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.jmx.mbeans.shard.ShardStats;
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.yangtools.concepts.Identifier;
67 import org.opendaylight.yangtools.concepts.ListenerRegistration;
68 import org.opendaylight.yangtools.yang.data.api.YangInstanceIdentifier;
69 import org.opendaylight.yangtools.yang.data.api.schema.NormalizedNode;
70 import org.opendaylight.yangtools.yang.data.api.schema.tree.ConflictingModificationAppliedException;
71 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeCandidate;
72 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeCandidateTip;
73 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeCandidates;
74 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeModification;
75 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeSnapshot;
76 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeTip;
77 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataValidationFailedException;
78 import org.opendaylight.yangtools.yang.data.api.schema.tree.TipProducingDataTree;
79 import org.opendaylight.yangtools.yang.data.api.schema.tree.TipProducingDataTreeTip;
80 import org.opendaylight.yangtools.yang.data.api.schema.tree.TreeType;
81 import org.opendaylight.yangtools.yang.data.impl.schema.tree.InMemoryDataTreeFactory;
82 import org.opendaylight.yangtools.yang.model.api.SchemaContext;
83 import org.slf4j.Logger;
84 import org.slf4j.LoggerFactory;
85 import scala.concurrent.duration.Duration;
88 * Internal shard state, similar to a DOMStore, but optimized for use in the actor system,
89 * e.g. it does not expose public interfaces and assumes it is only ever called from a
93 * This class is not part of the API contract and is subject to change at any time.
96 public class ShardDataTree extends ShardDataTreeTransactionParent {
97 private static final class CommitEntry {
98 final SimpleShardDataTreeCohort cohort;
101 CommitEntry(final SimpleShardDataTreeCohort cohort, final long now) {
102 this.cohort = Preconditions.checkNotNull(cohort);
107 private static final Timeout COMMIT_STEP_TIMEOUT = new Timeout(Duration.create(5, TimeUnit.SECONDS));
108 private static final Logger LOG = LoggerFactory.getLogger(ShardDataTree.class);
110 private final Map<LocalHistoryIdentifier, ShardDataTreeTransactionChain> transactionChains = new HashMap<>();
112 private final DataTreeCohortActorRegistry cohortRegistry = new DataTreeCohortActorRegistry();
113 private final Queue<CommitEntry> pendingTransactions = new ArrayDeque<>();
114 private final Queue<CommitEntry> pendingCommits = new ArrayDeque<>();
115 private final Queue<CommitEntry> pendingFinishCommits = new ArrayDeque<>();
118 * Callbacks that need to be invoked once a payload is replicated.
120 private final Map<Payload, Runnable> replicationCallbacks = new HashMap<>();
122 private final ShardDataTreeChangeListenerPublisher treeChangeListenerPublisher;
123 private final ShardDataChangeListenerPublisher dataChangeListenerPublisher;
124 private final Collection<ShardDataTreeMetadata<?>> metadata;
125 private final TipProducingDataTree dataTree;
126 private final String logContext;
127 private final Shard shard;
128 private Runnable runOnPendingTransactionsComplete;
131 * Optimistic {@link DataTreeCandidate} preparation. Since our DataTree implementation is a
132 * {@link TipProducingDataTree}, each {@link DataTreeCandidate} is also a {@link DataTreeTip}, e.g. another
133 * candidate can be prepared on top of it. They still need to be committed in sequence. Here we track the current
134 * tip of the data tree, which is the last DataTreeCandidate we have in flight, or the DataTree itself.
