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.Verify;
18 import com.google.common.collect.ImmutableList;
19 import com.google.common.collect.ImmutableMap;
20 import com.google.common.collect.ImmutableMap.Builder;
21 import com.google.common.collect.Iterables;
22 import com.google.common.primitives.UnsignedLong;
23 import edu.umd.cs.findbugs.annotations.SuppressFBWarnings;
25 import java.io.IOException;
26 import java.util.ArrayDeque;
27 import java.util.ArrayList;
28 import java.util.Collection;
29 import java.util.HashMap;
30 import java.util.Iterator;
32 import java.util.Map.Entry;
33 import java.util.Queue;
34 import java.util.concurrent.ExecutionException;
35 import java.util.concurrent.TimeUnit;
36 import java.util.concurrent.TimeoutException;
37 import java.util.function.Consumer;
38 import java.util.function.Function;
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);
111 * Process this many transactions in a single batched run. If we exceed this limit, we need to schedule later
112 * execution to finish up the batch. This is necessary in case of a long list of transactions which progress
113 * immediately through their preCommit phase -- if that happens, their completion eats up stack frames and could
114 * result in StackOverflowError.
116 private static final int MAX_TRANSACTION_BATCH = 100;
118 private final Map<LocalHistoryIdentifier, ShardDataTreeTransactionChain> transactionChains = new HashMap<>();
119 private final DataTreeCohortActorRegistry cohortRegistry = new DataTreeCohortActorRegistry();
120 private final Queue<CommitEntry> pendingTransactions = new ArrayDeque<>();
121 private final Queue<CommitEntry> pendingCommits = new ArrayDeque<>();
122 private final Queue<CommitEntry> pendingFinishCommits = new ArrayDeque<>();
125 * Callbacks that need to be invoked once a payload is replicated.
127 private final Map<Payload, Runnable> replicationCallbacks = new HashMap<>();
129 private final ShardDataTreeChangeListenerPublisher treeChangeListenerPublisher;
130 private final ShardDataChangeListenerPublisher dataChangeListenerPublisher;
131 private final Collection<ShardDataTreeMetadata<?>> metadata;
132 private final TipProducingDataTree dataTree;
133 private final String logContext;
134 private final Shard shard;
135 private Runnable runOnPendingTransactionsComplete;
138 * Optimistic {@link DataTreeCandidate} preparation. Since our DataTree implementation is a
139 * {@link TipProducingDataTree}, each {@link DataTreeCandidate} is also a {@link DataTreeTip}, e.g. another
140 * candidate can be prepared on top of it. They still need to be committed in sequence. Here we track the current
141 * tip of the data tree, which is the last DataTreeCandidate we have in flight, or the DataTree itself.
143 private TipProducingDataTreeTip tip;
145 private SchemaContext schemaContext;
147 private int currentTransactionBatch;
149 ShardDataTree(final Shard shard, final SchemaContext schemaContext, final TipProducingDataTree dataTree,
150 final ShardDataTreeChangeListenerPublisher treeChangeListenerPublisher,
151 final ShardDataChangeListenerPublisher dataChangeListenerPublisher, final String logContext,
152 final ShardDataTreeMetadata<?>... metadata) {
153 this.dataTree = Preconditions.checkNotNull(dataTree);
154 updateSchemaContext(schemaContext);
156 this.shard = Preconditions.checkNotNull(shard);
157 this.treeChangeListenerPublisher = Preconditions.checkNotNull(treeChangeListenerPublisher);
158 this.dataChangeListenerPublisher = Preconditions.checkNotNull(dataChangeListenerPublisher);
159 this.logContext = Preconditions.checkNotNull(logContext);
160 this.metadata = ImmutableList.copyOf(metadata);
164 ShardDataTree(final Shard shard, final SchemaContext schemaContext, final TreeType treeType,
165 final YangInstanceIdentifier root,
166 final ShardDataTreeChangeListenerPublisher treeChangeListenerPublisher,
167 final ShardDataChangeListenerPublisher dataChangeListenerPublisher, final String logContext,
168 final ShardDataTreeMetadata<?>... metadata) {
169 this(shard, schemaContext, InMemoryDataTreeFactory.getInstance().create(treeType, root),
170 treeChangeListenerPublisher, dataChangeListenerPublisher, logContext, metadata);
174 public ShardDataTree(final Shard shard, final SchemaContext schemaContext, final TreeType treeType) {
175 this(shard, schemaContext, treeType, YangInstanceIdentifier.EMPTY,
176 new DefaultShardDataTreeChangeListenerPublisher(""),
177 new DefaultShardDataChangeListenerPublisher(""), "");
180 final String logContext() {
184 final long readTime() {
185 return shard.ticker().read();
188 public TipProducingDataTree getDataTree() {
192 SchemaContext getSchemaContext() {
193 return schemaContext;
196 void updateSchemaContext(final SchemaContext newSchemaContext) {
197 dataTree.setSchemaContext(newSchemaContext);
198 this.schemaContext = Preconditions.checkNotNull(newSchemaContext);
201 void resetTransactionBatch() {
202 currentTransactionBatch = 0;
206 * Take a snapshot of current state for later recovery.
