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.UnaryOperator;
39 import javax.annotation.Nonnull;
40 import javax.annotation.Nullable;
41 import javax.annotation.concurrent.NotThreadSafe;
42 import org.opendaylight.controller.cluster.access.concepts.LocalHistoryIdentifier;
43 import org.opendaylight.controller.cluster.access.concepts.TransactionIdentifier;
44 import org.opendaylight.controller.cluster.datastore.DataTreeCohortActorRegistry.CohortRegistryCommand;
45 import org.opendaylight.controller.cluster.datastore.ShardDataTreeCohort.State;
46 import org.opendaylight.controller.cluster.datastore.jmx.mbeans.shard.ShardStats;
47 import org.opendaylight.controller.cluster.datastore.persisted.AbortTransactionPayload;
48 import org.opendaylight.controller.cluster.datastore.persisted.AbstractIdentifiablePayload;
49 import org.opendaylight.controller.cluster.datastore.persisted.CloseLocalHistoryPayload;
50 import org.opendaylight.controller.cluster.datastore.persisted.CommitTransactionPayload;
51 import org.opendaylight.controller.cluster.datastore.persisted.CreateLocalHistoryPayload;
52 import org.opendaylight.controller.cluster.datastore.persisted.MetadataShardDataTreeSnapshot;
53 import org.opendaylight.controller.cluster.datastore.persisted.PurgeLocalHistoryPayload;
54 import org.opendaylight.controller.cluster.datastore.persisted.PurgeTransactionPayload;
55 import org.opendaylight.controller.cluster.datastore.persisted.ShardDataTreeSnapshot;
56 import org.opendaylight.controller.cluster.datastore.persisted.ShardDataTreeSnapshotMetadata;
57 import org.opendaylight.controller.cluster.datastore.utils.DataTreeModificationOutput;
58 import org.opendaylight.controller.cluster.datastore.utils.PruningDataTreeModification;
59 import org.opendaylight.controller.cluster.raft.protobuff.client.messages.Payload;
60 import org.opendaylight.controller.md.sal.common.api.data.AsyncDataBroker.DataChangeScope;
61 import org.opendaylight.controller.md.sal.common.api.data.AsyncDataChangeListener;
62 import org.opendaylight.controller.md.sal.common.api.data.OptimisticLockFailedException;
63 import org.opendaylight.controller.md.sal.common.api.data.TransactionCommitFailedException;
64 import org.opendaylight.controller.md.sal.dom.api.DOMDataTreeChangeListener;
65 import org.opendaylight.yangtools.concepts.Identifier;
66 import org.opendaylight.yangtools.concepts.ListenerRegistration;
67 import org.opendaylight.yangtools.yang.data.api.YangInstanceIdentifier;
68 import org.opendaylight.yangtools.yang.data.api.schema.NormalizedNode;
69 import org.opendaylight.yangtools.yang.data.api.schema.tree.ConflictingModificationAppliedException;
70 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeCandidate;
71 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeCandidateTip;
72 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeCandidates;
73 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeModification;
74 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeSnapshot;
75 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeTip;
76 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataValidationFailedException;
77 import org.opendaylight.yangtools.yang.data.api.schema.tree.TipProducingDataTree;
78 import org.opendaylight.yangtools.yang.data.api.schema.tree.TipProducingDataTreeTip;
79 import org.opendaylight.yangtools.yang.data.api.schema.tree.TreeType;
80 import org.opendaylight.yangtools.yang.data.impl.schema.tree.InMemoryDataTreeFactory;
81 import org.opendaylight.yangtools.yang.model.api.SchemaContext;
82 import org.slf4j.Logger;
83 import org.slf4j.LoggerFactory;
84 import scala.concurrent.duration.Duration;
87 * Internal shard state, similar to a DOMStore, but optimized for use in the actor system,
88 * e.g. it does not expose public interfaces and assumes it is only ever called from a
92 * This class is not part of the API contract and is subject to change at any time.
95 public class ShardDataTree extends ShardDataTreeTransactionParent {
96 private static final class CommitEntry {
97 final SimpleShardDataTreeCohort cohort;
100 CommitEntry(final SimpleShardDataTreeCohort cohort, final long now) {
101 this.cohort = Preconditions.checkNotNull(cohort);
106 private static final Timeout COMMIT_STEP_TIMEOUT = new Timeout(Duration.create(5, TimeUnit.SECONDS));
107 private static final Logger LOG = LoggerFactory.getLogger(ShardDataTree.class);
110 * Process this many transactions in a single batched run. If we exceed this limit, we need to schedule later
111 * execution to finish up the batch. This is necessary in case of a long list of transactions which progress
112 * immediately through their preCommit phase -- if that happens, their completion eats up stack frames and could
113 * result in StackOverflowError.
115 private static final int MAX_TRANSACTION_BATCH = 100;
117 private final Map<LocalHistoryIdentifier, ShardDataTreeTransactionChain> transactionChains = new HashMap<>();
118 private final DataTreeCohortActorRegistry cohortRegistry = new DataTreeCohortActorRegistry();
119 private final Queue<CommitEntry> pendingTransactions = new ArrayDeque<>();
120 private final Queue<CommitEntry> pendingCommits = new ArrayDeque<>();
121 private final Queue<CommitEntry> pendingFinishCommits = new ArrayDeque<>();
124 * Callbacks that need to be invoked once a payload is replicated.
