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.serialization.Serialization;
16 import com.google.common.annotations.VisibleForTesting;
17 import com.google.common.base.Optional;
18 import com.google.common.base.Preconditions;
19 import com.google.common.base.Ticker;
20 import java.io.IOException;
21 import java.util.Collection;
22 import java.util.Collections;
24 import java.util.concurrent.TimeUnit;
25 import javax.annotation.Nonnull;
26 import org.opendaylight.controller.cluster.access.concepts.TransactionIdentifier;
27 import org.opendaylight.controller.cluster.common.actor.CommonConfig;
28 import org.opendaylight.controller.cluster.common.actor.MessageTracker;
29 import org.opendaylight.controller.cluster.common.actor.MessageTracker.Error;
30 import org.opendaylight.controller.cluster.common.actor.MeteringBehavior;
31 import org.opendaylight.controller.cluster.datastore.exceptions.NoShardLeaderException;
32 import org.opendaylight.controller.cluster.datastore.identifiers.ShardIdentifier;
33 import org.opendaylight.controller.cluster.datastore.jmx.mbeans.shard.ShardMBeanFactory;
34 import org.opendaylight.controller.cluster.datastore.jmx.mbeans.shard.ShardStats;
35 import org.opendaylight.controller.cluster.datastore.messages.AbortTransaction;
36 import org.opendaylight.controller.cluster.datastore.messages.ActorInitialized;
37 import org.opendaylight.controller.cluster.datastore.messages.BatchedModifications;
38 import org.opendaylight.controller.cluster.datastore.messages.CanCommitTransaction;
39 import org.opendaylight.controller.cluster.datastore.messages.CloseTransactionChain;
40 import org.opendaylight.controller.cluster.datastore.messages.CommitTransaction;
41 import org.opendaylight.controller.cluster.datastore.messages.CreateTransaction;
42 import org.opendaylight.controller.cluster.datastore.messages.CreateTransactionReply;
43 import org.opendaylight.controller.cluster.datastore.messages.DatastoreSnapshot;
44 import org.opendaylight.controller.cluster.datastore.messages.DatastoreSnapshot.ShardSnapshot;
45 import org.opendaylight.controller.cluster.datastore.messages.ForwardedReadyTransaction;
46 import org.opendaylight.controller.cluster.datastore.messages.GetShardDataTree;
47 import org.opendaylight.controller.cluster.datastore.messages.OnDemandShardState;
48 import org.opendaylight.controller.cluster.datastore.messages.PeerAddressResolved;
49 import org.opendaylight.controller.cluster.datastore.messages.ReadyLocalTransaction;
50 import org.opendaylight.controller.cluster.datastore.messages.RegisterChangeListener;
51 import org.opendaylight.controller.cluster.datastore.messages.RegisterDataTreeChangeListener;
52 import org.opendaylight.controller.cluster.datastore.messages.ShardLeaderStateChanged;
53 import org.opendaylight.controller.cluster.datastore.messages.UpdateSchemaContext;
54 import org.opendaylight.controller.cluster.datastore.utils.Dispatchers;
55 import org.opendaylight.controller.cluster.notifications.LeaderStateChanged;
56 import org.opendaylight.controller.cluster.notifications.RegisterRoleChangeListener;
57 import org.opendaylight.controller.cluster.notifications.RoleChangeNotifier;
58 import org.opendaylight.controller.cluster.raft.RaftActor;
59 import org.opendaylight.controller.cluster.raft.RaftActorRecoveryCohort;
60 import org.opendaylight.controller.cluster.raft.RaftActorSnapshotCohort;
61 import org.opendaylight.controller.cluster.raft.RaftState;
62 import org.opendaylight.controller.cluster.raft.base.messages.FollowerInitialSyncUpStatus;
63 import org.opendaylight.controller.cluster.raft.client.messages.OnDemandRaftState;
64 import org.opendaylight.controller.cluster.raft.messages.AppendEntriesReply;
65 import org.opendaylight.controller.cluster.raft.messages.ServerRemoved;
66 import org.opendaylight.controller.cluster.raft.protobuff.client.messages.Payload;
67 import org.opendaylight.yangtools.concepts.Identifier;
68 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataValidationFailedException;
69 import org.opendaylight.yangtools.yang.data.api.schema.tree.TipProducingDataTree;
70 import org.opendaylight.yangtools.yang.data.api.schema.tree.TreeType;
71 import org.opendaylight.yangtools.yang.model.api.SchemaContext;
72 import scala.concurrent.duration.Duration;
73 import scala.concurrent.duration.FiniteDuration;
76 * A Shard represents a portion of the logical data tree <br/>
78 * Our Shard uses InMemoryDataTree as it's internal representation and delegates all requests it
81 public class Shard extends RaftActor {
84 static final Object TX_COMMIT_TIMEOUT_CHECK_MESSAGE = new Object() {
86 public String toString() {
87 return "txCommitTimeoutCheck";
92 static final Object GET_SHARD_MBEAN_MESSAGE = new Object() {
94 public String toString() {
95 return "getShardMBeanMessage";
99 // FIXME: shard names should be encapsulated in their own class and this should be exposed as a constant.
