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.persistence.RecoveryFailure;
16 import akka.serialization.Serialization;
17 import com.google.common.annotations.VisibleForTesting;
18 import com.google.common.base.Optional;
19 import com.google.common.base.Preconditions;
20 import java.io.IOException;
21 import java.util.Collections;
23 import java.util.concurrent.TimeUnit;
24 import javax.annotation.Nonnull;
25 import org.opendaylight.controller.cluster.common.actor.CommonConfig;
26 import org.opendaylight.controller.cluster.common.actor.MeteringBehavior;
27 import org.opendaylight.controller.cluster.datastore.ShardCommitCoordinator.CohortEntry;
28 import org.opendaylight.controller.cluster.datastore.exceptions.NoShardLeaderException;
29 import org.opendaylight.controller.cluster.datastore.identifiers.ShardIdentifier;
30 import org.opendaylight.controller.cluster.datastore.identifiers.ShardTransactionIdentifier;
31 import org.opendaylight.controller.cluster.datastore.jmx.mbeans.shard.ShardMBeanFactory;
32 import org.opendaylight.controller.cluster.datastore.jmx.mbeans.shard.ShardStats;
33 import org.opendaylight.controller.cluster.datastore.messages.AbortTransaction;
34 import org.opendaylight.controller.cluster.datastore.messages.ActorInitialized;
35 import org.opendaylight.controller.cluster.datastore.messages.BatchedModifications;
36 import org.opendaylight.controller.cluster.datastore.messages.CanCommitTransaction;
37 import org.opendaylight.controller.cluster.datastore.messages.CloseTransactionChain;
38 import org.opendaylight.controller.cluster.datastore.messages.CommitTransaction;
39 import org.opendaylight.controller.cluster.datastore.messages.CommitTransactionReply;
40 import org.opendaylight.controller.cluster.datastore.messages.CreateTransaction;
41 import org.opendaylight.controller.cluster.datastore.messages.CreateTransactionReply;
42 import org.opendaylight.controller.cluster.datastore.messages.DatastoreSnapshot;
43 import org.opendaylight.controller.cluster.datastore.messages.DatastoreSnapshot.ShardSnapshot;
44 import org.opendaylight.controller.cluster.datastore.messages.ForwardedReadyTransaction;
45 import org.opendaylight.controller.cluster.datastore.messages.GetShardDataTree;
46 import org.opendaylight.controller.cluster.datastore.messages.PeerAddressResolved;
47 import org.opendaylight.controller.cluster.datastore.messages.ReadyLocalTransaction;
48 import org.opendaylight.controller.cluster.datastore.messages.RegisterChangeListener;
49 import org.opendaylight.controller.cluster.datastore.messages.RegisterDataTreeChangeListener;
50 import org.opendaylight.controller.cluster.datastore.messages.ShardLeaderStateChanged;
51 import org.opendaylight.controller.cluster.datastore.messages.UpdateSchemaContext;
52 import org.opendaylight.controller.cluster.datastore.modification.Modification;
53 import org.opendaylight.controller.cluster.datastore.modification.ModificationPayload;
54 import org.opendaylight.controller.cluster.datastore.modification.MutableCompositeModification;
55 import org.opendaylight.controller.cluster.datastore.utils.Dispatchers;
56 import org.opendaylight.controller.cluster.datastore.utils.MessageTracker;
57 import org.opendaylight.controller.cluster.notifications.LeaderStateChanged;
58 import org.opendaylight.controller.cluster.notifications.RegisterRoleChangeListener;
59 import org.opendaylight.controller.cluster.notifications.RoleChangeNotifier;
60 import org.opendaylight.controller.cluster.raft.RaftActor;
61 import org.opendaylight.controller.cluster.raft.RaftActorRecoveryCohort;
62 import org.opendaylight.controller.cluster.raft.RaftActorSnapshotCohort;
63 import org.opendaylight.controller.cluster.raft.RaftState;
64 import org.opendaylight.controller.cluster.raft.base.messages.FollowerInitialSyncUpStatus;
65 import org.opendaylight.controller.cluster.raft.messages.AppendEntriesReply;