136 private TipProducingDataTreeTip tip;
138 private SchemaContext schemaContext;
140 ShardDataTree(final Shard shard, final SchemaContext schemaContext, final TipProducingDataTree dataTree,
141 final ShardDataTreeChangeListenerPublisher treeChangeListenerPublisher,
142 final ShardDataChangeListenerPublisher dataChangeListenerPublisher, final String logContext,
143 final ShardDataTreeMetadata<?>... metadata) {
144 this.dataTree = Preconditions.checkNotNull(dataTree);
145 updateSchemaContext(schemaContext);
147 this.shard = Preconditions.checkNotNull(shard);
148 this.treeChangeListenerPublisher = Preconditions.checkNotNull(treeChangeListenerPublisher);
149 this.dataChangeListenerPublisher = Preconditions.checkNotNull(dataChangeListenerPublisher);
150 this.logContext = Preconditions.checkNotNull(logContext);
151 this.metadata = ImmutableList.copyOf(metadata);
155 ShardDataTree(final Shard shard, final SchemaContext schemaContext, final TreeType treeType,
156 final YangInstanceIdentifier root,
157 final ShardDataTreeChangeListenerPublisher treeChangeListenerPublisher,
158 final ShardDataChangeListenerPublisher dataChangeListenerPublisher, final String logContext,
159 final ShardDataTreeMetadata<?>... metadata) {
160 this(shard, schemaContext, InMemoryDataTreeFactory.getInstance().create(treeType, root),
161 treeChangeListenerPublisher, dataChangeListenerPublisher, logContext, metadata);
165 public ShardDataTree(final Shard shard, final SchemaContext schemaContext, final TreeType treeType) {
166 this(shard, schemaContext, treeType, YangInstanceIdentifier.EMPTY,
167 new DefaultShardDataTreeChangeListenerPublisher(), 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 %s", 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());
338 } else if (payload instanceof DataTreeCandidatePayload) {
339 applyRecoveryCandidate(((DataTreeCandidatePayload) payload).getCandidate());
341 LOG.debug("{}: ignoring unhandled payload {}", logContext, payload);
345 private void applyReplicatedCandidate(final Identifier identifier, final DataTreeCandidate foreign)
346 throws DataValidationFailedException {
347 LOG.debug("{}: Applying foreign transaction {}", logContext, identifier);
349 final DataTreeModification mod = dataTree.takeSnapshot().newModification();
350 DataTreeCandidates.applyToModification(mod, foreign);
353 LOG.trace("{}: Applying foreign modification {}", logContext, mod);
354 dataTree.validate(mod);
355 final DataTreeCandidate candidate = dataTree.prepare(mod);
356 dataTree.commit(candidate);
358 notifyListeners(candidate);
362 * Apply a payload coming from the leader, which could actually be us. This method assumes the leader and follower
363 * SchemaContexts match and does not perform any pruning.
365 * @param identifier Payload identifier as returned from RaftActor
366 * @param payload Payload
367 * @throws IOException when the snapshot fails to deserialize
368 * @throws DataValidationFailedException when the snapshot fails to apply
370 void applyReplicatedPayload(final Identifier identifier, final Payload payload) throws IOException,
371 DataValidationFailedException {
373 * 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
374 * if we are the leader and it has originated with us.
376 * The identifier will only ever be non-null when we were the leader which achieved consensus. Unfortunately,
377 * though, this may not be the case anymore, as we are being called some time afterwards and we may not be
378 * acting in that capacity anymore.
380 * In any case, we know that this is an entry coming from replication, hence we can be sure we will not observe
381 * pre-Boron state -- which limits the number of options here.
383 if (payload instanceof CommitTransactionPayload) {
384 if (identifier == null) {
385 final Entry<TransactionIdentifier, DataTreeCandidate> e =
386 ((CommitTransactionPayload) payload).getCandidate();
387 applyReplicatedCandidate(e.getKey(), e.getValue());
388 allMetadataCommittedTransaction(e.getKey());
390 Verify.verify(identifier instanceof TransactionIdentifier);
391 payloadReplicationComplete((TransactionIdentifier) identifier);
393 } else if (payload instanceof AbortTransactionPayload) {
394 if (identifier != null) {
395 payloadReplicationComplete((AbortTransactionPayload) payload);
397 allMetadataAbortedTransaction(((AbortTransactionPayload) payload).getIdentifier());
399 } else if (payload instanceof PurgeTransactionPayload) {
400 if (identifier != null) {
401 payloadReplicationComplete((PurgeTransactionPayload) payload);
403 allMetadataPurgedTransaction(((PurgeTransactionPayload) payload).getIdentifier());
405 } else if (payload instanceof CloseLocalHistoryPayload) {
406 if (identifier != null) {
407 payloadReplicationComplete((CloseLocalHistoryPayload) payload);
409 allMetadataClosedLocalHistory(((CloseLocalHistoryPayload) payload).getIdentifier());
411 } else if (payload instanceof CreateLocalHistoryPayload) {
