2 * Copyright (c) 2015 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
8 package org.opendaylight.controller.cluster.datastore;
10 import akka.actor.ActorRef;
11 import akka.util.Timeout;
12 import com.google.common.annotations.VisibleForTesting;
13 import com.google.common.base.MoreObjects;
14 import com.google.common.base.Optional;
15 import com.google.common.base.Preconditions;
16 import com.google.common.base.Stopwatch;
17 import com.google.common.base.Ticker;
18 import com.google.common.base.Verify;
19 import com.google.common.collect.ImmutableList;
20 import com.google.common.collect.ImmutableMap;
21 import com.google.common.collect.ImmutableMap.Builder;
22 import com.google.common.collect.Iterables;
23 import com.google.common.primitives.UnsignedLong;
24 import edu.umd.cs.findbugs.annotations.SuppressFBWarnings;
26 import java.io.IOException;
27 import java.util.AbstractMap.SimpleEntry;
28 import java.util.ArrayDeque;
29 import java.util.ArrayList;
30 import java.util.Collection;
31 import java.util.HashMap;
32 import java.util.Iterator;
34 import java.util.Map.Entry;
35 import java.util.Queue;
36 import java.util.concurrent.ExecutionException;
37 import java.util.concurrent.TimeUnit;
38 import java.util.concurrent.TimeoutException;
39 import java.util.function.Consumer;
40 import java.util.function.UnaryOperator;
41 import javax.annotation.Nonnull;
42 import javax.annotation.Nullable;
43 import javax.annotation.concurrent.NotThreadSafe;
44 import org.opendaylight.controller.cluster.access.concepts.LocalHistoryIdentifier;
45 import org.opendaylight.controller.cluster.access.concepts.TransactionIdentifier;
46 import org.opendaylight.controller.cluster.datastore.DataTreeCohortActorRegistry.CohortRegistryCommand;
47 import org.opendaylight.controller.cluster.datastore.ShardDataTreeCohort.State;
48 import org.opendaylight.controller.cluster.datastore.persisted.AbortTransactionPayload;
49 import org.opendaylight.controller.cluster.datastore.persisted.AbstractIdentifiablePayload;
50 import org.opendaylight.controller.cluster.datastore.persisted.CloseLocalHistoryPayload;
51 import org.opendaylight.controller.cluster.datastore.persisted.CommitTransactionPayload;
52 import org.opendaylight.controller.cluster.datastore.persisted.CreateLocalHistoryPayload;
53 import org.opendaylight.controller.cluster.datastore.persisted.MetadataShardDataTreeSnapshot;
54 import org.opendaylight.controller.cluster.datastore.persisted.PurgeLocalHistoryPayload;
55 import org.opendaylight.controller.cluster.datastore.persisted.PurgeTransactionPayload;
56 import org.opendaylight.controller.cluster.datastore.persisted.ShardDataTreeSnapshot;
57 import org.opendaylight.controller.cluster.datastore.persisted.ShardDataTreeSnapshotMetadata;
58 import org.opendaylight.controller.cluster.datastore.utils.DataTreeModificationOutput;
59 import org.opendaylight.controller.cluster.datastore.utils.PruningDataTreeModification;
60 import org.opendaylight.controller.cluster.raft.protobuff.client.messages.Payload;
61 import org.opendaylight.controller.md.sal.common.api.data.AsyncDataBroker.DataChangeScope;
62 import org.opendaylight.controller.md.sal.common.api.data.AsyncDataChangeListener;
63 import org.opendaylight.controller.md.sal.common.api.data.OptimisticLockFailedException;
64 import org.opendaylight.controller.md.sal.common.api.data.TransactionCommitFailedException;
65 import org.opendaylight.controller.md.sal.dom.api.DOMDataTreeChangeListener;
66 import org.opendaylight.controller.md.sal.dom.store.impl.DataChangeListenerRegistration;
67 import org.opendaylight.yangtools.concepts.Identifier;
68 import org.opendaylight.yangtools.concepts.ListenerRegistration;
69 import org.opendaylight.yangtools.yang.data.api.YangInstanceIdentifier;
70 import org.opendaylight.yangtools.yang.data.api.schema.NormalizedNode;
71 import org.opendaylight.yangtools.yang.data.api.schema.tree.ConflictingModificationAppliedException;
72 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeCandidate;
73 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeCandidateTip;
74 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeCandidates;
75 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeModification;
76 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeSnapshot;
77 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeTip;
78 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataValidationFailedException;
79 import org.opendaylight.yangtools.yang.data.api.schema.tree.TipProducingDataTree;
80 import org.opendaylight.yangtools.yang.data.api.schema.tree.TipProducingDataTreeTip;
81 import org.opendaylight.yangtools.yang.data.api.schema.tree.TreeType;
82 import org.opendaylight.yangtools.yang.data.impl.schema.tree.InMemoryDataTreeFactory;
83 import org.opendaylight.yangtools.yang.model.api.SchemaContext;
84 import org.slf4j.Logger;
85 import org.slf4j.LoggerFactory;
86 import scala.concurrent.duration.Duration;
89 * Internal shard state, similar to a DOMStore, but optimized for use in the actor system,
90 * e.g. it does not expose public interfaces and assumes it is only ever called from a
94 * This class is not part of the API contract and is subject to change at any time.
97 public class ShardDataTree extends ShardDataTreeTransactionParent {
98 private static final class CommitEntry {
99 final SimpleShardDataTreeCohort cohort;
102 CommitEntry(final SimpleShardDataTreeCohort cohort, final long now) {
103 this.cohort = Preconditions.checkNotNull(cohort);
108 private static final Timeout COMMIT_STEP_TIMEOUT = new Timeout(Duration.create(5, TimeUnit.SECONDS));
109 private static final Logger LOG = LoggerFactory.getLogger(ShardDataTree.class);
111 private final Map<LocalHistoryIdentifier, ShardDataTreeTransactionChain> transactionChains = new HashMap<>();
113 private final DataTreeCohortActorRegistry cohortRegistry = new DataTreeCohortActorRegistry();
114 private final Queue<CommitEntry> pendingTransactions = new ArrayDeque<>();
115 private final Queue<CommitEntry> pendingCommits = new ArrayDeque<>();
116 private final Queue<CommitEntry> pendingFinishCommits = new ArrayDeque<>();
119 * Callbacks that need to be invoked once a payload is replicated.
