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 static akka.actor.ActorRef.noSender;
11 import static com.google.common.base.Preconditions.checkState;
12 import static com.google.common.base.Verify.verify;
13 import static com.google.common.base.Verify.verifyNotNull;
14 import static java.util.Objects.requireNonNull;
15 import static java.util.Objects.requireNonNullElse;
17 import akka.actor.ActorRef;
18 import akka.util.Timeout;
19 import com.google.common.annotations.VisibleForTesting;
20 import com.google.common.base.Stopwatch;
21 import com.google.common.collect.ImmutableList;
22 import com.google.common.collect.ImmutableMap;
23 import com.google.common.collect.ImmutableMap.Builder;
24 import com.google.common.collect.Iterables;
25 import com.google.common.primitives.UnsignedLong;
26 import com.google.common.util.concurrent.FutureCallback;
27 import edu.umd.cs.findbugs.annotations.SuppressFBWarnings;
29 import java.io.IOException;
30 import java.util.ArrayDeque;
31 import java.util.ArrayList;
32 import java.util.Collection;
33 import java.util.Collections;
34 import java.util.Deque;
35 import java.util.HashMap;
36 import java.util.Iterator;
38 import java.util.Map.Entry;
39 import java.util.Optional;
40 import java.util.OptionalLong;
41 import java.util.Queue;
42 import java.util.SortedSet;
43 import java.util.concurrent.TimeUnit;
44 import java.util.concurrent.TimeoutException;
45 import java.util.function.Consumer;
46 import java.util.function.Function;
47 import java.util.function.UnaryOperator;
48 import org.eclipse.jdt.annotation.NonNull;
49 import org.eclipse.jdt.annotation.Nullable;
50 import org.opendaylight.controller.cluster.access.concepts.LocalHistoryIdentifier;
51 import org.opendaylight.controller.cluster.access.concepts.TransactionIdentifier;
52 import org.opendaylight.controller.cluster.datastore.DataTreeCohortActorRegistry.CohortRegistryCommand;
53 import org.opendaylight.controller.cluster.datastore.ShardDataTreeCohort.State;
54 import org.opendaylight.controller.cluster.datastore.node.utils.transformer.ReusableNormalizedNodePruner;
55 import org.opendaylight.controller.cluster.datastore.persisted.AbortTransactionPayload;
56 import org.opendaylight.controller.cluster.datastore.persisted.AbstractIdentifiablePayload;
57 import org.opendaylight.controller.cluster.datastore.persisted.CloseLocalHistoryPayload;
58 import org.opendaylight.controller.cluster.datastore.persisted.CommitTransactionPayload;
59 import org.opendaylight.controller.cluster.datastore.persisted.CreateLocalHistoryPayload;
60 import org.opendaylight.controller.cluster.datastore.persisted.DataTreeCandidateInputOutput.DataTreeCandidateWithVersion;
61 import org.opendaylight.controller.cluster.datastore.persisted.MetadataShardDataTreeSnapshot;
62 import org.opendaylight.controller.cluster.datastore.persisted.PayloadVersion;
63 import org.opendaylight.controller.cluster.datastore.persisted.PurgeLocalHistoryPayload;
64 import org.opendaylight.controller.cluster.datastore.persisted.PurgeTransactionPayload;
65 import org.opendaylight.controller.cluster.datastore.persisted.ShardDataTreeSnapshot;
66 import org.opendaylight.controller.cluster.datastore.persisted.ShardDataTreeSnapshotMetadata;
67 import org.opendaylight.controller.cluster.datastore.persisted.ShardSnapshotState;
68 import org.opendaylight.controller.cluster.datastore.utils.DataTreeModificationOutput;
69 import org.opendaylight.controller.cluster.datastore.utils.PruningDataTreeModification;
70 import org.opendaylight.controller.cluster.raft.base.messages.InitiateCaptureSnapshot;
71 import org.opendaylight.controller.cluster.raft.protobuff.client.messages.Payload;
72 import org.opendaylight.mdsal.common.api.OptimisticLockFailedException;
73 import org.opendaylight.mdsal.common.api.TransactionCommitFailedException;
74 import org.opendaylight.mdsal.dom.api.DOMDataTreeChangeListener;
75 import org.opendaylight.yangtools.concepts.Identifier;
76 import org.opendaylight.yangtools.concepts.ListenerRegistration;
77 import org.opendaylight.yangtools.yang.data.api.YangInstanceIdentifier;
78 import org.opendaylight.yangtools.yang.data.api.schema.NormalizedNode;
79 import org.opendaylight.yangtools.yang.data.api.schema.tree.ConflictingModificationAppliedException;
80 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTree;
81 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeCandidate;
82 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeCandidateTip;
83 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeCandidates;
84 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeConfiguration;
85 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeModification;
86 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeSnapshot;
87 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeTip;
88 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataValidationFailedException;
89 import org.opendaylight.yangtools.yang.data.api.schema.tree.ModificationType;
90 import org.opendaylight.yangtools.yang.data.api.schema.tree.TreeType;
91 import org.opendaylight.yangtools.yang.data.codec.binfmt.NormalizedNodeStreamVersion;
92 import org.opendaylight.yangtools.yang.data.impl.schema.tree.InMemoryDataTreeFactory;
93 import org.opendaylight.yangtools.yang.data.util.DataSchemaContextTree;
94 import org.opendaylight.yangtools.yang.model.api.EffectiveModelContext;
95 import org.opendaylight.yangtools.yang.model.api.SchemaContext;
96 import org.slf4j.Logger;
97 import org.slf4j.LoggerFactory;
98 import scala.concurrent.duration.FiniteDuration;
101 * Internal shard state, similar to a DOMStore, but optimized for use in the actor system, e.g. it does not expose
102 * public interfaces and assumes it is only ever called from a single thread.
105 * This class is not part of the API contract and is subject to change at any time. It is NOT thread-safe.
108 // non-final for mocking
109 public class ShardDataTree extends ShardDataTreeTransactionParent {
110 private static final class CommitEntry {
111 final SimpleShardDataTreeCohort cohort;
114 CommitEntry(final SimpleShardDataTreeCohort cohort, final long now) {
115 this.cohort = requireNonNull(cohort);
120 public String toString() {
121 return "CommitEntry [tx=" + cohort.getIdentifier() + ", state=" + cohort.getState() + "]";
125 private static final Timeout COMMIT_STEP_TIMEOUT = new Timeout(FiniteDuration.create(5, TimeUnit.SECONDS));
126 private static final Logger LOG = LoggerFactory.getLogger(ShardDataTree.class);
129 * Process this many transactions in a single batched run. If we exceed this limit, we need to schedule later
130 * execution to finish up the batch. This is necessary in case of a long list of transactions which progress
131 * immediately through their preCommit phase -- if that happens, their completion eats up stack frames and could
132 * result in StackOverflowError.
