2 * Copyright (c) 2014 Cisco Systems, Inc. and others. All rights reserved.
4 * This program and the accompanying materials are made available under the
5 * terms of the Eclipse Public License v1.0 which accompanies this distribution,
6 * and is available at http://www.eclipse.org/legal/epl-v10.html
9 package org.opendaylight.controller.cluster.datastore;
11 import static akka.pattern.Patterns.ask;
12 import akka.actor.ActorRef;
13 import akka.actor.Address;
14 import akka.actor.Cancellable;
15 import akka.actor.OneForOneStrategy;
16 import akka.actor.PoisonPill;
17 import akka.actor.Props;
18 import akka.actor.Status;
19 import akka.actor.SupervisorStrategy;
20 import akka.cluster.ClusterEvent;
21 import akka.dispatch.Futures;
22 import akka.dispatch.OnComplete;
23 import akka.japi.Function;
24 import akka.pattern.Patterns;
25 import akka.persistence.RecoveryCompleted;
26 import akka.persistence.SaveSnapshotFailure;
27 import akka.persistence.SaveSnapshotSuccess;
28 import akka.persistence.SnapshotOffer;
29 import akka.persistence.SnapshotSelectionCriteria;
30 import akka.serialization.Serialization;
31 import akka.util.Timeout;
32 import com.google.common.annotations.VisibleForTesting;
33 import com.google.common.base.Objects;
34 import com.google.common.base.Optional;
35 import com.google.common.base.Preconditions;
36 import com.google.common.base.Strings;
37 import com.google.common.base.Supplier;
38 import com.google.common.collect.Sets;
39 import java.io.ByteArrayInputStream;
40 import java.io.ObjectInputStream;
41 import java.io.Serializable;
42 import java.util.ArrayList;
43 import java.util.Collection;
44 import java.util.Collections;
45 import java.util.HashMap;
46 import java.util.HashSet;
47 import java.util.Iterator;
48 import java.util.List;
51 import java.util.concurrent.CountDownLatch;
52 import java.util.concurrent.TimeUnit;
53 import java.util.concurrent.TimeoutException;
54 import javax.annotation.Nonnull;
55 import javax.annotation.Nullable;
56 import org.apache.commons.lang3.SerializationUtils;
57 import org.opendaylight.controller.cluster.common.actor.AbstractUntypedPersistentActorWithMetering;
58 import org.opendaylight.controller.cluster.datastore.config.Configuration;
59 import org.opendaylight.controller.cluster.datastore.config.ModuleShardConfiguration;
60 import org.opendaylight.controller.cluster.datastore.exceptions.AlreadyExistsException;
61 import org.opendaylight.controller.cluster.datastore.exceptions.NoShardLeaderException;
62 import org.opendaylight.controller.cluster.datastore.exceptions.NotInitializedException;
63 import org.opendaylight.controller.cluster.datastore.exceptions.PrimaryNotFoundException;
64 import org.opendaylight.controller.cluster.datastore.identifiers.ShardIdentifier;
65 import org.opendaylight.controller.cluster.datastore.jmx.mbeans.shardmanager.ShardManagerInfo;
66 import org.opendaylight.controller.cluster.datastore.jmx.mbeans.shardmanager.ShardManagerInfoMBean;
67 import org.opendaylight.controller.cluster.datastore.messages.ActorInitialized;
68 import org.opendaylight.controller.cluster.datastore.messages.AddShardReplica;
69 import org.opendaylight.controller.cluster.datastore.messages.CreateShard;
70 import org.opendaylight.controller.cluster.datastore.messages.DatastoreSnapshot;
71 import org.opendaylight.controller.cluster.datastore.messages.FindLocalShard;
72 import org.opendaylight.controller.cluster.datastore.messages.FindPrimary;
73 import org.opendaylight.controller.cluster.datastore.messages.LocalPrimaryShardFound;
74 import org.opendaylight.controller.cluster.datastore.messages.LocalShardFound;
75 import org.opendaylight.controller.cluster.datastore.messages.LocalShardNotFound;
76 import org.opendaylight.controller.cluster.datastore.messages.PeerAddressResolved;
77 import org.opendaylight.controller.cluster.datastore.messages.PeerDown;
78 import org.opendaylight.controller.cluster.datastore.messages.PeerUp;
79 import org.opendaylight.controller.cluster.datastore.messages.RemoteFindPrimary;
80 import org.opendaylight.controller.cluster.datastore.messages.RemotePrimaryShardFound;
81 import org.opendaylight.controller.cluster.datastore.messages.RemoveShardReplica;
82 import org.opendaylight.controller.cluster.datastore.messages.ShardLeaderStateChanged;
83 import org.opendaylight.controller.cluster.datastore.messages.SwitchShardBehavior;
84 import org.opendaylight.controller.cluster.datastore.messages.UpdateSchemaContext;
85 import org.opendaylight.controller.cluster.datastore.utils.Dispatchers;
86 import org.opendaylight.controller.cluster.datastore.utils.PrimaryShardInfoFutureCache;
87 import org.opendaylight.controller.cluster.notifications.RegisterRoleChangeListener;
88 import org.opendaylight.controller.cluster.notifications.RoleChangeNotification;
89 import org.opendaylight.controller.cluster.raft.RaftState;
90 import org.opendaylight.controller.cluster.raft.base.messages.FollowerInitialSyncUpStatus;
91 import org.opendaylight.controller.cluster.raft.base.messages.SwitchBehavior;
92 import org.opendaylight.controller.cluster.raft.client.messages.GetSnapshot;
93 import org.opendaylight.controller.cluster.raft.client.messages.Shutdown;
94 import org.opendaylight.controller.cluster.raft.messages.AddServer;
95 import org.opendaylight.controller.cluster.raft.messages.AddServerReply;
96 import org.opendaylight.controller.cluster.raft.messages.RemoveServer;
97 import org.opendaylight.controller.cluster.raft.messages.RemoveServerReply;
98 import org.opendaylight.controller.cluster.raft.messages.ServerChangeStatus;
99 import org.opendaylight.controller.cluster.raft.messages.ServerRemoved;
100 import org.opendaylight.controller.cluster.raft.policy.DisableElectionsRaftPolicy;
101 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTree;
102 import org.opendaylight.yangtools.yang.model.api.SchemaContext;
103 import org.slf4j.Logger;
104 import org.slf4j.LoggerFactory;
105 import scala.concurrent.ExecutionContext;
106 import scala.concurrent.Future;
107 import scala.concurrent.duration.Duration;
108 import scala.concurrent.duration.FiniteDuration;
111 * The ShardManager has the following jobs,
113 * <li> Create all the local shard replicas that belong on this cluster member
114 * <li> Find the address of the local shard
115 * <li> Find the primary replica for any given shard
116 * <li> Monitor the cluster members and store their addresses
119 public class ShardManager extends AbstractUntypedPersistentActorWithMetering {
121 private static final Logger LOG = LoggerFactory.getLogger(ShardManager.class);
123 // Stores a mapping between a shard name and it's corresponding information
124 // Shard names look like inventory, topology etc and are as specified in
126 private final Map<String, ShardInformation> localShards = new HashMap<>();
128 // The type of a ShardManager reflects the type of the datastore itself
129 // A data store could be of type config/operational
130 private final String type;
132 private final ClusterWrapper cluster;
134 private final Configuration configuration;
136 private final String shardDispatcherPath;
138 private final ShardManagerInfo mBean;
140 private DatastoreContextFactory datastoreContextFactory;
142 private final CountDownLatch waitTillReadyCountdownLatch;
144 private final PrimaryShardInfoFutureCache primaryShardInfoCache;
146 private final ShardPeerAddressResolver peerAddressResolver;
148 private SchemaContext schemaContext;
150 private DatastoreSnapshot restoreFromSnapshot;
152 private ShardManagerSnapshot currentSnapshot;
154 private final Set<String> shardReplicaOperationsInProgress = new HashSet<>();
156 private final String persistenceId;
160 protected ShardManager(AbstractBuilder<?> builder) {
162 this.cluster = builder.cluster;
163 this.configuration = builder.configuration;
164 this.datastoreContextFactory = builder.datastoreContextFactory;
165 this.type = builder.datastoreContextFactory.getBaseDatastoreContext().getDataStoreName();
166 this.shardDispatcherPath =
167 new Dispatchers(context().system().dispatchers()).getDispatcherPath(Dispatchers.DispatcherType.Shard);
168 this.waitTillReadyCountdownLatch = builder.waitTillReadyCountdownLatch;
