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;
124 private final ShardTransactionMessageRetrySupport messageRetrySupport;
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);
166 messageRetrySupport = new ShardTransactionMessageRetrySupport(this);
169 private void setTransactionCommitTimeout() {
170 transactionCommitTimeout = TimeUnit.MILLISECONDS.convert(
171 datastoreContext.getShardTransactionCommitTimeoutInSeconds(), TimeUnit.SECONDS) / 2;
174 private Optional<ActorRef> createRoleChangeNotifier(String shardId) {
175 ActorRef shardRoleChangeNotifier = this.getContext().actorOf(
176 RoleChangeNotifier.getProps(shardId), shardId + "-notifier");
177 return Optional.of(shardRoleChangeNotifier);
181 public void postStop() {
182 LOG.info("Stopping Shard {}", persistenceId());
186 messageRetrySupport.close();
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 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());
269 } else if(ShardTransactionMessageRetrySupport.TIMER_MESSAGE_CLASS.isInstance(message)) {
270 messageRetrySupport.onTimerMessage(message);
272 super.onReceiveCommand(message);
279 private boolean hasLeader() {
280 return getLeaderId() != null;
283 public int getPendingTxCommitQueueSize() {
284 return commitCoordinator.getQueueSize();
288 protected Optional<ActorRef> getRoleChangeNotifier() {
289 return roleChangeNotifier;
293 protected LeaderStateChanged newLeaderStateChanged(String memberId, String leaderId, short leaderPayloadVersion) {
294 return new ShardLeaderStateChanged(memberId, leaderId,
295 isLeader() ? Optional.<DataTree>of(store.getDataTree()) : Optional.<DataTree>absent(),
296 leaderPayloadVersion);
299 protected void onDatastoreContext(DatastoreContext context) {
300 datastoreContext = context;
302 commitCoordinator.setQueueCapacity(datastoreContext.getShardTransactionCommitQueueCapacity());
304 setTransactionCommitTimeout();
306 if(datastoreContext.isPersistent() && !persistence().isRecoveryApplicable()) {
307 setPersistence(true);
308 } else if(!datastoreContext.isPersistent() && persistence().isRecoveryApplicable()) {
309 setPersistence(false);
312 updateConfigParams(datastoreContext.getShardRaftConfig());
315 private void handleTransactionCommitTimeoutCheck() {
316 CohortEntry cohortEntry = commitCoordinator.getCurrentCohortEntry();
317 if(cohortEntry != null) {
318 if(cohortEntry.isExpired(transactionCommitTimeout)) {
319 LOG.warn("{}: Current transaction {} has timed out after {} ms - aborting",
320 persistenceId(), cohortEntry.getTransactionID(), transactionCommitTimeout);
322 doAbortTransaction(cohortEntry.getTransactionID(), null);
326 commitCoordinator.cleanupExpiredCohortEntries();
329 private static boolean isEmptyCommit(final DataTreeCandidate candidate) {
330 return ModificationType.UNMODIFIED.equals(candidate.getRootNode().getModificationType());
333 void continueCommit(final CohortEntry cohortEntry) {
334 final DataTreeCandidate candidate = cohortEntry.getCandidate();
336 // If we do not have any followers and we are not using persistence
337 // or if cohortEntry has no modifications
338 // we can apply modification to the state immediately
339 if ((!hasFollowers() && !persistence().isRecoveryApplicable()) || isEmptyCommit(candidate)) {
340 applyModificationToState(cohortEntry.getReplySender(), cohortEntry.getTransactionID(), candidate);
342 Shard.this.persistData(cohortEntry.getReplySender(), cohortEntry.getTransactionID(),
343 DataTreeCandidatePayload.create(candidate));
347 private void handleCommitTransaction(final CommitTransaction commit) {
348 if(!commitCoordinator.handleCommit(commit.getTransactionID(), getSender(), this)) {
349 shardMBean.incrementFailedTransactionsCount();
353 private void finishCommit(@Nonnull final ActorRef sender, @Nonnull final String transactionID, @Nonnull final CohortEntry cohortEntry) {
354 LOG.debug("{}: Finishing commit for transaction {}", persistenceId(), cohortEntry.getTransactionID());
357 cohortEntry.commit();
359 sender.tell(CommitTransactionReply.INSTANCE.toSerializable(), getSelf());
361 shardMBean.incrementCommittedTransactionCount();
362 shardMBean.setLastCommittedTransactionTime(System.currentTimeMillis());
364 } catch (Exception e) {
365 sender.tell(new akka.actor.Status.Failure(e), getSelf());
367 LOG.error("{}, An exception occurred while committing transaction {}", persistenceId(),
369 shardMBean.incrementFailedTransactionsCount();
371 commitCoordinator.currentTransactionComplete(transactionID, true);
375 private void finishCommit(@Nonnull final ActorRef sender, final @Nonnull String transactionID) {
376 // With persistence enabled, this method is called via applyState by the leader strategy
377 // after the commit has been replicated to a majority of the followers.