412 if (identifier != null) {
413 payloadReplicationComplete((CreateLocalHistoryPayload)payload);
415 allMetadataCreatedLocalHistory(((CreateLocalHistoryPayload) payload).getIdentifier());
417 } else if (payload instanceof PurgeLocalHistoryPayload) {
418 if (identifier != null) {
419 payloadReplicationComplete((PurgeLocalHistoryPayload)payload);
421 allMetadataPurgedLocalHistory(((PurgeLocalHistoryPayload) payload).getIdentifier());
424 LOG.warn("{}: ignoring unhandled identifier {} payload {}", logContext, identifier, payload);
428 private void replicatePayload(final Identifier id, final Payload payload, @Nullable final Runnable callback) {
429 if (callback != null) {
430 replicationCallbacks.put(payload, callback);
432 shard.persistPayload(id, payload, true);
435 private void payloadReplicationComplete(final AbstractIdentifiablePayload<?> payload) {
436 final Runnable callback = replicationCallbacks.remove(payload);
437 if (callback != null) {
438 LOG.debug("{}: replication of {} completed, invoking {}", logContext, payload.getIdentifier(), callback);
441 LOG.debug("{}: replication of {} has no callback", logContext, payload.getIdentifier());
445 private void payloadReplicationComplete(final TransactionIdentifier txId) {
446 final CommitEntry current = pendingFinishCommits.peek();
447 if (current == null) {
448 LOG.warn("{}: No outstanding transactions, ignoring consensus on transaction {}", logContext, txId);
452 if (!current.cohort.getIdentifier().equals(txId)) {
453 LOG.debug("{}: Head of pendingFinishCommits queue is {}, ignoring consensus on transaction {}", logContext,
454 current.cohort.getIdentifier(), txId);
458 finishCommit(current.cohort);
461 private void allMetadataAbortedTransaction(final TransactionIdentifier txId) {
462 for (ShardDataTreeMetadata<?> m : metadata) {
463 m.onTransactionAborted(txId);
467 private void allMetadataCommittedTransaction(final TransactionIdentifier txId) {
468 for (ShardDataTreeMetadata<?> m : metadata) {
469 m.onTransactionCommitted(txId);
473 private void allMetadataPurgedTransaction(final TransactionIdentifier txId) {
474 for (ShardDataTreeMetadata<?> m : metadata) {
475 m.onTransactionPurged(txId);
479 private void allMetadataCreatedLocalHistory(final LocalHistoryIdentifier historyId) {
480 for (ShardDataTreeMetadata<?> m : metadata) {
481 m.onHistoryCreated(historyId);
485 private void allMetadataClosedLocalHistory(final LocalHistoryIdentifier historyId) {
486 for (ShardDataTreeMetadata<?> m : metadata) {
487 m.onHistoryClosed(historyId);
491 private void allMetadataPurgedLocalHistory(final LocalHistoryIdentifier historyId) {
492 for (ShardDataTreeMetadata<?> m : metadata) {
493 m.onHistoryPurged(historyId);
498 * Create a transaction chain for specified history. Unlike {@link #ensureTransactionChain(LocalHistoryIdentifier)},
499 * this method is used for re-establishing state when we are taking over
501 * @param historyId Local history identifier
502 * @param closed True if the chain should be created in closed state (i.e. pending purge)
503 * @return Transaction chain handle
505 ShardDataTreeTransactionChain recreateTransactionChain(final LocalHistoryIdentifier historyId,
506 final boolean closed) {
507 final ShardDataTreeTransactionChain ret = new ShardDataTreeTransactionChain(historyId, this);
508 final ShardDataTreeTransactionChain existing = transactionChains.putIfAbsent(historyId, ret);
509 Preconditions.checkState(existing == null, "Attempted to recreate chain %s, but %s already exists", historyId,
514 ShardDataTreeTransactionChain ensureTransactionChain(final LocalHistoryIdentifier historyId) {
515 ShardDataTreeTransactionChain chain = transactionChains.get(historyId);
517 chain = new ShardDataTreeTransactionChain(historyId, this);
518 transactionChains.put(historyId, chain);
519 shard.persistPayload(historyId, CreateLocalHistoryPayload.create(historyId), true);
525 ReadOnlyShardDataTreeTransaction newReadOnlyTransaction(final TransactionIdentifier txId) {
526 if (txId.getHistoryId().getHistoryId() == 0) {
527 return new ReadOnlyShardDataTreeTransaction(this, txId, dataTree.takeSnapshot());
530 return ensureTransactionChain(txId.getHistoryId()).newReadOnlyTransaction(txId);
533 ReadWriteShardDataTreeTransaction newReadWriteTransaction(final TransactionIdentifier txId) {
534 if (txId.getHistoryId().getHistoryId() == 0) {
535 return new ReadWriteShardDataTreeTransaction(ShardDataTree.this, txId, dataTree.takeSnapshot()
539 return ensureTransactionChain(txId.getHistoryId()).newReadWriteTransaction(txId);
543 public void notifyListeners(final DataTreeCandidate candidate) {
544 treeChangeListenerPublisher.publishChanges(candidate, logContext);
545 dataChangeListenerPublisher.publishChanges(candidate, logContext);
549 * Immediately purge all state relevant to leader. This includes all transaction chains and any scheduled
550 * replication callbacks.