169 this.primaryShardInfoCache = builder.primaryShardInfoCache;
170 this.restoreFromSnapshot = builder.restoreFromSnapshot;
172 String possiblePersistenceId = datastoreContextFactory.getBaseDatastoreContext().getShardManagerPersistenceId();
173 persistenceId = possiblePersistenceId != null ? possiblePersistenceId : "shard-manager-" + type;
175 peerAddressResolver = new ShardPeerAddressResolver(type, cluster.getCurrentMemberName());
177 // Subscribe this actor to cluster member events
178 cluster.subscribeToMemberEvents(getSelf());
180 List<String> localShardActorNames = new ArrayList<>();
181 mBean = ShardManagerInfo.createShardManagerMBean(cluster.getCurrentMemberName(),
182 "shard-manager-" + this.type,
183 datastoreContextFactory.getBaseDatastoreContext().getDataStoreMXBeanType(),
184 localShardActorNames);
185 mBean.setShardManager(this);
189 public void postStop() {
190 LOG.info("Stopping ShardManager {}", persistenceId());
192 mBean.unregisterMBean();
196 public void handleCommand(Object message) throws Exception {
197 if (message instanceof FindPrimary) {
198 findPrimary((FindPrimary)message);
199 } else if(message instanceof FindLocalShard){
200 findLocalShard((FindLocalShard) message);
201 } else if (message instanceof UpdateSchemaContext) {
202 updateSchemaContext(message);
203 } else if(message instanceof ActorInitialized) {
204 onActorInitialized(message);
205 } else if (message instanceof ClusterEvent.MemberUp){
206 memberUp((ClusterEvent.MemberUp) message);
207 } else if (message instanceof ClusterEvent.MemberExited){
208 memberExited((ClusterEvent.MemberExited) message);
209 } else if(message instanceof ClusterEvent.MemberRemoved) {
210 memberRemoved((ClusterEvent.MemberRemoved) message);
211 } else if(message instanceof ClusterEvent.UnreachableMember) {
212 memberUnreachable((ClusterEvent.UnreachableMember)message);
213 } else if(message instanceof ClusterEvent.ReachableMember) {
214 memberReachable((ClusterEvent.ReachableMember) message);
215 } else if(message instanceof DatastoreContextFactory) {
216 onDatastoreContextFactory((DatastoreContextFactory)message);
217 } else if(message instanceof RoleChangeNotification) {
218 onRoleChangeNotification((RoleChangeNotification) message);
219 } else if(message instanceof FollowerInitialSyncUpStatus){
220 onFollowerInitialSyncStatus((FollowerInitialSyncUpStatus) message);
221 } else if(message instanceof ShardNotInitializedTimeout) {
222 onShardNotInitializedTimeout((ShardNotInitializedTimeout)message);
223 } else if(message instanceof ShardLeaderStateChanged) {
224 onLeaderStateChanged((ShardLeaderStateChanged) message);
225 } else if(message instanceof SwitchShardBehavior){
226 onSwitchShardBehavior((SwitchShardBehavior) message);
227 } else if(message instanceof CreateShard) {
228 onCreateShard((CreateShard)message);
229 } else if(message instanceof AddShardReplica){
230 onAddShardReplica((AddShardReplica)message);
231 } else if(message instanceof ForwardedAddServerReply) {
232 ForwardedAddServerReply msg = (ForwardedAddServerReply)message;
233 onAddServerReply(msg.shardInfo, msg.addServerReply, getSender(), msg.leaderPath,
234 msg.removeShardOnFailure);
235 } else if(message instanceof ForwardedAddServerFailure) {
236 ForwardedAddServerFailure msg = (ForwardedAddServerFailure)message;
237 onAddServerFailure(msg.shardName, msg.failureMessage, msg.failure, getSender(), msg.removeShardOnFailure);
238 } else if(message instanceof PrimaryShardFoundForContext) {
239 PrimaryShardFoundForContext primaryShardFoundContext = (PrimaryShardFoundForContext)message;
240 onPrimaryShardFoundContext(primaryShardFoundContext);
241 } else if(message instanceof RemoveShardReplica) {
242 onRemoveShardReplica((RemoveShardReplica) message);
243 } else if(message instanceof WrappedShardResponse){
244 onWrappedShardResponse((WrappedShardResponse) message);
245 } else if(message instanceof GetSnapshot) {
247 } else if(message instanceof ServerRemoved){
248 onShardReplicaRemoved((ServerRemoved) message);
249 } else if(message instanceof SaveSnapshotSuccess) {
250 onSaveSnapshotSuccess((SaveSnapshotSuccess)message);
251 } else if(message instanceof SaveSnapshotFailure) {
252 LOG.error("{}: SaveSnapshotFailure received for saving snapshot of shards",
253 persistenceId(), ((SaveSnapshotFailure) message).cause());
254 } else if(message instanceof Shutdown) {
257 unknownMessage(message);
261 private void onShutDown() {
262 Shutdown shutdown = new Shutdown();
263 List<Future<Boolean>> stopFutures = new ArrayList<>(localShards.size());
264 for (ShardInformation info : localShards.values()) {
265 if (info.getActor() != null) {
266 LOG.debug("{}: Issuing gracefulStop to shard {}", persistenceId(), info.getShardId());
268 FiniteDuration duration = info.getDatastoreContext().getShardRaftConfig().getElectionTimeOutInterval().$times(2);
269 stopFutures.add(Patterns.gracefulStop(info.getActor(), duration, shutdown));
273 LOG.info("Shutting down ShardManager {} - waiting on {} shards", persistenceId(), stopFutures.size());
275 ExecutionContext dispatcher = new Dispatchers(context().system().dispatchers()).getDispatcher(Dispatchers.DispatcherType.Client);
276 Future<Iterable<Boolean>> combinedFutures = Futures.sequence(stopFutures, dispatcher);
278 combinedFutures.onComplete(new OnComplete<Iterable<Boolean>>() {
280 public void onComplete(Throwable failure, Iterable<Boolean> results) {
281 LOG.debug("{}: All shards shutdown - sending PoisonPill to self", persistenceId());
283 self().tell(PoisonPill.getInstance(), self());
285 if(failure != null) {
286 LOG.warn("{}: An error occurred attempting to shut down the shards", persistenceId(), failure);
289 for(Boolean r: results) {
296 LOG.warn("{}: {} shards did not shut down gracefully", persistenceId(), nfailed);
303 private void onWrappedShardResponse(WrappedShardResponse message) {
304 if (message.getResponse() instanceof RemoveServerReply) {
305 onRemoveServerReply(getSender(), message.getShardId(), (RemoveServerReply) message.getResponse(),
306 message.getLeaderPath());
310 private void onRemoveServerReply(ActorRef originalSender, ShardIdentifier shardId, RemoveServerReply replyMsg,
312 shardReplicaOperationsInProgress.remove(shardId);
314 LOG.debug ("{}: Received {} for shard {}", persistenceId(), replyMsg, shardId.getShardName());
316 if (replyMsg.getStatus() == ServerChangeStatus.OK) {
317 LOG.debug ("{}: Leader shard successfully removed the replica shard {}", persistenceId(),
318 shardId.getShardName());
319 originalSender.tell(new akka.actor.Status.Success(null), getSelf());
321 LOG.warn ("{}: Leader failed to remove shard replica {} with status {}",
322 persistenceId(), shardId, replyMsg.getStatus());
324 Exception failure = getServerChangeException(RemoveServer.class, replyMsg.getStatus(),
325 leaderPath, shardId);
326 originalSender.tell(new akka.actor.Status.Failure(failure), getSelf());
330 private void onPrimaryShardFoundContext(PrimaryShardFoundForContext primaryShardFoundContext) {
331 if(primaryShardFoundContext.getContextMessage() instanceof AddShardReplica) {
332 addShard(primaryShardFoundContext.getShardName(), primaryShardFoundContext.getRemotePrimaryShardFound(),
334 } else if(primaryShardFoundContext.getContextMessage() instanceof RemoveShardReplica){
335 removeShardReplica((RemoveShardReplica) primaryShardFoundContext.getContextMessage(),
336 primaryShardFoundContext.getShardName(), primaryShardFoundContext.getPrimaryPath(), getSender());
340 private void removeShardReplica(RemoveShardReplica contextMessage, final String shardName, final String primaryPath,
341 final ActorRef sender) {
342 if(isShardReplicaOperationInProgress(shardName, sender)) {
346 shardReplicaOperationsInProgress.add(shardName);
348 final ShardIdentifier shardId = getShardIdentifier(contextMessage.getMemberName(), shardName);
350 final DatastoreContext datastoreContext = newShardDatastoreContextBuilder(shardName).build();
352 //inform ShardLeader to remove this shard as a replica by sending an RemoveServer message
353 LOG.debug ("{}: Sending RemoveServer message to peer {} for shard {}", persistenceId(),
354 primaryPath, shardId);