379 CohortEntry cohortEntry = commitCoordinator.getCohortEntryIfCurrent(transactionID);
380 if (cohortEntry == null) {
381 // The transaction is no longer the current commit. This can happen if the transaction
382 // was aborted prior, most likely due to timeout in the front-end. We need to finish
383 // committing the transaction though since it was successfully persisted and replicated
384 // however we can't use the original cohort b/c it was already preCommitted and may
385 // conflict with the current commit or may have been aborted so we commit with a new
387 cohortEntry = commitCoordinator.getAndRemoveCohortEntry(transactionID);
388 if(cohortEntry != null) {
390 store.applyForeignCandidate(transactionID, cohortEntry.getCandidate());
391 } catch (DataValidationFailedException e) {
392 shardMBean.incrementFailedTransactionsCount();
393 LOG.error("{}: Failed to re-apply transaction {}", persistenceId(), transactionID, e);
396 sender.tell(CommitTransactionReply.INSTANCE.toSerializable(), getSelf());
398 // This really shouldn't happen - it likely means that persistence or replication
399 // took so long to complete such that the cohort entry was expired from the cache.
400 IllegalStateException ex = new IllegalStateException(
401 String.format("%s: Could not finish committing transaction %s - no CohortEntry found",
402 persistenceId(), transactionID));
403 LOG.error(ex.getMessage());
404 sender.tell(new akka.actor.Status.Failure(ex), getSelf());
407 finishCommit(sender, transactionID, cohortEntry);
411 private void handleCanCommitTransaction(final CanCommitTransaction canCommit) {
412 LOG.debug("{}: Can committing transaction {}", persistenceId(), canCommit.getTransactionID());
413 commitCoordinator.handleCanCommit(canCommit.getTransactionID(), getSender(), this);
416 protected void handleBatchedModificationsLocal(BatchedModifications batched, ActorRef sender) {
418 commitCoordinator.handleBatchedModifications(batched, sender, this);
419 } catch (Exception e) {
420 LOG.error("{}: Error handling BatchedModifications for Tx {}", persistenceId(),
421 batched.getTransactionID(), e);
422 sender.tell(new akka.actor.Status.Failure(e), getSelf());
426 private void handleBatchedModifications(BatchedModifications batched) {
427 // This message is sent to prepare the modifications transaction directly on the Shard as an
428 // optimization to avoid the extra overhead of a separate ShardTransaction actor. On the last
429 // BatchedModifications message, the caller sets the ready flag in the message indicating
430 // modifications are complete. The reply contains the cohort actor path (this actor) for the caller
431 // to initiate the 3-phase commit. This also avoids the overhead of sending an additional
432 // ReadyTransaction message.
434 // If we're not the leader then forward to the leader. This is a safety measure - we shouldn't
435 // normally get here if we're not the leader as the front-end (TransactionProxy) should determine
436 // the primary/leader shard. However with timing and caching on the front-end, there's a small
437 // window where it could have a stale leader during leadership transitions.
439 boolean isIsolatedLeader = isIsolatedLeader();
440 if (isLeader() && !isIsolatedLeader) {
441 handleBatchedModificationsLocal(batched, getSender());
443 ActorSelection leader = getLeader();
444 if (isIsolatedLeader || leader == null) {
445 messageRetrySupport.addMessageToRetry(batched, getSender(),
446 "Could not commit transaction " + batched.getTransactionID());
448 // TODO: what if this is not the first batch and leadership changed in between batched messages?
449 // We could check if the commitCoordinator already has a cached entry and forward all the previous
450 // batched modifications.
