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.ArrayDeque;
28 import java.util.ArrayList;
29 import java.util.Collection;
30 import java.util.HashMap;
31 import java.util.Iterator;
33 import java.util.Map.Entry;
34 import java.util.Queue;
35 import java.util.concurrent.ExecutionException;
36 import java.util.concurrent.TimeUnit;
37 import java.util.concurrent.TimeoutException;
38 import java.util.function.Consumer;
39 import java.util.function.UnaryOperator;
40 import javax.annotation.Nonnull;
41 import javax.annotation.Nullable;
42 import javax.annotation.concurrent.NotThreadSafe;
43 import org.opendaylight.controller.cluster.access.concepts.LocalHistoryIdentifier;
44 import org.opendaylight.controller.cluster.access.concepts.TransactionIdentifier;
45 import org.opendaylight.controller.cluster.datastore.DataTreeCohortActorRegistry.CohortRegistryCommand;
46 import org.opendaylight.controller.cluster.datastore.ShardDataTreeCohort.State;
47 import org.opendaylight.controller.cluster.datastore.persisted.AbortTransactionPayload;
48 import org.opendaylight.controller.cluster.datastore.persisted.AbstractIdentifiablePayload;
49 import org.opendaylight.controller.cluster.datastore.persisted.CloseLocalHistoryPayload;
50 import org.opendaylight.controller.cluster.datastore.persisted.CommitTransactionPayload;
51 import org.opendaylight.controller.cluster.datastore.persisted.CreateLocalHistoryPayload;
52 import org.opendaylight.controller.cluster.datastore.persisted.MetadataShardDataTreeSnapshot;
53 import org.opendaylight.controller.cluster.datastore.persisted.PurgeLocalHistoryPayload;
54 import org.opendaylight.controller.cluster.datastore.persisted.PurgeTransactionPayload;
55 import org.opendaylight.controller.cluster.datastore.persisted.ShardDataTreeSnapshot;
56 import org.opendaylight.controller.cluster.datastore.persisted.ShardDataTreeSnapshotMetadata;
57 import org.opendaylight.controller.cluster.datastore.utils.DataTreeModificationOutput;
58 import org.opendaylight.controller.cluster.datastore.utils.PruningDataTreeModification;
59 import org.opendaylight.controller.cluster.raft.protobuff.client.messages.Payload;
60 import org.opendaylight.controller.md.sal.common.api.data.AsyncDataBroker.DataChangeScope;
61 import org.opendaylight.controller.md.sal.common.api.data.AsyncDataChangeListener;
62 import org.opendaylight.controller.md.sal.common.api.data.OptimisticLockFailedException;
63 import org.opendaylight.controller.md.sal.common.api.data.TransactionCommitFailedException;
64 import org.opendaylight.controller.md.sal.dom.api.DOMDataTreeChangeListener;
65 import org.opendaylight.yangtools.concepts.Identifier;
66 import org.opendaylight.yangtools.concepts.ListenerRegistration;
67 import org.opendaylight.yangtools.yang.data.api.YangInstanceIdentifier;
68 import org.opendaylight.yangtools.yang.data.api.schema.NormalizedNode;
69 import org.opendaylight.yangtools.yang.data.api.schema.tree.ConflictingModificationAppliedException;
70 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeCandidate;
71 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeCandidateTip;
72 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeCandidates;
73 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeModification;
74 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeSnapshot;
75 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeTip;
76 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataValidationFailedException;
77 import org.opendaylight.yangtools.yang.data.api.schema.tree.TipProducingDataTree;
78 import org.opendaylight.yangtools.yang.data.api.schema.tree.TipProducingDataTreeTip;
79 import org.opendaylight.yangtools.yang.data.api.schema.tree.TreeType;
80 import org.opendaylight.yangtools.yang.data.impl.schema.tree.InMemoryDataTreeFactory;
81 import org.opendaylight.yangtools.yang.model.api.SchemaContext;
82 import org.slf4j.Logger;
83 import org.slf4j.LoggerFactory;
84 import scala.concurrent.duration.Duration;
87 * Internal shard state, similar to a DOMStore, but optimized for use in the actor system,
88 * e.g. it does not expose public interfaces and assumes it is only ever called from a
92 * This class is not part of the API contract and is subject to change at any time.
95 public class ShardDataTree extends ShardDataTreeTransactionParent {
96 private static final class CommitEntry {
97 final SimpleShardDataTreeCohort cohort;
100 CommitEntry(final SimpleShardDataTreeCohort cohort, final long now) {
101 this.cohort = Preconditions.checkNotNull(cohort);
106 private static final Timeout COMMIT_STEP_TIMEOUT = new Timeout(Duration.create(5, TimeUnit.SECONDS));
107 private static final Logger LOG = LoggerFactory.getLogger(ShardDataTree.class);
109 private final Map<LocalHistoryIdentifier, ShardDataTreeTransactionChain> transactionChains = new HashMap<>();
111 private final DataTreeCohortActorRegistry cohortRegistry = new DataTreeCohortActorRegistry();
112 private final Queue<CommitEntry> pendingTransactions = new ArrayDeque<>();
113 private final Queue<CommitEntry> pendingCommits = new ArrayDeque<>();
114 private final Queue<CommitEntry> pendingFinishCommits = new ArrayDeque<>();
117 * Callbacks that need to be invoked once a payload is replicated.