134 private static final int MAX_TRANSACTION_BATCH = 100;
136 private final Map<LocalHistoryIdentifier, ShardDataTreeTransactionChain> transactionChains = new HashMap<>();
137 private final DataTreeCohortActorRegistry cohortRegistry = new DataTreeCohortActorRegistry();
138 private final Deque<CommitEntry> pendingTransactions = new ArrayDeque<>();
139 private final Queue<CommitEntry> pendingCommits = new ArrayDeque<>();
140 private final Queue<CommitEntry> pendingFinishCommits = new ArrayDeque<>();
143 * Callbacks that need to be invoked once a payload is replicated.
145 private final Map<Payload, Runnable> replicationCallbacks = new HashMap<>();
147 private final ShardDataTreeChangeListenerPublisher treeChangeListenerPublisher;
148 private final Collection<ShardDataTreeMetadata<?>> metadata;
149 private final DataTree dataTree;
150 private final String logContext;
151 private final Shard shard;
152 private Runnable runOnPendingTransactionsComplete;
155 * Optimistic {@link DataTreeCandidate} preparation. Since our DataTree implementation is a
156 * {@link DataTree}, each {@link DataTreeCandidate} is also a {@link DataTreeTip}, e.g. another
157 * candidate can be prepared on top of it. They still need to be committed in sequence. Here we track the current
158 * tip of the data tree, which is the last DataTreeCandidate we have in flight, or the DataTree itself.
160 private DataTreeTip tip;
162 private SchemaContext schemaContext;
163 private DataSchemaContextTree dataSchemaContext;
165 private int currentTransactionBatch;
167 ShardDataTree(final Shard shard, final EffectiveModelContext schemaContext, final DataTree dataTree,
168 final ShardDataTreeChangeListenerPublisher treeChangeListenerPublisher,
169 final String logContext,
170 final ShardDataTreeMetadata<?>... metadata) {
171 this.dataTree = requireNonNull(dataTree);
172 updateSchemaContext(schemaContext);
174 this.shard = requireNonNull(shard);
175 this.treeChangeListenerPublisher = requireNonNull(treeChangeListenerPublisher);
176 this.logContext = requireNonNull(logContext);
177 this.metadata = ImmutableList.copyOf(metadata);
181 ShardDataTree(final Shard shard, final EffectiveModelContext schemaContext, final TreeType treeType,
182 final YangInstanceIdentifier root,
183 final ShardDataTreeChangeListenerPublisher treeChangeListenerPublisher,
184 final String logContext,
185 final ShardDataTreeMetadata<?>... metadata) {
186 this(shard, schemaContext, createDataTree(treeType, root), treeChangeListenerPublisher, logContext, metadata);
189 private static DataTree createDataTree(final TreeType treeType, final YangInstanceIdentifier root) {
190 final DataTreeConfiguration baseConfig = DataTreeConfiguration.getDefault(treeType);
191 return new InMemoryDataTreeFactory().create(new DataTreeConfiguration.Builder(baseConfig.getTreeType())
192 .setMandatoryNodesValidation(baseConfig.isMandatoryNodesValidationEnabled())
193 .setUniqueIndexes(baseConfig.isUniqueIndexEnabled())
199 public ShardDataTree(final Shard shard, final EffectiveModelContext schemaContext, final TreeType treeType) {
200 this(shard, schemaContext, treeType, YangInstanceIdentifier.empty(),
201 new DefaultShardDataTreeChangeListenerPublisher(""), "");
204 final String logContext() {
208 final long readTime() {
209 return shard.ticker().read();
212 final DataTree getDataTree() {
217 final SchemaContext getSchemaContext() {
218 return schemaContext;
221 final void updateSchemaContext(final @NonNull EffectiveModelContext newSchemaContext) {
222 dataTree.setEffectiveModelContext(newSchemaContext);
223 this.schemaContext = newSchemaContext;
224 this.dataSchemaContext = DataSchemaContextTree.from(newSchemaContext);
227 final void resetTransactionBatch() {
228 currentTransactionBatch = 0;
232 * Take a snapshot of current state for later recovery.
234 * @return A state snapshot
236 @NonNull ShardDataTreeSnapshot takeStateSnapshot() {
237 final NormalizedNode rootNode = dataTree.takeSnapshot().readNode(YangInstanceIdentifier.empty()).get();
238 final Builder<Class<? extends ShardDataTreeSnapshotMetadata<?>>, ShardDataTreeSnapshotMetadata<?>> metaBuilder =
239 ImmutableMap.builder();
241 for (ShardDataTreeMetadata<?> m : metadata) {
242 final ShardDataTreeSnapshotMetadata<?> meta = m.toSnapshot();
244 metaBuilder.put(meta.getType(), meta);
248 return new MetadataShardDataTreeSnapshot(rootNode, metaBuilder.build());
251 private boolean anyPendingTransactions() {
252 return !pendingTransactions.isEmpty() || !pendingCommits.isEmpty() || !pendingFinishCommits.isEmpty();
255 private void applySnapshot(final @NonNull ShardDataTreeSnapshot snapshot,
256 final UnaryOperator<DataTreeModification> wrapper) throws DataValidationFailedException {
257 final Stopwatch elapsed = Stopwatch.createStarted();
259 if (anyPendingTransactions()) {
260 LOG.warn("{}: applying state snapshot with pending transactions", logContext);
263 final Map<Class<? extends ShardDataTreeSnapshotMetadata<?>>, ShardDataTreeSnapshotMetadata<?>> snapshotMeta;
264 if (snapshot instanceof MetadataShardDataTreeSnapshot) {
265 snapshotMeta = ((MetadataShardDataTreeSnapshot) snapshot).getMetadata();
267 snapshotMeta = ImmutableMap.of();
270 for (ShardDataTreeMetadata<?> m : metadata) {
271 final ShardDataTreeSnapshotMetadata<?> s = snapshotMeta.get(m.getSupportedType());
279 final DataTreeModification unwrapped = dataTree.takeSnapshot().newModification();
280 final DataTreeModification mod = wrapper.apply(unwrapped);
281 // delete everything first
282 mod.delete(YangInstanceIdentifier.empty());
284 final Optional<NormalizedNode> maybeNode = snapshot.getRootNode();
285 if (maybeNode.isPresent()) {
286 // Add everything from the remote node back
287 mod.write(YangInstanceIdentifier.empty(), maybeNode.get());
291 dataTree.validate(unwrapped);
292 DataTreeCandidateTip candidate = dataTree.prepare(unwrapped);
293 dataTree.commit(candidate);
294 notifyListeners(candidate);
296 LOG.debug("{}: state snapshot applied in {}", logContext, elapsed);
300 * Apply a snapshot coming from the leader. This method assumes the leader and follower SchemaContexts match and
301 * does not perform any pruning.
303 * @param snapshot Snapshot that needs to be applied
304 * @throws DataValidationFailedException when the snapshot fails to apply
306 final void applySnapshot(final @NonNull ShardDataTreeSnapshot snapshot) throws DataValidationFailedException {
307 // TODO: we should be taking ShardSnapshotState here and performing forward-compatibility translation
308 applySnapshot(snapshot, UnaryOperator.identity());
312 * Apply a snapshot coming from recovery. This method does not assume the SchemaContexts match and performs data
313 * pruning in an attempt to adjust the state to our current SchemaContext.
