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.Failure;
16 import akka.serialization.Serialization;
17 import com.google.common.annotations.VisibleForTesting;
18 import com.google.common.base.Optional;
19 import com.google.common.base.Preconditions;
20 import com.google.common.base.Ticker;
21 import com.google.common.collect.ImmutableList;
22 import com.google.common.collect.Range;
23 import java.io.IOException;
24 import java.util.Arrays;
25 import java.util.Collection;
26 import java.util.Collections;
27 import java.util.HashMap;
29 import java.util.concurrent.TimeUnit;
30 import javax.annotation.Nonnull;
31 import javax.annotation.Nullable;
32 import org.opendaylight.controller.cluster.access.ABIVersion;
33 import org.opendaylight.controller.cluster.access.commands.ConnectClientRequest;
34 import org.opendaylight.controller.cluster.access.commands.ConnectClientSuccess;
35 import org.opendaylight.controller.cluster.access.commands.LocalHistoryRequest;
36 import org.opendaylight.controller.cluster.access.commands.NotLeaderException;
37 import org.opendaylight.controller.cluster.access.commands.TransactionRequest;
38 import org.opendaylight.controller.cluster.access.concepts.ClientIdentifier;
39 import org.opendaylight.controller.cluster.access.concepts.FrontendIdentifier;
40 import org.opendaylight.controller.cluster.access.concepts.Request;
41 import org.opendaylight.controller.cluster.access.concepts.RequestEnvelope;
42 import org.opendaylight.controller.cluster.access.concepts.RequestException;
43 import org.opendaylight.controller.cluster.access.concepts.RequestSuccess;
44 import org.opendaylight.controller.cluster.access.concepts.RetiredGenerationException;
45 import org.opendaylight.controller.cluster.access.concepts.RuntimeRequestException;
46 import org.opendaylight.controller.cluster.access.concepts.TransactionIdentifier;
47 import org.opendaylight.controller.cluster.access.concepts.UnsupportedRequestException;
48 import org.opendaylight.controller.cluster.common.actor.CommonConfig;
49 import org.opendaylight.controller.cluster.common.actor.MessageTracker;
50 import org.opendaylight.controller.cluster.common.actor.MessageTracker.Error;
51 import org.opendaylight.controller.cluster.common.actor.MeteringBehavior;
52 import org.opendaylight.controller.cluster.datastore.exceptions.NoShardLeaderException;
53 import org.opendaylight.controller.cluster.datastore.identifiers.ShardIdentifier;
54 import org.opendaylight.controller.cluster.datastore.jmx.mbeans.shard.ShardMBeanFactory;
55 import org.opendaylight.controller.cluster.datastore.jmx.mbeans.shard.ShardStats;
56 import org.opendaylight.controller.cluster.datastore.messages.AbortTransaction;
57 import org.opendaylight.controller.cluster.datastore.messages.ActorInitialized;
58 import org.opendaylight.controller.cluster.datastore.messages.BatchedModifications;
59 import org.opendaylight.controller.cluster.datastore.messages.CanCommitTransaction;
60 import org.opendaylight.controller.cluster.datastore.messages.CloseTransactionChain;
61 import org.opendaylight.controller.cluster.datastore.messages.CommitTransaction;
62 import org.opendaylight.controller.cluster.datastore.messages.CreateTransaction;
63 import org.opendaylight.controller.cluster.datastore.messages.CreateTransactionReply;
64 import org.opendaylight.controller.cluster.datastore.messages.DatastoreSnapshot;
65 import org.opendaylight.controller.cluster.datastore.messages.DatastoreSnapshot.ShardSnapshot;
66 import org.opendaylight.controller.cluster.datastore.messages.ForwardedReadyTransaction;
67 import org.opendaylight.controller.cluster.datastore.messages.GetShardDataTree;
68 import org.opendaylight.controller.cluster.datastore.messages.PeerAddressResolved;
69 import org.opendaylight.controller.cluster.datastore.messages.ReadyLocalTransaction;
70 import org.opendaylight.controller.cluster.datastore.messages.RegisterChangeListener;
71 import org.opendaylight.controller.cluster.datastore.messages.RegisterDataTreeChangeListener;
72 import org.opendaylight.controller.cluster.datastore.messages.ShardLeaderStateChanged;
73 import org.opendaylight.controller.cluster.datastore.messages.UpdateSchemaContext;
74 import org.opendaylight.controller.cluster.datastore.utils.Dispatchers;
75 import org.opendaylight.controller.cluster.notifications.LeaderStateChanged;
76 import org.opendaylight.controller.cluster.notifications.RegisterRoleChangeListener;
77 import org.opendaylight.controller.cluster.notifications.RoleChangeNotifier;
78 import org.opendaylight.controller.cluster.raft.RaftActor;
79 import org.opendaylight.controller.cluster.raft.RaftActorRecoveryCohort;
80 import org.opendaylight.controller.cluster.raft.RaftActorSnapshotCohort;
81 import org.opendaylight.controller.cluster.raft.RaftState;
82 import org.opendaylight.controller.cluster.raft.base.messages.FollowerInitialSyncUpStatus;
83 import org.opendaylight.controller.cluster.raft.messages.AppendEntriesReply;
84 import org.opendaylight.controller.cluster.raft.messages.ServerRemoved;
85 import org.opendaylight.controller.cluster.raft.protobuff.client.messages.Payload;
86 import org.opendaylight.yangtools.concepts.Identifier;
87 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataValidationFailedException;
88 import org.opendaylight.yangtools.yang.data.api.schema.tree.TipProducingDataTree;
89 import org.opendaylight.yangtools.yang.data.api.schema.tree.TreeType;
90 import org.opendaylight.yangtools.yang.model.api.SchemaContext;
91 import scala.concurrent.duration.Duration;
92 import scala.concurrent.duration.FiniteDuration;
95 * A Shard represents a portion of the logical data tree.
