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());
353 LOG.debug("{}: ignoring unhandled payload {}", logContext, payload);
357 private void applyReplicatedCandidate(final Identifier identifier, final DataTreeCandidate foreign)
358 throws DataValidationFailedException {
359 LOG.debug("{}: Applying foreign transaction {}", logContext, identifier);
361 final DataTreeModification mod = dataTree.takeSnapshot().newModification();
362 DataTreeCandidates.applyToModification(mod, foreign);
365 LOG.trace("{}: Applying foreign modification {}", logContext, mod);
366 dataTree.validate(mod);
367 final DataTreeCandidate candidate = dataTree.prepare(mod);
368 dataTree.commit(candidate);
370 notifyListeners(candidate);
374 * Apply a payload coming from the leader, which could actually be us. This method assumes the leader and follower
375 * SchemaContexts match and does not perform any pruning.
377 * @param identifier Payload identifier as returned from RaftActor
378 * @param payload Payload
379 * @throws IOException when the snapshot fails to deserialize
380 * @throws DataValidationFailedException when the snapshot fails to apply
382 void applyReplicatedPayload(final Identifier identifier, final Payload payload) throws IOException,
383 DataValidationFailedException {
385 * 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
386 * if we are the leader and it has originated with us.
388 * The identifier will only ever be non-null when we were the leader which achieved consensus. Unfortunately,
389 * though, this may not be the case anymore, as we are being called some time afterwards and we may not be
390 * acting in that capacity anymore.
392 * In any case, we know that this is an entry coming from replication, hence we can be sure we will not observe
393 * pre-Boron state -- which limits the number of options here.
395 if (payload instanceof CommitTransactionPayload) {
396 final TransactionIdentifier txId;
397 if (identifier == null) {
398 final Entry<TransactionIdentifier, DataTreeCandidate> e =
399 ((CommitTransactionPayload) payload).getCandidate();
401 applyReplicatedCandidate(txId, e.getValue());
403 Verify.verify(identifier instanceof TransactionIdentifier);
404 txId = (TransactionIdentifier) identifier;
405 payloadReplicationComplete(txId);
407 allMetadataCommittedTransaction(txId);
408 } else if (payload instanceof AbortTransactionPayload) {
409 if (identifier != null) {
410 payloadReplicationComplete((AbortTransactionPayload) payload);
412 allMetadataAbortedTransaction(((AbortTransactionPayload) payload).getIdentifier());
413 } else if (payload instanceof PurgeTransactionPayload) {
414 if (identifier != null) {
415 payloadReplicationComplete((PurgeTransactionPayload) payload);
417 allMetadataPurgedTransaction(((PurgeTransactionPayload) payload).getIdentifier());
418 } else if (payload instanceof CloseLocalHistoryPayload) {
419 if (identifier != null) {
420 payloadReplicationComplete((CloseLocalHistoryPayload) payload);
422 allMetadataClosedLocalHistory(((CloseLocalHistoryPayload) payload).getIdentifier());
423 } else if (payload instanceof CreateLocalHistoryPayload) {
424 if (identifier != null) {
425 payloadReplicationComplete((CreateLocalHistoryPayload)payload);
427 allMetadataCreatedLocalHistory(((CreateLocalHistoryPayload) payload).getIdentifier());
428 } else if (payload instanceof PurgeLocalHistoryPayload) {
429 if (identifier != null) {
430 payloadReplicationComplete((PurgeLocalHistoryPayload)payload);
432 allMetadataPurgedLocalHistory(((PurgeLocalHistoryPayload) payload).getIdentifier());
434 LOG.warn("{}: ignoring unhandled identifier {} payload {}", logContext, identifier, payload);
438 private void replicatePayload(final Identifier id, final Payload payload, @Nullable final Runnable callback) {
439 if (callback != null) {
440 replicationCallbacks.put(payload, callback);
442 shard.persistPayload(id, payload, true);
445 private void payloadReplicationComplete(final AbstractIdentifiablePayload<?> payload) {
446 final Runnable callback = replicationCallbacks.remove(payload);