552 void purgeLeaderState() {
553 for (ShardDataTreeTransactionChain chain : transactionChains.values()) {
557 transactionChains.clear();
558 replicationCallbacks.clear();
562 * Close a single transaction chain.
564 * @param id History identifier
565 * @param callback Callback to invoke upon completion, may be null
567 void closeTransactionChain(final LocalHistoryIdentifier id, @Nullable final Runnable callback) {
568 final ShardDataTreeTransactionChain chain = transactionChains.get(id);
570 LOG.debug("{}: Closing non-existent transaction chain {}", logContext, id);
571 if (callback != null) {
578 replicatePayload(id, CloseLocalHistoryPayload.create(id), callback);
582 * Purge a single transaction chain.
584 * @param id History identifier
585 * @param callback Callback to invoke upon completion, may be null
587 void purgeTransactionChain(final LocalHistoryIdentifier id, @Nullable final Runnable callback) {
588 final ShardDataTreeTransactionChain chain = transactionChains.remove(id);
590 LOG.debug("{}: Purging non-existent transaction chain {}", logContext, id);
591 if (callback != null) {
597 replicatePayload(id, PurgeLocalHistoryPayload.create(id), callback);
600 void registerDataChangeListener(YangInstanceIdentifier path,
601 AsyncDataChangeListener<YangInstanceIdentifier, NormalizedNode<?, ?>> listener, DataChangeScope scope,
602 Optional<DataTreeCandidate> initialState,
603 Consumer<ListenerRegistration<AsyncDataChangeListener<YangInstanceIdentifier, NormalizedNode<?, ?>>>>
605 dataChangeListenerPublisher.registerDataChangeListener(path, listener, scope, initialState, onRegistration);
608 Optional<DataTreeCandidate> readCurrentData() {
609 final Optional<NormalizedNode<?, ?>> currentState =
610 dataTree.takeSnapshot().readNode(YangInstanceIdentifier.EMPTY);
611 return currentState.isPresent() ? Optional.of(DataTreeCandidates.fromNormalizedNode(
612 YangInstanceIdentifier.EMPTY, currentState.get())) : Optional.<DataTreeCandidate>absent();
615 public void registerTreeChangeListener(YangInstanceIdentifier path, DOMDataTreeChangeListener listener,
616 Optional<DataTreeCandidate> initialState,
617 Consumer<ListenerRegistration<DOMDataTreeChangeListener>> onRegistration) {
618 treeChangeListenerPublisher.registerTreeChangeListener(path, listener, initialState, onRegistration);
622 return pendingTransactions.size() + pendingCommits.size() + pendingFinishCommits.size();
626 void abortTransaction(final AbstractShardDataTreeTransaction<?> transaction, final Runnable callback) {
627 final TransactionIdentifier id = transaction.getIdentifier();
628 LOG.debug("{}: aborting transaction {}", logContext, id);
629 replicatePayload(id, AbortTransactionPayload.create(id), callback);
633 void abortFromTransactionActor(final AbstractShardDataTreeTransaction<?> transaction) {
634 // No-op for free-standing transactions
639 ShardDataTreeCohort finishTransaction(final ReadWriteShardDataTreeTransaction transaction) {
640 final DataTreeModification snapshot = transaction.getSnapshot();
643 return createReadyCohort(transaction.getIdentifier(), snapshot);
646 void purgeTransaction(final TransactionIdentifier id, final Runnable callback) {
647 LOG.debug("{}: purging transaction {}", logContext, id);
648 replicatePayload(id, PurgeTransactionPayload.create(id), callback);
651 public Optional<NormalizedNode<?, ?>> readNode(final YangInstanceIdentifier path) {
652 return dataTree.takeSnapshot().readNode(path);
655 DataTreeSnapshot takeSnapshot() {
656 return dataTree.takeSnapshot();
660 public DataTreeModification newModification() {
661 return dataTree.takeSnapshot().newModification();
665 * Commits a modification.
