diff] [blame] | 223 | .collect(Collectors.toList()); |
| 224 | } |
Madan Jampani | 3354745 | 2016-02-29 16:45:04 -0800 | [diff] [blame] | 225 | |
Jordan Halterman | 07f052b | 2017-10-08 14:22:41 -0700 | [diff] [blame] | 226 | /** |
| 227 | * Returns a list of nodes sorted by time ordered oldest to newest. |
| 228 | * |
| 229 | * @return a list of nodes sorted by time |
| 230 | */ |
| 231 | private List<NodeId> getLocalNodes() { |
| 232 | return membershipService.getLocalGroup() |
| 233 | .members() |
| 234 | .stream() |
| 235 | .map(Member::nodeId) |
| 236 | .collect(Collectors.toList()); |
| 237 | } |
| 238 | |
| 239 | /** |
| 240 | * Computes an initial forked partition from the given source partition. |
| 241 | * |
| 242 | * @param sourcePartition the source partition from which to compute the partition |
| 243 | * @param targetVersion the target partition version |
| 244 | * @param members the set of members available to the partition |
| 245 | * @return the computed forked partition |
| 246 | */ |
| 247 | protected static Partition computeInitialPartition( |
| 248 | Partition sourcePartition, |
| 249 | Version targetVersion, |
| 250 | List<NodeId> members) { |
| 251 | return computePartition(sourcePartition, targetVersion, members, 1); |
| 252 | } |
| 253 | |
| 254 | /** |
| 255 | * Computes a final forked partition from the given source partition. |
| 256 | * |
| 257 | * @param sourcePartition the source partition from which to compute the partition |
| 258 | * @param targetVersion the target partition version |
| 259 | * @param members the set of members available to the partition |
| 260 | * @return the computed forked partition |
| 261 | */ |
| 262 | protected static Partition computeFinalPartition( |
| 263 | Partition sourcePartition, |
| 264 | Version targetVersion, |
| 265 | List<NodeId> members) { |
| 266 | return computePartition(sourcePartition, targetVersion, members, 0); |
| 267 | } |
| 268 | |
| 269 | /** |
| 270 | * Computes a forked partition from the given source partition. |
| 271 | * |
| 272 | * @param sourcePartition the source partition from which to compute the partition |
| 273 | * @param targetVersion the target partition version |
| 274 | * @param members the set of members available to the partition |
| 275 | * @param delta the number of additional members to preserve outside the partition |
| 276 | * @return the computed forked partition |
| 277 | */ |
| 278 | private static Partition computePartition( |
| 279 | Partition sourcePartition, |
| 280 | Version targetVersion, |
| 281 | List<NodeId> members, |
| 282 | int delta) { |
| 283 | // Create a collection of members of the forked/isolated partition. Initial membership |
| 284 | // will include up to n upgraded nodes until all n nodes in the partition have been upgraded. |
| 285 | List<NodeId> sortedMembers = members.stream() |
| 286 | .sorted() |
| 287 | .collect(Collectors.toList()); |
| 288 | |
| 289 | // Create a list of members of the partition that have been upgraded according to the |
| 290 | // version isolated cluster membership. |
| 291 | List<NodeId> partitionMembers = sortedMembers.stream() |
| 292 | .filter(nodeId -> sourcePartition.getMembers().contains(nodeId)) |
| 293 | .collect(Collectors.toList()); |
| 294 | |
| 295 | // If additional members need to be added to the partition to make up a full member list, |
| 296 | // add members in sorted order to create deterministic rebalancing of nodes. |
| 297 | int totalMembers = sourcePartition.getMembers().size() + delta; |
| 298 | if (partitionMembers.size() < totalMembers) { |
| 299 | for (int i = partitionMembers.size(); i < totalMembers; i++) { |
| 300 | Optional<NodeId> nextMember = sortedMembers.stream() |
| 301 | .filter(nodeId -> !partitionMembers.contains(nodeId)) |
| 302 | .findFirst(); |
| 303 | if (nextMember.isPresent()) { |
| 304 | partitionMembers.add(nextMember.get()); |
| 305 | } else { |
| 306 | break; |
| 307 | } |
| 308 | } |
| 309 | } |
| 310 | |
| 311 | return new DefaultPartition( |
