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 akka.actor.ActorRef;
12 import akka.actor.ActorSelection;
13 import akka.actor.Cancellable;
14 import akka.actor.Props;
15 import akka.japi.Creator;
16 import akka.persistence.RecoveryFailure;
17 import akka.serialization.Serialization;
18 import com.google.common.annotations.VisibleForTesting;
19 import com.google.common.base.Optional;
20 import com.google.common.base.Preconditions;
21 import java.io.IOException;
22 import java.util.HashMap;
24 import java.util.concurrent.TimeUnit;
25 import javax.annotation.Nonnull;
26 import org.opendaylight.controller.cluster.common.actor.CommonConfig;
27 import org.opendaylight.controller.cluster.common.actor.MeteringBehavior;
28 import org.opendaylight.controller.cluster.datastore.ShardCommitCoordinator.CohortEntry;
29 import org.opendaylight.controller.cluster.datastore.exceptions.NoShardLeaderException;
30 import org.opendaylight.controller.cluster.datastore.identifiers.ShardIdentifier;
31 import org.opendaylight.controller.cluster.datastore.identifiers.ShardTransactionIdentifier;
32 import org.opendaylight.controller.cluster.datastore.jmx.mbeans.shard.ShardMBeanFactory;
33 import org.opendaylight.controller.cluster.datastore.jmx.mbeans.shard.ShardStats;
34 import org.opendaylight.controller.cluster.datastore.messages.AbortTransaction;
35 import org.opendaylight.controller.cluster.datastore.messages.ActorInitialized;
36 import org.opendaylight.controller.cluster.datastore.messages.BatchedModifications;
37 import org.opendaylight.controller.cluster.datastore.messages.CanCommitTransaction;
38 import org.opendaylight.controller.cluster.datastore.messages.CloseTransactionChain;
39 import org.opendaylight.controller.cluster.datastore.messages.CommitTransaction;
40 import org.opendaylight.controller.cluster.datastore.messages.CommitTransactionReply;
41 import org.opendaylight.controller.cluster.datastore.messages.CreateTransaction;
42 import org.opendaylight.controller.cluster.datastore.messages.CreateTransactionReply;
43 import org.opendaylight.controller.cluster.datastore.messages.ForwardedReadyTransaction;
44 import org.opendaylight.controller.cluster.datastore.messages.PeerAddressResolved;
45 import org.opendaylight.controller.cluster.datastore.messages.ReadyLocalTransaction;
46 import org.opendaylight.controller.cluster.datastore.messages.RegisterChangeListener;
47 import org.opendaylight.controller.cluster.datastore.messages.RegisterDataTreeChangeListener;
48 import org.opendaylight.controller.cluster.datastore.messages.ShardLeaderStateChanged;
49 import org.opendaylight.controller.cluster.datastore.messages.UpdateSchemaContext;
50 import org.opendaylight.controller.cluster.datastore.modification.Modification;
51 import org.opendaylight.controller.cluster.datastore.modification.ModificationPayload;
52 import org.opendaylight.controller.cluster.datastore.modification.MutableCompositeModification;
53 import org.opendaylight.controller.cluster.datastore.utils.Dispatchers;
54 import org.opendaylight.controller.cluster.datastore.utils.MessageTracker;
55 import org.opendaylight.controller.cluster.notifications.LeaderStateChanged;
56 import org.opendaylight.controller.cluster.notifications.RegisterRoleChangeListener;
57 import org.opendaylight.controller.cluster.notifications.RoleChangeNotifier;
58 import org.opendaylight.controller.cluster.raft.RaftActor;
59 import org.opendaylight.controller.cluster.raft.RaftActorRecoveryCohort;
60 import org.opendaylight.controller.cluster.raft.RaftActorSnapshotCohort;
61 import org.opendaylight.controller.cluster.raft.RaftState;
62 import org.opendaylight.controller.cluster.raft.base.messages.FollowerInitialSyncUpStatus;
63 import org.opendaylight.controller.cluster.raft.messages.AppendEntriesReply;
64 import org.opendaylight.controller.cluster.raft.protobuff.client.messages.CompositeModificationByteStringPayload;
65 import org.opendaylight.controller.cluster.raft.protobuff.client.messages.CompositeModificationPayload;
66 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTree;
67 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeCandidate;
68 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataValidationFailedException;
69 import org.opendaylight.yangtools.yang.data.api.schema.tree.ModificationType;
70 import org.opendaylight.yangtools.yang.model.api.SchemaContext;
71 import scala.concurrent.duration.Duration;
72 import scala.concurrent.duration.FiniteDuration;
75 * A Shard represents a portion of the logical data tree <br/>
77 * Our Shard uses InMemoryDataTree as it's internal representation and delegates all requests it