126 private final Map<Payload, Runnable> replicationCallbacks = new HashMap<>();
128 private final ShardDataTreeChangeListenerPublisher treeChangeListenerPublisher;
129 private final ShardDataChangeListenerPublisher dataChangeListenerPublisher;
130 private final Collection<ShardDataTreeMetadata<?>> metadata;
131 private final TipProducingDataTree dataTree;
132 private final String logContext;
133 private final Shard shard;
134 private Runnable runOnPendingTransactionsComplete;
137 * Optimistic {@link DataTreeCandidate} preparation. Since our DataTree implementation is a
138 * {@link TipProducingDataTree}, each {@link DataTreeCandidate} is also a {@link DataTreeTip}, e.g. another
139 * candidate can be prepared on top of it. They still need to be committed in sequence. Here we track the current
140 * tip of the data tree, which is the last DataTreeCandidate we have in flight, or the DataTree itself.
142 private TipProducingDataTreeTip tip;
144 private SchemaContext schemaContext;
146 private int currentTransactionBatch;
148 ShardDataTree(final Shard shard, final SchemaContext schemaContext, final TipProducingDataTree dataTree,
149 final ShardDataTreeChangeListenerPublisher treeChangeListenerPublisher,
150 final ShardDataChangeListenerPublisher dataChangeListenerPublisher, final String logContext,
151 final ShardDataTreeMetadata<?>... metadata) {
152 this.dataTree = Preconditions.checkNotNull(dataTree);
153 updateSchemaContext(schemaContext);
155 this.shard = Preconditions.checkNotNull(shard);
156 this.treeChangeListenerPublisher = Preconditions.checkNotNull(treeChangeListenerPublisher);
157 this.dataChangeListenerPublisher = Preconditions.checkNotNull(dataChangeListenerPublisher);
158 this.logContext = Preconditions.checkNotNull(logContext);
159 this.metadata = ImmutableList.copyOf(metadata);
163 ShardDataTree(final Shard shard, final SchemaContext schemaContext, final TreeType treeType,
164 final YangInstanceIdentifier root,
165 final ShardDataTreeChangeListenerPublisher treeChangeListenerPublisher,
166 final ShardDataChangeListenerPublisher dataChangeListenerPublisher, final String logContext,
167 final ShardDataTreeMetadata<?>... metadata) {
168 this(shard, schemaContext, InMemoryDataTreeFactory.getInstance().create(treeType, root),
169 treeChangeListenerPublisher, dataChangeListenerPublisher, logContext, metadata);
173 public ShardDataTree(final Shard shard, final SchemaContext schemaContext, final TreeType treeType) {
174 this(shard, schemaContext, treeType, YangInstanceIdentifier.EMPTY,
175 new DefaultShardDataTreeChangeListenerPublisher(""),
176 new DefaultShardDataChangeListenerPublisher(""), "");
179 final String logContext() {
183 final long readTime() {
184 return shard.ticker().read();
187 public TipProducingDataTree getDataTree() {
191 SchemaContext getSchemaContext() {
192 return schemaContext;
195 void updateSchemaContext(final SchemaContext newSchemaContext) {
196 dataTree.setSchemaContext(newSchemaContext);
197 this.schemaContext = Preconditions.checkNotNull(newSchemaContext);
200 void resetTransactionBatch() {
201 currentTransactionBatch = 0;
205 * Take a snapshot of current state for later recovery.
207 * @return A state snapshot
209 @Nonnull ShardDataTreeSnapshot takeStateSnapshot() {
210 final NormalizedNode<?, ?> rootNode = dataTree.takeSnapshot().readNode(YangInstanceIdentifier.EMPTY).get();
211 final Builder<Class<? extends ShardDataTreeSnapshotMetadata<?>>, ShardDataTreeSnapshotMetadata<?>> metaBuilder =
212 ImmutableMap.builder();
214 for (ShardDataTreeMetadata<?> m : metadata) {
215 final ShardDataTreeSnapshotMetadata<?> meta = m.toSnapshot();
217 metaBuilder.put(meta.getType(), meta);
221 return new MetadataShardDataTreeSnapshot(rootNode, metaBuilder.build());
224 private boolean anyPendingTransactions() {
225 return !pendingTransactions.isEmpty() || !pendingCommits.isEmpty() || !pendingFinishCommits.isEmpty();
228 private void applySnapshot(@Nonnull final ShardDataTreeSnapshot snapshot,
229 final UnaryOperator<DataTreeModification> wrapper) throws DataValidationFailedException {
230 final Stopwatch elapsed = Stopwatch.createStarted();
232 if (anyPendingTransactions()) {
233 LOG.warn("{}: applying state snapshot with pending transactions", logContext);
236 final Map<Class<? extends ShardDataTreeSnapshotMetadata<?>>, ShardDataTreeSnapshotMetadata<?>> snapshotMeta;
237 if (snapshot instanceof MetadataShardDataTreeSnapshot) {
238 snapshotMeta = ((MetadataShardDataTreeSnapshot) snapshot).getMetadata();
240 snapshotMeta = ImmutableMap.of();
243 for (ShardDataTreeMetadata<?> m : metadata) {
244 final ShardDataTreeSnapshotMetadata<?> s = snapshotMeta.get(m.getSupportedType());
252 final DataTreeModification mod = wrapper.apply(dataTree.takeSnapshot().newModification());
253 // delete everything first
254 mod.delete(YangInstanceIdentifier.EMPTY);
256 final java.util.Optional<NormalizedNode<?, ?>> maybeNode = snapshot.getRootNode();
257 if (maybeNode.isPresent()) {
258 // Add everything from the remote node back
259 mod.write(YangInstanceIdentifier.EMPTY, maybeNode.get());
263 final DataTreeModification unwrapped = unwrap(mod);
264 dataTree.validate(unwrapped);
265 DataTreeCandidateTip candidate = dataTree.prepare(unwrapped);
266 dataTree.commit(candidate);
267 notifyListeners(candidate);
269 LOG.debug("{}: state snapshot applied in {}", logContext, elapsed);
273 * Apply a snapshot coming from the leader. This method assumes the leader and follower SchemaContexts match and
274 * does not perform any pruning.