447 if (callback != null) {
448 LOG.debug("{}: replication of {} completed, invoking {}", logContext, payload.getIdentifier(), callback);
451 LOG.debug("{}: replication of {} has no callback", logContext, payload.getIdentifier());
455 private void payloadReplicationComplete(final TransactionIdentifier txId) {
456 final CommitEntry current = pendingFinishCommits.peek();
457 if (current == null) {
458 LOG.warn("{}: No outstanding transactions, ignoring consensus on transaction {}", logContext, txId);
462 if (!current.cohort.getIdentifier().equals(txId)) {
463 LOG.debug("{}: Head of pendingFinishCommits queue is {}, ignoring consensus on transaction {}", logContext,
464 current.cohort.getIdentifier(), txId);
468 finishCommit(current.cohort);
471 private void allMetadataAbortedTransaction(final TransactionIdentifier txId) {
472 for (ShardDataTreeMetadata<?> m : metadata) {
473 m.onTransactionAborted(txId);
477 private void allMetadataCommittedTransaction(final TransactionIdentifier txId) {
478 for (ShardDataTreeMetadata<?> m : metadata) {
479 m.onTransactionCommitted(txId);
483 private void allMetadataPurgedTransaction(final TransactionIdentifier txId) {
484 for (ShardDataTreeMetadata<?> m : metadata) {
485 m.onTransactionPurged(txId);
489 private void allMetadataCreatedLocalHistory(final LocalHistoryIdentifier historyId) {
490 for (ShardDataTreeMetadata<?> m : metadata) {
491 m.onHistoryCreated(historyId);
495 private void allMetadataClosedLocalHistory(final LocalHistoryIdentifier historyId) {
496 for (ShardDataTreeMetadata<?> m : metadata) {
497 m.onHistoryClosed(historyId);
501 private void allMetadataPurgedLocalHistory(final LocalHistoryIdentifier historyId) {
502 for (ShardDataTreeMetadata<?> m : metadata) {
503 m.onHistoryPurged(historyId);
508 * Create a transaction chain for specified history. Unlike {@link #ensureTransactionChain(LocalHistoryIdentifier)},
509 * this method is used for re-establishing state when we are taking over
511 * @param historyId Local history identifier
512 * @param closed True if the chain should be created in closed state (i.e. pending purge)
513 * @return Transaction chain handle
515 ShardDataTreeTransactionChain recreateTransactionChain(final LocalHistoryIdentifier historyId,
516 final boolean closed) {
517 final ShardDataTreeTransactionChain ret = new ShardDataTreeTransactionChain(historyId, this);
518 final ShardDataTreeTransactionChain existing = transactionChains.putIfAbsent(historyId, ret);
519 Preconditions.checkState(existing == null, "Attempted to recreate chain %s, but %s already exists", historyId,
524 ShardDataTreeTransactionChain ensureTransactionChain(final LocalHistoryIdentifier historyId,
525 @Nullable final Runnable callback) {
526 ShardDataTreeTransactionChain chain = transactionChains.get(historyId);
528 chain = new ShardDataTreeTransactionChain(historyId, this);
529 transactionChains.put(historyId, chain);
530 replicatePayload(historyId, CreateLocalHistoryPayload.create(historyId), callback);
531 } else if (callback != null) {
538 ReadOnlyShardDataTreeTransaction newReadOnlyTransaction(final TransactionIdentifier txId) {
539 if (txId.getHistoryId().getHistoryId() == 0) {
540 return new ReadOnlyShardDataTreeTransaction(this, txId, dataTree.takeSnapshot());
543 return ensureTransactionChain(txId.getHistoryId(), null).newReadOnlyTransaction(txId);
546 ReadWriteShardDataTreeTransaction newReadWriteTransaction(final TransactionIdentifier txId) {
547 if (txId.getHistoryId().getHistoryId() == 0) {
548 return new ReadWriteShardDataTreeTransaction(ShardDataTree.this, txId, dataTree.takeSnapshot()
552 return ensureTransactionChain(txId.getHistoryId(), null).newReadWriteTransaction(txId);
556 public void notifyListeners(final DataTreeCandidate candidate) {
557 treeChangeListenerPublisher.publishChanges(candidate);
558 dataChangeListenerPublisher.publishChanges(candidate);
562 * Immediately purge all state relevant to leader. This includes all transaction chains and any scheduled
563 * replication callbacks.