80 public class Shard extends RaftActor {
82 protected static final Object TX_COMMIT_TIMEOUT_CHECK_MESSAGE = "txCommitTimeoutCheck";
85 static final Object GET_SHARD_MBEAN_MESSAGE = "getShardMBeanMessage";
88 static final String DEFAULT_NAME = "default";
90 // The state of this Shard
91 private final ShardDataTree store;
93 /// The name of this shard
94 private final String name;
96 private final ShardStats shardMBean;
98 private DatastoreContext datastoreContext;
100 private final ShardCommitCoordinator commitCoordinator;
102 private long transactionCommitTimeout;
104 private Cancellable txCommitTimeoutCheckSchedule;
106 private final Optional<ActorRef> roleChangeNotifier;
108 private final MessageTracker appendEntriesReplyTracker;
110 private final ShardTransactionActorFactory transactionActorFactory;
112 private final ShardSnapshotCohort snapshotCohort;
114 private final DataTreeChangeListenerSupport treeChangeSupport = new DataTreeChangeListenerSupport(this);
115 private final DataChangeListenerSupport changeSupport = new DataChangeListenerSupport(this);
117 protected Shard(final ShardIdentifier name, final Map<String, String> peerAddresses,
118 final DatastoreContext datastoreContext, final SchemaContext schemaContext) {
119 super(name.toString(), new HashMap<>(peerAddresses), Optional.of(datastoreContext.getShardRaftConfig()),
120 DataStoreVersions.CURRENT_VERSION);
122 this.name = name.toString();
123 this.datastoreContext = datastoreContext;
125 setPersistence(datastoreContext.isPersistent());
127 LOG.info("Shard created : {}, persistent : {}", name, datastoreContext.isPersistent());
129 store = new ShardDataTree(schemaContext);
131 shardMBean = ShardMBeanFactory.getShardStatsMBean(name.toString(),
132 datastoreContext.getDataStoreMXBeanType());
133 shardMBean.setShard(this);
135 if (isMetricsCaptureEnabled()) {
136 getContext().become(new MeteringBehavior(this));
139 commitCoordinator = new ShardCommitCoordinator(store,
140 datastoreContext.getShardCommitQueueExpiryTimeoutInMillis(),
141 datastoreContext.getShardTransactionCommitQueueCapacity(), self(), LOG, this.name);
143 setTransactionCommitTimeout();
145 // create a notifier actor for each cluster member
146 roleChangeNotifier = createRoleChangeNotifier(name.toString());
148 appendEntriesReplyTracker = new MessageTracker(AppendEntriesReply.class,
149 getRaftActorContext().getConfigParams().getIsolatedCheckIntervalInMillis());
151 transactionActorFactory = new ShardTransactionActorFactory(store, datastoreContext,
152 new Dispatchers(context().system().dispatchers()).getDispatcherPath(
153 Dispatchers.DispatcherType.Transaction), self(), getContext(), shardMBean);
155 snapshotCohort = new ShardSnapshotCohort(transactionActorFactory, store, LOG, this.name);
160 private void setTransactionCommitTimeout() {
161 transactionCommitTimeout = TimeUnit.MILLISECONDS.convert(
162 datastoreContext.getShardTransactionCommitTimeoutInSeconds(), TimeUnit.SECONDS) / 2;
165 public static Props props(final ShardIdentifier name,
166 final Map<String, String> peerAddresses,
167 final DatastoreContext datastoreContext, final SchemaContext schemaContext) {
168 Preconditions.checkNotNull(name, "name should not be null");
169 Preconditions.checkNotNull(peerAddresses, "peerAddresses should not be null");
170 Preconditions.checkNotNull(datastoreContext, "dataStoreContext should not be null");
171 Preconditions.checkNotNull(schemaContext, "schemaContext should not be null");
173 return Props.create(new ShardCreator(name, peerAddresses, datastoreContext, schemaContext));
176 private Optional<ActorRef> createRoleChangeNotifier(String shardId) {
177 ActorRef shardRoleChangeNotifier = this.getContext().actorOf(
178 RoleChangeNotifier.getProps(shardId), shardId + "-notifier");
179 return Optional.of(shardRoleChangeNotifier);
183 public void postStop() {
184 LOG.info("Stopping Shard {}", persistenceId());
188 if(txCommitTimeoutCheckSchedule != null) {
189 txCommitTimeoutCheckSchedule.cancel();
192 shardMBean.unregisterMBean();
196 public void onReceiveRecover(final Object message) throws Exception {
197 if(LOG.isDebugEnabled()) {
198 LOG.debug("{}: onReceiveRecover: Received message {} from {}", persistenceId(),
199 message.getClass().toString(), getSender());
202 if (message instanceof RecoveryFailure){
203 LOG.error("{}: Recovery failed because of this cause",
204 persistenceId(), ((RecoveryFailure) message).cause());
206 // Even though recovery failed, we still need to finish our recovery, eg send the
207 // ActorInitialized message and start the txCommitTimeoutCheckSchedule.
