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.shardmanager;
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.Status;
18 import akka.actor.SupervisorStrategy;
19 import akka.actor.SupervisorStrategy.Directive;
20 import akka.cluster.ClusterEvent;
21 import akka.cluster.Member;
22 import akka.dispatch.Futures;
23 import akka.dispatch.OnComplete;
24 import akka.japi.Function;
25 import akka.pattern.Patterns;
26 import akka.persistence.RecoveryCompleted;
27 import akka.persistence.SaveSnapshotFailure;
28 import akka.persistence.SaveSnapshotSuccess;
29 import akka.persistence.SnapshotOffer;
30 import akka.persistence.SnapshotSelectionCriteria;
31 import akka.util.Timeout;
32 import com.google.common.annotations.VisibleForTesting;
33 import com.google.common.base.Preconditions;
34 import java.io.ByteArrayInputStream;
35 import java.io.ObjectInputStream;
36 import java.util.ArrayList;
37 import java.util.Collection;
38 import java.util.Collections;
39 import java.util.HashMap;
40 import java.util.HashSet;
41 import java.util.List;
44 import java.util.concurrent.CountDownLatch;
45 import java.util.concurrent.TimeUnit;
46 import java.util.concurrent.TimeoutException;
47 import java.util.function.Supplier;
48 import javax.annotation.Nonnull;
49 import javax.annotation.Nullable;
50 import org.apache.commons.lang3.SerializationUtils;
51 import org.opendaylight.controller.cluster.access.concepts.MemberName;
52 import org.opendaylight.controller.cluster.common.actor.AbstractUntypedPersistentActorWithMetering;
53 import org.opendaylight.controller.cluster.datastore.ClusterWrapper;
54 import org.opendaylight.controller.cluster.datastore.DatastoreContext;
55 import org.opendaylight.controller.cluster.datastore.DatastoreContextFactory;
56 import org.opendaylight.controller.cluster.datastore.Shard;
57 import org.opendaylight.controller.cluster.datastore.config.Configuration;
58 import org.opendaylight.controller.cluster.datastore.config.ModuleShardConfiguration;
59 import org.opendaylight.controller.cluster.datastore.exceptions.AlreadyExistsException;
60 import org.opendaylight.controller.cluster.datastore.exceptions.NoShardLeaderException;
61 import org.opendaylight.controller.cluster.datastore.exceptions.NotInitializedException;
62 import org.opendaylight.controller.cluster.datastore.exceptions.PrimaryNotFoundException;
63 import org.opendaylight.controller.cluster.datastore.identifiers.ShardIdentifier;
64 import org.opendaylight.controller.cluster.datastore.messages.ActorInitialized;
65 import org.opendaylight.controller.cluster.datastore.messages.AddShardReplica;
66 import org.opendaylight.controller.cluster.datastore.messages.CreateShard;
67 import org.opendaylight.controller.cluster.datastore.messages.DatastoreSnapshot;
68 import org.opendaylight.controller.cluster.datastore.messages.FindLocalShard;
69 import org.opendaylight.controller.cluster.datastore.messages.FindPrimary;
70 import org.opendaylight.controller.cluster.datastore.messages.LocalPrimaryShardFound;
71 import org.opendaylight.controller.cluster.datastore.messages.LocalShardFound;
72 import org.opendaylight.controller.cluster.datastore.messages.LocalShardNotFound;
73 import org.opendaylight.controller.cluster.datastore.messages.RemoteFindPrimary;
74 import org.opendaylight.controller.cluster.datastore.messages.RemotePrimaryShardFound;
75 import org.opendaylight.controller.cluster.datastore.messages.RemoveShardReplica;
76 import org.opendaylight.controller.cluster.datastore.messages.ShardLeaderStateChanged;
77 import org.opendaylight.controller.cluster.datastore.messages.UpdateSchemaContext;
78 import org.opendaylight.controller.cluster.datastore.utils.Dispatchers;
79 import org.opendaylight.controller.cluster.datastore.utils.PrimaryShardInfoFutureCache;
80 import org.opendaylight.controller.cluster.notifications.RegisterRoleChangeListener;
81 import org.opendaylight.controller.cluster.notifications.RoleChangeNotification;
82 import org.opendaylight.controller.cluster.raft.base.messages.FollowerInitialSyncUpStatus;
83 import org.opendaylight.controller.cluster.raft.base.messages.SwitchBehavior;
84 import org.opendaylight.controller.cluster.raft.client.messages.GetSnapshot;
85 import org.opendaylight.controller.cluster.raft.client.messages.Shutdown;
86 import org.opendaylight.controller.cluster.raft.messages.AddServer;
87 import org.opendaylight.controller.cluster.raft.messages.AddServerReply;
88 import org.opendaylight.controller.cluster.raft.messages.RemoveServer;
89 import org.opendaylight.controller.cluster.raft.messages.RemoveServerReply;
90 import org.opendaylight.controller.cluster.raft.messages.ServerChangeStatus;
91 import org.opendaylight.controller.cluster.raft.messages.ServerRemoved;
92 import org.opendaylight.controller.cluster.raft.policy.DisableElectionsRaftPolicy;
93 import org.opendaylight.yangtools.yang.model.api.SchemaContext;
94 import org.slf4j.Logger;
95 import org.slf4j.LoggerFactory;
96 import scala.concurrent.ExecutionContext;
97 import scala.concurrent.Future;
98 import scala.concurrent.duration.Duration;
99 import scala.concurrent.duration.FiniteDuration;
102 * The ShardManager has the following jobs,
104 * <li> Create all the local shard replicas that belong on this cluster member
105 * <li> Find the address of the local shard
106 * <li> Find the primary replica for any given shard
107 * <li> Monitor the cluster members and store their addresses
110 class ShardManager extends AbstractUntypedPersistentActorWithMetering {
112 private static final Logger LOG = LoggerFactory.getLogger(ShardManager.class);
114 // Stores a mapping between a shard name and it's corresponding information
115 // Shard names look like inventory, topology etc and are as specified in
117 private final Map<String, ShardInformation> localShards = new HashMap<>();
119 // The type of a ShardManager reflects the type of the datastore itself
120 // A data store could be of type config/operational
121 private final String type;
123 private final ClusterWrapper cluster;
125 private final Configuration configuration;
127 private final String shardDispatcherPath;
129 private final ShardManagerInfo mBean;
131 private DatastoreContextFactory datastoreContextFactory;
133 private final CountDownLatch waitTillReadyCountdownLatch;
135 private final PrimaryShardInfoFutureCache primaryShardInfoCache;
137 private final ShardPeerAddressResolver peerAddressResolver;
139 private SchemaContext schemaContext;
141 private DatastoreSnapshot restoreFromSnapshot;
143 private ShardManagerSnapshot currentSnapshot;
145 private final Set<String> shardReplicaOperationsInProgress = new HashSet<>();
147 private final String persistenceId;
149 ShardManager(AbstractShardManagerCreator<?> builder) {
150 this.cluster = builder.getCluster();
151 this.configuration = builder.getConfiguration();
152 this.datastoreContextFactory = builder.getDatastoreContextFactory();
153 this.type = datastoreContextFactory.getBaseDatastoreContext().getDataStoreName();
154 this.shardDispatcherPath =
155 new Dispatchers(context().system().dispatchers()).getDispatcherPath(Dispatchers.DispatcherType.Shard);
156 this.waitTillReadyCountdownLatch = builder.getWaitTillReadyCountdownLatch();
157 this.primaryShardInfoCache = builder.getPrimaryShardInfoCache();
158 this.restoreFromSnapshot = builder.getRestoreFromSnapshot();
160 String possiblePersistenceId = datastoreContextFactory.getBaseDatastoreContext().getShardManagerPersistenceId();
161 persistenceId = possiblePersistenceId != null ? possiblePersistenceId : "shard-manager-" + type;
163 peerAddressResolver = new ShardPeerAddressResolver(type, cluster.getCurrentMemberName());