276 * @param snapshot Snapshot that needs to be applied
277 * @throws DataValidationFailedException when the snapshot fails to apply
279 void applySnapshot(@Nonnull final ShardDataTreeSnapshot snapshot) throws DataValidationFailedException {
280 applySnapshot(snapshot, UnaryOperator.identity());
283 private PruningDataTreeModification wrapWithPruning(final DataTreeModification delegate) {
284 return new PruningDataTreeModification(delegate, dataTree, schemaContext);
287 private static DataTreeModification unwrap(final DataTreeModification modification) {
288 if (modification instanceof PruningDataTreeModification) {
289 return ((PruningDataTreeModification)modification).delegate();
295 * Apply a snapshot coming from recovery. This method does not assume the SchemaContexts match and performs data
296 * pruning in an attempt to adjust the state to our current SchemaContext.
298 * @param snapshot Snapshot that needs to be applied
299 * @throws DataValidationFailedException when the snapshot fails to apply
301 void applyRecoverySnapshot(final @Nonnull ShardDataTreeSnapshot snapshot) throws DataValidationFailedException {
302 applySnapshot(snapshot, this::wrapWithPruning);
305 @SuppressWarnings("checkstyle:IllegalCatch")
306 private void applyRecoveryCandidate(final DataTreeCandidate candidate) throws DataValidationFailedException {
307 final PruningDataTreeModification mod = wrapWithPruning(dataTree.takeSnapshot().newModification());
308 DataTreeCandidates.applyToModification(mod, candidate);
311 final DataTreeModification unwrapped = mod.delegate();
312 LOG.trace("{}: Applying recovery modification {}", logContext, unwrapped);
315 dataTree.validate(unwrapped);
316 dataTree.commit(dataTree.prepare(unwrapped));
317 } catch (Exception e) {
318 File file = new File(System.getProperty("karaf.data", "."),
319 "failed-recovery-payload-" + logContext + ".out");
320 DataTreeModificationOutput.toFile(file, unwrapped);
321 throw new IllegalStateException(String.format(
322 "%s: Failed to apply recovery payload. Modification data was written to file %s",
323 logContext, file), e);
328 * Apply a payload coming from recovery. This method does not assume the SchemaContexts match and performs data
329 * pruning in an attempt to adjust the state to our current SchemaContext.
331 * @param payload Payload
332 * @throws IOException when the snapshot fails to deserialize
333 * @throws DataValidationFailedException when the snapshot fails to apply
335 void applyRecoveryPayload(final @Nonnull Payload payload) throws IOException, DataValidationFailedException {
336 if (payload instanceof CommitTransactionPayload) {
337 final Entry<TransactionIdentifier, DataTreeCandidate> e =
338 ((CommitTransactionPayload) payload).getCandidate();
339 applyRecoveryCandidate(e.getValue());
340 allMetadataCommittedTransaction(e.getKey());
341 } else if (payload instanceof AbortTransactionPayload) {
342 allMetadataAbortedTransaction(((AbortTransactionPayload) payload).getIdentifier());
343 } else if (payload instanceof PurgeTransactionPayload) {
344 allMetadataPurgedTransaction(((PurgeTransactionPayload) payload).getIdentifier());
345 } else if (payload instanceof CreateLocalHistoryPayload) {
346 allMetadataCreatedLocalHistory(((CreateLocalHistoryPayload) payload).getIdentifier());
347 } else if (payload instanceof CloseLocalHistoryPayload) {
348 allMetadataClosedLocalHistory(((CloseLocalHistoryPayload) payload).getIdentifier());
349 } else if (payload instanceof PurgeLocalHistoryPayload) {
350 allMetadataPurgedLocalHistory(((PurgeLocalHistoryPayload) payload).getIdentifier());
352 LOG.debug("{}: ignoring unhandled payload {}", logContext, payload);
356 private void applyReplicatedCandidate(final Identifier identifier, final DataTreeCandidate foreign)
357 throws DataValidationFailedException {
358 LOG.debug("{}: Applying foreign transaction {}", logContext, identifier);
360 final DataTreeModification mod = dataTree.takeSnapshot().newModification();
361 DataTreeCandidates.applyToModification(mod, foreign);
364 LOG.trace("{}: Applying foreign modification {}", logContext, mod);
365 dataTree.validate(mod);
366 final DataTreeCandidate candidate = dataTree.prepare(mod);
367 dataTree.commit(candidate);
369 notifyListeners(candidate);
373 * Apply a payload coming from the leader, which could actually be us. This method assumes the leader and follower
374 * SchemaContexts match and does not perform any pruning.
376 * @param identifier Payload identifier as returned from RaftActor
377 * @param payload Payload
378 * @throws IOException when the snapshot fails to deserialize
379 * @throws DataValidationFailedException when the snapshot fails to apply
381 void applyReplicatedPayload(final Identifier identifier, final Payload payload) throws IOException,
382 DataValidationFailedException {
384 * 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
385 * if we are the leader and it has originated with us.
387 * The identifier will only ever be non-null when we were the leader which achieved consensus. Unfortunately,
388 * though, this may not be the case anymore, as we are being called some time afterwards and we may not be
389 * acting in that capacity anymore.
391 * In any case, we know that this is an entry coming from replication, hence we can be sure we will not observe
392 * pre-Boron state -- which limits the number of options here.