121 private final Map<Payload, Runnable> replicationCallbacks = new HashMap<>();
123 private final ShardDataTreeChangeListenerPublisher treeChangeListenerPublisher;
124 private final ShardDataChangeListenerPublisher dataChangeListenerPublisher;
125 private final Collection<ShardDataTreeMetadata<?>> metadata;
126 private final TipProducingDataTree dataTree;
127 private final String logContext;
128 private final Shard shard;
129 private Runnable runOnPendingTransactionsComplete;
132 * Optimistic {@link DataTreeCandidate} preparation. Since our DataTree implementation is a
133 * {@link TipProducingDataTree}, each {@link DataTreeCandidate} is also a {@link DataTreeTip}, e.g. another
134 * candidate can be prepared on top of it. They still need to be committed in sequence. Here we track the current
135 * tip of the data tree, which is the last DataTreeCandidate we have in flight, or the DataTree itself.
137 private TipProducingDataTreeTip tip;
139 private SchemaContext schemaContext;
141 ShardDataTree(final Shard shard, final SchemaContext schemaContext, final TipProducingDataTree dataTree,
142 final ShardDataTreeChangeListenerPublisher treeChangeListenerPublisher,
143 final ShardDataChangeListenerPublisher dataChangeListenerPublisher, final String logContext,
144 final ShardDataTreeMetadata<?>... metadata) {
145 this.dataTree = Preconditions.checkNotNull(dataTree);
146 updateSchemaContext(schemaContext);
148 this.shard = Preconditions.checkNotNull(shard);
149 this.treeChangeListenerPublisher = Preconditions.checkNotNull(treeChangeListenerPublisher);
150 this.dataChangeListenerPublisher = Preconditions.checkNotNull(dataChangeListenerPublisher);
151 this.logContext = Preconditions.checkNotNull(logContext);
152 this.metadata = ImmutableList.copyOf(metadata);
156 ShardDataTree(final Shard shard, final SchemaContext schemaContext, final TreeType treeType,
157 final YangInstanceIdentifier root,
158 final ShardDataTreeChangeListenerPublisher treeChangeListenerPublisher,
159 final ShardDataChangeListenerPublisher dataChangeListenerPublisher, final String logContext,
160 final ShardDataTreeMetadata<?>... metadata) {
161 this(shard, schemaContext, InMemoryDataTreeFactory.getInstance().create(treeType, root),
162 treeChangeListenerPublisher, dataChangeListenerPublisher, logContext, metadata);
166 public ShardDataTree(final Shard shard, final SchemaContext schemaContext, final TreeType treeType) {
167 this(shard, schemaContext, treeType, YangInstanceIdentifier.EMPTY,
168 new DefaultShardDataTreeChangeListenerPublisher(), new DefaultShardDataChangeListenerPublisher(), "");
171 final String logContext() {
175 final Ticker ticker() {
176 return shard.ticker();
179 public TipProducingDataTree getDataTree() {
183 SchemaContext getSchemaContext() {
184 return schemaContext;
187 void updateSchemaContext(final SchemaContext newSchemaContext) {
188 dataTree.setSchemaContext(newSchemaContext);
189 this.schemaContext = Preconditions.checkNotNull(newSchemaContext);
193 * Take a snapshot of current state for later recovery.
195 * @return A state snapshot
197 @Nonnull ShardDataTreeSnapshot takeStateSnapshot() {
198 final NormalizedNode<?, ?> rootNode = dataTree.takeSnapshot().readNode(YangInstanceIdentifier.EMPTY).get();
199 final Builder<Class<? extends ShardDataTreeSnapshotMetadata<?>>, ShardDataTreeSnapshotMetadata<?>> metaBuilder =
200 ImmutableMap.builder();
202 for (ShardDataTreeMetadata<?> m : metadata) {
203 final ShardDataTreeSnapshotMetadata<?> meta = m.toSnapshot();
205 metaBuilder.put(meta.getType(), meta);
209 return new MetadataShardDataTreeSnapshot(rootNode, metaBuilder.build());
212 private boolean anyPendingTransactions() {
213 return !pendingTransactions.isEmpty() || !pendingCommits.isEmpty() || !pendingFinishCommits.isEmpty();
216 private void applySnapshot(@Nonnull final ShardDataTreeSnapshot snapshot,
217 final UnaryOperator<DataTreeModification> wrapper) throws DataValidationFailedException {
218 final Stopwatch elapsed = Stopwatch.createStarted();
220 if (anyPendingTransactions()) {
221 LOG.warn("{}: applying state snapshot with pending transactions", logContext);
224 final Map<Class<? extends ShardDataTreeSnapshotMetadata<?>>, ShardDataTreeSnapshotMetadata<?>> snapshotMeta;
225 if (snapshot instanceof MetadataShardDataTreeSnapshot) {
226 snapshotMeta = ((MetadataShardDataTreeSnapshot) snapshot).getMetadata();
228 snapshotMeta = ImmutableMap.of();
231 for (ShardDataTreeMetadata<?> m : metadata) {
232 final ShardDataTreeSnapshotMetadata<?> s = snapshotMeta.get(m.getSupportedType());
240 final DataTreeModification mod = wrapper.apply(dataTree.takeSnapshot().newModification());
241 // delete everything first
242 mod.delete(YangInstanceIdentifier.EMPTY);
244 final java.util.Optional<NormalizedNode<?, ?>> maybeNode = snapshot.getRootNode();
245 if (maybeNode.isPresent()) {
246 // Add everything from the remote node back
247 mod.write(YangInstanceIdentifier.EMPTY, maybeNode.get());
251 final DataTreeModification unwrapped = unwrap(mod);
252 dataTree.validate(unwrapped);
253 DataTreeCandidateTip candidate = dataTree.prepare(unwrapped);
254 dataTree.commit(candidate);
255 notifyListeners(candidate);
257 LOG.debug("{}: state snapshot applied in %s", logContext, elapsed);
261 * Apply a snapshot coming from the leader. This method assumes the leader and follower SchemaContexts match and
262 * does not perform any pruning.
