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.actor.Status;
16 import akka.actor.Status.Failure;
17 import akka.serialization.Serialization;
18 import com.google.common.annotations.VisibleForTesting;
19 import com.google.common.base.Optional;
20 import com.google.common.base.Preconditions;
21 import com.google.common.base.Ticker;
22 import com.google.common.base.Verify;
23 import com.google.common.collect.ImmutableList;
24 import com.google.common.collect.ImmutableMap;
25 import com.google.common.collect.Range;
26 import java.io.IOException;
27 import java.util.Arrays;
28 import java.util.Collection;
29 import java.util.Collections;
31 import java.util.concurrent.TimeUnit;
32 import javax.annotation.Nonnull;
33 import javax.annotation.Nullable;
34 import org.opendaylight.controller.cluster.access.ABIVersion;
35 import org.opendaylight.controller.cluster.access.commands.ConnectClientRequest;
36 import org.opendaylight.controller.cluster.access.commands.ConnectClientSuccess;
37 import org.opendaylight.controller.cluster.access.commands.LocalHistoryRequest;
38 import org.opendaylight.controller.cluster.access.commands.NotLeaderException;
39 import org.opendaylight.controller.cluster.access.commands.TransactionRequest;
40 import org.opendaylight.controller.cluster.access.concepts.ClientIdentifier;
41 import org.opendaylight.controller.cluster.access.concepts.FrontendIdentifier;
42 import org.opendaylight.controller.cluster.access.concepts.LocalHistoryIdentifier;
43 import org.opendaylight.controller.cluster.access.concepts.Request;
44 import org.opendaylight.controller.cluster.access.concepts.RequestEnvelope;
45 import org.opendaylight.controller.cluster.access.concepts.RequestException;
46 import org.opendaylight.controller.cluster.access.concepts.RequestSuccess;
47 import org.opendaylight.controller.cluster.access.concepts.RetiredGenerationException;
48 import org.opendaylight.controller.cluster.access.concepts.RuntimeRequestException;
49 import org.opendaylight.controller.cluster.access.concepts.SliceableMessage;
50 import org.opendaylight.controller.cluster.access.concepts.TransactionIdentifier;
51 import org.opendaylight.controller.cluster.access.concepts.UnsupportedRequestException;
52 import org.opendaylight.controller.cluster.common.actor.CommonConfig;
53 import org.opendaylight.controller.cluster.common.actor.Dispatchers;
54 import org.opendaylight.controller.cluster.common.actor.Dispatchers.DispatcherType;
55 import org.opendaylight.controller.cluster.common.actor.MessageTracker;
56 import org.opendaylight.controller.cluster.common.actor.MessageTracker.Error;
57 import org.opendaylight.controller.cluster.common.actor.MeteringBehavior;
58 import org.opendaylight.controller.cluster.datastore.exceptions.NoShardLeaderException;
59 import org.opendaylight.controller.cluster.datastore.identifiers.ShardIdentifier;
60 import org.opendaylight.controller.cluster.datastore.jmx.mbeans.shard.ShardMBeanFactory;
61 import org.opendaylight.controller.cluster.datastore.jmx.mbeans.shard.ShardStats;
62 import org.opendaylight.controller.cluster.datastore.messages.AbortTransaction;
63 import org.opendaylight.controller.cluster.datastore.messages.ActorInitialized;
64 import org.opendaylight.controller.cluster.datastore.messages.BatchedModifications;
65 import org.opendaylight.controller.cluster.datastore.messages.CanCommitTransaction;
66 import org.opendaylight.controller.cluster.datastore.messages.CloseTransactionChain;
67 import org.opendaylight.controller.cluster.datastore.messages.CommitTransaction;
68 import org.opendaylight.controller.cluster.datastore.messages.CreateTransaction;
69 import org.opendaylight.controller.cluster.datastore.messages.CreateTransactionReply;
70 import org.opendaylight.controller.cluster.datastore.messages.ForwardedReadyTransaction;
71 import org.opendaylight.controller.cluster.datastore.messages.GetShardDataTree;
72 import org.opendaylight.controller.cluster.datastore.messages.MakeLeaderLocal;
73 import org.opendaylight.controller.cluster.datastore.messages.OnDemandShardState;
74 import org.opendaylight.controller.cluster.datastore.messages.PeerAddressResolved;
75 import org.opendaylight.controller.cluster.datastore.messages.PersistAbortTransactionPayload;
76 import org.opendaylight.controller.cluster.datastore.messages.ReadyLocalTransaction;
77 import org.opendaylight.controller.cluster.datastore.messages.RegisterChangeListener;
78 import org.opendaylight.controller.cluster.datastore.messages.RegisterDataTreeChangeListener;
79 import org.opendaylight.controller.cluster.datastore.messages.ShardLeaderStateChanged;
80 import org.opendaylight.controller.cluster.datastore.messages.UpdateSchemaContext;