66 import org.opendaylight.controller.cluster.raft.messages.ServerRemoved;
67 import org.opendaylight.controller.cluster.raft.protobuff.client.messages.CompositeModificationByteStringPayload;
68 import org.opendaylight.controller.cluster.raft.protobuff.client.messages.CompositeModificationPayload;
69 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTree;
70 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeCandidate;
71 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataValidationFailedException;
72 import org.opendaylight.yangtools.yang.data.api.schema.tree.ModificationType;
73 import org.opendaylight.yangtools.yang.model.api.SchemaContext;
74 import scala.concurrent.duration.Duration;
75 import scala.concurrent.duration.FiniteDuration;
78 * A Shard represents a portion of the logical data tree <br/>
80 * Our Shard uses InMemoryDataTree as it's internal representation and delegates all requests it
83 public class Shard extends RaftActor {
85 protected static final Object TX_COMMIT_TIMEOUT_CHECK_MESSAGE = "txCommitTimeoutCheck";
88 static final Object GET_SHARD_MBEAN_MESSAGE = "getShardMBeanMessage";
91 static final String DEFAULT_NAME = "default";
93 // The state of this Shard
94 private final ShardDataTree store;
96 /// The name of this shard
97 private final String name;
99 private final ShardStats shardMBean;
101 private DatastoreContext datastoreContext;
103 private final ShardCommitCoordinator commitCoordinator;
105 private long transactionCommitTimeout;
107 private Cancellable txCommitTimeoutCheckSchedule;
109 private final Optional<ActorRef> roleChangeNotifier;
111 private final MessageTracker appendEntriesReplyTracker;
113 private final ShardTransactionActorFactory transactionActorFactory;
115 private final ShardSnapshotCohort snapshotCohort;
117 private final DataTreeChangeListenerSupport treeChangeSupport = new DataTreeChangeListenerSupport(this);
118 private final DataChangeListenerSupport changeSupport = new DataChangeListenerSupport(this);
121 private ShardSnapshot restoreFromSnapshot;
125 protected Shard(AbstractBuilder<?, ?> builder) {
126 super(builder.getId().toString(), builder.getPeerAddresses(),
127 Optional.of(builder.getDatastoreContext().getShardRaftConfig()), DataStoreVersions.CURRENT_VERSION);
129 this.name = builder.getId().toString();
130 this.datastoreContext = builder.getDatastoreContext();
131 this.restoreFromSnapshot = builder.getRestoreFromSnapshot();
133 setPersistence(datastoreContext.isPersistent());
135 LOG.info("Shard created : {}, persistent : {}", name, datastoreContext.isPersistent());
137 // FIXME: BUG-1014: pass down the proper TreeType
138 store = new ShardDataTree(builder.getSchemaContext());
140 shardMBean = ShardMBeanFactory.getShardStatsMBean(name.toString(),
141 datastoreContext.getDataStoreMXBeanType());
142 shardMBean.setShard(this);
144 if (isMetricsCaptureEnabled()) {
145 getContext().become(new MeteringBehavior(this));
148 commitCoordinator = new ShardCommitCoordinator(store,
149 datastoreContext.getShardCommitQueueExpiryTimeoutInMillis(),
150 datastoreContext.getShardTransactionCommitQueueCapacity(), LOG, this.name);
152 setTransactionCommitTimeout();
154 // create a notifier actor for each cluster member
155 roleChangeNotifier = createRoleChangeNotifier(name.toString());
157 appendEntriesReplyTracker = new MessageTracker(AppendEntriesReply.class,
158 getRaftActorContext().getConfigParams().getIsolatedCheckIntervalInMillis());
160 transactionActorFactory = new ShardTransactionActorFactory(store, datastoreContext,
161 new Dispatchers(context().system().dispatchers()).getDispatcherPath(
162 Dispatchers.DispatcherType.Transaction), self(), getContext(), shardMBean);
164 snapshotCohort = new ShardSnapshotCohort(transactionActorFactory, store, LOG, this.name);
170 private void setTransactionCommitTimeout() {
171 transactionCommitTimeout = TimeUnit.MILLISECONDS.convert(