315 * @param snapshot Snapshot that needs to be applied
316 * @throws DataValidationFailedException when the snapshot fails to apply
318 final void applyRecoverySnapshot(final @NonNull ShardSnapshotState snapshot) throws DataValidationFailedException {
319 // TODO: we should be able to reuse the pruner, provided we are not reentrant
320 final ReusableNormalizedNodePruner pruner = ReusableNormalizedNodePruner.forDataSchemaContext(
322 if (snapshot.needsMigration()) {
323 final ReusableNormalizedNodePruner uintPruner = pruner.withUintAdaption();
324 applySnapshot(snapshot.getSnapshot(),
325 delegate -> new PruningDataTreeModification.Proactive(delegate, dataTree, uintPruner));
327 applySnapshot(snapshot.getSnapshot(),
328 delegate -> new PruningDataTreeModification.Reactive(delegate, dataTree, pruner));
332 @SuppressWarnings("checkstyle:IllegalCatch")
333 private void applyRecoveryCandidate(final CommitTransactionPayload payload) throws IOException {
334 final Entry<TransactionIdentifier, DataTreeCandidateWithVersion> entry = payload.acquireCandidate();
335 final DataTreeModification unwrapped = dataTree.takeSnapshot().newModification();
336 final PruningDataTreeModification mod = createPruningModification(unwrapped,
337 NormalizedNodeStreamVersion.MAGNESIUM.compareTo(entry.getValue().getVersion()) > 0);
339 DataTreeCandidates.applyToModification(mod, entry.getValue().getCandidate());
341 LOG.trace("{}: Applying recovery modification {}", logContext, unwrapped);
344 dataTree.validate(unwrapped);
345 dataTree.commit(dataTree.prepare(unwrapped));
346 } catch (Exception e) {
347 File file = new File(System.getProperty("karaf.data", "."),
348 "failed-recovery-payload-" + logContext + ".out");
349 DataTreeModificationOutput.toFile(file, unwrapped);
350 throw new IllegalStateException(String.format(
351 "%s: Failed to apply recovery payload. Modification data was written to file %s",
352 logContext, file), e);
355 allMetadataCommittedTransaction(entry.getKey());
358 private PruningDataTreeModification createPruningModification(final DataTreeModification unwrapped,
359 final boolean uintAdapting) {
360 // TODO: we should be able to reuse the pruner, provided we are not reentrant
361 final ReusableNormalizedNodePruner pruner = ReusableNormalizedNodePruner.forDataSchemaContext(
363 return uintAdapting ? new PruningDataTreeModification.Proactive(unwrapped, dataTree, pruner.withUintAdaption())
364 : new PruningDataTreeModification.Reactive(unwrapped, dataTree, pruner);
368 * Apply a payload coming from recovery. This method does not assume the SchemaContexts match and performs data
369 * pruning in an attempt to adjust the state to our current SchemaContext.
371 * @param payload Payload
372 * @throws IOException when the snapshot fails to deserialize
373 * @throws DataValidationFailedException when the snapshot fails to apply
375 final void applyRecoveryPayload(final @NonNull Payload payload) throws IOException {
376 if (payload instanceof CommitTransactionPayload) {
377 applyRecoveryCandidate((CommitTransactionPayload) payload);
378 } else if (payload instanceof AbortTransactionPayload) {
379 allMetadataAbortedTransaction(((AbortTransactionPayload) payload).getIdentifier());
380 } else if (payload instanceof PurgeTransactionPayload) {
381 allMetadataPurgedTransaction(((PurgeTransactionPayload) payload).getIdentifier());
382 } else if (payload instanceof CreateLocalHistoryPayload) {
383 allMetadataCreatedLocalHistory(((CreateLocalHistoryPayload) payload).getIdentifier());
384 } else if (payload instanceof CloseLocalHistoryPayload) {
385 allMetadataClosedLocalHistory(((CloseLocalHistoryPayload) payload).getIdentifier());
386 } else if (payload instanceof PurgeLocalHistoryPayload) {
387 allMetadataPurgedLocalHistory(((PurgeLocalHistoryPayload) payload).getIdentifier());
389 LOG.debug("{}: ignoring unhandled payload {}", logContext, payload);
393 private void applyReplicatedCandidate(final CommitTransactionPayload payload)
394 throws DataValidationFailedException, IOException {
395 final Entry<TransactionIdentifier, DataTreeCandidateWithVersion> entry = payload.acquireCandidate();
396 final TransactionIdentifier identifier = entry.getKey();
397 LOG.debug("{}: Applying foreign transaction {}", logContext, identifier);
399 final DataTreeModification mod = dataTree.takeSnapshot().newModification();
400 // TODO: check version here, which will enable us to perform forward-compatibility transformations
401 DataTreeCandidates.applyToModification(mod, entry.getValue().getCandidate());
404 LOG.trace("{}: Applying foreign modification {}", logContext, mod);
405 dataTree.validate(mod);
406 final DataTreeCandidate candidate = dataTree.prepare(mod);
407 dataTree.commit(candidate);
409 allMetadataCommittedTransaction(identifier);
410 notifyListeners(candidate);
414 * Apply a payload coming from the leader, which could actually be us. This method assumes the leader and follower
415 * SchemaContexts match and does not perform any pruning.
417 * @param identifier Payload identifier as returned from RaftActor
418 * @param payload Payload
419 * @throws IOException when the snapshot fails to deserialize
420 * @throws DataValidationFailedException when the snapshot fails to apply
422 final void applyReplicatedPayload(final Identifier identifier, final Payload payload) throws IOException,
423 DataValidationFailedException {
425 * 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
426 * if we are the leader and it has originated with us.
428 * The identifier will only ever be non-null when we were the leader which achieved consensus. Unfortunately,
429 * though, this may not be the case anymore, as we are being called some time afterwards and we may not be
430 * acting in that capacity anymore.
432 * In any case, we know that this is an entry coming from replication, hence we can be sure we will not observe
433 * pre-Boron state -- which limits the number of options here.
435 if (payload instanceof CommitTransactionPayload) {
436 if (identifier == null) {
437 applyReplicatedCandidate((CommitTransactionPayload) payload);
439 verify(identifier instanceof TransactionIdentifier);
440 // if we did not track this transaction before, it means that it came from another leader and we are in
441 // the process of commiting it while in PreLeader state. That means that it hasnt yet been committed to
442 // the local DataTree and would be lost if it was only applied via payloadReplicationComplete().
443 if (!payloadReplicationComplete((TransactionIdentifier) identifier)) {
444 applyReplicatedCandidate((CommitTransactionPayload) payload);
448 // make sure acquireCandidate() is the last call touching the payload data as we want it to be GC-ed.