565 void purgeLeaderState() {
566 for (ShardDataTreeTransactionChain chain : transactionChains.values()) {
570 transactionChains.clear();
571 replicationCallbacks.clear();
575 * Close a single transaction chain.
577 * @param id History identifier
578 * @param callback Callback to invoke upon completion, may be null
580 void closeTransactionChain(final LocalHistoryIdentifier id, @Nullable final Runnable callback) {
581 final ShardDataTreeTransactionChain chain = transactionChains.get(id);
583 LOG.debug("{}: Closing non-existent transaction chain {}", logContext, id);
584 if (callback != null) {
591 replicatePayload(id, CloseLocalHistoryPayload.create(id), callback);
595 * Purge a single transaction chain.
597 * @param id History identifier
598 * @param callback Callback to invoke upon completion, may be null
600 void purgeTransactionChain(final LocalHistoryIdentifier id, @Nullable final Runnable callback) {
601 final ShardDataTreeTransactionChain chain = transactionChains.remove(id);
603 LOG.debug("{}: Purging non-existent transaction chain {}", logContext, id);
604 if (callback != null) {
610 replicatePayload(id, PurgeLocalHistoryPayload.create(id), callback);
613 void registerDataChangeListener(final YangInstanceIdentifier path,
614 final AsyncDataChangeListener<YangInstanceIdentifier, NormalizedNode<?, ?>> listener,
615 final DataChangeScope scope, final Optional<DataTreeCandidate> initialState,
616 final Consumer<ListenerRegistration<AsyncDataChangeListener<YangInstanceIdentifier, NormalizedNode<?, ?>>>>
618 dataChangeListenerPublisher.registerDataChangeListener(path, listener, scope, initialState, onRegistration);
621 Optional<DataTreeCandidate> readCurrentData() {
622 final Optional<NormalizedNode<?, ?>> currentState =
623 dataTree.takeSnapshot().readNode(YangInstanceIdentifier.EMPTY);
624 return currentState.isPresent() ? Optional.of(DataTreeCandidates.fromNormalizedNode(
625 YangInstanceIdentifier.EMPTY, currentState.get())) : Optional.<DataTreeCandidate>absent();
628 public void registerTreeChangeListener(final YangInstanceIdentifier path, final DOMDataTreeChangeListener listener,
629 final Optional<DataTreeCandidate> initialState,
630 final Consumer<ListenerRegistration<DOMDataTreeChangeListener>> onRegistration) {
631 treeChangeListenerPublisher.registerTreeChangeListener(path, listener, initialState, onRegistration);
635 return pendingTransactions.size() + pendingCommits.size() + pendingFinishCommits.size();
639 void abortTransaction(final AbstractShardDataTreeTransaction<?> transaction, final Runnable callback) {
640 final TransactionIdentifier id = transaction.getIdentifier();
641 LOG.debug("{}: aborting transaction {}", logContext, id);
642 replicatePayload(id, AbortTransactionPayload.create(id), callback);
646 void abortFromTransactionActor(final AbstractShardDataTreeTransaction<?> transaction) {
647 // No-op for free-standing transactions
652 ShardDataTreeCohort finishTransaction(final ReadWriteShardDataTreeTransaction transaction) {
653 final DataTreeModification snapshot = transaction.getSnapshot();
656 return createReadyCohort(transaction.getIdentifier(), snapshot);
659 void purgeTransaction(final TransactionIdentifier id, final Runnable callback) {
660 LOG.debug("{}: purging transaction {}", logContext, id);
661 replicatePayload(id, PurgeTransactionPayload.create(id), callback);
664 public Optional<NormalizedNode<?, ?>> readNode(final YangInstanceIdentifier path) {
665 return dataTree.takeSnapshot().readNode(path);
668 DataTreeSnapshot takeSnapshot() {
669 return dataTree.takeSnapshot();
673 public DataTreeModification newModification() {
674 return dataTree.takeSnapshot().newModification();
677 public Collection<ShardDataTreeCohort> getAndClearPendingTransactions() {
678 Collection<ShardDataTreeCohort> ret = new ArrayList<>(getQueueSize());
680 for (CommitEntry entry: pendingFinishCommits) {
681 ret.add(entry.cohort);
684 for (CommitEntry entry: pendingCommits) {
685 ret.add(entry.cohort);
688 for (CommitEntry entry: pendingTransactions) {
689 ret.add(entry.cohort);
692 pendingFinishCommits.clear();
693 pendingCommits.clear();
694 pendingTransactions.clear();
700 * Called some time after {@link #processNextPendingTransaction()} decides to stop processing.
