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 public 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 public 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 this(shard, schemaContext, InMemoryDataTreeFactory.getInstance().create(treeType, root),
161 treeChangeListenerPublisher, dataChangeListenerPublisher, logContext);
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 CloseLocalHistoryPayload) {
412 if (identifier != null) {
413 payloadReplicationComplete((CloseLocalHistoryPayload) payload);
415 allMetadataClosedLocalHistory(((CloseLocalHistoryPayload) payload).getIdentifier());
417 } else if (payload instanceof CreateLocalHistoryPayload) {
418 if (identifier != null) {
419 payloadReplicationComplete((CreateLocalHistoryPayload)payload);
421 allMetadataCreatedLocalHistory(((CreateLocalHistoryPayload) payload).getIdentifier());
423 } else if (payload instanceof PurgeLocalHistoryPayload) {
424 if (identifier != null) {
425 payloadReplicationComplete((PurgeLocalHistoryPayload)payload);
427 allMetadataPurgedLocalHistory(((PurgeLocalHistoryPayload) payload).getIdentifier());
430 LOG.warn("{}: ignoring unhandled identifier {} payload {}", logContext, identifier, payload);
434 private void replicatePayload(final Identifier id, final Payload payload, @Nullable final Runnable callback) {
435 if (callback != null) {
436 replicationCallbacks.put(payload, callback);
438 shard.persistPayload(id, payload, true);
441 private void payloadReplicationComplete(final AbstractIdentifiablePayload<?> payload) {
442 final Runnable callback = replicationCallbacks.remove(payload);
443 if (callback != null) {
444 LOG.debug("{}: replication of {} completed, invoking {}", logContext, payload.getIdentifier(), callback);
447 LOG.debug("{}: replication of {} has no callback", logContext, payload.getIdentifier());
451 private void payloadReplicationComplete(final TransactionIdentifier txId) {
452 final CommitEntry current = pendingFinishCommits.peek();
453 if (current == null) {
454 LOG.warn("{}: No outstanding transactions, ignoring consensus on transaction {}", logContext, txId);
458 if (!current.cohort.getIdentifier().equals(txId)) {
459 LOG.debug("{}: Head of pendingFinishCommits queue is {}, ignoring consensus on transaction {}", logContext,
460 current.cohort.getIdentifier(), txId);
464 finishCommit(current.cohort);
467 private void allMetadataAbortedTransaction(final TransactionIdentifier txId) {
468 for (ShardDataTreeMetadata<?> m : metadata) {
469 m.onTransactionAborted(txId);
473 private void allMetadataCommittedTransaction(final TransactionIdentifier txId) {
474 for (ShardDataTreeMetadata<?> m : metadata) {
475 m.onTransactionCommitted(txId);
479 private void allMetadataPurgedTransaction(final TransactionIdentifier txId) {
480 for (ShardDataTreeMetadata<?> m : metadata) {
481 m.onTransactionPurged(txId);
485 private void allMetadataCreatedLocalHistory(final LocalHistoryIdentifier historyId) {
486 for (ShardDataTreeMetadata<?> m : metadata) {
487 m.onHistoryCreated(historyId);
491 private void allMetadataClosedLocalHistory(final LocalHistoryIdentifier historyId) {
492 for (ShardDataTreeMetadata<?> m : metadata) {
493 m.onHistoryClosed(historyId);
497 private void allMetadataPurgedLocalHistory(final LocalHistoryIdentifier historyId) {
498 for (ShardDataTreeMetadata<?> m : metadata) {
499 m.onHistoryPurged(historyId);
504 * Create a transaction chain for specified history. Unlike {@link #ensureTransactionChain(LocalHistoryIdentifier)},
505 * this method is used for re-establishing state when we are taking over
507 * @param historyId Local history identifier