356 Timeout removeServerTimeout = new Timeout(datastoreContext.getShardLeaderElectionTimeout().
358 Future<Object> futureObj = ask(getContext().actorSelection(primaryPath),
359 new RemoveServer(shardId.toString()), removeServerTimeout);
361 futureObj.onComplete(new OnComplete<Object>() {
363 public void onComplete(Throwable failure, Object response) {
364 if (failure != null) {
365 String msg = String.format("RemoveServer request to leader %s for shard %s failed",
366 primaryPath, shardName);
368 LOG.debug ("{}: {}", persistenceId(), msg, failure);
371 sender.tell(new Status.Failure(new RuntimeException(msg, failure)), self());
374 self().tell(new WrappedShardResponse(shardId, response, primaryPath), sender);
377 }, new Dispatchers(context().system().dispatchers()).getDispatcher(Dispatchers.DispatcherType.Client));
380 private void onShardReplicaRemoved(ServerRemoved message) {
381 final ShardIdentifier shardId = new ShardIdentifier.Builder().fromShardIdString(message.getServerId()).build();
382 final ShardInformation shardInformation = localShards.remove(shardId.getShardName());
383 if(shardInformation == null) {
384 LOG.debug("{} : Shard replica {} is not present in list", persistenceId(), shardId.toString());
386 } else if(shardInformation.getActor() != null) {
387 LOG.debug("{} : Sending Shutdown to Shard actor {}", persistenceId(), shardInformation.getActor());
388 shardInformation.getActor().tell(new Shutdown(), self());
390 LOG.debug("{} : Local Shard replica for shard {} has been removed", persistenceId(), shardId.getShardName());
394 private void onGetSnapshot() {
395 LOG.debug("{}: onGetSnapshot", persistenceId());
397 List<String> notInitialized = null;
398 for(ShardInformation shardInfo: localShards.values()) {
399 if(!shardInfo.isShardInitialized()) {
400 if(notInitialized == null) {
401 notInitialized = new ArrayList<>();
404 notInitialized.add(shardInfo.getShardName());
408 if(notInitialized != null) {
409 getSender().tell(new akka.actor.Status.Failure(new IllegalStateException(String.format(
410 "%d shard(s) %s are not initialized", notInitialized.size(), notInitialized))), getSelf());
414 byte[] shardManagerSnapshot = null;
415 if(currentSnapshot != null) {
416 shardManagerSnapshot = SerializationUtils.serialize(currentSnapshot);
419 ActorRef replyActor = getContext().actorOf(ShardManagerGetSnapshotReplyActor.props(
420 new ArrayList<>(localShards.keySet()), type, shardManagerSnapshot , getSender(), persistenceId(),
421 datastoreContextFactory.getBaseDatastoreContext().getShardInitializationTimeout().duration()));
423 for(ShardInformation shardInfo: localShards.values()) {
424 shardInfo.getActor().tell(GetSnapshot.INSTANCE, replyActor);
428 private void onCreateShard(CreateShard createShard) {
429 LOG.debug("{}: onCreateShard: {}", persistenceId(), createShard);
433 String shardName = createShard.getModuleShardConfig().getShardName();
434 if(localShards.containsKey(shardName)) {
435 LOG.debug("{}: Shard {} already exists", persistenceId(), shardName);
436 reply = new akka.actor.Status.Success(String.format("Shard with name %s already exists", shardName));
438 doCreateShard(createShard);
439 reply = new akka.actor.Status.Success(null);
441 } catch (Exception e) {
442 LOG.error("{}: onCreateShard failed", persistenceId(), e);
443 reply = new akka.actor.Status.Failure(e);
446 if(getSender() != null && !getContext().system().deadLetters().equals(getSender())) {
447 getSender().tell(reply, getSelf());
451 private void doCreateShard(CreateShard createShard) {
452 ModuleShardConfiguration moduleShardConfig = createShard.getModuleShardConfig();
453 String shardName = moduleShardConfig.getShardName();
455 configuration.addModuleShardConfiguration(moduleShardConfig);
457 DatastoreContext shardDatastoreContext = createShard.getDatastoreContext();
458 if(shardDatastoreContext == null) {
459 shardDatastoreContext = newShardDatastoreContext(shardName);
461 shardDatastoreContext = DatastoreContext.newBuilderFrom(shardDatastoreContext).shardPeerAddressResolver(
462 peerAddressResolver).build();
465 ShardIdentifier shardId = getShardIdentifier(cluster.getCurrentMemberName(), shardName);
467 boolean shardWasInRecoveredSnapshot = currentSnapshot != null &&
468 currentSnapshot.getShardList().contains(shardName);
470 Map<String, String> peerAddresses;
471 boolean isActiveMember;
472 if(shardWasInRecoveredSnapshot || configuration.getMembersFromShardName(shardName).
473 contains(cluster.getCurrentMemberName())) {
474 peerAddresses = getPeerAddresses(shardName);
475 isActiveMember = true;
477 // The local member is not in the static shard member configuration and the shard did not
478 // previously exist (ie !shardWasInRecoveredSnapshot). In this case we'll create
479 // the shard with no peers and with elections disabled so it stays as follower. A
480 // subsequent AddServer request will be needed to make it an active member.
481 isActiveMember = false;
482 peerAddresses = Collections.emptyMap();
483 shardDatastoreContext = DatastoreContext.newBuilderFrom(shardDatastoreContext).
484 customRaftPolicyImplementation(DisableElectionsRaftPolicy.class.getName()).build();
487 LOG.debug("{} doCreateShard: shardId: {}, memberNames: {}, peerAddresses: {}, isActiveMember: {}",
488 persistenceId(), shardId, moduleShardConfig.getShardMemberNames(), peerAddresses,
491 ShardInformation info = new ShardInformation(shardName, shardId, peerAddresses,
492 shardDatastoreContext, createShard.getShardBuilder(), peerAddressResolver);
493 info.setActiveMember(isActiveMember);
494 localShards.put(info.getShardName(), info);
496 mBean.addLocalShard(shardId.toString());
498 if(schemaContext != null) {
499 info.setActor(newShardActor(schemaContext, info));
503 private DatastoreContext.Builder newShardDatastoreContextBuilder(String shardName) {
504 return DatastoreContext.newBuilderFrom(datastoreContextFactory.getShardDatastoreContext(shardName)).
505 shardPeerAddressResolver(peerAddressResolver);
508 private DatastoreContext newShardDatastoreContext(String shardName) {
509 return newShardDatastoreContextBuilder(shardName).build();
512 private void checkReady(){
513 if (isReadyWithLeaderId()) {
514 LOG.info("{}: All Shards are ready - data store {} is ready, available count is {}",
515 persistenceId(), type, waitTillReadyCountdownLatch.getCount());
517 waitTillReadyCountdownLatch.countDown();
521 private void onLeaderStateChanged(ShardLeaderStateChanged leaderStateChanged) {
522 LOG.info("{}: Received LeaderStateChanged message: {}", persistenceId(), leaderStateChanged);
524 ShardInformation shardInformation = findShardInformation(leaderStateChanged.getMemberId());
525 if(shardInformation != null) {
526 shardInformation.setLocalDataTree(leaderStateChanged.getLocalShardDataTree());
527 shardInformation.setLeaderVersion(leaderStateChanged.getLeaderPayloadVersion());
528 if(shardInformation.setLeaderId(leaderStateChanged.getLeaderId())) {
529 primaryShardInfoCache.remove(shardInformation.getShardName());
534 LOG.debug("No shard found with member Id {}", leaderStateChanged.getMemberId());
538 private void onShardNotInitializedTimeout(ShardNotInitializedTimeout message) {
539 ShardInformation shardInfo = message.getShardInfo();
541 LOG.debug("{}: Received ShardNotInitializedTimeout message for shard {}", persistenceId(),
542 shardInfo.getShardName());
544 shardInfo.removeOnShardInitialized(message.getOnShardInitialized());
546 if(!shardInfo.isShardInitialized()) {
547 LOG.debug("{}: Returning NotInitializedException for shard {}", persistenceId(), shardInfo.getShardName());
548 message.getSender().tell(createNotInitializedException(shardInfo.shardId), getSelf());
550 LOG.debug("{}: Returning NoShardLeaderException for shard {}", persistenceId(), shardInfo.getShardName());
551 message.getSender().tell(createNoShardLeaderException(shardInfo.shardId), getSelf());
555 private void onFollowerInitialSyncStatus(FollowerInitialSyncUpStatus status) {
556 LOG.info("{} Received follower initial sync status for {} status sync done {}", persistenceId(),
557 status.getName(), status.isInitialSyncDone());
559 ShardInformation shardInformation = findShardInformation(status.getName());
561 if(shardInformation != null) {
562 shardInformation.setFollowerSyncStatus(status.isInitialSyncDone());
564 mBean.setSyncStatus(isInSync());
569 private void onRoleChangeNotification(RoleChangeNotification roleChanged) {
570 LOG.info("{}: Received role changed for {} from {} to {}", persistenceId(), roleChanged.getMemberId(),
571 roleChanged.getOldRole(), roleChanged.getNewRole());
573 ShardInformation shardInformation = findShardInformation(roleChanged.getMemberId());
574 if(shardInformation != null) {
575 shardInformation.setRole(roleChanged.getNewRole());
577 mBean.setSyncStatus(isInSync());
582 private ShardInformation findShardInformation(String memberId) {
583 for(ShardInformation info : localShards.values()){
584 if(info.getShardId().toString().equals(memberId)){
592 private boolean isReadyWithLeaderId() {
593 boolean isReady = true;
594 for (ShardInformation info : localShards.values()) {
595 if(!info.isShardReadyWithLeaderId()){
603 private boolean isInSync(){
604 for (ShardInformation info : localShards.values()) {
605 if(!info.isInSync()){
612 private void onActorInitialized(Object message) {
613 final ActorRef sender = getSender();
615 if (sender == null) {
616 return; //why is a non-actor sending this message? Just ignore.
