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.PoisonPill;
15 import akka.actor.Props;
16 import akka.event.Logging;
17 import akka.event.LoggingAdapter;
18 import akka.japi.Creator;
19 import akka.persistence.RecoveryFailure;
20 import akka.serialization.Serialization;
21 import com.google.common.annotations.VisibleForTesting;
22 import com.google.common.base.Optional;
23 import com.google.common.base.Preconditions;
24 import com.google.common.collect.Lists;
25 import com.google.common.util.concurrent.FutureCallback;
26 import com.google.common.util.concurrent.Futures;
27 import com.google.common.util.concurrent.ListenableFuture;
28 import com.google.protobuf.ByteString;
29 import com.google.protobuf.InvalidProtocolBufferException;
30 import java.util.Collection;
31 import java.util.HashMap;
32 import java.util.List;
34 import java.util.concurrent.ExecutionException;
35 import java.util.concurrent.TimeUnit;
36 import javax.annotation.Nonnull;
37 import org.opendaylight.controller.cluster.DataPersistenceProvider;
38 import org.opendaylight.controller.cluster.common.actor.CommonConfig;
39 import org.opendaylight.controller.cluster.common.actor.MeteringBehavior;
40 import org.opendaylight.controller.cluster.datastore.ShardCommitCoordinator.CohortEntry;
41 import org.opendaylight.controller.cluster.datastore.compat.BackwardsCompatibleThreePhaseCommitCohort;
42 import org.opendaylight.controller.cluster.datastore.exceptions.NoShardLeaderException;
43 import org.opendaylight.controller.cluster.datastore.identifiers.ShardIdentifier;
44 import org.opendaylight.controller.cluster.datastore.identifiers.ShardTransactionIdentifier;
45 import org.opendaylight.controller.cluster.datastore.jmx.mbeans.shard.ShardMBeanFactory;
46 import org.opendaylight.controller.cluster.datastore.jmx.mbeans.shard.ShardStats;
47 import org.opendaylight.controller.cluster.datastore.messages.AbortTransaction;
48 import org.opendaylight.controller.cluster.datastore.messages.AbortTransactionReply;
49 import org.opendaylight.controller.cluster.datastore.messages.ActorInitialized;
50 import org.opendaylight.controller.cluster.datastore.messages.CanCommitTransaction;
51 import org.opendaylight.controller.cluster.datastore.messages.CloseTransactionChain;
52 import org.opendaylight.controller.cluster.datastore.messages.CommitTransaction;
53 import org.opendaylight.controller.cluster.datastore.messages.CommitTransactionReply;
54 import org.opendaylight.controller.cluster.datastore.messages.CreateTransaction;
55 import org.opendaylight.controller.cluster.datastore.messages.CreateTransactionReply;
56 import org.opendaylight.controller.cluster.datastore.messages.EnableNotification;
57 import org.opendaylight.controller.cluster.datastore.messages.ForwardedReadyTransaction;
58 import org.opendaylight.controller.cluster.datastore.messages.PeerAddressResolved;
59 import org.opendaylight.controller.cluster.datastore.messages.ReadData;
60 import org.opendaylight.controller.cluster.datastore.messages.ReadDataReply;
61 import org.opendaylight.controller.cluster.datastore.messages.ReadyTransactionReply;
62 import org.opendaylight.controller.cluster.datastore.messages.RegisterChangeListener;
63 import org.opendaylight.controller.cluster.datastore.messages.RegisterChangeListenerReply;
64 import org.opendaylight.controller.cluster.datastore.messages.UpdateSchemaContext;
65 import org.opendaylight.controller.cluster.datastore.modification.Modification;
66 import org.opendaylight.controller.cluster.datastore.modification.MutableCompositeModification;
67 import org.opendaylight.controller.cluster.datastore.node.NormalizedNodeToNodeCodec;
68 import org.opendaylight.controller.cluster.notifications.RoleChangeNotifier;
69 import org.opendaylight.controller.cluster.raft.RaftActor;
70 import org.opendaylight.controller.cluster.raft.ReplicatedLogEntry;
71 import org.opendaylight.controller.cluster.raft.base.messages.CaptureSnapshotReply;
72 import org.opendaylight.controller.cluster.raft.protobuff.client.messages.CompositeModificationPayload;
73 import org.opendaylight.controller.cluster.raft.protobuff.client.messages.Payload;
74 import org.opendaylight.controller.md.sal.common.api.data.AsyncDataChangeListener;
75 import org.opendaylight.controller.md.sal.dom.store.impl.InMemoryDOMDataStore;
76 import org.opendaylight.controller.md.sal.dom.store.impl.InMemoryDOMDataStoreFactory;