165 // Subscribe this actor to cluster member events
166 cluster.subscribeToMemberEvents(getSelf());
168 mBean = new ShardManagerInfo(getSelf(), cluster.getCurrentMemberName(), "shard-manager-" + this.type,
169 datastoreContextFactory.getBaseDatastoreContext().getDataStoreMXBeanType());
170 mBean.registerMBean();
174 public void postStop() {
175 LOG.info("Stopping ShardManager {}", persistenceId());
177 mBean.unregisterMBean();
181 public void handleCommand(Object message) throws Exception {
182 if (message instanceof FindPrimary) {
183 findPrimary((FindPrimary)message);
184 } else if(message instanceof FindLocalShard){
185 findLocalShard((FindLocalShard) message);
186 } else if (message instanceof UpdateSchemaContext) {
187 updateSchemaContext(message);
188 } else if(message instanceof ActorInitialized) {
189 onActorInitialized(message);
190 } else if (message instanceof ClusterEvent.MemberUp){
191 memberUp((ClusterEvent.MemberUp) message);
192 } else if (message instanceof ClusterEvent.MemberExited){
193 memberExited((ClusterEvent.MemberExited) message);
194 } else if(message instanceof ClusterEvent.MemberRemoved) {
195 memberRemoved((ClusterEvent.MemberRemoved) message);
196 } else if(message instanceof ClusterEvent.UnreachableMember) {
197 memberUnreachable((ClusterEvent.UnreachableMember)message);
198 } else if(message instanceof ClusterEvent.ReachableMember) {
199 memberReachable((ClusterEvent.ReachableMember) message);
200 } else if(message instanceof DatastoreContextFactory) {
201 onDatastoreContextFactory((DatastoreContextFactory)message);
202 } else if(message instanceof RoleChangeNotification) {
203 onRoleChangeNotification((RoleChangeNotification) message);
204 } else if(message instanceof FollowerInitialSyncUpStatus){
205 onFollowerInitialSyncStatus((FollowerInitialSyncUpStatus) message);
206 } else if(message instanceof ShardNotInitializedTimeout) {
207 onShardNotInitializedTimeout((ShardNotInitializedTimeout)message);
208 } else if(message instanceof ShardLeaderStateChanged) {
209 onLeaderStateChanged((ShardLeaderStateChanged) message);
210 } else if(message instanceof SwitchShardBehavior){
211 onSwitchShardBehavior((SwitchShardBehavior) message);
212 } else if(message instanceof CreateShard) {
213 onCreateShard((CreateShard)message);
214 } else if(message instanceof AddShardReplica){
215 onAddShardReplica((AddShardReplica)message);
216 } else if(message instanceof ForwardedAddServerReply) {
217 ForwardedAddServerReply msg = (ForwardedAddServerReply)message;
218 onAddServerReply(msg.shardInfo, msg.addServerReply, getSender(), msg.leaderPath,
219 msg.removeShardOnFailure);
220 } else if(message instanceof ForwardedAddServerFailure) {
221 ForwardedAddServerFailure msg = (ForwardedAddServerFailure)message;
222 onAddServerFailure(msg.shardName, msg.failureMessage, msg.failure, getSender(), msg.removeShardOnFailure);
223 } else if(message instanceof PrimaryShardFoundForContext) {
224 PrimaryShardFoundForContext primaryShardFoundContext = (PrimaryShardFoundForContext)message;
225 onPrimaryShardFoundContext(primaryShardFoundContext);
226 } else if(message instanceof RemoveShardReplica) {
227 onRemoveShardReplica((RemoveShardReplica) message);
228 } else if(message instanceof WrappedShardResponse){
229 onWrappedShardResponse((WrappedShardResponse) message);
230 } else if(message instanceof GetSnapshot) {
232 } else if(message instanceof ServerRemoved){
233 onShardReplicaRemoved((ServerRemoved) message);
234 } else if(message instanceof SaveSnapshotSuccess) {
235 onSaveSnapshotSuccess((SaveSnapshotSuccess)message);
236 } else if(message instanceof SaveSnapshotFailure) {
237 LOG.error("{}: SaveSnapshotFailure received for saving snapshot of shards",
238 persistenceId(), ((SaveSnapshotFailure) message).cause());
239 } else if(message instanceof Shutdown) {
241 } else if (message instanceof GetLocalShardIds) {
242 onGetLocalShardIds();
244 unknownMessage(message);
248 private void onShutDown() {
249 List<Future<Boolean>> stopFutures = new ArrayList<>(localShards.size());
250 for (ShardInformation info : localShards.values()) {
251 if (info.getActor() != null) {
252 LOG.debug("{}: Issuing gracefulStop to shard {}", persistenceId(), info.getShardId());
254 FiniteDuration duration = info.getDatastoreContext().getShardRaftConfig().getElectionTimeOutInterval().$times(2);
255 stopFutures.add(Patterns.gracefulStop(info.getActor(), duration, Shutdown.INSTANCE));
259 LOG.info("Shutting down ShardManager {} - waiting on {} shards", persistenceId(), stopFutures.size());
261 ExecutionContext dispatcher = new Dispatchers(context().system().dispatchers()).getDispatcher(Dispatchers.DispatcherType.Client);
262 Future<Iterable<Boolean>> combinedFutures = Futures.sequence(stopFutures, dispatcher);
264 combinedFutures.onComplete(new OnComplete<Iterable<Boolean>>() {
266 public void onComplete(Throwable failure, Iterable<Boolean> results) {
267 LOG.debug("{}: All shards shutdown - sending PoisonPill to self", persistenceId());
269 self().tell(PoisonPill.getInstance(), self());
271 if(failure != null) {
272 LOG.warn("{}: An error occurred attempting to shut down the shards", persistenceId(), failure);
275 for(Boolean r: results) {
282 LOG.warn("{}: {} shards did not shut down gracefully", persistenceId(), nfailed);
289 private void onWrappedShardResponse(WrappedShardResponse message) {
290 if (message.getResponse() instanceof RemoveServerReply) {
291 onRemoveServerReply(getSender(), message.getShardId(), (RemoveServerReply) message.getResponse(),
292 message.getLeaderPath());
296 private void onRemoveServerReply(ActorRef originalSender, ShardIdentifier shardId, RemoveServerReply replyMsg,
298 shardReplicaOperationsInProgress.remove(shardId);
300 LOG.debug ("{}: Received {} for shard {}", persistenceId(), replyMsg, shardId.getShardName());
302 if (replyMsg.getStatus() == ServerChangeStatus.OK) {
303 LOG.debug ("{}: Leader shard successfully removed the replica shard {}", persistenceId(),
304 shardId.getShardName());
305 originalSender.tell(new Status.Success(null), getSelf());
307 LOG.warn ("{}: Leader failed to remove shard replica {} with status {}",
308 persistenceId(), shardId, replyMsg.getStatus());
310 Exception failure = getServerChangeException(RemoveServer.class, replyMsg.getStatus(), leaderPath, shardId);
311 originalSender.tell(new Status.Failure(failure), getSelf());
315 private void onPrimaryShardFoundContext(PrimaryShardFoundForContext primaryShardFoundContext) {
316 if(primaryShardFoundContext.getContextMessage() instanceof AddShardReplica) {
317 addShard(primaryShardFoundContext.getShardName(), primaryShardFoundContext.getRemotePrimaryShardFound(),
319 } else if(primaryShardFoundContext.getContextMessage() instanceof RemoveShardReplica){
320 removeShardReplica((RemoveShardReplica) primaryShardFoundContext.getContextMessage(),
321 primaryShardFoundContext.getShardName(), primaryShardFoundContext.getPrimaryPath(), getSender());
325 private void removeShardReplica(RemoveShardReplica contextMessage, final String shardName, final String primaryPath,
326 final ActorRef sender) {
327 if(isShardReplicaOperationInProgress(shardName, sender)) {
331 shardReplicaOperationsInProgress.add(shardName);
333 final ShardIdentifier shardId = getShardIdentifier(contextMessage.getMemberName(), shardName);
335 final DatastoreContext datastoreContext = newShardDatastoreContextBuilder(shardName).build();
337 //inform ShardLeader to remove this shard as a replica by sending an RemoveServer message
338 LOG.debug ("{}: Sending RemoveServer message to peer {} for shard {}", persistenceId(),
339 primaryPath, shardId);