208 * @return A state snapshot
210 @Nonnull ShardDataTreeSnapshot takeStateSnapshot() {
211 final NormalizedNode<?, ?> rootNode = dataTree.takeSnapshot().readNode(YangInstanceIdentifier.EMPTY).get();
212 final Builder<Class<? extends ShardDataTreeSnapshotMetadata<?>>, ShardDataTreeSnapshotMetadata<?>> metaBuilder =
213 ImmutableMap.builder();
215 for (ShardDataTreeMetadata<?> m : metadata) {
216 final ShardDataTreeSnapshotMetadata<?> meta = m.toSnapshot();
218 metaBuilder.put(meta.getType(), meta);
222 return new MetadataShardDataTreeSnapshot(rootNode, metaBuilder.build());
225 private boolean anyPendingTransactions() {
226 return !pendingTransactions.isEmpty() || !pendingCommits.isEmpty() || !pendingFinishCommits.isEmpty();
229 private void applySnapshot(@Nonnull final ShardDataTreeSnapshot snapshot,
230 final UnaryOperator<DataTreeModification> wrapper) throws DataValidationFailedException {
231 final Stopwatch elapsed = Stopwatch.createStarted();
233 if (anyPendingTransactions()) {
234 LOG.warn("{}: applying state snapshot with pending transactions", logContext);
237 final Map<Class<? extends ShardDataTreeSnapshotMetadata<?>>, ShardDataTreeSnapshotMetadata<?>> snapshotMeta;
238 if (snapshot instanceof MetadataShardDataTreeSnapshot) {
239 snapshotMeta = ((MetadataShardDataTreeSnapshot) snapshot).getMetadata();
241 snapshotMeta = ImmutableMap.of();
244 for (ShardDataTreeMetadata<?> m : metadata) {
245 final ShardDataTreeSnapshotMetadata<?> s = snapshotMeta.get(m.getSupportedType());
253 final DataTreeModification mod = wrapper.apply(dataTree.takeSnapshot().newModification());
254 // delete everything first
255 mod.delete(YangInstanceIdentifier.EMPTY);
257 final java.util.Optional<NormalizedNode<?, ?>> maybeNode = snapshot.getRootNode();
258 if (maybeNode.isPresent()) {
259 // Add everything from the remote node back
260 mod.write(YangInstanceIdentifier.EMPTY, maybeNode.get());
264 final DataTreeModification unwrapped = unwrap(mod);
265 dataTree.validate(unwrapped);
266 DataTreeCandidateTip candidate = dataTree.prepare(unwrapped);
267 dataTree.commit(candidate);
268 notifyListeners(candidate);
270 LOG.debug("{}: state snapshot applied in {}", logContext, elapsed);
274 * Apply a snapshot coming from the leader. This method assumes the leader and follower SchemaContexts match and
275 * does not perform any pruning.
277 * @param snapshot Snapshot that needs to be applied
278 * @throws DataValidationFailedException when the snapshot fails to apply
280 void applySnapshot(@Nonnull final ShardDataTreeSnapshot snapshot) throws DataValidationFailedException {
281 applySnapshot(snapshot, UnaryOperator.identity());
284 private PruningDataTreeModification wrapWithPruning(final DataTreeModification delegate) {
285 return new PruningDataTreeModification(delegate, dataTree, schemaContext);
288 private static DataTreeModification unwrap(final DataTreeModification modification) {
289 if (modification instanceof PruningDataTreeModification) {
290 return ((PruningDataTreeModification)modification).delegate();
296 * Apply a snapshot coming from recovery. This method does not assume the SchemaContexts match and performs data
297 * pruning in an attempt to adjust the state to our current SchemaContext.
299 * @param snapshot Snapshot that needs to be applied
300 * @throws DataValidationFailedException when the snapshot fails to apply
302 void applyRecoverySnapshot(final @Nonnull ShardDataTreeSnapshot snapshot) throws DataValidationFailedException {
303 applySnapshot(snapshot, this::wrapWithPruning);
306 @SuppressWarnings("checkstyle:IllegalCatch")
307 private void applyRecoveryCandidate(final DataTreeCandidate candidate) throws DataValidationFailedException {
308 final PruningDataTreeModification mod = wrapWithPruning(dataTree.takeSnapshot().newModification());
309 DataTreeCandidates.applyToModification(mod, candidate);
312 final DataTreeModification unwrapped = mod.delegate();
313 LOG.trace("{}: Applying recovery modification {}", logContext, unwrapped);
316 dataTree.validate(unwrapped);
317 dataTree.commit(dataTree.prepare(unwrapped));
318 } catch (Exception e) {
319 File file = new File(System.getProperty("karaf.data", "."),
320 "failed-recovery-payload-" + logContext + ".out");
321 DataTreeModificationOutput.toFile(file, unwrapped);
322 throw new IllegalStateException(String.format(
323 "%s: Failed to apply recovery payload. Modification data was written to file %s",
324 logContext, file), e);
329 * Apply a payload coming from recovery. This method does not assume the SchemaContexts match and performs data
330 * pruning in an attempt to adjust the state to our current SchemaContext.