98 * Our Shard uses InMemoryDataTree as it's internal representation and delegates all requests it
100 public class Shard extends RaftActor {
103 static final Object TX_COMMIT_TIMEOUT_CHECK_MESSAGE = new Object() {
105 public String toString() {
106 return "txCommitTimeoutCheck";
111 static final Object GET_SHARD_MBEAN_MESSAGE = new Object() {
113 public String toString() {
114 return "getShardMBeanMessage";
118 // FIXME: shard names should be encapsulated in their own class and this should be exposed as a constant.
119 public static final String DEFAULT_NAME = "default";
121 private static final Collection<ABIVersion> SUPPORTED_ABIVERSIONS;
124 final ABIVersion[] values = ABIVersion.values();
125 final ABIVersion[] real = Arrays.copyOfRange(values, 1, values.length - 1);
126 SUPPORTED_ABIVERSIONS = ImmutableList.copyOf(real).reverse();
129 // FIXME: make this a dynamic property based on mailbox size and maximum number of clients
130 private static final int CLIENT_MAX_MESSAGES = 1000;
132 // The state of this Shard
133 private final ShardDataTree store;
135 /// The name of this shard
136 private final String name;
138 private final ShardStats shardMBean;
140 private DatastoreContext datastoreContext;
142 private final ShardCommitCoordinator commitCoordinator;
144 private long transactionCommitTimeout;
146 private Cancellable txCommitTimeoutCheckSchedule;
148 private final Optional<ActorRef> roleChangeNotifier;
150 private final MessageTracker appendEntriesReplyTracker;
152 private final ShardTransactionActorFactory transactionActorFactory;
154 private final ShardSnapshotCohort snapshotCohort;
156 private final DataTreeChangeListenerSupport treeChangeSupport = new DataTreeChangeListenerSupport(this);
157 private final DataChangeListenerSupport changeSupport = new DataChangeListenerSupport(this);
160 private ShardSnapshot restoreFromSnapshot;
162 private final ShardTransactionMessageRetrySupport messageRetrySupport;
164 private final FrontendMetadata frontendMetadata = new FrontendMetadata();
165 private final Map<FrontendIdentifier, LeaderFrontendState> knownFrontends = new HashMap<>();
167 protected Shard(final AbstractBuilder<?, ?> builder) {
168 super(builder.getId().toString(), builder.getPeerAddresses(),
169 Optional.of(builder.getDatastoreContext().getShardRaftConfig()), DataStoreVersions.CURRENT_VERSION);
171 this.name = builder.getId().toString();
172 this.datastoreContext = builder.getDatastoreContext();
173 this.restoreFromSnapshot = builder.getRestoreFromSnapshot();
175 setPersistence(datastoreContext.isPersistent());
177 LOG.info("Shard created : {}, persistent : {}", name, datastoreContext.isPersistent());
179 ShardDataTreeChangeListenerPublisherActorProxy treeChangeListenerPublisher =
180 new ShardDataTreeChangeListenerPublisherActorProxy(getContext(), name + "-DTCL-publisher");
181 ShardDataChangeListenerPublisherActorProxy dataChangeListenerPublisher =
182 new ShardDataChangeListenerPublisherActorProxy(getContext(), name + "-DCL-publisher");
183 if (builder.getDataTree() != null) {
184 store = new ShardDataTree(this, builder.getSchemaContext(), builder.getDataTree(),
185 treeChangeListenerPublisher, dataChangeListenerPublisher, name);
187 store = new ShardDataTree(this, builder.getSchemaContext(), builder.getTreeType(),
188 builder.getDatastoreContext().getStoreRoot(), treeChangeListenerPublisher,
189 dataChangeListenerPublisher, name);
192 shardMBean = ShardMBeanFactory.getShardStatsMBean(name, datastoreContext.getDataStoreMXBeanType(), this);
194 if (isMetricsCaptureEnabled()) {
195 getContext().become(new MeteringBehavior(this));
198 commitCoordinator = new ShardCommitCoordinator(store, LOG, this.name);
200 setTransactionCommitTimeout();
202 // create a notifier actor for each cluster member
203 roleChangeNotifier = createRoleChangeNotifier(name);
205 appendEntriesReplyTracker = new MessageTracker(AppendEntriesReply.class,
206 getRaftActorContext().getConfigParams().getIsolatedCheckIntervalInMillis());
208 transactionActorFactory = new ShardTransactionActorFactory(store, datastoreContext,