77 import org.opendaylight.controller.protobuff.messages.common.NormalizedNodeMessages;
78 import org.opendaylight.controller.sal.core.spi.data.DOMStoreThreePhaseCommitCohort;
79 import org.opendaylight.controller.sal.core.spi.data.DOMStoreTransactionChain;
80 import org.opendaylight.controller.sal.core.spi.data.DOMStoreTransactionFactory;
81 import org.opendaylight.controller.sal.core.spi.data.DOMStoreWriteTransaction;
82 import org.opendaylight.yangtools.concepts.ListenerRegistration;
83 import org.opendaylight.yangtools.yang.data.api.YangInstanceIdentifier;
84 import org.opendaylight.yangtools.yang.data.api.schema.NormalizedNode;
85 import org.opendaylight.yangtools.yang.model.api.SchemaContext;
86 import scala.concurrent.duration.Duration;
87 import scala.concurrent.duration.FiniteDuration;
90 * A Shard represents a portion of the logical data tree <br/>
92 * Our Shard uses InMemoryDataStore as it's internal representation and delegates all requests it
95 public class Shard extends RaftActor {
97 private static final Object COMMIT_TRANSACTION_REPLY = new CommitTransactionReply().toSerializable();
99 private static final Object TX_COMMIT_TIMEOUT_CHECK_MESSAGE = "txCommitTimeoutCheck";
101 public static final String DEFAULT_NAME = "default";
103 // The state of this Shard
104 private final InMemoryDOMDataStore store;
106 private final LoggingAdapter LOG =
107 Logging.getLogger(getContext().system(), this);
109 /// The name of this shard
110 private final ShardIdentifier name;
112 private final ShardStats shardMBean;
114 private final List<ActorSelection> dataChangeListeners = Lists.newArrayList();
116 private final List<DelayedListenerRegistration> delayedListenerRegistrations =
117 Lists.newArrayList();
119 private final DatastoreContext datastoreContext;
121 private final DataPersistenceProvider dataPersistenceProvider;
123 private SchemaContext schemaContext;
125 private ActorRef createSnapshotTransaction;
127 private int createSnapshotTransactionCounter;
129 private final ShardCommitCoordinator commitCoordinator;
131 private final long transactionCommitTimeout;
133 private Cancellable txCommitTimeoutCheckSchedule;
135 private Optional<ActorRef> roleChangeNotifier;
138 * Coordinates persistence recovery on startup.
140 private ShardRecoveryCoordinator recoveryCoordinator;
141 private List<Object> currentLogRecoveryBatch;
143 private final Map<String, DOMStoreTransactionChain> transactionChains = new HashMap<>();
145 protected Shard(final ShardIdentifier name, final Map<ShardIdentifier, String> peerAddresses,
146 final DatastoreContext datastoreContext, final SchemaContext schemaContext) {
147 super(name.toString(), mapPeerAddresses(peerAddresses),
148 Optional.of(datastoreContext.getShardRaftConfig()));
151 this.datastoreContext = datastoreContext;
152 this.schemaContext = schemaContext;
153 this.dataPersistenceProvider = (datastoreContext.isPersistent()) ? new PersistentDataProvider() : new NonPersistentRaftDataProvider();
155 LOG.info("Shard created : {} persistent : {}", name, datastoreContext.isPersistent());
157 store = InMemoryDOMDataStoreFactory.create(name.toString(), null,
158 datastoreContext.getDataStoreProperties());
160 if(schemaContext != null) {
161 store.onGlobalContextUpdated(schemaContext);
164 shardMBean = ShardMBeanFactory.getShardStatsMBean(name.toString(),
165 datastoreContext.getDataStoreMXBeanType());
166 shardMBean.setNotificationManager(store.getDataChangeListenerNotificationManager());
168 if (isMetricsCaptureEnabled()) {
169 getContext().become(new MeteringBehavior(this));
172 commitCoordinator = new ShardCommitCoordinator(TimeUnit.SECONDS.convert(1, TimeUnit.MINUTES),
173 datastoreContext.getShardTransactionCommitQueueCapacity());
175 transactionCommitTimeout = TimeUnit.MILLISECONDS.convert(
176 datastoreContext.getShardTransactionCommitTimeoutInSeconds(), TimeUnit.SECONDS);
178 // create a notifier actor for each cluster member
179 roleChangeNotifier = createRoleChangeNotifier(name.toString());
182 private static Map<String, String> mapPeerAddresses(
183 final Map<ShardIdentifier, String> peerAddresses) {
184 Map<String, String> map = new HashMap<>();
186 for (Map.Entry<ShardIdentifier, String> entry : peerAddresses
188 map.put(entry.getKey().toString(), entry.getValue());
194 public static Props props(final ShardIdentifier name,