332 * @param payload Payload
333 * @throws IOException when the snapshot fails to deserialize
334 * @throws DataValidationFailedException when the snapshot fails to apply
336 void applyRecoveryPayload(final @Nonnull Payload payload) throws IOException, DataValidationFailedException {
337 if (payload instanceof CommitTransactionPayload) {
338 final Entry<TransactionIdentifier, DataTreeCandidate> e =
339 ((CommitTransactionPayload) payload).getCandidate();
340 applyRecoveryCandidate(e.getValue());
341 allMetadataCommittedTransaction(e.getKey());
342 } else if (payload instanceof AbortTransactionPayload) {
343 allMetadataAbortedTransaction(((AbortTransactionPayload) payload).getIdentifier());
344 } else if (payload instanceof PurgeTransactionPayload) {
345 allMetadataPurgedTransaction(((PurgeTransactionPayload) payload).getIdentifier());
346 } else if (payload instanceof CreateLocalHistoryPayload) {
347 allMetadataCreatedLocalHistory(((CreateLocalHistoryPayload) payload).getIdentifier());
348 } else if (payload instanceof CloseLocalHistoryPayload) {
349 allMetadataClosedLocalHistory(((CloseLocalHistoryPayload) payload).getIdentifier());
350 } else if (payload instanceof PurgeLocalHistoryPayload) {
351 allMetadataPurgedLocalHistory(((PurgeLocalHistoryPayload) payload).getIdentifier());
352 } else if (payload instanceof DataTreeCandidatePayload) {
353 applyRecoveryCandidate(((DataTreeCandidatePayload) payload).getCandidate());
355 LOG.debug("{}: ignoring unhandled payload {}", logContext, payload);
359 private void applyReplicatedCandidate(final Identifier identifier, final DataTreeCandidate foreign)
360 throws DataValidationFailedException {
361 LOG.debug("{}: Applying foreign transaction {}", logContext, identifier);
363 final DataTreeModification mod = dataTree.takeSnapshot().newModification();
364 DataTreeCandidates.applyToModification(mod, foreign);
367 LOG.trace("{}: Applying foreign modification {}", logContext, mod);
368 dataTree.validate(mod);
369 final DataTreeCandidate candidate = dataTree.prepare(mod);
370 dataTree.commit(candidate);
372 notifyListeners(candidate);
376 * Apply a payload coming from the leader, which could actually be us. This method assumes the leader and follower
377 * SchemaContexts match and does not perform any pruning.
379 * @param identifier Payload identifier as returned from RaftActor
380 * @param payload Payload
381 * @throws IOException when the snapshot fails to deserialize
382 * @throws DataValidationFailedException when the snapshot fails to apply
384 void applyReplicatedPayload(final Identifier identifier, final Payload payload) throws IOException,
385 DataValidationFailedException {
387 * 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
388 * if we are the leader and it has originated with us.
390 * The identifier will only ever be non-null when we were the leader which achieved consensus. Unfortunately,
391 * though, this may not be the case anymore, as we are being called some time afterwards and we may not be
392 * acting in that capacity anymore.
394 * In any case, we know that this is an entry coming from replication, hence we can be sure we will not observe
395 * pre-Boron state -- which limits the number of options here.
397 if (payload instanceof CommitTransactionPayload) {
398 final TransactionIdentifier txId;
399 if (identifier == null) {
400 final Entry<TransactionIdentifier, DataTreeCandidate> e =
401 ((CommitTransactionPayload) payload).getCandidate();
403 applyReplicatedCandidate(txId, e.getValue());
405 Verify.verify(identifier instanceof TransactionIdentifier);
406 txId = (TransactionIdentifier) identifier;
407 payloadReplicationComplete(txId);
409 allMetadataCommittedTransaction(txId);
410 } else if (payload instanceof AbortTransactionPayload) {
411 if (identifier != null) {
412 payloadReplicationComplete((AbortTransactionPayload) payload);
414 allMetadataAbortedTransaction(((AbortTransactionPayload) payload).getIdentifier());
415 } else if (payload instanceof PurgeTransactionPayload) {
416 if (identifier != null) {
417 payloadReplicationComplete((PurgeTransactionPayload) payload);
419 allMetadataPurgedTransaction(((PurgeTransactionPayload) payload).getIdentifier());
420 } else if (payload instanceof CloseLocalHistoryPayload) {
421 if (identifier != null) {
422 payloadReplicationComplete((CloseLocalHistoryPayload) payload);
424 allMetadataClosedLocalHistory(((CloseLocalHistoryPayload) payload).getIdentifier());
425 } else if (payload instanceof CreateLocalHistoryPayload) {
426 if (identifier != null) {
427 payloadReplicationComplete((CreateLocalHistoryPayload)payload);
429 allMetadataCreatedLocalHistory(((CreateLocalHistoryPayload) payload).getIdentifier());
430 } else if (payload instanceof PurgeLocalHistoryPayload) {
431 if (identifier != null) {
432 payloadReplicationComplete((PurgeLocalHistoryPayload)payload);
434 allMetadataPurgedLocalHistory(((PurgeLocalHistoryPayload) payload).getIdentifier());
436 LOG.warn("{}: ignoring unhandled identifier {} payload {}", logContext, identifier, payload);