394 if (payload instanceof CommitTransactionPayload) {
395 final TransactionIdentifier txId;
396 if (identifier == null) {
397 final Entry<TransactionIdentifier, DataTreeCandidate> e =
398 ((CommitTransactionPayload) payload).getCandidate();
400 applyReplicatedCandidate(txId, e.getValue());
402 Verify.verify(identifier instanceof TransactionIdentifier);
403 txId = (TransactionIdentifier) identifier;
404 payloadReplicationComplete(txId);
406 allMetadataCommittedTransaction(txId);
407 } else if (payload instanceof AbortTransactionPayload) {
408 if (identifier != null) {
409 payloadReplicationComplete((AbortTransactionPayload) payload);
411 allMetadataAbortedTransaction(((AbortTransactionPayload) payload).getIdentifier());
412 } else if (payload instanceof PurgeTransactionPayload) {
413 if (identifier != null) {
414 payloadReplicationComplete((PurgeTransactionPayload) payload);
416 allMetadataPurgedTransaction(((PurgeTransactionPayload) payload).getIdentifier());
417 } else if (payload instanceof CloseLocalHistoryPayload) {
418 if (identifier != null) {
419 payloadReplicationComplete((CloseLocalHistoryPayload) payload);
421 allMetadataClosedLocalHistory(((CloseLocalHistoryPayload) payload).getIdentifier());
422 } else if (payload instanceof CreateLocalHistoryPayload) {
423 if (identifier != null) {
424 payloadReplicationComplete((CreateLocalHistoryPayload)payload);
426 allMetadataCreatedLocalHistory(((CreateLocalHistoryPayload) payload).getIdentifier());
427 } else if (payload instanceof PurgeLocalHistoryPayload) {
428 if (identifier != null) {
429 payloadReplicationComplete((PurgeLocalHistoryPayload)payload);
431 allMetadataPurgedLocalHistory(((PurgeLocalHistoryPayload) payload).getIdentifier());
433 LOG.warn("{}: ignoring unhandled identifier {} payload {}", logContext, identifier, payload);
437 private void replicatePayload(final Identifier id, final Payload payload, @Nullable final Runnable callback) {
438 if (callback != null) {
439 replicationCallbacks.put(payload, callback);
441 shard.persistPayload(id, payload, true);
444 private void payloadReplicationComplete(final AbstractIdentifiablePayload<?> payload) {
445 final Runnable callback = replicationCallbacks.remove(payload);
446 if (callback != null) {
447 LOG.debug("{}: replication of {} completed, invoking {}", logContext, payload.getIdentifier(), callback);
450 LOG.debug("{}: replication of {} has no callback", logContext, payload.getIdentifier());
454 private void payloadReplicationComplete(final TransactionIdentifier txId) {
455 final CommitEntry current = pendingFinishCommits.peek();
456 if (current == null) {
457 LOG.warn("{}: No outstanding transactions, ignoring consensus on transaction {}", logContext, txId);
461 if (!current.cohort.getIdentifier().equals(txId)) {
462 LOG.debug("{}: Head of pendingFinishCommits queue is {}, ignoring consensus on transaction {}", logContext,
463 current.cohort.getIdentifier(), txId);
467 finishCommit(current.cohort);
470 private void allMetadataAbortedTransaction(final TransactionIdentifier txId) {
471 for (ShardDataTreeMetadata<?> m : metadata) {
472 m.onTransactionAborted(txId);
476 private void allMetadataCommittedTransaction(final TransactionIdentifier txId) {
477 for (ShardDataTreeMetadata<?> m : metadata) {
478 m.onTransactionCommitted(txId);
482 private void allMetadataPurgedTransaction(final TransactionIdentifier txId) {
483 for (ShardDataTreeMetadata<?> m : metadata) {
484 m.onTransactionPurged(txId);
488 private void allMetadataCreatedLocalHistory(final LocalHistoryIdentifier historyId) {
489 for (ShardDataTreeMetadata<?> m : metadata) {
490 m.onHistoryCreated(historyId);
494 private void allMetadataClosedLocalHistory(final LocalHistoryIdentifier historyId) {
495 for (ShardDataTreeMetadata<?> m : metadata) {
496 m.onHistoryClosed(historyId);
500 private void allMetadataPurgedLocalHistory(final LocalHistoryIdentifier historyId) {
501 for (ShardDataTreeMetadata<?> m : metadata) {
502 m.onHistoryPurged(historyId);
507 * Create a transaction chain for specified history. Unlike {@link #ensureTransactionChain(LocalHistoryIdentifier)},
508 * this method is used for re-establishing state when we are taking over
510 * @param historyId Local history identifier
511 * @param closed True if the chain should be created in closed state (i.e. pending purge)
512 * @return Transaction chain handle
514 ShardDataTreeTransactionChain recreateTransactionChain(final LocalHistoryIdentifier historyId,
515 final boolean closed) {
516 final ShardDataTreeTransactionChain ret = new ShardDataTreeTransactionChain(historyId, this);
517 final ShardDataTreeTransactionChain existing = transactionChains.putIfAbsent(historyId, ret);
518 Preconditions.checkState(existing == null, "Attempted to recreate chain %s, but %s already exists", historyId,
523 ShardDataTreeTransactionChain ensureTransactionChain(final LocalHistoryIdentifier historyId,
524 @Nullable final Runnable callback) {
525 ShardDataTreeTransactionChain chain = transactionChains.get(historyId);
527 chain = new ShardDataTreeTransactionChain(historyId, this);
528 transactionChains.put(historyId, chain);
529 replicatePayload(historyId, CreateLocalHistoryPayload.create(historyId), callback);
530 } else if (callback != null) {
537 ReadOnlyShardDataTreeTransaction newReadOnlyTransaction(final TransactionIdentifier txId) {
538 if (txId.getHistoryId().getHistoryId() == 0) {
539 return new ReadOnlyShardDataTreeTransaction(this, txId, dataTree.takeSnapshot());
542 return ensureTransactionChain(txId.getHistoryId(), null).newReadOnlyTransaction(txId);
545 ReadWriteShardDataTreeTransaction newReadWriteTransaction(final TransactionIdentifier txId) {
546 if (txId.getHistoryId().getHistoryId() == 0) {
547 return new ReadWriteShardDataTreeTransaction(ShardDataTree.this, txId, dataTree.takeSnapshot()
551 return ensureTransactionChain(txId.getHistoryId(), null).newReadWriteTransaction(txId);
555 public void notifyListeners(final DataTreeCandidate candidate) {
556 treeChangeListenerPublisher.publishChanges(candidate);
557 dataChangeListenerPublisher.publishChanges(candidate);
561 * Immediately purge all state relevant to leader. This includes all transaction chains and any scheduled
562 * replication callbacks.