667 * @deprecated This method violates DataTree containment and will be removed.
671 public DataTreeCandidate commit(final DataTreeModification modification) throws DataValidationFailedException {
672 // Direct modification commit is a utility, which cannot be used while we have transactions in-flight
673 Preconditions.checkState(tip == dataTree, "Cannot modify data tree while transacgitons are pending");
675 modification.ready();
676 dataTree.validate(modification);
677 DataTreeCandidate candidate = dataTree.prepare(modification);
678 dataTree.commit(candidate);
682 public Collection<ShardDataTreeCohort> getAndClearPendingTransactions() {
683 Collection<ShardDataTreeCohort> ret = new ArrayList<>(getQueueSize());
685 for (CommitEntry entry: pendingFinishCommits) {
686 ret.add(entry.cohort);
689 for (CommitEntry entry: pendingCommits) {
690 ret.add(entry.cohort);
693 for (CommitEntry entry: pendingTransactions) {
694 ret.add(entry.cohort);
697 pendingFinishCommits.clear();
698 pendingCommits.clear();
699 pendingTransactions.clear();
704 @SuppressWarnings("checkstyle:IllegalCatch")
705 private void processNextPendingTransaction() {
706 processNextPending(pendingTransactions, State.CAN_COMMIT_PENDING, entry -> {
707 final SimpleShardDataTreeCohort cohort = entry.cohort;
708 final DataTreeModification modification = cohort.getDataTreeModification();
710 LOG.debug("{}: Validating transaction {}", logContext, cohort.getIdentifier());
713 cohort.throwCanCommitFailure();
715 tip.validate(modification);
716 LOG.debug("{}: Transaction {} validated", logContext, cohort.getIdentifier());
717 cohort.successfulCanCommit();
718 entry.lastAccess = ticker().read();
720 } catch (ConflictingModificationAppliedException e) {
721 LOG.warn("{}: Store Tx {}: Conflicting modification for path {}.", logContext, cohort.getIdentifier(),
723 cause = new OptimisticLockFailedException("Optimistic lock failed.", e);
724 } catch (DataValidationFailedException e) {
725 LOG.warn("{}: Store Tx {}: Data validation failed for path {}.", logContext, cohort.getIdentifier(),
728 // For debugging purposes, allow dumping of the modification. Coupled with the above
729 // precondition log, it should allow us to understand what went on.
730 LOG.debug("{}: Store Tx {}: modifications: {} tree: {}", cohort.getIdentifier(), modification,
732 cause = new TransactionCommitFailedException("Data did not pass validation.", e);
733 } catch (Exception e) {
734 LOG.warn("{}: Unexpected failure in validation phase", logContext, e);
738 // Failure path: propagate the failure, remove the transaction from the queue and loop to the next one
739 pendingTransactions.poll().cohort.failedCanCommit(cause);
743 private void processNextPending() {
744 processNextPendingCommit();
745 processNextPendingTransaction();
748 private void processNextPending(final Queue<CommitEntry> queue, final State allowedState,
749 final Consumer<CommitEntry> processor) {
750 while (!queue.isEmpty()) {
751 final CommitEntry entry = queue.peek();
752 final SimpleShardDataTreeCohort cohort = entry.cohort;
754 if (cohort.isFailed()) {
755 LOG.debug("{}: Removing failed transaction {}", logContext, cohort.getIdentifier());
760 if (cohort.getState() == allowedState) {
761 processor.accept(entry);
767 maybeRunOperationOnPendingTransactionsComplete();
770 private void processNextPendingCommit() {
771 processNextPending(pendingCommits, State.COMMIT_PENDING,
772 entry -> startCommit(entry.cohort, entry.cohort.getCandidate()));
775 private boolean peekNextPendingCommit() {
776 final CommitEntry first = pendingCommits.peek();
777 return first != null && first.cohort.getState() == State.COMMIT_PENDING;
780 void startCanCommit(final SimpleShardDataTreeCohort cohort) {
781 final SimpleShardDataTreeCohort current = pendingTransactions.peek().cohort;
782 if (!cohort.equals(current)) {
783 LOG.debug("{}: Transaction {} scheduled for canCommit step", logContext, cohort.getIdentifier());
787 processNextPendingTransaction();
790 private void failPreCommit(final Exception cause) {
791 shard.getShardMBean().incrementFailedTransactionsCount();
792 pendingTransactions.poll().cohort.failedPreCommit(cause);
793 processNextPendingTransaction();
796 @SuppressWarnings("checkstyle:IllegalCatch")
797 void startPreCommit(final SimpleShardDataTreeCohort cohort) {
798 final CommitEntry entry = pendingTransactions.peek();
799 Preconditions.checkState(entry != null, "Attempted to pre-commit of %s when no transactions pending", cohort);
801 final SimpleShardDataTreeCohort current = entry.cohort;
802 Verify.verify(cohort.equals(current), "Attempted to pre-commit %s while %s is pending", cohort, current);
804 LOG.debug("{}: Preparing transaction {}", logContext, current.getIdentifier());
806 final DataTreeCandidateTip candidate;
808 candidate = tip.prepare(cohort.getDataTreeModification());
809 cohort.userPreCommit(candidate);
810 } catch (ExecutionException | TimeoutException | RuntimeException e) {
815 // Set the tip of the data tree.