172 datastoreContext.getShardTransactionCommitTimeoutInSeconds(), TimeUnit.SECONDS) / 2;
175 private Optional<ActorRef> createRoleChangeNotifier(String shardId) {
176 ActorRef shardRoleChangeNotifier = this.getContext().actorOf(
177 RoleChangeNotifier.getProps(shardId), shardId + "-notifier");
178 return Optional.of(shardRoleChangeNotifier);
182 public void postStop() {
183 LOG.info("Stopping Shard {}", persistenceId());
187 if(txCommitTimeoutCheckSchedule != null) {
188 txCommitTimeoutCheckSchedule.cancel();
191 shardMBean.unregisterMBean();
195 public void onReceiveRecover(final Object message) throws Exception {
196 if(LOG.isDebugEnabled()) {
197 LOG.debug("{}: onReceiveRecover: Received message {} from {}", persistenceId(),
198 message.getClass().toString(), getSender());
201 if (message instanceof RecoveryFailure){
202 LOG.error("{}: Recovery failed because of this cause",
203 persistenceId(), ((RecoveryFailure) message).cause());
205 // Even though recovery failed, we still need to finish our recovery, eg send the
206 // ActorInitialized message and start the txCommitTimeoutCheckSchedule.
207 onRecoveryComplete();
209 super.onReceiveRecover(message);
210 if(LOG.isTraceEnabled()) {
211 appendEntriesReplyTracker.begin();
217 public void onReceiveCommand(final Object message) throws Exception {
219 MessageTracker.Context context = appendEntriesReplyTracker.received(message);
221 if(context.error().isPresent()){
222 LOG.trace("{} : AppendEntriesReply failed to arrive at the expected interval {}", persistenceId(),
227 if (CreateTransaction.SERIALIZABLE_CLASS.isInstance(message)) {
228 handleCreateTransaction(message);
229 } else if (BatchedModifications.class.isInstance(message)) {
230 handleBatchedModifications((BatchedModifications)message);
231 } else if (message instanceof ForwardedReadyTransaction) {
232 commitCoordinator.handleForwardedReadyTransaction((ForwardedReadyTransaction) message,
234 } else if (message instanceof ReadyLocalTransaction) {
235 handleReadyLocalTransaction((ReadyLocalTransaction)message);
236 } else if (CanCommitTransaction.SERIALIZABLE_CLASS.isInstance(message)) {
237 handleCanCommitTransaction(CanCommitTransaction.fromSerializable(message));
238 } else if (CommitTransaction.SERIALIZABLE_CLASS.isInstance(message)) {
239 handleCommitTransaction(CommitTransaction.fromSerializable(message));
240 } else if (AbortTransaction.SERIALIZABLE_CLASS.isInstance(message)) {
241 handleAbortTransaction(AbortTransaction.fromSerializable(message));
242 } else if (CloseTransactionChain.SERIALIZABLE_CLASS.isInstance(message)) {
243 closeTransactionChain(CloseTransactionChain.fromSerializable(message));
244 } else if (message instanceof RegisterChangeListener) {
245 changeSupport.onMessage((RegisterChangeListener) message, isLeader(), hasLeader());
246 } else if (message instanceof RegisterDataTreeChangeListener) {
247 treeChangeSupport.onMessage((RegisterDataTreeChangeListener) message, isLeader(), hasLeader());
248 } else if (message instanceof UpdateSchemaContext) {
249 updateSchemaContext((UpdateSchemaContext) message);
250 } else if (message instanceof PeerAddressResolved) {
251 PeerAddressResolved resolved = (PeerAddressResolved) message;
252 setPeerAddress(resolved.getPeerId().toString(),
253 resolved.getPeerAddress());
254 } else if (message.equals(TX_COMMIT_TIMEOUT_CHECK_MESSAGE)) {
255 handleTransactionCommitTimeoutCheck();
256 } else if(message instanceof DatastoreContext) {
257 onDatastoreContext((DatastoreContext)message);
258 } else if(message instanceof RegisterRoleChangeListener){
259 roleChangeNotifier.get().forward(message, context());
260 } else if (message instanceof FollowerInitialSyncUpStatus) {
261 shardMBean.setFollowerInitialSyncStatus(((FollowerInitialSyncUpStatus) message).isInitialSyncDone());
262 context().parent().tell(message, self());
263 } else if(GET_SHARD_MBEAN_MESSAGE.equals(message)){