264 * @param snapshot Snapshot that needs to be applied
265 * @throws DataValidationFailedException when the snapshot fails to apply
267 void applySnapshot(@Nonnull final ShardDataTreeSnapshot snapshot) throws DataValidationFailedException {
268 applySnapshot(snapshot, UnaryOperator.identity());
271 private PruningDataTreeModification wrapWithPruning(final DataTreeModification delegate) {
272 return new PruningDataTreeModification(delegate, dataTree, schemaContext);
275 private static DataTreeModification unwrap(final DataTreeModification modification) {
276 if (modification instanceof PruningDataTreeModification) {
277 return ((PruningDataTreeModification)modification).delegate();
283 * Apply a snapshot coming from recovery. This method does not assume the SchemaContexts match and performs data
284 * pruning in an attempt to adjust the state to our current SchemaContext.
286 * @param snapshot Snapshot that needs to be applied
287 * @throws DataValidationFailedException when the snapshot fails to apply
289 void applyRecoverySnapshot(final @Nonnull ShardDataTreeSnapshot snapshot) throws DataValidationFailedException {
290 applySnapshot(snapshot, this::wrapWithPruning);
293 @SuppressWarnings("checkstyle:IllegalCatch")
294 private void applyRecoveryCandidate(final DataTreeCandidate candidate) throws DataValidationFailedException {
295 final PruningDataTreeModification mod = wrapWithPruning(dataTree.takeSnapshot().newModification());
296 DataTreeCandidates.applyToModification(mod, candidate);
299 final DataTreeModification unwrapped = mod.delegate();
300 LOG.trace("{}: Applying recovery modification {}", logContext, unwrapped);
303 dataTree.validate(unwrapped);
304 dataTree.commit(dataTree.prepare(unwrapped));
305 } catch (Exception e) {
306 File file = new File(System.getProperty("karaf.data", "."),
307 "failed-recovery-payload-" + logContext + ".out");
308 DataTreeModificationOutput.toFile(file, unwrapped);
309 throw new IllegalStateException(String.format(
310 "%s: Failed to apply recovery payload. Modification data was written to file %s",
311 logContext, file), e);
316 * Apply a payload coming from recovery. This method does not assume the SchemaContexts match and performs data
317 * pruning in an attempt to adjust the state to our current SchemaContext.
319 * @param payload Payload
320 * @throws IOException when the snapshot fails to deserialize
321 * @throws DataValidationFailedException when the snapshot fails to apply
323 void applyRecoveryPayload(final @Nonnull Payload payload) throws IOException, DataValidationFailedException {
324 if (payload instanceof CommitTransactionPayload) {
325 final Entry<TransactionIdentifier, DataTreeCandidate> e =
326 ((CommitTransactionPayload) payload).getCandidate();
327 applyRecoveryCandidate(e.getValue());
328 allMetadataCommittedTransaction(e.getKey());
329 } else if (payload instanceof AbortTransactionPayload) {
330 allMetadataAbortedTransaction(((AbortTransactionPayload) payload).getIdentifier());
331 } else if (payload instanceof PurgeTransactionPayload) {
332 allMetadataPurgedTransaction(((PurgeTransactionPayload) payload).getIdentifier());
333 } else if (payload instanceof CreateLocalHistoryPayload) {
334 allMetadataCreatedLocalHistory(((CreateLocalHistoryPayload) payload).getIdentifier());
335 } else if (payload instanceof CloseLocalHistoryPayload) {
336 allMetadataClosedLocalHistory(((CloseLocalHistoryPayload) payload).getIdentifier());
337 } else if (payload instanceof PurgeLocalHistoryPayload) {
338 allMetadataPurgedLocalHistory(((PurgeLocalHistoryPayload) payload).getIdentifier());
340 LOG.debug("{}: ignoring unhandled payload {}", logContext, payload);
344 private void applyReplicatedCandidate(final Identifier identifier, final DataTreeCandidate foreign)
345 throws DataValidationFailedException {
346 LOG.debug("{}: Applying foreign transaction {}", logContext, identifier);
348 final DataTreeModification mod = dataTree.takeSnapshot().newModification();
349 DataTreeCandidates.applyToModification(mod, foreign);
352 LOG.trace("{}: Applying foreign modification {}", logContext, mod);
353 dataTree.validate(mod);
354 final DataTreeCandidate candidate = dataTree.prepare(mod);
355 dataTree.commit(candidate);
357 notifyListeners(candidate);
361 * Apply a payload coming from the leader, which could actually be us. This method assumes the leader and follower
362 * SchemaContexts match and does not perform any pruning.
364 * @param identifier Payload identifier as returned from RaftActor
365 * @param payload Payload
366 * @throws IOException when the snapshot fails to deserialize
367 * @throws DataValidationFailedException when the snapshot fails to apply
369 void applyReplicatedPayload(final Identifier identifier, final Payload payload) throws IOException,
370 DataValidationFailedException {
372 * This is a bit more involved than it needs to be due to to the fact we do not want to be touching the payload
373 * if we are the leader and it has originated with us.
375 * The identifier will only ever be non-null when we were the leader which achieved consensus. Unfortunately,
376 * though, this may not be the case anymore, as we are being called some time afterwards and we may not be
377 * acting in that capacity anymore.