209 new Dispatchers(context().system().dispatchers()).getDispatcherPath(
210 Dispatchers.DispatcherType.Transaction), self(), getContext(), shardMBean);
212 snapshotCohort = ShardSnapshotCohort.create(getContext(), builder.getId().getMemberName(), store, LOG,
215 messageRetrySupport = new ShardTransactionMessageRetrySupport(this);
218 private void setTransactionCommitTimeout() {
219 transactionCommitTimeout = TimeUnit.MILLISECONDS.convert(
220 datastoreContext.getShardTransactionCommitTimeoutInSeconds(), TimeUnit.SECONDS) / 2;
223 private Optional<ActorRef> createRoleChangeNotifier(final String shardId) {
224 ActorRef shardRoleChangeNotifier = this.getContext().actorOf(
225 RoleChangeNotifier.getProps(shardId), shardId + "-notifier");
226 return Optional.of(shardRoleChangeNotifier);
230 public void postStop() {
231 LOG.info("Stopping Shard {}", persistenceId());
235 messageRetrySupport.close();
237 if (txCommitTimeoutCheckSchedule != null) {
238 txCommitTimeoutCheckSchedule.cancel();
241 commitCoordinator.abortPendingTransactions("Transaction aborted due to shutdown.", this);
243 shardMBean.unregisterMBean();
247 protected void handleRecover(final Object message) {
248 LOG.debug("{}: onReceiveRecover: Received message {} from {}", persistenceId(), message.getClass(),
251 super.handleRecover(message);
252 if (LOG.isTraceEnabled()) {
253 appendEntriesReplyTracker.begin();
257 @SuppressWarnings("checkstyle:IllegalCatch")
259 protected void handleNonRaftCommand(final Object message) {
260 try (final MessageTracker.Context context = appendEntriesReplyTracker.received(message)) {
261 final Optional<Error> maybeError = context.error();
262 if (maybeError.isPresent()) {
263 LOG.trace("{} : AppendEntriesReply failed to arrive at the expected interval {}", persistenceId(),
267 if (message instanceof RequestEnvelope) {
268 final long now = ticker().read();
269 final RequestEnvelope envelope = (RequestEnvelope)message;
272 final RequestSuccess<?, ?> success = handleRequest(envelope, now);
273 if (success != null) {
274 envelope.sendSuccess(success, ticker().read() - now);
276 } catch (RequestException e) {
277 LOG.debug("{}: request {} failed", persistenceId(), envelope, e);
278 envelope.sendFailure(e, ticker().read() - now);
279 } catch (Exception e) {
280 LOG.debug("{}: request {} caused failure", persistenceId(), envelope, e);
281 envelope.sendFailure(new RuntimeRequestException("Request failed to process", e),
282 ticker().read() - now);
284 } else if (message instanceof ConnectClientRequest) {
285 handleConnectClient((ConnectClientRequest)message);
286 } else if (CreateTransaction.isSerializedType(message)) {
287 handleCreateTransaction(message);
288 } else if (message instanceof BatchedModifications) {
289 handleBatchedModifications((BatchedModifications)message);
290 } else if (message instanceof ForwardedReadyTransaction) {
291 handleForwardedReadyTransaction((ForwardedReadyTransaction) message);
292 } else if (message instanceof ReadyLocalTransaction) {
293 handleReadyLocalTransaction((ReadyLocalTransaction)message);
294 } else if (CanCommitTransaction.isSerializedType(message)) {
295 handleCanCommitTransaction(CanCommitTransaction.fromSerializable(message));
296 } else if (CommitTransaction.isSerializedType(message)) {
297 handleCommitTransaction(CommitTransaction.fromSerializable(message));
298 } else if (AbortTransaction.isSerializedType(message)) {
299 handleAbortTransaction(AbortTransaction.fromSerializable(message));
300 } else if (CloseTransactionChain.isSerializedType(message)) {
301 closeTransactionChain(CloseTransactionChain.fromSerializable(message));
302 } else if (message instanceof RegisterChangeListener) {
303 changeSupport.onMessage((RegisterChangeListener) message, isLeader(), hasLeader());
304 } else if (message instanceof RegisterDataTreeChangeListener) {
305 treeChangeSupport.onMessage((RegisterDataTreeChangeListener) message, isLeader(), hasLeader());
306 } else if (message instanceof UpdateSchemaContext) {
307 updateSchemaContext((UpdateSchemaContext) message);