119 private final Map<Payload, Runnable> replicationCallbacks = new HashMap<>();
121 private final ShardDataTreeChangeListenerPublisher treeChangeListenerPublisher;
122 private final ShardDataChangeListenerPublisher dataChangeListenerPublisher;
123 private final Collection<ShardDataTreeMetadata<?>> metadata;
124 private final TipProducingDataTree dataTree;
125 private final String logContext;
126 private final Shard shard;
127 private Runnable runOnPendingTransactionsComplete;
130 * Optimistic {@link DataTreeCandidate} preparation. Since our DataTree implementation is a
131 * {@link TipProducingDataTree}, each {@link DataTreeCandidate} is also a {@link DataTreeTip}, e.g. another
132 * candidate can be prepared on top of it. They still need to be committed in sequence. Here we track the current
133 * tip of the data tree, which is the last DataTreeCandidate we have in flight, or the DataTree itself.
135 private TipProducingDataTreeTip tip;
137 private SchemaContext schemaContext;
139 ShardDataTree(final Shard shard, final SchemaContext schemaContext, final TipProducingDataTree dataTree,
140 final ShardDataTreeChangeListenerPublisher treeChangeListenerPublisher,
141 final ShardDataChangeListenerPublisher dataChangeListenerPublisher, final String logContext,
142 final ShardDataTreeMetadata<?>... metadata) {
143 this.dataTree = Preconditions.checkNotNull(dataTree);
144 updateSchemaContext(schemaContext);
146 this.shard = Preconditions.checkNotNull(shard);
147 this.treeChangeListenerPublisher = Preconditions.checkNotNull(treeChangeListenerPublisher);
148 this.dataChangeListenerPublisher = Preconditions.checkNotNull(dataChangeListenerPublisher);
149 this.logContext = Preconditions.checkNotNull(logContext);
150 this.metadata = ImmutableList.copyOf(metadata);
154 ShardDataTree(final Shard shard, final SchemaContext schemaContext, final TreeType treeType,
155 final YangInstanceIdentifier root,
156 final ShardDataTreeChangeListenerPublisher treeChangeListenerPublisher,
157 final ShardDataChangeListenerPublisher dataChangeListenerPublisher, final String logContext,
158 final ShardDataTreeMetadata<?>... metadata) {
159 this(shard, schemaContext, InMemoryDataTreeFactory.getInstance().create(treeType, root),
160 treeChangeListenerPublisher, dataChangeListenerPublisher, logContext, metadata);
164 public ShardDataTree(final Shard shard, final SchemaContext schemaContext, final TreeType treeType) {
165 this(shard, schemaContext, treeType, YangInstanceIdentifier.EMPTY,
166 new DefaultShardDataTreeChangeListenerPublisher(""),
167 new DefaultShardDataChangeListenerPublisher(""), "");
170 final String logContext() {
174 final Ticker ticker() {
175 return shard.ticker();
178 public TipProducingDataTree getDataTree() {
182 SchemaContext getSchemaContext() {
183 return schemaContext;
186 void updateSchemaContext(final SchemaContext newSchemaContext) {
187 dataTree.setSchemaContext(newSchemaContext);
188 this.schemaContext = Preconditions.checkNotNull(newSchemaContext);
192 * Take a snapshot of current state for later recovery.
194 * @return A state snapshot
196 @Nonnull ShardDataTreeSnapshot takeStateSnapshot() {
197 final NormalizedNode<?, ?> rootNode = dataTree.takeSnapshot().readNode(YangInstanceIdentifier.EMPTY).get();
198 final Builder<Class<? extends ShardDataTreeSnapshotMetadata<?>>, ShardDataTreeSnapshotMetadata<?>> metaBuilder =
199 ImmutableMap.builder();
201 for (ShardDataTreeMetadata<?> m : metadata) {
202 final ShardDataTreeSnapshotMetadata<?> meta = m.toSnapshot();
204 metaBuilder.put(meta.getType(), meta);
208 return new MetadataShardDataTreeSnapshot(rootNode, metaBuilder.build());
211 private boolean anyPendingTransactions() {
212 return !pendingTransactions.isEmpty() || !pendingCommits.isEmpty() || !pendingFinishCommits.isEmpty();
215 private void applySnapshot(@Nonnull final ShardDataTreeSnapshot snapshot,
216 final UnaryOperator<DataTreeModification> wrapper) throws DataValidationFailedException {
217 final Stopwatch elapsed = Stopwatch.createStarted();
219 if (anyPendingTransactions()) {
220 LOG.warn("{}: applying state snapshot with pending transactions", logContext);
223 final Map<Class<? extends ShardDataTreeSnapshotMetadata<?>>, ShardDataTreeSnapshotMetadata<?>> snapshotMeta;
224 if (snapshot instanceof MetadataShardDataTreeSnapshot) {
225 snapshotMeta = ((MetadataShardDataTreeSnapshot) snapshot).getMetadata();
227 snapshotMeta = ImmutableMap.of();
230 for (ShardDataTreeMetadata<?> m : metadata) {
231 final ShardDataTreeSnapshotMetadata<?> s = snapshotMeta.get(m.getSupportedType());
239 final DataTreeModification mod = wrapper.apply(dataTree.takeSnapshot().newModification());
240 // delete everything first
241 mod.delete(YangInstanceIdentifier.EMPTY);
243 final java.util.Optional<NormalizedNode<?, ?>> maybeNode = snapshot.getRootNode();
244 if (maybeNode.isPresent()) {
245 // Add everything from the remote node back
246 mod.write(YangInstanceIdentifier.EMPTY, maybeNode.get());
250 final DataTreeModification unwrapped = unwrap(mod);
251 dataTree.validate(unwrapped);
252 DataTreeCandidateTip candidate = dataTree.prepare(unwrapped);
253 dataTree.commit(candidate);
254 notifyListeners(candidate);
256 LOG.debug("{}: state snapshot applied in {}", logContext, elapsed);
260 * Apply a snapshot coming from the leader. This method assumes the leader and follower SchemaContexts match and
261 * does not perform any pruning.