264 sender().tell(getShardMBean(), self());
265 } else if(message instanceof GetShardDataTree) {
266 sender().tell(store.getDataTree(), self());
267 } else if(message instanceof ServerRemoved){
268 context().parent().forward(message, context());
270 super.onReceiveCommand(message);
277 private boolean hasLeader() {
278 return getLeaderId() != null;
281 public int getPendingTxCommitQueueSize() {
282 return commitCoordinator.getQueueSize();
286 protected Optional<ActorRef> getRoleChangeNotifier() {
287 return roleChangeNotifier;
291 protected LeaderStateChanged newLeaderStateChanged(String memberId, String leaderId, short leaderPayloadVersion) {
292 return new ShardLeaderStateChanged(memberId, leaderId,
293 isLeader() ? Optional.<DataTree>of(store.getDataTree()) : Optional.<DataTree>absent(),
294 leaderPayloadVersion);
297 protected void onDatastoreContext(DatastoreContext context) {
298 datastoreContext = context;
300 commitCoordinator.setQueueCapacity(datastoreContext.getShardTransactionCommitQueueCapacity());
302 setTransactionCommitTimeout();
304 if(datastoreContext.isPersistent() && !persistence().isRecoveryApplicable()) {
305 setPersistence(true);
306 } else if(!datastoreContext.isPersistent() && persistence().isRecoveryApplicable()) {
307 setPersistence(false);
310 updateConfigParams(datastoreContext.getShardRaftConfig());
313 private void handleTransactionCommitTimeoutCheck() {
314 CohortEntry cohortEntry = commitCoordinator.getCurrentCohortEntry();
315 if(cohortEntry != null) {
316 if(cohortEntry.isExpired(transactionCommitTimeout)) {
317 LOG.warn("{}: Current transaction {} has timed out after {} ms - aborting",
318 persistenceId(), cohortEntry.getTransactionID(), transactionCommitTimeout);
320 doAbortTransaction(cohortEntry.getTransactionID(), null);
324 commitCoordinator.cleanupExpiredCohortEntries();
327 private static boolean isEmptyCommit(final DataTreeCandidate candidate) {
328 return ModificationType.UNMODIFIED.equals(candidate.getRootNode().getModificationType());
331 void continueCommit(final CohortEntry cohortEntry) {
332 final DataTreeCandidate candidate = cohortEntry.getCandidate();
334 // If we do not have any followers and we are not using persistence
335 // or if cohortEntry has no modifications
336 // we can apply modification to the state immediately
337 if ((!hasFollowers() && !persistence().isRecoveryApplicable()) || isEmptyCommit(candidate)) {
338 applyModificationToState(cohortEntry.getReplySender(), cohortEntry.getTransactionID(), candidate);
340 Shard.this.persistData(cohortEntry.getReplySender(), cohortEntry.getTransactionID(),
341 DataTreeCandidatePayload.create(candidate));
345 private void handleCommitTransaction(final CommitTransaction commit) {
346 if(!commitCoordinator.handleCommit(commit.getTransactionID(), getSender(), this)) {
347 shardMBean.incrementFailedTransactionsCount();
351 private void finishCommit(@Nonnull final ActorRef sender, @Nonnull final String transactionID, @Nonnull final CohortEntry cohortEntry) {
352 LOG.debug("{}: Finishing commit for transaction {}", persistenceId(), cohortEntry.getTransactionID());
355 cohortEntry.commit();
357 sender.tell(CommitTransactionReply.INSTANCE.toSerializable(), getSelf());
359 shardMBean.incrementCommittedTransactionCount();
360 shardMBean.setLastCommittedTransactionTime(System.currentTimeMillis());
362 } catch (Exception e) {
363 sender.tell(new akka.actor.Status.Failure(e), getSelf());
365 LOG.error("{}, An exception occurred while committing transaction {}", persistenceId(),
367 shardMBean.incrementFailedTransactionsCount();
369 commitCoordinator.currentTransactionComplete(transactionID, true);
373 private void finishCommit(@Nonnull final ActorRef sender, final @Nonnull String transactionID) {
374 // With persistence enabled, this method is called via applyState by the leader strategy
375 // after the commit has been replicated to a majority of the followers.