208 onRecoveryComplete();
210 super.onReceiveRecover(message);
211 if(LOG.isTraceEnabled()) {
212 appendEntriesReplyTracker.begin();
218 public void onReceiveCommand(final Object message) throws Exception {
220 MessageTracker.Context context = appendEntriesReplyTracker.received(message);
222 if(context.error().isPresent()){
223 LOG.trace("{} : AppendEntriesReply failed to arrive at the expected interval {}", persistenceId(),
228 if (CreateTransaction.SERIALIZABLE_CLASS.isInstance(message)) {
229 handleCreateTransaction(message);
230 } else if (BatchedModifications.class.isInstance(message)) {
231 handleBatchedModifications((BatchedModifications)message);
232 } else if (message instanceof ForwardedReadyTransaction) {
233 commitCoordinator.handleForwardedReadyTransaction((ForwardedReadyTransaction) message,
235 } else if (message instanceof ReadyLocalTransaction) {
236 handleReadyLocalTransaction((ReadyLocalTransaction)message);
237 } else if (CanCommitTransaction.SERIALIZABLE_CLASS.isInstance(message)) {
238 handleCanCommitTransaction(CanCommitTransaction.fromSerializable(message));
239 } else if (CommitTransaction.SERIALIZABLE_CLASS.isInstance(message)) {
240 handleCommitTransaction(CommitTransaction.fromSerializable(message));
241 } else if (AbortTransaction.SERIALIZABLE_CLASS.isInstance(message)) {
242 handleAbortTransaction(AbortTransaction.fromSerializable(message));
243 } else if (CloseTransactionChain.SERIALIZABLE_CLASS.isInstance(message)) {
244 closeTransactionChain(CloseTransactionChain.fromSerializable(message));
245 } else if (message instanceof RegisterChangeListener) {
246 changeSupport.onMessage((RegisterChangeListener) message, isLeader());
247 } else if (message instanceof RegisterDataTreeChangeListener) {
248 treeChangeSupport.onMessage((RegisterDataTreeChangeListener) message, isLeader());
249 } else if (message instanceof UpdateSchemaContext) {
250 updateSchemaContext((UpdateSchemaContext) message);
251 } else if (message instanceof PeerAddressResolved) {
252 PeerAddressResolved resolved = (PeerAddressResolved) message;
253 setPeerAddress(resolved.getPeerId().toString(),
254 resolved.getPeerAddress());
255 } else if (message.equals(TX_COMMIT_TIMEOUT_CHECK_MESSAGE)) {
256 handleTransactionCommitTimeoutCheck();
257 } else if(message instanceof DatastoreContext) {
258 onDatastoreContext((DatastoreContext)message);
259 } else if(message instanceof RegisterRoleChangeListener){
260 roleChangeNotifier.get().forward(message, context());
261 } else if (message instanceof FollowerInitialSyncUpStatus) {
262 shardMBean.setFollowerInitialSyncStatus(((FollowerInitialSyncUpStatus) message).isInitialSyncDone());
263 context().parent().tell(message, self());
264 } else if(GET_SHARD_MBEAN_MESSAGE.equals(message)){
265 sender().tell(getShardMBean(), self());
267 super.onReceiveCommand(message);
274 public int getPendingTxCommitQueueSize() {
275 return commitCoordinator.getQueueSize();
279 protected Optional<ActorRef> getRoleChangeNotifier() {
280 return roleChangeNotifier;
284 protected LeaderStateChanged newLeaderStateChanged(String memberId, String leaderId, short leaderPayloadVersion) {
285 return new ShardLeaderStateChanged(memberId, leaderId,