263 * @param snapshot Snapshot that needs to be applied
264 * @throws DataValidationFailedException when the snapshot fails to apply
266 void applySnapshot(@Nonnull final ShardDataTreeSnapshot snapshot) throws DataValidationFailedException {
267 applySnapshot(snapshot, UnaryOperator.identity());
270 private PruningDataTreeModification wrapWithPruning(final DataTreeModification delegate) {
271 return new PruningDataTreeModification(delegate, dataTree, schemaContext);
274 private static DataTreeModification unwrap(final DataTreeModification modification) {
275 if (modification instanceof PruningDataTreeModification) {
276 return ((PruningDataTreeModification)modification).delegate();
282 * Apply a snapshot coming from recovery. This method does not assume the SchemaContexts match and performs data
283 * pruning in an attempt to adjust the state to our current SchemaContext.
285 * @param snapshot Snapshot that needs to be applied
286 * @throws DataValidationFailedException when the snapshot fails to apply
288 void applyRecoverySnapshot(final @Nonnull ShardDataTreeSnapshot snapshot) throws DataValidationFailedException {
289 applySnapshot(snapshot, this::wrapWithPruning);
292 @SuppressWarnings("checkstyle:IllegalCatch")
293 private void applyRecoveryCandidate(final DataTreeCandidate candidate) throws DataValidationFailedException {
294 final PruningDataTreeModification mod = wrapWithPruning(dataTree.takeSnapshot().newModification());
295 DataTreeCandidates.applyToModification(mod, candidate);
298 final DataTreeModification unwrapped = mod.delegate();
299 LOG.trace("{}: Applying recovery modification {}", logContext, unwrapped);
302 dataTree.validate(unwrapped);
303 dataTree.commit(dataTree.prepare(unwrapped));
304 } catch (Exception e) {
305 File file = new File(System.getProperty("karaf.data", "."),
306 "failed-recovery-payload-" + logContext + ".out");
307 DataTreeModificationOutput.toFile(file, unwrapped);
308 throw new IllegalStateException(String.format(
309 "%s: Failed to apply recovery payload. Modification data was written to file %s",
310 logContext, file), e);
315 * Apply a payload coming from recovery. This method does not assume the SchemaContexts match and performs data
316 * pruning in an attempt to adjust the state to our current SchemaContext.
318 * @param payload Payload
319 * @throws IOException when the snapshot fails to deserialize
320 * @throws DataValidationFailedException when the snapshot fails to apply
322 void applyRecoveryPayload(final @Nonnull Payload payload) throws IOException, DataValidationFailedException {
323 if (payload instanceof CommitTransactionPayload) {
324 final Entry<TransactionIdentifier, DataTreeCandidate> e =
325 ((CommitTransactionPayload) payload).getCandidate();
326 applyRecoveryCandidate(e.getValue());
327 allMetadataCommittedTransaction(e.getKey());
328 } else if (payload instanceof AbortTransactionPayload) {
329 allMetadataAbortedTransaction(((AbortTransactionPayload) payload).getIdentifier());
330 } else if (payload instanceof PurgeTransactionPayload) {
331 allMetadataPurgedTransaction(((PurgeTransactionPayload) payload).getIdentifier());
332 } else if (payload instanceof CreateLocalHistoryPayload) {
333 allMetadataCreatedLocalHistory(((CreateLocalHistoryPayload) payload).getIdentifier());
334 } else if (payload instanceof CloseLocalHistoryPayload) {
335 allMetadataClosedLocalHistory(((CloseLocalHistoryPayload) payload).getIdentifier());
336 } else if (payload instanceof PurgeLocalHistoryPayload) {
337 allMetadataPurgedLocalHistory(((PurgeLocalHistoryPayload) payload).getIdentifier());
339 LOG.debug("{}: ignoring unhandled payload {}", logContext, payload);
343 private void applyReplicatedCandidate(final Identifier identifier, final DataTreeCandidate foreign)
344 throws DataValidationFailedException {
345 LOG.debug("{}: Applying foreign transaction {}", logContext, identifier);
347 final DataTreeModification mod = dataTree.takeSnapshot().newModification();
348 DataTreeCandidates.applyToModification(mod, foreign);
351 LOG.trace("{}: Applying foreign modification {}", logContext, mod);
352 dataTree.validate(mod);
353 final DataTreeCandidate candidate = dataTree.prepare(mod);
354 dataTree.commit(candidate);
356 notifyListeners(candidate);
360 * Apply a payload coming from the leader, which could actually be us. This method assumes the leader and follower
361 * SchemaContexts match and does not perform any pruning.
363 * @param identifier Payload identifier as returned from RaftActor
364 * @param payload Payload
365 * @throws IOException when the snapshot fails to deserialize
366 * @throws DataValidationFailedException when the snapshot fails to apply
368 void applyReplicatedPayload(final Identifier identifier, final Payload payload) throws IOException,
369 DataValidationFailedException {
371 * 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
372 * if we are the leader and it has originated with us.
374 * The identifier will only ever be non-null when we were the leader which achieved consensus. Unfortunately,
375 * though, this may not be the case anymore, as we are being called some time afterwards and we may not be
376 * acting in that capacity anymore.
378 * In any case, we know that this is an entry coming from replication, hence we can be sure we will not observe
379 * pre-Boron state -- which limits the number of options here.
