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.data.api.schema.tree.TreeType;
74 import org.opendaylight.yangtools.yang.model.api.SchemaContext;
75 import scala.concurrent.duration.Duration;
76 import scala.concurrent.duration.FiniteDuration;
79 * A Shard represents a portion of the logical data tree <br/>
81 * Our Shard uses InMemoryDataTree as it's internal representation and delegates all requests it
84 public class Shard extends RaftActor {
86 protected static final Object TX_COMMIT_TIMEOUT_CHECK_MESSAGE = "txCommitTimeoutCheck";
89 static final Object GET_SHARD_MBEAN_MESSAGE = "getShardMBeanMessage";
92 static final String DEFAULT_NAME = "default";
94 // The state of this Shard
95 private final ShardDataTree store;
97 /// The name of this shard
98 private final String name;
100 private final ShardStats shardMBean;
102 private DatastoreContext datastoreContext;
104 private final ShardCommitCoordinator commitCoordinator;
106 private long transactionCommitTimeout;
108 private Cancellable txCommitTimeoutCheckSchedule;
110 private final Optional<ActorRef> roleChangeNotifier;
112 private final MessageTracker appendEntriesReplyTracker;
114 private final ShardTransactionActorFactory transactionActorFactory;
116 private final ShardSnapshotCohort snapshotCohort;
118 private final DataTreeChangeListenerSupport treeChangeSupport = new DataTreeChangeListenerSupport(this);
119 private final DataChangeListenerSupport changeSupport = new DataChangeListenerSupport(this);
122 private ShardSnapshot restoreFromSnapshot;
126 protected Shard(AbstractBuilder<?, ?> builder) {
127 super(builder.getId().toString(), builder.getPeerAddresses(),
128 Optional.of(builder.getDatastoreContext().getShardRaftConfig()), DataStoreVersions.CURRENT_VERSION);
130 this.name = builder.getId().toString();
131 this.datastoreContext = builder.getDatastoreContext();
132 this.restoreFromSnapshot = builder.getRestoreFromSnapshot();
134 setPersistence(datastoreContext.isPersistent());
136 LOG.info("Shard created : {}, persistent : {}", name, datastoreContext.isPersistent());
138 store = new ShardDataTree(builder.getSchemaContext(), builder.getTreeType());
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 handleForwardedReadyTransaction((ForwardedReadyTransaction) message);
233 } else if (message instanceof ReadyLocalTransaction) {
234 handleReadyLocalTransaction((ReadyLocalTransaction)message);
235 } else if (CanCommitTransaction.SERIALIZABLE_CLASS.isInstance(message)) {
236 handleCanCommitTransaction(CanCommitTransaction.fromSerializable(message));
237 } else if (CommitTransaction.SERIALIZABLE_CLASS.isInstance(message)) {
238 handleCommitTransaction(CommitTransaction.fromSerializable(message));
239 } else if (AbortTransaction.SERIALIZABLE_CLASS.isInstance(message)) {
240 handleAbortTransaction(AbortTransaction.fromSerializable(message));
241 } else if (CloseTransactionChain.SERIALIZABLE_CLASS.isInstance(message)) {
242 closeTransactionChain(CloseTransactionChain.fromSerializable(message));
243 } else if (message instanceof RegisterChangeListener) {
244 changeSupport.onMessage((RegisterChangeListener) message, isLeader(), hasLeader());
245 } else if (message instanceof RegisterDataTreeChangeListener) {
246 treeChangeSupport.onMessage((RegisterDataTreeChangeListener) message, isLeader(), hasLeader());
247 } else if (message instanceof UpdateSchemaContext) {
248 updateSchemaContext((UpdateSchemaContext) message);
249 } else if (message instanceof PeerAddressResolved) {
250 PeerAddressResolved resolved = (PeerAddressResolved) message;
251 setPeerAddress(resolved.getPeerId().toString(),
252 resolved.getPeerAddress());
253 } else if (message.equals(TX_COMMIT_TIMEOUT_CHECK_MESSAGE)) {
254 handleTransactionCommitTimeoutCheck();
255 } else if(message instanceof DatastoreContext) {
256 onDatastoreContext((DatastoreContext)message);
257 } else if(message instanceof RegisterRoleChangeListener){
258 roleChangeNotifier.get().forward(message, context());
259 } else if (message instanceof FollowerInitialSyncUpStatus) {