377 CohortEntry cohortEntry = commitCoordinator.getCohortEntryIfCurrent(transactionID);
378 if (cohortEntry == null) {
379 // The transaction is no longer the current commit. This can happen if the transaction
380 // was aborted prior, most likely due to timeout in the front-end. We need to finish
381 // committing the transaction though since it was successfully persisted and replicated
382 // however we can't use the original cohort b/c it was already preCommitted and may
383 // conflict with the current commit or may have been aborted so we commit with a new
385 cohortEntry = commitCoordinator.getAndRemoveCohortEntry(transactionID);
386 if(cohortEntry != null) {
388 store.applyForeignCandidate(transactionID, cohortEntry.getCandidate());
389 } catch (DataValidationFailedException e) {
390 shardMBean.incrementFailedTransactionsCount();
391 LOG.error("{}: Failed to re-apply transaction {}", persistenceId(), transactionID, e);
394 sender.tell(CommitTransactionReply.INSTANCE.toSerializable(), getSelf());
396 // This really shouldn't happen - it likely means that persistence or replication
397 // took so long to complete such that the cohort entry was expired from the cache.
398 IllegalStateException ex = new IllegalStateException(
399 String.format("%s: Could not finish committing transaction %s - no CohortEntry found",
400 persistenceId(), transactionID));
401 LOG.error(ex.getMessage());
402 sender.tell(new akka.actor.Status.Failure(ex), getSelf());
405 finishCommit(sender, transactionID, cohortEntry);
409 private void handleCanCommitTransaction(final CanCommitTransaction canCommit) {
410 LOG.debug("{}: Can committing transaction {}", persistenceId(), canCommit.getTransactionID());
411 commitCoordinator.handleCanCommit(canCommit.getTransactionID(), getSender(), this);
414 private void noLeaderError(String errMessage, Object message) {
415 // TODO: rather than throwing an immediate exception, we could schedule a timer to try again to make
416 // it more resilient in case we're in the process of electing a new leader.
417 getSender().tell(new akka.actor.Status.Failure(new NoShardLeaderException(errMessage, persistenceId())), getSelf());
420 protected void handleBatchedModificationsLocal(BatchedModifications batched, ActorRef sender) {
422 commitCoordinator.handleBatchedModifications(batched, sender, this);
423 } catch (Exception e) {
424 LOG.error("{}: Error handling BatchedModifications for Tx {}", persistenceId(),
425 batched.getTransactionID(), e);
426 sender.tell(new akka.actor.Status.Failure(e), getSelf());
430 private void handleBatchedModifications(BatchedModifications batched) {
431 // This message is sent to prepare the modifications transaction directly on the Shard as an
432 // optimization to avoid the extra overhead of a separate ShardTransaction actor. On the last
433 // BatchedModifications message, the caller sets the ready flag in the message indicating
434 // modifications are complete. The reply contains the cohort actor path (this actor) for the caller
435 // to initiate the 3-phase commit. This also avoids the overhead of sending an additional
436 // ReadyTransaction message.