341 Timeout removeServerTimeout = new Timeout(datastoreContext.getShardLeaderElectionTimeout().
343 Future<Object> futureObj = ask(getContext().actorSelection(primaryPath),
344 new RemoveServer(shardId.toString()), removeServerTimeout);
346 futureObj.onComplete(new OnComplete<Object>() {
348 public void onComplete(Throwable failure, Object response) {
349 if (failure != null) {
350 String msg = String.format("RemoveServer request to leader %s for shard %s failed",
351 primaryPath, shardName);
353 LOG.debug ("{}: {}", persistenceId(), msg, failure);
356 sender.tell(new Status.Failure(new RuntimeException(msg, failure)), self());
359 self().tell(new WrappedShardResponse(shardId, response, primaryPath), sender);
362 }, new Dispatchers(context().system().dispatchers()).getDispatcher(Dispatchers.DispatcherType.Client));
365 private void onShardReplicaRemoved(ServerRemoved message) {
366 final ShardIdentifier shardId = new ShardIdentifier.Builder().fromShardIdString(message.getServerId()).build();
367 final ShardInformation shardInformation = localShards.remove(shardId.getShardName());
368 if(shardInformation == null) {
369 LOG.debug("{} : Shard replica {} is not present in list", persistenceId(), shardId.toString());
371 } else if(shardInformation.getActor() != null) {
372 LOG.debug("{} : Sending Shutdown to Shard actor {}", persistenceId(), shardInformation.getActor());
373 shardInformation.getActor().tell(Shutdown.INSTANCE, self());
375 LOG.debug("{} : Local Shard replica for shard {} has been removed", persistenceId(), shardId.getShardName());
379 private void onGetSnapshot() {
380 LOG.debug("{}: onGetSnapshot", persistenceId());
382 List<String> notInitialized = null;
383 for(ShardInformation shardInfo: localShards.values()) {
384 if(!shardInfo.isShardInitialized()) {
385 if(notInitialized == null) {
386 notInitialized = new ArrayList<>();
389 notInitialized.add(shardInfo.getShardName());
393 if(notInitialized != null) {
394 getSender().tell(new Status.Failure(new IllegalStateException(String.format(
395 "%d shard(s) %s are not initialized", notInitialized.size(), notInitialized))), getSelf());
399 byte[] shardManagerSnapshot = null;
400 if(currentSnapshot != null) {
401 shardManagerSnapshot = SerializationUtils.serialize(currentSnapshot);
404 ActorRef replyActor = getContext().actorOf(ShardManagerGetSnapshotReplyActor.props(
405 new ArrayList<>(localShards.keySet()), type, shardManagerSnapshot , getSender(), persistenceId(),
406 datastoreContextFactory.getBaseDatastoreContext().getShardInitializationTimeout().duration()));
408 for(ShardInformation shardInfo: localShards.values()) {
409 shardInfo.getActor().tell(GetSnapshot.INSTANCE, replyActor);
413 private void onCreateShard(CreateShard createShard) {
414 LOG.debug("{}: onCreateShard: {}", persistenceId(), createShard);
418 String shardName = createShard.getModuleShardConfig().getShardName();
419 if(localShards.containsKey(shardName)) {
420 LOG.debug("{}: Shard {} already exists", persistenceId(), shardName);
421 reply = new Status.Success(String.format("Shard with name %s already exists", shardName));
423 doCreateShard(createShard);
424 reply = new Status.Success(null);
426 } catch (Exception e) {
427 LOG.error("{}: onCreateShard failed", persistenceId(), e);
428 reply = new Status.Failure(e);
431 if(getSender() != null && !getContext().system().deadLetters().equals(getSender())) {
432 getSender().tell(reply, getSelf());
436 private void doCreateShard(CreateShard createShard) {
437 ModuleShardConfiguration moduleShardConfig = createShard.getModuleShardConfig();
438 String shardName = moduleShardConfig.getShardName();
440 configuration.addModuleShardConfiguration(moduleShardConfig);
442 DatastoreContext shardDatastoreContext = createShard.getDatastoreContext();
443 if(shardDatastoreContext == null) {
444 shardDatastoreContext = newShardDatastoreContext(shardName);
446 shardDatastoreContext = DatastoreContext.newBuilderFrom(shardDatastoreContext).shardPeerAddressResolver(
447 peerAddressResolver).build();
450 ShardIdentifier shardId = getShardIdentifier(cluster.getCurrentMemberName(), shardName);
452 boolean shardWasInRecoveredSnapshot = currentSnapshot != null &&
453 currentSnapshot.getShardList().contains(shardName);
455 Map<String, String> peerAddresses;
456 boolean isActiveMember;
457 if(shardWasInRecoveredSnapshot || configuration.getMembersFromShardName(shardName).
458 contains(cluster.getCurrentMemberName())) {
459 peerAddresses = getPeerAddresses(shardName);
460 isActiveMember = true;
462 // The local member is not in the static shard member configuration and the shard did not
463 // previously exist (ie !shardWasInRecoveredSnapshot). In this case we'll create
464 // the shard with no peers and with elections disabled so it stays as follower. A
465 // subsequent AddServer request will be needed to make it an active member.
466 isActiveMember = false;
467 peerAddresses = Collections.emptyMap();
468 shardDatastoreContext = DatastoreContext.newBuilderFrom(shardDatastoreContext).
469 customRaftPolicyImplementation(DisableElectionsRaftPolicy.class.getName()).build();
472 LOG.debug("{} doCreateShard: shardId: {}, memberNames: {}, peerAddresses: {}, isActiveMember: {}",
473 persistenceId(), shardId, moduleShardConfig.getShardMemberNames(), peerAddresses,
476 ShardInformation info = new ShardInformation(shardName, shardId, peerAddresses,
477 shardDatastoreContext, createShard.getShardBuilder(), peerAddressResolver);
478 info.setActiveMember(isActiveMember);
479 localShards.put(info.getShardName(), info);
481 if(schemaContext != null) {
482 info.setActor(newShardActor(schemaContext, info));
486 private DatastoreContext.Builder newShardDatastoreContextBuilder(String shardName) {
487 return DatastoreContext.newBuilderFrom(datastoreContextFactory.getShardDatastoreContext(shardName)).