195 final Map<ShardIdentifier, String> peerAddresses,
196 final DatastoreContext datastoreContext, final SchemaContext schemaContext) {
197 Preconditions.checkNotNull(name, "name should not be null");
198 Preconditions.checkNotNull(peerAddresses, "peerAddresses should not be null");
199 Preconditions.checkNotNull(datastoreContext, "dataStoreContext should not be null");
200 Preconditions.checkNotNull(schemaContext, "schemaContext should not be null");
202 return Props.create(new ShardCreator(name, peerAddresses, datastoreContext, schemaContext));
205 private Optional<ActorRef> createRoleChangeNotifier(String shardId) {
206 ActorRef shardRoleChangeNotifier = this.getContext().actorOf(
207 RoleChangeNotifier.getProps(shardId), shardId + "-notifier");
208 return Optional.<ActorRef>of(shardRoleChangeNotifier);
212 public void postStop() {
215 if(txCommitTimeoutCheckSchedule != null) {
216 txCommitTimeoutCheckSchedule.cancel();
221 public void onReceiveRecover(final Object message) throws Exception {
222 if(LOG.isDebugEnabled()) {
223 LOG.debug("onReceiveRecover: Received message {} from {}",
224 message.getClass().toString(),
228 if (message instanceof RecoveryFailure){
229 LOG.error(((RecoveryFailure) message).cause(), "Recovery failed because of this cause");
231 // Even though recovery failed, we still need to finish our recovery, eg send the
232 // ActorInitialized message and start the txCommitTimeoutCheckSchedule.
233 onRecoveryComplete();
235 super.onReceiveRecover(message);
240 public void onReceiveCommand(final Object message) throws Exception {
241 if(LOG.isDebugEnabled()) {
242 LOG.debug("onReceiveCommand: Received message {} from {}", message, getSender());
245 if(message.getClass().equals(ReadDataReply.SERIALIZABLE_CLASS)) {
246 handleReadDataReply(message);
247 } else if (message.getClass().equals(CreateTransaction.SERIALIZABLE_CLASS)) {
248 handleCreateTransaction(message);
249 } else if(message instanceof ForwardedReadyTransaction) {
250 handleForwardedReadyTransaction((ForwardedReadyTransaction)message);
251 } else if(message.getClass().equals(CanCommitTransaction.SERIALIZABLE_CLASS)) {
252 handleCanCommitTransaction(CanCommitTransaction.fromSerializable(message));
253 } else if(message.getClass().equals(CommitTransaction.SERIALIZABLE_CLASS)) {
254 handleCommitTransaction(CommitTransaction.fromSerializable(message));
255 } else if(message.getClass().equals(AbortTransaction.SERIALIZABLE_CLASS)) {
256 handleAbortTransaction(AbortTransaction.fromSerializable(message));
257 } else if (message.getClass().equals(CloseTransactionChain.SERIALIZABLE_CLASS)){
258 closeTransactionChain(CloseTransactionChain.fromSerializable(message));
259 } else if (message instanceof RegisterChangeListener) {
260 registerChangeListener((RegisterChangeListener) message);
261 } else if (message instanceof UpdateSchemaContext) {
262 updateSchemaContext((UpdateSchemaContext) message);
263 } else if (message instanceof PeerAddressResolved) {
264 PeerAddressResolved resolved = (PeerAddressResolved) message;
265 setPeerAddress(resolved.getPeerId().toString(),
266 resolved.getPeerAddress());
267 } else if(message.equals(TX_COMMIT_TIMEOUT_CHECK_MESSAGE)) {
268 handleTransactionCommitTimeoutCheck();
270 super.onReceiveCommand(message);
275 protected Optional<ActorRef> getRoleChangeNotifier() {
276 return roleChangeNotifier;
279 private void handleTransactionCommitTimeoutCheck() {
280 CohortEntry cohortEntry = commitCoordinator.getCurrentCohortEntry();
281 if(cohortEntry != null) {
282 long elapsed = System.currentTimeMillis() - cohortEntry.getLastAccessTime();
283 if(elapsed > transactionCommitTimeout) {
284 LOG.warning("Current transaction {} has timed out after {} ms - aborting",
285 cohortEntry.getTransactionID(), transactionCommitTimeout);
287 doAbortTransaction(cohortEntry.getTransactionID(), null);
292 private void handleCommitTransaction(final CommitTransaction commit) {
293 final String transactionID = commit.getTransactionID();
295 LOG.debug("Committing transaction {}", transactionID);
297 // Get the current in-progress cohort entry in the commitCoordinator if it corresponds to
299 final CohortEntry cohortEntry = commitCoordinator.getCohortEntryIfCurrent(transactionID);
300 if(cohortEntry == null) {
301 // We're not the current Tx - the Tx was likely expired b/c it took too long in
302 // between the canCommit and commit messages.