286 isLeader() ? Optional.<DataTree>of(store.getDataTree()) : Optional.<DataTree>absent(),
287 leaderPayloadVersion);
290 private void onDatastoreContext(DatastoreContext context) {
291 datastoreContext = context;
293 commitCoordinator.setQueueCapacity(datastoreContext.getShardTransactionCommitQueueCapacity());
295 setTransactionCommitTimeout();
297 if(datastoreContext.isPersistent() && !persistence().isRecoveryApplicable()) {
298 setPersistence(true);
299 } else if(!datastoreContext.isPersistent() && persistence().isRecoveryApplicable()) {
300 setPersistence(false);
303 updateConfigParams(datastoreContext.getShardRaftConfig());
306 private void handleTransactionCommitTimeoutCheck() {
307 CohortEntry cohortEntry = commitCoordinator.getCurrentCohortEntry();
308 if(cohortEntry != null) {
309 if(cohortEntry.isExpired(transactionCommitTimeout)) {
310 LOG.warn("{}: Current transaction {} has timed out after {} ms - aborting",
311 persistenceId(), cohortEntry.getTransactionID(), transactionCommitTimeout);
313 doAbortTransaction(cohortEntry.getTransactionID(), null);
317 commitCoordinator.cleanupExpiredCohortEntries();
320 private static boolean isEmptyCommit(final DataTreeCandidate candidate) {
321 return ModificationType.UNMODIFIED.equals(candidate.getRootNode().getModificationType());
324 void continueCommit(final CohortEntry cohortEntry) throws Exception {
325 final DataTreeCandidate candidate = cohortEntry.getCandidate();
327 // If we do not have any followers and we are not using persistence
328 // or if cohortEntry has no modifications
329 // we can apply modification to the state immediately
330 if ((!hasFollowers() && !persistence().isRecoveryApplicable()) || isEmptyCommit(candidate)) {
331 applyModificationToState(cohortEntry.getReplySender(), cohortEntry.getTransactionID(), candidate);
333 Shard.this.persistData(cohortEntry.getReplySender(), cohortEntry.getTransactionID(),
334 DataTreeCandidatePayload.create(candidate));
338 private void handleCommitTransaction(final CommitTransaction commit) {
339 if(!commitCoordinator.handleCommit(commit.getTransactionID(), getSender(), this)) {
340 shardMBean.incrementFailedTransactionsCount();
344 private void finishCommit(@Nonnull final ActorRef sender, @Nonnull final String transactionID, @Nonnull final CohortEntry cohortEntry) {
345 LOG.debug("{}: Finishing commit for transaction {}", persistenceId(), cohortEntry.getTransactionID());
348 cohortEntry.commit();
350 sender.tell(CommitTransactionReply.INSTANCE.toSerializable(), getSelf());
352 shardMBean.incrementCommittedTransactionCount();
353 shardMBean.setLastCommittedTransactionTime(System.currentTimeMillis());
355 } catch (Exception e) {
356 sender.tell(new akka.actor.Status.Failure(e), getSelf());
358 LOG.error("{}, An exception occurred while committing transaction {}", persistenceId(),
360 shardMBean.incrementFailedTransactionsCount();
362 commitCoordinator.currentTransactionComplete(transactionID, true);
366 private void finishCommit(@Nonnull final ActorRef sender, final @Nonnull String transactionID) {
367 // With persistence enabled, this method is called via applyState by the leader strategy
368 // after the commit has been replicated to a majority of the followers.