816 tip = Verify.verifyNotNull(candidate);
818 entry.lastAccess = ticker().read();
820 pendingTransactions.remove();
821 pendingCommits.add(entry);
823 LOG.debug("{}: Transaction {} prepared", logContext, current.getIdentifier());
825 cohort.successfulPreCommit(candidate);
827 processNextPendingTransaction();
830 private void failCommit(final Exception cause) {
831 shard.getShardMBean().incrementFailedTransactionsCount();
832 pendingFinishCommits.poll().cohort.failedCommit(cause);
833 processNextPending();
836 @SuppressWarnings("checkstyle:IllegalCatch")
837 private void finishCommit(final SimpleShardDataTreeCohort cohort) {
838 final TransactionIdentifier txId = cohort.getIdentifier();
839 final DataTreeCandidate candidate = cohort.getCandidate();
841 LOG.debug("{}: Resuming commit of transaction {}", logContext, txId);
843 if (tip == candidate) {
844 // All pending candidates have been committed, reset the tip to the data tree.
849 dataTree.commit(candidate);
850 } catch (Exception e) {
851 LOG.error("{}: Failed to commit transaction {}", logContext, txId, e);
856 shard.getShardMBean().incrementCommittedTransactionCount();
857 shard.getShardMBean().setLastCommittedTransactionTime(System.currentTimeMillis());
859 // FIXME: propagate journal index
860 pendingFinishCommits.poll().cohort.successfulCommit(UnsignedLong.ZERO);
862 LOG.trace("{}: Transaction {} committed, proceeding to notify", logContext, txId);
863 notifyListeners(candidate);
865 processNextPending();
868 void startCommit(final SimpleShardDataTreeCohort cohort, final DataTreeCandidate candidate) {
869 final CommitEntry entry = pendingCommits.peek();
870 Preconditions.checkState(entry != null, "Attempted to start commit of %s when no transactions pending", cohort);
872 final SimpleShardDataTreeCohort current = entry.cohort;
873 if (!cohort.equals(current)) {
874 LOG.debug("{}: Transaction {} scheduled for commit step", logContext, cohort.getIdentifier());
878 LOG.debug("{}: Starting commit for transaction {}", logContext, current.getIdentifier());
880 final TransactionIdentifier txId = cohort.getIdentifier();
881 final Payload payload;
883 payload = CommitTransactionPayload.create(txId, candidate);
884 } catch (IOException e) {
885 LOG.error("{}: Failed to encode transaction {} candidate {}", logContext, txId, candidate, e);
886 pendingCommits.poll().cohort.failedCommit(e);
887 processNextPending();
891 // We process next transactions pending canCommit before we call persistPayload to possibly progress subsequent
892 // transactions to the COMMIT_PENDING state so the payloads can be batched for replication. This is done for
893 // single-shard transactions that immediately transition from canCommit to preCommit to commit. Note that
894 // if the next pending transaction is progressed to COMMIT_PENDING and this method (startCommit) is called,
895 // the next transaction will not attempt to replicate b/c the current transaction is still at the head of the
896 // pendingCommits queue.
897 processNextPendingTransaction();
899 // After processing next pending transactions, we can now remove the current transaction from pendingCommits.