451 LOG.debug("{}: Forwarding BatchedModifications to leader {}", persistenceId(), leader);
452 leader.forward(batched, getContext());
457 private boolean failIfIsolatedLeader(ActorRef sender) {
458 if(isIsolatedLeader()) {
459 sender.tell(new akka.actor.Status.Failure(new NoShardLeaderException(String.format(
460 "Shard %s was the leader but has lost contact with all of its followers. Either all" +
461 " other follower nodes are down or this node is isolated by a network partition.",
462 persistenceId()))), getSelf());
469 protected boolean isIsolatedLeader() {
470 return getRaftState() == RaftState.IsolatedLeader;
473 private void handleReadyLocalTransaction(final ReadyLocalTransaction message) {
474 LOG.debug("{}: handleReadyLocalTransaction for {}", persistenceId(), message.getTransactionID());
476 boolean isIsolatedLeader = isIsolatedLeader();
477 if (isLeader() && !isIsolatedLeader) {
479 commitCoordinator.handleReadyLocalTransaction(message, getSender(), this);
480 } catch (Exception e) {
481 LOG.error("{}: Error handling ReadyLocalTransaction for Tx {}", persistenceId(),
482 message.getTransactionID(), e);
483 getSender().tell(new akka.actor.Status.Failure(e), getSelf());
486 ActorSelection leader = getLeader();
487 if (isIsolatedLeader || leader == null) {
488 messageRetrySupport.addMessageToRetry(message, getSender(),
489 "Could not commit transaction " + message.getTransactionID());
491 LOG.debug("{}: Forwarding ReadyLocalTransaction to leader {}", persistenceId(), leader);
492 message.setRemoteVersion(getCurrentBehavior().getLeaderPayloadVersion());
493 leader.forward(message, getContext());
498 private void handleForwardedReadyTransaction(ForwardedReadyTransaction forwardedReady) {
499 LOG.debug("{}: handleForwardedReadyTransaction for {}", persistenceId(), forwardedReady.getTransactionID());
501 boolean isIsolatedLeader = isIsolatedLeader();
502 if (isLeader() && !isIsolatedLeader) {
503 commitCoordinator.handleForwardedReadyTransaction(forwardedReady, getSender(), this);
505 ActorSelection leader = getLeader();
506 if (isIsolatedLeader || leader == null) {
507 messageRetrySupport.addMessageToRetry(forwardedReady, getSender(),
508 "Could not commit transaction " + forwardedReady.getTransactionID());
510 LOG.debug("{}: Forwarding ForwardedReadyTransaction to leader {}", persistenceId(), leader);
512 ReadyLocalTransaction readyLocal = new ReadyLocalTransaction(forwardedReady.getTransactionID(),
513 forwardedReady.getTransaction().getSnapshot(), forwardedReady.isDoImmediateCommit());
514 readyLocal.setRemoteVersion(getCurrentBehavior().getLeaderPayloadVersion());
515 leader.forward(readyLocal, getContext());
520 private void handleAbortTransaction(final AbortTransaction abort) {
521 doAbortTransaction(abort.getTransactionID(), getSender());
524 void doAbortTransaction(final String transactionID, final ActorRef sender) {
525 commitCoordinator.handleAbort(transactionID, sender, this);
528 private void handleCreateTransaction(final Object message) {
530 createTransaction(CreateTransaction.fromSerializable(message));
531 } else if (getLeader() != null) {
532 getLeader().forward(message, getContext());
534 getSender().tell(new akka.actor.Status.Failure(new NoShardLeaderException(
535 "Could not create a shard transaction", persistenceId())), getSelf());
539 private void closeTransactionChain(final CloseTransactionChain closeTransactionChain) {
540 store.closeTransactionChain(closeTransactionChain.getTransactionChainId());
543 private ActorRef createTypedTransactionActor(int transactionType,
544 ShardTransactionIdentifier transactionId, String transactionChainId,
545 short clientVersion ) {
547 return transactionActorFactory.newShardTransaction(TransactionType.fromInt(transactionType),
548 transactionId, transactionChainId, clientVersion);
551 private void createTransaction(CreateTransaction createTransaction) {
553 if(TransactionType.fromInt(createTransaction.getTransactionType()) != TransactionType.READ_ONLY &&
554 failIfIsolatedLeader(getSender())) {
558 ActorRef transactionActor = createTransaction(createTransaction.getTransactionType(),
559 createTransaction.getTransactionId(), createTransaction.getTransactionChainId(),
560 createTransaction.getVersion());
562 getSender().tell(new CreateTransactionReply(Serialization.serializedActorPath(transactionActor),
563 createTransaction.getTransactionId()).toSerializable(), getSelf());
564 } catch (Exception e) {
565 getSender().tell(new akka.actor.Status.Failure(e), getSelf());