619 String actorName = sender.path().name();
620 //find shard name from actor name; actor name is stringified shardId
621 ShardIdentifier shardId = ShardIdentifier.builder().fromShardIdString(actorName).build();
623 if (shardId.getShardName() == null) {
627 markShardAsInitialized(shardId.getShardName());
630 private void markShardAsInitialized(String shardName) {
631 LOG.debug("{}: Initializing shard [{}]", persistenceId(), shardName);
633 ShardInformation shardInformation = localShards.get(shardName);
634 if (shardInformation != null) {
635 shardInformation.setActorInitialized();
637 shardInformation.getActor().tell(new RegisterRoleChangeListener(), self());
642 protected void handleRecover(Object message) throws Exception {
643 if (message instanceof RecoveryCompleted) {
644 onRecoveryCompleted();
645 } else if (message instanceof SnapshotOffer) {
646 applyShardManagerSnapshot((ShardManagerSnapshot)((SnapshotOffer) message).snapshot());
650 private void onRecoveryCompleted() {
651 LOG.info("Recovery complete : {}", persistenceId());
653 // We no longer persist SchemaContext modules so delete all the prior messages from the akka
654 // journal on upgrade from Helium.
655 deleteMessages(lastSequenceNr());
657 if(currentSnapshot == null && restoreFromSnapshot != null &&
658 restoreFromSnapshot.getShardManagerSnapshot() != null) {
659 try(ObjectInputStream ois = new ObjectInputStream(new ByteArrayInputStream(
660 restoreFromSnapshot.getShardManagerSnapshot()))) {
661 ShardManagerSnapshot snapshot = (ShardManagerSnapshot) ois.readObject();
663 LOG.debug("{}: Deserialized restored ShardManagerSnapshot: {}", persistenceId(), snapshot);
665 applyShardManagerSnapshot(snapshot);
666 } catch(Exception e) {
667 LOG.error("{}: Error deserializing restored ShardManagerSnapshot", persistenceId(), e);
674 private void findLocalShard(FindLocalShard message) {
675 final ShardInformation shardInformation = localShards.get(message.getShardName());
677 if(shardInformation == null){
678 getSender().tell(new LocalShardNotFound(message.getShardName()), getSelf());
682 sendResponse(shardInformation, message.isWaitUntilInitialized(), false, new Supplier<Object>() {
684 public Object get() {
685 return new LocalShardFound(shardInformation.getActor());
690 private void sendResponse(ShardInformation shardInformation, boolean doWait,
691 boolean wantShardReady, final Supplier<Object> messageSupplier) {
692 if (!shardInformation.isShardInitialized() || (wantShardReady && !shardInformation.isShardReadyWithLeaderId())) {
694 final ActorRef sender = getSender();
695 final ActorRef self = self();
697 Runnable replyRunnable = new Runnable() {
700 sender.tell(messageSupplier.get(), self);
704 OnShardInitialized onShardInitialized = wantShardReady ? new OnShardReady(replyRunnable) :
705 new OnShardInitialized(replyRunnable);
707 shardInformation.addOnShardInitialized(onShardInitialized);
709 FiniteDuration timeout = shardInformation.getDatastoreContext().getShardInitializationTimeout().duration();
710 if(shardInformation.isShardInitialized()) {
711 // If the shard is already initialized then we'll wait enough time for the shard to
712 // elect a leader, ie 2 times the election timeout.
713 timeout = FiniteDuration.create(shardInformation.getDatastoreContext().getShardRaftConfig()
714 .getElectionTimeOutInterval().toMillis() * 2, TimeUnit.MILLISECONDS);
717 LOG.debug("{}: Scheduling {} ms timer to wait for shard {}", persistenceId(), timeout.toMillis(),
718 shardInformation.getShardName());
720 Cancellable timeoutSchedule = getContext().system().scheduler().scheduleOnce(
722 new ShardNotInitializedTimeout(shardInformation, onShardInitialized, sender),
723 getContext().dispatcher(), getSelf());
725 onShardInitialized.setTimeoutSchedule(timeoutSchedule);
727 } else if (!shardInformation.isShardInitialized()) {
728 LOG.debug("{}: Returning NotInitializedException for shard {}", persistenceId(),
729 shardInformation.getShardName());
730 getSender().tell(createNotInitializedException(shardInformation.shardId), getSelf());
732 LOG.debug("{}: Returning NoShardLeaderException for shard {}", persistenceId(),
733 shardInformation.getShardName());
734 getSender().tell(createNoShardLeaderException(shardInformation.shardId), getSelf());
740 getSender().tell(messageSupplier.get(), getSelf());
743 private static NoShardLeaderException createNoShardLeaderException(ShardIdentifier shardId) {
744 return new NoShardLeaderException(null, shardId.toString());
747 private static NotInitializedException createNotInitializedException(ShardIdentifier shardId) {
748 return new NotInitializedException(String.format(
749 "Found primary shard %s but it's not initialized yet. Please try again later", shardId));
752 private void memberRemoved(ClusterEvent.MemberRemoved message) {
753 String memberName = message.member().roles().iterator().next();
755 LOG.debug("{}: Received MemberRemoved: memberName: {}, address: {}", persistenceId(), memberName,
756 message.member().address());
758 peerAddressResolver.removePeerAddress(memberName);
760 for(ShardInformation info : localShards.values()){
761 info.peerDown(memberName, getShardIdentifier(memberName, info.getShardName()).toString(), getSelf());
765 private void memberExited(ClusterEvent.MemberExited message) {
766 String memberName = message.member().roles().iterator().next();
768 LOG.debug("{}: Received MemberExited: memberName: {}, address: {}", persistenceId(), memberName,
769 message.member().address());
771 peerAddressResolver.removePeerAddress(memberName);
773 for(ShardInformation info : localShards.values()){
774 info.peerDown(memberName, getShardIdentifier(memberName, info.getShardName()).toString(), getSelf());
778 private void memberUp(ClusterEvent.MemberUp message) {
779 String memberName = message.member().roles().iterator().next();
781 LOG.debug("{}: Received MemberUp: memberName: {}, address: {}", persistenceId(), memberName,
782 message.member().address());
784 addPeerAddress(memberName, message.member().address());
789 private void addPeerAddress(String memberName, Address address) {
790 peerAddressResolver.addPeerAddress(memberName, address);
792 for(ShardInformation info : localShards.values()){
793 String shardName = info.getShardName();
794 String peerId = getShardIdentifier(memberName, shardName).toString();
795 info.updatePeerAddress(peerId, peerAddressResolver.getShardActorAddress(shardName, memberName), getSelf());
797 info.peerUp(memberName, peerId, getSelf());
801 private void memberReachable(ClusterEvent.ReachableMember message) {
802 String memberName = message.member().roles().iterator().next();
803 LOG.debug("Received ReachableMember: memberName {}, address: {}", memberName, message.member().address());
805 addPeerAddress(memberName, message.member().address());
807 markMemberAvailable(memberName);
810 private void memberUnreachable(ClusterEvent.UnreachableMember message) {
811 String memberName = message.member().roles().iterator().next();
812 LOG.debug("Received UnreachableMember: memberName {}, address: {}", memberName, message.member().address());
814 markMemberUnavailable(memberName);
817 private void markMemberUnavailable(final String memberName) {
818 for(ShardInformation info : localShards.values()){
819 String leaderId = info.getLeaderId();
820 if(leaderId != null && leaderId.contains(memberName)) {
821 LOG.debug("Marking Leader {} as unavailable.", leaderId);
822 info.setLeaderAvailable(false);
824 primaryShardInfoCache.remove(info.getShardName());
827 info.peerDown(memberName, getShardIdentifier(memberName, info.getShardName()).toString(), getSelf());
831 private void markMemberAvailable(final String memberName) {
832 for(ShardInformation info : localShards.values()){
833 String leaderId = info.getLeaderId();
834 if(leaderId != null && leaderId.contains(memberName)) {
835 LOG.debug("Marking Leader {} as available.", leaderId);
836 info.setLeaderAvailable(true);
839 info.peerUp(memberName, getShardIdentifier(memberName, info.getShardName()).toString(), getSelf());
843 private void onDatastoreContextFactory(DatastoreContextFactory factory) {
844 datastoreContextFactory = factory;
845 for (ShardInformation info : localShards.values()) {
846 info.setDatastoreContext(newShardDatastoreContext(info.getShardName()), getSelf());
850 private void onSwitchShardBehavior(SwitchShardBehavior message) {
851 ShardIdentifier identifier = ShardIdentifier.builder().fromShardIdString(message.getShardName()).build();
853 ShardInformation shardInformation = localShards.get(identifier.getShardName());
855 if(shardInformation != null && shardInformation.getActor() != null) {
856 shardInformation.getActor().tell(
857 new SwitchBehavior(RaftState.valueOf(message.getNewState()), message.getTerm()), getSelf());
859 LOG.warn("Could not switch the behavior of shard {} to {} - shard is not yet available",