508 * @param closed True if the chain should be created in closed state (i.e. pending purge)
509 * @return Transaction chain handle
511 ShardDataTreeTransactionChain recreateTransactionChain(final LocalHistoryIdentifier historyId,
512 final boolean closed) {
513 final ShardDataTreeTransactionChain ret = new ShardDataTreeTransactionChain(historyId, this);
514 final ShardDataTreeTransactionChain existing = transactionChains.putIfAbsent(historyId, ret);
515 Preconditions.checkState(existing == null, "Attempted to recreate chain %s, but %s already exists", historyId,
520 ShardDataTreeTransactionChain ensureTransactionChain(final LocalHistoryIdentifier historyId) {
521 ShardDataTreeTransactionChain chain = transactionChains.get(historyId);
523 chain = new ShardDataTreeTransactionChain(historyId, this);
524 transactionChains.put(historyId, chain);
525 shard.persistPayload(historyId, CreateLocalHistoryPayload.create(historyId), true);
531 ReadOnlyShardDataTreeTransaction newReadOnlyTransaction(final TransactionIdentifier txId) {
532 if (txId.getHistoryId().getHistoryId() == 0) {
533 return new ReadOnlyShardDataTreeTransaction(this, txId, dataTree.takeSnapshot());
536 return ensureTransactionChain(txId.getHistoryId()).newReadOnlyTransaction(txId);
539 ReadWriteShardDataTreeTransaction newReadWriteTransaction(final TransactionIdentifier txId) {
540 if (txId.getHistoryId().getHistoryId() == 0) {
541 return new ReadWriteShardDataTreeTransaction(ShardDataTree.this, txId, dataTree.takeSnapshot()
545 return ensureTransactionChain(txId.getHistoryId()).newReadWriteTransaction(txId);
549 public void notifyListeners(final DataTreeCandidate candidate) {
550 treeChangeListenerPublisher.publishChanges(candidate, logContext);
551 dataChangeListenerPublisher.publishChanges(candidate, logContext);
554 void notifyOfInitialData(final DataChangeListenerRegistration<AsyncDataChangeListener<YangInstanceIdentifier,
555 NormalizedNode<?, ?>>> listenerReg, final Optional<DataTreeCandidate> currentState) {
556 if (currentState.isPresent()) {
557 ShardDataChangeListenerPublisher localPublisher = dataChangeListenerPublisher.newInstance();
558 localPublisher.registerDataChangeListener(listenerReg.getPath(), listenerReg.getInstance(),
559 listenerReg.getScope());
560 localPublisher.publishChanges(currentState.get(), logContext);
564 void notifyOfInitialData(final YangInstanceIdentifier path, final DOMDataTreeChangeListener listener,
565 final Optional<DataTreeCandidate> currentState) {
566 if (currentState.isPresent()) {
567 ShardDataTreeChangeListenerPublisher localPublisher = treeChangeListenerPublisher.newInstance();
568 localPublisher.registerTreeChangeListener(path, listener);
569 localPublisher.publishChanges(currentState.get(), logContext);
574 * Immediately purge all state relevant to leader. This includes all transaction chains and any scheduled
575 * replication callbacks.
577 void purgeLeaderState() {
578 for (ShardDataTreeTransactionChain chain : transactionChains.values()) {
582 transactionChains.clear();
583 replicationCallbacks.clear();
587 * Close a single transaction chain.
589 * @param id History identifier
590 * @param callback Callback to invoke upon completion, may be null
592 void closeTransactionChain(final LocalHistoryIdentifier id, @Nullable final Runnable callback) {
593 final ShardDataTreeTransactionChain chain = transactionChains.get(id);
595 LOG.debug("{}: Closing non-existent transaction chain {}", logContext, id);
596 if (callback != null) {
603 replicatePayload(id, CloseLocalHistoryPayload.create(id), callback);
607 * Purge a single transaction chain.