438 // If we're not the leader then forward to the leader. This is a safety measure - we shouldn't
439 // normally get here if we're not the leader as the front-end (TransactionProxy) should determine
440 // the primary/leader shard. However with timing and caching on the front-end, there's a small
441 // window where it could have a stale leader during leadership transitions.
444 failIfIsolatedLeader(getSender());
446 handleBatchedModificationsLocal(batched, getSender());
448 ActorSelection leader = getLeader();
450 // TODO: what if this is not the first batch and leadership changed in between batched messages?
451 // We could check if the commitCoordinator already has a cached entry and forward all the previous
452 // batched modifications.
453 LOG.debug("{}: Forwarding BatchedModifications to leader {}", persistenceId(), leader);
454 leader.forward(batched, getContext());
456 noLeaderError("Could not commit transaction " + batched.getTransactionID(), batched);
461 private boolean failIfIsolatedLeader(ActorRef sender) {
462 if(isIsolatedLeader()) {
463 sender.tell(new akka.actor.Status.Failure(new NoShardLeaderException(String.format(
464 "Shard %s was the leader but has lost contact with all of its followers. Either all" +
465 " other follower nodes are down or this node is isolated by a network partition.",
466 persistenceId()))), getSelf());
473 protected boolean isIsolatedLeader() {
474 return getRaftState() == RaftState.IsolatedLeader;
477 private void handleReadyLocalTransaction(final ReadyLocalTransaction message) {
479 failIfIsolatedLeader(getSender());
482 commitCoordinator.handleReadyLocalTransaction(message, getSender(), this);
483 } catch (Exception e) {
484 LOG.error("{}: Error handling ReadyLocalTransaction for Tx {}", persistenceId(),
485 message.getTransactionID(), e);
486 getSender().tell(new akka.actor.Status.Failure(e), getSelf());
489 ActorSelection leader = getLeader();
490 if (leader != null) {
491 LOG.debug("{}: Forwarding ReadyLocalTransaction to leader {}", persistenceId(), leader);
492 message.setRemoteVersion(getCurrentBehavior().getLeaderPayloadVersion());
493 leader.forward(message, getContext());
495 noLeaderError("Could not commit transaction " + message.getTransactionID(), message);
500 private void handleAbortTransaction(final AbortTransaction abort) {
501 doAbortTransaction(abort.getTransactionID(), getSender());
504 void doAbortTransaction(final String transactionID, final ActorRef sender) {
505 commitCoordinator.handleAbort(transactionID, sender, this);
508 private void handleCreateTransaction(final Object message) {
510 createTransaction(CreateTransaction.fromSerializable(message));
511 } else if (getLeader() != null) {
512 getLeader().forward(message, getContext());
514 getSender().tell(new akka.actor.Status.Failure(new NoShardLeaderException(
515 "Could not create a shard transaction", persistenceId())), getSelf());
519 private void closeTransactionChain(final CloseTransactionChain closeTransactionChain) {
520 store.closeTransactionChain(closeTransactionChain.getTransactionChainId());
523 private ActorRef createTypedTransactionActor(int transactionType,
524 ShardTransactionIdentifier transactionId, String transactionChainId,
525 short clientVersion ) {
527 return transactionActorFactory.newShardTransaction(TransactionType.fromInt(transactionType),
528 transactionId, transactionChainId, clientVersion);
531 private void createTransaction(CreateTransaction createTransaction) {
533 if(TransactionType.fromInt(createTransaction.getTransactionType()) != TransactionType.READ_ONLY &&
534 failIfIsolatedLeader(getSender())) {
538 ActorRef transactionActor = createTransaction(createTransaction.getTransactionType(),
539 createTransaction.getTransactionId(), createTransaction.getTransactionChainId(),
540 createTransaction.getVersion());
542 getSender().tell(new CreateTransactionReply(Serialization.serializedActorPath(transactionActor),
543 createTransaction.getTransactionId()).toSerializable(), getSelf());
544 } catch (Exception e) {
545 getSender().tell(new akka.actor.Status.Failure(e), getSelf());
549 private ActorRef createTransaction(int transactionType, String remoteTransactionId,
550 String transactionChainId, short clientVersion) {
553 ShardTransactionIdentifier transactionId = new ShardTransactionIdentifier(remoteTransactionId);