381 if (payload instanceof CommitTransactionPayload) {
382 if (identifier == null) {
383 final Entry<TransactionIdentifier, DataTreeCandidate> e =
384 ((CommitTransactionPayload) payload).getCandidate();
385 applyReplicatedCandidate(e.getKey(), e.getValue());
386 allMetadataCommittedTransaction(e.getKey());
388 Verify.verify(identifier instanceof TransactionIdentifier);
389 payloadReplicationComplete((TransactionIdentifier) identifier);
391 } else if (payload instanceof AbortTransactionPayload) {
392 if (identifier != null) {
393 payloadReplicationComplete((AbortTransactionPayload) payload);
395 allMetadataAbortedTransaction(((AbortTransactionPayload) payload).getIdentifier());
397 } else if (payload instanceof PurgeTransactionPayload) {
398 if (identifier != null) {
399 payloadReplicationComplete((PurgeTransactionPayload) payload);
401 allMetadataPurgedTransaction(((PurgeTransactionPayload) payload).getIdentifier());
403 } else if (payload instanceof CloseLocalHistoryPayload) {
404 if (identifier != null) {
405 payloadReplicationComplete((CloseLocalHistoryPayload) payload);
407 allMetadataClosedLocalHistory(((CloseLocalHistoryPayload) payload).getIdentifier());
409 } else if (payload instanceof CreateLocalHistoryPayload) {
410 if (identifier != null) {
411 payloadReplicationComplete((CreateLocalHistoryPayload)payload);
413 allMetadataCreatedLocalHistory(((CreateLocalHistoryPayload) payload).getIdentifier());
415 } else if (payload instanceof PurgeLocalHistoryPayload) {
416 if (identifier != null) {
417 payloadReplicationComplete((PurgeLocalHistoryPayload)payload);
419 allMetadataPurgedLocalHistory(((PurgeLocalHistoryPayload) payload).getIdentifier());
422 LOG.warn("{}: ignoring unhandled identifier {} payload {}", logContext, identifier, payload);
426 private void replicatePayload(final Identifier id, final Payload payload, @Nullable final Runnable callback) {
427 if (callback != null) {
428 replicationCallbacks.put(payload, callback);
430 shard.persistPayload(id, payload, true);
433 private void payloadReplicationComplete(final AbstractIdentifiablePayload<?> payload) {
434 final Runnable callback = replicationCallbacks.remove(payload);
435 if (callback != null) {
436 LOG.debug("{}: replication of {} completed, invoking {}", logContext, payload.getIdentifier(), callback);
439 LOG.debug("{}: replication of {} has no callback", logContext, payload.getIdentifier());
443 private void payloadReplicationComplete(final TransactionIdentifier txId) {
444 final CommitEntry current = pendingFinishCommits.peek();
445 if (current == null) {
446 LOG.warn("{}: No outstanding transactions, ignoring consensus on transaction {}", logContext, txId);
450 if (!current.cohort.getIdentifier().equals(txId)) {
451 LOG.debug("{}: Head of pendingFinishCommits queue is {}, ignoring consensus on transaction {}", logContext,
452 current.cohort.getIdentifier(), txId);
456 finishCommit(current.cohort);
459 private void allMetadataAbortedTransaction(final TransactionIdentifier txId) {
460 for (ShardDataTreeMetadata<?> m : metadata) {
461 m.onTransactionAborted(txId);
465 private void allMetadataCommittedTransaction(final TransactionIdentifier txId) {
466 for (ShardDataTreeMetadata<?> m : metadata) {
467 m.onTransactionCommitted(txId);
471 private void allMetadataPurgedTransaction(final TransactionIdentifier txId) {
472 for (ShardDataTreeMetadata<?> m : metadata) {
473 m.onTransactionPurged(txId);
477 private void allMetadataCreatedLocalHistory(final LocalHistoryIdentifier historyId) {
478 for (ShardDataTreeMetadata<?> m : metadata) {
479 m.onHistoryCreated(historyId);
483 private void allMetadataClosedLocalHistory(final LocalHistoryIdentifier historyId) {
484 for (ShardDataTreeMetadata<?> m : metadata) {
485 m.onHistoryClosed(historyId);
489 private void allMetadataPurgedLocalHistory(final LocalHistoryIdentifier historyId) {
490 for (ShardDataTreeMetadata<?> m : metadata) {
491 m.onHistoryPurged(historyId);
496 * Create a transaction chain for specified history. Unlike {@link #ensureTransactionChain(LocalHistoryIdentifier)},
497 * this method is used for re-establishing state when we are taking over
499 * @param historyId Local history identifier
500 * @param closed True if the chain should be created in closed state (i.e. pending purge)
501 * @return Transaction chain handle
503 ShardDataTreeTransactionChain recreateTransactionChain(final LocalHistoryIdentifier historyId,
504 final boolean closed) {
505 final ShardDataTreeTransactionChain ret = new ShardDataTreeTransactionChain(historyId, this);
506 final ShardDataTreeTransactionChain existing = transactionChains.putIfAbsent(historyId, ret);
507 Preconditions.checkState(existing == null, "Attempted to recreate chain %s, but %s already exists", historyId,
512 ShardDataTreeTransactionChain ensureTransactionChain(final LocalHistoryIdentifier historyId) {
513 ShardDataTreeTransactionChain chain = transactionChains.get(historyId);
515 chain = new ShardDataTreeTransactionChain(historyId, this);
516 transactionChains.put(historyId, chain);
517 shard.persistPayload(historyId, CreateLocalHistoryPayload.create(historyId), true);
523 ReadOnlyShardDataTreeTransaction newReadOnlyTransaction(final TransactionIdentifier txId) {
524 if (txId.getHistoryId().getHistoryId() == 0) {
525 return new ReadOnlyShardDataTreeTransaction(this, txId, dataTree.takeSnapshot());
528 return ensureTransactionChain(txId.getHistoryId()).newReadOnlyTransaction(txId);
531 ReadWriteShardDataTreeTransaction newReadWriteTransaction(final TransactionIdentifier txId) {
532 if (txId.getHistoryId().getHistoryId() == 0) {
533 return new ReadWriteShardDataTreeTransaction(ShardDataTree.this, txId, dataTree.takeSnapshot()
537 return ensureTransactionChain(txId.getHistoryId()).newReadWriteTransaction(txId);
541 public void notifyListeners(final DataTreeCandidate candidate) {
542 treeChangeListenerPublisher.publishChanges(candidate);
543 dataChangeListenerPublisher.publishChanges(candidate);
547 * Immediately purge all state relevant to leader. This includes all transaction chains and any scheduled
548 * replication callbacks.