440 private void replicatePayload(final Identifier id, final Payload payload, @Nullable final Runnable callback) {
441 if (callback != null) {
442 replicationCallbacks.put(payload, callback);
444 shard.persistPayload(id, payload, true);
447 private void payloadReplicationComplete(final AbstractIdentifiablePayload<?> payload) {
448 final Runnable callback = replicationCallbacks.remove(payload);
449 if (callback != null) {
450 LOG.debug("{}: replication of {} completed, invoking {}", logContext, payload.getIdentifier(), callback);
453 LOG.debug("{}: replication of {} has no callback", logContext, payload.getIdentifier());
457 private void payloadReplicationComplete(final TransactionIdentifier txId) {
458 final CommitEntry current = pendingFinishCommits.peek();
459 if (current == null) {
460 LOG.warn("{}: No outstanding transactions, ignoring consensus on transaction {}", logContext, txId);
464 if (!current.cohort.getIdentifier().equals(txId)) {
465 LOG.debug("{}: Head of pendingFinishCommits queue is {}, ignoring consensus on transaction {}", logContext,
466 current.cohort.getIdentifier(), txId);
470 finishCommit(current.cohort);
473 private void allMetadataAbortedTransaction(final TransactionIdentifier txId) {
474 for (ShardDataTreeMetadata<?> m : metadata) {
475 m.onTransactionAborted(txId);
479 private void allMetadataCommittedTransaction(final TransactionIdentifier txId) {
480 for (ShardDataTreeMetadata<?> m : metadata) {
481 m.onTransactionCommitted(txId);
485 private void allMetadataPurgedTransaction(final TransactionIdentifier txId) {
486 for (ShardDataTreeMetadata<?> m : metadata) {
487 m.onTransactionPurged(txId);
491 private void allMetadataCreatedLocalHistory(final LocalHistoryIdentifier historyId) {
492 for (ShardDataTreeMetadata<?> m : metadata) {
493 m.onHistoryCreated(historyId);
497 private void allMetadataClosedLocalHistory(final LocalHistoryIdentifier historyId) {
498 for (ShardDataTreeMetadata<?> m : metadata) {
499 m.onHistoryClosed(historyId);
503 private void allMetadataPurgedLocalHistory(final LocalHistoryIdentifier historyId) {
504 for (ShardDataTreeMetadata<?> m : metadata) {
505 m.onHistoryPurged(historyId);
510 * Create a transaction chain for specified history. Unlike {@link #ensureTransactionChain(LocalHistoryIdentifier)},
511 * this method is used for re-establishing state when we are taking over
513 * @param historyId Local history identifier
514 * @param closed True if the chain should be created in closed state (i.e. pending purge)
515 * @return Transaction chain handle
517 ShardDataTreeTransactionChain recreateTransactionChain(final LocalHistoryIdentifier historyId,
518 final boolean closed) {
519 final ShardDataTreeTransactionChain ret = new ShardDataTreeTransactionChain(historyId, this);
520 final ShardDataTreeTransactionChain existing = transactionChains.putIfAbsent(historyId, ret);
521 Preconditions.checkState(existing == null, "Attempted to recreate chain %s, but %s already exists", historyId,
526 ShardDataTreeTransactionChain ensureTransactionChain(final LocalHistoryIdentifier historyId,
527 @Nullable final Runnable callback) {
528 ShardDataTreeTransactionChain chain = transactionChains.get(historyId);
530 chain = new ShardDataTreeTransactionChain(historyId, this);
531 transactionChains.put(historyId, chain);
532 replicatePayload(historyId, CreateLocalHistoryPayload.create(historyId), callback);
533 } else if (callback != null) {
540 ReadOnlyShardDataTreeTransaction newReadOnlyTransaction(final TransactionIdentifier txId) {
541 if (txId.getHistoryId().getHistoryId() == 0) {
542 return new ReadOnlyShardDataTreeTransaction(this, txId, dataTree.takeSnapshot());
545 return ensureTransactionChain(txId.getHistoryId(), null).newReadOnlyTransaction(txId);
548 ReadWriteShardDataTreeTransaction newReadWriteTransaction(final TransactionIdentifier txId) {
549 if (txId.getHistoryId().getHistoryId() == 0) {
550 return new ReadWriteShardDataTreeTransaction(ShardDataTree.this, txId, dataTree.takeSnapshot()
554 return ensureTransactionChain(txId.getHistoryId(), null).newReadWriteTransaction(txId);
558 public void notifyListeners(final DataTreeCandidate candidate) {
559 treeChangeListenerPublisher.publishChanges(candidate);
560 dataChangeListenerPublisher.publishChanges(candidate);
564 * Immediately purge all state relevant to leader. This includes all transaction chains and any scheduled
565 * replication callbacks.
567 void purgeLeaderState() {
568 for (ShardDataTreeTransactionChain chain : transactionChains.values()) {
572 transactionChains.clear();
573 replicationCallbacks.clear();
577 * Close a single transaction chain.
579 * @param id History identifier
580 * @param callback Callback to invoke upon completion, may be null
582 void closeTransactionChain(final LocalHistoryIdentifier id, @Nullable final Runnable callback) {
583 final ShardDataTreeTransactionChain chain = transactionChains.get(id);
585 LOG.debug("{}: Closing non-existent transaction chain {}", logContext, id);
586 if (callback != null) {
593 replicatePayload(id, CloseLocalHistoryPayload.create(id), callback);
597 * Purge a single transaction chain.