900 // Note this must be done before the call to peekNextPendingCommit below so we check the next transaction
901 // in order to properly determine the batchHint flag for the call to persistPayload.
902 pendingCommits.remove();
903 pendingFinishCommits.add(entry);
905 // See if the next transaction is pending commit (ie in the COMMIT_PENDING state) so it can be batched with
906 // this transaction for replication.
907 boolean replicationBatchHint = peekNextPendingCommit();
909 // Once completed, we will continue via payloadReplicationComplete
910 shard.persistPayload(txId, payload, replicationBatchHint);
912 entry.lastAccess = shard.ticker().read();
914 LOG.debug("{}: Transaction {} submitted to persistence", logContext, txId);
916 // Process the next transaction pending commit, if any. If there is one it will be batched with this
917 // transaction for replication.
918 processNextPendingCommit();
921 Collection<ActorRef> getCohortActors() {
922 return cohortRegistry.getCohortActors();
925 void processCohortRegistryCommand(final ActorRef sender, final CohortRegistryCommand message) {
926 cohortRegistry.process(sender, message);
931 ShardDataTreeCohort createFailedCohort(final TransactionIdentifier txId, final DataTreeModification mod,
932 final Exception failure) {
933 SimpleShardDataTreeCohort cohort = new SimpleShardDataTreeCohort.DeadOnArrival(this, mod, txId, failure);
934 pendingTransactions.add(new CommitEntry(cohort, ticker().read()));
939 ShardDataTreeCohort createReadyCohort(final TransactionIdentifier txId, final DataTreeModification mod) {
940 SimpleShardDataTreeCohort cohort = new SimpleShardDataTreeCohort.Normal(this, mod, txId,
941 cohortRegistry.createCohort(schemaContext, txId, COMMIT_STEP_TIMEOUT));
942 pendingTransactions.add(new CommitEntry(cohort, ticker().read()));
946 @SuppressFBWarnings(value = "DB_DUPLICATE_SWITCH_CLAUSES", justification = "See inline comments below.")
947 void checkForExpiredTransactions(final long transactionCommitTimeoutMillis) {
948 final long timeout = TimeUnit.MILLISECONDS.toNanos(transactionCommitTimeoutMillis);
949 final long now = ticker().read();
951 final Queue<CommitEntry> currentQueue = !pendingFinishCommits.isEmpty() ? pendingFinishCommits :
952 !pendingCommits.isEmpty() ? pendingCommits : pendingTransactions;
953 final CommitEntry currentTx = currentQueue.peek();
954 if (currentTx != null && currentTx.lastAccess + timeout < now) {
955 LOG.warn("{}: Current transaction {} has timed out after {} ms in state {}", logContext,
956 currentTx.cohort.getIdentifier(), transactionCommitTimeoutMillis, currentTx.cohort.getState());
957 boolean processNext = true;
958 switch (currentTx.cohort.getState()) {
959 case CAN_COMMIT_PENDING:
960 currentQueue.remove().cohort.failedCanCommit(new TimeoutException());
962 case CAN_COMMIT_COMPLETE:
963 // The suppression of the FindBugs "DB_DUPLICATE_SWITCH_CLAUSES" warning pertains to this clause
964 // whose code is duplicated with PRE_COMMIT_COMPLETE. The clauses aren't combined in case the code
965 // in PRE_COMMIT_COMPLETE is changed.
966 currentQueue.remove().cohort.reportFailure(new TimeoutException());
968 case PRE_COMMIT_PENDING:
969 currentQueue.remove().cohort.failedPreCommit(new TimeoutException());
971 case PRE_COMMIT_COMPLETE:
972 // FIXME: this is a legacy behavior problem. Three-phase commit protocol specifies that after we
973 // are ready we should commit the transaction, not abort it. Our current software stack does
974 // not allow us to do that consistently, because we persist at the time of commit, hence
975 // we can end up in a state where we have pre-committed a transaction, then a leader failover
976 // occurred ... the new leader does not see the pre-committed transaction and does not have
977 // a running timer. To fix this we really need two persistence events.
979 // The first one, done at pre-commit time will hold the transaction payload. When consensus
980 // is reached, we exit the pre-commit phase and start the pre-commit timer. Followers do not
981 // apply the state in this event.
983 // The second one, done at commit (or abort) time holds only the transaction identifier and
984 // signals to followers that the state should (or should not) be applied.