81 import org.opendaylight.controller.cluster.datastore.persisted.AbortTransactionPayload;
82 import org.opendaylight.controller.cluster.datastore.persisted.DatastoreSnapshot;
83 import org.opendaylight.controller.cluster.datastore.persisted.DatastoreSnapshot.ShardSnapshot;
84 import org.opendaylight.controller.cluster.messaging.MessageSlicer;
85 import org.opendaylight.controller.cluster.messaging.SliceOptions;
86 import org.opendaylight.controller.cluster.notifications.LeaderStateChanged;
87 import org.opendaylight.controller.cluster.notifications.RegisterRoleChangeListener;
88 import org.opendaylight.controller.cluster.notifications.RoleChangeNotifier;
89 import org.opendaylight.controller.cluster.raft.LeadershipTransferFailedException;
90 import org.opendaylight.controller.cluster.raft.RaftActor;
91 import org.opendaylight.controller.cluster.raft.RaftActorRecoveryCohort;
92 import org.opendaylight.controller.cluster.raft.RaftActorSnapshotCohort;
93 import org.opendaylight.controller.cluster.raft.RaftState;
94 import org.opendaylight.controller.cluster.raft.base.messages.FollowerInitialSyncUpStatus;
95 import org.opendaylight.controller.cluster.raft.client.messages.OnDemandRaftState;
96 import org.opendaylight.controller.cluster.raft.messages.AppendEntriesReply;
97 import org.opendaylight.controller.cluster.raft.messages.RequestLeadership;
98 import org.opendaylight.controller.cluster.raft.messages.ServerRemoved;
99 import org.opendaylight.controller.cluster.raft.protobuff.client.messages.Payload;
100 import org.opendaylight.yangtools.concepts.Identifier;
101 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataValidationFailedException;
102 import org.opendaylight.yangtools.yang.data.api.schema.tree.TipProducingDataTree;
103 import org.opendaylight.yangtools.yang.data.api.schema.tree.TreeType;
104 import org.opendaylight.yangtools.yang.model.api.SchemaContext;
105 import org.opendaylight.yangtools.yang.model.api.SchemaContextProvider;
106 import scala.concurrent.duration.Duration;
107 import scala.concurrent.duration.FiniteDuration;
110 * A Shard represents a portion of the logical data tree.
113 * Our Shard uses InMemoryDataTree as it's internal representation and delegates all requests it
115 public class Shard extends RaftActor {
118 static final Object TX_COMMIT_TIMEOUT_CHECK_MESSAGE = new Object() {
120 public String toString() {
121 return "txCommitTimeoutCheck";
126 static final Object GET_SHARD_MBEAN_MESSAGE = new Object() {
128 public String toString() {
129 return "getShardMBeanMessage";
133 static final Object RESUME_NEXT_PENDING_TRANSACTION = new Object() {
135 public String toString() {
136 return "resumeNextPendingTransaction";
140 // FIXME: shard names should be encapsulated in their own class and this should be exposed as a constant.
141 public static final String DEFAULT_NAME = "default";
143 private static final Collection<ABIVersion> SUPPORTED_ABIVERSIONS;
146 final ABIVersion[] values = ABIVersion.values();
147 final ABIVersion[] real = Arrays.copyOfRange(values, 1, values.length - 1);
148 SUPPORTED_ABIVERSIONS = ImmutableList.copyOf(real).reverse();
151 // FIXME: make this a dynamic property based on mailbox size and maximum number of clients
152 private static final int CLIENT_MAX_MESSAGES = 1000;
154 // The state of this Shard
155 private final ShardDataTree store;
157 /// The name of this shard
158 private final String name;
160 private final ShardStats shardMBean;
162 private DatastoreContext datastoreContext;
164 private final ShardCommitCoordinator commitCoordinator;
166 private long transactionCommitTimeout;
168 private Cancellable txCommitTimeoutCheckSchedule;
170 private final Optional<ActorRef> roleChangeNotifier;
172 private final MessageTracker appendEntriesReplyTracker;
174 private final ShardTransactionActorFactory transactionActorFactory;
176 private final ShardSnapshotCohort snapshotCohort;
178 private final DataTreeChangeListenerSupport treeChangeSupport = new DataTreeChangeListenerSupport(this);
179 private final DataChangeListenerSupport changeSupport = new DataChangeListenerSupport(this);
182 private ShardSnapshot restoreFromSnapshot;
184 private final ShardTransactionMessageRetrySupport messageRetrySupport;
186 private final FrontendMetadata frontendMetadata;
187 private Map<FrontendIdentifier, LeaderFrontendState> knownFrontends = ImmutableMap.of();
189 private final MessageSlicer responseMessageSlicer;
190 private final Dispatchers dispatchers;
192 protected Shard(final AbstractBuilder<?, ?> builder) {
193 super(builder.getId().toString(), builder.getPeerAddresses(),
194 Optional.of(builder.getDatastoreContext().getShardRaftConfig()), DataStoreVersions.CURRENT_VERSION);