308 } else if (message instanceof PeerAddressResolved) {
309 PeerAddressResolved resolved = (PeerAddressResolved) message;
310 setPeerAddress(resolved.getPeerId(), resolved.getPeerAddress());
311 } else if (TX_COMMIT_TIMEOUT_CHECK_MESSAGE.equals(message)) {
312 store.checkForExpiredTransactions(transactionCommitTimeout);
313 commitCoordinator.checkForExpiredTransactions(transactionCommitTimeout, this);
314 } else if (message instanceof DatastoreContext) {
315 onDatastoreContext((DatastoreContext)message);
316 } else if (message instanceof RegisterRoleChangeListener) {
317 roleChangeNotifier.get().forward(message, context());
318 } else if (message instanceof FollowerInitialSyncUpStatus) {
319 shardMBean.setFollowerInitialSyncStatus(((FollowerInitialSyncUpStatus) message).isInitialSyncDone());
320 context().parent().tell(message, self());
321 } else if (GET_SHARD_MBEAN_MESSAGE.equals(message)) {
322 sender().tell(getShardMBean(), self());
323 } else if (message instanceof GetShardDataTree) {
324 sender().tell(store.getDataTree(), self());
325 } else if (message instanceof ServerRemoved) {
326 context().parent().forward(message, context());
327 } else if (ShardTransactionMessageRetrySupport.TIMER_MESSAGE_CLASS.isInstance(message)) {
328 messageRetrySupport.onTimerMessage(message);
329 } else if (message instanceof DataTreeCohortActorRegistry.CohortRegistryCommand) {
330 store.processCohortRegistryCommand(getSender(),
331 (DataTreeCohortActorRegistry.CohortRegistryCommand) message);
333 super.handleNonRaftCommand(message);
338 // Acquire our frontend tracking handle and verify generation matches
339 private LeaderFrontendState getFrontend(final ClientIdentifier clientId) throws RequestException {
340 final LeaderFrontendState existing = knownFrontends.get(clientId.getFrontendId());
341 if (existing != null) {
342 final int cmp = Long.compareUnsigned(existing.getIdentifier().getGeneration(), clientId.getGeneration());
347 LOG.debug("{}: rejecting request from outdated client {}", persistenceId(), clientId);
348 throw new RetiredGenerationException(existing.getIdentifier().getGeneration());
351 LOG.info("{}: retiring state {}, outdated by request from client {}", persistenceId(), existing, clientId);
353 knownFrontends.remove(clientId.getFrontendId());
355 LOG.debug("{}: client {} is not yet known", persistenceId(), clientId);
358 final LeaderFrontendState ret = new LeaderFrontendState(persistenceId(), clientId, store);
359 knownFrontends.put(clientId.getFrontendId(), ret);
360 LOG.debug("{}: created state {} for client {}", persistenceId(), ret, clientId);
364 private static @Nonnull ABIVersion selectVersion(final ConnectClientRequest message) {
365 final Range<ABIVersion> clientRange = Range.closed(message.getMinVersion(), message.getMaxVersion());
366 for (ABIVersion v : SUPPORTED_ABIVERSIONS) {
367 if (clientRange.contains(v)) {
372 throw new IllegalArgumentException(String.format(
373 "No common version between backend versions %s and client versions %s", SUPPORTED_ABIVERSIONS,
377 @SuppressWarnings("checkstyle:IllegalCatch")
378 private void handleConnectClient(final ConnectClientRequest message) {
380 if (!isLeader() || !isLeaderActive()) {
381 LOG.debug("{}: not currently leader, rejecting request {}", persistenceId(), message);
382 throw new NotLeaderException(getSelf());
385 final ABIVersion selectedVersion = selectVersion(message);
386 final LeaderFrontendState frontend = getFrontend(message.getTarget());
387 frontend.reconnect();
388 message.getReplyTo().tell(new ConnectClientSuccess(message.getTarget(), message.getSequence(), getSelf(),
389 ImmutableList.of(), store.getDataTree(), CLIENT_MAX_MESSAGES).toVersion(selectedVersion),
390 ActorRef.noSender());
391 } catch (RequestException | RuntimeException e) {
392 message.getReplyTo().tell(new Failure(e), ActorRef.noSender());
396 private @Nullable RequestSuccess<?, ?> handleRequest(final RequestEnvelope envelope, final long now)
397 throws RequestException {
398 // We are not the leader, hence we want to fail-fast.