555 if(LOG.isDebugEnabled()) {
556 LOG.debug("{}: Creating transaction : {} ", persistenceId(), transactionId);
559 ActorRef transactionActor = createTypedTransactionActor(transactionType, transactionId,
560 transactionChainId, clientVersion);
562 return transactionActor;
565 private void commitWithNewTransaction(final Modification modification) {
566 ReadWriteShardDataTreeTransaction tx = store.newReadWriteTransaction(modification.toString(), null);
567 modification.apply(tx.getSnapshot());
569 snapshotCohort.syncCommitTransaction(tx);
570 shardMBean.incrementCommittedTransactionCount();
571 shardMBean.setLastCommittedTransactionTime(System.currentTimeMillis());
572 } catch (Exception e) {
573 shardMBean.incrementFailedTransactionsCount();
574 LOG.error("{}: Failed to commit", persistenceId(), e);
578 private void updateSchemaContext(final UpdateSchemaContext message) {
579 updateSchemaContext(message.getSchemaContext());
583 void updateSchemaContext(final SchemaContext schemaContext) {
584 store.updateSchemaContext(schemaContext);
587 private boolean isMetricsCaptureEnabled() {
588 CommonConfig config = new CommonConfig(getContext().system().settings().config());
589 return config.isMetricCaptureEnabled();
594 public RaftActorSnapshotCohort getRaftActorSnapshotCohort() {
595 return snapshotCohort;
600 protected RaftActorRecoveryCohort getRaftActorRecoveryCohort() {
601 return new ShardRecoveryCoordinator(store, store.getSchemaContext(),
602 restoreFromSnapshot != null ? restoreFromSnapshot.getSnapshot() : null, persistenceId(), LOG);
606 protected void onRecoveryComplete() {
607 restoreFromSnapshot = null;
609 //notify shard manager
610 getContext().parent().tell(new ActorInitialized(), getSelf());
612 // Being paranoid here - this method should only be called once but just in case...
613 if(txCommitTimeoutCheckSchedule == null) {
614 // Schedule a message to be periodically sent to check if the current in-progress
615 // transaction should be expired and aborted.
616 FiniteDuration period = Duration.create(transactionCommitTimeout / 3, TimeUnit.MILLISECONDS);
617 txCommitTimeoutCheckSchedule = getContext().system().scheduler().schedule(
618 period, period, getSelf(),
619 TX_COMMIT_TIMEOUT_CHECK_MESSAGE, getContext().dispatcher(), ActorRef.noSender());
624 protected void applyState(final ActorRef clientActor, final String identifier, final Object data) {
625 if (data instanceof DataTreeCandidatePayload) {
626 if (clientActor == null) {
627 // No clientActor indicates a replica coming from the leader
629 store.applyForeignCandidate(identifier, ((DataTreeCandidatePayload)data).getCandidate());
630 } catch (DataValidationFailedException | IOException e) {
631 LOG.error("{}: Error applying replica {}", persistenceId(), identifier, e);
634 // Replication consensus reached, proceed to commit
635 finishCommit(clientActor, identifier);
637 } else if (data instanceof ModificationPayload) {
639 applyModificationToState(clientActor, identifier, ((ModificationPayload) data).getModification());
640 } catch (ClassNotFoundException | IOException e) {
641 LOG.error("{}: Error extracting ModificationPayload", persistenceId(), e);
643 } else if (data instanceof CompositeModificationPayload) {
644 Object modification = ((CompositeModificationPayload) data).getModification();
646 applyModificationToState(clientActor, identifier, modification);
647 } else if(data instanceof CompositeModificationByteStringPayload ){
648 Object modification = ((CompositeModificationByteStringPayload) data).getModification();
650 applyModificationToState(clientActor, identifier, modification);
652 LOG.error("{}: Unknown state received {} Class loader = {} CompositeNodeMod.ClassLoader = {}",
653 persistenceId(), data, data.getClass().getClassLoader(),
654 CompositeModificationPayload.class.getClassLoader());
658 private void applyModificationToState(ActorRef clientActor, String identifier, Object modification) {
659 if(modification == null) {
661 "{}: modification is null - this is very unexpected, clientActor = {}, identifier = {}",
662 persistenceId(), identifier, clientActor != null ? clientActor.path().toString() : null);
663 } else if(clientActor == null) {
664 // There's no clientActor to which to send a commit reply so we must be applying
665 // replicated state from the leader.