449 checkRootOverwrite(((CommitTransactionPayload) payload).acquireCandidate().getValue()
451 } else if (payload instanceof AbortTransactionPayload) {
452 if (identifier != null) {
453 payloadReplicationComplete((AbortTransactionPayload) payload);
455 allMetadataAbortedTransaction(((AbortTransactionPayload) payload).getIdentifier());
456 } else if (payload instanceof PurgeTransactionPayload) {
457 if (identifier != null) {
458 payloadReplicationComplete((PurgeTransactionPayload) payload);
460 allMetadataPurgedTransaction(((PurgeTransactionPayload) payload).getIdentifier());
461 } else if (payload instanceof CloseLocalHistoryPayload) {
462 if (identifier != null) {
463 payloadReplicationComplete((CloseLocalHistoryPayload) payload);
465 allMetadataClosedLocalHistory(((CloseLocalHistoryPayload) payload).getIdentifier());
466 } else if (payload instanceof CreateLocalHistoryPayload) {
467 if (identifier != null) {
468 payloadReplicationComplete((CreateLocalHistoryPayload)payload);
470 allMetadataCreatedLocalHistory(((CreateLocalHistoryPayload) payload).getIdentifier());
471 } else if (payload instanceof PurgeLocalHistoryPayload) {
472 if (identifier != null) {
473 payloadReplicationComplete((PurgeLocalHistoryPayload)payload);
475 allMetadataPurgedLocalHistory(((PurgeLocalHistoryPayload) payload).getIdentifier());
477 LOG.warn("{}: ignoring unhandled identifier {} payload {}", logContext, identifier, payload);
481 private void checkRootOverwrite(final DataTreeCandidate candidate) {
482 final DatastoreContext datastoreContext = shard.getDatastoreContext();
483 if (!datastoreContext.isSnapshotOnRootOverwrite()) {
487 if (!datastoreContext.isPersistent()) {
488 // FIXME: why don't we want a snapshot in non-persistent state?
492 // top level container ie "/"
493 if (candidate.getRootPath().isEmpty()
494 && candidate.getRootNode().getModificationType() == ModificationType.WRITE) {
495 LOG.debug("{}: shard root overwritten, enqueuing snapshot", logContext);
496 shard.self().tell(new InitiateCaptureSnapshot(), noSender());
500 private void replicatePayload(final Identifier id, final Payload payload, final @Nullable Runnable callback) {
501 if (callback != null) {
502 replicationCallbacks.put(payload, callback);
504 shard.persistPayload(id, payload, true);
507 private void payloadReplicationComplete(final AbstractIdentifiablePayload<?> payload) {
508 final Runnable callback = replicationCallbacks.remove(payload);
509 if (callback != null) {
510 LOG.debug("{}: replication of {} completed, invoking {}", logContext, payload.getIdentifier(), callback);
513 LOG.debug("{}: replication of {} has no callback", logContext, payload.getIdentifier());
517 private boolean payloadReplicationComplete(final TransactionIdentifier txId) {
518 final CommitEntry current = pendingFinishCommits.peek();
519 if (current == null) {
520 LOG.warn("{}: No outstanding transactions, ignoring consensus on transaction {}", logContext, txId);
521 allMetadataCommittedTransaction(txId);
525 if (!current.cohort.getIdentifier().equals(txId)) {
526 LOG.debug("{}: Head of pendingFinishCommits queue is {}, ignoring consensus on transaction {}", logContext,
527 current.cohort.getIdentifier(), txId);
528 allMetadataCommittedTransaction(txId);
532 finishCommit(current.cohort);
536 private void allMetadataAbortedTransaction(final TransactionIdentifier txId) {
537 for (ShardDataTreeMetadata<?> m : metadata) {
538 m.onTransactionAborted(txId);
542 private void allMetadataCommittedTransaction(final TransactionIdentifier txId) {
543 for (ShardDataTreeMetadata<?> m : metadata) {
544 m.onTransactionCommitted(txId);
548 private void allMetadataPurgedTransaction(final TransactionIdentifier txId) {
549 for (ShardDataTreeMetadata<?> m : metadata) {
550 m.onTransactionPurged(txId);
554 private void allMetadataCreatedLocalHistory(final LocalHistoryIdentifier historyId) {
555 for (ShardDataTreeMetadata<?> m : metadata) {
556 m.onHistoryCreated(historyId);
560 private void allMetadataClosedLocalHistory(final LocalHistoryIdentifier historyId) {
561 for (ShardDataTreeMetadata<?> m : metadata) {
562 m.onHistoryClosed(historyId);
566 private void allMetadataPurgedLocalHistory(final LocalHistoryIdentifier historyId) {
567 for (ShardDataTreeMetadata<?> m : metadata) {
568 m.onHistoryPurged(historyId);
573 * Create a transaction chain for specified history. Unlike {@link #ensureTransactionChain(LocalHistoryIdentifier)},
574 * this method is used for re-establishing state when we are taking over
576 * @param historyId Local history identifier
577 * @param closed True if the chain should be created in closed state (i.e. pending purge)
578 * @return Transaction chain handle
580 final ShardDataTreeTransactionChain recreateTransactionChain(final LocalHistoryIdentifier historyId,
581 final boolean closed) {
582 final ShardDataTreeTransactionChain ret = new ShardDataTreeTransactionChain(historyId, this);
583 final ShardDataTreeTransactionChain existing = transactionChains.putIfAbsent(historyId, ret);
584 checkState(existing == null, "Attempted to recreate chain %s, but %s already exists", historyId, existing);
588 final ShardDataTreeTransactionChain ensureTransactionChain(final LocalHistoryIdentifier historyId,
589 final @Nullable Runnable callback) {
590 ShardDataTreeTransactionChain chain = transactionChains.get(historyId);
592 chain = new ShardDataTreeTransactionChain(historyId, this);
593 transactionChains.put(historyId, chain);
594 replicatePayload(historyId, CreateLocalHistoryPayload.create(
595 historyId, shard.getDatastoreContext().getInitialPayloadSerializedBufferCapacity()), callback);
596 } else if (callback != null) {
603 final ReadOnlyShardDataTreeTransaction newReadOnlyTransaction(final TransactionIdentifier txId) {
604 shard.getShardMBean().incrementReadOnlyTransactionCount();
606 if (txId.getHistoryId().getHistoryId() == 0) {
607 return new ReadOnlyShardDataTreeTransaction(this, txId, dataTree.takeSnapshot());
610 return ensureTransactionChain(txId.getHistoryId(), null).newReadOnlyTransaction(txId);
613 final ReadWriteShardDataTreeTransaction newReadWriteTransaction(final TransactionIdentifier txId) {
614 shard.getShardMBean().incrementReadWriteTransactionCount();
616 if (txId.getHistoryId().getHistoryId() == 0) {
617 return new ReadWriteShardDataTreeTransaction(ShardDataTree.this, txId, dataTree.takeSnapshot()
621 return ensureTransactionChain(txId.getHistoryId(), null).newReadWriteTransaction(txId);
625 final void notifyListeners(final DataTreeCandidate candidate) {
626 treeChangeListenerPublisher.publishChanges(candidate);
630 * Immediately purge all state relevant to leader. This includes all transaction chains and any scheduled
631 * replication callbacks.
633 final void purgeLeaderState() {
634 for (ShardDataTreeTransactionChain chain : transactionChains.values()) {
638 transactionChains.clear();
639 replicationCallbacks.clear();
643 * Close a single transaction chain.