609 * @param id History identifier
610 * @param callback Callback to invoke upon completion, may be null
612 void purgeTransactionChain(final LocalHistoryIdentifier id, @Nullable final Runnable callback) {
613 final ShardDataTreeTransactionChain chain = transactionChains.remove(id);
615 LOG.debug("{}: Purging non-existent transaction chain {}", logContext, id);
616 if (callback != null) {
622 replicatePayload(id, PurgeLocalHistoryPayload.create(id), callback);
625 Entry<DataChangeListenerRegistration<AsyncDataChangeListener<YangInstanceIdentifier, NormalizedNode<?, ?>>>,
626 Optional<DataTreeCandidate>> registerChangeListener(final YangInstanceIdentifier path,
627 final AsyncDataChangeListener<YangInstanceIdentifier, NormalizedNode<?, ?>> listener,
628 final DataChangeScope scope) {
629 DataChangeListenerRegistration<AsyncDataChangeListener<YangInstanceIdentifier, NormalizedNode<?, ?>>> reg =
630 dataChangeListenerPublisher.registerDataChangeListener(path, listener, scope);
632 return new SimpleEntry<>(reg, readCurrentData());
635 private Optional<DataTreeCandidate> readCurrentData() {
636 final Optional<NormalizedNode<?, ?>> currentState =
637 dataTree.takeSnapshot().readNode(YangInstanceIdentifier.EMPTY);
638 return currentState.isPresent() ? Optional.of(DataTreeCandidates.fromNormalizedNode(
639 YangInstanceIdentifier.EMPTY, currentState.get())) : Optional.<DataTreeCandidate>absent();
642 public Entry<ListenerRegistration<DOMDataTreeChangeListener>, Optional<DataTreeCandidate>>
643 registerTreeChangeListener(final YangInstanceIdentifier path, final DOMDataTreeChangeListener listener) {
644 final ListenerRegistration<DOMDataTreeChangeListener> reg =
645 treeChangeListenerPublisher.registerTreeChangeListener(path, listener);
647 return new SimpleEntry<>(reg, readCurrentData());
651 return pendingTransactions.size() + pendingCommits.size() + pendingFinishCommits.size();
655 void abortTransaction(final AbstractShardDataTreeTransaction<?> transaction, final Runnable callback) {
656 final TransactionIdentifier id = transaction.getIdentifier();
657 LOG.debug("{}: aborting transaction {}", logContext, id);
658 replicatePayload(id, AbortTransactionPayload.create(id), callback);
663 void purgeTransaction(final TransactionIdentifier id, final Runnable callback) {
664 LOG.debug("{}: purging transaction {}", logContext, id);
665 replicatePayload(id, PurgeTransactionPayload.create(id), callback);
669 ShardDataTreeCohort finishTransaction(final ReadWriteShardDataTreeTransaction transaction) {
670 final DataTreeModification snapshot = transaction.getSnapshot();
673 return createReadyCohort(transaction.getIdentifier(), snapshot);
676 public Optional<NormalizedNode<?, ?>> readNode(final YangInstanceIdentifier path) {
677 return dataTree.takeSnapshot().readNode(path);
680 DataTreeSnapshot takeSnapshot() {
681 return dataTree.takeSnapshot();
685 public DataTreeModification newModification() {
686 return dataTree.takeSnapshot().newModification();
690 * Commits a modification.
692 * @deprecated This method violates DataTree containment and will be removed.