860 message.getShardName(), message.getNewState());
865 * Notifies all the local shards of a change in the schema context
869 private void updateSchemaContext(final Object message) {
870 schemaContext = ((UpdateSchemaContext) message).getSchemaContext();
872 LOG.debug("Got updated SchemaContext: # of modules {}", schemaContext.getAllModuleIdentifiers().size());
874 for (ShardInformation info : localShards.values()) {
875 if (info.getActor() == null) {
876 LOG.debug("Creating Shard {}", info.getShardId());
877 info.setActor(newShardActor(schemaContext, info));
879 info.getActor().tell(message, getSelf());
885 protected ClusterWrapper getCluster() {
890 protected ActorRef newShardActor(final SchemaContext schemaContext, ShardInformation info) {
891 return getContext().actorOf(info.newProps(schemaContext)
892 .withDispatcher(shardDispatcherPath), info.getShardId().toString());
895 private void findPrimary(FindPrimary message) {
896 LOG.debug("{}: In findPrimary: {}", persistenceId(), message);
898 final String shardName = message.getShardName();
899 final boolean canReturnLocalShardState = !(message instanceof RemoteFindPrimary);
901 // First see if the there is a local replica for the shard
902 final ShardInformation info = localShards.get(shardName);
903 if (info != null && info.isActiveMember()) {
904 sendResponse(info, message.isWaitUntilReady(), true, new Supplier<Object>() {
906 public Object get() {
907 String primaryPath = info.getSerializedLeaderActor();
908 Object found = canReturnLocalShardState && info.isLeader() ?
909 new LocalPrimaryShardFound(primaryPath, info.getLocalShardDataTree().get()) :
910 new RemotePrimaryShardFound(primaryPath, info.getLeaderVersion());
912 if(LOG.isDebugEnabled()) {
913 LOG.debug("{}: Found primary for {}: {}", persistenceId(), shardName, found);
923 Collection<String> visitedAddresses;
924 if(message instanceof RemoteFindPrimary) {
925 visitedAddresses = ((RemoteFindPrimary)message).getVisitedAddresses();
927 visitedAddresses = new ArrayList<>();
930 visitedAddresses.add(peerAddressResolver.getShardManagerActorPathBuilder(cluster.getSelfAddress()).toString());
932 for(String address: peerAddressResolver.getShardManagerPeerActorAddresses()) {
933 if(visitedAddresses.contains(address)) {
937 LOG.debug("{}: findPrimary for {} forwarding to remote ShardManager {}", persistenceId(),
940 getContext().actorSelection(address).forward(new RemoteFindPrimary(shardName,
941 message.isWaitUntilReady(), visitedAddresses), getContext());
945 LOG.debug("{}: No shard found for {}", persistenceId(), shardName);
947 getSender().tell(new PrimaryNotFoundException(
948 String.format("No primary shard found for %s.", shardName)), getSelf());
952 * Construct the name of the shard actor given the name of the member on
953 * which the shard resides and the name of the shard
959 private ShardIdentifier getShardIdentifier(String memberName, String shardName){
960 return peerAddressResolver.getShardIdentifier(memberName, shardName);
964 * Create shards that are local to the member on which the ShardManager
968 private void createLocalShards() {
969 String memberName = this.cluster.getCurrentMemberName();
970 Collection<String> memberShardNames = this.configuration.getMemberShardNames(memberName);
972 Map<String, DatastoreSnapshot.ShardSnapshot> shardSnapshots = new HashMap<>();
973 if(restoreFromSnapshot != null)
975 for(DatastoreSnapshot.ShardSnapshot snapshot: restoreFromSnapshot.getShardSnapshots()) {
976 shardSnapshots.put(snapshot.getName(), snapshot);
980 restoreFromSnapshot = null; // null out to GC
982 for(String shardName : memberShardNames){
983 ShardIdentifier shardId = getShardIdentifier(memberName, shardName);
985 LOG.debug("{}: Creating local shard: {}", persistenceId(), shardId);
987 Map<String, String> peerAddresses = getPeerAddresses(shardName);
988 localShards.put(shardName, new ShardInformation(shardName, shardId, peerAddresses,
989 newShardDatastoreContext(shardName), Shard.builder().restoreFromSnapshot(
990 shardSnapshots.get(shardName)), peerAddressResolver));
991 mBean.addLocalShard(shardId.toString());
996 * Given the name of the shard find the addresses of all it's peers
1000 private Map<String, String> getPeerAddresses(String shardName) {
1001 Collection<String> members = configuration.getMembersFromShardName(shardName);
1002 Map<String, String> peerAddresses = new HashMap<>();
1004 String currentMemberName = this.cluster.getCurrentMemberName();
1006 for(String memberName : members) {
1007 if(!currentMemberName.equals(memberName)) {
1008 ShardIdentifier shardId = getShardIdentifier(memberName, shardName);
1009 String address = peerAddressResolver.getShardActorAddress(shardName, memberName);
1010 peerAddresses.put(shardId.toString(), address);
1013 return peerAddresses;
1017 public SupervisorStrategy supervisorStrategy() {
1019 return new OneForOneStrategy(10, Duration.create("1 minute"),
1020 new Function<Throwable, SupervisorStrategy.Directive>() {
1022 public SupervisorStrategy.Directive apply(Throwable t) {
1023 LOG.warn("Supervisor Strategy caught unexpected exception - resuming", t);
1024 return SupervisorStrategy.resume();
1032 public String persistenceId() {
1033 return persistenceId;
1037 ShardManagerInfoMBean getMBean(){
1041 private boolean isShardReplicaOperationInProgress(final String shardName, final ActorRef sender) {
1042 if (shardReplicaOperationsInProgress.contains(shardName)) {
1043 String msg = String.format("A shard replica operation for %s is already in progress", shardName);
1044 LOG.debug ("{}: {}", persistenceId(), msg);
1045 sender.tell(new akka.actor.Status.Failure(new IllegalStateException(msg)), getSelf());
1052 private void onAddShardReplica (final AddShardReplica shardReplicaMsg) {
1053 final String shardName = shardReplicaMsg.getShardName();
1055 LOG.debug("{}: onAddShardReplica: {}", persistenceId(), shardReplicaMsg);
1057 // verify the shard with the specified name is present in the cluster configuration
1058 if (!(this.configuration.isShardConfigured(shardName))) {
1059 String msg = String.format("No module configuration exists for shard %s", shardName);
1060 LOG.debug ("{}: {}", persistenceId(), msg);
1061 getSender().tell(new akka.actor.Status.Failure(new IllegalArgumentException(msg)), getSelf());
1065 // Create the localShard
1066 if (schemaContext == null) {
1067 String msg = String.format(
1068 "No SchemaContext is available in order to create a local shard instance for %s", shardName);
1069 LOG.debug ("{}: {}", persistenceId(), msg);
1070 getSender().tell(new akka.actor.Status.Failure(new IllegalStateException(msg)), getSelf());
1074 findPrimary(shardName, new AutoFindPrimaryFailureResponseHandler(getSender(), shardName, persistenceId(), getSelf()) {
1076 public void onRemotePrimaryShardFound(RemotePrimaryShardFound response) {
1077 getSelf().tell(new PrimaryShardFoundForContext(getShardName(), shardReplicaMsg, response), getTargetActor());
1081 public void onLocalPrimaryFound(LocalPrimaryShardFound message) {
1082 sendLocalReplicaAlreadyExistsReply(getShardName(), getTargetActor());
1088 private void sendLocalReplicaAlreadyExistsReply(String shardName, ActorRef sender) {
1089 String msg = String.format("Local shard %s already exists", shardName);
1090 LOG.debug ("{}: {}", persistenceId(), msg);
1091 sender.tell(new akka.actor.Status.Failure(new AlreadyExistsException(msg)), getSelf());
1094 private void addShard(final String shardName, final RemotePrimaryShardFound response, final ActorRef sender) {
1095 if(isShardReplicaOperationInProgress(shardName, sender)) {
1099 shardReplicaOperationsInProgress.add(shardName);
1101 final ShardInformation shardInfo;
1102 final boolean removeShardOnFailure;
1103 ShardInformation existingShardInfo = localShards.get(shardName);
1104 if(existingShardInfo == null) {
1105 removeShardOnFailure = true;
1106 ShardIdentifier shardId = getShardIdentifier(cluster.getCurrentMemberName(), shardName);
1108 DatastoreContext datastoreContext = newShardDatastoreContextBuilder(shardName).customRaftPolicyImplementation(
1109 DisableElectionsRaftPolicy.class.getName()).build();
1111 shardInfo = new ShardInformation(shardName, shardId, getPeerAddresses(shardName), datastoreContext,
1112 Shard.builder(), peerAddressResolver);
1113 shardInfo.setActiveMember(false);
1114 localShards.put(shardName, shardInfo);
1115 shardInfo.setActor(newShardActor(schemaContext, shardInfo));
1117 removeShardOnFailure = false;
1118 shardInfo = existingShardInfo;
1121 String localShardAddress = peerAddressResolver.getShardActorAddress(shardName, cluster.getCurrentMemberName());
1123 //inform ShardLeader to add this shard as a replica by sending an AddServer message
1124 LOG.debug ("{}: Sending AddServer message to peer {} for shard {}", persistenceId(),
1125 response.getPrimaryPath(), shardInfo.getShardId());