702 void resumeNextPendingTransaction() {
703 LOG.debug("{}: attempting to resume transaction processing", logContext);
704 processNextPending();
707 @SuppressWarnings("checkstyle:IllegalCatch")
708 private void processNextPendingTransaction() {
709 ++currentTransactionBatch;
710 if (currentTransactionBatch > MAX_TRANSACTION_BATCH) {
711 LOG.debug("{}: Already processed {}, scheduling continuation", logContext, currentTransactionBatch);
712 shard.scheduleNextPendingTransaction();
716 processNextPending(pendingTransactions, State.CAN_COMMIT_PENDING, entry -> {
717 final SimpleShardDataTreeCohort cohort = entry.cohort;
718 final DataTreeModification modification = cohort.getDataTreeModification();
720 LOG.debug("{}: Validating transaction {}", logContext, cohort.getIdentifier());
723 tip.validate(modification);
724 LOG.debug("{}: Transaction {} validated", logContext, cohort.getIdentifier());
725 cohort.successfulCanCommit();
726 entry.lastAccess = readTime();
728 } catch (ConflictingModificationAppliedException e) {
729 LOG.warn("{}: Store Tx {}: Conflicting modification for path {}.", logContext, cohort.getIdentifier(),
731 cause = new OptimisticLockFailedException("Optimistic lock failed.", e);
732 } catch (DataValidationFailedException e) {
733 LOG.warn("{}: Store Tx {}: Data validation failed for path {}.", logContext, cohort.getIdentifier(),
736 // For debugging purposes, allow dumping of the modification. Coupled with the above
737 // precondition log, it should allow us to understand what went on.