599 * @param id History identifier
600 * @param callback Callback to invoke upon completion, may be null
602 void purgeTransactionChain(final LocalHistoryIdentifier id, @Nullable final Runnable callback) {
603 final ShardDataTreeTransactionChain chain = transactionChains.remove(id);
605 LOG.debug("{}: Purging non-existent transaction chain {}", logContext, id);
606 if (callback != null) {
612 replicatePayload(id, PurgeLocalHistoryPayload.create(id), callback);
615 void registerDataChangeListener(final YangInstanceIdentifier path,
616 final AsyncDataChangeListener<YangInstanceIdentifier, NormalizedNode<?, ?>> listener,
617 final DataChangeScope scope, final Optional<DataTreeCandidate> initialState,
618 final Consumer<ListenerRegistration<AsyncDataChangeListener<YangInstanceIdentifier, NormalizedNode<?, ?>>>>
620 dataChangeListenerPublisher.registerDataChangeListener(path, listener, scope, initialState, onRegistration);
623 Optional<DataTreeCandidate> readCurrentData() {
624 final Optional<NormalizedNode<?, ?>> currentState =
625 dataTree.takeSnapshot().readNode(YangInstanceIdentifier.EMPTY);
626 return currentState.isPresent() ? Optional.of(DataTreeCandidates.fromNormalizedNode(
627 YangInstanceIdentifier.EMPTY, currentState.get())) : Optional.<DataTreeCandidate>absent();
630 public void registerTreeChangeListener(final YangInstanceIdentifier path, final DOMDataTreeChangeListener listener,
631 final Optional<DataTreeCandidate> initialState,
632 final Consumer<ListenerRegistration<DOMDataTreeChangeListener>> onRegistration) {
633 treeChangeListenerPublisher.registerTreeChangeListener(path, listener, initialState, onRegistration);
637 return pendingTransactions.size() + pendingCommits.size() + pendingFinishCommits.size();
641 void abortTransaction(final AbstractShardDataTreeTransaction<?> transaction, final Runnable callback) {
642 final TransactionIdentifier id = transaction.getIdentifier();
643 LOG.debug("{}: aborting transaction {}", logContext, id);
644 replicatePayload(id, AbortTransactionPayload.create(id), callback);
648 void abortFromTransactionActor(final AbstractShardDataTreeTransaction<?> transaction) {
649 // No-op for free-standing transactions
654 ShardDataTreeCohort finishTransaction(final ReadWriteShardDataTreeTransaction transaction) {
655 final DataTreeModification snapshot = transaction.getSnapshot();
658 return createReadyCohort(transaction.getIdentifier(), snapshot);
661 void purgeTransaction(final TransactionIdentifier id, final Runnable callback) {
662 LOG.debug("{}: purging transaction {}", logContext, id);
663 replicatePayload(id, PurgeTransactionPayload.create(id), callback);
666 public Optional<NormalizedNode<?, ?>> readNode(final YangInstanceIdentifier path) {
667 return dataTree.takeSnapshot().readNode(path);
670 DataTreeSnapshot takeSnapshot() {
671 return dataTree.takeSnapshot();
675 public DataTreeModification newModification() {
676 return dataTree.takeSnapshot().newModification();
680 * Commits a modification.
682 * @deprecated This method violates DataTree containment and will be removed.
686 public DataTreeCandidate commit(final DataTreeModification modification) throws DataValidationFailedException {
687 // Direct modification commit is a utility, which cannot be used while we have transactions in-flight
688 Preconditions.checkState(tip == dataTree, "Cannot modify data tree while transacgitons are pending");
690 modification.ready();
691 dataTree.validate(modification);
692 DataTreeCandidate candidate = dataTree.prepare(modification);
693 dataTree.commit(candidate);
697 public Collection<ShardDataTreeCohort> getAndClearPendingTransactions() {
698 Collection<ShardDataTreeCohort> ret = new ArrayList<>(getQueueSize());
700 for (CommitEntry entry: pendingFinishCommits) {
701 ret.add(entry.cohort);
704 for (CommitEntry entry: pendingCommits) {
705 ret.add(entry.cohort);
708 for (CommitEntry entry: pendingTransactions) {
709 ret.add(entry.cohort);
712 pendingFinishCommits.clear();
713 pendingCommits.clear();
714 pendingTransactions.clear();
720 * Called some time after {@link #processNextPendingTransaction()} decides to stop processing.
722 void resumeNextPendingTransaction() {
723 LOG.debug("{}: attempting to resume transaction processing", logContext);
724 processNextPending();
727 @SuppressWarnings("checkstyle:IllegalCatch")
728 private void processNextPendingTransaction() {
729 ++currentTransactionBatch;
730 if (currentTransactionBatch > MAX_TRANSACTION_BATCH) {
731 LOG.debug("{}: Already processed {}, scheduling continuation", logContext, currentTransactionBatch);
732 shard.scheduleNextPendingTransaction();
736 processNextPending(pendingTransactions, State.CAN_COMMIT_PENDING, entry -> {
737 final SimpleShardDataTreeCohort cohort = entry.cohort;
738 final DataTreeModification modification = cohort.getDataTreeModification();
740 LOG.debug("{}: Validating transaction {}", logContext, cohort.getIdentifier());
743 tip.validate(modification);
744 LOG.debug("{}: Transaction {} validated", logContext, cohort.getIdentifier());
745 cohort.successfulCanCommit();
746 entry.lastAccess = readTime();
748 } catch (ConflictingModificationAppliedException e) {
749 LOG.warn("{}: Store Tx {}: Conflicting modification for path {}.", logContext, cohort.getIdentifier(),
751 cause = new OptimisticLockFailedException("Optimistic lock failed.", e);
752 } catch (DataValidationFailedException e) {
753 LOG.warn("{}: Store Tx {}: Data validation failed for path {}.", logContext, cohort.getIdentifier(),
756 // For debugging purposes, allow dumping of the modification. Coupled with the above
757 // precondition log, it should allow us to understand what went on.