696 public DataTreeCandidate commit(final DataTreeModification modification) throws DataValidationFailedException {
697 // Direct modification commit is a utility, which cannot be used while we have transactions in-flight
698 Preconditions.checkState(tip == dataTree, "Cannot modify data tree while transacgitons are pending");
700 modification.ready();
701 dataTree.validate(modification);
702 DataTreeCandidate candidate = dataTree.prepare(modification);
703 dataTree.commit(candidate);
707 public Collection<ShardDataTreeCohort> getAndClearPendingTransactions() {
708 Collection<ShardDataTreeCohort> ret = new ArrayList<>(getQueueSize());
710 for (CommitEntry entry: pendingFinishCommits) {
711 ret.add(entry.cohort);
714 for (CommitEntry entry: pendingCommits) {
715 ret.add(entry.cohort);
718 for (CommitEntry entry: pendingTransactions) {
719 ret.add(entry.cohort);
722 pendingFinishCommits.clear();
723 pendingCommits.clear();
724 pendingTransactions.clear();
729 @SuppressWarnings("checkstyle:IllegalCatch")
730 private void processNextPendingTransaction() {
731 processNextPending(pendingTransactions, State.CAN_COMMIT_PENDING, entry -> {
732 final SimpleShardDataTreeCohort cohort = entry.cohort;
733 final DataTreeModification modification = cohort.getDataTreeModification();
735 LOG.debug("{}: Validating transaction {}", logContext, cohort.getIdentifier());
738 tip.validate(modification);
739 LOG.debug("{}: Transaction {} validated", logContext, cohort.getIdentifier());
740 cohort.successfulCanCommit();
741 entry.lastAccess = ticker().read();
743 } catch (ConflictingModificationAppliedException e) {
744 LOG.warn("{}: Store Tx {}: Conflicting modification for path {}.", logContext, cohort.getIdentifier(),
746 cause = new OptimisticLockFailedException("Optimistic lock failed.", e);
747 } catch (DataValidationFailedException e) {
748 LOG.warn("{}: Store Tx {}: Data validation failed for path {}.", logContext, cohort.getIdentifier(),
751 // For debugging purposes, allow dumping of the modification. Coupled with the above
752 // precondition log, it should allow us to understand what went on.
753 LOG.debug("{}: Store Tx {}: modifications: {} tree: {}", cohort.getIdentifier(), modification,
755 cause = new TransactionCommitFailedException("Data did not pass validation.", e);
756 } catch (Exception e) {
757 LOG.warn("{}: Unexpected failure in validation phase", logContext, e);
761 // Failure path: propagate the failure, remove the transaction from the queue and loop to the next one
762 pendingTransactions.poll().cohort.failedCanCommit(cause);
766 private void processNextPending() {
767 processNextPendingCommit();
768 processNextPendingTransaction();
771 private void processNextPending(final Queue<CommitEntry> queue, final State allowedState,
772 final Consumer<CommitEntry> processor) {
773 while (!queue.isEmpty()) {
774 final CommitEntry entry = queue.peek();
775 final SimpleShardDataTreeCohort cohort = entry.cohort;
777 if (cohort.isFailed()) {
778 LOG.debug("{}: Removing failed transaction {}", logContext, cohort.getIdentifier());
783 if (cohort.getState() == allowedState) {
784 processor.accept(entry);
790 maybeRunOperationOnPendingTransactionsComplete();
793 private void processNextPendingCommit() {
794 processNextPending(pendingCommits, State.COMMIT_PENDING,
795 entry -> startCommit(entry.cohort, entry.cohort.getCandidate()));
798 private boolean peekNextPendingCommit() {
799 final CommitEntry first = pendingCommits.peek();
800 return first != null && first.cohort.getState() == State.COMMIT_PENDING;
803 void startCanCommit(final SimpleShardDataTreeCohort cohort) {
804 final SimpleShardDataTreeCohort current = pendingTransactions.peek().cohort;
805 if (!cohort.equals(current)) {
806 LOG.debug("{}: Transaction {} scheduled for canCommit step", logContext, cohort.getIdentifier());
810 processNextPendingTransaction();
813 private void failPreCommit(final Exception cause) {
814 shard.getShardMBean().incrementFailedTransactionsCount();
815 pendingTransactions.poll().cohort.failedPreCommit(cause);
816 processNextPendingTransaction();
819 @SuppressWarnings("checkstyle:IllegalCatch")
820 void startPreCommit(final SimpleShardDataTreeCohort cohort) {
821 final CommitEntry entry = pendingTransactions.peek();
822 Preconditions.checkState(entry != null, "Attempted to pre-commit of %s when no transactions pending", cohort);
824 final SimpleShardDataTreeCohort current = entry.cohort;
825 Verify.verify(cohort.equals(current), "Attempted to pre-commit %s while %s is pending", cohort, current);
827 LOG.debug("{}: Preparing transaction {}", logContext, current.getIdentifier());
829 final DataTreeCandidateTip candidate;
831 candidate = tip.prepare(cohort.getDataTreeModification());
832 cohort.userPreCommit(candidate);
833 } catch (ExecutionException | TimeoutException | RuntimeException e) {
838 // Set the tip of the data tree.