738 LOG.debug("{}: Store Tx {}: modifications: {} tree: {}", cohort.getIdentifier(), modification,
740 cause = new TransactionCommitFailedException("Data did not pass validation.", e);
741 } catch (Exception e) {
742 LOG.warn("{}: Unexpected failure in validation phase", logContext, e);
746 // Failure path: propagate the failure, remove the transaction from the queue and loop to the next one
747 pendingTransactions.poll().cohort.failedCanCommit(cause);
751 private void processNextPending() {
752 processNextPendingCommit();
753 processNextPendingTransaction();
756 private void processNextPending(final Queue<CommitEntry> queue, final State allowedState,
757 final Consumer<CommitEntry> processor) {
758 while (!queue.isEmpty()) {
759 final CommitEntry entry = queue.peek();
760 final SimpleShardDataTreeCohort cohort = entry.cohort;
762 if (cohort.isFailed()) {
763 LOG.debug("{}: Removing failed transaction {}", logContext, cohort.getIdentifier());
768 if (cohort.getState() == allowedState) {
769 processor.accept(entry);
775 maybeRunOperationOnPendingTransactionsComplete();
778 private void processNextPendingCommit() {
779 processNextPending(pendingCommits, State.COMMIT_PENDING,
780 entry -> startCommit(entry.cohort, entry.cohort.getCandidate()));
783 private boolean peekNextPendingCommit() {
784 final CommitEntry first = pendingCommits.peek();
785 return first != null && first.cohort.getState() == State.COMMIT_PENDING;
788 void startCanCommit(final SimpleShardDataTreeCohort cohort) {
789 final CommitEntry head = pendingTransactions.peek();
791 LOG.warn("{}: No transactions enqueued while attempting to start canCommit on {}", logContext, cohort);
794 if (!cohort.equals(head.cohort)) {
795 LOG.debug("{}: Transaction {} scheduled for canCommit step", logContext, cohort.getIdentifier());
799 processNextPendingTransaction();
802 private void failPreCommit(final Exception cause) {
803 shard.getShardMBean().incrementFailedTransactionsCount();
804 pendingTransactions.poll().cohort.failedPreCommit(cause);
805 processNextPendingTransaction();
808 @SuppressWarnings("checkstyle:IllegalCatch")
809 void startPreCommit(final SimpleShardDataTreeCohort cohort) {
810 final CommitEntry entry = pendingTransactions.peek();
811 Preconditions.checkState(entry != null, "Attempted to pre-commit of %s when no transactions pending", cohort);
813 final SimpleShardDataTreeCohort current = entry.cohort;
814 Verify.verify(cohort.equals(current), "Attempted to pre-commit %s while %s is pending", cohort, current);
816 LOG.debug("{}: Preparing transaction {}", logContext, current.getIdentifier());
818 final DataTreeCandidateTip candidate;
820 candidate = tip.prepare(cohort.getDataTreeModification());
821 cohort.userPreCommit(candidate);
822 } catch (ExecutionException | TimeoutException | RuntimeException e) {
827 // Set the tip of the data tree.
828 tip = Verify.verifyNotNull(candidate);
830 entry.lastAccess = readTime();
832 pendingTransactions.remove();
833 pendingCommits.add(entry);
835 LOG.debug("{}: Transaction {} prepared", logContext, current.getIdentifier());
837 cohort.successfulPreCommit(candidate);
839 processNextPendingTransaction();
842 private void failCommit(final Exception cause) {
843 shard.getShardMBean().incrementFailedTransactionsCount();
844 pendingFinishCommits.poll().cohort.failedCommit(cause);
845 processNextPending();
848 @SuppressWarnings("checkstyle:IllegalCatch")
849 private void finishCommit(final SimpleShardDataTreeCohort cohort) {
850 final TransactionIdentifier txId = cohort.getIdentifier();
851 final DataTreeCandidate candidate = cohort.getCandidate();
853 LOG.debug("{}: Resuming commit of transaction {}", logContext, txId);
855 if (tip == candidate) {
856 // All pending candidates have been committed, reset the tip to the data tree.
861 dataTree.commit(candidate);
862 } catch (Exception e) {
863 LOG.error("{}: Failed to commit transaction {}", logContext, txId, e);
868 shard.getShardMBean().incrementCommittedTransactionCount();
869 shard.getShardMBean().setLastCommittedTransactionTime(System.currentTimeMillis());
871 // FIXME: propagate journal index
872 pendingFinishCommits.poll().cohort.successfulCommit(UnsignedLong.ZERO);
874 LOG.trace("{}: Transaction {} committed, proceeding to notify", logContext, txId);
875 notifyListeners(candidate);
877 processNextPending();
880 void startCommit(final SimpleShardDataTreeCohort cohort, final DataTreeCandidate candidate) {
881 final CommitEntry entry = pendingCommits.peek();
882 Preconditions.checkState(entry != null, "Attempted to start commit of %s when no transactions pending", cohort);
884 final SimpleShardDataTreeCohort current = entry.cohort;
885 if (!cohort.equals(current)) {
886 LOG.debug("{}: Transaction {} scheduled for commit step", logContext, cohort.getIdentifier());
890 LOG.debug("{}: Starting commit for transaction {}", logContext, current.getIdentifier());
892 final TransactionIdentifier txId = cohort.getIdentifier();
893 final Payload payload;
895 payload = CommitTransactionPayload.create(txId, candidate);
896 } catch (IOException e) {
897 LOG.error("{}: Failed to encode transaction {} candidate {}", logContext, txId, candidate, e);
898 pendingCommits.poll().cohort.failedCommit(e);
899 processNextPending();
903 // We process next transactions pending canCommit before we call persistPayload to possibly progress subsequent
904 // transactions to the COMMIT_PENDING state so the payloads can be batched for replication. This is done for
905 // single-shard transactions that immediately transition from canCommit to preCommit to commit. Note that
906 // if the next pending transaction is progressed to COMMIT_PENDING and this method (startCommit) is called,
907 // the next transaction will not attempt to replicate b/c the current transaction is still at the head of the
908 // pendingCommits queue.