1127 Timeout addServerTimeout = new Timeout(shardInfo.getDatastoreContext().getShardLeaderElectionTimeout().
1129 Future<Object> futureObj = ask(getContext().actorSelection(response.getPrimaryPath()),
1130 new AddServer(shardInfo.getShardId().toString(), localShardAddress, true), addServerTimeout);
1132 futureObj.onComplete(new OnComplete<Object>() {
1134 public void onComplete(Throwable failure, Object addServerResponse) {
1135 if (failure != null) {
1136 LOG.debug ("{}: AddServer request to {} for {} failed", persistenceId(),
1137 response.getPrimaryPath(), shardName, failure);
1139 String msg = String.format("AddServer request to leader %s for shard %s failed",
1140 response.getPrimaryPath(), shardName);
1141 self().tell(new ForwardedAddServerFailure(shardName, msg, failure, removeShardOnFailure), sender);
1143 self().tell(new ForwardedAddServerReply(shardInfo, (AddServerReply)addServerResponse,
1144 response.getPrimaryPath(), removeShardOnFailure), sender);
1147 }, new Dispatchers(context().system().dispatchers()).getDispatcher(Dispatchers.DispatcherType.Client));
1150 private void onAddServerFailure(String shardName, String message, Throwable failure, ActorRef sender,
1151 boolean removeShardOnFailure) {
1152 shardReplicaOperationsInProgress.remove(shardName);
1154 if(removeShardOnFailure) {
1155 ShardInformation shardInfo = localShards.remove(shardName);
1156 if (shardInfo.getActor() != null) {
1157 shardInfo.getActor().tell(PoisonPill.getInstance(), getSelf());
1161 sender.tell(new akka.actor.Status.Failure(message == null ? failure :
1162 new RuntimeException(message, failure)), getSelf());
1165 private void onAddServerReply(ShardInformation shardInfo, AddServerReply replyMsg, ActorRef sender,
1166 String leaderPath, boolean removeShardOnFailure) {
1167 String shardName = shardInfo.getShardName();
1168 shardReplicaOperationsInProgress.remove(shardName);
1170 LOG.debug ("{}: Received {} for shard {} from leader {}", persistenceId(), replyMsg, shardName, leaderPath);
1172 if (replyMsg.getStatus() == ServerChangeStatus.OK) {
1173 LOG.debug ("{}: Leader shard successfully added the replica shard {}", persistenceId(), shardName);
1175 // Make the local shard voting capable
1176 shardInfo.setDatastoreContext(newShardDatastoreContext(shardName), getSelf());
1177 shardInfo.setActiveMember(true);
1180 mBean.addLocalShard(shardInfo.getShardId().toString());
1181 sender.tell(new akka.actor.Status.Success(null), getSelf());
1182 } else if(replyMsg.getStatus() == ServerChangeStatus.ALREADY_EXISTS) {
1183 sendLocalReplicaAlreadyExistsReply(shardName, sender);
1185 LOG.warn ("{}: Leader failed to add shard replica {} with status {}",
1186 persistenceId(), shardName, replyMsg.getStatus());
1188 Exception failure = getServerChangeException(AddServer.class, replyMsg.getStatus(), leaderPath, shardInfo.getShardId());
1190 onAddServerFailure(shardName, null, failure, sender, removeShardOnFailure);
1194 private Exception getServerChangeException(Class<?> serverChange, ServerChangeStatus serverChangeStatus,
1195 String leaderPath, ShardIdentifier shardId) {
1197 switch (serverChangeStatus) {
1199 failure = new TimeoutException(String.format(
1200 "The shard leader %s timed out trying to replicate the initial data to the new shard %s." +
1201 "Possible causes - there was a problem replicating the data or shard leadership changed while replicating the shard data",
1202 leaderPath, shardId.getShardName()));
1205 failure = createNoShardLeaderException(shardId);
1208 failure = new UnsupportedOperationException(String.format("%s request is not supported for shard %s",
1209 serverChange.getSimpleName(), shardId.getShardName()));
1212 failure = new RuntimeException(String.format(
1213 "%s request to leader %s for shard %s failed with status %s",
1214 serverChange.getSimpleName(), leaderPath, shardId.getShardName(), serverChangeStatus));
1219 private void onRemoveShardReplica (final RemoveShardReplica shardReplicaMsg) {
1220 LOG.debug("{}: onRemoveShardReplica: {}", persistenceId(), shardReplicaMsg);
1222 findPrimary(shardReplicaMsg.getShardName(), new AutoFindPrimaryFailureResponseHandler(getSender(),
1223 shardReplicaMsg.getShardName(), persistenceId(), getSelf()) {
1225 public void onRemotePrimaryShardFound(RemotePrimaryShardFound response) {
1226 getSelf().tell(new PrimaryShardFoundForContext(getShardName(), shardReplicaMsg, response), getTargetActor());
1230 public void onLocalPrimaryFound(LocalPrimaryShardFound response) {
1231 getSelf().tell(new PrimaryShardFoundForContext(getShardName(), shardReplicaMsg, response), getTargetActor());
1236 private void persistShardList() {
1237 List<String> shardList = new ArrayList<>(localShards.keySet());
1238 for (ShardInformation shardInfo : localShards.values()) {
1239 if (!shardInfo.isActiveMember()) {
1240 shardList.remove(shardInfo.getShardName());
1243 LOG.debug ("{}: persisting the shard list {}", persistenceId(), shardList);
1244 saveSnapshot(updateShardManagerSnapshot(shardList));
1247 private ShardManagerSnapshot updateShardManagerSnapshot(List<String> shardList) {
1248 currentSnapshot = new ShardManagerSnapshot(shardList);
1249 return currentSnapshot;
1252 private void applyShardManagerSnapshot(ShardManagerSnapshot snapshot) {
1253 currentSnapshot = snapshot;
1255 LOG.debug ("{}: onSnapshotOffer: {}", persistenceId(), currentSnapshot);
1257 String currentMember = cluster.getCurrentMemberName();
1258 Set<String> configuredShardList =
1259 new HashSet<>(configuration.getMemberShardNames(currentMember));
1260 for (String shard : currentSnapshot.getShardList()) {
1261 if (!configuredShardList.contains(shard)) {
1262 // add the current member as a replica for the shard
1263 LOG.debug ("{}: adding shard {}", persistenceId(), shard);
1264 configuration.addMemberReplicaForShard(shard, currentMember);
1266 configuredShardList.remove(shard);
1269 for (String shard : configuredShardList) {
1270 // remove the member as a replica for the shard
1271 LOG.debug ("{}: removing shard {}", persistenceId(), shard);
1272 configuration.removeMemberReplicaForShard(shard, currentMember);
1276 private void onSaveSnapshotSuccess (SaveSnapshotSuccess successMessage) {
1277 LOG.debug ("{} saved ShardManager snapshot successfully. Deleting the prev snapshot if available",
1279 deleteSnapshots(new SnapshotSelectionCriteria(scala.Long.MaxValue(), (successMessage.metadata().timestamp() - 1)));
1282 private static class ForwardedAddServerReply {
1283 ShardInformation shardInfo;
1284 AddServerReply addServerReply;
1286 boolean removeShardOnFailure;
1288 ForwardedAddServerReply(ShardInformation shardInfo, AddServerReply addServerReply, String leaderPath,
1289 boolean removeShardOnFailure) {
1290 this.shardInfo = shardInfo;
1291 this.addServerReply = addServerReply;
1292 this.leaderPath = leaderPath;
1293 this.removeShardOnFailure = removeShardOnFailure;
1297 private static class ForwardedAddServerFailure {
1299 String failureMessage;
1301 boolean removeShardOnFailure;
1303 ForwardedAddServerFailure(String shardName, String failureMessage, Throwable failure,
1304 boolean removeShardOnFailure) {
1305 this.shardName = shardName;
1306 this.failureMessage = failureMessage;