260 shardMBean.setFollowerInitialSyncStatus(((FollowerInitialSyncUpStatus) message).isInitialSyncDone());
261 context().parent().tell(message, self());
262 } else if(GET_SHARD_MBEAN_MESSAGE.equals(message)){
263 sender().tell(getShardMBean(), self());
264 } else if(message instanceof GetShardDataTree) {
265 sender().tell(store.getDataTree(), self());
266 } else if(message instanceof ServerRemoved){
267 context().parent().forward(message, context());
269 super.onReceiveCommand(message);
276 private boolean hasLeader() {
277 return getLeaderId() != null;
280 public int getPendingTxCommitQueueSize() {
281 return commitCoordinator.getQueueSize();
285 protected Optional<ActorRef> getRoleChangeNotifier() {
286 return roleChangeNotifier;
290 protected LeaderStateChanged newLeaderStateChanged(String memberId, String leaderId, short leaderPayloadVersion) {
291 return new ShardLeaderStateChanged(memberId, leaderId,
292 isLeader() ? Optional.<DataTree>of(store.getDataTree()) : Optional.<DataTree>absent(),
293 leaderPayloadVersion);
296 protected void onDatastoreContext(DatastoreContext context) {
297 datastoreContext = context;
299 commitCoordinator.setQueueCapacity(datastoreContext.getShardTransactionCommitQueueCapacity());
301 setTransactionCommitTimeout();
303 if(datastoreContext.isPersistent() && !persistence().isRecoveryApplicable()) {
304 setPersistence(true);
305 } else if(!datastoreContext.isPersistent() && persistence().isRecoveryApplicable()) {
306 setPersistence(false);
309 updateConfigParams(datastoreContext.getShardRaftConfig());
312 private void handleTransactionCommitTimeoutCheck() {
313 CohortEntry cohortEntry = commitCoordinator.getCurrentCohortEntry();
314 if(cohortEntry != null) {
315 if(cohortEntry.isExpired(transactionCommitTimeout)) {
316 LOG.warn("{}: Current transaction {} has timed out after {} ms - aborting",
317 persistenceId(), cohortEntry.getTransactionID(), transactionCommitTimeout);
319 doAbortTransaction(cohortEntry.getTransactionID(), null);
323 commitCoordinator.cleanupExpiredCohortEntries();
326 private static boolean isEmptyCommit(final DataTreeCandidate candidate) {
327 return ModificationType.UNMODIFIED.equals(candidate.getRootNode().getModificationType());
330 void continueCommit(final CohortEntry cohortEntry) {
331 final DataTreeCandidate candidate = cohortEntry.getCandidate();
333 // If we do not have any followers and we are not using persistence
334 // or if cohortEntry has no modifications
335 // we can apply modification to the state immediately
336 if ((!hasFollowers() && !persistence().isRecoveryApplicable()) || isEmptyCommit(candidate)) {
337 applyModificationToState(cohortEntry.getReplySender(), cohortEntry.getTransactionID(), candidate);
339 Shard.this.persistData(cohortEntry.getReplySender(), cohortEntry.getTransactionID(),
340 DataTreeCandidatePayload.create(candidate));
344 private void handleCommitTransaction(final CommitTransaction commit) {
345 if(!commitCoordinator.handleCommit(commit.getTransactionID(), getSender(), this)) {
346 shardMBean.incrementFailedTransactionsCount();
350 private void finishCommit(@Nonnull final ActorRef sender, @Nonnull final String transactionID, @Nonnull final CohortEntry cohortEntry) {
351 LOG.debug("{}: Finishing commit for transaction {}", persistenceId(), cohortEntry.getTransactionID());
354 cohortEntry.commit();
356 sender.tell(CommitTransactionReply.INSTANCE.toSerializable(), getSelf());
358 shardMBean.incrementCommittedTransactionCount();
359 shardMBean.setLastCommittedTransactionTime(System.currentTimeMillis());
361 } catch (Exception e) {
362 sender.tell(new akka.actor.Status.Failure(e), getSelf());
364 LOG.error("{}, An exception occurred while committing transaction {}", persistenceId(),
366 shardMBean.incrementFailedTransactionsCount();
368 commitCoordinator.currentTransactionComplete(transactionID, true);
372 private void finishCommit(@Nonnull final ActorRef sender, final @Nonnull String transactionID) {
373 // With persistence enabled, this method is called via applyState by the leader strategy
374 // after the commit has been replicated to a majority of the followers.