909 processNextPendingTransaction();
911 // After processing next pending transactions, we can now remove the current transaction from pendingCommits.
912 // Note this must be done before the call to peekNextPendingCommit below so we check the next transaction
913 // in order to properly determine the batchHint flag for the call to persistPayload.
914 pendingCommits.remove();
915 pendingFinishCommits.add(entry);
917 // See if the next transaction is pending commit (ie in the COMMIT_PENDING state) so it can be batched with
918 // this transaction for replication.
919 boolean replicationBatchHint = peekNextPendingCommit();
921 // Once completed, we will continue via payloadReplicationComplete
922 shard.persistPayload(txId, payload, replicationBatchHint);
924 entry.lastAccess = shard.ticker().read();
926 LOG.debug("{}: Transaction {} submitted to persistence", logContext, txId);
928 // Process the next transaction pending commit, if any. If there is one it will be batched with this
929 // transaction for replication.
930 processNextPendingCommit();
933 Collection<ActorRef> getCohortActors() {
934 return cohortRegistry.getCohortActors();
937 void processCohortRegistryCommand(final ActorRef sender, final CohortRegistryCommand message) {
938 cohortRegistry.process(sender, message);
942 ShardDataTreeCohort createFailedCohort(final TransactionIdentifier txId, final DataTreeModification mod,
943 final Exception failure) {
944 final SimpleShardDataTreeCohort cohort = new SimpleShardDataTreeCohort(this, mod, txId, failure);
945 pendingTransactions.add(new CommitEntry(cohort, readTime()));
950 ShardDataTreeCohort createReadyCohort(final TransactionIdentifier txId, final DataTreeModification mod) {
951 SimpleShardDataTreeCohort cohort = new SimpleShardDataTreeCohort(this, mod, txId,
952 cohortRegistry.createCohort(schemaContext, txId, COMMIT_STEP_TIMEOUT));
953 pendingTransactions.add(new CommitEntry(cohort, readTime()));
957 // Exposed for ShardCommitCoordinator so it does not have deal with local histories (it does not care), this mimics
958 // the newReadWriteTransaction()
959 ShardDataTreeCohort newReadyCohort(final TransactionIdentifier txId, final DataTreeModification mod) {
960 if (txId.getHistoryId().getHistoryId() == 0) {
961 return createReadyCohort(txId, mod);
964 return ensureTransactionChain(txId.getHistoryId(), null).createReadyCohort(txId, mod);
967 @SuppressFBWarnings(value = "DB_DUPLICATE_SWITCH_CLAUSES", justification = "See inline comments below.")