645 * @param id History identifier
646 * @param callback Callback to invoke upon completion, may be null
648 final void closeTransactionChain(final LocalHistoryIdentifier id, final @Nullable Runnable callback) {
649 if (commonCloseTransactionChain(id, callback)) {
650 replicatePayload(id, CloseLocalHistoryPayload.create(id,
651 shard.getDatastoreContext().getInitialPayloadSerializedBufferCapacity()), callback);
656 * Close a single transaction chain which is received through ask-based protocol. It does not keep a commit record.
658 * @param id History identifier
660 final void closeTransactionChain(final LocalHistoryIdentifier id) {
661 commonCloseTransactionChain(id, null);
664 private boolean commonCloseTransactionChain(final LocalHistoryIdentifier id, final @Nullable Runnable callback) {
665 final ShardDataTreeTransactionChain chain = transactionChains.get(id);
667 LOG.debug("{}: Closing non-existent transaction chain {}", logContext, id);
668 if (callback != null) {
679 * Purge a single transaction chain.
681 * @param id History identifier
682 * @param callback Callback to invoke upon completion, may be null
684 final void purgeTransactionChain(final LocalHistoryIdentifier id, final @Nullable Runnable callback) {
685 final ShardDataTreeTransactionChain chain = transactionChains.remove(id);
687 LOG.debug("{}: Purging non-existent transaction chain {}", logContext, id);
688 if (callback != null) {
694 replicatePayload(id, PurgeLocalHistoryPayload.create(
695 id, shard.getDatastoreContext().getInitialPayloadSerializedBufferCapacity()), callback);
698 final Optional<DataTreeCandidate> readCurrentData() {
699 return dataTree.takeSnapshot().readNode(YangInstanceIdentifier.empty())
700 .map(state -> DataTreeCandidates.fromNormalizedNode(YangInstanceIdentifier.empty(), state));
703 final void registerTreeChangeListener(final YangInstanceIdentifier path, final DOMDataTreeChangeListener listener,
704 final Optional<DataTreeCandidate> initialState,
705 final Consumer<ListenerRegistration<DOMDataTreeChangeListener>> onRegistration) {
706 treeChangeListenerPublisher.registerTreeChangeListener(path, listener, initialState, onRegistration);
709 final int getQueueSize() {
710 return pendingTransactions.size() + pendingCommits.size() + pendingFinishCommits.size();
714 final void abortTransaction(final AbstractShardDataTreeTransaction<?> transaction, final Runnable callback) {
715 final TransactionIdentifier id = transaction.getIdentifier();
716 LOG.debug("{}: aborting transaction {}", logContext, id);
717 replicatePayload(id, AbortTransactionPayload.create(
718 id, shard.getDatastoreContext().getInitialPayloadSerializedBufferCapacity()), callback);
722 final void abortFromTransactionActor(final AbstractShardDataTreeTransaction<?> transaction) {
723 // No-op for free-standing transactions
727 final ShardDataTreeCohort finishTransaction(final ReadWriteShardDataTreeTransaction transaction,
728 final Optional<SortedSet<String>> participatingShardNames) {
729 final DataTreeModification snapshot = transaction.getSnapshot();
730 final TransactionIdentifier id = transaction.getIdentifier();
731 LOG.debug("{}: readying transaction {}", logContext, id);
733 LOG.debug("{}: transaction {} ready", logContext, id);
735 return createReadyCohort(transaction.getIdentifier(), snapshot, participatingShardNames);
738 final void purgeTransaction(final TransactionIdentifier id, final Runnable callback) {
739 LOG.debug("{}: purging transaction {}", logContext, id);
740 replicatePayload(id, PurgeTransactionPayload.create(
741 id, shard.getDatastoreContext().getInitialPayloadSerializedBufferCapacity()), callback);
745 public final Optional<NormalizedNode> readNode(final YangInstanceIdentifier path) {
746 return dataTree.takeSnapshot().readNode(path);
749 final DataTreeSnapshot takeSnapshot() {
750 return dataTree.takeSnapshot();
754 final DataTreeModification newModification() {
755 return dataTree.takeSnapshot().newModification();
758 final Collection<ShardDataTreeCohort> getAndClearPendingTransactions() {
759 Collection<ShardDataTreeCohort> ret = new ArrayList<>(getQueueSize());
761 for (CommitEntry entry: pendingFinishCommits) {
762 ret.add(entry.cohort);
765 for (CommitEntry entry: pendingCommits) {
766 ret.add(entry.cohort);
769 for (CommitEntry entry: pendingTransactions) {
770 ret.add(entry.cohort);
773 pendingFinishCommits.clear();
774 pendingCommits.clear();
775 pendingTransactions.clear();
781 * Called some time after {@link #processNextPendingTransaction()} decides to stop processing.
783 final void resumeNextPendingTransaction() {
784 LOG.debug("{}: attempting to resume transaction processing", logContext);
785 processNextPending();
788 @SuppressWarnings("checkstyle:IllegalCatch")
789 private void processNextPendingTransaction() {
790 ++currentTransactionBatch;
791 if (currentTransactionBatch > MAX_TRANSACTION_BATCH) {
792 LOG.debug("{}: Already processed {}, scheduling continuation", logContext, currentTransactionBatch);
793 shard.scheduleNextPendingTransaction();
797 processNextPending(pendingTransactions, State.CAN_COMMIT_PENDING, entry -> {
798 final SimpleShardDataTreeCohort cohort = entry.cohort;
799 final DataTreeModification modification = cohort.getDataTreeModification();
801 LOG.debug("{}: Validating transaction {}", logContext, cohort.getIdentifier());
804 tip.validate(modification);
805 LOG.debug("{}: Transaction {} validated", logContext, cohort.getIdentifier());
806 cohort.successfulCanCommit();
807 entry.lastAccess = readTime();
809 } catch (ConflictingModificationAppliedException e) {
810 LOG.warn("{}: Store Tx {}: Conflicting modification for path {}.", logContext, cohort.getIdentifier(),
812 cause = new OptimisticLockFailedException("Optimistic lock failed for path " + e.getPath(), e);
813 } catch (DataValidationFailedException e) {
814 LOG.warn("{}: Store Tx {}: Data validation failed for path {}.", logContext, cohort.getIdentifier(),
817 // For debugging purposes, allow dumping of the modification. Coupled with the above
818 // precondition log, it should allow us to understand what went on.