196 this.name = builder.getId().toString();
197 this.datastoreContext = builder.getDatastoreContext();
198 this.restoreFromSnapshot = builder.getRestoreFromSnapshot();
199 this.frontendMetadata = new FrontendMetadata(name);
201 setPersistence(datastoreContext.isPersistent());
203 LOG.info("Shard created : {}, persistent : {}", name, datastoreContext.isPersistent());
205 ShardDataTreeChangeListenerPublisherActorProxy treeChangeListenerPublisher =
206 new ShardDataTreeChangeListenerPublisherActorProxy(getContext(), name + "-DTCL-publisher", name);
207 ShardDataChangeListenerPublisherActorProxy dataChangeListenerPublisher =
208 new ShardDataChangeListenerPublisherActorProxy(getContext(), name + "-DCL-publisher", name);
209 if (builder.getDataTree() != null) {
210 store = new ShardDataTree(this, builder.getSchemaContext(), builder.getDataTree(),
211 treeChangeListenerPublisher, dataChangeListenerPublisher, name, frontendMetadata);
213 store = new ShardDataTree(this, builder.getSchemaContext(), builder.getTreeType(),
214 builder.getDatastoreContext().getStoreRoot(), treeChangeListenerPublisher,
215 dataChangeListenerPublisher, name, frontendMetadata);
218 shardMBean = ShardMBeanFactory.getShardStatsMBean(name, datastoreContext.getDataStoreMXBeanType(), this);
220 if (isMetricsCaptureEnabled()) {
221 getContext().become(new MeteringBehavior(this));
224 commitCoordinator = new ShardCommitCoordinator(store, LOG, this.name);
226 setTransactionCommitTimeout();
228 // create a notifier actor for each cluster member
229 roleChangeNotifier = createRoleChangeNotifier(name);
231 appendEntriesReplyTracker = new MessageTracker(AppendEntriesReply.class,
232 getRaftActorContext().getConfigParams().getIsolatedCheckIntervalInMillis());
234 dispatchers = new Dispatchers(context().system().dispatchers());
235 transactionActorFactory = new ShardTransactionActorFactory(store, datastoreContext,
236 dispatchers.getDispatcherPath(Dispatchers.DispatcherType.Transaction),
237 self(), getContext(), shardMBean, builder.getId().getShardName());
239 snapshotCohort = ShardSnapshotCohort.create(getContext(), builder.getId().getMemberName(), store, LOG,
242 messageRetrySupport = new ShardTransactionMessageRetrySupport(this);
244 responseMessageSlicer = MessageSlicer.builder().logContext(this.name)
245 .messageSliceSize(datastoreContext.getMaximumMessageSliceSize())
246 .fileBackedStreamFactory(getRaftActorContext().getFileBackedOutputStreamFactory())
247 .expireStateAfterInactivity(2, TimeUnit.MINUTES).build();
250 private void setTransactionCommitTimeout() {
251 transactionCommitTimeout = TimeUnit.MILLISECONDS.convert(
252 datastoreContext.getShardTransactionCommitTimeoutInSeconds(), TimeUnit.SECONDS) / 2;
255 private Optional<ActorRef> createRoleChangeNotifier(final String shardId) {
256 ActorRef shardRoleChangeNotifier = this.getContext().actorOf(
257 RoleChangeNotifier.getProps(shardId), shardId + "-notifier");
258 return Optional.of(shardRoleChangeNotifier);
262 public void postStop() {
263 LOG.info("Stopping Shard {}", persistenceId());
267 messageRetrySupport.close();
269 if (txCommitTimeoutCheckSchedule != null) {
270 txCommitTimeoutCheckSchedule.cancel();
273 commitCoordinator.abortPendingTransactions("Transaction aborted due to shutdown.", this);
275 shardMBean.unregisterMBean();
279 protected void handleRecover(final Object message) {
280 LOG.debug("{}: onReceiveRecover: Received message {} from {}", persistenceId(), message.getClass(),
283 super.handleRecover(message);
284 if (LOG.isTraceEnabled()) {
285 appendEntriesReplyTracker.begin();
290 protected void handleNonRaftCommand(final Object message) {
291 try (MessageTracker.Context context = appendEntriesReplyTracker.received(message)) {
292 final Optional<Error> maybeError = context.error();
293 if (maybeError.isPresent()) {
294 LOG.trace("{} : AppendEntriesReply failed to arrive at the expected interval {}", persistenceId(),
298 store.resetTransactionBatch();
300 if (message instanceof RequestEnvelope) {
301 handleRequestEnvelope((RequestEnvelope)message);
302 } else if (message instanceof ConnectClientRequest) {
303 handleConnectClient((ConnectClientRequest)message);
304 } else if (CreateTransaction.isSerializedType(message)) {
305 handleCreateTransaction(message);
306 } else if (message instanceof BatchedModifications) {
307 handleBatchedModifications((BatchedModifications)message);
308 } else if (message instanceof ForwardedReadyTransaction) {
309 handleForwardedReadyTransaction((ForwardedReadyTransaction) message);
310 } else if (message instanceof ReadyLocalTransaction) {
311 handleReadyLocalTransaction((ReadyLocalTransaction)message);
312 } else if (CanCommitTransaction.isSerializedType(message)) {