303 IllegalStateException ex = new IllegalStateException(
304 String.format("Cannot commit transaction %s - it is not the current transaction",
306 LOG.error(ex.getMessage());
307 shardMBean.incrementFailedTransactionsCount();
308 getSender().tell(new akka.actor.Status.Failure(ex), getSelf());
312 // We perform the preCommit phase here atomically with the commit phase. This is an
313 // optimization to eliminate the overhead of an extra preCommit message. We lose front-end
314 // coordination of preCommit across shards in case of failure but preCommit should not
315 // normally fail since we ensure only one concurrent 3-phase commit.
318 // We block on the future here so we don't have to worry about possibly accessing our
319 // state on a different thread outside of our dispatcher. Also, the data store
320 // currently uses a same thread executor anyway.
321 cohortEntry.getCohort().preCommit().get();
323 Shard.this.persistData(getSender(), transactionID,
324 new CompositeModificationPayload(cohortEntry.getModification().toSerializable()));
325 } catch (InterruptedException | ExecutionException e) {
326 LOG.error(e, "An exception occurred while preCommitting transaction {}",
327 cohortEntry.getTransactionID());
328 shardMBean.incrementFailedTransactionsCount();
329 getSender().tell(new akka.actor.Status.Failure(e), getSelf());
332 cohortEntry.updateLastAccessTime();
335 private void finishCommit(@Nonnull final ActorRef sender, final @Nonnull String transactionID) {
336 // With persistence enabled, this method is called via applyState by the leader strategy
337 // after the commit has been replicated to a majority of the followers.
339 CohortEntry cohortEntry = commitCoordinator.getCohortEntryIfCurrent(transactionID);
340 if(cohortEntry == null) {
341 // The transaction is no longer the current commit. This can happen if the transaction
342 // was aborted prior, most likely due to timeout in the front-end. We need to finish
343 // committing the transaction though since it was successfully persisted and replicated
344 // however we can't use the original cohort b/c it was already preCommitted and may
345 // conflict with the current commit or may have been aborted so we commit with a new
347 cohortEntry = commitCoordinator.getAndRemoveCohortEntry(transactionID);
348 if(cohortEntry != null) {
349 commitWithNewTransaction(cohortEntry.getModification());
350 sender.tell(COMMIT_TRANSACTION_REPLY, getSelf());
352 // This really shouldn't happen - it likely means that persistence or replication
353 // took so long to complete such that the cohort entry was expired from the cache.
354 IllegalStateException ex = new IllegalStateException(
355 String.format("Could not finish committing transaction %s - no CohortEntry found",
357 LOG.error(ex.getMessage());
358 sender.tell(new akka.actor.Status.Failure(ex), getSelf());
364 LOG.debug("Finishing commit for transaction {}", cohortEntry.getTransactionID());
367 // We block on the future here so we don't have to worry about possibly accessing our
368 // state on a different thread outside of our dispatcher. Also, the data store
369 // currently uses a same thread executor anyway.
370 cohortEntry.getCohort().commit().get();
372 sender.tell(COMMIT_TRANSACTION_REPLY, getSelf());
374 shardMBean.incrementCommittedTransactionCount();
375 shardMBean.setLastCommittedTransactionTime(System.currentTimeMillis());
377 } catch (InterruptedException | ExecutionException e) {
378 sender.tell(new akka.actor.Status.Failure(e), getSelf());
380 LOG.error(e, "An exception occurred while committing transaction {}", transactionID);
381 shardMBean.incrementFailedTransactionsCount();
384 commitCoordinator.currentTransactionComplete(transactionID, true);
387 private void handleCanCommitTransaction(final CanCommitTransaction canCommit) {
388 LOG.debug("Can committing transaction {}", canCommit.getTransactionID());
389 commitCoordinator.handleCanCommit(canCommit, getSender(), self());
392 private void handleForwardedReadyTransaction(ForwardedReadyTransaction ready) {
393 LOG.debug("Readying transaction {}, client version {}", ready.getTransactionID(),
394 ready.getTxnClientVersion());
396 // This message is forwarded by the ShardTransaction on ready. We cache the cohort in the
397 // commitCoordinator in preparation for the subsequent three phase commit initiated by
399 commitCoordinator.transactionReady(ready.getTransactionID(), ready.getCohort(),
400 ready.getModification());
402 // Return our actor path as we'll handle the three phase commit, except if the Tx client
403 // version < 1 (Helium-1 version). This means the Tx was initiated by a base Helium version
404 // node. In that case, the subsequent 3-phase commit messages won't contain the
405 // transactionId so to maintain backwards compatibility, we create a separate cohort actor
406 // to provide the compatible behavior.