550 void purgeLeaderState() {
551 for (ShardDataTreeTransactionChain chain : transactionChains.values()) {
555 transactionChains.clear();
556 replicationCallbacks.clear();
560 * Close a single transaction chain.
562 * @param id History identifier
563 * @param callback Callback to invoke upon completion, may be null
565 void closeTransactionChain(final LocalHistoryIdentifier id, @Nullable final Runnable callback) {
566 final ShardDataTreeTransactionChain chain = transactionChains.get(id);
568 LOG.debug("{}: Closing non-existent transaction chain {}", logContext, id);
569 if (callback != null) {
576 replicatePayload(id, CloseLocalHistoryPayload.create(id), callback);
580 * Purge a single transaction chain.
582 * @param id History identifier
583 * @param callback Callback to invoke upon completion, may be null
585 void purgeTransactionChain(final LocalHistoryIdentifier id, @Nullable final Runnable callback) {
586 final ShardDataTreeTransactionChain chain = transactionChains.remove(id);
588 LOG.debug("{}: Purging non-existent transaction chain {}", logContext, id);
589 if (callback != null) {
595 replicatePayload(id, PurgeLocalHistoryPayload.create(id), callback);
598 void registerDataChangeListener(final YangInstanceIdentifier path,
599 final AsyncDataChangeListener<YangInstanceIdentifier, NormalizedNode<?, ?>> listener,
600 final DataChangeScope scope, final Optional<DataTreeCandidate> initialState,
601 final Consumer<ListenerRegistration<AsyncDataChangeListener<YangInstanceIdentifier, NormalizedNode<?, ?>>>>
603 dataChangeListenerPublisher.registerDataChangeListener(path, listener, scope, initialState, onRegistration);
606 Optional<DataTreeCandidate> readCurrentData() {
607 final Optional<NormalizedNode<?, ?>> currentState =
608 dataTree.takeSnapshot().readNode(YangInstanceIdentifier.EMPTY);
609 return currentState.isPresent() ? Optional.of(DataTreeCandidates.fromNormalizedNode(
610 YangInstanceIdentifier.EMPTY, currentState.get())) : Optional.<DataTreeCandidate>absent();
613 public void registerTreeChangeListener(final YangInstanceIdentifier path, final DOMDataTreeChangeListener listener,
614 final Optional<DataTreeCandidate> initialState,
615 final Consumer<ListenerRegistration<DOMDataTreeChangeListener>> onRegistration) {
616 treeChangeListenerPublisher.registerTreeChangeListener(path, listener, initialState, onRegistration);
620 return pendingTransactions.size() + pendingCommits.size() + pendingFinishCommits.size();
624 void abortTransaction(final AbstractShardDataTreeTransaction<?> transaction, final Runnable callback) {
625 final TransactionIdentifier id = transaction.getIdentifier();
626 LOG.debug("{}: aborting transaction {}", logContext, id);
627 replicatePayload(id, AbortTransactionPayload.create(id), callback);
631 void abortFromTransactionActor(final AbstractShardDataTreeTransaction<?> transaction) {
632 // No-op for free-standing transactions
637 ShardDataTreeCohort finishTransaction(final ReadWriteShardDataTreeTransaction transaction) {
638 final DataTreeModification snapshot = transaction.getSnapshot();
641 return createReadyCohort(transaction.getIdentifier(), snapshot);
644 void purgeTransaction(final TransactionIdentifier id, final Runnable callback) {
645 LOG.debug("{}: purging transaction {}", logContext, id);
646 replicatePayload(id, PurgeTransactionPayload.create(id), callback);
649 public Optional<NormalizedNode<?, ?>> readNode(final YangInstanceIdentifier path) {
650 return dataTree.takeSnapshot().readNode(path);
653 DataTreeSnapshot takeSnapshot() {
654 return dataTree.takeSnapshot();
658 public DataTreeModification newModification() {
659 return dataTree.takeSnapshot().newModification();
663 * Commits a modification.
665 * @deprecated This method violates DataTree containment and will be removed.