313 handleCanCommitTransaction(CanCommitTransaction.fromSerializable(message));
314 } else if (CommitTransaction.isSerializedType(message)) {
315 handleCommitTransaction(CommitTransaction.fromSerializable(message));
316 } else if (AbortTransaction.isSerializedType(message)) {
317 handleAbortTransaction(AbortTransaction.fromSerializable(message));
318 } else if (CloseTransactionChain.isSerializedType(message)) {
319 closeTransactionChain(CloseTransactionChain.fromSerializable(message));
320 } else if (message instanceof RegisterChangeListener) {
321 changeSupport.onMessage((RegisterChangeListener) message, isLeader(), hasLeader());
322 } else if (message instanceof RegisterDataTreeChangeListener) {
323 treeChangeSupport.onMessage((RegisterDataTreeChangeListener) message, isLeader(), hasLeader());
324 } else if (message instanceof UpdateSchemaContext) {
325 updateSchemaContext((UpdateSchemaContext) message);
326 } else if (message instanceof PeerAddressResolved) {
327 PeerAddressResolved resolved = (PeerAddressResolved) message;
328 setPeerAddress(resolved.getPeerId(), resolved.getPeerAddress());
329 } else if (TX_COMMIT_TIMEOUT_CHECK_MESSAGE.equals(message)) {
330 store.checkForExpiredTransactions(transactionCommitTimeout);
331 commitCoordinator.checkForExpiredTransactions(transactionCommitTimeout, this);
332 } else if (message instanceof DatastoreContext) {
333 onDatastoreContext((DatastoreContext)message);
334 } else if (message instanceof RegisterRoleChangeListener) {
335 roleChangeNotifier.get().forward(message, context());
336 } else if (message instanceof FollowerInitialSyncUpStatus) {
337 shardMBean.setFollowerInitialSyncStatus(((FollowerInitialSyncUpStatus) message).isInitialSyncDone());
338 context().parent().tell(message, self());
339 } else if (GET_SHARD_MBEAN_MESSAGE.equals(message)) {
340 sender().tell(getShardMBean(), self());
341 } else if (message instanceof GetShardDataTree) {
342 sender().tell(store.getDataTree(), self());
343 } else if (message instanceof ServerRemoved) {
344 context().parent().forward(message, context());
345 } else if (ShardTransactionMessageRetrySupport.TIMER_MESSAGE_CLASS.isInstance(message)) {
346 messageRetrySupport.onTimerMessage(message);
347 } else if (message instanceof DataTreeCohortActorRegistry.CohortRegistryCommand) {
348 store.processCohortRegistryCommand(getSender(),
349 (DataTreeCohortActorRegistry.CohortRegistryCommand) message);
350 } else if (message instanceof PersistAbortTransactionPayload) {
351 final TransactionIdentifier txId = ((PersistAbortTransactionPayload) message).getTransactionId();
352 persistPayload(txId, AbortTransactionPayload.create(txId), true);
353 } else if (message instanceof MakeLeaderLocal) {
355 } else if (RESUME_NEXT_PENDING_TRANSACTION.equals(message)) {
356 store.resumeNextPendingTransaction();
357 } else if (!responseMessageSlicer.handleMessage(message)) {
358 super.handleNonRaftCommand(message);
363 @SuppressWarnings("checkstyle:IllegalCatch")
364 private void handleRequestEnvelope(final RequestEnvelope envelope) {
365 final long now = ticker().read();
367 final RequestSuccess<?, ?> success = handleRequest(envelope, now);
368 if (success != null) {
369 final long executionTimeNanos = ticker().read() - now;
370 if (success instanceof SliceableMessage) {
371 dispatchers.getDispatcher(DispatcherType.Serialization).execute(() ->
372 responseMessageSlicer.slice(SliceOptions.builder().identifier(success.getTarget())
373 .message(envelope.newSuccessEnvelope(success, executionTimeNanos))
374 .sendTo(envelope.getMessage().getReplyTo()).replyTo(self())
375 .onFailureCallback(t -> {
376 LOG.warn("Error slicing response {}", success, t);
379 envelope.sendSuccess(success, executionTimeNanos);
382 } catch (RequestException e) {
383 LOG.debug("{}: request {} failed", persistenceId(), envelope, e);
384 envelope.sendFailure(e, ticker().read() - now);
385 } catch (Exception e) {
386 LOG.debug("{}: request {} caused failure", persistenceId(), envelope, e);
387 envelope.sendFailure(new RuntimeRequestException("Request failed to process", e),
388 ticker().read() - now);
392 private void onMakeLeaderLocal() {
393 LOG.debug("{}: onMakeLeaderLocal received", persistenceId());
395 getSender().tell(new Status.Success(null), getSelf());
399 final ActorSelection leader = getLeader();
401 if (leader == null) {
402 // Leader is not present. The cluster is most likely trying to
403 // elect a leader and we should let that run its normal course
405 // TODO we can wait for the election to complete and retry the
406 // request. We can also let the caller retry by sending a flag
407 // in the response indicating the request is "reTryable".