488 shardPeerAddressResolver(peerAddressResolver);
491 private DatastoreContext newShardDatastoreContext(String shardName) {
492 return newShardDatastoreContextBuilder(shardName).build();
495 private void checkReady(){
496 if (isReadyWithLeaderId()) {
497 LOG.info("{}: All Shards are ready - data store {} is ready, available count is {}",
498 persistenceId(), type, waitTillReadyCountdownLatch.getCount());
500 waitTillReadyCountdownLatch.countDown();
504 private void onLeaderStateChanged(ShardLeaderStateChanged leaderStateChanged) {
505 LOG.info("{}: Received LeaderStateChanged message: {}", persistenceId(), leaderStateChanged);
507 ShardInformation shardInformation = findShardInformation(leaderStateChanged.getMemberId());
508 if(shardInformation != null) {
509 shardInformation.setLocalDataTree(leaderStateChanged.getLocalShardDataTree());
510 shardInformation.setLeaderVersion(leaderStateChanged.getLeaderPayloadVersion());
511 if(shardInformation.setLeaderId(leaderStateChanged.getLeaderId())) {
512 primaryShardInfoCache.remove(shardInformation.getShardName());
517 LOG.debug("No shard found with member Id {}", leaderStateChanged.getMemberId());
521 private void onShardNotInitializedTimeout(ShardNotInitializedTimeout message) {
522 ShardInformation shardInfo = message.getShardInfo();
524 LOG.debug("{}: Received ShardNotInitializedTimeout message for shard {}", persistenceId(),
525 shardInfo.getShardName());
527 shardInfo.removeOnShardInitialized(message.getOnShardInitialized());
529 if(!shardInfo.isShardInitialized()) {
530 LOG.debug("{}: Returning NotInitializedException for shard {}", persistenceId(), shardInfo.getShardName());
531 message.getSender().tell(createNotInitializedException(shardInfo.getShardId()), getSelf());
533 LOG.debug("{}: Returning NoShardLeaderException for shard {}", persistenceId(), shardInfo.getShardName());
534 message.getSender().tell(createNoShardLeaderException(shardInfo.getShardId()), getSelf());
538 private void onFollowerInitialSyncStatus(FollowerInitialSyncUpStatus status) {
539 LOG.info("{} Received follower initial sync status for {} status sync done {}", persistenceId(),
540 status.getName(), status.isInitialSyncDone());
542 ShardInformation shardInformation = findShardInformation(status.getName());
544 if(shardInformation != null) {
545 shardInformation.setFollowerSyncStatus(status.isInitialSyncDone());
547 mBean.setSyncStatus(isInSync());
552 private void onRoleChangeNotification(RoleChangeNotification roleChanged) {
553 LOG.info("{}: Received role changed for {} from {} to {}", persistenceId(), roleChanged.getMemberId(),
554 roleChanged.getOldRole(), roleChanged.getNewRole());
556 ShardInformation shardInformation = findShardInformation(roleChanged.getMemberId());
557 if(shardInformation != null) {
558 shardInformation.setRole(roleChanged.getNewRole());
560 mBean.setSyncStatus(isInSync());
565 private ShardInformation findShardInformation(String memberId) {
566 for(ShardInformation info : localShards.values()){
567 if(info.getShardId().toString().equals(memberId)){
575 private boolean isReadyWithLeaderId() {
576 boolean isReady = true;
577 for (ShardInformation info : localShards.values()) {
578 if(!info.isShardReadyWithLeaderId()){
586 private boolean isInSync(){
587 for (ShardInformation info : localShards.values()) {
588 if(!info.isInSync()){
595 private void onActorInitialized(Object message) {
596 final ActorRef sender = getSender();
598 if (sender == null) {
599 return; //why is a non-actor sending this message? Just ignore.
602 String actorName = sender.path().name();
603 //find shard name from actor name; actor name is stringified shardId
604 ShardIdentifier shardId = ShardIdentifier.builder().fromShardIdString(actorName).build();
606 if (shardId.getShardName() == null) {
610 markShardAsInitialized(shardId.getShardName());
613 private void markShardAsInitialized(String shardName) {
614 LOG.debug("{}: Initializing shard [{}]", persistenceId(), shardName);
616 ShardInformation shardInformation = localShards.get(shardName);
617 if (shardInformation != null) {
618 shardInformation.setActorInitialized();
620 shardInformation.getActor().tell(new RegisterRoleChangeListener(), self());
625 protected void handleRecover(Object message) throws Exception {
626 if (message instanceof RecoveryCompleted) {
627 onRecoveryCompleted();
628 } else if (message instanceof SnapshotOffer) {
629 applyShardManagerSnapshot((ShardManagerSnapshot)((SnapshotOffer) message).snapshot());
633 private void onRecoveryCompleted() {
634 LOG.info("Recovery complete : {}", persistenceId());
636 // We no longer persist SchemaContext modules so delete all the prior messages from the akka
637 // journal on upgrade from Helium.
638 deleteMessages(lastSequenceNr());
640 if(currentSnapshot == null && restoreFromSnapshot != null &&
641 restoreFromSnapshot.getShardManagerSnapshot() != null) {
642 try(ObjectInputStream ois = new ObjectInputStream(new ByteArrayInputStream(
643 restoreFromSnapshot.getShardManagerSnapshot()))) {
644 ShardManagerSnapshot snapshot = (ShardManagerSnapshot) ois.readObject();
646 LOG.debug("{}: Deserialized restored ShardManagerSnapshot: {}", persistenceId(), snapshot);
648 applyShardManagerSnapshot(snapshot);
649 } catch(Exception e) {
650 LOG.error("{}: Error deserializing restored ShardManagerSnapshot", persistenceId(), e);
657 private void findLocalShard(FindLocalShard message) {
658 final ShardInformation shardInformation = localShards.get(message.getShardName());
660 if(shardInformation == null){
661 getSender().tell(new LocalShardNotFound(message.getShardName()), getSelf());
665 sendResponse(shardInformation, message.isWaitUntilInitialized(), false, () -> new LocalShardFound(shardInformation.getActor()));
668 private void sendResponse(ShardInformation shardInformation, boolean doWait,
669 boolean wantShardReady, final Supplier<Object> messageSupplier) {
670 if (!shardInformation.isShardInitialized() || (wantShardReady && !shardInformation.isShardReadyWithLeaderId())) {
672 final ActorRef sender = getSender();
673 final ActorRef self = self();
675 Runnable replyRunnable = () -> sender.tell(messageSupplier.get(), self);
677 OnShardInitialized onShardInitialized = wantShardReady ? new OnShardReady(replyRunnable) :
678 new OnShardInitialized(replyRunnable);
680 shardInformation.addOnShardInitialized(onShardInitialized);
682 FiniteDuration timeout = shardInformation.getDatastoreContext().getShardInitializationTimeout().duration();
683 if(shardInformation.isShardInitialized()) {
684 // If the shard is already initialized then we'll wait enough time for the shard to
685 // elect a leader, ie 2 times the election timeout.