370 CohortEntry cohortEntry = commitCoordinator.getCohortEntryIfCurrent(transactionID);
371 if (cohortEntry == null) {
372 // The transaction is no longer the current commit. This can happen if the transaction
373 // was aborted prior, most likely due to timeout in the front-end. We need to finish
374 // committing the transaction though since it was successfully persisted and replicated
375 // however we can't use the original cohort b/c it was already preCommitted and may
376 // conflict with the current commit or may have been aborted so we commit with a new
378 cohortEntry = commitCoordinator.getAndRemoveCohortEntry(transactionID);
379 if(cohortEntry != null) {
381 store.applyForeignCandidate(transactionID, cohortEntry.getCandidate());
382 } catch (DataValidationFailedException e) {
383 shardMBean.incrementFailedTransactionsCount();
384 LOG.error("{}: Failed to re-apply transaction {}", persistenceId(), transactionID, e);
387 sender.tell(CommitTransactionReply.INSTANCE.toSerializable(), getSelf());
389 // This really shouldn't happen - it likely means that persistence or replication
390 // took so long to complete such that the cohort entry was expired from the cache.
391 IllegalStateException ex = new IllegalStateException(
392 String.format("%s: Could not finish committing transaction %s - no CohortEntry found",
393 persistenceId(), transactionID));
394 LOG.error(ex.getMessage());
395 sender.tell(new akka.actor.Status.Failure(ex), getSelf());
398 finishCommit(sender, transactionID, cohortEntry);
402 private void handleCanCommitTransaction(final CanCommitTransaction canCommit) {
403 LOG.debug("{}: Can committing transaction {}", persistenceId(), canCommit.getTransactionID());
404 commitCoordinator.handleCanCommit(canCommit.getTransactionID(), getSender(), this);
407 private void noLeaderError(String errMessage, Object message) {
408 // TODO: rather than throwing an immediate exception, we could schedule a timer to try again to make
409 // it more resilient in case we're in the process of electing a new leader.
410 getSender().tell(new akka.actor.Status.Failure(new NoShardLeaderException(errMessage, persistenceId())), getSelf());
413 private void handleBatchedModifications(BatchedModifications batched) {
414 // This message is sent to prepare the modifications transaction directly on the Shard as an
415 // optimization to avoid the extra overhead of a separate ShardTransaction actor. On the last
416 // BatchedModifications message, the caller sets the ready flag in the message indicating
417 // modifications are complete. The reply contains the cohort actor path (this actor) for the caller
418 // to initiate the 3-phase commit. This also avoids the overhead of sending an additional
419 // ReadyTransaction message.
421 // If we're not the leader then forward to the leader. This is a safety measure - we shouldn't
422 // normally get here if we're not the leader as the front-end (TransactionProxy) should determine
423 // the primary/leader shard. However with timing and caching on the front-end, there's a small
424 // window where it could have a stale leader during leadership transitions.
427 failIfIsolatedLeader(getSender());
430 commitCoordinator.handleBatchedModifications(batched, getSender(), this);
431 } catch (Exception e) {
432 LOG.error("{}: Error handling BatchedModifications for Tx {}", persistenceId(),
433 batched.getTransactionID(), e);
434 getSender().tell(new akka.actor.Status.Failure(e), getSelf());
437 ActorSelection leader = getLeader();
439 // TODO: what if this is not the first batch and leadership changed in between batched messages?
440 // We could check if the commitCoordinator already has a cached entry and forward all the previous
441 // batched modifications.