968 void checkForExpiredTransactions(final long transactionCommitTimeoutMillis,
969 final Function<SimpleShardDataTreeCohort, Optional<Long>> accessTimeUpdater) {
970 final long timeout = TimeUnit.MILLISECONDS.toNanos(transactionCommitTimeoutMillis);
971 final long now = readTime();
973 final Queue<CommitEntry> currentQueue = !pendingFinishCommits.isEmpty() ? pendingFinishCommits :
974 !pendingCommits.isEmpty() ? pendingCommits : pendingTransactions;
975 final CommitEntry currentTx = currentQueue.peek();
976 if (currentTx == null) {
977 // Empty queue, no-op
981 long delta = now - currentTx.lastAccess;
982 if (delta < timeout) {
983 // Not expired yet, bail
987 final Optional<Long> updateOpt = accessTimeUpdater.apply(currentTx.cohort);
988 if (updateOpt.isPresent()) {
989 final long newAccess = updateOpt.get().longValue();
990 final long newDelta = now - newAccess;
991 if (newDelta < delta) {
992 LOG.debug("{}: Updated current transaction {} access time", logContext,
993 currentTx.cohort.getIdentifier());
994 currentTx.lastAccess = newAccess;
998 if (delta < timeout) {
999 // Not expired yet, bail
1004 final long deltaMillis = TimeUnit.NANOSECONDS.toMillis(delta);
1005 final State state = currentTx.cohort.getState();
1007 LOG.warn("{}: Current transaction {} has timed out after {} ms in state {}", logContext,
1008 currentTx.cohort.getIdentifier(), deltaMillis, state);
1009 boolean processNext = true;
1010 final TimeoutException cohortFailure = new TimeoutException("Backend timeout in state " + state + " after "
1011 + deltaMillis + "ms");
1014 case CAN_COMMIT_PENDING:
1015 currentQueue.remove().cohort.failedCanCommit(cohortFailure);
1017 case CAN_COMMIT_COMPLETE:
1018 // The suppression of the FindBugs "DB_DUPLICATE_SWITCH_CLAUSES" warning pertains to this clause
1019 // whose code is duplicated with PRE_COMMIT_COMPLETE. The clauses aren't combined in case the code
1020 // in PRE_COMMIT_COMPLETE is changed.
1021 currentQueue.remove().cohort.reportFailure(cohortFailure);
1023 case PRE_COMMIT_PENDING:
1024 currentQueue.remove().cohort.failedPreCommit(cohortFailure);
1026 case PRE_COMMIT_COMPLETE:
1027 // FIXME: this is a legacy behavior problem. Three-phase commit protocol specifies that after we
1028 // are ready we should commit the transaction, not abort it. Our current software stack does
1029 // not allow us to do that consistently, because we persist at the time of commit, hence
1030 // we can end up in a state where we have pre-committed a transaction, then a leader failover
1031 // occurred ... the new leader does not see the pre-committed transaction and does not have
1032 // a running timer. To fix this we really need two persistence events.
1034 // The first one, done at pre-commit time will hold the transaction payload. When consensus
1035 // is reached, we exit the pre-commit phase and start the pre-commit timer. Followers do not
1036 // apply the state in this event.
1038 // The second one, done at commit (or abort) time holds only the transaction identifier and
1039 // signals to followers that the state should (or should not) be applied.
1041 // In order to make the pre-commit timer working across failovers, though, we need
1042 // a per-shard cluster-wide monotonic time, so a follower becoming the leader can accurately
1043 // restart the timer.