686 timeout = FiniteDuration.create(shardInformation.getDatastoreContext().getShardRaftConfig()
687 .getElectionTimeOutInterval().toMillis() * 2, TimeUnit.MILLISECONDS);
690 LOG.debug("{}: Scheduling {} ms timer to wait for shard {}", persistenceId(), timeout.toMillis(),
691 shardInformation.getShardName());
693 Cancellable timeoutSchedule = getContext().system().scheduler().scheduleOnce(
695 new ShardNotInitializedTimeout(shardInformation, onShardInitialized, sender),
696 getContext().dispatcher(), getSelf());
698 onShardInitialized.setTimeoutSchedule(timeoutSchedule);
700 } else if (!shardInformation.isShardInitialized()) {
701 LOG.debug("{}: Returning NotInitializedException for shard {}", persistenceId(),
702 shardInformation.getShardName());
703 getSender().tell(createNotInitializedException(shardInformation.getShardId()), getSelf());
705 LOG.debug("{}: Returning NoShardLeaderException for shard {}", persistenceId(),
706 shardInformation.getShardName());
707 getSender().tell(createNoShardLeaderException(shardInformation.getShardId()), getSelf());
713 getSender().tell(messageSupplier.get(), getSelf());
716 private static NoShardLeaderException createNoShardLeaderException(ShardIdentifier shardId) {
717 return new NoShardLeaderException(null, shardId.toString());
720 private static NotInitializedException createNotInitializedException(ShardIdentifier shardId) {
721 return new NotInitializedException(String.format(
722 "Found primary shard %s but it's not initialized yet. Please try again later", shardId));
726 static MemberName memberToName(final Member member) {
727 return MemberName.forName(member.roles().iterator().next());
730 private void memberRemoved(ClusterEvent.MemberRemoved message) {
731 MemberName memberName = memberToName(message.member());
733 LOG.debug("{}: Received MemberRemoved: memberName: {}, address: {}", persistenceId(), memberName,
734 message.member().address());
736 peerAddressResolver.removePeerAddress(memberName);
738 for(ShardInformation info : localShards.values()){
739 info.peerDown(memberName, getShardIdentifier(memberName, info.getShardName()).toString(), getSelf());
743 private void memberExited(ClusterEvent.MemberExited message) {
744 MemberName memberName = memberToName(message.member());
746 LOG.debug("{}: Received MemberExited: memberName: {}, address: {}", persistenceId(), memberName,
747 message.member().address());
749 peerAddressResolver.removePeerAddress(memberName);
751 for(ShardInformation info : localShards.values()){
752 info.peerDown(memberName, getShardIdentifier(memberName, info.getShardName()).toString(), getSelf());
756 private void memberUp(ClusterEvent.MemberUp message) {
757 MemberName memberName = memberToName(message.member());
759 LOG.debug("{}: Received MemberUp: memberName: {}, address: {}", persistenceId(), memberName,
760 message.member().address());
762 addPeerAddress(memberName, message.member().address());
767 private void addPeerAddress(MemberName memberName, Address address) {
768 peerAddressResolver.addPeerAddress(memberName, address);
770 for(ShardInformation info : localShards.values()){
771 String shardName = info.getShardName();
772 String peerId = getShardIdentifier(memberName, shardName).toString();
773 info.updatePeerAddress(peerId, peerAddressResolver.getShardActorAddress(shardName, memberName), getSelf());
775 info.peerUp(memberName, peerId, getSelf());
779 private void memberReachable(ClusterEvent.ReachableMember message) {
780 MemberName memberName = memberToName(message.member());
781 LOG.debug("Received ReachableMember: memberName {}, address: {}", memberName, message.member().address());
783 addPeerAddress(memberName, message.member().address());
785 markMemberAvailable(memberName);
788 private void memberUnreachable(ClusterEvent.UnreachableMember message) {
789 MemberName memberName = memberToName(message.member());
790 LOG.debug("Received UnreachableMember: memberName {}, address: {}", memberName, message.member().address());
792 markMemberUnavailable(memberName);
795 private void markMemberUnavailable(final MemberName memberName) {
796 final String memberStr = memberName.getName();
797 for (ShardInformation info : localShards.values()) {
798 String leaderId = info.getLeaderId();
799 // XXX: why are we using String#contains() here?
800 if (leaderId != null && leaderId.contains(memberStr)) {
801 LOG.debug("Marking Leader {} as unavailable.", leaderId);
802 info.setLeaderAvailable(false);
804 primaryShardInfoCache.remove(info.getShardName());
807 info.peerDown(memberName, getShardIdentifier(memberName, info.getShardName()).toString(), getSelf());
811 private void markMemberAvailable(final MemberName memberName) {
812 final String memberStr = memberName.getName();
813 for (ShardInformation info : localShards.values()) {
814 String leaderId = info.getLeaderId();
815 // XXX: why are we using String#contains() here?
816 if (leaderId != null && leaderId.contains(memberStr)) {
817 LOG.debug("Marking Leader {} as available.", leaderId);
818 info.setLeaderAvailable(true);
821 info.peerUp(memberName, getShardIdentifier(memberName, info.getShardName()).toString(), getSelf());
825 private void onDatastoreContextFactory(DatastoreContextFactory factory) {
826 datastoreContextFactory = factory;
827 for (ShardInformation info : localShards.values()) {
828 info.setDatastoreContext(newShardDatastoreContext(info.getShardName()), getSelf());
832 private void onGetLocalShardIds() {
833 final List<String> response = new ArrayList<>(localShards.size());
835 for (ShardInformation info : localShards.values()) {
836 response.add(info.getShardId().toString());
839 getSender().tell(new Status.Success(response), getSelf());
842 private void onSwitchShardBehavior(final SwitchShardBehavior message) {
843 final ShardIdentifier identifier = message.getShardId();
845 if (identifier != null) {
846 final ShardInformation info = localShards.get(identifier.getShardName());
848 getSender().tell(new Status.Failure(
849 new IllegalArgumentException("Shard " + identifier + " is not local")), getSelf());
853 switchShardBehavior(info, new SwitchBehavior(message.getNewState(), message.getTerm()));
855 for (ShardInformation info : localShards.values()) {
856 switchShardBehavior(info, new SwitchBehavior(message.getNewState(), message.getTerm()));
860 getSender().tell(new Status.Success(null), getSelf());
863 private void switchShardBehavior(final ShardInformation info, final SwitchBehavior switchBehavior) {
864 final ActorRef actor = info.getActor();
866 actor.tell(switchBehavior, getSelf());
868 LOG.warn("Could not switch the behavior of shard {} to {} - shard is not yet available",
869 info.getShardName(), switchBehavior.getNewState());
874 * Notifies all the local shards of a change in the schema context
878 private void updateSchemaContext(final Object message) {
879 schemaContext = ((UpdateSchemaContext) message).getSchemaContext();
881 LOG.debug("Got updated SchemaContext: # of modules {}", schemaContext.getAllModuleIdentifiers().size());
883 for (ShardInformation info : localShards.values()) {
884 if (info.getActor() == null) {
885 LOG.debug("Creating Shard {}", info.getShardId());
886 info.setActor(newShardActor(schemaContext, info));
888 info.getActor().tell(message, getSelf());
894 protected ClusterWrapper getCluster() {
899 protected ActorRef newShardActor(final SchemaContext schemaContext, ShardInformation info) {
900 return getContext().actorOf(info.newProps(schemaContext)
901 .withDispatcher(shardDispatcherPath), info.getShardId().toString());
904 private void findPrimary(FindPrimary message) {
905 LOG.debug("{}: In findPrimary: {}", persistenceId(), message);
907 final String shardName = message.getShardName();
908 final boolean canReturnLocalShardState = !(message instanceof RemoteFindPrimary);
910 // First see if the there is a local replica for the shard
911 final ShardInformation info = localShards.get(shardName);