669 public DataTreeCandidate commit(final DataTreeModification modification) throws DataValidationFailedException {
670 // Direct modification commit is a utility, which cannot be used while we have transactions in-flight
671 Preconditions.checkState(tip == dataTree, "Cannot modify data tree while transacgitons are pending");
673 modification.ready();
674 dataTree.validate(modification);
675 DataTreeCandidate candidate = dataTree.prepare(modification);
676 dataTree.commit(candidate);
680 public Collection<ShardDataTreeCohort> getAndClearPendingTransactions() {
681 Collection<ShardDataTreeCohort> ret = new ArrayList<>(getQueueSize());
683 for (CommitEntry entry: pendingFinishCommits) {
684 ret.add(entry.cohort);
687 for (CommitEntry entry: pendingCommits) {
688 ret.add(entry.cohort);
691 for (CommitEntry entry: pendingTransactions) {
692 ret.add(entry.cohort);
695 pendingFinishCommits.clear();
696 pendingCommits.clear();
697 pendingTransactions.clear();
702 @SuppressWarnings("checkstyle:IllegalCatch")
703 private void processNextPendingTransaction() {
704 processNextPending(pendingTransactions, State.CAN_COMMIT_PENDING, entry -> {
705 final SimpleShardDataTreeCohort cohort = entry.cohort;
706 final DataTreeModification modification = cohort.getDataTreeModification();
708 LOG.debug("{}: Validating transaction {}", logContext, cohort.getIdentifier());
711 cohort.throwCanCommitFailure();
713 tip.validate(modification);
714 LOG.debug("{}: Transaction {} validated", logContext, cohort.getIdentifier());
715 cohort.successfulCanCommit();
716 entry.lastAccess = ticker().read();
718 } catch (ConflictingModificationAppliedException e) {
719 LOG.warn("{}: Store Tx {}: Conflicting modification for path {}.", logContext, cohort.getIdentifier(),
721 cause = new OptimisticLockFailedException("Optimistic lock failed.", e);
722 } catch (DataValidationFailedException e) {
723 LOG.warn("{}: Store Tx {}: Data validation failed for path {}.", logContext, cohort.getIdentifier(),
726 // For debugging purposes, allow dumping of the modification. Coupled with the above
727 // precondition log, it should allow us to understand what went on.
728 LOG.debug("{}: Store Tx {}: modifications: {} tree: {}", cohort.getIdentifier(), modification,
730 cause = new TransactionCommitFailedException("Data did not pass validation.", e);
731 } catch (Exception e) {
732 LOG.warn("{}: Unexpected failure in validation phase", logContext, e);
736 // Failure path: propagate the failure, remove the transaction from the queue and loop to the next one
737 pendingTransactions.poll().cohort.failedCanCommit(cause);
741 private void processNextPending() {
742 processNextPendingCommit();
743 processNextPendingTransaction();
746 private void processNextPending(final Queue<CommitEntry> queue, final State allowedState,
747 final Consumer<CommitEntry> processor) {
748 while (!queue.isEmpty()) {
749 final CommitEntry entry = queue.peek();
750 final SimpleShardDataTreeCohort cohort = entry.cohort;
752 if (cohort.isFailed()) {
753 LOG.debug("{}: Removing failed transaction {}", logContext, cohort.getIdentifier());
758 if (cohort.getState() == allowedState) {
759 processor.accept(entry);
765 maybeRunOperationOnPendingTransactionsComplete();
768 private void processNextPendingCommit() {
769 processNextPending(pendingCommits, State.COMMIT_PENDING,
770 entry -> startCommit(entry.cohort, entry.cohort.getCandidate()));
773 private boolean peekNextPendingCommit() {
774 final CommitEntry first = pendingCommits.peek();
775 return first != null && first.cohort.getState() == State.COMMIT_PENDING;
778 void startCanCommit(final SimpleShardDataTreeCohort cohort) {
779 final SimpleShardDataTreeCohort current = pendingTransactions.peek().cohort;
780 if (!cohort.equals(current)) {
781 LOG.debug("{}: Transaction {} scheduled for canCommit step", logContext, cohort.getIdentifier());
785 processNextPendingTransaction();
788 private void failPreCommit(final Exception cause) {
789 shard.getShardMBean().incrementFailedTransactionsCount();
790 pendingTransactions.poll().cohort.failedPreCommit(cause);
791 processNextPendingTransaction();
794 @SuppressWarnings("checkstyle:IllegalCatch")
795 void startPreCommit(final SimpleShardDataTreeCohort cohort) {
796 final CommitEntry entry = pendingTransactions.peek();
797 Preconditions.checkState(entry != null, "Attempted to pre-commit of %s when no transactions pending", cohort);
799 final SimpleShardDataTreeCohort current = entry.cohort;
800 Verify.verify(cohort.equals(current), "Attempted to pre-commit %s while %s is pending", cohort, current);
802 LOG.debug("{}: Preparing transaction {}", logContext, current.getIdentifier());
804 final DataTreeCandidateTip candidate;
806 candidate = tip.prepare(cohort.getDataTreeModification());
807 cohort.userPreCommit(candidate);
808 } catch (ExecutionException | TimeoutException | RuntimeException e) {
813 // Set the tip of the data tree.
814 tip = Verify.verifyNotNull(candidate);
816 entry.lastAccess = ticker().read();
818 pendingTransactions.remove();
819 pendingCommits.add(entry);
821 LOG.debug("{}: Transaction {} prepared", logContext, current.getIdentifier());
823 cohort.successfulPreCommit(candidate);
825 processNextPendingTransaction();
828 private void failCommit(final Exception cause) {
829 shard.getShardMBean().incrementFailedTransactionsCount();
830 pendingFinishCommits.poll().cohort.failedCommit(cause);
831 processNextPending();
834 @SuppressWarnings("checkstyle:IllegalCatch")
835 private void finishCommit(final SimpleShardDataTreeCohort cohort) {
836 final TransactionIdentifier txId = cohort.getIdentifier();
837 final DataTreeCandidate candidate = cohort.getCandidate();
839 LOG.debug("{}: Resuming commit of transaction {}", logContext, txId);
841 if (tip == candidate) {
842 // All pending candidates have been committed, reset the tip to the data tree.