819 LOG.debug("{}: Store Tx {}: modifications: {}", logContext, cohort.getIdentifier(), modification);
820 LOG.trace("{}: Current tree: {}", logContext, dataTree);
821 cause = new TransactionCommitFailedException("Data did not pass validation for path " + e.getPath(), e);
822 } catch (Exception e) {
823 LOG.warn("{}: Unexpected failure in validation phase", logContext, e);
827 // Failure path: propagate the failure, remove the transaction from the queue and loop to the next one
828 pendingTransactions.poll().cohort.failedCanCommit(cause);
832 private void processNextPending() {
833 processNextPendingCommit();
834 processNextPendingTransaction();
837 private void processNextPending(final Queue<CommitEntry> queue, final State allowedState,
838 final Consumer<CommitEntry> processor) {
839 while (!queue.isEmpty()) {
840 final CommitEntry entry = queue.peek();
841 final SimpleShardDataTreeCohort cohort = entry.cohort;
843 if (cohort.isFailed()) {
844 LOG.debug("{}: Removing failed transaction {}", logContext, cohort.getIdentifier());
849 if (cohort.getState() == allowedState) {
850 processor.accept(entry);
856 maybeRunOperationOnPendingTransactionsComplete();
859 private void processNextPendingCommit() {
860 processNextPending(pendingCommits, State.COMMIT_PENDING,
861 entry -> startCommit(entry.cohort, entry.cohort.getCandidate()));
864 private boolean peekNextPendingCommit() {
865 final CommitEntry first = pendingCommits.peek();
866 return first != null && first.cohort.getState() == State.COMMIT_PENDING;
869 // non-final for mocking
870 void startCanCommit(final SimpleShardDataTreeCohort cohort) {
871 final CommitEntry head = pendingTransactions.peek();
873 LOG.warn("{}: No transactions enqueued while attempting to start canCommit on {}", logContext, cohort);
876 if (!cohort.equals(head.cohort)) {
877 // The tx isn't at the head of the queue so we can't start canCommit at this point. Here we check if this
878 // tx should be moved ahead of other tx's in the READY state in the pendingTransactions queue. If this tx
879 // has other participating shards, it could deadlock with other tx's accessing the same shards
880 // depending on the order the tx's are readied on each shard
881 // (see https://jira.opendaylight.org/browse/CONTROLLER-1836). Therefore, if the preceding participating
882 // shard names for a preceding pending tx, call it A, in the queue matches that of this tx, then this tx
883 // is allowed to be moved ahead of tx A in the queue so it is processed first to avoid potential deadlock
884 // if tx A is behind this tx in the pendingTransactions queue for a preceding shard. In other words, since
885 // canCommmit for this tx was requested before tx A, honor that request. If this tx is moved to the head of
886 // the queue as a result, then proceed with canCommit.
888 Collection<String> precedingShardNames = extractPrecedingShardNames(cohort.getParticipatingShardNames());
889 if (precedingShardNames.isEmpty()) {
890 LOG.debug("{}: Tx {} is scheduled for canCommit step", logContext, cohort.getIdentifier());
894 LOG.debug("{}: Evaluating tx {} for canCommit - preceding participating shard names {}",
895 logContext, cohort.getIdentifier(), precedingShardNames);
896 final Iterator<CommitEntry> iter = pendingTransactions.iterator();
898 int moveToIndex = -1;
899 while (iter.hasNext()) {
900 final CommitEntry entry = iter.next();
903 if (cohort.equals(entry.cohort)) {
904 if (moveToIndex < 0) {
905 LOG.debug("{}: Not moving tx {} - cannot proceed with canCommit",
906 logContext, cohort.getIdentifier());
910 LOG.debug("{}: Moving {} to index {} in the pendingTransactions queue",
911 logContext, cohort.getIdentifier(), moveToIndex);
913 insertEntry(pendingTransactions, entry, moveToIndex);
915 if (!cohort.equals(pendingTransactions.peek().cohort)) {
916 LOG.debug("{}: Tx {} is not at the head of the queue - cannot proceed with canCommit",
917 logContext, cohort.getIdentifier());
921 LOG.debug("{}: Tx {} is now at the head of the queue - proceeding with canCommit",
922 logContext, cohort.getIdentifier());
926 if (entry.cohort.getState() != State.READY) {
927 LOG.debug("{}: Skipping pending transaction {} in state {}",
928 logContext, entry.cohort.getIdentifier(), entry.cohort.getState());
932 final Collection<String> pendingPrecedingShardNames = extractPrecedingShardNames(
933 entry.cohort.getParticipatingShardNames());
935 if (precedingShardNames.equals(pendingPrecedingShardNames)) {
936 if (moveToIndex < 0) {
937 LOG.debug("{}: Preceding shard names {} for pending tx {} match - saving moveToIndex {}",
938 logContext, pendingPrecedingShardNames, entry.cohort.getIdentifier(), index);
942 "{}: Preceding shard names {} for pending tx {} match but moveToIndex already set to {}",
943 logContext, pendingPrecedingShardNames, entry.cohort.getIdentifier(), moveToIndex);
946 LOG.debug("{}: Preceding shard names {} for pending tx {} differ - skipping",
947 logContext, pendingPrecedingShardNames, entry.cohort.getIdentifier());
952 processNextPendingTransaction();
955 private static void insertEntry(final Deque<CommitEntry> queue, final CommitEntry entry, final int atIndex) {
957 queue.addFirst(entry);
961 LOG.trace("Inserting into Deque at index {}", atIndex);
963 Deque<CommitEntry> tempStack = new ArrayDeque<>(atIndex);
964 for (int i = 0; i < atIndex; i++) {
965 tempStack.push(queue.poll());
968 queue.addFirst(entry);
970 tempStack.forEach(queue::addFirst);
973 private Collection<String> extractPrecedingShardNames(final Optional<SortedSet<String>> participatingShardNames) {
974 return participatingShardNames.map((Function<SortedSet<String>, Collection<String>>)
975 set -> set.headSet(shard.getShardName())).orElse(Collections.<String>emptyList());
978 private void failPreCommit(final Throwable cause) {
979 shard.getShardMBean().incrementFailedTransactionsCount();
980 pendingTransactions.poll().cohort.failedPreCommit(cause);
981 processNextPendingTransaction();
984 // non-final for mocking
985 @SuppressWarnings("checkstyle:IllegalCatch")
986 void startPreCommit(final SimpleShardDataTreeCohort cohort) {
987 final CommitEntry entry = pendingTransactions.peek();
988 checkState(entry != null, "Attempted to pre-commit of %s when no transactions pending", cohort);
990 final SimpleShardDataTreeCohort current = entry.cohort;
991 verify(cohort.equals(current), "Attempted to pre-commit %s while %s is pending", cohort, current);
993 final TransactionIdentifier currentId = current.getIdentifier();
994 LOG.debug("{}: Preparing transaction {}", logContext, currentId);
996 final DataTreeCandidateTip candidate;
998 candidate = tip.prepare(cohort.getDataTreeModification());
999 LOG.debug("{}: Transaction {} candidate ready", logContext, currentId);
1000 } catch (DataValidationFailedException | RuntimeException e) {
1005 cohort.userPreCommit(candidate, new FutureCallback<Void>() {
1007 public void onSuccess(final Void noop) {
1008 // Set the tip of the data tree.