442 LOG.debug("{}: Forwarding BatchedModifications to leader {}", persistenceId(), leader);
443 leader.forward(batched, getContext());
445 noLeaderError("Could not commit transaction " + batched.getTransactionID(), batched);
450 private boolean failIfIsolatedLeader(ActorRef sender) {
451 if(getRaftState() == RaftState.IsolatedLeader) {
452 sender.tell(new akka.actor.Status.Failure(new NoShardLeaderException(String.format(
453 "Shard %s was the leader but has lost contact with all of its followers. Either all" +
454 " other follower nodes are down or this node is isolated by a network partition.",
455 persistenceId()))), getSelf());
462 private void handleReadyLocalTransaction(final ReadyLocalTransaction message) {
464 failIfIsolatedLeader(getSender());
467 commitCoordinator.handleReadyLocalTransaction(message, getSender(), this);
468 } catch (Exception e) {
469 LOG.error("{}: Error handling ReadyLocalTransaction for Tx {}", persistenceId(),
470 message.getTransactionID(), e);
471 getSender().tell(new akka.actor.Status.Failure(e), getSelf());
474 ActorSelection leader = getLeader();
475 if (leader != null) {
476 LOG.debug("{}: Forwarding ReadyLocalTransaction to leader {}", persistenceId(), leader);
477 message.setRemoteVersion(getCurrentBehavior().getLeaderPayloadVersion());
478 leader.forward(message, getContext());
480 noLeaderError("Could not commit transaction " + message.getTransactionID(), message);
485 private void handleAbortTransaction(final AbortTransaction abort) {
486 doAbortTransaction(abort.getTransactionID(), getSender());
489 void doAbortTransaction(final String transactionID, final ActorRef sender) {
490 commitCoordinator.handleAbort(transactionID, sender, this);
493 private void handleCreateTransaction(final Object message) {
495 createTransaction(CreateTransaction.fromSerializable(message));
496 } else if (getLeader() != null) {
497 getLeader().forward(message, getContext());
499 getSender().tell(new akka.actor.Status.Failure(new NoShardLeaderException(
500 "Could not create a shard transaction", persistenceId())), getSelf());
504 private void closeTransactionChain(final CloseTransactionChain closeTransactionChain) {
505 store.closeTransactionChain(closeTransactionChain.getTransactionChainId());
508 private ActorRef createTypedTransactionActor(int transactionType,
509 ShardTransactionIdentifier transactionId, String transactionChainId,
510 short clientVersion ) {
512 return transactionActorFactory.newShardTransaction(TransactionType.fromInt(transactionType),
513 transactionId, transactionChainId, clientVersion);
516 private void createTransaction(CreateTransaction createTransaction) {
518 if(TransactionType.fromInt(createTransaction.getTransactionType()) != TransactionType.READ_ONLY &&
519 failIfIsolatedLeader(getSender())) {
523 ActorRef transactionActor = createTransaction(createTransaction.getTransactionType(),
524 createTransaction.getTransactionId(), createTransaction.getTransactionChainId(),
525 createTransaction.getVersion());
527 getSender().tell(new CreateTransactionReply(Serialization.serializedActorPath(transactionActor),
528 createTransaction.getTransactionId()).toSerializable(), getSelf());
529 } catch (Exception e) {
530 getSender().tell(new akka.actor.Status.Failure(e), getSelf());
534 private ActorRef createTransaction(int transactionType, String remoteTransactionId,
535 String transactionChainId, short clientVersion) {
538 ShardTransactionIdentifier transactionId = new ShardTransactionIdentifier(remoteTransactionId);
540 if(LOG.isDebugEnabled()) {
541 LOG.debug("{}: Creating transaction : {} ", persistenceId(), transactionId);
544 ActorRef transactionActor = createTypedTransactionActor(transactionType, transactionId,
545 transactionChainId, clientVersion);
547 return transactionActor;
550 private void commitWithNewTransaction(final Modification modification) {
551 ReadWriteShardDataTreeTransaction tx = store.newReadWriteTransaction(modification.toString(), null);
552 modification.apply(tx.getSnapshot());
554 snapshotCohort.syncCommitTransaction(tx);
555 shardMBean.incrementCommittedTransactionCount();
556 shardMBean.setLastCommittedTransactionTime(System.currentTimeMillis());
557 } catch (Exception e) {
558 shardMBean.incrementFailedTransactionsCount();
559 LOG.error("{}: Failed to commit", persistenceId(), e);
563 private void updateSchemaContext(final UpdateSchemaContext message) {
564 updateSchemaContext(message.getSchemaContext());
568 void updateSchemaContext(final SchemaContext schemaContext) {
569 store.updateSchemaContext(schemaContext);
572 private boolean isMetricsCaptureEnabled() {
573 CommonConfig config = new CommonConfig(getContext().system().settings().config());
574 return config.isMetricCaptureEnabled();
579 public RaftActorSnapshotCohort getRaftActorSnapshotCohort() {
580 return snapshotCohort;
585 protected RaftActorRecoveryCohort getRaftActorRecoveryCohort() {
586 return new ShardRecoveryCoordinator(store, store.getSchemaContext(), persistenceId(), LOG);
590 protected void onRecoveryComplete() {
591 //notify shard manager
592 getContext().parent().tell(new ActorInitialized(), getSelf());
594 // Being paranoid here - this method should only be called once but just in case...
