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 com.google.common.base.Preconditions.checkState;
11 import static com.google.common.base.Verify.verify;
12 import static com.google.common.base.Verify.verifyNotNull;
13 import static java.util.Objects.requireNonNull;
15 import akka.actor.ActorRef;
16 import akka.util.Timeout;
17 import com.google.common.annotations.VisibleForTesting;
18 import com.google.common.base.MoreObjects;
19 import com.google.common.base.Stopwatch;
20 import com.google.common.collect.ImmutableList;
21 import com.google.common.collect.ImmutableMap;
22 import com.google.common.collect.ImmutableMap.Builder;
23 import com.google.common.collect.Iterables;
24 import com.google.common.primitives.UnsignedLong;
25 import com.google.common.util.concurrent.FutureCallback;
26 import edu.umd.cs.findbugs.annotations.SuppressFBWarnings;
28 import java.io.IOException;
29 import java.util.ArrayDeque;
30 import java.util.ArrayList;
31 import java.util.Collection;
32 import java.util.Collections;
33 import java.util.Deque;
34 import java.util.HashMap;
35 import java.util.Iterator;
37 import java.util.Map.Entry;
38 import java.util.Optional;
39 import java.util.OptionalLong;
40 import java.util.Queue;
41 import java.util.SortedSet;
42 import java.util.concurrent.TimeUnit;
43 import java.util.concurrent.TimeoutException;
44 import java.util.function.Consumer;
45 import java.util.function.Function;
46 import java.util.function.UnaryOperator;
47 import org.eclipse.jdt.annotation.NonNull;
48 import org.eclipse.jdt.annotation.Nullable;
49 import org.opendaylight.controller.cluster.access.concepts.LocalHistoryIdentifier;
50 import org.opendaylight.controller.cluster.access.concepts.TransactionIdentifier;
51 import org.opendaylight.controller.cluster.datastore.DataTreeCohortActorRegistry.CohortRegistryCommand;
52 import org.opendaylight.controller.cluster.datastore.ShardDataTreeCohort.State;
53 import org.opendaylight.controller.cluster.datastore.jmx.mbeans.shard.ShardStats;
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.protobuff.client.messages.Payload;
71 import org.opendaylight.mdsal.common.api.OptimisticLockFailedException;
72 import org.opendaylight.mdsal.common.api.TransactionCommitFailedException;
73 import org.opendaylight.mdsal.dom.api.DOMDataTreeChangeListener;
74 import org.opendaylight.yangtools.concepts.Identifier;
75 import org.opendaylight.yangtools.concepts.ListenerRegistration;
76 import org.opendaylight.yangtools.yang.data.api.YangInstanceIdentifier;
77 import org.opendaylight.yangtools.yang.data.api.schema.NormalizedNode;
78 import org.opendaylight.yangtools.yang.data.api.schema.tree.ConflictingModificationAppliedException;
79 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTree;
80 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeCandidate;
81 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeCandidateTip;
82 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeCandidates;
83 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeConfiguration;
84 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeModification;
85 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeSnapshot;
86 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeTip;
87 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataValidationFailedException;
88 import org.opendaylight.yangtools.yang.data.api.schema.tree.TreeType;
89 import org.opendaylight.yangtools.yang.data.codec.binfmt.NormalizedNodeStreamVersion;
90 import org.opendaylight.yangtools.yang.data.impl.schema.tree.InMemoryDataTreeFactory;
91 import org.opendaylight.yangtools.yang.data.util.DataSchemaContextTree;
92 import org.opendaylight.yangtools.yang.model.api.SchemaContext;
93 import org.slf4j.Logger;
94 import org.slf4j.LoggerFactory;
95 import scala.concurrent.duration.FiniteDuration;
98 * Internal shard state, similar to a DOMStore, but optimized for use in the actor system, e.g. it does not expose
99 * public interfaces and assumes it is only ever called from a single thread.
102 * This class is not part of the API contract and is subject to change at any time. It is NOT thread-safe.
104 public class ShardDataTree extends ShardDataTreeTransactionParent {
105 private static final class CommitEntry {
106 final SimpleShardDataTreeCohort cohort;
109 CommitEntry(final SimpleShardDataTreeCohort cohort, final long now) {
110 this.cohort = requireNonNull(cohort);
115 public String toString() {
116 return "CommitEntry [tx=" + cohort.getIdentifier() + ", state=" + cohort.getState() + "]";
120 private static final Timeout COMMIT_STEP_TIMEOUT = new Timeout(FiniteDuration.create(5, TimeUnit.SECONDS));
121 private static final Logger LOG = LoggerFactory.getLogger(ShardDataTree.class);
124 * Process this many transactions in a single batched run. If we exceed this limit, we need to schedule later
125 * execution to finish up the batch. This is necessary in case of a long list of transactions which progress
126 * immediately through their preCommit phase -- if that happens, their completion eats up stack frames and could
127 * result in StackOverflowError.
129 private static final int MAX_TRANSACTION_BATCH = 100;
131 private final Map<LocalHistoryIdentifier, ShardDataTreeTransactionChain> transactionChains = new HashMap<>();
132 private final DataTreeCohortActorRegistry cohortRegistry = new DataTreeCohortActorRegistry();
133 private final Deque<CommitEntry> pendingTransactions = new ArrayDeque<>();
134 private final Queue<CommitEntry> pendingCommits = new ArrayDeque<>();
135 private final Queue<CommitEntry> pendingFinishCommits = new ArrayDeque<>();
138 * Callbacks that need to be invoked once a payload is replicated.
140 private final Map<Payload, Runnable> replicationCallbacks = new HashMap<>();
142 private final ShardDataTreeChangeListenerPublisher treeChangeListenerPublisher;
143 private final Collection<ShardDataTreeMetadata<?>> metadata;
144 private final DataTree dataTree;
145 private final String logContext;
146 private final Shard shard;
147 private Runnable runOnPendingTransactionsComplete;
150 * Optimistic {@link DataTreeCandidate} preparation. Since our DataTree implementation is a
151 * {@link DataTree}, each {@link DataTreeCandidate} is also a {@link DataTreeTip}, e.g. another
152 * candidate can be prepared on top of it. They still need to be committed in sequence. Here we track the current
153 * tip of the data tree, which is the last DataTreeCandidate we have in flight, or the DataTree itself.