758 LOG.debug("{}: Store Tx {}: modifications: {} tree: {}", cohort.getIdentifier(), modification,
760 cause = new TransactionCommitFailedException("Data did not pass validation.", e);
761 } catch (Exception e) {
762 LOG.warn("{}: Unexpected failure in validation phase", logContext, e);
766 // Failure path: propagate the failure, remove the transaction from the queue and loop to the next one
767 pendingTransactions.poll().cohort.failedCanCommit(cause);
771 private void processNextPending() {
772 processNextPendingCommit();
773 processNextPendingTransaction();
776 private void processNextPending(final Queue<CommitEntry> queue, final State allowedState,
777 final Consumer<CommitEntry> processor) {
778 while (!queue.isEmpty()) {
779 final CommitEntry entry = queue.peek();
780 final SimpleShardDataTreeCohort cohort = entry.cohort;
782 if (cohort.isFailed()) {
783 LOG.debug("{}: Removing failed transaction {}", logContext, cohort.getIdentifier());
788 if (cohort.getState() == allowedState) {
789 processor.accept(entry);
795 maybeRunOperationOnPendingTransactionsComplete();
798 private void processNextPendingCommit() {
799 processNextPending(pendingCommits, State.COMMIT_PENDING,
800 entry -> startCommit(entry.cohort, entry.cohort.getCandidate()));
803 private boolean peekNextPendingCommit() {
804 final CommitEntry first = pendingCommits.peek();
805 return first != null && first.cohort.getState() == State.COMMIT_PENDING;
808 void startCanCommit(final SimpleShardDataTreeCohort cohort) {
809 final CommitEntry head = pendingTransactions.peek();
811 LOG.warn("{}: No transactions enqueued while attempting to start canCommit on {}", logContext, cohort);
814 if (!cohort.equals(head.cohort)) {
815 LOG.debug("{}: Transaction {} scheduled for canCommit step", logContext, cohort.getIdentifier());
819 processNextPendingTransaction();
822 private void failPreCommit(final Exception cause) {
823 shard.getShardMBean().incrementFailedTransactionsCount();
824 pendingTransactions.poll().cohort.failedPreCommit(cause);
825 processNextPendingTransaction();
828 @SuppressWarnings("checkstyle:IllegalCatch")
829 void startPreCommit(final SimpleShardDataTreeCohort cohort) {
830 final CommitEntry entry = pendingTransactions.peek();
831 Preconditions.checkState(entry != null, "Attempted to pre-commit of %s when no transactions pending", cohort);
833 final SimpleShardDataTreeCohort current = entry.cohort;
834 Verify.verify(cohort.equals(current), "Attempted to pre-commit %s while %s is pending", cohort, current);
836 LOG.debug("{}: Preparing transaction {}", logContext, current.getIdentifier());
838 final DataTreeCandidateTip candidate;
840 candidate = tip.prepare(cohort.getDataTreeModification());
841 cohort.userPreCommit(candidate);
842 } catch (ExecutionException | TimeoutException | RuntimeException e) {
847 // Set the tip of the data tree.
848 tip = Verify.verifyNotNull(candidate);
850 entry.lastAccess = readTime();
852 pendingTransactions.remove();
853 pendingCommits.add(entry);
855 LOG.debug("{}: Transaction {} prepared", logContext, current.getIdentifier());
857 cohort.successfulPreCommit(candidate);
859 processNextPendingTransaction();
862 private void failCommit(final Exception cause) {
863 shard.getShardMBean().incrementFailedTransactionsCount();
864 pendingFinishCommits.poll().cohort.failedCommit(cause);
865 processNextPending();
868 @SuppressWarnings("checkstyle:IllegalCatch")
869 private void finishCommit(final SimpleShardDataTreeCohort cohort) {
870 final TransactionIdentifier txId = cohort.getIdentifier();
871 final DataTreeCandidate candidate = cohort.getCandidate();
873 LOG.debug("{}: Resuming commit of transaction {}", logContext, txId);
875 if (tip == candidate) {
876 // All pending candidates have been committed, reset the tip to the data tree.
881 dataTree.commit(candidate);
882 } catch (Exception e) {
883 LOG.error("{}: Failed to commit transaction {}", logContext, txId, e);
888 shard.getShardMBean().incrementCommittedTransactionCount();
889 shard.getShardMBean().setLastCommittedTransactionTime(System.currentTimeMillis());
891 // FIXME: propagate journal index
892 pendingFinishCommits.poll().cohort.successfulCommit(UnsignedLong.ZERO);
894 LOG.trace("{}: Transaction {} committed, proceeding to notify", logContext, txId);
895 notifyListeners(candidate);
897 processNextPending();
900 void startCommit(final SimpleShardDataTreeCohort cohort, final DataTreeCandidate candidate) {
901 final CommitEntry entry = pendingCommits.peek();
902 Preconditions.checkState(entry != null, "Attempted to start commit of %s when no transactions pending", cohort);
904 final SimpleShardDataTreeCohort current = entry.cohort;
905 if (!cohort.equals(current)) {
906 LOG.debug("{}: Transaction {} scheduled for commit step", logContext, cohort.getIdentifier());
910 LOG.debug("{}: Starting commit for transaction {}", logContext, current.getIdentifier());
912 final TransactionIdentifier txId = cohort.getIdentifier();
913 final Payload payload;
915 payload = CommitTransactionPayload.create(txId, candidate);
916 } catch (IOException e) {
917 LOG.error("{}: Failed to encode transaction {} candidate {}", logContext, txId, candidate, e);
918 pendingCommits.poll().cohort.failedCommit(e);
919 processNextPending();
923 // We process next transactions pending canCommit before we call persistPayload to possibly progress subsequent
924 // transactions to the COMMIT_PENDING state so the payloads can be batched for replication. This is done for
925 // single-shard transactions that immediately transition from canCommit to preCommit to commit. Note that
926 // if the next pending transaction is progressed to COMMIT_PENDING and this method (startCommit) is called,
927 // the next transaction will not attempt to replicate b/c the current transaction is still at the head of the
928 // pendingCommits queue.