379 * In any case, we know that this is an entry coming from replication, hence we can be sure we will not observe
380 * pre-Boron state -- which limits the number of options here.
382 if (payload instanceof CommitTransactionPayload) {
383 if (identifier == null) {
384 final Entry<TransactionIdentifier, DataTreeCandidate> e =
385 ((CommitTransactionPayload) payload).getCandidate();
386 applyReplicatedCandidate(e.getKey(), e.getValue());
387 allMetadataCommittedTransaction(e.getKey());
389 Verify.verify(identifier instanceof TransactionIdentifier);
390 payloadReplicationComplete((TransactionIdentifier) identifier);
392 } else if (payload instanceof AbortTransactionPayload) {
393 if (identifier != null) {
394 payloadReplicationComplete((AbortTransactionPayload) payload);
396 allMetadataAbortedTransaction(((AbortTransactionPayload) payload).getIdentifier());
398 } else if (payload instanceof PurgeTransactionPayload) {
399 if (identifier != null) {
400 payloadReplicationComplete((PurgeTransactionPayload) payload);
402 allMetadataPurgedTransaction(((PurgeTransactionPayload) payload).getIdentifier());
404 } else if (payload instanceof CloseLocalHistoryPayload) {
405 if (identifier != null) {
406 payloadReplicationComplete((CloseLocalHistoryPayload) payload);
408 allMetadataClosedLocalHistory(((CloseLocalHistoryPayload) payload).getIdentifier());
410 } else if (payload instanceof CreateLocalHistoryPayload) {
411 if (identifier != null) {
412 payloadReplicationComplete((CreateLocalHistoryPayload)payload);
414 allMetadataCreatedLocalHistory(((CreateLocalHistoryPayload) payload).getIdentifier());
416 } else if (payload instanceof PurgeLocalHistoryPayload) {
417 if (identifier != null) {
418 payloadReplicationComplete((PurgeLocalHistoryPayload)payload);
420 allMetadataPurgedLocalHistory(((PurgeLocalHistoryPayload) payload).getIdentifier());
423 LOG.warn("{}: ignoring unhandled identifier {} payload {}", logContext, identifier, payload);
427 private void replicatePayload(final Identifier id, final Payload payload, @Nullable final Runnable callback) {
428 if (callback != null) {
429 replicationCallbacks.put(payload, callback);
431 shard.persistPayload(id, payload, true);
434 private void payloadReplicationComplete(final AbstractIdentifiablePayload<?> payload) {
435 final Runnable callback = replicationCallbacks.remove(payload);
436 if (callback != null) {
437 LOG.debug("{}: replication of {} completed, invoking {}", logContext, payload.getIdentifier(), callback);
440 LOG.debug("{}: replication of {} has no callback", logContext, payload.getIdentifier());
444 private void payloadReplicationComplete(final TransactionIdentifier txId) {
445 final CommitEntry current = pendingFinishCommits.peek();
446 if (current == null) {
447 LOG.warn("{}: No outstanding transactions, ignoring consensus on transaction {}", logContext, txId);
451 if (!current.cohort.getIdentifier().equals(txId)) {
452 LOG.debug("{}: Head of pendingFinishCommits queue is {}, ignoring consensus on transaction {}", logContext,
453 current.cohort.getIdentifier(), txId);
457 finishCommit(current.cohort);
460 private void allMetadataAbortedTransaction(final TransactionIdentifier txId) {
461 for (ShardDataTreeMetadata<?> m : metadata) {
462 m.onTransactionAborted(txId);
466 private void allMetadataCommittedTransaction(final TransactionIdentifier txId) {
467 for (ShardDataTreeMetadata<?> m : metadata) {
468 m.onTransactionCommitted(txId);
472 private void allMetadataPurgedTransaction(final TransactionIdentifier txId) {
473 for (ShardDataTreeMetadata<?> m : metadata) {
474 m.onTransactionPurged(txId);
478 private void allMetadataCreatedLocalHistory(final LocalHistoryIdentifier historyId) {
479 for (ShardDataTreeMetadata<?> m : metadata) {
480 m.onHistoryCreated(historyId);
484 private void allMetadataClosedLocalHistory(final LocalHistoryIdentifier historyId) {
485 for (ShardDataTreeMetadata<?> m : metadata) {
486 m.onHistoryClosed(historyId);
490 private void allMetadataPurgedLocalHistory(final LocalHistoryIdentifier historyId) {
491 for (ShardDataTreeMetadata<?> m : metadata) {
492 m.onHistoryPurged(historyId);
497 * Create a transaction chain for specified history. Unlike {@link #ensureTransactionChain(LocalHistoryIdentifier)},
498 * this method is used for re-establishing state when we are taking over
500 * @param historyId Local history identifier
501 * @param closed True if the chain should be created in closed state (i.e. pending purge)
502 * @return Transaction chain handle
504 ShardDataTreeTransactionChain recreateTransactionChain(final LocalHistoryIdentifier historyId,
505 final boolean closed) {
506 final ShardDataTreeTransactionChain ret = new ShardDataTreeTransactionChain(historyId, this);
507 final ShardDataTreeTransactionChain existing = transactionChains.putIfAbsent(historyId, ret);
508 Preconditions.checkState(existing == null, "Attempted to recreate chain %s, but %s already exists", historyId,
513 ShardDataTreeTransactionChain ensureTransactionChain(final LocalHistoryIdentifier historyId) {
514 ShardDataTreeTransactionChain chain = transactionChains.get(historyId);
516 chain = new ShardDataTreeTransactionChain(historyId, this);
517 transactionChains.put(historyId, chain);
518 shard.persistPayload(historyId, CreateLocalHistoryPayload.create(historyId), true);
524 ReadOnlyShardDataTreeTransaction newReadOnlyTransaction(final TransactionIdentifier txId) {
525 if (txId.getHistoryId().getHistoryId() == 0) {
526 return new ReadOnlyShardDataTreeTransaction(this, txId, dataTree.takeSnapshot());