839 tip = Verify.verifyNotNull(candidate);
841 entry.lastAccess = ticker().read();
843 pendingTransactions.remove();
844 pendingCommits.add(entry);
846 LOG.debug("{}: Transaction {} prepared", logContext, current.getIdentifier());
848 cohort.successfulPreCommit(candidate);
850 processNextPendingTransaction();
853 private void failCommit(final Exception cause) {
854 shard.getShardMBean().incrementFailedTransactionsCount();
855 pendingFinishCommits.poll().cohort.failedCommit(cause);
856 processNextPending();
859 @SuppressWarnings("checkstyle:IllegalCatch")
860 private void finishCommit(final SimpleShardDataTreeCohort cohort) {
861 final TransactionIdentifier txId = cohort.getIdentifier();
862 final DataTreeCandidate candidate = cohort.getCandidate();
864 LOG.debug("{}: Resuming commit of transaction {}", logContext, txId);
866 if (tip == candidate) {
867 // All pending candidates have been committed, reset the tip to the data tree.
872 dataTree.commit(candidate);
873 } catch (Exception e) {
874 LOG.error("{}: Failed to commit transaction {}", logContext, txId, e);
879 shard.getShardMBean().incrementCommittedTransactionCount();
880 shard.getShardMBean().setLastCommittedTransactionTime(System.currentTimeMillis());
882 // FIXME: propagate journal index
883 pendingFinishCommits.poll().cohort.successfulCommit(UnsignedLong.ZERO);
885 LOG.trace("{}: Transaction {} committed, proceeding to notify", logContext, txId);
886 notifyListeners(candidate);
888 processNextPending();
891 void startCommit(final SimpleShardDataTreeCohort cohort, final DataTreeCandidate candidate) {
892 final CommitEntry entry = pendingCommits.peek();
893 Preconditions.checkState(entry != null, "Attempted to start commit of %s when no transactions pending", cohort);
895 final SimpleShardDataTreeCohort current = entry.cohort;
896 if (!cohort.equals(current)) {
897 LOG.debug("{}: Transaction {} scheduled for commit step", logContext, cohort.getIdentifier());
901 LOG.debug("{}: Starting commit for transaction {}", logContext, current.getIdentifier());
903 final TransactionIdentifier txId = cohort.getIdentifier();
904 final Payload payload;
906 payload = CommitTransactionPayload.create(txId, candidate);
907 } catch (IOException e) {
908 LOG.error("{}: Failed to encode transaction {} candidate {}", logContext, txId, candidate, e);
909 pendingCommits.poll().cohort.failedCommit(e);
910 processNextPending();
914 // We process next transactions pending canCommit before we call persistPayload to possibly progress subsequent
915 // transactions to the COMMIT_PENDING state so the payloads can be batched for replication. This is done for
916 // single-shard transactions that immediately transition from canCommit to preCommit to commit. Note that
917 // if the next pending transaction is progressed to COMMIT_PENDING and this method (startCommit) is called,
918 // the next transaction will not attempt to replicate b/c the current transaction is still at the head of the
919 // pendingCommits queue.
920 processNextPendingTransaction();
922 // After processing next pending transactions, we can now remove the current transaction from pendingCommits.
923 // Note this must be done before the call to peekNextPendingCommit below so we check the next transaction
924 // in order to properly determine the batchHint flag for the call to persistPayload.
925 pendingCommits.remove();
926 pendingFinishCommits.add(entry);
928 // See if the next transaction is pending commit (ie in the COMMIT_PENDING state) so it can be batched with
929 // this transaction for replication.