155 private DataTreeTip tip;
157 private SchemaContext schemaContext;
158 private DataSchemaContextTree dataSchemaContext;
160 private int currentTransactionBatch;
162 ShardDataTree(final Shard shard, final SchemaContext schemaContext, final DataTree dataTree,
163 final ShardDataTreeChangeListenerPublisher treeChangeListenerPublisher,
164 final String logContext,
165 final ShardDataTreeMetadata<?>... metadata) {
166 this.dataTree = requireNonNull(dataTree);
167 updateSchemaContext(schemaContext);
169 this.shard = requireNonNull(shard);
170 this.treeChangeListenerPublisher = requireNonNull(treeChangeListenerPublisher);
171 this.logContext = requireNonNull(logContext);
172 this.metadata = ImmutableList.copyOf(metadata);
176 ShardDataTree(final Shard shard, final SchemaContext schemaContext, final TreeType treeType,
177 final YangInstanceIdentifier root,
178 final ShardDataTreeChangeListenerPublisher treeChangeListenerPublisher,
179 final String logContext,
180 final ShardDataTreeMetadata<?>... metadata) {
181 this(shard, schemaContext, createDataTree(treeType, root), treeChangeListenerPublisher, logContext, metadata);
184 private static DataTree createDataTree(final TreeType treeType, final YangInstanceIdentifier root) {
185 final DataTreeConfiguration baseConfig = DataTreeConfiguration.getDefault(treeType);
186 return new InMemoryDataTreeFactory().create(new DataTreeConfiguration.Builder(baseConfig.getTreeType())
187 .setMandatoryNodesValidation(baseConfig.isMandatoryNodesValidationEnabled())
188 .setUniqueIndexes(baseConfig.isUniqueIndexEnabled())
194 public ShardDataTree(final Shard shard, final SchemaContext schemaContext, final TreeType treeType) {
195 this(shard, schemaContext, treeType, YangInstanceIdentifier.empty(),
196 new DefaultShardDataTreeChangeListenerPublisher(""), "");
199 final String logContext() {
203 final long readTime() {
204 return shard.ticker().read();
207 public DataTree getDataTree() {
211 SchemaContext getSchemaContext() {
212 return schemaContext;
215 void updateSchemaContext(final SchemaContext newSchemaContext) {
216 dataTree.setSchemaContext(newSchemaContext);
217 this.schemaContext = requireNonNull(newSchemaContext);
218 this.dataSchemaContext = DataSchemaContextTree.from(newSchemaContext);
221 void resetTransactionBatch() {
222 currentTransactionBatch = 0;
226 * Take a snapshot of current state for later recovery.
228 * @return A state snapshot
230 @NonNull ShardDataTreeSnapshot takeStateSnapshot() {
231 final NormalizedNode<?, ?> rootNode = dataTree.takeSnapshot().readNode(YangInstanceIdentifier.empty()).get();
232 final Builder<Class<? extends ShardDataTreeSnapshotMetadata<?>>, ShardDataTreeSnapshotMetadata<?>> metaBuilder =
233 ImmutableMap.builder();
235 for (ShardDataTreeMetadata<?> m : metadata) {
236 final ShardDataTreeSnapshotMetadata<?> meta = m.toSnapshot();
238 metaBuilder.put(meta.getType(), meta);
242 return new MetadataShardDataTreeSnapshot(rootNode, metaBuilder.build());
245 private boolean anyPendingTransactions() {
246 return !pendingTransactions.isEmpty() || !pendingCommits.isEmpty() || !pendingFinishCommits.isEmpty();
249 private void applySnapshot(final @NonNull ShardDataTreeSnapshot snapshot,
250 final UnaryOperator<DataTreeModification> wrapper) throws DataValidationFailedException {
251 final Stopwatch elapsed = Stopwatch.createStarted();
253 if (anyPendingTransactions()) {
254 LOG.warn("{}: applying state snapshot with pending transactions", logContext);
257 final Map<Class<? extends ShardDataTreeSnapshotMetadata<?>>, ShardDataTreeSnapshotMetadata<?>> snapshotMeta;
258 if (snapshot instanceof MetadataShardDataTreeSnapshot) {
259 snapshotMeta = ((MetadataShardDataTreeSnapshot) snapshot).getMetadata();
261 snapshotMeta = ImmutableMap.of();
264 for (ShardDataTreeMetadata<?> m : metadata) {
265 final ShardDataTreeSnapshotMetadata<?> s = snapshotMeta.get(m.getSupportedType());
273 final DataTreeModification unwrapped = dataTree.takeSnapshot().newModification();
274 final DataTreeModification mod = wrapper.apply(unwrapped);
275 // delete everything first
276 mod.delete(YangInstanceIdentifier.empty());
278 final Optional<NormalizedNode<?, ?>> maybeNode = snapshot.getRootNode();
279 if (maybeNode.isPresent()) {
280 // Add everything from the remote node back
281 mod.write(YangInstanceIdentifier.empty(), maybeNode.get());
285 dataTree.validate(unwrapped);
286 DataTreeCandidateTip candidate = dataTree.prepare(unwrapped);
287 dataTree.commit(candidate);
288 notifyListeners(candidate);
290 LOG.debug("{}: state snapshot applied in {}", logContext, elapsed);
294 * Apply a snapshot coming from the leader. This method assumes the leader and follower SchemaContexts match and
295 * does not perform any pruning.
297 * @param snapshot Snapshot that needs to be applied
298 * @throws DataValidationFailedException when the snapshot fails to apply
300 void applySnapshot(final @NonNull ShardDataTreeSnapshot snapshot) throws DataValidationFailedException {
301 // TODO: we should be taking ShardSnapshotState here and performing forward-compatibility translation
302 applySnapshot(snapshot, UnaryOperator.identity());
306 * Apply a snapshot coming from recovery. This method does not assume the SchemaContexts match and performs data
307 * pruning in an attempt to adjust the state to our current SchemaContext.