529 return ensureTransactionChain(txId.getHistoryId()).newReadOnlyTransaction(txId);
532 ReadWriteShardDataTreeTransaction newReadWriteTransaction(final TransactionIdentifier txId) {
533 if (txId.getHistoryId().getHistoryId() == 0) {
534 return new ReadWriteShardDataTreeTransaction(ShardDataTree.this, txId, dataTree.takeSnapshot()
538 return ensureTransactionChain(txId.getHistoryId()).newReadWriteTransaction(txId);
542 public void notifyListeners(final DataTreeCandidate candidate) {
543 treeChangeListenerPublisher.publishChanges(candidate, logContext);
544 dataChangeListenerPublisher.publishChanges(candidate, logContext);
547 void notifyOfInitialData(final DataChangeListenerRegistration<AsyncDataChangeListener<YangInstanceIdentifier,
548 NormalizedNode<?, ?>>> listenerReg, final Optional<DataTreeCandidate> currentState) {
549 if (currentState.isPresent()) {
550 ShardDataChangeListenerPublisher localPublisher = dataChangeListenerPublisher.newInstance();
551 localPublisher.registerDataChangeListener(listenerReg.getPath(), listenerReg.getInstance(),
552 listenerReg.getScope());
553 localPublisher.publishChanges(currentState.get(), logContext);
557 void notifyOfInitialData(final YangInstanceIdentifier path, final DOMDataTreeChangeListener listener,
558 final Optional<DataTreeCandidate> currentState) {
559 if (currentState.isPresent()) {
560 ShardDataTreeChangeListenerPublisher localPublisher = treeChangeListenerPublisher.newInstance();
561 localPublisher.registerTreeChangeListener(path, listener);
562 localPublisher.publishChanges(currentState.get(), logContext);
567 * Immediately purge all state relevant to leader. This includes all transaction chains and any scheduled
568 * replication callbacks.
570 void purgeLeaderState() {
571 for (ShardDataTreeTransactionChain chain : transactionChains.values()) {
575 transactionChains.clear();
576 replicationCallbacks.clear();
580 * Close a single transaction chain.
582 * @param id History identifier
583 * @param callback Callback to invoke upon completion, may be null
585 void closeTransactionChain(final LocalHistoryIdentifier id, @Nullable final Runnable callback) {
586 final ShardDataTreeTransactionChain chain = transactionChains.get(id);
588 LOG.debug("{}: Closing non-existent transaction chain {}", logContext, id);
589 if (callback != null) {
596 replicatePayload(id, CloseLocalHistoryPayload.create(id), callback);
600 * Purge a single transaction chain.
602 * @param id History identifier
603 * @param callback Callback to invoke upon completion, may be null
605 void purgeTransactionChain(final LocalHistoryIdentifier id, @Nullable final Runnable callback) {
606 final ShardDataTreeTransactionChain chain = transactionChains.remove(id);
608 LOG.debug("{}: Purging non-existent transaction chain {}", logContext, id);
609 if (callback != null) {
615 replicatePayload(id, PurgeLocalHistoryPayload.create(id), callback);
618 Entry<DataChangeListenerRegistration<AsyncDataChangeListener<YangInstanceIdentifier, NormalizedNode<?, ?>>>,
619 Optional<DataTreeCandidate>> registerChangeListener(final YangInstanceIdentifier path,
620 final AsyncDataChangeListener<YangInstanceIdentifier, NormalizedNode<?, ?>> listener,
621 final DataChangeScope scope) {
622 DataChangeListenerRegistration<AsyncDataChangeListener<YangInstanceIdentifier, NormalizedNode<?, ?>>> reg =
623 dataChangeListenerPublisher.registerDataChangeListener(path, listener, scope);
625 return new SimpleEntry<>(reg, readCurrentData());
628 private Optional<DataTreeCandidate> readCurrentData() {
629 final Optional<NormalizedNode<?, ?>> currentState =
630 dataTree.takeSnapshot().readNode(YangInstanceIdentifier.EMPTY);
631 return currentState.isPresent() ? Optional.of(DataTreeCandidates.fromNormalizedNode(
632 YangInstanceIdentifier.EMPTY, currentState.get())) : Optional.<DataTreeCandidate>absent();
635 public Entry<ListenerRegistration<DOMDataTreeChangeListener>, Optional<DataTreeCandidate>>
636 registerTreeChangeListener(final YangInstanceIdentifier path, final DOMDataTreeChangeListener listener) {
637 final ListenerRegistration<DOMDataTreeChangeListener> reg =
638 treeChangeListenerPublisher.registerTreeChangeListener(path, listener);
640 return new SimpleEntry<>(reg, readCurrentData());
644 return pendingTransactions.size() + pendingCommits.size() + pendingFinishCommits.size();
648 void abortTransaction(final AbstractShardDataTreeTransaction<?> transaction, final Runnable callback) {
649 final TransactionIdentifier id = transaction.getIdentifier();
650 LOG.debug("{}: aborting transaction {}", logContext, id);
651 replicatePayload(id, AbortTransactionPayload.create(id), callback);
655 void abortFromTransactionActor(final AbstractShardDataTreeTransaction<?> transaction) {
656 // No-op for free-standing transactions
661 ShardDataTreeCohort finishTransaction(final ReadWriteShardDataTreeTransaction transaction) {
662 final DataTreeModification snapshot = transaction.getSnapshot();
665 return createReadyCohort(transaction.getIdentifier(), snapshot);
668 void purgeTransaction(final TransactionIdentifier id, final Runnable callback) {
669 LOG.debug("{}: purging transaction {}", logContext, id);
670 replicatePayload(id, PurgeTransactionPayload.create(id), callback);
673 public Optional<NormalizedNode<?, ?>> readNode(final YangInstanceIdentifier path) {
674 return dataTree.takeSnapshot().readNode(path);
677 DataTreeSnapshot takeSnapshot() {
678 return dataTree.takeSnapshot();
682 public DataTreeModification newModification() {
683 return dataTree.takeSnapshot().newModification();
687 * Commits a modification.