100 public static final String DEFAULT_NAME = "default";
102 // The state of this Shard
103 private final ShardDataTree store;
105 /// The name of this shard
106 private final String name;
108 private final ShardStats shardMBean;
110 private DatastoreContext datastoreContext;
112 private final ShardCommitCoordinator commitCoordinator;
114 private long transactionCommitTimeout;
116 private Cancellable txCommitTimeoutCheckSchedule;
118 private final Optional<ActorRef> roleChangeNotifier;
120 private final MessageTracker appendEntriesReplyTracker;
122 private final ShardTransactionActorFactory transactionActorFactory;
124 private final ShardSnapshotCohort snapshotCohort;
126 private final DataTreeChangeListenerSupport treeChangeSupport = new DataTreeChangeListenerSupport(this);
127 private final DataChangeListenerSupport changeSupport = new DataChangeListenerSupport(this);
130 private ShardSnapshot restoreFromSnapshot;
132 private final ShardTransactionMessageRetrySupport messageRetrySupport;
134 private final FrontendMetadata frontendMetadata = new FrontendMetadata();
136 protected Shard(final AbstractBuilder<?, ?> builder) {
137 super(builder.getId().toString(), builder.getPeerAddresses(),
138 Optional.of(builder.getDatastoreContext().getShardRaftConfig()), DataStoreVersions.CURRENT_VERSION);
140 this.name = builder.getId().toString();
141 this.datastoreContext = builder.getDatastoreContext();
142 this.restoreFromSnapshot = builder.getRestoreFromSnapshot();
144 setPersistence(datastoreContext.isPersistent());
146 LOG.info("Shard created : {}, persistent : {}", name, datastoreContext.isPersistent());
148 ShardDataTreeChangeListenerPublisherActorProxy treeChangeListenerPublisher =
149 new ShardDataTreeChangeListenerPublisherActorProxy(getContext(), name + "-DTCL-publisher");
150 ShardDataChangeListenerPublisherActorProxy dataChangeListenerPublisher =
151 new ShardDataChangeListenerPublisherActorProxy(getContext(), name + "-DCL-publisher");
152 if(builder.getDataTree() != null) {
153 store = new ShardDataTree(this, builder.getSchemaContext(), builder.getDataTree(),
154 treeChangeListenerPublisher, dataChangeListenerPublisher, name);
156 store = new ShardDataTree(this, builder.getSchemaContext(), builder.getTreeType(),
157 treeChangeListenerPublisher, dataChangeListenerPublisher, name);
160 shardMBean = ShardMBeanFactory.getShardStatsMBean(name.toString(),
161 datastoreContext.getDataStoreMXBeanType());
162 shardMBean.setShard(this);
164 if (isMetricsCaptureEnabled()) {
165 getContext().become(new MeteringBehavior(this));
168 commitCoordinator = new ShardCommitCoordinator(store, LOG, this.name);
170 setTransactionCommitTimeout();
172 // create a notifier actor for each cluster member
173 roleChangeNotifier = createRoleChangeNotifier(name.toString());
175 appendEntriesReplyTracker = new MessageTracker(AppendEntriesReply.class,
176 getRaftActorContext().getConfigParams().getIsolatedCheckIntervalInMillis());
178 transactionActorFactory = new ShardTransactionActorFactory(store, datastoreContext,
179 new Dispatchers(context().system().dispatchers()).getDispatcherPath(
180 Dispatchers.DispatcherType.Transaction), self(), getContext(), shardMBean);
182 snapshotCohort = ShardSnapshotCohort.create(getContext(), builder.getId().getMemberName(), store, LOG,
185 messageRetrySupport = new ShardTransactionMessageRetrySupport(this);
188 private void setTransactionCommitTimeout() {
189 transactionCommitTimeout = TimeUnit.MILLISECONDS.convert(
190 datastoreContext.getShardTransactionCommitTimeoutInSeconds(), TimeUnit.SECONDS) / 2;
193 private Optional<ActorRef> createRoleChangeNotifier(final String shardId) {
194 ActorRef shardRoleChangeNotifier = this.getContext().actorOf(
195 RoleChangeNotifier.getProps(shardId), shardId + "-notifier");
196 return Optional.of(shardRoleChangeNotifier);
200 public void postStop() {