408 getSender().tell(new Failure(
409 new LeadershipTransferFailedException("We cannot initiate leadership transfer to local node. "
410 + "Currently there is no leader for " + persistenceId())),
415 leader.tell(new RequestLeadership(getId(), getSender()), getSelf());
418 // Acquire our frontend tracking handle and verify generation matches
419 private LeaderFrontendState getFrontend(final ClientIdentifier clientId) throws RequestException {
420 final LeaderFrontendState existing = knownFrontends.get(clientId.getFrontendId());
421 if (existing != null) {
422 final int cmp = Long.compareUnsigned(existing.getIdentifier().getGeneration(), clientId.getGeneration());
427 LOG.debug("{}: rejecting request from outdated client {}", persistenceId(), clientId);
428 throw new RetiredGenerationException(existing.getIdentifier().getGeneration());
431 LOG.info("{}: retiring state {}, outdated by request from client {}", persistenceId(), existing, clientId);
433 knownFrontends.remove(clientId.getFrontendId());
435 LOG.debug("{}: client {} is not yet known", persistenceId(), clientId);
438 final LeaderFrontendState ret = new LeaderFrontendState(persistenceId(), clientId, store);
439 knownFrontends.put(clientId.getFrontendId(), ret);
440 LOG.debug("{}: created state {} for client {}", persistenceId(), ret, clientId);
444 private static @Nonnull ABIVersion selectVersion(final ConnectClientRequest message) {
445 final Range<ABIVersion> clientRange = Range.closed(message.getMinVersion(), message.getMaxVersion());
446 for (ABIVersion v : SUPPORTED_ABIVERSIONS) {
447 if (clientRange.contains(v)) {
452 throw new IllegalArgumentException(String.format(
453 "No common version between backend versions %s and client versions %s", SUPPORTED_ABIVERSIONS,
457 @SuppressWarnings("checkstyle:IllegalCatch")
458 private void handleConnectClient(final ConnectClientRequest message) {
460 if (!isLeader() || !isLeaderActive()) {
461 LOG.info("{}: not currently leader, rejecting request {}. isLeader: {}, isLeaderActive: {},"
462 + "isLeadershipTransferInProgress: {}.",
463 persistenceId(), message, isLeader(), isLeaderActive(), isLeadershipTransferInProgress());
464 throw new NotLeaderException(getSelf());
467 final ABIVersion selectedVersion = selectVersion(message);
468 final LeaderFrontendState frontend = getFrontend(message.getTarget());
469 frontend.reconnect();
470 message.getReplyTo().tell(new ConnectClientSuccess(message.getTarget(), message.getSequence(), getSelf(),
471 ImmutableList.of(), store.getDataTree(), CLIENT_MAX_MESSAGES).toVersion(selectedVersion),
472 ActorRef.noSender());
473 } catch (RequestException | RuntimeException e) {
474 message.getReplyTo().tell(new Failure(e), ActorRef.noSender());
478 private @Nullable RequestSuccess<?, ?> handleRequest(final RequestEnvelope envelope, final long now)
479 throws RequestException {
480 // We are not the leader, hence we want to fail-fast.
481 if (!isLeader() || !isLeaderActive()) {
482 LOG.info("{}: not currently leader, rejecting request {}. isLeader: {}, isLeaderActive: {},"
483 + "isLeadershipTransferInProgress: {}.",
484 persistenceId(), envelope, isLeader(), isLeaderActive(), isLeadershipTransferInProgress());
485 throw new NotLeaderException(getSelf());
488 final Request<?, ?> request = envelope.getMessage();
489 if (request instanceof TransactionRequest) {
490 final TransactionRequest<?> txReq = (TransactionRequest<?>)request;
491 final ClientIdentifier clientId = txReq.getTarget().getHistoryId().getClientId();
492 return getFrontend(clientId).handleTransactionRequest(txReq, envelope, now);
493 } else if (request instanceof LocalHistoryRequest) {
494 final LocalHistoryRequest<?> lhReq = (LocalHistoryRequest<?>)request;
495 final ClientIdentifier clientId = lhReq.getTarget().getClientId();
496 return getFrontend(clientId).handleLocalHistoryRequest(lhReq, envelope, now);
498 LOG.warn("{}: rejecting unsupported request {}", persistenceId(), request);
499 throw new UnsupportedRequestException(request);
503 private boolean hasLeader() {
504 return getLeaderId() != null;
507 public int getPendingTxCommitQueueSize() {
508 return store.getQueueSize();
511 public int getCohortCacheSize() {
512 return commitCoordinator.getCohortCacheSize();
516 protected Optional<ActorRef> getRoleChangeNotifier() {
517 return roleChangeNotifier;
521 protected LeaderStateChanged newLeaderStateChanged(final String memberId, final String leaderId,
522 final short leaderPayloadVersion) {
523 return isLeader() ? new ShardLeaderStateChanged(memberId, leaderId, store.getDataTree(), leaderPayloadVersion)
524 : new ShardLeaderStateChanged(memberId, leaderId, leaderPayloadVersion);
527 protected void onDatastoreContext(final DatastoreContext context) {
528 datastoreContext = context;
530 setTransactionCommitTimeout();
532 setPersistence(datastoreContext.isPersistent());
534 updateConfigParams(datastoreContext.getShardRaftConfig());
537 // applyState() will be invoked once consensus is reached on the payload
538 void persistPayload(final Identifier id, final Payload payload, final boolean batchHint) {
539 boolean canSkipPayload = !hasFollowers() && !persistence().isRecoveryApplicable();
540 if (canSkipPayload) {
541 applyState(self(), id, payload);
543 // We are faking the sender
544 persistData(self(), id, payload, batchHint);
548 private void handleCommitTransaction(final CommitTransaction commit) {
550 commitCoordinator.handleCommit(commit.getTransactionId(), getSender(), this);
552 ActorSelection leader = getLeader();
553 if (leader == null) {
554 messageRetrySupport.addMessageToRetry(commit, getSender(),
555 "Could not commit transaction " + commit.getTransactionId());
557 LOG.debug("{}: Forwarding CommitTransaction to leader {}", persistenceId(), leader);
558 leader.forward(commit, getContext());
563 private void handleCanCommitTransaction(final CanCommitTransaction canCommit) {
564 LOG.debug("{}: Can committing transaction {}", persistenceId(), canCommit.getTransactionId());
567 commitCoordinator.handleCanCommit(canCommit.getTransactionId(), getSender(), this);
569 ActorSelection leader = getLeader();
570 if (leader == null) {
571 messageRetrySupport.addMessageToRetry(canCommit, getSender(),
572 "Could not canCommit transaction " + canCommit.getTransactionId());
574 LOG.debug("{}: Forwarding CanCommitTransaction to leader {}", persistenceId(), leader);
575 leader.forward(canCommit, getContext());
580 @SuppressWarnings("checkstyle:IllegalCatch")
581 protected void handleBatchedModificationsLocal(final BatchedModifications batched, final ActorRef sender) {
583 commitCoordinator.handleBatchedModifications(batched, sender, this);
584 } catch (Exception e) {
585 LOG.error("{}: Error handling BatchedModifications for Tx {}", persistenceId(),
586 batched.getTransactionId(), e);
587 sender.tell(new Failure(e), getSelf());
591 private void handleBatchedModifications(final BatchedModifications batched) {
592 // This message is sent to prepare the modifications transaction directly on the Shard as an
593 // optimization to avoid the extra overhead of a separate ShardTransaction actor. On the last
594 // BatchedModifications message, the caller sets the ready flag in the message indicating
595 // modifications are complete. The reply contains the cohort actor path (this actor) for the caller
596 // to initiate the 3-phase commit. This also avoids the overhead of sending an additional
597 // ReadyTransaction message.