912 if (info != null && info.isActiveMember()) {
913 sendResponse(info, message.isWaitUntilReady(), true, () -> {
914 String primaryPath = info.getSerializedLeaderActor();
915 Object found = canReturnLocalShardState && info.isLeader() ?
916 new LocalPrimaryShardFound(primaryPath, info.getLocalShardDataTree().get()) :
917 new RemotePrimaryShardFound(primaryPath, info.getLeaderVersion());
919 if(LOG.isDebugEnabled()) {
920 LOG.debug("{}: Found primary for {}: {}", persistenceId(), shardName, found);
929 final Collection<String> visitedAddresses;
930 if(message instanceof RemoteFindPrimary) {
931 visitedAddresses = ((RemoteFindPrimary)message).getVisitedAddresses();
933 visitedAddresses = new ArrayList<>(1);
936 visitedAddresses.add(peerAddressResolver.getShardManagerActorPathBuilder(cluster.getSelfAddress()).toString());
938 for(String address: peerAddressResolver.getShardManagerPeerActorAddresses()) {
939 if(visitedAddresses.contains(address)) {
943 LOG.debug("{}: findPrimary for {} forwarding to remote ShardManager {}, visitedAddresses: {}",
944 persistenceId(), shardName, address, visitedAddresses);
946 getContext().actorSelection(address).forward(new RemoteFindPrimary(shardName,
947 message.isWaitUntilReady(), visitedAddresses), getContext());
951 LOG.debug("{}: No shard found for {}", persistenceId(), shardName);
953 getSender().tell(new PrimaryNotFoundException(
954 String.format("No primary shard found for %s.", shardName)), getSelf());
958 * Construct the name of the shard actor given the name of the member on
959 * which the shard resides and the name of the shard
965 private ShardIdentifier getShardIdentifier(MemberName memberName, String shardName){
966 return peerAddressResolver.getShardIdentifier(memberName, shardName);
970 * Create shards that are local to the member on which the ShardManager
974 private void createLocalShards() {
975 MemberName memberName = this.cluster.getCurrentMemberName();
976 Collection<String> memberShardNames = this.configuration.getMemberShardNames(memberName);
978 Map<String, DatastoreSnapshot.ShardSnapshot> shardSnapshots = new HashMap<>();
979 if(restoreFromSnapshot != null)
981 for(DatastoreSnapshot.ShardSnapshot snapshot: restoreFromSnapshot.getShardSnapshots()) {
982 shardSnapshots.put(snapshot.getName(), snapshot);
986 restoreFromSnapshot = null; // null out to GC
988 for(String shardName : memberShardNames){
989 ShardIdentifier shardId = getShardIdentifier(memberName, shardName);
991 LOG.debug("{}: Creating local shard: {}", persistenceId(), shardId);
993 Map<String, String> peerAddresses = getPeerAddresses(shardName);
994 localShards.put(shardName, new ShardInformation(shardName, shardId, peerAddresses,
995 newShardDatastoreContext(shardName), Shard.builder().restoreFromSnapshot(
996 shardSnapshots.get(shardName)), peerAddressResolver));
1001 * Given the name of the shard find the addresses of all it's peers
1005 private Map<String, String> getPeerAddresses(String shardName) {
1006 Collection<MemberName> members = configuration.getMembersFromShardName(shardName);
1007 Map<String, String> peerAddresses = new HashMap<>();
1009 MemberName currentMemberName = this.cluster.getCurrentMemberName();
1011 for (MemberName memberName : members) {
1012 if (!currentMemberName.equals(memberName)) {
1013 ShardIdentifier shardId = getShardIdentifier(memberName, shardName);
1014 String address = peerAddressResolver.getShardActorAddress(shardName, memberName);
1015 peerAddresses.put(shardId.toString(), address);
1018 return peerAddresses;
1022 public SupervisorStrategy supervisorStrategy() {
1024 return new OneForOneStrategy(10, Duration.create("1 minute"),
1025 (Function<Throwable, Directive>) t -> {
1026 LOG.warn("Supervisor Strategy caught unexpected exception - resuming", t);
1027 return SupervisorStrategy.resume();
1034 public String persistenceId() {
1035 return persistenceId;
1039 ShardManagerInfoMBean getMBean(){
1043 private boolean isShardReplicaOperationInProgress(final String shardName, final ActorRef sender) {
1044 if (shardReplicaOperationsInProgress.contains(shardName)) {
1045 String msg = String.format("A shard replica operation for %s is already in progress", shardName);
1046 LOG.debug ("{}: {}", persistenceId(), msg);
1047 sender.tell(new Status.Failure(new IllegalStateException(msg)), getSelf());
1054 private void onAddShardReplica (final AddShardReplica shardReplicaMsg) {
1055 final String shardName = shardReplicaMsg.getShardName();
1057 LOG.debug("{}: onAddShardReplica: {}", persistenceId(), shardReplicaMsg);
1059 // verify the shard with the specified name is present in the cluster configuration
1060 if (!(this.configuration.isShardConfigured(shardName))) {
1061 String msg = String.format("No module configuration exists for shard %s", shardName);
1062 LOG.debug ("{}: {}", persistenceId(), msg);
1063 getSender().tell(new Status.Failure(new IllegalArgumentException(msg)), getSelf());
1067 // Create the localShard
1068 if (schemaContext == null) {
1069 String msg = String.format(
1070 "No SchemaContext is available in order to create a local shard instance for %s", shardName);
1071 LOG.debug ("{}: {}", persistenceId(), msg);
1072 getSender().tell(new Status.Failure(new IllegalStateException(msg)), getSelf());
1076 findPrimary(shardName, new AutoFindPrimaryFailureResponseHandler(getSender(), shardName, persistenceId(), getSelf()) {
1078 public void onRemotePrimaryShardFound(RemotePrimaryShardFound response) {
1079 getSelf().tell(new PrimaryShardFoundForContext(getShardName(), shardReplicaMsg, response), getTargetActor());
1083 public void onLocalPrimaryFound(LocalPrimaryShardFound message) {
1084 sendLocalReplicaAlreadyExistsReply(getShardName(), getTargetActor());
1090 private void sendLocalReplicaAlreadyExistsReply(String shardName, ActorRef sender) {
1091 String msg = String.format("Local shard %s already exists", shardName);
1092 LOG.debug ("{}: {}", persistenceId(), msg);
1093 sender.tell(new Status.Failure(new AlreadyExistsException(msg)), getSelf());
1096 private void addShard(final String shardName, final RemotePrimaryShardFound response, final ActorRef sender) {
1097 if(isShardReplicaOperationInProgress(shardName, sender)) {
1101 shardReplicaOperationsInProgress.add(shardName);
1103 final ShardInformation shardInfo;
1104 final boolean removeShardOnFailure;
1105 ShardInformation existingShardInfo = localShards.get(shardName);
1106 if(existingShardInfo == null) {
1107 removeShardOnFailure = true;
1108 ShardIdentifier shardId = getShardIdentifier(cluster.getCurrentMemberName(), shardName);
1110 DatastoreContext datastoreContext = newShardDatastoreContextBuilder(shardName).customRaftPolicyImplementation(
1111 DisableElectionsRaftPolicy.class.getName()).build();
1113 shardInfo = new ShardInformation(shardName, shardId, getPeerAddresses(shardName), datastoreContext,
1114 Shard.builder(), peerAddressResolver);
1115 shardInfo.setActiveMember(false);
1116 localShards.put(shardName, shardInfo);
1117 shardInfo.setActor(newShardActor(schemaContext, shardInfo));
1119 removeShardOnFailure = false;
1120 shardInfo = existingShardInfo;
1123 String localShardAddress = peerAddressResolver.getShardActorAddress(shardName, cluster.getCurrentMemberName());
1125 //inform ShardLeader to add this shard as a replica by sending an AddServer message
1126 LOG.debug ("{}: Sending AddServer message to peer {} for shard {}", persistenceId(),
1127 response.getPrimaryPath(), shardInfo.getShardId());