407 ActorRef replyActorPath = self();
408 if(ready.getTxnClientVersion() < CreateTransaction.HELIUM_1_VERSION) {
409 LOG.debug("Creating BackwardsCompatibleThreePhaseCommitCohort");
410 replyActorPath = getContext().actorOf(BackwardsCompatibleThreePhaseCommitCohort.props(
411 ready.getTransactionID()));
414 ReadyTransactionReply readyTransactionReply = new ReadyTransactionReply(
415 Serialization.serializedActorPath(replyActorPath));
416 getSender().tell(ready.isReturnSerialized() ? readyTransactionReply.toSerializable() :
417 readyTransactionReply, getSelf());
420 private void handleAbortTransaction(final AbortTransaction abort) {
421 doAbortTransaction(abort.getTransactionID(), getSender());
424 private void doAbortTransaction(final String transactionID, final ActorRef sender) {
425 final CohortEntry cohortEntry = commitCoordinator.getCohortEntryIfCurrent(transactionID);
426 if(cohortEntry != null) {
427 LOG.debug("Aborting transaction {}", transactionID);
429 // We don't remove the cached cohort entry here (ie pass false) in case the Tx was
430 // aborted during replication in which case we may still commit locally if replication
432 commitCoordinator.currentTransactionComplete(transactionID, false);
434 final ListenableFuture<Void> future = cohortEntry.getCohort().abort();
435 final ActorRef self = getSelf();
437 Futures.addCallback(future, new FutureCallback<Void>() {
439 public void onSuccess(final Void v) {
440 shardMBean.incrementAbortTransactionsCount();
443 sender.tell(new AbortTransactionReply().toSerializable(), self);
448 public void onFailure(final Throwable t) {
449 LOG.error(t, "An exception happened during abort");
452 sender.tell(new akka.actor.Status.Failure(t), self);
459 private void handleCreateTransaction(final Object message) {
461 createTransaction(CreateTransaction.fromSerializable(message));
462 } else if (getLeader() != null) {
463 getLeader().forward(message, getContext());
465 getSender().tell(new akka.actor.Status.Failure(new NoShardLeaderException(
466 "Could not find shard leader so transaction cannot be created. This typically happens" +
467 " when the system is coming up or recovering and a leader is being elected. Try again" +
468 " later.")), getSelf());
472 private void handleReadDataReply(final Object message) {
473 // This must be for install snapshot. Don't want to open this up and trigger
476 self().tell(new CaptureSnapshotReply(ReadDataReply.getNormalizedNodeByteString(message)),
479 createSnapshotTransaction = null;
481 // Send a PoisonPill instead of sending close transaction because we do not really need
483 getSender().tell(PoisonPill.getInstance(), self());
486 private void closeTransactionChain(final CloseTransactionChain closeTransactionChain) {
487 DOMStoreTransactionChain chain =
488 transactionChains.remove(closeTransactionChain.getTransactionChainId());
495 private ActorRef createTypedTransactionActor(int transactionType,
496 ShardTransactionIdentifier transactionId, String transactionChainId, int clientVersion ) {
498 DOMStoreTransactionFactory factory = store;
500 if(!transactionChainId.isEmpty()) {
501 factory = transactionChains.get(transactionChainId);
503 DOMStoreTransactionChain transactionChain = store.createTransactionChain();
504 transactionChains.put(transactionChainId, transactionChain);
505 factory = transactionChain;
509 if(this.schemaContext == null) {
510 throw new IllegalStateException("SchemaContext is not set");
513 if (transactionType == TransactionProxy.TransactionType.READ_ONLY.ordinal()) {
515 shardMBean.incrementReadOnlyTransactionCount();
517 return getContext().actorOf(
518 ShardTransaction.props(factory.newReadOnlyTransaction(), getSelf(),
519 schemaContext,datastoreContext, shardMBean,
520 transactionId.getRemoteTransactionId(), clientVersion),
521 transactionId.toString());
523 } else if (transactionType == TransactionProxy.TransactionType.READ_WRITE.ordinal()) {
525 shardMBean.incrementReadWriteTransactionCount();
527 return getContext().actorOf(
528 ShardTransaction.props(factory.newReadWriteTransaction(), getSelf(),
529 schemaContext, datastoreContext, shardMBean,
530 transactionId.getRemoteTransactionId(), clientVersion),
531 transactionId.toString());