595 if(txCommitTimeoutCheckSchedule == null) {
596 // Schedule a message to be periodically sent to check if the current in-progress
597 // transaction should be expired and aborted.
598 FiniteDuration period = Duration.create(transactionCommitTimeout / 3, TimeUnit.MILLISECONDS);
599 txCommitTimeoutCheckSchedule = getContext().system().scheduler().schedule(
600 period, period, getSelf(),
601 TX_COMMIT_TIMEOUT_CHECK_MESSAGE, getContext().dispatcher(), ActorRef.noSender());
606 protected void applyState(final ActorRef clientActor, final String identifier, final Object data) {
607 if (data instanceof DataTreeCandidatePayload) {
608 if (clientActor == null) {
609 // No clientActor indicates a replica coming from the leader
611 store.applyForeignCandidate(identifier, ((DataTreeCandidatePayload)data).getCandidate());
612 } catch (DataValidationFailedException | IOException e) {
613 LOG.error("{}: Error applying replica {}", persistenceId(), identifier, e);
616 // Replication consensus reached, proceed to commit
617 finishCommit(clientActor, identifier);
619 } else if (data instanceof ModificationPayload) {
621 applyModificationToState(clientActor, identifier, ((ModificationPayload) data).getModification());
622 } catch (ClassNotFoundException | IOException e) {
623 LOG.error("{}: Error extracting ModificationPayload", persistenceId(), e);
625 } else if (data instanceof CompositeModificationPayload) {
626 Object modification = ((CompositeModificationPayload) data).getModification();
628 applyModificationToState(clientActor, identifier, modification);
629 } else if(data instanceof CompositeModificationByteStringPayload ){
630 Object modification = ((CompositeModificationByteStringPayload) data).getModification();
632 applyModificationToState(clientActor, identifier, modification);
634 LOG.error("{}: Unknown state received {} Class loader = {} CompositeNodeMod.ClassLoader = {}",
635 persistenceId(), data, data.getClass().getClassLoader(),
636 CompositeModificationPayload.class.getClassLoader());
640 private void applyModificationToState(ActorRef clientActor, String identifier, Object modification) {
641 if(modification == null) {
643 "{}: modification is null - this is very unexpected, clientActor = {}, identifier = {}",
644 persistenceId(), identifier, clientActor != null ? clientActor.path().toString() : null);
645 } else if(clientActor == null) {
646 // There's no clientActor to which to send a commit reply so we must be applying
647 // replicated state from the leader.
648 commitWithNewTransaction(MutableCompositeModification.fromSerializable(modification));
650 // This must be the OK to commit after replication consensus.
651 finishCommit(clientActor, identifier);
656 protected void onStateChanged() {
657 boolean isLeader = isLeader();
658 changeSupport.onLeadershipChange(isLeader);
659 treeChangeSupport.onLeadershipChange(isLeader);
661 // If this actor is no longer the leader close all the transaction chains
663 if(LOG.isDebugEnabled()) {
665 "{}: onStateChanged: Closing all transaction chains because shard {} is no longer the leader",
666 persistenceId(), getId());
669 store.closeAllTransactionChains();
674 protected void onLeaderChanged(String oldLeader, String newLeader) {
675 shardMBean.incrementLeadershipChangeCount();
679 public String persistenceId() {
684 ShardCommitCoordinator getCommitCoordinator() {
685 return commitCoordinator;
689 private static class ShardCreator implements Creator<Shard> {
691 private static final long serialVersionUID = 1L;
693 final ShardIdentifier name;
694 final Map<String, String> peerAddresses;
695 final DatastoreContext datastoreContext;
696 final SchemaContext schemaContext;
698 ShardCreator(final ShardIdentifier name, final Map<String, String> peerAddresses,
699 final DatastoreContext datastoreContext, final SchemaContext schemaContext) {
701 this.peerAddresses = peerAddresses;
702 this.datastoreContext = datastoreContext;
703 this.schemaContext = schemaContext;
707 public Shard create() throws Exception {
708 return new Shard(name, peerAddresses, datastoreContext, schemaContext);
713 public ShardDataTree getDataStore() {
718 ShardStats getShardMBean() {