1129 Timeout addServerTimeout = new Timeout(shardInfo.getDatastoreContext().getShardLeaderElectionTimeout().
1131 Future<Object> futureObj = ask(getContext().actorSelection(response.getPrimaryPath()),
1132 new AddServer(shardInfo.getShardId().toString(), localShardAddress, true), addServerTimeout);
1134 futureObj.onComplete(new OnComplete<Object>() {
1136 public void onComplete(Throwable failure, Object addServerResponse) {
1137 if (failure != null) {
1138 LOG.debug ("{}: AddServer request to {} for {} failed", persistenceId(),
1139 response.getPrimaryPath(), shardName, failure);
1141 String msg = String.format("AddServer request to leader %s for shard %s failed",
1142 response.getPrimaryPath(), shardName);
1143 self().tell(new ForwardedAddServerFailure(shardName, msg, failure, removeShardOnFailure), sender);
1145 self().tell(new ForwardedAddServerReply(shardInfo, (AddServerReply)addServerResponse,
1146 response.getPrimaryPath(), removeShardOnFailure), sender);
1149 }, new Dispatchers(context().system().dispatchers()).getDispatcher(Dispatchers.DispatcherType.Client));
1152 private void onAddServerFailure(String shardName, String message, Throwable failure, ActorRef sender,
1153 boolean removeShardOnFailure) {
1154 shardReplicaOperationsInProgress.remove(shardName);
1156 if(removeShardOnFailure) {
1157 ShardInformation shardInfo = localShards.remove(shardName);
1158 if (shardInfo.getActor() != null) {
1159 shardInfo.getActor().tell(PoisonPill.getInstance(), getSelf());
1163 sender.tell(new Status.Failure(message == null ? failure :
1164 new RuntimeException(message, failure)), getSelf());
1167 private void onAddServerReply(ShardInformation shardInfo, AddServerReply replyMsg, ActorRef sender,
1168 String leaderPath, boolean removeShardOnFailure) {
1169 String shardName = shardInfo.getShardName();
1170 shardReplicaOperationsInProgress.remove(shardName);
1172 LOG.debug ("{}: Received {} for shard {} from leader {}", persistenceId(), replyMsg, shardName, leaderPath);
1174 if (replyMsg.getStatus() == ServerChangeStatus.OK) {
1175 LOG.debug ("{}: Leader shard successfully added the replica shard {}", persistenceId(), shardName);
1177 // Make the local shard voting capable
1178 shardInfo.setDatastoreContext(newShardDatastoreContext(shardName), getSelf());
1179 shardInfo.setActiveMember(true);
1182 sender.tell(new Status.Success(null), getSelf());
1183 } else if(replyMsg.getStatus() == ServerChangeStatus.ALREADY_EXISTS) {
1184 sendLocalReplicaAlreadyExistsReply(shardName, sender);
1186 LOG.warn ("{}: Leader failed to add shard replica {} with status {}",
1187 persistenceId(), shardName, replyMsg.getStatus());
1189 Exception failure = getServerChangeException(AddServer.class, replyMsg.getStatus(), leaderPath, shardInfo.getShardId());
1191 onAddServerFailure(shardName, null, failure, sender, removeShardOnFailure);
1195 private static Exception getServerChangeException(Class<?> serverChange, ServerChangeStatus serverChangeStatus,
1196 String leaderPath, ShardIdentifier shardId) {
1198 switch (serverChangeStatus) {
1200 failure = new TimeoutException(String.format(
1201 "The shard leader %s timed out trying to replicate the initial data to the new shard %s." +
1202 "Possible causes - there was a problem replicating the data or shard leadership changed while replicating the shard data",
1203 leaderPath, shardId.getShardName()));
1206 failure = createNoShardLeaderException(shardId);
1209 failure = new UnsupportedOperationException(String.format("%s request is not supported for shard %s",
1210 serverChange.getSimpleName(), shardId.getShardName()));
1213 failure = new RuntimeException(String.format(
1214 "%s request to leader %s for shard %s failed with status %s",
1215 serverChange.getSimpleName(), leaderPath, shardId.getShardName(), serverChangeStatus));
1220 private void onRemoveShardReplica (final RemoveShardReplica shardReplicaMsg) {
1221 LOG.debug("{}: onRemoveShardReplica: {}", persistenceId(), shardReplicaMsg);
1223 findPrimary(shardReplicaMsg.getShardName(), new AutoFindPrimaryFailureResponseHandler(getSender(),
1224 shardReplicaMsg.getShardName(), persistenceId(), getSelf()) {
1226 public void onRemotePrimaryShardFound(RemotePrimaryShardFound response) {
1227 getSelf().tell(new PrimaryShardFoundForContext(getShardName(), shardReplicaMsg, response), getTargetActor());
1231 public void onLocalPrimaryFound(LocalPrimaryShardFound response) {
1232 getSelf().tell(new PrimaryShardFoundForContext(getShardName(), shardReplicaMsg, response), getTargetActor());
1237 private void persistShardList() {
1238 List<String> shardList = new ArrayList<>(localShards.keySet());
1239 for (ShardInformation shardInfo : localShards.values()) {
1240 if (!shardInfo.isActiveMember()) {
1241 shardList.remove(shardInfo.getShardName());
1244 LOG.debug ("{}: persisting the shard list {}", persistenceId(), shardList);
1245 saveSnapshot(updateShardManagerSnapshot(shardList));
1248 private ShardManagerSnapshot updateShardManagerSnapshot(List<String> shardList) {
1249 currentSnapshot = new ShardManagerSnapshot(shardList);
1250 return currentSnapshot;
1253 private void applyShardManagerSnapshot(ShardManagerSnapshot snapshot) {
1254 currentSnapshot = snapshot;
1256 LOG.debug ("{}: onSnapshotOffer: {}", persistenceId(), currentSnapshot);
1258 final MemberName currentMember = cluster.getCurrentMemberName();
1259 Set<String> configuredShardList =
1260 new HashSet<>(configuration.getMemberShardNames(currentMember));
1261 for (String shard : currentSnapshot.getShardList()) {
1262 if (!configuredShardList.contains(shard)) {
1263 // add the current member as a replica for the shard
1264 LOG.debug ("{}: adding shard {}", persistenceId(), shard);
1265 configuration.addMemberReplicaForShard(shard, currentMember);
1267 configuredShardList.remove(shard);
1270 for (String shard : configuredShardList) {
1271 // remove the member as a replica for the shard
1272 LOG.debug ("{}: removing shard {}", persistenceId(), shard);
1273 configuration.removeMemberReplicaForShard(shard, currentMember);
1277 private void onSaveSnapshotSuccess (SaveSnapshotSuccess successMessage) {
1278 LOG.debug ("{} saved ShardManager snapshot successfully. Deleting the prev snapshot if available",
1280 deleteSnapshots(new SnapshotSelectionCriteria(scala.Long.MaxValue(), successMessage.metadata().timestamp() - 1,
1284 private static final class ForwardedAddServerReply {
1285 ShardInformation shardInfo;
1286 AddServerReply addServerReply;
1288 boolean removeShardOnFailure;
1290 ForwardedAddServerReply(ShardInformation shardInfo, AddServerReply addServerReply, String leaderPath,
1291 boolean removeShardOnFailure) {
1292 this.shardInfo = shardInfo;
1293 this.addServerReply = addServerReply;
1294 this.leaderPath = leaderPath;
1295 this.removeShardOnFailure = removeShardOnFailure;
1299 private static final class ForwardedAddServerFailure {
1301 String failureMessage;
1303 boolean removeShardOnFailure;
1305 ForwardedAddServerFailure(String shardName, String failureMessage, Throwable failure,
1306 boolean removeShardOnFailure) {
1307 this.shardName = shardName;
1308 this.failureMessage = failureMessage;
1309 this.failure = failure;
1310 this.removeShardOnFailure = removeShardOnFailure;
1314 static class OnShardInitialized {
1315 private final Runnable replyRunnable;
1316 private Cancellable timeoutSchedule;
1318 OnShardInitialized(Runnable replyRunnable) {
1319 this.replyRunnable = replyRunnable;
1322 Runnable getReplyRunnable() {
1323 return replyRunnable;
1326 Cancellable getTimeoutSchedule() {
1327 return timeoutSchedule;
1330 void setTimeoutSchedule(Cancellable timeoutSchedule) {
1331 this.timeoutSchedule = timeoutSchedule;
1335 static class OnShardReady extends OnShardInitialized {
1336 OnShardReady(Runnable replyRunnable) {
1337 super(replyRunnable);
1341 private void findPrimary(final String shardName, final FindPrimaryResponseHandler handler) {