309 * @param snapshot Snapshot that needs to be applied
310 * @throws DataValidationFailedException when the snapshot fails to apply
312 void applyRecoverySnapshot(final @NonNull ShardSnapshotState snapshot) throws DataValidationFailedException {
313 // TODO: we should be able to reuse the pruner, provided we are not reentrant
314 ReusableNormalizedNodePruner pruner = ReusableNormalizedNodePruner.forDataSchemaContext(dataSchemaContext);
315 if (snapshot.needsMigration()) {
316 pruner = pruner.withUintAdaption();
320 final ReusableNormalizedNodePruner finalPruner = pruner;
321 applySnapshot(snapshot.getSnapshot(),
322 delegate -> new PruningDataTreeModification(delegate, dataTree, finalPruner));
325 @SuppressWarnings("checkstyle:IllegalCatch")
326 private void applyRecoveryCandidate(final CommitTransactionPayload payload) throws IOException {
327 final Entry<TransactionIdentifier, DataTreeCandidateWithVersion> entry = payload.getCandidate();
328 final DataTreeModification unwrapped = dataTree.takeSnapshot().newModification();
330 // TODO: we should be able to reuse the pruner, provided we are not reentrant
331 ReusableNormalizedNodePruner pruner = ReusableNormalizedNodePruner.forDataSchemaContext(dataSchemaContext);
332 if (NormalizedNodeStreamVersion.MAGNESIUM.compareTo(entry.getValue().getVersion()) > 0) {
333 pruner = pruner.withUintAdaption();
336 final PruningDataTreeModification mod = new PruningDataTreeModification(unwrapped, dataTree, pruner);
337 DataTreeCandidates.applyToModification(mod, entry.getValue().getCandidate());
340 LOG.trace("{}: Applying recovery modification {}", logContext, unwrapped);
343 dataTree.validate(unwrapped);
344 dataTree.commit(dataTree.prepare(unwrapped));
345 } catch (Exception e) {
346 File file = new File(System.getProperty("karaf.data", "."),
347 "failed-recovery-payload-" + logContext + ".out");
348 DataTreeModificationOutput.toFile(file, unwrapped);
349 throw new IllegalStateException(String.format(
350 "%s: Failed to apply recovery payload. Modification data was written to file %s",
351 logContext, file), e);
354 allMetadataCommittedTransaction(entry.getKey());
358 * Apply a payload coming from recovery. This method does not assume the SchemaContexts match and performs data
359 * pruning in an attempt to adjust the state to our current SchemaContext.
361 * @param payload Payload
362 * @throws IOException when the snapshot fails to deserialize
363 * @throws DataValidationFailedException when the snapshot fails to apply
365 void applyRecoveryPayload(final @NonNull Payload payload) throws IOException {
366 if (payload instanceof CommitTransactionPayload) {
367 applyRecoveryCandidate((CommitTransactionPayload) payload);
368 } else if (payload instanceof AbortTransactionPayload) {
369 allMetadataAbortedTransaction(((AbortTransactionPayload) payload).getIdentifier());
370 } else if (payload instanceof PurgeTransactionPayload) {
371 allMetadataPurgedTransaction(((PurgeTransactionPayload) payload).getIdentifier());
372 } else if (payload instanceof CreateLocalHistoryPayload) {
373 allMetadataCreatedLocalHistory(((CreateLocalHistoryPayload) payload).getIdentifier());
374 } else if (payload instanceof CloseLocalHistoryPayload) {
375 allMetadataClosedLocalHistory(((CloseLocalHistoryPayload) payload).getIdentifier());
376 } else if (payload instanceof PurgeLocalHistoryPayload) {
377 allMetadataPurgedLocalHistory(((PurgeLocalHistoryPayload) payload).getIdentifier());
379 LOG.debug("{}: ignoring unhandled payload {}", logContext, payload);
383 private void applyReplicatedCandidate(final CommitTransactionPayload payload)
384 throws DataValidationFailedException, IOException {
385 final Entry<TransactionIdentifier, DataTreeCandidateWithVersion> entry = payload.getCandidate();
386 final TransactionIdentifier identifier = entry.getKey();
387 LOG.debug("{}: Applying foreign transaction {}", logContext, identifier);
389 final DataTreeModification mod = dataTree.takeSnapshot().newModification();
390 // TODO: check version here, which will enable us to perform forward-compatibility transformations
391 DataTreeCandidates.applyToModification(mod, entry.getValue().getCandidate());
394 LOG.trace("{}: Applying foreign modification {}", logContext, mod);
395 dataTree.validate(mod);
396 final DataTreeCandidate candidate = dataTree.prepare(mod);
397 dataTree.commit(candidate);
399 allMetadataCommittedTransaction(identifier);
400 notifyListeners(candidate);
404 * Apply a payload coming from the leader, which could actually be us. This method assumes the leader and follower
405 * SchemaContexts match and does not perform any pruning.
407 * @param identifier Payload identifier as returned from RaftActor
408 * @param payload Payload
409 * @throws IOException when the snapshot fails to deserialize
410 * @throws DataValidationFailedException when the snapshot fails to apply
412 void applyReplicatedPayload(final Identifier identifier, final Payload payload) throws IOException,
413 DataValidationFailedException {
415 * 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
416 * if we are the leader and it has originated with us.
418 * The identifier will only ever be non-null when we were the leader which achieved consensus. Unfortunately,
419 * though, this may not be the case anymore, as we are being called some time afterwards and we may not be
420 * acting in that capacity anymore.
422 * In any case, we know that this is an entry coming from replication, hence we can be sure we will not observe
423 * pre-Boron state -- which limits the number of options here.