399 if (!isLeader() || !isLeaderActive()) {
400 LOG.debug("{}: not currently leader, rejecting request {}", persistenceId(), envelope);
401 throw new NotLeaderException(getSelf());
404 final Request<?, ?> request = envelope.getMessage();
405 if (request instanceof TransactionRequest) {
406 final TransactionRequest<?> txReq = (TransactionRequest<?>)request;
407 final ClientIdentifier clientId = txReq.getTarget().getHistoryId().getClientId();
408 return getFrontend(clientId).handleTransactionRequest(txReq, envelope, now);
409 } else if (request instanceof LocalHistoryRequest) {
410 final LocalHistoryRequest<?> lhReq = (LocalHistoryRequest<?>)request;
411 final ClientIdentifier clientId = lhReq.getTarget().getClientId();
412 return getFrontend(clientId).handleLocalHistoryRequest(lhReq, envelope, now);
414 LOG.debug("{}: rejecting unsupported request {}", persistenceId(), request);
415 throw new UnsupportedRequestException(request);
419 private boolean hasLeader() {
420 return getLeaderId() != null;
423 public int getPendingTxCommitQueueSize() {
424 return store.getQueueSize();
427 public int getCohortCacheSize() {
428 return commitCoordinator.getCohortCacheSize();
432 protected Optional<ActorRef> getRoleChangeNotifier() {
433 return roleChangeNotifier;
437 protected LeaderStateChanged newLeaderStateChanged(final String memberId, final String leaderId,
438 final short leaderPayloadVersion) {
439 return isLeader() ? new ShardLeaderStateChanged(memberId, leaderId, store.getDataTree(), leaderPayloadVersion)
440 : new ShardLeaderStateChanged(memberId, leaderId, leaderPayloadVersion);
443 protected void onDatastoreContext(final DatastoreContext context) {
444 datastoreContext = context;
446 setTransactionCommitTimeout();
448 setPersistence(datastoreContext.isPersistent());
450 updateConfigParams(datastoreContext.getShardRaftConfig());
453 // applyState() will be invoked once consensus is reached on the payload
454 void persistPayload(final TransactionIdentifier transactionId, final Payload payload, boolean batchHint) {
455 boolean canSkipPayload = !hasFollowers() && !persistence().isRecoveryApplicable();
456 if (canSkipPayload) {
457 applyState(self(), transactionId, payload);
459 // We are faking the sender
460 persistData(self(), transactionId, payload, batchHint);
464 private void handleCommitTransaction(final CommitTransaction commit) {
466 commitCoordinator.handleCommit(commit.getTransactionId(), getSender(), this);
468 ActorSelection leader = getLeader();
469 if (leader == null) {
470 messageRetrySupport.addMessageToRetry(commit, getSender(),
471 "Could not commit transaction " + commit.getTransactionId());
473 LOG.debug("{}: Forwarding CommitTransaction to leader {}", persistenceId(), leader);
474 leader.forward(commit, getContext());
479 private void handleCanCommitTransaction(final CanCommitTransaction canCommit) {
480 LOG.debug("{}: Can committing transaction {}", persistenceId(), canCommit.getTransactionId());
483 commitCoordinator.handleCanCommit(canCommit.getTransactionId(), getSender(), this);
485 ActorSelection leader = getLeader();
486 if (leader == null) {
487 messageRetrySupport.addMessageToRetry(canCommit, getSender(),
488 "Could not canCommit transaction " + canCommit.getTransactionId());
490 LOG.debug("{}: Forwarding CanCommitTransaction to leader {}", persistenceId(), leader);
491 leader.forward(canCommit, getContext());
496 @SuppressWarnings("checkstyle:IllegalCatch")
497 protected void handleBatchedModificationsLocal(final BatchedModifications batched, final ActorRef sender) {
499 commitCoordinator.handleBatchedModifications(batched, sender, this);
500 } catch (Exception e) {
501 LOG.error("{}: Error handling BatchedModifications for Tx {}", persistenceId(),
502 batched.getTransactionId(), e);
503 sender.tell(new Failure(e), getSelf());
507 private void handleBatchedModifications(final BatchedModifications batched) {
508 // This message is sent to prepare the modifications transaction directly on the Shard as an
509 // optimization to avoid the extra overhead of a separate ShardTransaction actor. On the last
510 // BatchedModifications message, the caller sets the ready flag in the message indicating
511 // modifications are complete. The reply contains the cohort actor path (this actor) for the caller
512 // to initiate the 3-phase commit. This also avoids the overhead of sending an additional
513 // ReadyTransaction message.