569 private ActorRef createTransaction(int transactionType, String remoteTransactionId,
570 String transactionChainId, short clientVersion) {
573 ShardTransactionIdentifier transactionId = new ShardTransactionIdentifier(remoteTransactionId);
575 if(LOG.isDebugEnabled()) {
576 LOG.debug("{}: Creating transaction : {} ", persistenceId(), transactionId);
579 ActorRef transactionActor = createTypedTransactionActor(transactionType, transactionId,
580 transactionChainId, clientVersion);
582 return transactionActor;
585 private void commitWithNewTransaction(final Modification modification) {
586 ReadWriteShardDataTreeTransaction tx = store.newReadWriteTransaction(modification.toString(), null);
587 modification.apply(tx.getSnapshot());
589 snapshotCohort.syncCommitTransaction(tx);
590 shardMBean.incrementCommittedTransactionCount();
591 shardMBean.setLastCommittedTransactionTime(System.currentTimeMillis());
592 } catch (Exception e) {
593 shardMBean.incrementFailedTransactionsCount();
594 LOG.error("{}: Failed to commit", persistenceId(), e);
598 private void updateSchemaContext(final UpdateSchemaContext message) {
599 updateSchemaContext(message.getSchemaContext());
603 void updateSchemaContext(final SchemaContext schemaContext) {
604 store.updateSchemaContext(schemaContext);
607 private boolean isMetricsCaptureEnabled() {
608 CommonConfig config = new CommonConfig(getContext().system().settings().config());
609 return config.isMetricCaptureEnabled();
614 public RaftActorSnapshotCohort getRaftActorSnapshotCohort() {
615 return snapshotCohort;
620 protected RaftActorRecoveryCohort getRaftActorRecoveryCohort() {
621 return new ShardRecoveryCoordinator(store, store.getSchemaContext(),
622 restoreFromSnapshot != null ? restoreFromSnapshot.getSnapshot() : null, persistenceId(), LOG);
626 protected void onRecoveryComplete() {
627 restoreFromSnapshot = null;
629 //notify shard manager
630 getContext().parent().tell(new ActorInitialized(), getSelf());
632 // Being paranoid here - this method should only be called once but just in case...
633 if(txCommitTimeoutCheckSchedule == null) {
634 // Schedule a message to be periodically sent to check if the current in-progress
635 // transaction should be expired and aborted.
636 FiniteDuration period = Duration.create(transactionCommitTimeout / 3, TimeUnit.MILLISECONDS);
637 txCommitTimeoutCheckSchedule = getContext().system().scheduler().schedule(
638 period, period, getSelf(),
639 TX_COMMIT_TIMEOUT_CHECK_MESSAGE, getContext().dispatcher(), ActorRef.noSender());
644 protected void applyState(final ActorRef clientActor, final String identifier, final Object data) {
645 if (data instanceof DataTreeCandidatePayload) {
646 if (clientActor == null) {
647 // No clientActor indicates a replica coming from the leader
649 store.applyForeignCandidate(identifier, ((DataTreeCandidatePayload)data).getCandidate());
650 } catch (DataValidationFailedException | IOException e) {
651 LOG.error("{}: Error applying replica {}", persistenceId(), identifier, e);
654 // Replication consensus reached, proceed to commit
655 finishCommit(clientActor, identifier);
657 } else if (data instanceof ModificationPayload) {
659 applyModificationToState(clientActor, identifier, ((ModificationPayload) data).getModification());
660 } catch (ClassNotFoundException | IOException e) {
661 LOG.error("{}: Error extracting ModificationPayload", persistenceId(), e);
663 } else if (data instanceof CompositeModificationPayload) {
664 Object modification = ((CompositeModificationPayload) data).getModification();
666 applyModificationToState(clientActor, identifier, modification);
667 } else if(data instanceof CompositeModificationByteStringPayload ){
668 Object modification = ((CompositeModificationByteStringPayload) data).getModification();
670 applyModificationToState(clientActor, identifier, modification);
672 LOG.error("{}: Unknown state received {} Class loader = {} CompositeNodeMod.ClassLoader = {}",
673 persistenceId(), data, data.getClass().getClassLoader(),
674 CompositeModificationPayload.class.getClassLoader());
678 private void applyModificationToState(ActorRef clientActor, String identifier, Object modification) {
679 if(modification == null) {
681 "{}: modification is null - this is very unexpected, clientActor = {}, identifier = {}",
682 persistenceId(), identifier, clientActor != null ? clientActor.path().toString() : null);
683 } else if(clientActor == null) {
684 // There's no clientActor to which to send a commit reply so we must be applying
685 // replicated state from the leader.