564 void purgeLeaderState() {
565 for (ShardDataTreeTransactionChain chain : transactionChains.values()) {
569 transactionChains.clear();
570 replicationCallbacks.clear();
574 * Close a single transaction chain.
576 * @param id History identifier
577 * @param callback Callback to invoke upon completion, may be null
579 void closeTransactionChain(final LocalHistoryIdentifier id, @Nullable final Runnable callback) {
580 final ShardDataTreeTransactionChain chain = transactionChains.get(id);
582 LOG.debug("{}: Closing non-existent transaction chain {}", logContext, id);
583 if (callback != null) {
590 replicatePayload(id, CloseLocalHistoryPayload.create(id), callback);
594 * Purge a single transaction chain.
596 * @param id History identifier
597 * @param callback Callback to invoke upon completion, may be null
599 void purgeTransactionChain(final LocalHistoryIdentifier id, @Nullable final Runnable callback) {
600 final ShardDataTreeTransactionChain chain = transactionChains.remove(id);
602 LOG.debug("{}: Purging non-existent transaction chain {}", logContext, id);
603 if (callback != null) {
609 replicatePayload(id, PurgeLocalHistoryPayload.create(id), callback);
612 void registerDataChangeListener(final YangInstanceIdentifier path,
613 final AsyncDataChangeListener<YangInstanceIdentifier, NormalizedNode<?, ?>> listener,
614 final DataChangeScope scope, final Optional<DataTreeCandidate> initialState,
615 final Consumer<ListenerRegistration<AsyncDataChangeListener<YangInstanceIdentifier, NormalizedNode<?, ?>>>>
617 dataChangeListenerPublisher.registerDataChangeListener(path, listener, scope, initialState, onRegistration);
620 Optional<DataTreeCandidate> readCurrentData() {
621 final Optional<NormalizedNode<?, ?>> currentState =
622 dataTree.takeSnapshot().readNode(YangInstanceIdentifier.EMPTY);
623 return currentState.isPresent() ? Optional.of(DataTreeCandidates.fromNormalizedNode(
624 YangInstanceIdentifier.EMPTY, currentState.get())) : Optional.<DataTreeCandidate>absent();
627 public void registerTreeChangeListener(final YangInstanceIdentifier path, final DOMDataTreeChangeListener listener,
628 final Optional<DataTreeCandidate> initialState,
629 final Consumer<ListenerRegistration<DOMDataTreeChangeListener>> onRegistration) {
630 treeChangeListenerPublisher.registerTreeChangeListener(path, listener, initialState, onRegistration);
634 return pendingTransactions.size() + pendingCommits.size() + pendingFinishCommits.size();
638 void abortTransaction(final AbstractShardDataTreeTransaction<?> transaction, final Runnable callback) {
639 final TransactionIdentifier id = transaction.getIdentifier();
640 LOG.debug("{}: aborting transaction {}", logContext, id);
641 replicatePayload(id, AbortTransactionPayload.create(id), callback);
645 void abortFromTransactionActor(final AbstractShardDataTreeTransaction<?> transaction) {
646 // No-op for free-standing transactions
651 ShardDataTreeCohort finishTransaction(final ReadWriteShardDataTreeTransaction transaction) {
652 final DataTreeModification snapshot = transaction.getSnapshot();
655 return createReadyCohort(transaction.getIdentifier(), snapshot);
658 void purgeTransaction(final TransactionIdentifier id, final Runnable callback) {
659 LOG.debug("{}: purging transaction {}", logContext, id);
660 replicatePayload(id, PurgeTransactionPayload.create(id), callback);
663 public Optional<NormalizedNode<?, ?>> readNode(final YangInstanceIdentifier path) {
664 return dataTree.takeSnapshot().readNode(path);
667 DataTreeSnapshot takeSnapshot() {
668 return dataTree.takeSnapshot();
672 public DataTreeModification newModification() {
673 return dataTree.takeSnapshot().newModification();
677 * Commits a modification.
679 * @deprecated This method violates DataTree containment and will be removed.
683 public DataTreeCandidate commit(final DataTreeModification modification) throws DataValidationFailedException {
684 // Direct modification commit is a utility, which cannot be used while we have transactions in-flight
685 Preconditions.checkState(tip == dataTree, "Cannot modify data tree while transacgitons are pending");
687 modification.ready();
688 dataTree.validate(modification);
689 DataTreeCandidate candidate = dataTree.prepare(modification);
690 dataTree.commit(candidate);
694 public Collection<ShardDataTreeCohort> getAndClearPendingTransactions() {
695 Collection<ShardDataTreeCohort> ret = new ArrayList<>(getQueueSize());
697 for (CommitEntry entry: pendingFinishCommits) {
698 ret.add(entry.cohort);
701 for (CommitEntry entry: pendingCommits) {
702 ret.add(entry.cohort);
705 for (CommitEntry entry: pendingTransactions) {
706 ret.add(entry.cohort);
709 pendingFinishCommits.clear();
710 pendingCommits.clear();
711 pendingTransactions.clear();
717 * Called some time after {@link #processNextPendingTransaction()} decides to stop processing.