599 // If we're not the leader then forward to the leader. This is a safety measure - we shouldn't
600 // normally get here if we're not the leader as the front-end (TransactionProxy) should determine
601 // the primary/leader shard. However with timing and caching on the front-end, there's a small
602 // window where it could have a stale leader during leadership transitions.
604 boolean isLeaderActive = isLeaderActive();
605 if (isLeader() && isLeaderActive) {
606 handleBatchedModificationsLocal(batched, getSender());
608 ActorSelection leader = getLeader();
609 if (!isLeaderActive || leader == null) {
610 messageRetrySupport.addMessageToRetry(batched, getSender(),
611 "Could not commit transaction " + batched.getTransactionId());
613 // If this is not the first batch and leadership changed in between batched messages,
614 // we need to reconstruct previous BatchedModifications from the transaction
615 // DataTreeModification, honoring the max batched modification count, and forward all the
616 // previous BatchedModifications to the new leader.
617 Collection<BatchedModifications> newModifications = commitCoordinator
618 .createForwardedBatchedModifications(batched,
619 datastoreContext.getShardBatchedModificationCount());
621 LOG.debug("{}: Forwarding {} BatchedModifications to leader {}", persistenceId(),
622 newModifications.size(), leader);
624 for (BatchedModifications bm : newModifications) {
625 leader.forward(bm, getContext());
631 private boolean failIfIsolatedLeader(final ActorRef sender) {
632 if (isIsolatedLeader()) {
633 sender.tell(new Failure(new NoShardLeaderException(String.format(
634 "Shard %s was the leader but has lost contact with all of its followers. Either all"
635 + " other follower nodes are down or this node is isolated by a network partition.",
636 persistenceId()))), getSelf());
643 protected boolean isIsolatedLeader() {
644 return getRaftState() == RaftState.IsolatedLeader;
647 @SuppressWarnings("checkstyle:IllegalCatch")
648 private void handleReadyLocalTransaction(final ReadyLocalTransaction message) {
649 LOG.debug("{}: handleReadyLocalTransaction for {}", persistenceId(), message.getTransactionId());
651 boolean isLeaderActive = isLeaderActive();
652 if (isLeader() && isLeaderActive) {
654 commitCoordinator.handleReadyLocalTransaction(message, getSender(), this);
655 } catch (Exception e) {
656 LOG.error("{}: Error handling ReadyLocalTransaction for Tx {}", persistenceId(),
657 message.getTransactionId(), e);
658 getSender().tell(new Failure(e), getSelf());
661 ActorSelection leader = getLeader();
662 if (!isLeaderActive || leader == null) {
663 messageRetrySupport.addMessageToRetry(message, getSender(),
664 "Could not commit transaction " + message.getTransactionId());
666 LOG.debug("{}: Forwarding ReadyLocalTransaction to leader {}", persistenceId(), leader);
667 message.setRemoteVersion(getCurrentBehavior().getLeaderPayloadVersion());
668 leader.forward(message, getContext());
673 private void handleForwardedReadyTransaction(final ForwardedReadyTransaction forwardedReady) {
674 LOG.debug("{}: handleForwardedReadyTransaction for {}", persistenceId(), forwardedReady.getTransactionId());
676 boolean isLeaderActive = isLeaderActive();
677 if (isLeader() && isLeaderActive) {
678 commitCoordinator.handleForwardedReadyTransaction(forwardedReady, getSender(), this);
680 ActorSelection leader = getLeader();
681 if (!isLeaderActive || leader == null) {
682 messageRetrySupport.addMessageToRetry(forwardedReady, getSender(),
683 "Could not commit transaction " + forwardedReady.getTransactionId());
685 LOG.debug("{}: Forwarding ForwardedReadyTransaction to leader {}", persistenceId(), leader);
687 ReadyLocalTransaction readyLocal = new ReadyLocalTransaction(forwardedReady.getTransactionId(),
688 forwardedReady.getTransaction().getSnapshot(), forwardedReady.isDoImmediateCommit());
689 readyLocal.setRemoteVersion(getCurrentBehavior().getLeaderPayloadVersion());
690 leader.forward(readyLocal, getContext());
695 private void handleAbortTransaction(final AbortTransaction abort) {
696 doAbortTransaction(abort.getTransactionId(), getSender());
699 void doAbortTransaction(final Identifier transactionID, final ActorRef sender) {
700 commitCoordinator.handleAbort(transactionID, sender, this);
703 private void handleCreateTransaction(final Object message) {
705 createTransaction(CreateTransaction.fromSerializable(message));
706 } else if (getLeader() != null) {
707 getLeader().forward(message, getContext());