1307 this.failure = failure;
1308 this.removeShardOnFailure = removeShardOnFailure;
1313 protected static class ShardInformation {
1314 private final ShardIdentifier shardId;
1315 private final String shardName;
1316 private ActorRef actor;
1317 private final Map<String, String> initialPeerAddresses;
1318 private Optional<DataTree> localShardDataTree;
1319 private boolean leaderAvailable = false;
1321 // flag that determines if the actor is ready for business
1322 private boolean actorInitialized = false;
1324 private boolean followerSyncStatus = false;
1326 private final Set<OnShardInitialized> onShardInitializedSet = Sets.newHashSet();
1327 private String role ;
1328 private String leaderId;
1329 private short leaderVersion;
1331 private DatastoreContext datastoreContext;
1332 private Shard.AbstractBuilder<?, ?> builder;
1333 private final ShardPeerAddressResolver addressResolver;
1334 private boolean isActiveMember = true;
1336 private ShardInformation(String shardName, ShardIdentifier shardId,
1337 Map<String, String> initialPeerAddresses, DatastoreContext datastoreContext,
1338 Shard.AbstractBuilder<?, ?> builder, ShardPeerAddressResolver addressResolver) {
1339 this.shardName = shardName;
1340 this.shardId = shardId;
1341 this.initialPeerAddresses = initialPeerAddresses;
1342 this.datastoreContext = datastoreContext;
1343 this.builder = builder;
1344 this.addressResolver = addressResolver;
1347 Props newProps(SchemaContext schemaContext) {
1348 Preconditions.checkNotNull(builder);
1349 Props props = builder.id(shardId).peerAddresses(initialPeerAddresses).datastoreContext(datastoreContext).
1350 schemaContext(schemaContext).props();
1355 String getShardName() {
1360 ActorRef getActor(){
1364 void setActor(ActorRef actor) {
1368 ShardIdentifier getShardId() {
1372 void setLocalDataTree(Optional<DataTree> localShardDataTree) {
1373 this.localShardDataTree = localShardDataTree;
1376 Optional<DataTree> getLocalShardDataTree() {
1377 return localShardDataTree;
1380 DatastoreContext getDatastoreContext() {
1381 return datastoreContext;
1384 void setDatastoreContext(DatastoreContext datastoreContext, ActorRef sender) {
1385 this.datastoreContext = datastoreContext;
1386 if (actor != null) {
1387 LOG.debug ("Sending new DatastoreContext to {}", shardId);
1388 actor.tell(this.datastoreContext, sender);
1392 void updatePeerAddress(String peerId, String peerAddress, ActorRef sender){
1393 LOG.info("updatePeerAddress for peer {} with address {}", peerId, peerAddress);
1396 if(LOG.isDebugEnabled()) {
1397 LOG.debug("Sending PeerAddressResolved for peer {} with address {} to {}",
1398 peerId, peerAddress, actor.path());
1401 actor.tell(new PeerAddressResolved(peerId, peerAddress), sender);
1404 notifyOnShardInitializedCallbacks();
1407 void peerDown(String memberName, String peerId, ActorRef sender) {
1409 actor.tell(new PeerDown(memberName, peerId), sender);
1413 void peerUp(String memberName, String peerId, ActorRef sender) {
1415 actor.tell(new PeerUp(memberName, peerId), sender);
1419 boolean isShardReady() {
1420 return !RaftState.Candidate.name().equals(role) && !Strings.isNullOrEmpty(role);
1423 boolean isShardReadyWithLeaderId() {
1424 return leaderAvailable && isShardReady() && !RaftState.IsolatedLeader.name().equals(role) &&
1425 (isLeader() || addressResolver.resolve(leaderId) != null);
1428 boolean isShardInitialized() {
1429 return getActor() != null && actorInitialized;
1432 boolean isLeader() {
1433 return Objects.equal(leaderId, shardId.toString());
1436 String getSerializedLeaderActor() {
1438 return Serialization.serializedActorPath(getActor());
1440 return addressResolver.resolve(leaderId);
1444 void setActorInitialized() {
1445 LOG.debug("Shard {} is initialized", shardId);
1447 this.actorInitialized = true;
1449 notifyOnShardInitializedCallbacks();
1452 private void notifyOnShardInitializedCallbacks() {
1453 if(onShardInitializedSet.isEmpty()) {
1457 boolean ready = isShardReadyWithLeaderId();
1459 if(LOG.isDebugEnabled()) {
1460 LOG.debug("Shard {} is {} - notifying {} OnShardInitialized callbacks", shardId,
1461 ready ? "ready" : "initialized", onShardInitializedSet.size());
1464 Iterator<OnShardInitialized> iter = onShardInitializedSet.iterator();
1465 while(iter.hasNext()) {
1466 OnShardInitialized onShardInitialized = iter.next();
1467 if(!(onShardInitialized instanceof OnShardReady) || ready) {
1469 onShardInitialized.getTimeoutSchedule().cancel();
1470 onShardInitialized.getReplyRunnable().run();
1475 void addOnShardInitialized(OnShardInitialized onShardInitialized) {
1476 onShardInitializedSet.add(onShardInitialized);
1479 void removeOnShardInitialized(OnShardInitialized onShardInitialized) {
1480 onShardInitializedSet.remove(onShardInitialized);
1483 void setRole(String newRole) {
1484 this.role = newRole;
1486 notifyOnShardInitializedCallbacks();
1489 void setFollowerSyncStatus(boolean syncStatus){
1490 this.followerSyncStatus = syncStatus;
1494 if(RaftState.Follower.name().equals(this.role)){
1495 return followerSyncStatus;
1496 } else if(RaftState.Leader.name().equals(this.role)){
1503 boolean setLeaderId(String leaderId) {
1504 boolean changed = !Objects.equal(this.leaderId, leaderId);
1505 this.leaderId = leaderId;
1506 if(leaderId != null) {
1507 this.leaderAvailable = true;
1509 notifyOnShardInitializedCallbacks();
1514 String getLeaderId() {
1518 void setLeaderAvailable(boolean leaderAvailable) {
1519 this.leaderAvailable = leaderAvailable;
1521 if(leaderAvailable) {
1522 notifyOnShardInitializedCallbacks();
1526 short getLeaderVersion() {
1527 return leaderVersion;
1530 void setLeaderVersion(short leaderVersion) {
1531 this.leaderVersion = leaderVersion;
1534 boolean isActiveMember() {
1535 return isActiveMember;
1538 void setActiveMember(boolean isActiveMember) {
1539 this.isActiveMember = isActiveMember;
1543 private static class OnShardInitialized {
1544 private final Runnable replyRunnable;
1545 private Cancellable timeoutSchedule;
1547 OnShardInitialized(Runnable replyRunnable) {
1548 this.replyRunnable = replyRunnable;
1551 Runnable getReplyRunnable() {
1552 return replyRunnable;
1555 Cancellable getTimeoutSchedule() {
1556 return timeoutSchedule;
1559 void setTimeoutSchedule(Cancellable timeoutSchedule) {
1560 this.timeoutSchedule = timeoutSchedule;
1564 private static class OnShardReady extends OnShardInitialized {
1565 OnShardReady(Runnable replyRunnable) {
1566 super(replyRunnable);
1570 private static class ShardNotInitializedTimeout {
1571 private final ActorRef sender;
1572 private final ShardInformation shardInfo;
1573 private final OnShardInitialized onShardInitialized;
1575 ShardNotInitializedTimeout(ShardInformation shardInfo, OnShardInitialized onShardInitialized, ActorRef sender) {
1576 this.sender = sender;
1577 this.shardInfo = shardInfo;
1578 this.onShardInitialized = onShardInitialized;
1581 ActorRef getSender() {
1585 ShardInformation getShardInfo() {
1589 OnShardInitialized getOnShardInitialized() {
1590 return onShardInitialized;
1595 * We no longer persist SchemaContextModules but keep this class around for now for backwards
1596 * compatibility so we don't get de-serialization failures on upgrade from Helium.