689 * @deprecated This method violates DataTree containment and will be removed.
693 public DataTreeCandidate commit(final DataTreeModification modification) throws DataValidationFailedException {
694 // Direct modification commit is a utility, which cannot be used while we have transactions in-flight
695 Preconditions.checkState(tip == dataTree, "Cannot modify data tree while transacgitons are pending");
697 modification.ready();
698 dataTree.validate(modification);
699 DataTreeCandidate candidate = dataTree.prepare(modification);
700 dataTree.commit(candidate);
704 public Collection<ShardDataTreeCohort> getAndClearPendingTransactions() {
705 Collection<ShardDataTreeCohort> ret = new ArrayList<>(getQueueSize());
707 for (CommitEntry entry: pendingFinishCommits) {
708 ret.add(entry.cohort);
711 for (CommitEntry entry: pendingCommits) {
712 ret.add(entry.cohort);
715 for (CommitEntry entry: pendingTransactions) {
716 ret.add(entry.cohort);
719 pendingFinishCommits.clear();
720 pendingCommits.clear();
721 pendingTransactions.clear();
726 @SuppressWarnings("checkstyle:IllegalCatch")
727 private void processNextPendingTransaction() {
728 processNextPending(pendingTransactions, State.CAN_COMMIT_PENDING, entry -> {
729 final SimpleShardDataTreeCohort cohort = entry.cohort;
730 final DataTreeModification modification = cohort.getDataTreeModification();
732 LOG.debug("{}: Validating transaction {}", logContext, cohort.getIdentifier());
735 cohort.throwCanCommitFailure();
737 tip.validate(modification);
738 LOG.debug("{}: Transaction {} validated", logContext, cohort.getIdentifier());
739 cohort.successfulCanCommit();
740 entry.lastAccess = ticker().read();
742 } catch (ConflictingModificationAppliedException e) {
743 LOG.warn("{}: Store Tx {}: Conflicting modification for path {}.", logContext, cohort.getIdentifier(),
745 cause = new OptimisticLockFailedException("Optimistic lock failed.", e);
746 } catch (DataValidationFailedException e) {
747 LOG.warn("{}: Store Tx {}: Data validation failed for path {}.", logContext, cohort.getIdentifier(),
750 // For debugging purposes, allow dumping of the modification. Coupled with the above
751 // precondition log, it should allow us to understand what went on.
752 LOG.debug("{}: Store Tx {}: modifications: {} tree: {}", cohort.getIdentifier(), modification,
754 cause = new TransactionCommitFailedException("Data did not pass validation.", e);
755 } catch (Exception e) {
756 LOG.warn("{}: Unexpected failure in validation phase", logContext, e);
760 // Failure path: propagate the failure, remove the transaction from the queue and loop to the next one
761 pendingTransactions.poll().cohort.failedCanCommit(cause);
765 private void processNextPending() {
766 processNextPendingCommit();
767 processNextPendingTransaction();
770 private void processNextPending(final Queue<CommitEntry> queue, final State allowedState,
771 final Consumer<CommitEntry> processor) {
772 while (!queue.isEmpty()) {
773 final CommitEntry entry = queue.peek();
774 final SimpleShardDataTreeCohort cohort = entry.cohort;
776 if (cohort.isFailed()) {
777 LOG.debug("{}: Removing failed transaction {}", logContext, cohort.getIdentifier());
782 if (cohort.getState() == allowedState) {
783 processor.accept(entry);
789 maybeRunOperationOnPendingTransactionsComplete();
792 private void processNextPendingCommit() {
793 processNextPending(pendingCommits, State.COMMIT_PENDING,
794 entry -> startCommit(entry.cohort, entry.cohort.getCandidate()));
797 private boolean peekNextPendingCommit() {
798 final CommitEntry first = pendingCommits.peek();
799 return first != null && first.cohort.getState() == State.COMMIT_PENDING;
802 void startCanCommit(final SimpleShardDataTreeCohort cohort) {
803 final SimpleShardDataTreeCohort current = pendingTransactions.peek().cohort;
804 if (!cohort.equals(current)) {
805 LOG.debug("{}: Transaction {} scheduled for canCommit step", logContext, cohort.getIdentifier());
809 processNextPendingTransaction();
812 private void failPreCommit(final Exception cause) {
813 shard.getShardMBean().incrementFailedTransactionsCount();
814 pendingTransactions.poll().cohort.failedPreCommit(cause);
815 processNextPendingTransaction();
818 @SuppressWarnings("checkstyle:IllegalCatch")
819 void startPreCommit(final SimpleShardDataTreeCohort cohort) {
820 final CommitEntry entry = pendingTransactions.peek();
821 Preconditions.checkState(entry != null, "Attempted to pre-commit of %s when no transactions pending", cohort);
823 final SimpleShardDataTreeCohort current = entry.cohort;
824 Verify.verify(cohort.equals(current), "Attempted to pre-commit %s while %s is pending", cohort, current);
826 LOG.debug("{}: Preparing transaction {}", logContext, current.getIdentifier());
828 final DataTreeCandidateTip candidate;
830 candidate = tip.prepare(cohort.getDataTreeModification());
831 cohort.userPreCommit(candidate);
832 } catch (ExecutionException | TimeoutException | RuntimeException e) {
837 // Set the tip of the data tree.