515 // If we're not the leader then forward to the leader. This is a safety measure - we shouldn't
516 // normally get here if we're not the leader as the front-end (TransactionProxy) should determine
517 // the primary/leader shard. However with timing and caching on the front-end, there's a small
518 // window where it could have a stale leader during leadership transitions.
520 boolean isLeaderActive = isLeaderActive();
521 if (isLeader() && isLeaderActive) {
522 handleBatchedModificationsLocal(batched, getSender());
524 ActorSelection leader = getLeader();
525 if (!isLeaderActive || leader == null) {
526 messageRetrySupport.addMessageToRetry(batched, getSender(),
527 "Could not commit transaction " + batched.getTransactionId());
529 // If this is not the first batch and leadership changed in between batched messages,
530 // we need to reconstruct previous BatchedModifications from the transaction
531 // DataTreeModification, honoring the max batched modification count, and forward all the
532 // previous BatchedModifications to the new leader.
533 Collection<BatchedModifications> newModifications = commitCoordinator
534 .createForwardedBatchedModifications(batched,
535 datastoreContext.getShardBatchedModificationCount());
537 LOG.debug("{}: Forwarding {} BatchedModifications to leader {}", persistenceId(),
538 newModifications.size(), leader);
540 for (BatchedModifications bm : newModifications) {
541 leader.forward(bm, getContext());
547 private boolean failIfIsolatedLeader(final ActorRef sender) {
548 if (isIsolatedLeader()) {
549 sender.tell(new Failure(new NoShardLeaderException(String.format(
550 "Shard %s was the leader but has lost contact with all of its followers. Either all"
551 + " other follower nodes are down or this node is isolated by a network partition.",
552 persistenceId()))), getSelf());
559 protected boolean isIsolatedLeader() {
560 return getRaftState() == RaftState.IsolatedLeader;
563 @SuppressWarnings("checkstyle:IllegalCatch")
564 private void handleReadyLocalTransaction(final ReadyLocalTransaction message) {
565 LOG.debug("{}: handleReadyLocalTransaction for {}", persistenceId(), message.getTransactionId());
567 boolean isLeaderActive = isLeaderActive();
568 if (isLeader() && isLeaderActive) {
570 commitCoordinator.handleReadyLocalTransaction(message, getSender(), this);
571 } catch (Exception e) {
572 LOG.error("{}: Error handling ReadyLocalTransaction for Tx {}", persistenceId(),
573 message.getTransactionId(), e);
574 getSender().tell(new Failure(e), getSelf());
577 ActorSelection leader = getLeader();
578 if (!isLeaderActive || leader == null) {
579 messageRetrySupport.addMessageToRetry(message, getSender(),
580 "Could not commit transaction " + message.getTransactionId());
582 LOG.debug("{}: Forwarding ReadyLocalTransaction to leader {}", persistenceId(), leader);
583 message.setRemoteVersion(getCurrentBehavior().getLeaderPayloadVersion());
584 leader.forward(message, getContext());
589 private void handleForwardedReadyTransaction(final ForwardedReadyTransaction forwardedReady) {
590 LOG.debug("{}: handleForwardedReadyTransaction for {}", persistenceId(), forwardedReady.getTransactionId());
592 boolean isLeaderActive = isLeaderActive();
593 if (isLeader() && isLeaderActive) {
594 commitCoordinator.handleForwardedReadyTransaction(forwardedReady, getSender(), this);
596 ActorSelection leader = getLeader();
597 if (!isLeaderActive || leader == null) {
598 messageRetrySupport.addMessageToRetry(forwardedReady, getSender(),
599 "Could not commit transaction " + forwardedReady.getTransactionId());
601 LOG.debug("{}: Forwarding ForwardedReadyTransaction to leader {}", persistenceId(), leader);
603 ReadyLocalTransaction readyLocal = new ReadyLocalTransaction(forwardedReady.getTransactionId(),
604 forwardedReady.getTransaction().getSnapshot(), forwardedReady.isDoImmediateCommit());
605 readyLocal.setRemoteVersion(getCurrentBehavior().getLeaderPayloadVersion());
606 leader.forward(readyLocal, getContext());
611 private void handleAbortTransaction(final AbortTransaction abort) {
612 doAbortTransaction(abort.getTransactionId(), getSender());
615 void doAbortTransaction(final TransactionIdentifier transactionID, final ActorRef sender) {
616 commitCoordinator.handleAbort(transactionID, sender, this);