201 LOG.info("Stopping Shard {}", persistenceId());
205 messageRetrySupport.close();
207 if (txCommitTimeoutCheckSchedule != null) {
208 txCommitTimeoutCheckSchedule.cancel();
211 commitCoordinator.abortPendingTransactions("Transaction aborted due to shutdown.", this);
213 shardMBean.unregisterMBean();
217 protected void handleRecover(final Object message) {
218 LOG.debug("{}: onReceiveRecover: Received message {} from {}", persistenceId(), message.getClass(),
221 super.handleRecover(message);
222 if (LOG.isTraceEnabled()) {
223 appendEntriesReplyTracker.begin();
228 protected void handleNonRaftCommand(final Object message) {
229 try (final MessageTracker.Context context = appendEntriesReplyTracker.received(message)) {
230 final Optional<Error> maybeError = context.error();
231 if (maybeError.isPresent()) {
232 LOG.trace("{} : AppendEntriesReply failed to arrive at the expected interval {}", persistenceId(),
236 if (CreateTransaction.isSerializedType(message)) {
237 handleCreateTransaction(message);
238 } else if (message instanceof BatchedModifications) {
239 handleBatchedModifications((BatchedModifications)message);
240 } else if (message instanceof ForwardedReadyTransaction) {
241 handleForwardedReadyTransaction((ForwardedReadyTransaction) message);
242 } else if (message instanceof ReadyLocalTransaction) {
243 handleReadyLocalTransaction((ReadyLocalTransaction)message);
244 } else if (CanCommitTransaction.isSerializedType(message)) {
245 handleCanCommitTransaction(CanCommitTransaction.fromSerializable(message));
246 } else if (CommitTransaction.isSerializedType(message)) {
247 handleCommitTransaction(CommitTransaction.fromSerializable(message));
248 } else if (AbortTransaction.isSerializedType(message)) {
249 handleAbortTransaction(AbortTransaction.fromSerializable(message));
250 } else if (CloseTransactionChain.isSerializedType(message)) {
251 closeTransactionChain(CloseTransactionChain.fromSerializable(message));
252 } else if (message instanceof RegisterChangeListener) {
253 changeSupport.onMessage((RegisterChangeListener) message, isLeader(), hasLeader());
254 } else if (message instanceof RegisterDataTreeChangeListener) {
255 treeChangeSupport.onMessage((RegisterDataTreeChangeListener) message, isLeader(), hasLeader());
256 } else if (message instanceof UpdateSchemaContext) {
257 updateSchemaContext((UpdateSchemaContext) message);
258 } else if (message instanceof PeerAddressResolved) {
259 PeerAddressResolved resolved = (PeerAddressResolved) message;
260 setPeerAddress(resolved.getPeerId().toString(),
261 resolved.getPeerAddress());
262 } else if (TX_COMMIT_TIMEOUT_CHECK_MESSAGE.equals(message)) {
263 store.checkForExpiredTransactions(transactionCommitTimeout);
264 commitCoordinator.checkForExpiredTransactions(transactionCommitTimeout, this);
265 } else if (message instanceof DatastoreContext) {
266 onDatastoreContext((DatastoreContext)message);
267 } else if (message instanceof RegisterRoleChangeListener){
268 roleChangeNotifier.get().forward(message, context());
269 } else if (message instanceof FollowerInitialSyncUpStatus) {
270 shardMBean.setFollowerInitialSyncStatus(((FollowerInitialSyncUpStatus) message).isInitialSyncDone());
271 context().parent().tell(message, self());
272 } else if (GET_SHARD_MBEAN_MESSAGE.equals(message)){
273 sender().tell(getShardMBean(), self());
274 } else if (message instanceof GetShardDataTree) {
275 sender().tell(store.getDataTree(), self());
276 } else if (message instanceof ServerRemoved){
277 context().parent().forward(message, context());
278 } else if (ShardTransactionMessageRetrySupport.TIMER_MESSAGE_CLASS.isInstance(message)) {
279 messageRetrySupport.onTimerMessage(message);
280 } else if (message instanceof DataTreeCohortActorRegistry.CohortRegistryCommand) {
281 store.processCohortRegistryCommand(getSender(),
282 (DataTreeCohortActorRegistry.CohortRegistryCommand) message);