709 getSender().tell(new Failure(new NoShardLeaderException(
710 "Could not create a shard transaction", persistenceId())), getSelf());
714 private void closeTransactionChain(final CloseTransactionChain closeTransactionChain) {
715 final LocalHistoryIdentifier id = closeTransactionChain.getIdentifier();
716 store.closeTransactionChain(id, null);
717 store.purgeTransactionChain(id, null);
720 @SuppressWarnings("checkstyle:IllegalCatch")
721 private void createTransaction(final CreateTransaction createTransaction) {
723 if (TransactionType.fromInt(createTransaction.getTransactionType()) != TransactionType.READ_ONLY
724 && failIfIsolatedLeader(getSender())) {
728 ActorRef transactionActor = createTransaction(createTransaction.getTransactionType(),
729 createTransaction.getTransactionId());
731 getSender().tell(new CreateTransactionReply(Serialization.serializedActorPath(transactionActor),
732 createTransaction.getTransactionId(), createTransaction.getVersion()).toSerializable(), getSelf());
733 } catch (Exception e) {
734 getSender().tell(new Failure(e), getSelf());
738 private ActorRef createTransaction(final int transactionType, final TransactionIdentifier transactionId) {
739 LOG.debug("{}: Creating transaction : {} ", persistenceId(), transactionId);
740 return transactionActorFactory.newShardTransaction(TransactionType.fromInt(transactionType),
744 private void updateSchemaContext(final UpdateSchemaContext message) {
745 updateSchemaContext(message.getSchemaContext());
749 void updateSchemaContext(final SchemaContext schemaContext) {
750 store.updateSchemaContext(schemaContext);
753 private boolean isMetricsCaptureEnabled() {
754 CommonConfig config = new CommonConfig(getContext().system().settings().config());
755 return config.isMetricCaptureEnabled();
760 public RaftActorSnapshotCohort getRaftActorSnapshotCohort() {
761 return snapshotCohort;
766 protected RaftActorRecoveryCohort getRaftActorRecoveryCohort() {
767 return new ShardRecoveryCoordinator(store,
768 restoreFromSnapshot != null ? restoreFromSnapshot.getSnapshot() : null, persistenceId(), LOG);
772 protected void onRecoveryComplete() {
773 restoreFromSnapshot = null;
775 //notify shard manager
776 getContext().parent().tell(new ActorInitialized(), getSelf());
778 // Being paranoid here - this method should only be called once but just in case...
779 if (txCommitTimeoutCheckSchedule == null) {
780 // Schedule a message to be periodically sent to check if the current in-progress
781 // transaction should be expired and aborted.
782 FiniteDuration period = Duration.create(transactionCommitTimeout / 3, TimeUnit.MILLISECONDS);
783 txCommitTimeoutCheckSchedule = getContext().system().scheduler().schedule(
784 period, period, getSelf(),
785 TX_COMMIT_TIMEOUT_CHECK_MESSAGE, getContext().dispatcher(), ActorRef.noSender());
790 protected void applyState(final ActorRef clientActor, final Identifier identifier, final Object data) {
791 if (data instanceof Payload) {
793 store.applyReplicatedPayload(identifier, (Payload)data);
794 } catch (DataValidationFailedException | IOException e) {
795 LOG.error("{}: Error applying replica {}", persistenceId(), identifier, e);
798 LOG.error("{}: Unknown state for {} received {}", persistenceId(), identifier, data);
803 protected void onStateChanged() {
804 boolean isLeader = isLeader();
805 boolean hasLeader = hasLeader();
806 changeSupport.onLeadershipChange(isLeader, hasLeader);
807 treeChangeSupport.onLeadershipChange(isLeader, hasLeader);
809 // If this actor is no longer the leader close all the transaction chains
811 if (LOG.isDebugEnabled()) {
813 "{}: onStateChanged: Closing all transaction chains because shard {} is no longer the leader",
814 persistenceId(), getId());
817 store.purgeLeaderState();
820 if (hasLeader && !isIsolatedLeader()) {
821 messageRetrySupport.retryMessages();
826 protected void onLeaderChanged(final String oldLeader, final String newLeader) {
827 shardMBean.incrementLeadershipChangeCount();
829 final boolean hasLeader = hasLeader();
831 // No leader implies we are not the leader, lose frontend state if we have any. This also places
832 // an explicit guard so the map will not get modified accidentally.
833 if (!knownFrontends.isEmpty()) {
834 LOG.debug("{}: removing frontend state for {}", persistenceId(), knownFrontends.keySet());
835 knownFrontends = ImmutableMap.of();
841 // Another leader was elected. If we were the previous leader and had pending transactions, convert
842 // them to transaction messages and send to the new leader.