376 CohortEntry cohortEntry = commitCoordinator.getCohortEntryIfCurrent(transactionID);
377 if (cohortEntry == null) {
378 // The transaction is no longer the current commit. This can happen if the transaction
379 // was aborted prior, most likely due to timeout in the front-end. We need to finish
380 // committing the transaction though since it was successfully persisted and replicated
381 // however we can't use the original cohort b/c it was already preCommitted and may
382 // conflict with the current commit or may have been aborted so we commit with a new
384 cohortEntry = commitCoordinator.getAndRemoveCohortEntry(transactionID);
385 if(cohortEntry != null) {
387 store.applyForeignCandidate(transactionID, cohortEntry.getCandidate());
388 } catch (DataValidationFailedException e) {
389 shardMBean.incrementFailedTransactionsCount();
390 LOG.error("{}: Failed to re-apply transaction {}", persistenceId(), transactionID, e);
393 sender.tell(CommitTransactionReply.INSTANCE.toSerializable(), getSelf());
395 // This really shouldn't happen - it likely means that persistence or replication
396 // took so long to complete such that the cohort entry was expired from the cache.
397 IllegalStateException ex = new IllegalStateException(
398 String.format("%s: Could not finish committing transaction %s - no CohortEntry found",
399 persistenceId(), transactionID));
400 LOG.error(ex.getMessage());
401 sender.tell(new akka.actor.Status.Failure(ex), getSelf());
404 finishCommit(sender, transactionID, cohortEntry);
408 private void handleCanCommitTransaction(final CanCommitTransaction canCommit) {
409 LOG.debug("{}: Can committing transaction {}", persistenceId(), canCommit.getTransactionID());
410 commitCoordinator.handleCanCommit(canCommit.getTransactionID(), getSender(), this);
413 private void noLeaderError(String errMessage) {
414 // TODO: rather than throwing an immediate exception, we could schedule a timer to try again to make
415 // it more resilient in case we're in the process of electing a new leader.
416 getSender().tell(new akka.actor.Status.Failure(new NoShardLeaderException(errMessage, persistenceId())), getSelf());
419 protected void handleBatchedModificationsLocal(BatchedModifications batched, ActorRef sender) {
421 commitCoordinator.handleBatchedModifications(batched, sender, this);
422 } catch (Exception e) {
423 LOG.error("{}: Error handling BatchedModifications for Tx {}", persistenceId(),
424 batched.getTransactionID(), e);
425 sender.tell(new akka.actor.Status.Failure(e), getSelf());
429 private void handleBatchedModifications(BatchedModifications batched) {
430 // This message is sent to prepare the modifications transaction directly on the Shard as an
431 // optimization to avoid the extra overhead of a separate ShardTransaction actor. On the last
432 // BatchedModifications message, the caller sets the ready flag in the message indicating
433 // modifications are complete. The reply contains the cohort actor path (this actor) for the caller
434 // to initiate the 3-phase commit. This also avoids the overhead of sending an additional
435 // ReadyTransaction message.
437 // If we're not the leader then forward to the leader. This is a safety measure - we shouldn't
438 // normally get here if we're not the leader as the front-end (TransactionProxy) should determine
439 // the primary/leader shard. However with timing and caching on the front-end, there's a small
440 // window where it could have a stale leader during leadership transitions.
443 failIfIsolatedLeader(getSender());
445 handleBatchedModificationsLocal(batched, getSender());
447 ActorSelection leader = getLeader();
449 // TODO: what if this is not the first batch and leadership changed in between batched messages?
450 // We could check if the commitCoordinator already has a cached entry and forward all the previous
451 // batched modifications.