425 if (payload instanceof CommitTransactionPayload) {
426 if (identifier == null) {
427 applyReplicatedCandidate((CommitTransactionPayload) payload);
429 verify(identifier instanceof TransactionIdentifier);
430 payloadReplicationComplete((TransactionIdentifier) identifier);
432 } else if (payload instanceof AbortTransactionPayload) {
433 if (identifier != null) {
434 payloadReplicationComplete((AbortTransactionPayload) payload);
436 allMetadataAbortedTransaction(((AbortTransactionPayload) payload).getIdentifier());
437 } else if (payload instanceof PurgeTransactionPayload) {
438 if (identifier != null) {
439 payloadReplicationComplete((PurgeTransactionPayload) payload);
441 allMetadataPurgedTransaction(((PurgeTransactionPayload) payload).getIdentifier());
442 } else if (payload instanceof CloseLocalHistoryPayload) {
443 if (identifier != null) {
444 payloadReplicationComplete((CloseLocalHistoryPayload) payload);
446 allMetadataClosedLocalHistory(((CloseLocalHistoryPayload) payload).getIdentifier());
447 } else if (payload instanceof CreateLocalHistoryPayload) {
448 if (identifier != null) {
449 payloadReplicationComplete((CreateLocalHistoryPayload)payload);
451 allMetadataCreatedLocalHistory(((CreateLocalHistoryPayload) payload).getIdentifier());
452 } else if (payload instanceof PurgeLocalHistoryPayload) {
453 if (identifier != null) {
454 payloadReplicationComplete((PurgeLocalHistoryPayload)payload);
456 allMetadataPurgedLocalHistory(((PurgeLocalHistoryPayload) payload).getIdentifier());
458 LOG.warn("{}: ignoring unhandled identifier {} payload {}", logContext, identifier, payload);
462 private void replicatePayload(final Identifier id, final Payload payload, final @Nullable Runnable callback) {
463 if (callback != null) {
464 replicationCallbacks.put(payload, callback);
466 shard.persistPayload(id, payload, true);
469 private void payloadReplicationComplete(final AbstractIdentifiablePayload<?> payload) {
470 final Runnable callback = replicationCallbacks.remove(payload);
471 if (callback != null) {
472 LOG.debug("{}: replication of {} completed, invoking {}", logContext, payload.getIdentifier(), callback);
475 LOG.debug("{}: replication of {} has no callback", logContext, payload.getIdentifier());
479 private void payloadReplicationComplete(final TransactionIdentifier txId) {
480 final CommitEntry current = pendingFinishCommits.peek();
481 if (current == null) {
482 LOG.warn("{}: No outstanding transactions, ignoring consensus on transaction {}", logContext, txId);
483 allMetadataCommittedTransaction(txId);
487 if (!current.cohort.getIdentifier().equals(txId)) {
488 LOG.debug("{}: Head of pendingFinishCommits queue is {}, ignoring consensus on transaction {}", logContext,
489 current.cohort.getIdentifier(), txId);
490 allMetadataCommittedTransaction(txId);
494 finishCommit(current.cohort);
497 private void allMetadataAbortedTransaction(final TransactionIdentifier txId) {
498 for (ShardDataTreeMetadata<?> m : metadata) {
499 m.onTransactionAborted(txId);
503 private void allMetadataCommittedTransaction(final TransactionIdentifier txId) {
504 for (ShardDataTreeMetadata<?> m : metadata) {
505 m.onTransactionCommitted(txId);
509 private void allMetadataPurgedTransaction(final TransactionIdentifier txId) {
510 for (ShardDataTreeMetadata<?> m : metadata) {
511 m.onTransactionPurged(txId);
515 private void allMetadataCreatedLocalHistory(final LocalHistoryIdentifier historyId) {
516 for (ShardDataTreeMetadata<?> m : metadata) {
517 m.onHistoryCreated(historyId);
521 private void allMetadataClosedLocalHistory(final LocalHistoryIdentifier historyId) {
522 for (ShardDataTreeMetadata<?> m : metadata) {
523 m.onHistoryClosed(historyId);
527 private void allMetadataPurgedLocalHistory(final LocalHistoryIdentifier historyId) {
528 for (ShardDataTreeMetadata<?> m : metadata) {
529 m.onHistoryPurged(historyId);
534 * Create a transaction chain for specified history. Unlike {@link #ensureTransactionChain(LocalHistoryIdentifier)},
535 * this method is used for re-establishing state when we are taking over
537 * @param historyId Local history identifier
538 * @param closed True if the chain should be created in closed state (i.e. pending purge)
539 * @return Transaction chain handle
541 ShardDataTreeTransactionChain recreateTransactionChain(final LocalHistoryIdentifier historyId,
542 final boolean closed) {
543 final ShardDataTreeTransactionChain ret = new ShardDataTreeTransactionChain(historyId, this);
544 final ShardDataTreeTransactionChain existing = transactionChains.putIfAbsent(historyId, ret);
545 checkState(existing == null, "Attempted to recreate chain %s, but %s already exists", historyId, existing);
549 ShardDataTreeTransactionChain ensureTransactionChain(final LocalHistoryIdentifier historyId,
550 final @Nullable Runnable callback) {
551 ShardDataTreeTransactionChain chain = transactionChains.get(historyId);
553 chain = new ShardDataTreeTransactionChain(historyId, this);
554 transactionChains.put(historyId, chain);
555 replicatePayload(historyId, CreateLocalHistoryPayload.create(
556 historyId, shard.getDatastoreContext().getInitialPayloadSerializedBufferCapacity()), callback);
557 } else if (callback != null) {
564 ReadOnlyShardDataTreeTransaction newReadOnlyTransaction(final TransactionIdentifier txId) {
565 shard.getShardMBean().incrementReadOnlyTransactionCount();
567 if (txId.getHistoryId().getHistoryId() == 0) {
568 return new ReadOnlyShardDataTreeTransaction(this, txId, dataTree.takeSnapshot());
571 return ensureTransactionChain(txId.getHistoryId(), null).newReadOnlyTransaction(txId);
574 ReadWriteShardDataTreeTransaction newReadWriteTransaction(final TransactionIdentifier txId) {
575 shard.getShardMBean().incrementReadWriteTransactionCount();
577 if (txId.getHistoryId().getHistoryId() == 0) {
578 return new ReadWriteShardDataTreeTransaction(ShardDataTree.this, txId, dataTree.takeSnapshot()
582 return ensureTransactionChain(txId.getHistoryId(), null).newReadWriteTransaction(txId);
586 public void notifyListeners(final DataTreeCandidate candidate) {
587 treeChangeListenerPublisher.publishChanges(candidate);
591 * Immediately purge all state relevant to leader. This includes all transaction chains and any scheduled
592 * replication callbacks.
594 void purgeLeaderState() {
595 for (ShardDataTreeTransactionChain chain : transactionChains.values()) {
599 transactionChains.clear();
600 replicationCallbacks.clear();
604 * Close a single transaction chain.
606 * @param id History identifier
607 * @param callback Callback to invoke upon completion, may be null
609 void closeTransactionChain(final LocalHistoryIdentifier id, final @Nullable Runnable callback) {
610 if (commonCloseTransactionChain(id, callback)) {
611 replicatePayload(id, CloseLocalHistoryPayload.create(id,
612 shard.getDatastoreContext().getInitialPayloadSerializedBufferCapacity()), callback);
617 * Close a single transaction chain which is received through ask-based protocol. It does not keep a commit record.
619 * @param id History identifier
621 void closeTransactionChain(final LocalHistoryIdentifier id) {
622 commonCloseTransactionChain(id, null);
625 private boolean commonCloseTransactionChain(final LocalHistoryIdentifier id, final @Nullable Runnable callback) {
626 final ShardDataTreeTransactionChain chain = transactionChains.get(id);
628 LOG.debug("{}: Closing non-existent transaction chain {}", logContext, id);
629 if (callback != null) {
640 * Purge a single transaction chain.