929 processNextPendingTransaction();
931 // After processing next pending transactions, we can now remove the current transaction from pendingCommits.
932 // Note this must be done before the call to peekNextPendingCommit below so we check the next transaction
933 // in order to properly determine the batchHint flag for the call to persistPayload.
934 pendingCommits.remove();
935 pendingFinishCommits.add(entry);
937 // See if the next transaction is pending commit (ie in the COMMIT_PENDING state) so it can be batched with
938 // this transaction for replication.
939 boolean replicationBatchHint = peekNextPendingCommit();
941 // Once completed, we will continue via payloadReplicationComplete
942 shard.persistPayload(txId, payload, replicationBatchHint);
944 entry.lastAccess = shard.ticker().read();
946 LOG.debug("{}: Transaction {} submitted to persistence", logContext, txId);
948 // Process the next transaction pending commit, if any. If there is one it will be batched with this
949 // transaction for replication.
950 processNextPendingCommit();
953 Collection<ActorRef> getCohortActors() {
954 return cohortRegistry.getCohortActors();
957 void processCohortRegistryCommand(final ActorRef sender, final CohortRegistryCommand message) {
958 cohortRegistry.process(sender, message);
962 ShardDataTreeCohort createFailedCohort(final TransactionIdentifier txId, final DataTreeModification mod,
963 final Exception failure) {
964 final SimpleShardDataTreeCohort cohort = new SimpleShardDataTreeCohort(this, mod, txId, failure);
965 pendingTransactions.add(new CommitEntry(cohort, readTime()));
970 ShardDataTreeCohort createReadyCohort(final TransactionIdentifier txId, final DataTreeModification mod) {
971 SimpleShardDataTreeCohort cohort = new SimpleShardDataTreeCohort(this, mod, txId,
972 cohortRegistry.createCohort(schemaContext, txId, COMMIT_STEP_TIMEOUT));
973 pendingTransactions.add(new CommitEntry(cohort, readTime()));
977 // Exposed for ShardCommitCoordinator so it does not have deal with local histories (it does not care), this mimics
978 // the newReadWriteTransaction()
979 ShardDataTreeCohort newReadyCohort(final TransactionIdentifier txId, final DataTreeModification mod) {
980 if (txId.getHistoryId().getHistoryId() == 0) {
981 return createReadyCohort(txId, mod);
984 return ensureTransactionChain(txId.getHistoryId(), null).createReadyCohort(txId, mod);
987 @SuppressFBWarnings(value = "DB_DUPLICATE_SWITCH_CLAUSES", justification = "See inline comments below.")
988 void checkForExpiredTransactions(final long transactionCommitTimeoutMillis,
989 final Function<SimpleShardDataTreeCohort, Optional<Long>> accessTimeUpdater) {
990 final long timeout = TimeUnit.MILLISECONDS.toNanos(transactionCommitTimeoutMillis);
991 final long now = readTime();
993 final Queue<CommitEntry> currentQueue = !pendingFinishCommits.isEmpty() ? pendingFinishCommits :
994 !pendingCommits.isEmpty() ? pendingCommits : pendingTransactions;
995 final CommitEntry currentTx = currentQueue.peek();
996 if (currentTx == null) {
997 // Empty queue, no-op
1001 long delta = now - currentTx.lastAccess;
1002 if (delta < timeout) {
1003 // Not expired yet, bail
1007 final Optional<Long> updateOpt = accessTimeUpdater.apply(currentTx.cohort);
1008 if (updateOpt.isPresent()) {
1009 final long newAccess = updateOpt.get().longValue();
1010 final long newDelta = now - newAccess;
1011 if (newDelta < delta) {
1012 LOG.debug("{}: Updated current transaction {} access time", logContext,
1013 currentTx.cohort.getIdentifier());
1014 currentTx.lastAccess = newAccess;
1018 if (delta < timeout) {
1019 // Not expired yet, bail
1024 final long deltaMillis = TimeUnit.NANOSECONDS.toMillis(delta);
1025 final State state = currentTx.cohort.getState();
1027 LOG.warn("{}: Current transaction {} has timed out after {} ms in state {}", logContext,
1028 currentTx.cohort.getIdentifier(), deltaMillis, state);
1029 boolean processNext = true;
1030 final TimeoutException cohortFailure = new TimeoutException("Backend timeout in state " + state + " after "
1031 + deltaMillis + "ms");
1034 case CAN_COMMIT_PENDING:
1035 currentQueue.remove().cohort.failedCanCommit(cohortFailure);
1037 case CAN_COMMIT_COMPLETE:
1038 // The suppression of the FindBugs "DB_DUPLICATE_SWITCH_CLAUSES" warning pertains to this clause
1039 // whose code is duplicated with PRE_COMMIT_COMPLETE. The clauses aren't combined in case the code
1040 // in PRE_COMMIT_COMPLETE is changed.
1041 currentQueue.remove().cohort.reportFailure(cohortFailure);
1043 case PRE_COMMIT_PENDING:
1044 currentQueue.remove().cohort.failedPreCommit(cohortFailure);
1046 case PRE_COMMIT_COMPLETE:
1047 // FIXME: this is a legacy behavior problem. Three-phase commit protocol specifies that after we
1048 // are ready we should commit the transaction, not abort it. Our current software stack does
1049 // not allow us to do that consistently, because we persist at the time of commit, hence
1050 // we can end up in a state where we have pre-committed a transaction, then a leader failover
1051 // occurred ... the new leader does not see the pre-committed transaction and does not have
1052 // a running timer. To fix this we really need two persistence events.