452 LOG.debug("{}: Forwarding BatchedModifications to leader {}", persistenceId(), leader);
453 leader.forward(batched, getContext());
455 noLeaderError("Could not commit transaction " + batched.getTransactionID());
460 private boolean failIfIsolatedLeader(ActorRef sender) {
461 if(isIsolatedLeader()) {
462 sender.tell(new akka.actor.Status.Failure(new NoShardLeaderException(String.format(
463 "Shard %s was the leader but has lost contact with all of its followers. Either all" +
464 " other follower nodes are down or this node is isolated by a network partition.",
465 persistenceId()))), getSelf());
472 protected boolean isIsolatedLeader() {
473 return getRaftState() == RaftState.IsolatedLeader;
476 private void handleReadyLocalTransaction(final ReadyLocalTransaction message) {
478 failIfIsolatedLeader(getSender());
481 commitCoordinator.handleReadyLocalTransaction(message, getSender(), this);
482 } catch (Exception e) {
483 LOG.error("{}: Error handling ReadyLocalTransaction for Tx {}", persistenceId(),
484 message.getTransactionID(), e);
485 getSender().tell(new akka.actor.Status.Failure(e), getSelf());
488 ActorSelection leader = getLeader();
489 if (leader != null) {
490 LOG.debug("{}: Forwarding ReadyLocalTransaction to leader {}", persistenceId(), leader);
491 message.setRemoteVersion(getCurrentBehavior().getLeaderPayloadVersion());
492 leader.forward(message, getContext());
494 noLeaderError("Could not commit transaction " + message.getTransactionID());
499 private void handleForwardedReadyTransaction(ForwardedReadyTransaction forwardedReady) {
501 failIfIsolatedLeader(getSender());
503 commitCoordinator.handleForwardedReadyTransaction(forwardedReady, getSender(), this);
505 ActorSelection leader = getLeader();
506 if (leader != null) {
507 LOG.debug("{}: Forwarding ForwardedReadyTransaction to leader {}", persistenceId(), leader);
509 ReadyLocalTransaction readyLocal = new ReadyLocalTransaction(forwardedReady.getTransactionID(),
510 forwardedReady.getTransaction().getSnapshot(), forwardedReady.isDoImmediateCommit());
511 readyLocal.setRemoteVersion(getCurrentBehavior().getLeaderPayloadVersion());
512 leader.forward(readyLocal, getContext());
514 noLeaderError("Could not commit transaction " + forwardedReady.getTransactionID());
519 private void handleAbortTransaction(final AbortTransaction abort) {
520 doAbortTransaction(abort.getTransactionID(), getSender());
523 void doAbortTransaction(final String transactionID, final ActorRef sender) {
524 commitCoordinator.handleAbort(transactionID, sender, this);
527 private void handleCreateTransaction(final Object message) {
529 createTransaction(CreateTransaction.fromSerializable(message));
530 } else if (getLeader() != null) {
531 getLeader().forward(message, getContext());
533 getSender().tell(new akka.actor.Status.Failure(new NoShardLeaderException(
534 "Could not create a shard transaction", persistenceId())), getSelf());
538 private void closeTransactionChain(final CloseTransactionChain closeTransactionChain) {
539 store.closeTransactionChain(closeTransactionChain.getTransactionChainId());
542 private ActorRef createTypedTransactionActor(int transactionType,
543 ShardTransactionIdentifier transactionId, String transactionChainId,
544 short clientVersion ) {
546 return transactionActorFactory.newShardTransaction(TransactionType.fromInt(transactionType),
547 transactionId, transactionChainId, clientVersion);
550 private void createTransaction(CreateTransaction createTransaction) {
552 if(TransactionType.fromInt(createTransaction.getTransactionType()) != TransactionType.READ_ONLY &&
553 failIfIsolatedLeader(getSender())) {
557 ActorRef transactionActor = createTransaction(createTransaction.getTransactionType(),
558 createTransaction.getTransactionId(), createTransaction.getTransactionChainId(),
559 createTransaction.getVersion());
561 getSender().tell(new CreateTransactionReply(Serialization.serializedActorPath(transactionActor),
562 createTransaction.getTransactionId()).toSerializable(), getSelf());
563 } catch (Exception e) {
564 getSender().tell(new akka.actor.Status.Failure(e), getSelf());
568 private ActorRef createTransaction(int transactionType, String remoteTransactionId,
569 String transactionChainId, short clientVersion) {
572 ShardTransactionIdentifier transactionId = new ShardTransactionIdentifier(remoteTransactionId);