847 dataTree.commit(candidate);
848 } catch (Exception e) {
849 LOG.error("{}: Failed to commit transaction {}", logContext, txId, e);
854 shard.getShardMBean().incrementCommittedTransactionCount();
855 shard.getShardMBean().setLastCommittedTransactionTime(System.currentTimeMillis());
857 // FIXME: propagate journal index
858 pendingFinishCommits.poll().cohort.successfulCommit(UnsignedLong.ZERO);
860 LOG.trace("{}: Transaction {} committed, proceeding to notify", logContext, txId);
861 notifyListeners(candidate);
863 processNextPending();
866 void startCommit(final SimpleShardDataTreeCohort cohort, final DataTreeCandidate candidate) {
867 final CommitEntry entry = pendingCommits.peek();
868 Preconditions.checkState(entry != null, "Attempted to start commit of %s when no transactions pending", cohort);
870 final SimpleShardDataTreeCohort current = entry.cohort;
871 if (!cohort.equals(current)) {
872 LOG.debug("{}: Transaction {} scheduled for commit step", logContext, cohort.getIdentifier());
876 LOG.debug("{}: Starting commit for transaction {}", logContext, current.getIdentifier());
878 final TransactionIdentifier txId = cohort.getIdentifier();
879 final Payload payload;
881 payload = CommitTransactionPayload.create(txId, candidate);
882 } catch (IOException e) {
883 LOG.error("{}: Failed to encode transaction {} candidate {}", logContext, txId, candidate, e);
884 pendingCommits.poll().cohort.failedCommit(e);
885 processNextPending();
889 // We process next transactions pending canCommit before we call persistPayload to possibly progress subsequent
890 // transactions to the COMMIT_PENDING state so the payloads can be batched for replication. This is done for
891 // single-shard transactions that immediately transition from canCommit to preCommit to commit. Note that
892 // if the next pending transaction is progressed to COMMIT_PENDING and this method (startCommit) is called,
893 // the next transaction will not attempt to replicate b/c the current transaction is still at the head of the
894 // pendingCommits queue.
895 processNextPendingTransaction();
897 // After processing next pending transactions, we can now remove the current transaction from pendingCommits.
898 // Note this must be done before the call to peekNextPendingCommit below so we check the next transaction
899 // in order to properly determine the batchHint flag for the call to persistPayload.
900 pendingCommits.remove();
901 pendingFinishCommits.add(entry);
903 // See if the next transaction is pending commit (ie in the COMMIT_PENDING state) so it can be batched with
904 // this transaction for replication.
905 boolean replicationBatchHint = peekNextPendingCommit();
907 // Once completed, we will continue via payloadReplicationComplete
908 shard.persistPayload(txId, payload, replicationBatchHint);
910 entry.lastAccess = shard.ticker().read();
912 LOG.debug("{}: Transaction {} submitted to persistence", logContext, txId);
914 // Process the next transaction pending commit, if any. If there is one it will be batched with this
915 // transaction for replication.
916 processNextPendingCommit();
919 Collection<ActorRef> getCohortActors() {
920 return cohortRegistry.getCohortActors();
923 void processCohortRegistryCommand(final ActorRef sender, final CohortRegistryCommand message) {
924 cohortRegistry.process(sender, message);
928 ShardDataTreeCohort createFailedCohort(final TransactionIdentifier txId, final DataTreeModification mod,
929 final Exception failure) {
930 SimpleShardDataTreeCohort cohort = new SimpleShardDataTreeCohort.DeadOnArrival(this, mod, txId, failure);
931 pendingTransactions.add(new CommitEntry(cohort, ticker().read()));
936 ShardDataTreeCohort createReadyCohort(final TransactionIdentifier txId,
937 final DataTreeModification mod) {
938 SimpleShardDataTreeCohort cohort = new SimpleShardDataTreeCohort.Normal(this, mod, txId,
939 cohortRegistry.createCohort(schemaContext, txId, COMMIT_STEP_TIMEOUT));
940 pendingTransactions.add(new CommitEntry(cohort, ticker().read()));
944 // Exposed for ShardCommitCoordinator so it does not have deal with local histories (it does not care), this mimics
945 // the newReadWriteTransaction()
946 ShardDataTreeCohort newReadyCohort(final TransactionIdentifier txId, final DataTreeModification mod) {
947 if (txId.getHistoryId().getHistoryId() == 0) {
948 return createReadyCohort(txId, mod);
951 return ensureTransactionChain(txId.getHistoryId()).createReadyCohort(txId, mod);
954 @SuppressFBWarnings(value = "DB_DUPLICATE_SWITCH_CLAUSES", justification = "See inline comments below.")
955 void checkForExpiredTransactions(final long transactionCommitTimeoutMillis) {
956 final long timeout = TimeUnit.MILLISECONDS.toNanos(transactionCommitTimeoutMillis);
957 final long now = ticker().read();
959 final Queue<CommitEntry> currentQueue = !pendingFinishCommits.isEmpty() ? pendingFinishCommits :
960 !pendingCommits.isEmpty() ? pendingCommits : pendingTransactions;
961 final CommitEntry currentTx = currentQueue.peek();
962 if (currentTx != null && currentTx.lastAccess + timeout < now) {
963 LOG.warn("{}: Current transaction {} has timed out after {} ms in state {}", logContext,
964 currentTx.cohort.getIdentifier(), transactionCommitTimeoutMillis, currentTx.cohort.getState());
965 boolean processNext = true;
966 switch (currentTx.cohort.getState()) {
967 case CAN_COMMIT_PENDING:
968 currentQueue.remove().cohort.failedCanCommit(new TimeoutException());
970 case CAN_COMMIT_COMPLETE:
971 // The suppression of the FindBugs "DB_DUPLICATE_SWITCH_CLAUSES" warning pertains to this clause
972 // whose code is duplicated with PRE_COMMIT_COMPLETE. The clauses aren't combined in case the code
973 // in PRE_COMMIT_COMPLETE is changed.