719 void resumeNextPendingTransaction() {
720 LOG.debug("{}: attempting to resume transaction processing", logContext);
721 processNextPending();
724 @SuppressWarnings("checkstyle:IllegalCatch")
725 private void processNextPendingTransaction() {
726 ++currentTransactionBatch;
727 if (currentTransactionBatch > MAX_TRANSACTION_BATCH) {
728 LOG.debug("{}: Already processed {}, scheduling continuation", logContext, currentTransactionBatch);
729 shard.scheduleNextPendingTransaction();
733 processNextPending(pendingTransactions, State.CAN_COMMIT_PENDING, entry -> {
734 final SimpleShardDataTreeCohort cohort = entry.cohort;
735 final DataTreeModification modification = cohort.getDataTreeModification();
737 LOG.debug("{}: Validating transaction {}", logContext, cohort.getIdentifier());
740 cohort.throwCanCommitFailure();
742 tip.validate(modification);
743 LOG.debug("{}: Transaction {} validated", logContext, cohort.getIdentifier());
744 cohort.successfulCanCommit();
745 entry.lastAccess = readTime();
747 } catch (ConflictingModificationAppliedException e) {
748 LOG.warn("{}: Store Tx {}: Conflicting modification for path {}.", logContext, cohort.getIdentifier(),
750 cause = new OptimisticLockFailedException("Optimistic lock failed.", e);
751 } catch (DataValidationFailedException e) {
752 LOG.warn("{}: Store Tx {}: Data validation failed for path {}.", logContext, cohort.getIdentifier(),
755 // For debugging purposes, allow dumping of the modification. Coupled with the above
756 // precondition log, it should allow us to understand what went on.
757 LOG.debug("{}: Store Tx {}: modifications: {} tree: {}", cohort.getIdentifier(), modification,
759 cause = new TransactionCommitFailedException("Data did not pass validation.", e);
760 } catch (Exception e) {
761 LOG.warn("{}: Unexpected failure in validation phase", logContext, e);
765 // Failure path: propagate the failure, remove the transaction from the queue and loop to the next one
766 pendingTransactions.poll().cohort.failedCanCommit(cause);
770 private void processNextPending() {
771 processNextPendingCommit();
772 processNextPendingTransaction();
775 private void processNextPending(final Queue<CommitEntry> queue, final State allowedState,
776 final Consumer<CommitEntry> processor) {
777 while (!queue.isEmpty()) {
778 final CommitEntry entry = queue.peek();
779 final SimpleShardDataTreeCohort cohort = entry.cohort;
781 if (cohort.isFailed()) {
782 LOG.debug("{}: Removing failed transaction {}", logContext, cohort.getIdentifier());
787 if (cohort.getState() == allowedState) {
788 processor.accept(entry);
794 maybeRunOperationOnPendingTransactionsComplete();
797 private void processNextPendingCommit() {
798 processNextPending(pendingCommits, State.COMMIT_PENDING,
799 entry -> startCommit(entry.cohort, entry.cohort.getCandidate()));
802 private boolean peekNextPendingCommit() {
803 final CommitEntry first = pendingCommits.peek();
804 return first != null && first.cohort.getState() == State.COMMIT_PENDING;
807 void startCanCommit(final SimpleShardDataTreeCohort cohort) {
808 final SimpleShardDataTreeCohort current = pendingTransactions.peek().cohort;
809 if (!cohort.equals(current)) {
810 LOG.debug("{}: Transaction {} scheduled for canCommit step", logContext, cohort.getIdentifier());
814 processNextPendingTransaction();
817 private void failPreCommit(final Exception cause) {
818 shard.getShardMBean().incrementFailedTransactionsCount();
819 pendingTransactions.poll().cohort.failedPreCommit(cause);
820 processNextPendingTransaction();
823 @SuppressWarnings("checkstyle:IllegalCatch")
824 void startPreCommit(final SimpleShardDataTreeCohort cohort) {
825 final CommitEntry entry = pendingTransactions.peek();
826 Preconditions.checkState(entry != null, "Attempted to pre-commit of %s when no transactions pending", cohort);
828 final SimpleShardDataTreeCohort current = entry.cohort;
829 Verify.verify(cohort.equals(current), "Attempted to pre-commit %s while %s is pending", cohort, current);
831 LOG.debug("{}: Preparing transaction {}", logContext, current.getIdentifier());
833 final DataTreeCandidateTip candidate;
835 candidate = tip.prepare(cohort.getDataTreeModification());
836 cohort.userPreCommit(candidate);
837 } catch (ExecutionException | TimeoutException | RuntimeException e) {
842 // Set the tip of the data tree.
843 tip = Verify.verifyNotNull(candidate);
845 entry.lastAccess = readTime();
847 pendingTransactions.remove();
848 pendingCommits.add(entry);
850 LOG.debug("{}: Transaction {} prepared", logContext, current.getIdentifier());
852 cohort.successfulPreCommit(candidate);
854 processNextPendingTransaction();
857 private void failCommit(final Exception cause) {
858 shard.getShardMBean().incrementFailedTransactionsCount();
859 pendingFinishCommits.poll().cohort.failedCommit(cause);
860 processNextPending();
863 @SuppressWarnings("checkstyle:IllegalCatch")
864 private void finishCommit(final SimpleShardDataTreeCohort cohort) {
865 final TransactionIdentifier txId = cohort.getIdentifier();
866 final DataTreeCandidate candidate = cohort.getCandidate();
868 LOG.debug("{}: Resuming commit of transaction {}", logContext, txId);
870 if (tip == candidate) {
871 // All pending candidates have been committed, reset the tip to the data tree.