574 if(LOG.isDebugEnabled()) {
575 LOG.debug("{}: Creating transaction : {} ", persistenceId(), transactionId);
578 ActorRef transactionActor = createTypedTransactionActor(transactionType, transactionId,
579 transactionChainId, clientVersion);
581 return transactionActor;
584 private void commitWithNewTransaction(final Modification modification) {
585 ReadWriteShardDataTreeTransaction tx = store.newReadWriteTransaction(modification.toString(), null);
586 modification.apply(tx.getSnapshot());
588 snapshotCohort.syncCommitTransaction(tx);
589 shardMBean.incrementCommittedTransactionCount();
590 shardMBean.setLastCommittedTransactionTime(System.currentTimeMillis());
591 } catch (Exception e) {
592 shardMBean.incrementFailedTransactionsCount();
593 LOG.error("{}: Failed to commit", persistenceId(), e);
597 private void updateSchemaContext(final UpdateSchemaContext message) {
598 updateSchemaContext(message.getSchemaContext());
602 void updateSchemaContext(final SchemaContext schemaContext) {
603 store.updateSchemaContext(schemaContext);
606 private boolean isMetricsCaptureEnabled() {
607 CommonConfig config = new CommonConfig(getContext().system().settings().config());
608 return config.isMetricCaptureEnabled();
613 public RaftActorSnapshotCohort getRaftActorSnapshotCohort() {
614 return snapshotCohort;
619 protected RaftActorRecoveryCohort getRaftActorRecoveryCohort() {
620 return new ShardRecoveryCoordinator(store, store.getSchemaContext(),
621 restoreFromSnapshot != null ? restoreFromSnapshot.getSnapshot() : null, persistenceId(), LOG);
625 protected void onRecoveryComplete() {
626 restoreFromSnapshot = null;
628 //notify shard manager
629 getContext().parent().tell(new ActorInitialized(), getSelf());
631 // Being paranoid here - this method should only be called once but just in case...
632 if(txCommitTimeoutCheckSchedule == null) {
633 // Schedule a message to be periodically sent to check if the current in-progress
634 // transaction should be expired and aborted.
635 FiniteDuration period = Duration.create(transactionCommitTimeout / 3, TimeUnit.MILLISECONDS);
636 txCommitTimeoutCheckSchedule = getContext().system().scheduler().schedule(
637 period, period, getSelf(),
638 TX_COMMIT_TIMEOUT_CHECK_MESSAGE, getContext().dispatcher(), ActorRef.noSender());
643 protected void applyState(final ActorRef clientActor, final String identifier, final Object data) {
644 if (data instanceof DataTreeCandidatePayload) {
645 if (clientActor == null) {
646 // No clientActor indicates a replica coming from the leader
648 store.applyForeignCandidate(identifier, ((DataTreeCandidatePayload)data).getCandidate());
649 } catch (DataValidationFailedException | IOException e) {
650 LOG.error("{}: Error applying replica {}", persistenceId(), identifier, e);
653 // Replication consensus reached, proceed to commit
654 finishCommit(clientActor, identifier);
656 } else if (data instanceof ModificationPayload) {
658 applyModificationToState(clientActor, identifier, ((ModificationPayload) data).getModification());
659 } catch (ClassNotFoundException | IOException e) {
660 LOG.error("{}: Error extracting ModificationPayload", persistenceId(), e);
662 } else if (data instanceof CompositeModificationPayload) {
663 Object modification = ((CompositeModificationPayload) data).getModification();
665 applyModificationToState(clientActor, identifier, modification);
666 } else if(data instanceof CompositeModificationByteStringPayload ){
667 Object modification = ((CompositeModificationByteStringPayload) data).getModification();
669 applyModificationToState(clientActor, identifier, modification);
671 LOG.error("{}: Unknown state received {} Class loader = {} CompositeNodeMod.ClassLoader = {}",
672 persistenceId(), data, data.getClass().getClassLoader(),
673 CompositeModificationPayload.class.getClassLoader());
677 private void applyModificationToState(ActorRef clientActor, String identifier, Object modification) {
678 if(modification == null) {
680 "{}: modification is null - this is very unexpected, clientActor = {}, identifier = {}",
681 persistenceId(), identifier, clientActor != null ? clientActor.path().toString() : null);
682 } else if(clientActor == null) {
683 // There's no clientActor to which to send a commit reply so we must be applying
684 // replicated state from the leader.