619 private void handleCreateTransaction(final Object message) {
621 createTransaction(CreateTransaction.fromSerializable(message));
622 } else if (getLeader() != null) {
623 getLeader().forward(message, getContext());
625 getSender().tell(new Failure(new NoShardLeaderException(
626 "Could not create a shard transaction", persistenceId())), getSelf());
630 private void closeTransactionChain(final CloseTransactionChain closeTransactionChain) {
631 store.closeTransactionChain(closeTransactionChain.getIdentifier());
634 @SuppressWarnings("checkstyle:IllegalCatch")
635 private void createTransaction(final CreateTransaction createTransaction) {
637 if (TransactionType.fromInt(createTransaction.getTransactionType()) != TransactionType.READ_ONLY
638 && failIfIsolatedLeader(getSender())) {
642 ActorRef transactionActor = createTransaction(createTransaction.getTransactionType(),
643 createTransaction.getTransactionId());
645 getSender().tell(new CreateTransactionReply(Serialization.serializedActorPath(transactionActor),
646 createTransaction.getTransactionId(), createTransaction.getVersion()).toSerializable(), getSelf());
647 } catch (Exception e) {
648 getSender().tell(new Failure(e), getSelf());
652 private ActorRef createTransaction(final int transactionType, final TransactionIdentifier transactionId) {
653 LOG.debug("{}: Creating transaction : {} ", persistenceId(), transactionId);
654 return transactionActorFactory.newShardTransaction(TransactionType.fromInt(transactionType),
658 private void updateSchemaContext(final UpdateSchemaContext message) {
659 updateSchemaContext(message.getSchemaContext());
663 void updateSchemaContext(final SchemaContext schemaContext) {
664 store.updateSchemaContext(schemaContext);
667 private boolean isMetricsCaptureEnabled() {
668 CommonConfig config = new CommonConfig(getContext().system().settings().config());
669 return config.isMetricCaptureEnabled();
674 public RaftActorSnapshotCohort getRaftActorSnapshotCohort() {
675 return snapshotCohort;
680 protected RaftActorRecoveryCohort getRaftActorRecoveryCohort() {
681 return new ShardRecoveryCoordinator(store,
682 restoreFromSnapshot != null ? restoreFromSnapshot.getSnapshot() : null, persistenceId(), LOG);
686 protected void onRecoveryComplete() {
687 restoreFromSnapshot = null;
689 //notify shard manager
690 getContext().parent().tell(new ActorInitialized(), getSelf());
692 // Being paranoid here - this method should only be called once but just in case...
693 if (txCommitTimeoutCheckSchedule == null) {
694 // Schedule a message to be periodically sent to check if the current in-progress
695 // transaction should be expired and aborted.
696 FiniteDuration period = Duration.create(transactionCommitTimeout / 3, TimeUnit.MILLISECONDS);
697 txCommitTimeoutCheckSchedule = getContext().system().scheduler().schedule(
698 period, period, getSelf(),
699 TX_COMMIT_TIMEOUT_CHECK_MESSAGE, getContext().dispatcher(), ActorRef.noSender());
704 protected void applyState(final ActorRef clientActor, final Identifier identifier, final Object data) {
705 if (data instanceof Payload) {
707 store.applyReplicatedPayload(identifier, (Payload)data);
708 } catch (DataValidationFailedException | IOException e) {
709 LOG.error("{}: Error applying replica {}", persistenceId(), identifier, e);
712 LOG.error("{}: Unknown state for {} received {}", persistenceId(), identifier, data);
717 protected void onStateChanged() {
718 boolean isLeader = isLeader();
719 boolean hasLeader = hasLeader();
720 changeSupport.onLeadershipChange(isLeader, hasLeader);
721 treeChangeSupport.onLeadershipChange(isLeader, hasLeader);
723 // If this actor is no longer the leader close all the transaction chains
725 if (LOG.isDebugEnabled()) {
727 "{}: onStateChanged: Closing all transaction chains because shard {} is no longer the leader",
728 persistenceId(), getId());
731 store.closeAllTransactionChains();
734 if (hasLeader && !isIsolatedLeader()) {
735 messageRetrySupport.retryMessages();
740 protected void onLeaderChanged(final String oldLeader, final String newLeader) {
741 shardMBean.incrementLeadershipChangeCount();
743 boolean hasLeader = hasLeader();
744 if (hasLeader && !isLeader()) {
745 // Another leader was elected. If we were the previous leader and had pending transactions, convert
746 // them to transaction messages and send to the new leader.