843 ActorSelection leader = getLeader();
844 if (leader != null) {
845 Collection<?> messagesToForward = convertPendingTransactionsToMessages();
847 if (!messagesToForward.isEmpty()) {
848 LOG.debug("{}: Forwarding {} pending transaction messages to leader {}", persistenceId(),
849 messagesToForward.size(), leader);
851 for (Object message : messagesToForward) {
852 leader.tell(message, self());
856 commitCoordinator.abortPendingTransactions("The transacton was aborted due to inflight leadership "
857 + "change and the leader address isn't available.", this);
860 // We have become the leader, we need to reconstruct frontend state
861 knownFrontends = Verify.verifyNotNull(frontendMetadata.toLeaderState(this));
862 LOG.debug("{}: became leader with frontend state for {}", persistenceId(), knownFrontends.keySet());
865 if (!isIsolatedLeader()) {
866 messageRetrySupport.retryMessages();
871 * Clears all pending transactions and converts them to messages to be forwarded to a new leader.
873 * @return the converted messages
875 public Collection<?> convertPendingTransactionsToMessages() {
876 return commitCoordinator.convertPendingTransactionsToMessages(
877 datastoreContext.getShardBatchedModificationCount());
881 protected void pauseLeader(final Runnable operation) {
882 LOG.debug("{}: In pauseLeader, operation: {}", persistenceId(), operation);
883 store.setRunOnPendingTransactionsComplete(operation);
887 protected void unpauseLeader() {
888 LOG.debug("{}: In unpauseLeader", persistenceId());
889 store.setRunOnPendingTransactionsComplete(null);
893 protected OnDemandRaftState.AbstractBuilder<?> newOnDemandRaftStateBuilder() {
894 return OnDemandShardState.newBuilder().treeChangeListenerActors(treeChangeSupport.getListenerActors())
895 .dataChangeListenerActors(changeSupport.getListenerActors())
896 .commitCohortActors(store.getCohortActors());
900 public String persistenceId() {
905 ShardCommitCoordinator getCommitCoordinator() {
906 return commitCoordinator;
909 public DatastoreContext getDatastoreContext() {
910 return datastoreContext;
914 public ShardDataTree getDataStore() {
919 ShardStats getShardMBean() {
923 public static Builder builder() {
924 return new Builder();
927 public abstract static class AbstractBuilder<T extends AbstractBuilder<T, S>, S extends Shard> {
928 private final Class<S> shardClass;
929 private ShardIdentifier id;
930 private Map<String, String> peerAddresses = Collections.emptyMap();
931 private DatastoreContext datastoreContext;
932 private SchemaContextProvider schemaContextProvider;
933 private DatastoreSnapshot.ShardSnapshot restoreFromSnapshot;
934 private TipProducingDataTree dataTree;
935 private volatile boolean sealed;
937 protected AbstractBuilder(final Class<S> shardClass) {
938 this.shardClass = shardClass;
941 protected void checkSealed() {
942 Preconditions.checkState(!sealed, "Builder isalready sealed - further modifications are not allowed");
945 @SuppressWarnings("unchecked")
950 public T id(final ShardIdentifier newId) {
956 public T peerAddresses(final Map<String, String> newPeerAddresses) {
958 this.peerAddresses = newPeerAddresses;
962 public T datastoreContext(final DatastoreContext newDatastoreContext) {
964 this.datastoreContext = newDatastoreContext;
968 public T schemaContextProvider(final SchemaContextProvider schemaContextProvider) {
970 this.schemaContextProvider = Preconditions.checkNotNull(schemaContextProvider);
974 public T restoreFromSnapshot(final DatastoreSnapshot.ShardSnapshot newRestoreFromSnapshot) {
976 this.restoreFromSnapshot = newRestoreFromSnapshot;
980 public T dataTree(final TipProducingDataTree newDataTree) {
982 this.dataTree = newDataTree;
986 public ShardIdentifier getId() {
990 public Map<String, String> getPeerAddresses() {
991 return peerAddresses;
994 public DatastoreContext getDatastoreContext() {
995 return datastoreContext;
998 public SchemaContext getSchemaContext() {
999 return Verify.verifyNotNull(schemaContextProvider.getSchemaContext());
1002 public DatastoreSnapshot.ShardSnapshot getRestoreFromSnapshot() {
1003 return restoreFromSnapshot;
1006 public TipProducingDataTree getDataTree() {
1010 public TreeType getTreeType() {
1011 switch (datastoreContext.getLogicalStoreType()) {
1013 return TreeType.CONFIGURATION;
1015 return TreeType.OPERATIONAL;
1017 throw new IllegalStateException("Unhandled logical store type "
1018 + datastoreContext.getLogicalStoreType());
1022 protected void verify() {
1023 Preconditions.checkNotNull(id, "id should not be null");
1024 Preconditions.checkNotNull(peerAddresses, "peerAddresses should not be null");
1025 Preconditions.checkNotNull(datastoreContext, "dataStoreContext should not be null");
1026 Preconditions.checkNotNull(schemaContextProvider, "schemaContextProvider should not be null");
1029 public Props props() {
1032 return Props.create(shardClass, this);
1036 public static class Builder extends AbstractBuilder<Builder, Shard> {
1043 return Ticker.systemTicker();
1046 void scheduleNextPendingTransaction() {
1047 self().tell(RESUME_NEXT_PENDING_TRANSACTION, ActorRef.noSender());