686 commitWithNewTransaction(MutableCompositeModification.fromSerializable(modification));
688 // This must be the OK to commit after replication consensus.
689 finishCommit(clientActor, identifier);
694 protected void onStateChanged() {
695 boolean isLeader = isLeader();
696 boolean hasLeader = hasLeader();
697 changeSupport.onLeadershipChange(isLeader, hasLeader);
698 treeChangeSupport.onLeadershipChange(isLeader, hasLeader);
700 // If this actor is no longer the leader close all the transaction chains
702 if(LOG.isDebugEnabled()) {
704 "{}: onStateChanged: Closing all transaction chains because shard {} is no longer the leader",
705 persistenceId(), getId());
708 store.closeAllTransactionChains();
711 if(hasLeader && !isIsolatedLeader()) {
712 messageRetrySupport.retryMessages();
717 protected void onLeaderChanged(String oldLeader, String newLeader) {
718 shardMBean.incrementLeadershipChangeCount();
720 if(hasLeader() && !isIsolatedLeader()) {
721 messageRetrySupport.retryMessages();
726 public String persistenceId() {
731 ShardCommitCoordinator getCommitCoordinator() {
732 return commitCoordinator;
735 public DatastoreContext getDatastoreContext() {
736 return datastoreContext;
740 public ShardDataTree getDataStore() {
745 ShardStats getShardMBean() {
749 public static Builder builder() {
750 return new Builder();
753 public static abstract class AbstractBuilder<T extends AbstractBuilder<T, S>, S extends Shard> {
754 private final Class<S> shardClass;
755 private ShardIdentifier id;
756 private Map<String, String> peerAddresses = Collections.emptyMap();
757 private DatastoreContext datastoreContext;
758 private SchemaContext schemaContext;
759 private DatastoreSnapshot.ShardSnapshot restoreFromSnapshot;
760 private volatile boolean sealed;
762 protected AbstractBuilder(Class<S> shardClass) {
763 this.shardClass = shardClass;
766 protected void checkSealed() {
767 Preconditions.checkState(!sealed, "Builder isalready sealed - further modifications are not allowed");
770 @SuppressWarnings("unchecked")
775 public T id(ShardIdentifier id) {
781 public T peerAddresses(Map<String, String> peerAddresses) {
783 this.peerAddresses = peerAddresses;
787 public T datastoreContext(DatastoreContext datastoreContext) {
789 this.datastoreContext = datastoreContext;
793 public T schemaContext(SchemaContext schemaContext) {
795 this.schemaContext = schemaContext;
799 public T restoreFromSnapshot(DatastoreSnapshot.ShardSnapshot restoreFromSnapshot) {
801 this.restoreFromSnapshot = restoreFromSnapshot;
805 public ShardIdentifier getId() {
809 public Map<String, String> getPeerAddresses() {
810 return peerAddresses;
813 public DatastoreContext getDatastoreContext() {
814 return datastoreContext;
817 public SchemaContext getSchemaContext() {
818 return schemaContext;
821 public DatastoreSnapshot.ShardSnapshot getRestoreFromSnapshot() {
822 return restoreFromSnapshot;
825 public TreeType getTreeType() {
826 switch (datastoreContext.getLogicalStoreType()) {
828 return TreeType.CONFIGURATION;
830 return TreeType.OPERATIONAL;
833 throw new IllegalStateException("Unhandled logical store type " + datastoreContext.getLogicalStoreType());
836 protected void verify() {
837 Preconditions.checkNotNull(id, "id should not be null");
838 Preconditions.checkNotNull(peerAddresses, "peerAddresses should not be null");
839 Preconditions.checkNotNull(datastoreContext, "dataStoreContext should not be null");
840 Preconditions.checkNotNull(schemaContext, "schemaContext should not be null");
843 public Props props() {
846 return Props.create(shardClass, this);
850 public static class Builder extends AbstractBuilder<Builder, Shard> {