685 commitWithNewTransaction(MutableCompositeModification.fromSerializable(modification));
687 // This must be the OK to commit after replication consensus.
688 finishCommit(clientActor, identifier);
693 protected void onStateChanged() {
694 boolean isLeader = isLeader();
695 boolean hasLeader = hasLeader();
696 changeSupport.onLeadershipChange(isLeader, hasLeader);
697 treeChangeSupport.onLeadershipChange(isLeader, hasLeader);
699 // If this actor is no longer the leader close all the transaction chains
701 if(LOG.isDebugEnabled()) {
703 "{}: onStateChanged: Closing all transaction chains because shard {} is no longer the leader",
704 persistenceId(), getId());
707 store.closeAllTransactionChains();
712 protected void onLeaderChanged(String oldLeader, String newLeader) {
713 shardMBean.incrementLeadershipChangeCount();
717 public String persistenceId() {
722 ShardCommitCoordinator getCommitCoordinator() {
723 return commitCoordinator;
726 public DatastoreContext getDatastoreContext() {
727 return datastoreContext;
731 public ShardDataTree getDataStore() {
736 ShardStats getShardMBean() {
740 public static Builder builder() {
741 return new Builder();
744 public static abstract class AbstractBuilder<T extends AbstractBuilder<T, S>, S extends Shard> {
745 private final Class<S> shardClass;
746 private ShardIdentifier id;
747 private Map<String, String> peerAddresses = Collections.emptyMap();
748 private DatastoreContext datastoreContext;
749 private SchemaContext schemaContext;
750 private DatastoreSnapshot.ShardSnapshot restoreFromSnapshot;
751 private volatile boolean sealed;
753 protected AbstractBuilder(Class<S> shardClass) {
754 this.shardClass = shardClass;
757 protected void checkSealed() {
758 Preconditions.checkState(!sealed, "Builder isalready sealed - further modifications are not allowed");
761 @SuppressWarnings("unchecked")
766 public T id(ShardIdentifier id) {
772 public T peerAddresses(Map<String, String> peerAddresses) {
774 this.peerAddresses = peerAddresses;
778 public T datastoreContext(DatastoreContext datastoreContext) {
780 this.datastoreContext = datastoreContext;
784 public T schemaContext(SchemaContext schemaContext) {
786 this.schemaContext = schemaContext;
790 public T restoreFromSnapshot(DatastoreSnapshot.ShardSnapshot restoreFromSnapshot) {
792 this.restoreFromSnapshot = restoreFromSnapshot;
796 public ShardIdentifier getId() {
800 public Map<String, String> getPeerAddresses() {
801 return peerAddresses;
804 public DatastoreContext getDatastoreContext() {
805 return datastoreContext;
808 public SchemaContext getSchemaContext() {
809 return schemaContext;
812 public DatastoreSnapshot.ShardSnapshot getRestoreFromSnapshot() {
813 return restoreFromSnapshot;
816 public TreeType getTreeType() {
817 switch (datastoreContext.getLogicalStoreType()) {
819 return TreeType.CONFIGURATION;
821 return TreeType.OPERATIONAL;
824 throw new IllegalStateException("Unhandled logical store type " + datastoreContext.getLogicalStoreType());
827 protected void verify() {
828 Preconditions.checkNotNull(id, "id should not be null");
829 Preconditions.checkNotNull(peerAddresses, "peerAddresses should not be null");
830 Preconditions.checkNotNull(datastoreContext, "dataStoreContext should not be null");
831 Preconditions.checkNotNull(schemaContext, "schemaContext should not be null");
834 public Props props() {
837 return Props.create(shardClass, this);
841 public static class Builder extends AbstractBuilder<Builder, Shard> {