838 tip = Verify.verifyNotNull(candidate);
840 entry.lastAccess = ticker().read();
842 pendingTransactions.remove();
843 pendingCommits.add(entry);
845 LOG.debug("{}: Transaction {} prepared", logContext, current.getIdentifier());
847 cohort.successfulPreCommit(candidate);
849 processNextPendingTransaction();
852 private void failCommit(final Exception cause) {
853 shard.getShardMBean().incrementFailedTransactionsCount();
854 pendingFinishCommits.poll().cohort.failedCommit(cause);
855 processNextPending();
858 @SuppressWarnings("checkstyle:IllegalCatch")
859 private void finishCommit(final SimpleShardDataTreeCohort cohort) {
860 final TransactionIdentifier txId = cohort.getIdentifier();
861 final DataTreeCandidate candidate = cohort.getCandidate();
863 LOG.debug("{}: Resuming commit of transaction {}", logContext, txId);
865 if (tip == candidate) {
866 // All pending candidates have been committed, reset the tip to the data tree.
871 dataTree.commit(candidate);
872 } catch (Exception e) {
873 LOG.error("{}: Failed to commit transaction {}", logContext, txId, e);
878 shard.getShardMBean().incrementCommittedTransactionCount();
879 shard.getShardMBean().setLastCommittedTransactionTime(System.currentTimeMillis());
881 // FIXME: propagate journal index
882 pendingFinishCommits.poll().cohort.successfulCommit(UnsignedLong.ZERO);
884 LOG.trace("{}: Transaction {} committed, proceeding to notify", logContext, txId);
885 notifyListeners(candidate);
887 processNextPending();
890 void startCommit(final SimpleShardDataTreeCohort cohort, final DataTreeCandidate candidate) {
891 final CommitEntry entry = pendingCommits.peek();
892 Preconditions.checkState(entry != null, "Attempted to start commit of %s when no transactions pending", cohort);
894 final SimpleShardDataTreeCohort current = entry.cohort;
895 if (!cohort.equals(current)) {
896 LOG.debug("{}: Transaction {} scheduled for commit step", logContext, cohort.getIdentifier());
900 LOG.debug("{}: Starting commit for transaction {}", logContext, current.getIdentifier());
902 final TransactionIdentifier txId = cohort.getIdentifier();
903 final Payload payload;
905 payload = CommitTransactionPayload.create(txId, candidate);
906 } catch (IOException e) {
907 LOG.error("{}: Failed to encode transaction {} candidate {}", logContext, txId, candidate, e);
908 pendingCommits.poll().cohort.failedCommit(e);
909 processNextPending();
913 // We process next transactions pending canCommit before we call persistPayload to possibly progress subsequent
914 // transactions to the COMMIT_PENDING state so the payloads can be batched for replication. This is done for
915 // single-shard transactions that immediately transition from canCommit to preCommit to commit. Note that
916 // if the next pending transaction is progressed to COMMIT_PENDING and this method (startCommit) is called,
917 // the next transaction will not attempt to replicate b/c the current transaction is still at the head of the
918 // pendingCommits queue.
919 processNextPendingTransaction();
921 // After processing next pending transactions, we can now remove the current transaction from pendingCommits.
922 // Note this must be done before the call to peekNextPendingCommit below so we check the next transaction
923 // in order to properly determine the batchHint flag for the call to persistPayload.
924 pendingCommits.remove();
925 pendingFinishCommits.add(entry);
927 // See if the next transaction is pending commit (ie in the COMMIT_PENDING state) so it can be batched with
928 // this transaction for replication.
929 boolean replicationBatchHint = peekNextPendingCommit();
931 // Once completed, we will continue via payloadReplicationComplete
932 shard.persistPayload(txId, payload, replicationBatchHint);
934 entry.lastAccess = shard.ticker().read();
936 LOG.debug("{}: Transaction {} submitted to persistence", logContext, txId);
938 // Process the next transaction pending commit, if any. If there is one it will be batched with this
939 // transaction for replication.
940 processNextPendingCommit();
943 Collection<ActorRef> getCohortActors() {
944 return cohortRegistry.getCohortActors();
947 void processCohortRegistryCommand(final ActorRef sender, final CohortRegistryCommand message) {
948 cohortRegistry.process(sender, message);
953 ShardDataTreeCohort createFailedCohort(final TransactionIdentifier txId, final DataTreeModification mod,
954 final Exception failure) {
955 SimpleShardDataTreeCohort cohort = new SimpleShardDataTreeCohort.DeadOnArrival(this, mod, txId, failure);
956 pendingTransactions.add(new CommitEntry(cohort, ticker().read()));
961 ShardDataTreeCohort createReadyCohort(final TransactionIdentifier txId,
962 final DataTreeModification mod) {
963 SimpleShardDataTreeCohort cohort = new SimpleShardDataTreeCohort.Normal(this, mod, txId,
964 cohortRegistry.createCohort(schemaContext, txId, COMMIT_STEP_TIMEOUT));
965 pendingTransactions.add(new CommitEntry(cohort, ticker().read()));
969 @SuppressFBWarnings(value = "DB_DUPLICATE_SWITCH_CLAUSES", justification = "See inline comments below.")
970 void checkForExpiredTransactions(final long transactionCommitTimeoutMillis) {
971 final long timeout = TimeUnit.MILLISECONDS.toNanos(transactionCommitTimeoutMillis);
972 final long now = ticker().read();
974 final Queue<CommitEntry> currentQueue = !pendingFinishCommits.isEmpty() ? pendingFinishCommits :
975 !pendingCommits.isEmpty() ? pendingCommits : pendingTransactions;
976 final CommitEntry currentTx = currentQueue.peek();
977 if (currentTx != null && currentTx.lastAccess + timeout < now) {
978 LOG.warn("{}: Current transaction {} has timed out after {} ms in state {}", logContext,
979 currentTx.cohort.getIdentifier(), transactionCommitTimeoutMillis, currentTx.cohort.getState());
980 boolean processNext = true;
981 switch (currentTx.cohort.getState()) {
982 case CAN_COMMIT_PENDING:
983 currentQueue.remove().cohort.failedCanCommit(new TimeoutException());
985 case CAN_COMMIT_COMPLETE:
986 // The suppression of the FindBugs "DB_DUPLICATE_SWITCH_CLAUSES" warning pertains to this clause
987 // whose code is duplicated with PRE_COMMIT_COMPLETE. The clauses aren't combined in case the code
988 // in PRE_COMMIT_COMPLETE is changed.