284 super.handleNonRaftCommand(message);
289 private boolean hasLeader() {
290 return getLeaderId() != null;
293 public int getPendingTxCommitQueueSize() {
294 return store.getQueueSize();
297 public int getCohortCacheSize() {
298 return commitCoordinator.getCohortCacheSize();
302 protected Optional<ActorRef> getRoleChangeNotifier() {
303 return roleChangeNotifier;
307 protected LeaderStateChanged newLeaderStateChanged(final String memberId, final String leaderId, final short leaderPayloadVersion) {
308 return isLeader() ? new ShardLeaderStateChanged(memberId, leaderId, store.getDataTree(), leaderPayloadVersion)
309 : new ShardLeaderStateChanged(memberId, leaderId, leaderPayloadVersion);
312 protected void onDatastoreContext(final DatastoreContext context) {
313 datastoreContext = context;
315 setTransactionCommitTimeout();
317 setPersistence(datastoreContext.isPersistent());
319 updateConfigParams(datastoreContext.getShardRaftConfig());
322 boolean canSkipPayload() {
323 // If we do not have any followers and we are not using persistence we can apply modification to the state
325 return !hasFollowers() && !persistence().isRecoveryApplicable();
328 // applyState() will be invoked once consensus is reached on the payload
329 void persistPayload(final TransactionIdentifier transactionId, final Payload payload) {
330 // We are faking the sender
331 persistData(self(), transactionId, payload);
334 private void handleCommitTransaction(final CommitTransaction commit) {
336 commitCoordinator.handleCommit(commit.getTransactionID(), getSender(), this);
338 ActorSelection leader = getLeader();
339 if (leader == null) {
340 messageRetrySupport.addMessageToRetry(commit, getSender(),
341 "Could not commit transaction " + commit.getTransactionID());
343 LOG.debug("{}: Forwarding CommitTransaction to leader {}", persistenceId(), leader);
344 leader.forward(commit, getContext());
349 private void handleCanCommitTransaction(final CanCommitTransaction canCommit) {
350 LOG.debug("{}: Can committing transaction {}", persistenceId(), canCommit.getTransactionID());
353 commitCoordinator.handleCanCommit(canCommit.getTransactionID(), getSender(), this);
355 ActorSelection leader = getLeader();
356 if (leader == null) {
357 messageRetrySupport.addMessageToRetry(canCommit, getSender(),
358 "Could not canCommit transaction " + canCommit.getTransactionID());
360 LOG.debug("{}: Forwarding CanCommitTransaction to leader {}", persistenceId(), leader);
361 leader.forward(canCommit, getContext());
366 protected void handleBatchedModificationsLocal(final BatchedModifications batched, final ActorRef sender) {
368 commitCoordinator.handleBatchedModifications(batched, sender, this);
369 } catch (Exception e) {
370 LOG.error("{}: Error handling BatchedModifications for Tx {}", persistenceId(),
371 batched.getTransactionID(), e);
372 sender.tell(new akka.actor.Status.Failure(e), getSelf());
376 private void handleBatchedModifications(final BatchedModifications batched) {
377 // This message is sent to prepare the modifications transaction directly on the Shard as an
378 // optimization to avoid the extra overhead of a separate ShardTransaction actor. On the last
379 // BatchedModifications message, the caller sets the ready flag in the message indicating
380 // modifications are complete. The reply contains the cohort actor path (this actor) for the caller
381 // to initiate the 3-phase commit. This also avoids the overhead of sending an additional
382 // ReadyTransaction message.
384 // If we're not the leader then forward to the leader. This is a safety measure - we shouldn't
385 // normally get here if we're not the leader as the front-end (TransactionProxy) should determine
386 // the primary/leader shard. However with timing and caching on the front-end, there's a small
387 // window where it could have a stale leader during leadership transitions.