930 boolean replicationBatchHint = peekNextPendingCommit();
932 // Once completed, we will continue via payloadReplicationComplete
933 shard.persistPayload(txId, payload, replicationBatchHint);
935 entry.lastAccess = shard.ticker().read();
937 LOG.debug("{}: Transaction {} submitted to persistence", logContext, txId);
939 // Process the next transaction pending commit, if any. If there is one it will be batched with this
940 // transaction for replication.
941 processNextPendingCommit();
944 Collection<ActorRef> getCohortActors() {
945 return cohortRegistry.getCohortActors();
948 void processCohortRegistryCommand(final ActorRef sender, final CohortRegistryCommand message) {
949 cohortRegistry.process(sender, message);
953 ShardDataTreeCohort createReadyCohort(final TransactionIdentifier txId,
954 final DataTreeModification modification) {
955 SimpleShardDataTreeCohort cohort = new SimpleShardDataTreeCohort(this, modification, txId,
956 cohortRegistry.createCohort(schemaContext, txId, COMMIT_STEP_TIMEOUT));
957 pendingTransactions.add(new CommitEntry(cohort, ticker().read()));
961 @SuppressFBWarnings(value = "DB_DUPLICATE_SWITCH_CLAUSES", justification = "See inline comments below.")
962 void checkForExpiredTransactions(final long transactionCommitTimeoutMillis) {
963 final long timeout = TimeUnit.MILLISECONDS.toNanos(transactionCommitTimeoutMillis);
964 final long now = ticker().read();
966 final Queue<CommitEntry> currentQueue = !pendingFinishCommits.isEmpty() ? pendingFinishCommits :
967 !pendingCommits.isEmpty() ? pendingCommits : pendingTransactions;
968 final CommitEntry currentTx = currentQueue.peek();
969 if (currentTx != null && currentTx.lastAccess + timeout < now) {
970 LOG.warn("{}: Current transaction {} has timed out after {} ms in state {}", logContext,
971 currentTx.cohort.getIdentifier(), transactionCommitTimeoutMillis, currentTx.cohort.getState());
972 boolean processNext = true;
973 switch (currentTx.cohort.getState()) {
974 case CAN_COMMIT_PENDING:
975 currentQueue.remove().cohort.failedCanCommit(new TimeoutException());
977 case CAN_COMMIT_COMPLETE:
978 // The suppression of the FindBugs "DB_DUPLICATE_SWITCH_CLAUSES" warning pertains to this clause
979 // whose code is duplicated with PRE_COMMIT_COMPLETE. The clauses aren't combined in case the code
980 // in PRE_COMMIT_COMPLETE is changed.
981 currentQueue.remove().cohort.reportFailure(new TimeoutException());
983 case PRE_COMMIT_PENDING:
984 currentQueue.remove().cohort.failedPreCommit(new TimeoutException());
986 case PRE_COMMIT_COMPLETE:
987 // FIXME: this is a legacy behavior problem. Three-phase commit protocol specifies that after we
988 // are ready we should commit the transaction, not abort it. Our current software stack does
989 // not allow us to do that consistently, because we persist at the time of commit, hence
990 // we can end up in a state where we have pre-committed a transaction, then a leader failover
991 // occurred ... the new leader does not see the pre-committed transaction and does not have
992 // a running timer. To fix this we really need two persistence events.
994 // The first one, done at pre-commit time will hold the transaction payload. When consensus
995 // is reached, we exit the pre-commit phase and start the pre-commit timer. Followers do not
996 // apply the state in this event.
998 // The second one, done at commit (or abort) time holds only the transaction identifier and
999 // signals to followers that the state should (or should not) be applied.
1001 // In order to make the pre-commit timer working across failovers, though, we need
1002 // a per-shard cluster-wide monotonic time, so a follower becoming the leader can accurately
1003 // restart the timer.