| 312 | sourcePartition.getId(), |
| 313 | targetVersion, |
| 314 | partitionMembers); |
| 315 | } |
| 316 | |
| 317 | private void processInstanceReady(NodeId nodeId) { |
| 318 | if (upgradeService.isUpgrading() && upgradeService.isLocalUpgraded()) { |
| 319 | currentClusterMetadata.get() |
| 320 | .getPartitions() |
| 321 | .forEach(partition -> { |
| 322 | StoragePartition activePartition = activePartitions.get(partition.getId()); |
| 323 | if (activePartition != null) { |
| 324 | Partition newPartition = computeFinalPartition( |
| 325 | partition, |
| 326 | upgradeService.getState().target(), |
| 327 | getLocalNodes()); |
| 328 | log.info("Updating storage partition {}: {}", partition, newPartition); |
| 329 | activePartition.onUpdate(newPartition); |
| 330 | } |
| 331 | }); |
| 332 | } |
| 333 | } |
| 334 | |
| 335 | private void processUpgradeComplete(Upgrade upgrade) { |
| 336 | if (!inactivePartitions.isEmpty()) { |
| 337 | List<CompletableFuture<Void>> futures = inactivePartitions.values() |
| 338 | .stream() |
| 339 | .map(StoragePartition::delete) |
| 340 | .collect(Collectors.toList()); |
| 341 | CompletableFuture.allOf(futures.toArray(new CompletableFuture[futures.size()])).thenRun(() -> { |
| 342 | try { |
| 343 | Files.delete(new File(InactiveStoragePartition.INACTIVE_DIR).toPath()); |
| 344 | } catch (IOException e) { |
| 345 | log.error("Failed to delete partition archive"); |
| 346 | } |
| 347 | }); |
| 348 | inactivePartitions.clear(); |
| 349 | } |
| 350 | } |
| 351 | |
Madan Jampani | 3354745 | 2016-02-29 16:45:04 -0800 | [diff] [blame] | 352 | private void processMetadataUpdate(ClusterMetadata clusterMetadata) { |
| 353 | ClusterMetadataDiff diffExaminer = |
| 354 | new ClusterMetadataDiff(currentClusterMetadata.get(), clusterMetadata); |
| 355 | diffExaminer.partitionDiffs() |
| 356 | .values() |
| 357 | .stream() |
Madan Jampani | 3354745 | 2016-02-29 16:45:04 -0800 | [diff] [blame] | 358 | .filter(PartitionDiff::hasChanged) |
Jordan Halterman | 980a8c1 | 2017-09-22 18:01:19 -0700 | [diff] [blame] | 359 | .forEach(diff -> activePartitions.get(diff.partitionId()).onUpdate(diff.newValue())); |
jiangrui | 9e956a5 | 2017-11-07 10:33:24 +0800 | [diff] [blame] | 360 | currentClusterMetadata.set(clusterMetadata); |
Madan Jampani | 3354745 | 2016-02-29 16:45:04 -0800 | [diff] [blame] | 361 | } |
| 362 | |
Jordan Halterman | 07f052b | 2017-10-08 14:22:41 -0700 | [diff] [blame] | 363 | private class InternalClusterEventListener implements ClusterEventListener { |
| 364 | @Override |
| 365 | public void event(ClusterEvent event) { |
| 366 | if (event.type() == ClusterEvent.Type.INSTANCE_READY) { |
| 367 | processInstanceReady(event.subject().id()); |
| 368 | } |
| 369 | } |
| 370 | } |
| 371 | |
| 372 | private class InternalUpgradeEventListener implements UpgradeEventListener { |
| 373 | @Override |
| 374 | public void event(UpgradeEvent event) { |
| 375 | if (event.type() == UpgradeEvent.Type.COMMITTED) { |
| 376 | processUpgradeComplete(event.subject()); |
| 377 | } |
| 378 | } |
| 379 | } |
| 380 | |
Madan Jampani | 3354745 | 2016-02-29 16:45:04 -0800 | [diff] [blame] | 381 | private class InternalClusterMetadataListener implements ClusterMetadataEventListener { |
| 382 | @Override |
| 383 | public void event(ClusterMetadataEvent event) { |
| 384 | processMetadataUpdate(event.subject()); |
| 385 | } |
| 386 | } |
Madan Jampani | ccdf9da | 2016-05-05 14:37:27 -0700 | [diff] [blame] | 387 | |
| 388 | @Override |
| 389 | public List<PartitionClientInfo> partitionClientInfo() { |
Jordan Halterman | 980a8c1 | 2017-09-22 18:01:19 -0700 | [diff] [blame] | 390 | return activePartitions.values() |
Madan Jampani | ccdf9da | 2016-05-05 14:37:27 -0700 | [diff] [blame] | 391 | .stream() |
| 392 | .map(StoragePartition::client) |
| 393 | .map(StoragePartitionClient::clientInfo) |
| 394 | .collect(Collectors.toList()); |
| 395 | } |
Madan Jampani | 2f9cc71 | 2016-02-15 19:36:21 -0800 | [diff] [blame] | 396 | } |