974 currentQueue.remove().cohort.reportFailure(new TimeoutException());
976 case PRE_COMMIT_PENDING:
977 currentQueue.remove().cohort.failedPreCommit(new TimeoutException());
979 case PRE_COMMIT_COMPLETE:
980 // FIXME: this is a legacy behavior problem. Three-phase commit protocol specifies that after we
981 // are ready we should commit the transaction, not abort it. Our current software stack does
982 // not allow us to do that consistently, because we persist at the time of commit, hence
983 // we can end up in a state where we have pre-committed a transaction, then a leader failover
984 // occurred ... the new leader does not see the pre-committed transaction and does not have
985 // a running timer. To fix this we really need two persistence events.
987 // The first one, done at pre-commit time will hold the transaction payload. When consensus
988 // is reached, we exit the pre-commit phase and start the pre-commit timer. Followers do not
989 // apply the state in this event.
991 // The second one, done at commit (or abort) time holds only the transaction identifier and
992 // signals to followers that the state should (or should not) be applied.
994 // In order to make the pre-commit timer working across failovers, though, we need
995 // a per-shard cluster-wide monotonic time, so a follower becoming the leader can accurately
996 // restart the timer.
997 currentQueue.remove().cohort.reportFailure(new TimeoutException());
1000 LOG.warn("{}: Transaction {} is still committing, cannot abort", logContext,
1001 currentTx.cohort.getIdentifier());
1002 currentTx.lastAccess = now;
1003 processNext = false;
1010 currentQueue.remove();
1014 processNextPending();
1019 boolean startAbort(final SimpleShardDataTreeCohort cohort) {
1020 final Iterator<CommitEntry> it = Iterables.concat(pendingFinishCommits, pendingCommits,
1021 pendingTransactions).iterator();
1022 if (!it.hasNext()) {
1023 LOG.debug("{}: no open transaction while attempting to abort {}", logContext, cohort.getIdentifier());
1027 // First entry is special, as it may already be committing
1028 final CommitEntry first = it.next();
1029 if (cohort.equals(first.cohort)) {
1030 if (cohort.getState() != State.COMMIT_PENDING) {
1031 LOG.debug("{}: aborting head of queue {} in state {}", logContext, cohort.getIdentifier(),
1032 cohort.getIdentifier());
1035 if (cohort.getCandidate() != null) {
1036 rebaseTransactions(it, dataTree);
1039 processNextPending();
1043 LOG.warn("{}: transaction {} is committing, skipping abort", logContext, cohort.getIdentifier());
1047 TipProducingDataTreeTip newTip = MoreObjects.firstNonNull(first.cohort.getCandidate(), dataTree);
1048 while (it.hasNext()) {
1049 final CommitEntry e = it.next();
1050 if (cohort.equals(e.cohort)) {
1051 LOG.debug("{}: aborting queued transaction {}", logContext, cohort.getIdentifier());
1054 if (cohort.getCandidate() != null) {
1055 rebaseTransactions(it, newTip);
1060 newTip = MoreObjects.firstNonNull(e.cohort.getCandidate(), newTip);
1064 LOG.debug("{}: aborted transaction {} not found in the queue", logContext, cohort.getIdentifier());
1068 @SuppressWarnings("checkstyle:IllegalCatch")
1069 private void rebaseTransactions(final Iterator<CommitEntry> iter, @Nonnull final TipProducingDataTreeTip newTip) {
1070 tip = Preconditions.checkNotNull(newTip);
1071 while (iter.hasNext()) {
1072 final SimpleShardDataTreeCohort cohort = iter.next().cohort;
1073 if (cohort.getState() == State.CAN_COMMIT_COMPLETE) {
1074 LOG.debug("{}: Revalidating queued transaction {}", logContext, cohort.getIdentifier());
1077 tip.validate(cohort.getDataTreeModification());
1078 } catch (DataValidationFailedException | RuntimeException e) {
1079 LOG.debug("{}: Failed to revalidate queued transaction {}", logContext, cohort.getIdentifier(), e);
1080 cohort.reportFailure(e);
1082 } else if (cohort.getState() == State.PRE_COMMIT_COMPLETE) {
1083 LOG.debug("{}: Repreparing queued transaction {}", logContext, cohort.getIdentifier());
1086 tip.validate(cohort.getDataTreeModification());
1087 DataTreeCandidateTip candidate = tip.prepare(cohort.getDataTreeModification());
1088 cohort.userPreCommit(candidate);
1090 cohort.setNewCandidate(candidate);
1092 } catch (ExecutionException | TimeoutException | RuntimeException | DataValidationFailedException e) {
1093 LOG.debug("{}: Failed to reprepare queued transaction {}", logContext, cohort.getIdentifier(), e);
1094 cohort.reportFailure(e);
1100 void setRunOnPendingTransactionsComplete(final Runnable operation) {
1101 runOnPendingTransactionsComplete = operation;
1102 maybeRunOperationOnPendingTransactionsComplete();
1105 private void maybeRunOperationOnPendingTransactionsComplete() {
1106 if (runOnPendingTransactionsComplete != null && !anyPendingTransactions()) {
1107 LOG.debug("{}: Pending transactions complete - running operation {}", logContext,
1108 runOnPendingTransactionsComplete);
1110 runOnPendingTransactionsComplete.run();
1111 runOnPendingTransactionsComplete = null;