1054 // The first one, done at pre-commit time will hold the transaction payload. When consensus
1055 // is reached, we exit the pre-commit phase and start the pre-commit timer. Followers do not
1056 // apply the state in this event.
1058 // The second one, done at commit (or abort) time holds only the transaction identifier and
1059 // signals to followers that the state should (or should not) be applied.
1061 // In order to make the pre-commit timer working across failovers, though, we need
1062 // a per-shard cluster-wide monotonic time, so a follower becoming the leader can accurately
1063 // restart the timer.
1064 currentQueue.remove().cohort.reportFailure(cohortFailure);
1066 case COMMIT_PENDING:
1067 LOG.warn("{}: Transaction {} is still committing, cannot abort", logContext,
1068 currentTx.cohort.getIdentifier());
1069 currentTx.lastAccess = now;
1070 processNext = false;
1073 currentQueue.remove().cohort.reportFailure(cohortFailure);
1079 currentQueue.remove();
1083 processNextPending();
1087 boolean startAbort(final SimpleShardDataTreeCohort cohort) {
1088 final Iterator<CommitEntry> it = Iterables.concat(pendingFinishCommits, pendingCommits,
1089 pendingTransactions).iterator();
1090 if (!it.hasNext()) {
1091 LOG.debug("{}: no open transaction while attempting to abort {}", logContext, cohort.getIdentifier());
1095 // First entry is special, as it may already be committing
1096 final CommitEntry first = it.next();
1097 if (cohort.equals(first.cohort)) {
1098 if (cohort.getState() != State.COMMIT_PENDING) {
1099 LOG.debug("{}: aborting head of queue {} in state {}", logContext, cohort.getIdentifier(),
1100 cohort.getIdentifier());
1103 if (cohort.getCandidate() != null) {
1104 rebaseTransactions(it, dataTree);
1107 processNextPending();
1111 LOG.warn("{}: transaction {} is committing, skipping abort", logContext, cohort.getIdentifier());
1115 TipProducingDataTreeTip newTip = MoreObjects.firstNonNull(first.cohort.getCandidate(), dataTree);
1116 while (it.hasNext()) {
1117 final CommitEntry e = it.next();
1118 if (cohort.equals(e.cohort)) {
1119 LOG.debug("{}: aborting queued transaction {}", logContext, cohort.getIdentifier());
1122 if (cohort.getCandidate() != null) {
1123 rebaseTransactions(it, newTip);
1128 newTip = MoreObjects.firstNonNull(e.cohort.getCandidate(), newTip);
1132 LOG.debug("{}: aborted transaction {} not found in the queue", logContext, cohort.getIdentifier());
1136 @SuppressWarnings("checkstyle:IllegalCatch")
1137 private void rebaseTransactions(final Iterator<CommitEntry> iter, @Nonnull final TipProducingDataTreeTip newTip) {
1138 tip = Preconditions.checkNotNull(newTip);
1139 while (iter.hasNext()) {
1140 final SimpleShardDataTreeCohort cohort = iter.next().cohort;
1141 if (cohort.getState() == State.CAN_COMMIT_COMPLETE) {
1142 LOG.debug("{}: Revalidating queued transaction {}", logContext, cohort.getIdentifier());
1145 tip.validate(cohort.getDataTreeModification());
1146 } catch (DataValidationFailedException | RuntimeException e) {
1147 LOG.debug("{}: Failed to revalidate queued transaction {}", logContext, cohort.getIdentifier(), e);
1148 cohort.reportFailure(e);
1150 } else if (cohort.getState() == State.PRE_COMMIT_COMPLETE) {
1151 LOG.debug("{}: Repreparing queued transaction {}", logContext, cohort.getIdentifier());
1154 tip.validate(cohort.getDataTreeModification());
1155 DataTreeCandidateTip candidate = tip.prepare(cohort.getDataTreeModification());
1156 cohort.userPreCommit(candidate);
1158 cohort.setNewCandidate(candidate);
1160 } catch (ExecutionException | TimeoutException | RuntimeException | DataValidationFailedException e) {
1161 LOG.debug("{}: Failed to reprepare queued transaction {}", logContext, cohort.getIdentifier(), e);
1162 cohort.reportFailure(e);
1168 void setRunOnPendingTransactionsComplete(final Runnable operation) {
1169 runOnPendingTransactionsComplete = operation;
1170 maybeRunOperationOnPendingTransactionsComplete();
1173 private void maybeRunOperationOnPendingTransactionsComplete() {
1174 if (runOnPendingTransactionsComplete != null && !anyPendingTransactions()) {
1175 LOG.debug("{}: Pending transactions complete - running operation {}", logContext,
1176 runOnPendingTransactionsComplete);
1178 runOnPendingTransactionsComplete.run();
1179 runOnPendingTransactionsComplete = null;
1183 ShardStats getStats() {
1184 return shard.getShardMBean();
1187 Iterator<SimpleShardDataTreeCohort> cohortIterator() {
1188 return Iterables.transform(Iterables.concat(pendingFinishCommits, pendingCommits, pendingTransactions),
1189 e -> e.cohort).iterator();
1192 void removeTransactionChain(final LocalHistoryIdentifier id) {
1193 if (transactionChains.remove(id) != null) {
1194 LOG.debug("{}: Removed transaction chain {}", logContext, id);