1009 tip = verifyNotNull(candidate);
1011 entry.lastAccess = readTime();
1013 pendingTransactions.remove();
1014 pendingCommits.add(entry);
1016 LOG.debug("{}: Transaction {} prepared", logContext, currentId);
1018 cohort.successfulPreCommit(candidate);
1020 processNextPendingTransaction();
1024 public void onFailure(final Throwable failure) {
1025 failPreCommit(failure);
1030 private void failCommit(final Exception cause) {
1031 shard.getShardMBean().incrementFailedTransactionsCount();
1032 pendingFinishCommits.poll().cohort.failedCommit(cause);
1033 processNextPending();
1036 @SuppressWarnings("checkstyle:IllegalCatch")
1037 private void finishCommit(final SimpleShardDataTreeCohort cohort) {
1038 final TransactionIdentifier txId = cohort.getIdentifier();
1039 final DataTreeCandidate candidate = cohort.getCandidate();
1041 LOG.debug("{}: Resuming commit of transaction {}", logContext, txId);
1043 if (tip == candidate) {
1044 // All pending candidates have been committed, reset the tip to the data tree.
1049 dataTree.commit(candidate);
1050 } catch (Exception e) {
1051 LOG.error("{}: Failed to commit transaction {}", logContext, txId, e);
1056 allMetadataCommittedTransaction(txId);
1057 shard.getShardMBean().incrementCommittedTransactionCount();
1058 shard.getShardMBean().setLastCommittedTransactionTime(System.currentTimeMillis());
1060 // FIXME: propagate journal index
1061 pendingFinishCommits.poll().cohort.successfulCommit(UnsignedLong.ZERO, () -> {
1062 LOG.trace("{}: Transaction {} committed, proceeding to notify", logContext, txId);
1063 notifyListeners(candidate);
1065 processNextPending();
1069 // non-final for mocking
1070 void startCommit(final SimpleShardDataTreeCohort cohort, final DataTreeCandidate candidate) {
1071 final CommitEntry entry = pendingCommits.peek();
1072 checkState(entry != null, "Attempted to start commit of %s when no transactions pending", cohort);
1074 final SimpleShardDataTreeCohort current = entry.cohort;
1075 if (!cohort.equals(current)) {
1076 LOG.debug("{}: Transaction {} scheduled for commit step", logContext, cohort.getIdentifier());
1080 LOG.debug("{}: Starting commit for transaction {}", logContext, current.getIdentifier());
1082 final TransactionIdentifier txId = cohort.getIdentifier();
1083 final Payload payload;
1085 payload = CommitTransactionPayload.create(txId, candidate, PayloadVersion.current(),
1086 shard.getDatastoreContext().getInitialPayloadSerializedBufferCapacity());
1087 } catch (IOException e) {
1088 LOG.error("{}: Failed to encode transaction {} candidate {}", logContext, txId, candidate, e);
1089 pendingCommits.poll().cohort.failedCommit(e);
1090 processNextPending();
1094 // We process next transactions pending canCommit before we call persistPayload to possibly progress subsequent
1095 // transactions to the COMMIT_PENDING state so the payloads can be batched for replication. This is done for
1096 // single-shard transactions that immediately transition from canCommit to preCommit to commit. Note that
1097 // if the next pending transaction is progressed to COMMIT_PENDING and this method (startCommit) is called,
1098 // the next transaction will not attempt to replicate b/c the current transaction is still at the head of the
1099 // pendingCommits queue.
1100 processNextPendingTransaction();
1102 // After processing next pending transactions, we can now remove the current transaction from pendingCommits.
1103 // Note this must be done before the call to peekNextPendingCommit below so we check the next transaction
1104 // in order to properly determine the batchHint flag for the call to persistPayload.
1105 pendingCommits.remove();
1106 pendingFinishCommits.add(entry);
1108 // See if the next transaction is pending commit (ie in the COMMIT_PENDING state) so it can be batched with
1109 // this transaction for replication.
1110 boolean replicationBatchHint = peekNextPendingCommit();
1112 // Once completed, we will continue via payloadReplicationComplete
1113 shard.persistPayload(txId, payload, replicationBatchHint);
1115 entry.lastAccess = shard.ticker().read();
1117 LOG.debug("{}: Transaction {} submitted to persistence", logContext, txId);
1119 // Process the next transaction pending commit, if any. If there is one it will be batched with this
1120 // transaction for replication.
1121 processNextPendingCommit();
1124 final Collection<ActorRef> getCohortActors() {
1125 return cohortRegistry.getCohortActors();
1128 final void processCohortRegistryCommand(final ActorRef sender, final CohortRegistryCommand message) {
1129 cohortRegistry.process(sender, message);
1133 final ShardDataTreeCohort createFailedCohort(final TransactionIdentifier txId, final DataTreeModification mod,
1134 final Exception failure) {
1135 final SimpleShardDataTreeCohort cohort = new SimpleShardDataTreeCohort(this, mod, txId, failure);
1136 pendingTransactions.add(new CommitEntry(cohort, readTime()));
1141 final ShardDataTreeCohort createReadyCohort(final TransactionIdentifier txId, final DataTreeModification mod,
1142 final Optional<SortedSet<String>> participatingShardNames) {
1143 SimpleShardDataTreeCohort cohort = new SimpleShardDataTreeCohort(this, mod, txId,
1144 cohortRegistry.createCohort(schemaContext, txId, shard::executeInSelf,
1145 COMMIT_STEP_TIMEOUT), participatingShardNames);
1146 pendingTransactions.add(new CommitEntry(cohort, readTime()));
1150 // Exposed for ShardCommitCoordinator so it does not have deal with local histories (it does not care), this mimics
1151 // the newReadWriteTransaction()
1152 final ShardDataTreeCohort newReadyCohort(final TransactionIdentifier txId, final DataTreeModification mod,
1153 final Optional<SortedSet<String>> participatingShardNames) {
1154 if (txId.getHistoryId().getHistoryId() == 0) {
1155 return createReadyCohort(txId, mod, participatingShardNames);
1158 return ensureTransactionChain(txId.getHistoryId(), null).createReadyCohort(txId, mod, participatingShardNames);
1161 @SuppressFBWarnings(value = "DB_DUPLICATE_SWITCH_CLAUSES", justification = "See inline comments below.")