989 currentQueue.remove().cohort.reportFailure(new TimeoutException());
991 case PRE_COMMIT_PENDING:
992 currentQueue.remove().cohort.failedPreCommit(new TimeoutException());
994 case PRE_COMMIT_COMPLETE:
995 // FIXME: this is a legacy behavior problem. Three-phase commit protocol specifies that after we
996 // are ready we should commit the transaction, not abort it. Our current software stack does
997 // not allow us to do that consistently, because we persist at the time of commit, hence
998 // we can end up in a state where we have pre-committed a transaction, then a leader failover
999 // occurred ... the new leader does not see the pre-committed transaction and does not have
1000 // a running timer. To fix this we really need two persistence events.
1002 // The first one, done at pre-commit time will hold the transaction payload. When consensus
1003 // is reached, we exit the pre-commit phase and start the pre-commit timer. Followers do not
1004 // apply the state in this event.
1006 // The second one, done at commit (or abort) time holds only the transaction identifier and
1007 // signals to followers that the state should (or should not) be applied.
1009 // In order to make the pre-commit timer working across failovers, though, we need
1010 // a per-shard cluster-wide monotonic time, so a follower becoming the leader can accurately
1011 // restart the timer.
1012 currentQueue.remove().cohort.reportFailure(new TimeoutException());
1014 case COMMIT_PENDING:
1015 LOG.warn("{}: Transaction {} is still committing, cannot abort", logContext,
1016 currentTx.cohort.getIdentifier());
1017 currentTx.lastAccess = now;
1018 processNext = false;
1025 currentQueue.remove();
1029 processNextPending();
1034 boolean startAbort(final SimpleShardDataTreeCohort cohort) {
1035 final Iterator<CommitEntry> it = Iterables.concat(pendingFinishCommits, pendingCommits,
1036 pendingTransactions).iterator();
1037 if (!it.hasNext()) {
1038 LOG.debug("{}: no open transaction while attempting to abort {}", logContext, cohort.getIdentifier());
1042 // First entry is special, as it may already be committing
1043 final CommitEntry first = it.next();
1044 if (cohort.equals(first.cohort)) {
1045 if (cohort.getState() != State.COMMIT_PENDING) {
1046 LOG.debug("{}: aborting head of queue {} in state {}", logContext, cohort.getIdentifier(),
1047 cohort.getIdentifier());
1050 if (cohort.getCandidate() != null) {
1051 rebaseTransactions(it, dataTree);
1054 processNextPending();
1058 LOG.warn("{}: transaction {} is committing, skipping abort", logContext, cohort.getIdentifier());
1062 TipProducingDataTreeTip newTip = MoreObjects.firstNonNull(first.cohort.getCandidate(), dataTree);
1063 while (it.hasNext()) {
1064 final CommitEntry e = it.next();
1065 if (cohort.equals(e.cohort)) {
1066 LOG.debug("{}: aborting queued transaction {}", logContext, cohort.getIdentifier());
1069 if (cohort.getCandidate() != null) {
1070 rebaseTransactions(it, newTip);
1075 newTip = MoreObjects.firstNonNull(e.cohort.getCandidate(), newTip);
1079 LOG.debug("{}: aborted transaction {} not found in the queue", logContext, cohort.getIdentifier());
1083 @SuppressWarnings("checkstyle:IllegalCatch")
1084 private void rebaseTransactions(final Iterator<CommitEntry> iter, @Nonnull final TipProducingDataTreeTip newTip) {
1085 tip = Preconditions.checkNotNull(newTip);
1086 while (iter.hasNext()) {
1087 final SimpleShardDataTreeCohort cohort = iter.next().cohort;
1088 if (cohort.getState() == State.CAN_COMMIT_COMPLETE) {
1089 LOG.debug("{}: Revalidating queued transaction {}", logContext, cohort.getIdentifier());
1092 tip.validate(cohort.getDataTreeModification());
1093 } catch (DataValidationFailedException | RuntimeException e) {
1094 LOG.debug("{}: Failed to revalidate queued transaction {}", logContext, cohort.getIdentifier(), e);
1095 cohort.reportFailure(e);
1097 } else if (cohort.getState() == State.PRE_COMMIT_COMPLETE) {
1098 LOG.debug("{}: Repreparing queued transaction {}", logContext, cohort.getIdentifier());
1101 tip.validate(cohort.getDataTreeModification());
1102 DataTreeCandidateTip candidate = tip.prepare(cohort.getDataTreeModification());
1103 cohort.userPreCommit(candidate);
1105 cohort.setNewCandidate(candidate);
1107 } catch (ExecutionException | TimeoutException | RuntimeException | DataValidationFailedException e) {
1108 LOG.debug("{}: Failed to reprepare queued transaction {}", logContext, cohort.getIdentifier(), e);
1109 cohort.reportFailure(e);
1115 void setRunOnPendingTransactionsComplete(final Runnable operation) {
1116 runOnPendingTransactionsComplete = operation;
1117 maybeRunOperationOnPendingTransactionsComplete();
1120 private void maybeRunOperationOnPendingTransactionsComplete() {
1121 if (runOnPendingTransactionsComplete != null && !anyPendingTransactions()) {
1122 LOG.debug("{}: Pending transactions complete - running operation {}", logContext,
1123 runOnPendingTransactionsComplete);
1125 runOnPendingTransactionsComplete.run();
1126 runOnPendingTransactionsComplete = null;