1342 Timeout findPrimaryTimeout = new Timeout(datastoreContextFactory.getBaseDatastoreContext().
1343 getShardInitializationTimeout().duration().$times(2));
1346 Future<Object> futureObj = ask(getSelf(), new FindPrimary(shardName, true), findPrimaryTimeout);
1347 futureObj.onComplete(new OnComplete<Object>() {
1349 public void onComplete(Throwable failure, Object response) {
1350 if (failure != null) {
1351 handler.onFailure(failure);
1353 if(response instanceof RemotePrimaryShardFound) {
1354 handler.onRemotePrimaryShardFound((RemotePrimaryShardFound) response);
1355 } else if(response instanceof LocalPrimaryShardFound) {
1356 handler.onLocalPrimaryFound((LocalPrimaryShardFound) response);
1358 handler.onUnknownResponse(response);
1362 }, new Dispatchers(context().system().dispatchers()).getDispatcher(Dispatchers.DispatcherType.Client));
1366 * The FindPrimaryResponseHandler provides specific callback methods which are invoked when a response to the
1367 * a remote or local find primary message is processed
1369 private static interface FindPrimaryResponseHandler {
1371 * Invoked when a Failure message is received as a response
1375 void onFailure(Throwable failure);
1378 * Invoked when a RemotePrimaryShardFound response is received
1382 void onRemotePrimaryShardFound(RemotePrimaryShardFound response);
1385 * Invoked when a LocalPrimaryShardFound response is received
1388 void onLocalPrimaryFound(LocalPrimaryShardFound response);
1391 * Invoked when an unknown response is received. This is another type of failure.
1395 void onUnknownResponse(Object response);
1399 * The AutoFindPrimaryFailureResponseHandler automatically processes Failure responses when finding a primary
1400 * replica and sends a wrapped Failure response to some targetActor
1402 private static abstract class AutoFindPrimaryFailureResponseHandler implements FindPrimaryResponseHandler {
1403 private final ActorRef targetActor;
1404 private final String shardName;
1405 private final String persistenceId;
1406 private final ActorRef shardManagerActor;
1409 * @param targetActor The actor to whom the Failure response should be sent when a FindPrimary failure occurs
1410 * @param shardName The name of the shard for which the primary replica had to be found
1411 * @param persistenceId The persistenceId for the ShardManager
1412 * @param shardManagerActor The ShardManager actor which triggered the call to FindPrimary
1414 protected AutoFindPrimaryFailureResponseHandler(ActorRef targetActor, String shardName, String persistenceId, ActorRef shardManagerActor){
1415 this.targetActor = Preconditions.checkNotNull(targetActor);
1416 this.shardName = Preconditions.checkNotNull(shardName);
1417 this.persistenceId = Preconditions.checkNotNull(persistenceId);
1418 this.shardManagerActor = Preconditions.checkNotNull(shardManagerActor);
1421 public ActorRef getTargetActor() {
1425 public String getShardName() {
1430 public void onFailure(Throwable failure) {
1431 LOG.debug ("{}: Received failure from FindPrimary for shard {}", persistenceId, shardName, failure);
1432 targetActor.tell(new Status.Failure(new RuntimeException(
1433 String.format("Failed to find leader for shard %s", shardName), failure)), shardManagerActor);
1437 public void onUnknownResponse(Object response) {
1438 String msg = String.format("Failed to find leader for shard %s: received response: %s",
1439 shardName, response);
1440 LOG.debug ("{}: {}", persistenceId, msg);
1441 targetActor.tell(new Status.Failure(response instanceof Throwable ? (Throwable) response :
1442 new RuntimeException(msg)), shardManagerActor);
1447 * The PrimaryShardFoundForContext is a DTO which puts together a message (aka 'Context' message) which needs to be
1448 * forwarded to the primary replica of a shard and the message (aka 'PrimaryShardFound' message) that is received
1449 * as a successful response to find primary.
1451 private static class PrimaryShardFoundForContext {
1452 private final String shardName;
1453 private final Object contextMessage;
1454 private final RemotePrimaryShardFound remotePrimaryShardFound;
1455 private final LocalPrimaryShardFound localPrimaryShardFound;
1457 public PrimaryShardFoundForContext(@Nonnull String shardName, @Nonnull Object contextMessage,
1458 @Nonnull Object primaryFoundMessage) {
1459 this.shardName = Preconditions.checkNotNull(shardName);
1460 this.contextMessage = Preconditions.checkNotNull(contextMessage);
1461 Preconditions.checkNotNull(primaryFoundMessage);
1462 this.remotePrimaryShardFound = (primaryFoundMessage instanceof RemotePrimaryShardFound) ?
1463 (RemotePrimaryShardFound) primaryFoundMessage : null;
1464 this.localPrimaryShardFound = (primaryFoundMessage instanceof LocalPrimaryShardFound) ?
1465 (LocalPrimaryShardFound) primaryFoundMessage : null;
1469 String getPrimaryPath(){
1470 if(remotePrimaryShardFound != null) {
1471 return remotePrimaryShardFound.getPrimaryPath();
1473 return localPrimaryShardFound.getPrimaryPath();
1477 Object getContextMessage() {
1478 return contextMessage;
1482 RemotePrimaryShardFound getRemotePrimaryShardFound() {
1483 return remotePrimaryShardFound;
1487 String getShardName() {
1493 * The WrappedShardResponse class wraps a response from a Shard.
1495 private static final class WrappedShardResponse {
1496 private final ShardIdentifier shardId;
1497 private final Object response;
1498 private final String leaderPath;
1500 WrappedShardResponse(ShardIdentifier shardId, Object response, String leaderPath) {
1501 this.shardId = shardId;
1502 this.response = response;
1503 this.leaderPath = leaderPath;
1506 ShardIdentifier getShardId() {
1510 Object getResponse() {
1514 String getLeaderPath() {
1519 private static final class ShardNotInitializedTimeout {
1520 private final ActorRef sender;
1521 private final ShardInformation shardInfo;
1522 private final OnShardInitialized onShardInitialized;
1524 ShardNotInitializedTimeout(ShardInformation shardInfo, OnShardInitialized onShardInitialized, ActorRef sender) {
1525 this.sender = sender;
1526 this.shardInfo = shardInfo;
1527 this.onShardInitialized = onShardInitialized;
1530 ActorRef getSender() {
1534 ShardInformation getShardInfo() {
1538 OnShardInitialized getOnShardInitialized() {
1539 return onShardInitialized;