876 dataTree.commit(candidate);
877 } catch (Exception e) {
878 LOG.error("{}: Failed to commit transaction {}", logContext, txId, e);
883 shard.getShardMBean().incrementCommittedTransactionCount();
884 shard.getShardMBean().setLastCommittedTransactionTime(System.currentTimeMillis());
886 // FIXME: propagate journal index
887 pendingFinishCommits.poll().cohort.successfulCommit(UnsignedLong.ZERO);
889 LOG.trace("{}: Transaction {} committed, proceeding to notify", logContext, txId);
890 notifyListeners(candidate);
892 processNextPending();
895 void startCommit(final SimpleShardDataTreeCohort cohort, final DataTreeCandidate candidate) {
896 final CommitEntry entry = pendingCommits.peek();
897 Preconditions.checkState(entry != null, "Attempted to start commit of %s when no transactions pending", cohort);
899 final SimpleShardDataTreeCohort current = entry.cohort;
900 if (!cohort.equals(current)) {
901 LOG.debug("{}: Transaction {} scheduled for commit step", logContext, cohort.getIdentifier());
905 LOG.debug("{}: Starting commit for transaction {}", logContext, current.getIdentifier());
907 final TransactionIdentifier txId = cohort.getIdentifier();
908 final Payload payload;
910 payload = CommitTransactionPayload.create(txId, candidate);
911 } catch (IOException e) {
912 LOG.error("{}: Failed to encode transaction {} candidate {}", logContext, txId, candidate, e);
913 pendingCommits.poll().cohort.failedCommit(e);
914 processNextPending();
918 // We process next transactions pending canCommit before we call persistPayload to possibly progress subsequent
919 // transactions to the COMMIT_PENDING state so the payloads can be batched for replication. This is done for
920 // single-shard transactions that immediately transition from canCommit to preCommit to commit. Note that
921 // if the next pending transaction is progressed to COMMIT_PENDING and this method (startCommit) is called,
922 // the next transaction will not attempt to replicate b/c the current transaction is still at the head of the
923 // pendingCommits queue.
924 processNextPendingTransaction();
926 // After processing next pending transactions, we can now remove the current transaction from pendingCommits.
927 // Note this must be done before the call to peekNextPendingCommit below so we check the next transaction
928 // in order to properly determine the batchHint flag for the call to persistPayload.
929 pendingCommits.remove();
930 pendingFinishCommits.add(entry);
932 // See if the next transaction is pending commit (ie in the COMMIT_PENDING state) so it can be batched with
933 // this transaction for replication.
934 boolean replicationBatchHint = peekNextPendingCommit();
936 // Once completed, we will continue via payloadReplicationComplete
937 shard.persistPayload(txId, payload, replicationBatchHint);
939 entry.lastAccess = shard.ticker().read();
941 LOG.debug("{}: Transaction {} submitted to persistence", logContext, txId);
943 // Process the next transaction pending commit, if any. If there is one it will be batched with this
944 // transaction for replication.
945 processNextPendingCommit();
948 Collection<ActorRef> getCohortActors() {
949 return cohortRegistry.getCohortActors();
952 void processCohortRegistryCommand(final ActorRef sender, final CohortRegistryCommand message) {
953 cohortRegistry.process(sender, message);
957 ShardDataTreeCohort createFailedCohort(final TransactionIdentifier txId, final DataTreeModification mod,
958 final Exception failure) {
959 SimpleShardDataTreeCohort cohort = new SimpleShardDataTreeCohort.DeadOnArrival(this, mod, txId, failure);
960 pendingTransactions.add(new CommitEntry(cohort, readTime()));
965 ShardDataTreeCohort createReadyCohort(final TransactionIdentifier txId,
966 final DataTreeModification mod) {
967 SimpleShardDataTreeCohort cohort = new SimpleShardDataTreeCohort.Normal(this, mod, txId,
968 cohortRegistry.createCohort(schemaContext, txId, COMMIT_STEP_TIMEOUT));
969 pendingTransactions.add(new CommitEntry(cohort, readTime()));
973 // Exposed for ShardCommitCoordinator so it does not have deal with local histories (it does not care), this mimics
974 // the newReadWriteTransaction()
975 ShardDataTreeCohort newReadyCohort(final TransactionIdentifier txId, final DataTreeModification mod) {
976 if (txId.getHistoryId().getHistoryId() == 0) {
977 return createReadyCohort(txId, mod);
980 return ensureTransactionChain(txId.getHistoryId(), null).createReadyCohort(txId, mod);
983 @SuppressFBWarnings(value = "DB_DUPLICATE_SWITCH_CLAUSES", justification = "See inline comments below.")
984 void checkForExpiredTransactions(final long transactionCommitTimeoutMillis) {
985 final long timeout = TimeUnit.MILLISECONDS.toNanos(transactionCommitTimeoutMillis);
986 final long now = readTime();
988 final Queue<CommitEntry> currentQueue = !pendingFinishCommits.isEmpty() ? pendingFinishCommits :
989 !pendingCommits.isEmpty() ? pendingCommits : pendingTransactions;
990 final CommitEntry currentTx = currentQueue.peek();
991 if (currentTx != null && currentTx.lastAccess + timeout < now) {
992 LOG.warn("{}: Current transaction {} has timed out after {} ms in state {}", logContext,
993 currentTx.cohort.getIdentifier(), transactionCommitTimeoutMillis, currentTx.cohort.getState());
994 boolean processNext = true;
995 switch (currentTx.cohort.getState()) {
996 case CAN_COMMIT_PENDING:
997 currentQueue.remove().cohort.failedCanCommit(new TimeoutException());
999 case CAN_COMMIT_COMPLETE:
1000 // The suppression of the FindBugs "DB_DUPLICATE_SWITCH_CLAUSES" warning pertains to this clause
1001 // whose code is duplicated with PRE_COMMIT_COMPLETE. The clauses aren't combined in case the code
1002 // in PRE_COMMIT_COMPLETE is changed.