389 boolean isLeaderActive = isLeaderActive();
390 if (isLeader() && isLeaderActive) {
391 handleBatchedModificationsLocal(batched, getSender());
393 ActorSelection leader = getLeader();
394 if (!isLeaderActive || leader == null) {
395 messageRetrySupport.addMessageToRetry(batched, getSender(),
396 "Could not commit transaction " + batched.getTransactionID());
398 // If this is not the first batch and leadership changed in between batched messages,
399 // we need to reconstruct previous BatchedModifications from the transaction
400 // DataTreeModification, honoring the max batched modification count, and forward all the
401 // previous BatchedModifications to the new leader.
402 Collection<BatchedModifications> newModifications = commitCoordinator.createForwardedBatchedModifications(
403 batched, datastoreContext.getShardBatchedModificationCount());
405 LOG.debug("{}: Forwarding {} BatchedModifications to leader {}", persistenceId(),
406 newModifications.size(), leader);
408 for (BatchedModifications bm : newModifications) {
409 leader.forward(bm, getContext());
415 private boolean failIfIsolatedLeader(final ActorRef sender) {
416 if (isIsolatedLeader()) {
417 sender.tell(new akka.actor.Status.Failure(new NoShardLeaderException(String.format(
418 "Shard %s was the leader but has lost contact with all of its followers. Either all" +
419 " other follower nodes are down or this node is isolated by a network partition.",
420 persistenceId()))), getSelf());
427 protected boolean isIsolatedLeader() {
428 return getRaftState() == RaftState.IsolatedLeader;
431 private void handleReadyLocalTransaction(final ReadyLocalTransaction message) {
432 LOG.debug("{}: handleReadyLocalTransaction for {}", persistenceId(), message.getTransactionID());
434 boolean isLeaderActive = isLeaderActive();
435 if (isLeader() && isLeaderActive) {
437 commitCoordinator.handleReadyLocalTransaction(message, getSender(), this);
438 } catch (Exception e) {
439 LOG.error("{}: Error handling ReadyLocalTransaction for Tx {}", persistenceId(),
440 message.getTransactionID(), e);
441 getSender().tell(new akka.actor.Status.Failure(e), getSelf());
444 ActorSelection leader = getLeader();
445 if (!isLeaderActive || leader == null) {
446 messageRetrySupport.addMessageToRetry(message, getSender(),
447 "Could not commit transaction " + message.getTransactionID());
449 LOG.debug("{}: Forwarding ReadyLocalTransaction to leader {}", persistenceId(), leader);
450 message.setRemoteVersion(getCurrentBehavior().getLeaderPayloadVersion());
451 leader.forward(message, getContext());
456 private void handleForwardedReadyTransaction(final ForwardedReadyTransaction forwardedReady) {
457 LOG.debug("{}: handleForwardedReadyTransaction for {}", persistenceId(), forwardedReady.getTransactionID());
459 boolean isLeaderActive = isLeaderActive();
460 if (isLeader() && isLeaderActive) {
461 commitCoordinator.handleForwardedReadyTransaction(forwardedReady, getSender(), this);
463 ActorSelection leader = getLeader();
464 if (!isLeaderActive || leader == null) {
465 messageRetrySupport.addMessageToRetry(forwardedReady, getSender(),
466 "Could not commit transaction " + forwardedReady.getTransactionID());
468 LOG.debug("{}: Forwarding ForwardedReadyTransaction to leader {}", persistenceId(), leader);
470 ReadyLocalTransaction readyLocal = new ReadyLocalTransaction(forwardedReady.getTransactionID(),
471 forwardedReady.getTransaction().getSnapshot(), forwardedReady.isDoImmediateCommit());
472 readyLocal.setRemoteVersion(getCurrentBehavior().getLeaderPayloadVersion());
473 leader.forward(readyLocal, getContext());
478 private void handleAbortTransaction(final AbortTransaction abort) {
479 doAbortTransaction(abort.getTransactionID(), getSender());
482 void doAbortTransaction(final TransactionIdentifier transactionID, final ActorRef sender) {
483 commitCoordinator.handleAbort(transactionID, sender, this);
486 private void handleCreateTransaction(final Object message) {
488 createTransaction(CreateTransaction.fromSerializable(message));
489 } else if (getLeader() != null) {
490 getLeader().forward(message, getContext());