642 * @param id History identifier
643 * @param callback Callback to invoke upon completion, may be null
645 void purgeTransactionChain(final LocalHistoryIdentifier id, final @Nullable Runnable callback) {
646 final ShardDataTreeTransactionChain chain = transactionChains.remove(id);
648 LOG.debug("{}: Purging non-existent transaction chain {}", logContext, id);
649 if (callback != null) {
655 replicatePayload(id, PurgeLocalHistoryPayload.create(
656 id, shard.getDatastoreContext().getInitialPayloadSerializedBufferCapacity()), callback);
659 Optional<DataTreeCandidate> readCurrentData() {
660 return dataTree.takeSnapshot().readNode(YangInstanceIdentifier.empty())
661 .map(state -> DataTreeCandidates.fromNormalizedNode(YangInstanceIdentifier.empty(), state));
664 public void registerTreeChangeListener(final YangInstanceIdentifier path, final DOMDataTreeChangeListener listener,
665 final Optional<DataTreeCandidate> initialState,
666 final Consumer<ListenerRegistration<DOMDataTreeChangeListener>> onRegistration) {
667 treeChangeListenerPublisher.registerTreeChangeListener(path, listener, initialState, onRegistration);
671 return pendingTransactions.size() + pendingCommits.size() + pendingFinishCommits.size();
675 void abortTransaction(final AbstractShardDataTreeTransaction<?> transaction, final Runnable callback) {
676 final TransactionIdentifier id = transaction.getIdentifier();
677 LOG.debug("{}: aborting transaction {}", logContext, id);
678 replicatePayload(id, AbortTransactionPayload.create(
679 id, shard.getDatastoreContext().getInitialPayloadSerializedBufferCapacity()), callback);
683 void abortFromTransactionActor(final AbstractShardDataTreeTransaction<?> transaction) {
684 // No-op for free-standing transactions
689 ShardDataTreeCohort finishTransaction(final ReadWriteShardDataTreeTransaction transaction,
690 final Optional<SortedSet<String>> participatingShardNames) {
691 final DataTreeModification snapshot = transaction.getSnapshot();
692 final TransactionIdentifier id = transaction.getIdentifier();
693 LOG.debug("{}: readying transaction {}", logContext, id);
695 LOG.debug("{}: transaction {} ready", logContext, id);
697 return createReadyCohort(transaction.getIdentifier(), snapshot, participatingShardNames);
700 void purgeTransaction(final TransactionIdentifier id, final Runnable callback) {
701 LOG.debug("{}: purging transaction {}", logContext, id);
702 replicatePayload(id, PurgeTransactionPayload.create(
703 id, shard.getDatastoreContext().getInitialPayloadSerializedBufferCapacity()), callback);
706 public Optional<NormalizedNode<?, ?>> readNode(final YangInstanceIdentifier path) {
707 return dataTree.takeSnapshot().readNode(path);
710 DataTreeSnapshot takeSnapshot() {
711 return dataTree.takeSnapshot();
715 public DataTreeModification newModification() {
716 return dataTree.takeSnapshot().newModification();
719 public Collection<ShardDataTreeCohort> getAndClearPendingTransactions() {
720 Collection<ShardDataTreeCohort> ret = new ArrayList<>(getQueueSize());
722 for (CommitEntry entry: pendingFinishCommits) {
723 ret.add(entry.cohort);
726 for (CommitEntry entry: pendingCommits) {
727 ret.add(entry.cohort);
730 for (CommitEntry entry: pendingTransactions) {
731 ret.add(entry.cohort);
734 pendingFinishCommits.clear();
735 pendingCommits.clear();
736 pendingTransactions.clear();
742 * Called some time after {@link #processNextPendingTransaction()} decides to stop processing.
744 void resumeNextPendingTransaction() {
745 LOG.debug("{}: attempting to resume transaction processing", logContext);
746 processNextPending();
749 @SuppressWarnings("checkstyle:IllegalCatch")
750 private void processNextPendingTransaction() {
751 ++currentTransactionBatch;
752 if (currentTransactionBatch > MAX_TRANSACTION_BATCH) {
753 LOG.debug("{}: Already processed {}, scheduling continuation", logContext, currentTransactionBatch);
754 shard.scheduleNextPendingTransaction();
758 processNextPending(pendingTransactions, State.CAN_COMMIT_PENDING, entry -> {
759 final SimpleShardDataTreeCohort cohort = entry.cohort;
760 final DataTreeModification modification = cohort.getDataTreeModification();
762 LOG.debug("{}: Validating transaction {}", logContext, cohort.getIdentifier());
765 tip.validate(modification);
766 LOG.debug("{}: Transaction {} validated", logContext, cohort.getIdentifier());
767 cohort.successfulCanCommit();
768 entry.lastAccess = readTime();
770 } catch (ConflictingModificationAppliedException e) {
771 LOG.warn("{}: Store Tx {}: Conflicting modification for path {}.", logContext, cohort.getIdentifier(),
773 cause = new OptimisticLockFailedException("Optimistic lock failed for path " + e.getPath(), e);
774 } catch (DataValidationFailedException e) {
775 LOG.warn("{}: Store Tx {}: Data validation failed for path {}.", logContext, cohort.getIdentifier(),
778 // For debugging purposes, allow dumping of the modification. Coupled with the above
779 // precondition log, it should allow us to understand what went on.