986 // In order to make the pre-commit timer working across failovers, though, we need
987 // a per-shard cluster-wide monotonic time, so a follower becoming the leader can accurately
988 // restart the timer.
989 currentQueue.remove().cohort.reportFailure(new TimeoutException());
992 LOG.warn("{}: Transaction {} is still committing, cannot abort", logContext,
993 currentTx.cohort.getIdentifier());
994 currentTx.lastAccess = now;
1002 currentQueue.remove();
1006 processNextPending();
1011 boolean startAbort(final SimpleShardDataTreeCohort cohort) {
1012 final Iterator<CommitEntry> it = Iterables.concat(pendingFinishCommits, pendingCommits,
1013 pendingTransactions).iterator();
1014 if (!it.hasNext()) {
1015 LOG.debug("{}: no open transaction while attempting to abort {}", logContext, cohort.getIdentifier());
1019 // First entry is special, as it may already be committing
1020 final CommitEntry first = it.next();
1021 if (cohort.equals(first.cohort)) {
1022 if (cohort.getState() != State.COMMIT_PENDING) {
1023 LOG.debug("{}: aborting head of queue {} in state {}", logContext, cohort.getIdentifier(),
1024 cohort.getIdentifier());
1027 if (cohort.getCandidate() != null) {
1028 rebaseTransactions(it, dataTree);
1031 processNextPending();
1035 LOG.warn("{}: transaction {} is committing, skipping abort", logContext, cohort.getIdentifier());
1039 TipProducingDataTreeTip newTip = MoreObjects.firstNonNull(first.cohort.getCandidate(), dataTree);
1040 while (it.hasNext()) {
1041 final CommitEntry e = it.next();
1042 if (cohort.equals(e.cohort)) {
1043 LOG.debug("{}: aborting queued transaction {}", logContext, cohort.getIdentifier());
1046 if (cohort.getCandidate() != null) {
1047 rebaseTransactions(it, newTip);
1052 newTip = MoreObjects.firstNonNull(e.cohort.getCandidate(), newTip);
1056 LOG.debug("{}: aborted transaction {} not found in the queue", logContext, cohort.getIdentifier());
1060 @SuppressWarnings("checkstyle:IllegalCatch")
1061 private void rebaseTransactions(final Iterator<CommitEntry> iter, @Nonnull final TipProducingDataTreeTip newTip) {
1062 tip = Preconditions.checkNotNull(newTip);
1063 while (iter.hasNext()) {
1064 final SimpleShardDataTreeCohort cohort = iter.next().cohort;
1065 if (cohort.getState() == State.CAN_COMMIT_COMPLETE) {
1066 LOG.debug("{}: Revalidating queued transaction {}", logContext, cohort.getIdentifier());
1069 tip.validate(cohort.getDataTreeModification());
1070 } catch (DataValidationFailedException | RuntimeException e) {
1071 LOG.debug("{}: Failed to revalidate queued transaction {}", logContext, cohort.getIdentifier(), e);
1072 cohort.reportFailure(e);
1074 } else if (cohort.getState() == State.PRE_COMMIT_COMPLETE) {
1075 LOG.debug("{}: Repreparing queued transaction {}", logContext, cohort.getIdentifier());
1078 tip.validate(cohort.getDataTreeModification());
1079 DataTreeCandidateTip candidate = tip.prepare(cohort.getDataTreeModification());
1080 cohort.userPreCommit(candidate);
1082 cohort.setNewCandidate(candidate);
1084 } catch (ExecutionException | TimeoutException | RuntimeException | DataValidationFailedException e) {
1085 LOG.debug("{}: Failed to reprepare queued transaction {}", logContext, cohort.getIdentifier(), e);
1086 cohort.reportFailure(e);
1092 void setRunOnPendingTransactionsComplete(final Runnable operation) {
1093 runOnPendingTransactionsComplete = operation;
1094 maybeRunOperationOnPendingTransactionsComplete();
1097 private void maybeRunOperationOnPendingTransactionsComplete() {
1098 if (runOnPendingTransactionsComplete != null && !anyPendingTransactions()) {
1099 LOG.debug("{}: Pending transactions complete - running operation {}", logContext,
1100 runOnPendingTransactionsComplete);
1102 runOnPendingTransactionsComplete.run();
1103 runOnPendingTransactionsComplete = null;
1107 ShardStats getStats() {
1108 return shard.getShardMBean();