1004 currentQueue.remove().cohort.reportFailure(new TimeoutException());
1006 case COMMIT_PENDING:
1007 LOG.warn("{}: Transaction {} is still committing, cannot abort", logContext,
1008 currentTx.cohort.getIdentifier());
1009 currentTx.lastAccess = now;
1010 processNext = false;
1017 currentQueue.remove();
1021 processNextPending();
1026 boolean startAbort(final SimpleShardDataTreeCohort cohort) {
1027 final Iterator<CommitEntry> it = Iterables.concat(pendingFinishCommits, pendingCommits,
1028 pendingTransactions).iterator();
1029 if (!it.hasNext()) {
1030 LOG.debug("{}: no open transaction while attempting to abort {}", logContext, cohort.getIdentifier());
1034 // First entry is special, as it may already be committing
1035 final CommitEntry first = it.next();
1036 if (cohort.equals(first.cohort)) {
1037 if (cohort.getState() != State.COMMIT_PENDING) {
1038 LOG.debug("{}: aborting head of queue {} in state {}", logContext, cohort.getIdentifier(),
1039 cohort.getIdentifier());
1042 if (cohort.getCandidate() != null) {
1043 rebaseTransactions(it, dataTree);
1046 processNextPending();
1050 LOG.warn("{}: transaction {} is committing, skipping abort", logContext, cohort.getIdentifier());
1054 TipProducingDataTreeTip newTip = MoreObjects.firstNonNull(first.cohort.getCandidate(), dataTree);
1055 while (it.hasNext()) {
1056 final CommitEntry e = it.next();
1057 if (cohort.equals(e.cohort)) {
1058 LOG.debug("{}: aborting queued transaction {}", logContext, cohort.getIdentifier());
1061 if (cohort.getCandidate() != null) {
1062 rebaseTransactions(it, newTip);
1067 newTip = MoreObjects.firstNonNull(e.cohort.getCandidate(), newTip);
1071 LOG.debug("{}: aborted transaction {} not found in the queue", logContext, cohort.getIdentifier());
1075 @SuppressWarnings("checkstyle:IllegalCatch")
1076 private void rebaseTransactions(final Iterator<CommitEntry> iter, @Nonnull final TipProducingDataTreeTip newTip) {
1077 tip = Preconditions.checkNotNull(newTip);
1078 while (iter.hasNext()) {
1079 final SimpleShardDataTreeCohort cohort = iter.next().cohort;
1080 if (cohort.getState() == State.CAN_COMMIT_COMPLETE) {
1081 LOG.debug("{}: Revalidating queued transaction {}", logContext, cohort.getIdentifier());
1084 tip.validate(cohort.getDataTreeModification());
1085 } catch (DataValidationFailedException | RuntimeException e) {
1086 LOG.debug("{}: Failed to revalidate queued transaction {}", logContext, cohort.getIdentifier(), e);
1087 cohort.reportFailure(e);
1089 } else if (cohort.getState() == State.PRE_COMMIT_COMPLETE) {
1090 LOG.debug("{}: Repreparing queued transaction {}", logContext, cohort.getIdentifier());
1093 tip.validate(cohort.getDataTreeModification());
1094 DataTreeCandidateTip candidate = tip.prepare(cohort.getDataTreeModification());
1095 cohort.userPreCommit(candidate);
1097 cohort.setNewCandidate(candidate);
1099 } catch (ExecutionException | TimeoutException | RuntimeException | DataValidationFailedException e) {
1100 LOG.debug("{}: Failed to reprepare queued transaction {}", logContext, cohort.getIdentifier(), e);
1101 cohort.reportFailure(e);
1107 void setRunOnPendingTransactionsComplete(final Runnable operation) {
1108 runOnPendingTransactionsComplete = operation;
1109 maybeRunOperationOnPendingTransactionsComplete();
1112 private void maybeRunOperationOnPendingTransactionsComplete() {
1113 if (runOnPendingTransactionsComplete != null && !anyPendingTransactions()) {
1114 LOG.debug("{}: Pending transactions complete - running operation {}", logContext,
1115 runOnPendingTransactionsComplete);
1117 runOnPendingTransactionsComplete.run();
1118 runOnPendingTransactionsComplete = null;