780 LOG.debug("{}: Store Tx {}: modifications: {}", logContext, cohort.getIdentifier(), modification);
781 LOG.trace("{}: Current tree: {}", logContext, dataTree);
782 cause = new TransactionCommitFailedException("Data did not pass validation for path " + e.getPath(), e);
783 } catch (Exception e) {
784 LOG.warn("{}: Unexpected failure in validation phase", logContext, e);
788 // Failure path: propagate the failure, remove the transaction from the queue and loop to the next one
789 pendingTransactions.poll().cohort.failedCanCommit(cause);
793 private void processNextPending() {
794 processNextPendingCommit();
795 processNextPendingTransaction();
798 private void processNextPending(final Queue<CommitEntry> queue, final State allowedState,
799 final Consumer<CommitEntry> processor) {
800 while (!queue.isEmpty()) {
801 final CommitEntry entry = queue.peek();
802 final SimpleShardDataTreeCohort cohort = entry.cohort;
804 if (cohort.isFailed()) {
805 LOG.debug("{}: Removing failed transaction {}", logContext, cohort.getIdentifier());
810 if (cohort.getState() == allowedState) {
811 processor.accept(entry);
817 maybeRunOperationOnPendingTransactionsComplete();
820 private void processNextPendingCommit() {
821 processNextPending(pendingCommits, State.COMMIT_PENDING,
822 entry -> startCommit(entry.cohort, entry.cohort.getCandidate()));
825 private boolean peekNextPendingCommit() {
826 final CommitEntry first = pendingCommits.peek();
827 return first != null && first.cohort.getState() == State.COMMIT_PENDING;
830 void startCanCommit(final SimpleShardDataTreeCohort cohort) {
831 final CommitEntry head = pendingTransactions.peek();
833 LOG.warn("{}: No transactions enqueued while attempting to start canCommit on {}", logContext, cohort);
836 if (!cohort.equals(head.cohort)) {
837 // The tx isn't at the head of the queue so we can't start canCommit at this point. Here we check if this
838 // tx should be moved ahead of other tx's in the READY state in the pendingTransactions queue. If this tx
839 // has other participating shards, it could deadlock with other tx's accessing the same shards
840 // depending on the order the tx's are readied on each shard
841 // (see https://jira.opendaylight.org/browse/CONTROLLER-1836). Therefore, if the preceding participating
842 // shard names for a preceding pending tx, call it A, in the queue matches that of this tx, then this tx
843 // is allowed to be moved ahead of tx A in the queue so it is processed first to avoid potential deadlock
844 // if tx A is behind this tx in the pendingTransactions queue for a preceding shard. In other words, since
845 // canCommmit for this tx was requested before tx A, honor that request. If this tx is moved to the head of
846 // the queue as a result, then proceed with canCommit.
848 Collection<String> precedingShardNames = extractPrecedingShardNames(cohort.getParticipatingShardNames());
849 if (precedingShardNames.isEmpty()) {
850 LOG.debug("{}: Tx {} is scheduled for canCommit step", logContext, cohort.getIdentifier());
854 LOG.debug("{}: Evaluating tx {} for canCommit - preceding participating shard names {}",
855 logContext, cohort.getIdentifier(), precedingShardNames);
856 final Iterator<CommitEntry> iter = pendingTransactions.iterator();
858 int moveToIndex = -1;
859 while (iter.hasNext()) {
860 final CommitEntry entry = iter.next();
863 if (cohort.equals(entry.cohort)) {
864 if (moveToIndex < 0) {
865 LOG.debug("{}: Not moving tx {} - cannot proceed with canCommit",
866 logContext, cohort.getIdentifier());
870 LOG.debug("{}: Moving {} to index {} in the pendingTransactions queue",
871 logContext, cohort.getIdentifier(), moveToIndex);
873 insertEntry(pendingTransactions, entry, moveToIndex);
875 if (!cohort.equals(pendingTransactions.peek().cohort)) {
876 LOG.debug("{}: Tx {} is not at the head of the queue - cannot proceed with canCommit",
877 logContext, cohort.getIdentifier());
881 LOG.debug("{}: Tx {} is now at the head of the queue - proceeding with canCommit",
882 logContext, cohort.getIdentifier());
886 if (entry.cohort.getState() != State.READY) {
887 LOG.debug("{}: Skipping pending transaction {} in state {}",
888 logContext, entry.cohort.getIdentifier(), entry.cohort.getState());
892 final Collection<String> pendingPrecedingShardNames = extractPrecedingShardNames(
893 entry.cohort.getParticipatingShardNames());
895 if (precedingShardNames.equals(pendingPrecedingShardNames)) {
896 if (moveToIndex < 0) {
897 LOG.debug("{}: Preceding shard names {} for pending tx {} match - saving moveToIndex {}",
898 logContext, pendingPrecedingShardNames, entry.cohort.getIdentifier(), index);
902 "{}: Preceding shard names {} for pending tx {} match but moveToIndex already set to {}",
903 logContext, pendingPrecedingShardNames, entry.cohort.getIdentifier(), moveToIndex);
906 LOG.debug("{}: Preceding shard names {} for pending tx {} differ - skipping",
907 logContext, pendingPrecedingShardNames, entry.cohort.getIdentifier());
912 processNextPendingTransaction();
915 private static void insertEntry(final Deque<CommitEntry> queue, final CommitEntry entry, final int atIndex) {
917 queue.addFirst(entry);
921 LOG.trace("Inserting into Deque at index {}", atIndex);
923 Deque<CommitEntry> tempStack = new ArrayDeque<>(atIndex);
924 for (int i = 0; i < atIndex; i++) {
925 tempStack.push(queue.poll());
928 queue.addFirst(entry);
930 tempStack.forEach(queue::addFirst);
933 private Collection<String> extractPrecedingShardNames(final Optional<SortedSet<String>> participatingShardNames) {
934 return participatingShardNames.map((Function<SortedSet<String>, Collection<String>>)
935 set -> set.headSet(shard.getShardName())).orElse(Collections.<String>emptyList());
938 private void failPreCommit(final Throwable cause) {
939 shard.getShardMBean().incrementFailedTransactionsCount();
940 pendingTransactions.poll().cohort.failedPreCommit(cause);
941 processNextPendingTransaction();
944 @SuppressWarnings("checkstyle:IllegalCatch")
945 void startPreCommit(final SimpleShardDataTreeCohort cohort) {
946 final CommitEntry entry = pendingTransactions.peek();
947 checkState(entry != null, "Attempted to pre-commit of %s when no transactions pending", cohort);
949 final SimpleShardDataTreeCohort current = entry.cohort;
950 verify(cohort.equals(current), "Attempted to pre-commit %s while %s is pending", cohort, current);
952 final TransactionIdentifier currentId = current.getIdentifier();
953 LOG.debug("{}: Preparing transaction {}", logContext, currentId);
955 final DataTreeCandidateTip candidate;
957 candidate = tip.prepare(cohort.getDataTreeModification());
958 LOG.debug("{}: Transaction {} candidate ready", logContext, currentId);
959 } catch (DataValidationFailedException | RuntimeException e) {
964 cohort.userPreCommit(candidate, new FutureCallback<Void>() {
966 public void onSuccess(final Void noop) {
967 // Set the tip of the data tree.