1044 currentQueue.remove().cohort.reportFailure(cohortFailure);
1046 case COMMIT_PENDING:
1047 LOG.warn("{}: Transaction {} is still committing, cannot abort", logContext,
1048 currentTx.cohort.getIdentifier());
1049 currentTx.lastAccess = now;
1050 processNext = false;
1053 currentQueue.remove().cohort.reportFailure(cohortFailure);
1059 currentQueue.remove();
1063 processNextPending();
1067 boolean startAbort(final SimpleShardDataTreeCohort cohort) {
1068 final Iterator<CommitEntry> it = Iterables.concat(pendingFinishCommits, pendingCommits,
1069 pendingTransactions).iterator();
1070 if (!it.hasNext()) {
1071 LOG.debug("{}: no open transaction while attempting to abort {}", logContext, cohort.getIdentifier());
1075 // First entry is special, as it may already be committing
1076 final CommitEntry first = it.next();
1077 if (cohort.equals(first.cohort)) {
1078 if (cohort.getState() != State.COMMIT_PENDING) {
1079 LOG.debug("{}: aborting head of queue {} in state {}", logContext, cohort.getIdentifier(),
1080 cohort.getIdentifier());
1083 if (cohort.getCandidate() != null) {
1084 rebaseTransactions(it, dataTree);
1087 processNextPending();
1091 LOG.warn("{}: transaction {} is committing, skipping abort", logContext, cohort.getIdentifier());
1095 TipProducingDataTreeTip newTip = MoreObjects.firstNonNull(first.cohort.getCandidate(), dataTree);
1096 while (it.hasNext()) {
1097 final CommitEntry e = it.next();
1098 if (cohort.equals(e.cohort)) {
1099 LOG.debug("{}: aborting queued transaction {}", logContext, cohort.getIdentifier());
1102 if (cohort.getCandidate() != null) {
1103 rebaseTransactions(it, newTip);
1108 newTip = MoreObjects.firstNonNull(e.cohort.getCandidate(), newTip);
1112 LOG.debug("{}: aborted transaction {} not found in the queue", logContext, cohort.getIdentifier());
1116 @SuppressWarnings("checkstyle:IllegalCatch")
1117 private void rebaseTransactions(final Iterator<CommitEntry> iter, @Nonnull final TipProducingDataTreeTip newTip) {
1118 tip = Preconditions.checkNotNull(newTip);
1119 while (iter.hasNext()) {
1120 final SimpleShardDataTreeCohort cohort = iter.next().cohort;
1121 if (cohort.getState() == State.CAN_COMMIT_COMPLETE) {
1122 LOG.debug("{}: Revalidating queued transaction {}", logContext, cohort.getIdentifier());
1125 tip.validate(cohort.getDataTreeModification());
1126 } catch (DataValidationFailedException | RuntimeException e) {
1127 LOG.debug("{}: Failed to revalidate queued transaction {}", logContext, cohort.getIdentifier(), e);
1128 cohort.reportFailure(e);
1130 } else if (cohort.getState() == State.PRE_COMMIT_COMPLETE) {
1131 LOG.debug("{}: Repreparing queued transaction {}", logContext, cohort.getIdentifier());
1134 tip.validate(cohort.getDataTreeModification());
1135 DataTreeCandidateTip candidate = tip.prepare(cohort.getDataTreeModification());
1136 cohort.userPreCommit(candidate);
1138 cohort.setNewCandidate(candidate);
1140 } catch (ExecutionException | TimeoutException | RuntimeException | DataValidationFailedException e) {
1141 LOG.debug("{}: Failed to reprepare queued transaction {}", logContext, cohort.getIdentifier(), e);
1142 cohort.reportFailure(e);
1148 void setRunOnPendingTransactionsComplete(final Runnable operation) {
1149 runOnPendingTransactionsComplete = operation;
1150 maybeRunOperationOnPendingTransactionsComplete();
1153 private void maybeRunOperationOnPendingTransactionsComplete() {
1154 if (runOnPendingTransactionsComplete != null && !anyPendingTransactions()) {
1155 LOG.debug("{}: Pending transactions complete - running operation {}", logContext,
1156 runOnPendingTransactionsComplete);
1158 runOnPendingTransactionsComplete.run();
1159 runOnPendingTransactionsComplete = null;
1163 ShardStats getStats() {
1164 return shard.getShardMBean();
1167 Iterator<SimpleShardDataTreeCohort> cohortIterator() {
1168 return Iterables.transform(Iterables.concat(pendingFinishCommits, pendingCommits, pendingTransactions),
1169 e -> e.cohort).iterator();
1172 void removeTransactionChain(final LocalHistoryIdentifier id) {
1173 if (transactionChains.remove(id) != null) {
1174 LOG.debug("{}: Removed transaction chain {}", logContext, id);