666 commitWithNewTransaction(MutableCompositeModification.fromSerializable(modification));
668 // This must be the OK to commit after replication consensus.
669 finishCommit(clientActor, identifier);
674 protected void onStateChanged() {
675 boolean isLeader = isLeader();
676 boolean hasLeader = hasLeader();
677 changeSupport.onLeadershipChange(isLeader, hasLeader);
678 treeChangeSupport.onLeadershipChange(isLeader, hasLeader);
680 // If this actor is no longer the leader close all the transaction chains
682 if(LOG.isDebugEnabled()) {
684 "{}: onStateChanged: Closing all transaction chains because shard {} is no longer the leader",
685 persistenceId(), getId());
688 store.closeAllTransactionChains();
693 protected void onLeaderChanged(String oldLeader, String newLeader) {
694 shardMBean.incrementLeadershipChangeCount();
698 public String persistenceId() {
703 ShardCommitCoordinator getCommitCoordinator() {
704 return commitCoordinator;
707 public DatastoreContext getDatastoreContext() {
708 return datastoreContext;
712 public ShardDataTree getDataStore() {
717 ShardStats getShardMBean() {
721 public static Builder builder() {
722 return new Builder();
725 public static abstract class AbstractBuilder<T extends AbstractBuilder<T, S>, S extends Shard> {
726 private final Class<S> shardClass;
727 private ShardIdentifier id;
728 private Map<String, String> peerAddresses = Collections.emptyMap();
729 private DatastoreContext datastoreContext;
730 private SchemaContext schemaContext;
731 private DatastoreSnapshot.ShardSnapshot restoreFromSnapshot;
732 private volatile boolean sealed;
734 protected AbstractBuilder(Class<S> shardClass) {
735 this.shardClass = shardClass;
738 protected void checkSealed() {
739 Preconditions.checkState(!sealed, "Builder isalready sealed - further modifications are not allowed");
742 @SuppressWarnings("unchecked")
747 public T id(ShardIdentifier id) {
753 public T peerAddresses(Map<String, String> peerAddresses) {
755 this.peerAddresses = peerAddresses;
759 public T datastoreContext(DatastoreContext datastoreContext) {
761 this.datastoreContext = datastoreContext;
765 public T schemaContext(SchemaContext schemaContext) {
767 this.schemaContext = schemaContext;
771 public T restoreFromSnapshot(DatastoreSnapshot.ShardSnapshot restoreFromSnapshot) {
773 this.restoreFromSnapshot = restoreFromSnapshot;
777 public ShardIdentifier getId() {
781 public Map<String, String> getPeerAddresses() {
782 return peerAddresses;
785 public DatastoreContext getDatastoreContext() {
786 return datastoreContext;
789 public SchemaContext getSchemaContext() {
790 return schemaContext;
793 public DatastoreSnapshot.ShardSnapshot getRestoreFromSnapshot() {
794 return restoreFromSnapshot;
797 protected void verify() {
798 Preconditions.checkNotNull(id, "id should not be null");
799 Preconditions.checkNotNull(peerAddresses, "peerAddresses should not be null");
800 Preconditions.checkNotNull(datastoreContext, "dataStoreContext should not be null");
801 Preconditions.checkNotNull(schemaContext, "schemaContext should not be null");
804 public Props props() {
807 return Props.create(shardClass, this);
811 public static class Builder extends AbstractBuilder<Builder, Shard> {