534 } else if (transactionType == TransactionProxy.TransactionType.WRITE_ONLY.ordinal()) {
536 shardMBean.incrementWriteOnlyTransactionCount();
538 return getContext().actorOf(
539 ShardTransaction.props(factory.newWriteOnlyTransaction(), getSelf(),
540 schemaContext, datastoreContext, shardMBean,
541 transactionId.getRemoteTransactionId(), clientVersion),
542 transactionId.toString());
544 throw new IllegalArgumentException(
545 "Shard="+name + ":CreateTransaction message has unidentified transaction type="
550 private void createTransaction(CreateTransaction createTransaction) {
552 ActorRef transactionActor = createTransaction(createTransaction.getTransactionType(),
553 createTransaction.getTransactionId(), createTransaction.getTransactionChainId(),
554 createTransaction.getVersion());
556 getSender().tell(new CreateTransactionReply(Serialization.serializedActorPath(transactionActor),
557 createTransaction.getTransactionId()).toSerializable(), getSelf());
558 } catch (Exception e) {
559 getSender().tell(new akka.actor.Status.Failure(e), getSelf());
563 private ActorRef createTransaction(int transactionType, String remoteTransactionId,
564 String transactionChainId, int clientVersion) {
566 ShardTransactionIdentifier transactionId =
567 ShardTransactionIdentifier.builder()
568 .remoteTransactionId(remoteTransactionId)
571 if(LOG.isDebugEnabled()) {
572 LOG.debug("Creating transaction : {} ", transactionId);
575 ActorRef transactionActor = createTypedTransactionActor(transactionType, transactionId,
576 transactionChainId, clientVersion);
578 return transactionActor;
581 private void syncCommitTransaction(final DOMStoreWriteTransaction transaction)
582 throws ExecutionException, InterruptedException {
583 DOMStoreThreePhaseCommitCohort commitCohort = transaction.ready();
584 commitCohort.preCommit().get();
585 commitCohort.commit().get();
588 private void commitWithNewTransaction(final Modification modification) {
589 DOMStoreWriteTransaction tx = store.newWriteOnlyTransaction();
590 modification.apply(tx);
592 syncCommitTransaction(tx);
593 shardMBean.incrementCommittedTransactionCount();
594 shardMBean.setLastCommittedTransactionTime(System.currentTimeMillis());
595 } catch (InterruptedException | ExecutionException e) {
596 shardMBean.incrementFailedTransactionsCount();
597 LOG.error(e, "Failed to commit");
601 private void updateSchemaContext(final UpdateSchemaContext message) {
602 this.schemaContext = message.getSchemaContext();
603 updateSchemaContext(message.getSchemaContext());
604 store.onGlobalContextUpdated(message.getSchemaContext());
608 void updateSchemaContext(final SchemaContext schemaContext) {
609 store.onGlobalContextUpdated(schemaContext);
612 private void registerChangeListener(final RegisterChangeListener registerChangeListener) {
614 LOG.debug("registerDataChangeListener for {}", registerChangeListener.getPath());
616 ListenerRegistration<AsyncDataChangeListener<YangInstanceIdentifier,
617 NormalizedNode<?, ?>>> registration;
619 registration = doChangeListenerRegistration(registerChangeListener);
621 LOG.debug("Shard is not the leader - delaying registration");
623 DelayedListenerRegistration delayedReg =
624 new DelayedListenerRegistration(registerChangeListener);
625 delayedListenerRegistrations.add(delayedReg);
626 registration = delayedReg;
629 ActorRef listenerRegistration = getContext().actorOf(
630 DataChangeListenerRegistration.props(registration));
632 LOG.debug("registerDataChangeListener sending reply, listenerRegistrationPath = {} ",
633 listenerRegistration.path());
635 getSender().tell(new RegisterChangeListenerReply(listenerRegistration.path()), getSelf());
638 private ListenerRegistration<AsyncDataChangeListener<YangInstanceIdentifier,
639 NormalizedNode<?, ?>>> doChangeListenerRegistration(
640 final RegisterChangeListener registerChangeListener) {
642 ActorSelection dataChangeListenerPath = getContext().system().actorSelection(
643 registerChangeListener.getDataChangeListenerPath());
645 // Notify the listener if notifications should be enabled or not
646 // If this shard is the leader then it will enable notifications else
648 dataChangeListenerPath.tell(new EnableNotification(true), getSelf());