747 ActorSelection leader = getLeader();
748 if (leader != null) {
749 Collection<?> messagesToForward = convertPendingTransactionsToMessages();
751 if (!messagesToForward.isEmpty()) {
752 LOG.debug("{}: Forwarding {} pending transaction messages to leader {}", persistenceId(),
753 messagesToForward.size(), leader);
755 for (Object message : messagesToForward) {
756 leader.tell(message, self());
760 commitCoordinator.abortPendingTransactions("The transacton was aborted due to inflight leadership "
761 + "change and the leader address isn't available.", this);
765 if (hasLeader && !isIsolatedLeader()) {
766 messageRetrySupport.retryMessages();
771 * Clears all pending transactions and converts them to messages to be forwarded to a new leader.
773 * @return the converted messages
775 public Collection<?> convertPendingTransactionsToMessages() {
776 return commitCoordinator.convertPendingTransactionsToMessages(
777 datastoreContext.getShardBatchedModificationCount());
781 protected void pauseLeader(final Runnable operation) {
782 LOG.debug("{}: In pauseLeader, operation: {}", persistenceId(), operation);
783 store.setRunOnPendingTransactionsComplete(operation);
787 public String persistenceId() {
792 ShardCommitCoordinator getCommitCoordinator() {
793 return commitCoordinator;
796 public DatastoreContext getDatastoreContext() {
797 return datastoreContext;
801 public ShardDataTree getDataStore() {
806 ShardStats getShardMBean() {
810 public static Builder builder() {
811 return new Builder();
814 public abstract static class AbstractBuilder<T extends AbstractBuilder<T, S>, S extends Shard> {
815 private final Class<S> shardClass;
816 private ShardIdentifier id;
817 private Map<String, String> peerAddresses = Collections.emptyMap();
818 private DatastoreContext datastoreContext;
819 private SchemaContext schemaContext;
820 private DatastoreSnapshot.ShardSnapshot restoreFromSnapshot;
821 private TipProducingDataTree dataTree;
822 private volatile boolean sealed;
824 protected AbstractBuilder(final Class<S> shardClass) {
825 this.shardClass = shardClass;
828 protected void checkSealed() {
829 Preconditions.checkState(!sealed, "Builder isalready sealed - further modifications are not allowed");
832 @SuppressWarnings("unchecked")
837 public T id(final ShardIdentifier newId) {
843 public T peerAddresses(final Map<String, String> newPeerAddresses) {
845 this.peerAddresses = newPeerAddresses;
849 public T datastoreContext(final DatastoreContext newDatastoreContext) {
851 this.datastoreContext = newDatastoreContext;
855 public T schemaContext(final SchemaContext newSchemaContext) {
857 this.schemaContext = newSchemaContext;
861 public T restoreFromSnapshot(final DatastoreSnapshot.ShardSnapshot newRestoreFromSnapshot) {
863 this.restoreFromSnapshot = newRestoreFromSnapshot;
867 public T dataTree(final TipProducingDataTree newDataTree) {
869 this.dataTree = newDataTree;
873 public ShardIdentifier getId() {
877 public Map<String, String> getPeerAddresses() {
878 return peerAddresses;
881 public DatastoreContext getDatastoreContext() {
882 return datastoreContext;
885 public SchemaContext getSchemaContext() {
886 return schemaContext;
889 public DatastoreSnapshot.ShardSnapshot getRestoreFromSnapshot() {
890 return restoreFromSnapshot;
893 public TipProducingDataTree getDataTree() {
897 public TreeType getTreeType() {
898 switch (datastoreContext.getLogicalStoreType()) {
900 return TreeType.CONFIGURATION;
902 return TreeType.OPERATIONAL;
904 throw new IllegalStateException("Unhandled logical store type "
905 + datastoreContext.getLogicalStoreType());
909 protected void verify() {
910 Preconditions.checkNotNull(id, "id should not be null");
911 Preconditions.checkNotNull(peerAddresses, "peerAddresses should not be null");
912 Preconditions.checkNotNull(datastoreContext, "dataStoreContext should not be null");
913 Preconditions.checkNotNull(schemaContext, "schemaContext should not be null");
916 public Props props() {
919 return Props.create(shardClass, this);
923 public static class Builder extends AbstractBuilder<Builder, Shard> {
930 return Ticker.systemTicker();