492 getSender().tell(new akka.actor.Status.Failure(new NoShardLeaderException(
493 "Could not create a shard transaction", persistenceId())), getSelf());
497 private void closeTransactionChain(final CloseTransactionChain closeTransactionChain) {
498 store.closeTransactionChain(closeTransactionChain.getIdentifier());
501 private void createTransaction(final CreateTransaction createTransaction) {
503 if (TransactionType.fromInt(createTransaction.getTransactionType()) != TransactionType.READ_ONLY &&
504 failIfIsolatedLeader(getSender())) {
508 ActorRef transactionActor = createTransaction(createTransaction.getTransactionType(),
509 createTransaction.getTransactionId());
511 getSender().tell(new CreateTransactionReply(Serialization.serializedActorPath(transactionActor),
512 createTransaction.getTransactionId(), createTransaction.getVersion()).toSerializable(), getSelf());
513 } catch (Exception e) {
514 getSender().tell(new akka.actor.Status.Failure(e), getSelf());
518 private ActorRef createTransaction(final int transactionType, final TransactionIdentifier transactionId) {
519 LOG.debug("{}: Creating transaction : {} ", persistenceId(), transactionId);
520 return transactionActorFactory.newShardTransaction(TransactionType.fromInt(transactionType),
524 private void updateSchemaContext(final UpdateSchemaContext message) {
525 updateSchemaContext(message.getSchemaContext());
529 void updateSchemaContext(final SchemaContext schemaContext) {
530 store.updateSchemaContext(schemaContext);
533 private boolean isMetricsCaptureEnabled() {
534 CommonConfig config = new CommonConfig(getContext().system().settings().config());
535 return config.isMetricCaptureEnabled();
540 public RaftActorSnapshotCohort getRaftActorSnapshotCohort() {
541 return snapshotCohort;
546 protected RaftActorRecoveryCohort getRaftActorRecoveryCohort() {
547 return new ShardRecoveryCoordinator(store,
548 restoreFromSnapshot != null ? restoreFromSnapshot.getSnapshot() : null, persistenceId(), LOG);
552 protected void onRecoveryComplete() {
553 restoreFromSnapshot = null;
555 //notify shard manager
556 getContext().parent().tell(new ActorInitialized(), getSelf());
558 // Being paranoid here - this method should only be called once but just in case...
559 if (txCommitTimeoutCheckSchedule == null) {
560 // Schedule a message to be periodically sent to check if the current in-progress
561 // transaction should be expired and aborted.
562 FiniteDuration period = Duration.create(transactionCommitTimeout / 3, TimeUnit.MILLISECONDS);
563 txCommitTimeoutCheckSchedule = getContext().system().scheduler().schedule(
564 period, period, getSelf(),
565 TX_COMMIT_TIMEOUT_CHECK_MESSAGE, getContext().dispatcher(), ActorRef.noSender());
570 protected void applyState(final ActorRef clientActor, final Identifier identifier, final Object data) {
571 if (data instanceof Payload) {
573 store.applyReplicatedPayload(identifier, (Payload)data);
574 } catch (DataValidationFailedException | IOException e) {
575 LOG.error("{}: Error applying replica {}", persistenceId(), identifier, e);
578 LOG.error("{}: Unknown state for {} received {}", persistenceId(), identifier, data);
583 protected void onStateChanged() {
584 boolean isLeader = isLeader();
585 boolean hasLeader = hasLeader();
586 changeSupport.onLeadershipChange(isLeader, hasLeader);
587 treeChangeSupport.onLeadershipChange(isLeader, hasLeader);
589 // If this actor is no longer the leader close all the transaction chains
591 if (LOG.isDebugEnabled()) {
593 "{}: onStateChanged: Closing all transaction chains because shard {} is no longer the leader",
594 persistenceId(), getId());
597 store.closeAllTransactionChains();
600 if (hasLeader && !isIsolatedLeader()) {
601 messageRetrySupport.retryMessages();
606 protected void onLeaderChanged(final String oldLeader, final String newLeader) {
607 shardMBean.incrementLeadershipChangeCount();
609 boolean hasLeader = hasLeader();
610 if (hasLeader && !isLeader()) {
611 // Another leader was elected. If we were the previous leader and had pending transactions, convert
612 // them to transaction messages and send to the new leader.