650 // Now store a reference to the data change listener so it can be notified
651 // at a later point if notifications should be enabled or disabled
652 dataChangeListeners.add(dataChangeListenerPath);
654 AsyncDataChangeListener<YangInstanceIdentifier, NormalizedNode<?, ?>> listener =
655 new DataChangeListenerProxy(schemaContext, dataChangeListenerPath);
657 LOG.debug("Registering for path {}", registerChangeListener.getPath());
659 return store.registerChangeListener(registerChangeListener.getPath(), listener,
660 registerChangeListener.getScope());
663 private boolean isMetricsCaptureEnabled(){
664 CommonConfig config = new CommonConfig(getContext().system().settings().config());
665 return config.isMetricCaptureEnabled();
670 void startLogRecoveryBatch(final int maxBatchSize) {
671 currentLogRecoveryBatch = Lists.newArrayListWithCapacity(maxBatchSize);
673 if(LOG.isDebugEnabled()) {
674 LOG.debug("{} : starting log recovery batch with max size {}", persistenceId(), maxBatchSize);
679 protected void appendRecoveredLogEntry(final Payload data) {
680 if (data instanceof CompositeModificationPayload) {
681 currentLogRecoveryBatch.add(((CompositeModificationPayload) data).getModification());
683 LOG.error("Unknown state received {} during recovery", data);
688 protected void applyRecoverySnapshot(final ByteString snapshot) {
689 if(recoveryCoordinator == null) {
690 recoveryCoordinator = new ShardRecoveryCoordinator(persistenceId(), schemaContext);
693 recoveryCoordinator.submit(snapshot, store.newWriteOnlyTransaction());
695 if(LOG.isDebugEnabled()) {
696 LOG.debug("{} : submitted recovery sbapshot", persistenceId());
701 protected void applyCurrentLogRecoveryBatch() {
702 if(recoveryCoordinator == null) {
703 recoveryCoordinator = new ShardRecoveryCoordinator(persistenceId(), schemaContext);
706 recoveryCoordinator.submit(currentLogRecoveryBatch, store.newWriteOnlyTransaction());
708 if(LOG.isDebugEnabled()) {
709 LOG.debug("{} : submitted log recovery batch with size {}", persistenceId(),
710 currentLogRecoveryBatch.size());
715 protected void onRecoveryComplete() {
716 if(recoveryCoordinator != null) {
717 Collection<DOMStoreWriteTransaction> txList = recoveryCoordinator.getTransactions();
719 if(LOG.isDebugEnabled()) {
720 LOG.debug("{} : recovery complete - committing {} Tx's", persistenceId(), txList.size());
723 for(DOMStoreWriteTransaction tx: txList) {
725 syncCommitTransaction(tx);
726 shardMBean.incrementCommittedTransactionCount();
727 } catch (InterruptedException | ExecutionException e) {
728 shardMBean.incrementFailedTransactionsCount();
729 LOG.error(e, "Failed to commit");
734 recoveryCoordinator = null;
735 currentLogRecoveryBatch = null;
736 updateJournalStats();
738 //notify shard manager
739 getContext().parent().tell(new ActorInitialized(), getSelf());
741 // Being paranoid here - this method should only be called once but just in case...
742 if(txCommitTimeoutCheckSchedule == null) {
743 // Schedule a message to be periodically sent to check if the current in-progress
744 // transaction should be expired and aborted.
745 FiniteDuration period = Duration.create(transactionCommitTimeout / 3, TimeUnit.MILLISECONDS);
746 txCommitTimeoutCheckSchedule = getContext().system().scheduler().schedule(
747 period, period, getSelf(),
748 TX_COMMIT_TIMEOUT_CHECK_MESSAGE, getContext().dispatcher(), ActorRef.noSender());
753 protected void applyState(final ActorRef clientActor, final String identifier, final Object data) {
755 if (data instanceof CompositeModificationPayload) {
756 Object modification = ((CompositeModificationPayload) data).getModification();
758 if(modification == null) {
760 "modification is null - this is very unexpected, clientActor = {}, identifier = {}",
761 identifier, clientActor != null ? clientActor.path().toString() : null);
762 } else if(clientActor == null) {
763 // There's no clientActor to which to send a commit reply so we must be applying
764 // replicated state from the leader.
765 commitWithNewTransaction(MutableCompositeModification.fromSerializable(
766 modification, schemaContext));
768 // This must be the OK to commit after replication consensus.