1162 final void checkForExpiredTransactions(final long transactionCommitTimeoutMillis,
1163 final Function<SimpleShardDataTreeCohort, OptionalLong> accessTimeUpdater) {
1164 final long timeout = TimeUnit.MILLISECONDS.toNanos(transactionCommitTimeoutMillis);
1165 final long now = readTime();
1167 final Queue<CommitEntry> currentQueue = !pendingFinishCommits.isEmpty() ? pendingFinishCommits :
1168 !pendingCommits.isEmpty() ? pendingCommits : pendingTransactions;
1169 final CommitEntry currentTx = currentQueue.peek();
1170 if (currentTx == null) {
1171 // Empty queue, no-op
1175 long delta = now - currentTx.lastAccess;
1176 if (delta < timeout) {
1177 // Not expired yet, bail
1181 final OptionalLong updateOpt = accessTimeUpdater.apply(currentTx.cohort);
1182 if (updateOpt.isPresent()) {
1183 final long newAccess = updateOpt.getAsLong();
1184 final long newDelta = now - newAccess;
1185 if (newDelta < delta) {
1186 LOG.debug("{}: Updated current transaction {} access time", logContext,
1187 currentTx.cohort.getIdentifier());
1188 currentTx.lastAccess = newAccess;
1192 if (delta < timeout) {
1193 // Not expired yet, bail
1198 final long deltaMillis = TimeUnit.NANOSECONDS.toMillis(delta);
1199 final State state = currentTx.cohort.getState();
1201 LOG.warn("{}: Current transaction {} has timed out after {} ms in state {}", logContext,
1202 currentTx.cohort.getIdentifier(), deltaMillis, state);
1203 boolean processNext = true;
1204 final TimeoutException cohortFailure = new TimeoutException("Backend timeout in state " + state + " after "
1205 + deltaMillis + "ms");
1208 case CAN_COMMIT_PENDING:
1209 currentQueue.remove().cohort.failedCanCommit(cohortFailure);
1211 case CAN_COMMIT_COMPLETE:
1212 // The suppression of the FindBugs "DB_DUPLICATE_SWITCH_CLAUSES" warning pertains to this clause
1213 // whose code is duplicated with PRE_COMMIT_COMPLETE. The clauses aren't combined in case the code
1214 // in PRE_COMMIT_COMPLETE is changed.
1215 currentQueue.remove().cohort.reportFailure(cohortFailure);
1217 case PRE_COMMIT_PENDING:
1218 currentQueue.remove().cohort.failedPreCommit(cohortFailure);
1220 case PRE_COMMIT_COMPLETE:
1221 // FIXME: this is a legacy behavior problem. Three-phase commit protocol specifies that after we
1222 // are ready we should commit the transaction, not abort it. Our current software stack does
1223 // not allow us to do that consistently, because we persist at the time of commit, hence
1224 // we can end up in a state where we have pre-committed a transaction, then a leader failover
1225 // occurred ... the new leader does not see the pre-committed transaction and does not have
1226 // a running timer. To fix this we really need two persistence events.
1228 // The first one, done at pre-commit time will hold the transaction payload. When consensus
1229 // is reached, we exit the pre-commit phase and start the pre-commit timer. Followers do not
1230 // apply the state in this event.
1232 // The second one, done at commit (or abort) time holds only the transaction identifier and
1233 // signals to followers that the state should (or should not) be applied.
1235 // In order to make the pre-commit timer working across failovers, though, we need
1236 // a per-shard cluster-wide monotonic time, so a follower becoming the leader can accurately
1237 // restart the timer.
1238 currentQueue.remove().cohort.reportFailure(cohortFailure);
1240 case COMMIT_PENDING:
1241 LOG.warn("{}: Transaction {} is still committing, cannot abort", logContext,
1242 currentTx.cohort.getIdentifier());
1243 currentTx.lastAccess = now;
1244 processNext = false;
1247 currentQueue.remove().cohort.reportFailure(cohortFailure);
1253 currentQueue.remove();
1257 processNextPending();
1261 // non-final for mocking
1262 boolean startAbort(final SimpleShardDataTreeCohort cohort) {
1263 final Iterator<CommitEntry> it = Iterables.concat(pendingFinishCommits, pendingCommits,
1264 pendingTransactions).iterator();
1265 if (!it.hasNext()) {
1266 LOG.debug("{}: no open transaction while attempting to abort {}", logContext, cohort.getIdentifier());
1270 // First entry is special, as it may already be committing
1271 final CommitEntry first = it.next();
1272 if (cohort.equals(first.cohort)) {
1273 if (cohort.getState() != State.COMMIT_PENDING) {
1274 LOG.debug("{}: aborting head of queue {} in state {}", logContext, cohort.getIdentifier(),
1275 cohort.getIdentifier());
1278 if (cohort.getCandidate() != null) {
1279 rebaseTransactions(it, dataTree);
1282 processNextPending();
1286 LOG.warn("{}: transaction {} is committing, skipping abort", logContext, cohort.getIdentifier());
1290 DataTreeTip newTip = requireNonNullElse(first.cohort.getCandidate(), dataTree);
1291 while (it.hasNext()) {
1292 final CommitEntry e = it.next();
1293 if (cohort.equals(e.cohort)) {
1294 LOG.debug("{}: aborting queued transaction {}", logContext, cohort.getIdentifier());
1297 if (cohort.getCandidate() != null) {
1298 rebaseTransactions(it, newTip);
1303 newTip = requireNonNullElse(e.cohort.getCandidate(), newTip);
1307 LOG.debug("{}: aborted transaction {} not found in the queue", logContext, cohort.getIdentifier());
1311 @SuppressWarnings("checkstyle:IllegalCatch")
1312 private void rebaseTransactions(final Iterator<CommitEntry> iter, final @NonNull DataTreeTip newTip) {
1313 tip = requireNonNull(newTip);
1314 while (iter.hasNext()) {
1315 final SimpleShardDataTreeCohort cohort = iter.next().cohort;
1316 if (cohort.getState() == State.CAN_COMMIT_COMPLETE) {
1317 LOG.debug("{}: Revalidating queued transaction {}", logContext, cohort.getIdentifier());
1320 tip.validate(cohort.getDataTreeModification());
1321 } catch (DataValidationFailedException | RuntimeException e) {
1322 LOG.debug("{}: Failed to revalidate queued transaction {}", logContext, cohort.getIdentifier(), e);
1323 cohort.reportFailure(e);
1325 } else if (cohort.getState() == State.PRE_COMMIT_COMPLETE) {
1326 LOG.debug("{}: Repreparing queued transaction {}", logContext, cohort.getIdentifier());
1329 tip.validate(cohort.getDataTreeModification());
1330 DataTreeCandidateTip candidate = tip.prepare(cohort.getDataTreeModification());
1332 cohort.setNewCandidate(candidate);
1334 } catch (RuntimeException | DataValidationFailedException e) {
1335 LOG.debug("{}: Failed to reprepare queued transaction {}", logContext, cohort.getIdentifier(), e);
1336 cohort.reportFailure(e);
1342 final void setRunOnPendingTransactionsComplete(final Runnable operation) {
1343 runOnPendingTransactionsComplete = operation;
1344 maybeRunOperationOnPendingTransactionsComplete();
1347 private void maybeRunOperationOnPendingTransactionsComplete() {
1348 if (runOnPendingTransactionsComplete != null && !anyPendingTransactions()) {
1349 LOG.debug("{}: Pending transactions complete - running operation {}", logContext,
1350 runOnPendingTransactionsComplete);
1352 runOnPendingTransactionsComplete.run();
1353 runOnPendingTransactionsComplete = null;
1357 final ShardStats getStats() {
1358 return shard.getShardMBean();
1361 final Iterator<SimpleShardDataTreeCohort> cohortIterator() {
1362 return Iterables.transform(Iterables.concat(pendingFinishCommits, pendingCommits, pendingTransactions),
1363 e -> e.cohort).iterator();
1366 final void removeTransactionChain(final LocalHistoryIdentifier id) {
1367 if (transactionChains.remove(id) != null) {
1368 LOG.debug("{}: Removed transaction chain {}", logContext, id);