613 ActorSelection leader = getLeader();
614 if (leader != null) {
615 Collection<?> messagesToForward = convertPendingTransactionsToMessages();
617 if (!messagesToForward.isEmpty()) {
618 LOG.debug("{}: Forwarding {} pending transaction messages to leader {}", persistenceId(),
619 messagesToForward.size(), leader);
621 for (Object message : messagesToForward) {
622 leader.tell(message, self());
626 commitCoordinator.abortPendingTransactions(
627 "The transacton was aborted due to inflight leadership change and the leader address isn't available.",
632 if (hasLeader && !isIsolatedLeader()) {
633 messageRetrySupport.retryMessages();
638 * Clears all pending transactions and converts them to messages to be forwarded to a new leader.
640 * @return the converted messages
642 public Collection<?> convertPendingTransactionsToMessages() {
643 return commitCoordinator.convertPendingTransactionsToMessages(datastoreContext.getShardBatchedModificationCount());
647 protected void pauseLeader(final Runnable operation) {
648 LOG.debug("{}: In pauseLeader, operation: {}", persistenceId(), operation);
649 store.setRunOnPendingTransactionsComplete(operation);
653 protected OnDemandRaftState.AbstractBuilder<?> newOnDemandRaftStateBuilder() {
654 return OnDemandShardState.newBuilder().treeChangeListenerActors(treeChangeSupport.getListenerActors())
655 .dataChangeListenerActors(changeSupport.getListenerActors())
656 .commitCohortActors(store.getCohortActors());
660 public String persistenceId() {
665 ShardCommitCoordinator getCommitCoordinator() {
666 return commitCoordinator;
669 public DatastoreContext getDatastoreContext() {
670 return datastoreContext;
674 public ShardDataTree getDataStore() {
679 ShardStats getShardMBean() {
683 public static Builder builder() {
684 return new Builder();
687 public static abstract class AbstractBuilder<T extends AbstractBuilder<T, S>, S extends Shard> {
688 private final Class<S> shardClass;
689 private ShardIdentifier id;
690 private Map<String, String> peerAddresses = Collections.emptyMap();
691 private DatastoreContext datastoreContext;
692 private SchemaContext schemaContext;
693 private DatastoreSnapshot.ShardSnapshot restoreFromSnapshot;
694 private TipProducingDataTree dataTree;
695 private volatile boolean sealed;
697 protected AbstractBuilder(final Class<S> shardClass) {
698 this.shardClass = shardClass;
701 protected void checkSealed() {
702 Preconditions.checkState(!sealed, "Builder isalready sealed - further modifications are not allowed");
705 @SuppressWarnings("unchecked")
710 public T id(final ShardIdentifier id) {
716 public T peerAddresses(final Map<String, String> peerAddresses) {
718 this.peerAddresses = peerAddresses;
722 public T datastoreContext(final DatastoreContext datastoreContext) {
724 this.datastoreContext = datastoreContext;
728 public T schemaContext(final SchemaContext schemaContext) {
730 this.schemaContext = schemaContext;
734 public T restoreFromSnapshot(final DatastoreSnapshot.ShardSnapshot restoreFromSnapshot) {
736 this.restoreFromSnapshot = restoreFromSnapshot;
740 public T dataTree(final TipProducingDataTree dataTree) {
742 this.dataTree = dataTree;
746 public ShardIdentifier getId() {
750 public Map<String, String> getPeerAddresses() {
751 return peerAddresses;
754 public DatastoreContext getDatastoreContext() {
755 return datastoreContext;
758 public SchemaContext getSchemaContext() {
759 return schemaContext;
762 public DatastoreSnapshot.ShardSnapshot getRestoreFromSnapshot() {
763 return restoreFromSnapshot;
766 public TipProducingDataTree getDataTree() {
770 public TreeType getTreeType() {
771 switch (datastoreContext.getLogicalStoreType()) {
773 return TreeType.CONFIGURATION;
775 return TreeType.OPERATIONAL;
778 throw new IllegalStateException("Unhandled logical store type " + datastoreContext.getLogicalStoreType());
781 protected void verify() {
782 Preconditions.checkNotNull(id, "id should not be null");
783 Preconditions.checkNotNull(peerAddresses, "peerAddresses should not be null");
784 Preconditions.checkNotNull(datastoreContext, "dataStoreContext should not be null");
785 Preconditions.checkNotNull(schemaContext, "schemaContext should not be null");
788 public Props props() {
791 return Props.create(shardClass, this);
795 public static class Builder extends AbstractBuilder<Builder, Shard> {
802 return Ticker.systemTicker();