968 tip = verifyNotNull(candidate);
970 entry.lastAccess = readTime();
972 pendingTransactions.remove();
973 pendingCommits.add(entry);
975 LOG.debug("{}: Transaction {} prepared", logContext, currentId);
977 cohort.successfulPreCommit(candidate);
979 processNextPendingTransaction();
983 public void onFailure(final Throwable failure) {
984 failPreCommit(failure);
989 private void failCommit(final Exception cause) {
990 shard.getShardMBean().incrementFailedTransactionsCount();
991 pendingFinishCommits.poll().cohort.failedCommit(cause);
992 processNextPending();
995 @SuppressWarnings("checkstyle:IllegalCatch")
996 private void finishCommit(final SimpleShardDataTreeCohort cohort) {
997 final TransactionIdentifier txId = cohort.getIdentifier();
998 final DataTreeCandidate candidate = cohort.getCandidate();
1000 LOG.debug("{}: Resuming commit of transaction {}", logContext, txId);
1002 if (tip == candidate) {
1003 // All pending candidates have been committed, reset the tip to the data tree.
1008 dataTree.commit(candidate);
1009 } catch (Exception e) {
1010 LOG.error("{}: Failed to commit transaction {}", logContext, txId, e);
1015 allMetadataCommittedTransaction(txId);
1016 shard.getShardMBean().incrementCommittedTransactionCount();
1017 shard.getShardMBean().setLastCommittedTransactionTime(System.currentTimeMillis());
1019 // FIXME: propagate journal index
1020 pendingFinishCommits.poll().cohort.successfulCommit(UnsignedLong.ZERO, () -> {
1021 LOG.trace("{}: Transaction {} committed, proceeding to notify", logContext, txId);
1022 notifyListeners(candidate);
1024 processNextPending();
1028 void startCommit(final SimpleShardDataTreeCohort cohort, final DataTreeCandidate candidate) {
1029 final CommitEntry entry = pendingCommits.peek();
1030 checkState(entry != null, "Attempted to start commit of %s when no transactions pending", cohort);
1032 final SimpleShardDataTreeCohort current = entry.cohort;
1033 if (!cohort.equals(current)) {
1034 LOG.debug("{}: Transaction {} scheduled for commit step", logContext, cohort.getIdentifier());
1038 LOG.debug("{}: Starting commit for transaction {}", logContext, current.getIdentifier());
1040 final TransactionIdentifier txId = cohort.getIdentifier();
1041 final Payload payload;
1043 payload = CommitTransactionPayload.create(txId, candidate, PayloadVersion.current(),
1044 shard.getDatastoreContext().getInitialPayloadSerializedBufferCapacity());
1045 } catch (IOException e) {
1046 LOG.error("{}: Failed to encode transaction {} candidate {}", logContext, txId, candidate, e);
1047 pendingCommits.poll().cohort.failedCommit(e);
1048 processNextPending();
1052 // We process next transactions pending canCommit before we call persistPayload to possibly progress subsequent
1053 // transactions to the COMMIT_PENDING state so the payloads can be batched for replication. This is done for
1054 // single-shard transactions that immediately transition from canCommit to preCommit to commit. Note that
1055 // if the next pending transaction is progressed to COMMIT_PENDING and this method (startCommit) is called,
1056 // the next transaction will not attempt to replicate b/c the current transaction is still at the head of the
1057 // pendingCommits queue.
1058 processNextPendingTransaction();
1060 // After processing next pending transactions, we can now remove the current transaction from pendingCommits.
1061 // Note this must be done before the call to peekNextPendingCommit below so we check the next transaction
1062 // in order to properly determine the batchHint flag for the call to persistPayload.
1063 pendingCommits.remove();
1064 pendingFinishCommits.add(entry);
1066 // See if the next transaction is pending commit (ie in the COMMIT_PENDING state) so it can be batched with
1067 // this transaction for replication.
1068 boolean replicationBatchHint = peekNextPendingCommit();
1070 // Once completed, we will continue via payloadReplicationComplete
1071 shard.persistPayload(txId, payload, replicationBatchHint);
1073 entry.lastAccess = shard.ticker().read();
1075 LOG.debug("{}: Transaction {} submitted to persistence", logContext, txId);
1077 // Process the next transaction pending commit, if any. If there is one it will be batched with this
1078 // transaction for replication.
1079 processNextPendingCommit();
1082 Collection<ActorRef> getCohortActors() {
1083 return cohortRegistry.getCohortActors();
1086 void processCohortRegistryCommand(final ActorRef sender, final CohortRegistryCommand message) {
1087 cohortRegistry.process(sender, message);
1091 ShardDataTreeCohort createFailedCohort(final TransactionIdentifier txId, final DataTreeModification mod,
1092 final Exception failure) {
1093 final SimpleShardDataTreeCohort cohort = new SimpleShardDataTreeCohort(this, mod, txId, failure);
1094 pendingTransactions.add(new CommitEntry(cohort, readTime()));
1099 ShardDataTreeCohort createReadyCohort(final TransactionIdentifier txId, final DataTreeModification mod,
1100 final Optional<SortedSet<String>> participatingShardNames) {
1101 SimpleShardDataTreeCohort cohort = new SimpleShardDataTreeCohort(this, mod, txId,
1102 cohortRegistry.createCohort(schemaContext, txId, shard::executeInSelf,
1103 COMMIT_STEP_TIMEOUT), participatingShardNames);
1104 pendingTransactions.add(new CommitEntry(cohort, readTime()));
1108 // Exposed for ShardCommitCoordinator so it does not have deal with local histories (it does not care), this mimics
1109 // the newReadWriteTransaction()
1110 ShardDataTreeCohort newReadyCohort(final TransactionIdentifier txId, final DataTreeModification mod,
1111 final Optional<SortedSet<String>> participatingShardNames) {
1112 if (txId.getHistoryId().getHistoryId() == 0) {
1113 return createReadyCohort(txId, mod, participatingShardNames);
1116 return ensureTransactionChain(txId.getHistoryId(), null).createReadyCohort(txId, mod, participatingShardNames);
1119 @SuppressFBWarnings(value = "DB_DUPLICATE_SWITCH_CLAUSES", justification = "See inline comments below.")