1599 static class SchemaContextModules implements Serializable {
1600 private static final long serialVersionUID = -8884620101025936590L;
1602 private final Set<String> modules;
1604 SchemaContextModules(Set<String> modules){
1605 this.modules = modules;
1608 public Set<String> getModules() {
1613 public static Builder builder() {
1614 return new Builder();
1617 public static abstract class AbstractBuilder<T extends AbstractBuilder<T>> {
1618 private ClusterWrapper cluster;
1619 private Configuration configuration;
1620 private DatastoreContextFactory datastoreContextFactory;
1621 private CountDownLatch waitTillReadyCountdownLatch;
1622 private PrimaryShardInfoFutureCache primaryShardInfoCache;
1623 private DatastoreSnapshot restoreFromSnapshot;
1624 private volatile boolean sealed;
1626 @SuppressWarnings("unchecked")
1631 protected void checkSealed() {
1632 Preconditions.checkState(!sealed, "Builder is already sealed - further modifications are not allowed");
1635 public T cluster(ClusterWrapper cluster) {
1637 this.cluster = cluster;
1641 public T configuration(Configuration configuration) {
1643 this.configuration = configuration;
1647 public T datastoreContextFactory(DatastoreContextFactory datastoreContextFactory) {
1649 this.datastoreContextFactory = datastoreContextFactory;
1653 public T waitTillReadyCountdownLatch(CountDownLatch waitTillReadyCountdownLatch) {
1655 this.waitTillReadyCountdownLatch = waitTillReadyCountdownLatch;
1659 public T primaryShardInfoCache(PrimaryShardInfoFutureCache primaryShardInfoCache) {
1661 this.primaryShardInfoCache = primaryShardInfoCache;
1665 public T restoreFromSnapshot(DatastoreSnapshot restoreFromSnapshot) {
1667 this.restoreFromSnapshot = restoreFromSnapshot;
1671 protected void verify() {
1673 Preconditions.checkNotNull(cluster, "cluster should not be null");
1674 Preconditions.checkNotNull(configuration, "configuration should not be null");
1675 Preconditions.checkNotNull(datastoreContextFactory, "datastoreContextFactory should not be null");
1676 Preconditions.checkNotNull(waitTillReadyCountdownLatch, "waitTillReadyCountdownLatch should not be null");
1677 Preconditions.checkNotNull(primaryShardInfoCache, "primaryShardInfoCache should not be null");
1680 public Props props() {
1682 return Props.create(ShardManager.class, this);
1686 public static class Builder extends AbstractBuilder<Builder> {
1689 private void findPrimary(final String shardName, final FindPrimaryResponseHandler handler) {
1690 Timeout findPrimaryTimeout = new Timeout(datastoreContextFactory.getBaseDatastoreContext().
1691 getShardInitializationTimeout().duration().$times(2));
1694 Future<Object> futureObj = ask(getSelf(), new FindPrimary(shardName, true), findPrimaryTimeout);
1695 futureObj.onComplete(new OnComplete<Object>() {
1697 public void onComplete(Throwable failure, Object response) {
1698 if (failure != null) {
1699 handler.onFailure(failure);
1701 if(response instanceof RemotePrimaryShardFound) {
1702 handler.onRemotePrimaryShardFound((RemotePrimaryShardFound) response);
1703 } else if(response instanceof LocalPrimaryShardFound) {
1704 handler.onLocalPrimaryFound((LocalPrimaryShardFound) response);
1706 handler.onUnknownResponse(response);
1710 }, new Dispatchers(context().system().dispatchers()).getDispatcher(Dispatchers.DispatcherType.Client));
1714 * The FindPrimaryResponseHandler provides specific callback methods which are invoked when a response to the
1715 * a remote or local find primary message is processed
1717 private static interface FindPrimaryResponseHandler {
1719 * Invoked when a Failure message is received as a response
1723 void onFailure(Throwable failure);
1726 * Invoked when a RemotePrimaryShardFound response is received
1730 void onRemotePrimaryShardFound(RemotePrimaryShardFound response);
1733 * Invoked when a LocalPrimaryShardFound response is received
1736 void onLocalPrimaryFound(LocalPrimaryShardFound response);
1739 * Invoked when an unknown response is received. This is another type of failure.
1743 void onUnknownResponse(Object response);
1747 * The AutoFindPrimaryFailureResponseHandler automatically processes Failure responses when finding a primary
1748 * replica and sends a wrapped Failure response to some targetActor
1750 private static abstract class AutoFindPrimaryFailureResponseHandler implements FindPrimaryResponseHandler {
1751 private final ActorRef targetActor;
1752 private final String shardName;
1753 private final String persistenceId;
1754 private final ActorRef shardManagerActor;
1757 * @param targetActor The actor to whom the Failure response should be sent when a FindPrimary failure occurs
1758 * @param shardName The name of the shard for which the primary replica had to be found
1759 * @param persistenceId The persistenceId for the ShardManager
1760 * @param shardManagerActor The ShardManager actor which triggered the call to FindPrimary
1762 protected AutoFindPrimaryFailureResponseHandler(ActorRef targetActor, String shardName, String persistenceId, ActorRef shardManagerActor){
1763 this.targetActor = Preconditions.checkNotNull(targetActor);
1764 this.shardName = Preconditions.checkNotNull(shardName);
1765 this.persistenceId = Preconditions.checkNotNull(persistenceId);
1766 this.shardManagerActor = Preconditions.checkNotNull(shardManagerActor);
1769 public ActorRef getTargetActor() {
1773 public String getShardName() {
1778 public void onFailure(Throwable failure) {
1779 LOG.debug ("{}: Received failure from FindPrimary for shard {}", persistenceId, shardName, failure);
1780 targetActor.tell(new akka.actor.Status.Failure(new RuntimeException(
1781 String.format("Failed to find leader for shard %s", shardName), failure)), shardManagerActor);
1785 public void onUnknownResponse(Object response) {
1786 String msg = String.format("Failed to find leader for shard %s: received response: %s",
1787 shardName, response);
1788 LOG.debug ("{}: {}", persistenceId, msg);
1789 targetActor.tell(new akka.actor.Status.Failure(response instanceof Throwable ? (Throwable) response :
1790 new RuntimeException(msg)), shardManagerActor);
1796 * The PrimaryShardFoundForContext is a DTO which puts together a message (aka 'Context' message) which needs to be
1797 * forwarded to the primary replica of a shard and the message (aka 'PrimaryShardFound' message) that is received
1798 * as a successful response to find primary.
1800 private static class PrimaryShardFoundForContext {
1801 private final String shardName;
1802 private final Object contextMessage;
1803 private final RemotePrimaryShardFound remotePrimaryShardFound;
1804 private final LocalPrimaryShardFound localPrimaryShardFound;
1806 public PrimaryShardFoundForContext(@Nonnull String shardName, @Nonnull Object contextMessage,
1807 @Nonnull Object primaryFoundMessage) {
1808 this.shardName = Preconditions.checkNotNull(shardName);
1809 this.contextMessage = Preconditions.checkNotNull(contextMessage);
1810 Preconditions.checkNotNull(primaryFoundMessage);
1811 this.remotePrimaryShardFound = (primaryFoundMessage instanceof RemotePrimaryShardFound) ?
1812 (RemotePrimaryShardFound) primaryFoundMessage : null;
1813 this.localPrimaryShardFound = (primaryFoundMessage instanceof LocalPrimaryShardFound) ?
1814 (LocalPrimaryShardFound) primaryFoundMessage : null;
1818 String getPrimaryPath(){
1819 if(remotePrimaryShardFound != null) {
1820 return remotePrimaryShardFound.getPrimaryPath();
1822 return localPrimaryShardFound.getPrimaryPath();
1826 Object getContextMessage() {
1827 return contextMessage;
1831 RemotePrimaryShardFound getRemotePrimaryShardFound() {
1832 return remotePrimaryShardFound;
1836 String getShardName() {
1842 * The WrappedShardResponse class wraps a response from a Shard.
1844 private static class WrappedShardResponse {
1845 private final ShardIdentifier shardId;
1846 private final Object response;
1847 private final String leaderPath;
1849 private WrappedShardResponse(ShardIdentifier shardId, Object response, String leaderPath) {
1850 this.shardId = shardId;
1851 this.response = response;
1852 this.leaderPath = leaderPath;
1855 ShardIdentifier getShardId() {
1859 Object getResponse() {
1863 String getLeaderPath() {