1003 currentQueue.remove().cohort.reportFailure(new TimeoutException());
1005 case PRE_COMMIT_PENDING:
1006 currentQueue.remove().cohort.failedPreCommit(new TimeoutException());
1008 case PRE_COMMIT_COMPLETE:
1009 // FIXME: this is a legacy behavior problem. Three-phase commit protocol specifies that after we
1010 // are ready we should commit the transaction, not abort it. Our current software stack does
1011 // not allow us to do that consistently, because we persist at the time of commit, hence
1012 // we can end up in a state where we have pre-committed a transaction, then a leader failover
1013 // occurred ... the new leader does not see the pre-committed transaction and does not have
1014 // a running timer. To fix this we really need two persistence events.
1016 // The first one, done at pre-commit time will hold the transaction payload. When consensus
1017 // is reached, we exit the pre-commit phase and start the pre-commit timer. Followers do not
1018 // apply the state in this event.
1020 // The second one, done at commit (or abort) time holds only the transaction identifier and
1021 // signals to followers that the state should (or should not) be applied.
1023 // In order to make the pre-commit timer working across failovers, though, we need
1024 // a per-shard cluster-wide monotonic time, so a follower becoming the leader can accurately
1025 // restart the timer.
1026 currentQueue.remove().cohort.reportFailure(new TimeoutException());
1028 case COMMIT_PENDING:
1029 LOG.warn("{}: Transaction {} is still committing, cannot abort", logContext,
1030 currentTx.cohort.getIdentifier());
1031 currentTx.lastAccess = now;
1032 processNext = false;
1039 currentQueue.remove();
1043 processNextPending();
1048 boolean startAbort(final SimpleShardDataTreeCohort cohort) {
1049 final Iterator<CommitEntry> it = Iterables.concat(pendingFinishCommits, pendingCommits,
1050 pendingTransactions).iterator();
1051 if (!it.hasNext()) {
1052 LOG.debug("{}: no open transaction while attempting to abort {}", logContext, cohort.getIdentifier());
1056 // First entry is special, as it may already be committing
1057 final CommitEntry first = it.next();
1058 if (cohort.equals(first.cohort)) {
1059 if (cohort.getState() != State.COMMIT_PENDING) {
1060 LOG.debug("{}: aborting head of queue {} in state {}", logContext, cohort.getIdentifier(),
1061 cohort.getIdentifier());
1064 if (cohort.getCandidate() != null) {
1065 rebaseTransactions(it, dataTree);
1068 processNextPending();
1072 LOG.warn("{}: transaction {} is committing, skipping abort", logContext, cohort.getIdentifier());
1076 TipProducingDataTreeTip newTip = MoreObjects.firstNonNull(first.cohort.getCandidate(), dataTree);
1077 while (it.hasNext()) {
1078 final CommitEntry e = it.next();
1079 if (cohort.equals(e.cohort)) {
1080 LOG.debug("{}: aborting queued transaction {}", logContext, cohort.getIdentifier());
1083 if (cohort.getCandidate() != null) {
1084 rebaseTransactions(it, newTip);
1089 newTip = MoreObjects.firstNonNull(e.cohort.getCandidate(), newTip);
1093 LOG.debug("{}: aborted transaction {} not found in the queue", logContext, cohort.getIdentifier());
1097 @SuppressWarnings("checkstyle:IllegalCatch")
1098 private void rebaseTransactions(final Iterator<CommitEntry> iter, @Nonnull final TipProducingDataTreeTip newTip) {
1099 tip = Preconditions.checkNotNull(newTip);
1100 while (iter.hasNext()) {
1101 final SimpleShardDataTreeCohort cohort = iter.next().cohort;
1102 if (cohort.getState() == State.CAN_COMMIT_COMPLETE) {
1103 LOG.debug("{}: Revalidating queued transaction {}", logContext, cohort.getIdentifier());
1106 tip.validate(cohort.getDataTreeModification());
1107 } catch (DataValidationFailedException | RuntimeException e) {
1108 LOG.debug("{}: Failed to revalidate queued transaction {}", logContext, cohort.getIdentifier(), e);
1109 cohort.reportFailure(e);
1111 } else if (cohort.getState() == State.PRE_COMMIT_COMPLETE) {
1112 LOG.debug("{}: Repreparing queued transaction {}", logContext, cohort.getIdentifier());
1115 tip.validate(cohort.getDataTreeModification());
1116 DataTreeCandidateTip candidate = tip.prepare(cohort.getDataTreeModification());
1117 cohort.userPreCommit(candidate);
1119 cohort.setNewCandidate(candidate);
1121 } catch (ExecutionException | TimeoutException | RuntimeException | DataValidationFailedException e) {
1122 LOG.debug("{}: Failed to reprepare queued transaction {}", logContext, cohort.getIdentifier(), e);
1123 cohort.reportFailure(e);
1129 void setRunOnPendingTransactionsComplete(final Runnable operation) {
1130 runOnPendingTransactionsComplete = operation;
1131 maybeRunOperationOnPendingTransactionsComplete();
1134 private void maybeRunOperationOnPendingTransactionsComplete() {
1135 if (runOnPendingTransactionsComplete != null && !anyPendingTransactions()) {
1136 LOG.debug("{}: Pending transactions complete - running operation {}", logContext,
1137 runOnPendingTransactionsComplete);
1139 runOnPendingTransactionsComplete.run();
1140 runOnPendingTransactionsComplete = null;
1144 ShardStats getStats() {
1145 return shard.getShardMBean();