769 finishCommit(clientActor, identifier);
772 LOG.error("Unknown state received {} Class loader = {} CompositeNodeMod.ClassLoader = {}",
773 data, data.getClass().getClassLoader(),
774 CompositeModificationPayload.class.getClassLoader());
777 updateJournalStats();
781 private void updateJournalStats() {
782 ReplicatedLogEntry lastLogEntry = getLastLogEntry();
784 if (lastLogEntry != null) {
785 shardMBean.setLastLogIndex(lastLogEntry.getIndex());
786 shardMBean.setLastLogTerm(lastLogEntry.getTerm());
789 shardMBean.setCommitIndex(getCommitIndex());
790 shardMBean.setLastApplied(getLastApplied());
791 shardMBean.setInMemoryJournalDataSize(getRaftActorContext().getReplicatedLog().dataSize());
795 protected void createSnapshot() {
796 if (createSnapshotTransaction == null) {
798 // Create a transaction. We are really going to treat the transaction as a worker
799 // so that this actor does not get block building the snapshot
800 createSnapshotTransaction = createTransaction(
801 TransactionProxy.TransactionType.READ_ONLY.ordinal(),
802 "createSnapshot" + ++createSnapshotTransactionCounter, "",
803 CreateTransaction.CURRENT_VERSION);
805 createSnapshotTransaction.tell(
806 new ReadData(YangInstanceIdentifier.builder().build()).toSerializable(), self());
813 protected void applySnapshot(final ByteString snapshot) {
814 // Since this will be done only on Recovery or when this actor is a Follower
815 // we can safely commit everything in here. We not need to worry about event notifications
816 // as they would have already been disabled on the follower
818 LOG.info("Applying snapshot");
820 DOMStoreWriteTransaction transaction = store.newWriteOnlyTransaction();
821 NormalizedNodeMessages.Node serializedNode = NormalizedNodeMessages.Node.parseFrom(snapshot);
822 NormalizedNode<?, ?> node = new NormalizedNodeToNodeCodec(schemaContext)
823 .decode(serializedNode);
825 // delete everything first
826 transaction.delete(YangInstanceIdentifier.builder().build());
828 // Add everything from the remote node back
829 transaction.write(YangInstanceIdentifier.builder().build(), node);
830 syncCommitTransaction(transaction);
831 } catch (InvalidProtocolBufferException | InterruptedException | ExecutionException e) {
832 LOG.error(e, "An exception occurred when applying snapshot");
834 LOG.info("Done applying snapshot");
839 protected void onStateChanged() {
840 boolean isLeader = isLeader();
841 for (ActorSelection dataChangeListener : dataChangeListeners) {
842 dataChangeListener.tell(new EnableNotification(isLeader), getSelf());
846 for(DelayedListenerRegistration reg: delayedListenerRegistrations) {
847 if(!reg.isClosed()) {
848 reg.setDelegate(doChangeListenerRegistration(reg.getRegisterChangeListener()));
852 delayedListenerRegistrations.clear();
855 shardMBean.setRaftState(getRaftState().name());
856 shardMBean.setCurrentTerm(getCurrentTerm());
858 // If this actor is no longer the leader close all the transaction chains
860 for(Map.Entry<String, DOMStoreTransactionChain> entry : transactionChains.entrySet()){
861 if(LOG.isDebugEnabled()) {
863 "onStateChanged: Closing transaction chain {} because shard {} is no longer the leader",
864 entry.getKey(), getId());
866 entry.getValue().close();
869 transactionChains.clear();
874 protected DataPersistenceProvider persistence() {
875 return dataPersistenceProvider;
878 @Override protected void onLeaderChanged(final String oldLeader, final String newLeader) {
879 shardMBean.setLeader(newLeader);
882 @Override public String persistenceId() {
883 return this.name.toString();
887 DataPersistenceProvider getDataPersistenceProvider() {
888 return dataPersistenceProvider;
891 private static class ShardCreator implements Creator<Shard> {
893 private static final long serialVersionUID = 1L;
895 final ShardIdentifier name;
896 final Map<ShardIdentifier, String> peerAddresses;
897 final DatastoreContext datastoreContext;
898 final SchemaContext schemaContext;
900 ShardCreator(final ShardIdentifier name, final Map<ShardIdentifier, String> peerAddresses,
901 final DatastoreContext datastoreContext, final SchemaContext schemaContext) {
903 this.peerAddresses = peerAddresses;
904 this.datastoreContext = datastoreContext;
905 this.schemaContext = schemaContext;
909 public Shard create() throws Exception {
910 return new Shard(name, peerAddresses, datastoreContext, schemaContext);
915 InMemoryDOMDataStore getDataStore() {
920 ShardStats getShardMBean() {
924 private static class DelayedListenerRegistration implements
925 ListenerRegistration<AsyncDataChangeListener<YangInstanceIdentifier, NormalizedNode<?, ?>>> {
927 private volatile boolean closed;
929 private final RegisterChangeListener registerChangeListener;
931 private volatile ListenerRegistration<AsyncDataChangeListener<YangInstanceIdentifier,
932 NormalizedNode<?, ?>>> delegate;
934 DelayedListenerRegistration(final RegisterChangeListener registerChangeListener) {
935 this.registerChangeListener = registerChangeListener;
938 void setDelegate( final ListenerRegistration<AsyncDataChangeListener<YangInstanceIdentifier,
939 NormalizedNode<?, ?>>> registration) {
940 this.delegate = registration;
947 RegisterChangeListener getRegisterChangeListener() {
948 return registerChangeListener;
952 public AsyncDataChangeListener<YangInstanceIdentifier, NormalizedNode<?, ?>> getInstance() {
953 return delegate != null ? delegate.getInstance() : null;
957 public void close() {
959 if(delegate != null) {