1120 void checkForExpiredTransactions(final long transactionCommitTimeoutMillis,
1121 final Function<SimpleShardDataTreeCohort, OptionalLong> accessTimeUpdater) {
1122 final long timeout = TimeUnit.MILLISECONDS.toNanos(transactionCommitTimeoutMillis);
1123 final long now = readTime();
1125 final Queue<CommitEntry> currentQueue = !pendingFinishCommits.isEmpty() ? pendingFinishCommits :
1126 !pendingCommits.isEmpty() ? pendingCommits : pendingTransactions;
1127 final CommitEntry currentTx = currentQueue.peek();
1128 if (currentTx == null) {
1129 // Empty queue, no-op
1133 long delta = now - currentTx.lastAccess;
1134 if (delta < timeout) {
1135 // Not expired yet, bail
1139 final OptionalLong updateOpt = accessTimeUpdater.apply(currentTx.cohort);
1140 if (updateOpt.isPresent()) {
1141 final long newAccess = updateOpt.getAsLong();
1142 final long newDelta = now - newAccess;
1143 if (newDelta < delta) {
1144 LOG.debug("{}: Updated current transaction {} access time", logContext,
1145 currentTx.cohort.getIdentifier());
1146 currentTx.lastAccess = newAccess;
1150 if (delta < timeout) {
1151 // Not expired yet, bail
1156 final long deltaMillis = TimeUnit.NANOSECONDS.toMillis(delta);
1157 final State state = currentTx.cohort.getState();
1159 LOG.warn("{}: Current transaction {} has timed out after {} ms in state {}", logContext,
1160 currentTx.cohort.getIdentifier(), deltaMillis, state);
1161 boolean processNext = true;
1162 final TimeoutException cohortFailure = new TimeoutException("Backend timeout in state " + state + " after "
1163 + deltaMillis + "ms");
1166 case CAN_COMMIT_PENDING:
1167 currentQueue.remove().cohort.failedCanCommit(cohortFailure);
1169 case CAN_COMMIT_COMPLETE:
1170 // The suppression of the FindBugs "DB_DUPLICATE_SWITCH_CLAUSES" warning pertains to this clause
1171 // whose code is duplicated with PRE_COMMIT_COMPLETE. The clauses aren't combined in case the code
1172 // in PRE_COMMIT_COMPLETE is changed.
1173 currentQueue.remove().cohort.reportFailure(cohortFailure);
1175 case PRE_COMMIT_PENDING:
1176 currentQueue.remove().cohort.failedPreCommit(cohortFailure);
1178 case PRE_COMMIT_COMPLETE:
1179 // FIXME: this is a legacy behavior problem. Three-phase commit protocol specifies that after we
1180 // are ready we should commit the transaction, not abort it. Our current software stack does
1181 // not allow us to do that consistently, because we persist at the time of commit, hence
1182 // we can end up in a state where we have pre-committed a transaction, then a leader failover
1183 // occurred ... the new leader does not see the pre-committed transaction and does not have
1184 // a running timer. To fix this we really need two persistence events.
1186 // The first one, done at pre-commit time will hold the transaction payload. When consensus
1187 // is reached, we exit the pre-commit phase and start the pre-commit timer. Followers do not
1188 // apply the state in this event.
1190 // The second one, done at commit (or abort) time holds only the transaction identifier and
1191 // signals to followers that the state should (or should not) be applied.
1193 // In order to make the pre-commit timer working across failovers, though, we need
1194 // a per-shard cluster-wide monotonic time, so a follower becoming the leader can accurately
1195 // restart the timer.
1196 currentQueue.remove().cohort.reportFailure(cohortFailure);
1198 case COMMIT_PENDING:
1199 LOG.warn("{}: Transaction {} is still committing, cannot abort", logContext,
1200 currentTx.cohort.getIdentifier());
1201 currentTx.lastAccess = now;
1202 processNext = false;
1205 currentQueue.remove().cohort.reportFailure(cohortFailure);
1211 currentQueue.remove();
1215 processNextPending();
1219 boolean startAbort(final SimpleShardDataTreeCohort cohort) {
1220 final Iterator<CommitEntry> it = Iterables.concat(pendingFinishCommits, pendingCommits,
1221 pendingTransactions).iterator();
1222 if (!it.hasNext()) {
1223 LOG.debug("{}: no open transaction while attempting to abort {}", logContext, cohort.getIdentifier());
1227 // First entry is special, as it may already be committing
1228 final CommitEntry first = it.next();
1229 if (cohort.equals(first.cohort)) {
1230 if (cohort.getState() != State.COMMIT_PENDING) {
1231 LOG.debug("{}: aborting head of queue {} in state {}", logContext, cohort.getIdentifier(),
1232 cohort.getIdentifier());
1235 if (cohort.getCandidate() != null) {
1236 rebaseTransactions(it, dataTree);
1239 processNextPending();
1243 LOG.warn("{}: transaction {} is committing, skipping abort", logContext, cohort.getIdentifier());
1247 DataTreeTip newTip = MoreObjects.firstNonNull(first.cohort.getCandidate(), dataTree);
1248 while (it.hasNext()) {
1249 final CommitEntry e = it.next();
1250 if (cohort.equals(e.cohort)) {
1251 LOG.debug("{}: aborting queued transaction {}", logContext, cohort.getIdentifier());
1254 if (cohort.getCandidate() != null) {
1255 rebaseTransactions(it, newTip);
1260 newTip = MoreObjects.firstNonNull(e.cohort.getCandidate(), newTip);
1264 LOG.debug("{}: aborted transaction {} not found in the queue", logContext, cohort.getIdentifier());
1268 @SuppressWarnings("checkstyle:IllegalCatch")
1269 private void rebaseTransactions(final Iterator<CommitEntry> iter, final @NonNull DataTreeTip newTip) {
1270 tip = requireNonNull(newTip);
1271 while (iter.hasNext()) {
1272 final SimpleShardDataTreeCohort cohort = iter.next().cohort;
1273 if (cohort.getState() == State.CAN_COMMIT_COMPLETE) {
1274 LOG.debug("{}: Revalidating queued transaction {}", logContext, cohort.getIdentifier());
1277 tip.validate(cohort.getDataTreeModification());
1278 } catch (DataValidationFailedException | RuntimeException e) {
1279 LOG.debug("{}: Failed to revalidate queued transaction {}", logContext, cohort.getIdentifier(), e);
1280 cohort.reportFailure(e);
1282 } else if (cohort.getState() == State.PRE_COMMIT_COMPLETE) {
1283 LOG.debug("{}: Repreparing queued transaction {}", logContext, cohort.getIdentifier());
1286 tip.validate(cohort.getDataTreeModification());
1287 DataTreeCandidateTip candidate = tip.prepare(cohort.getDataTreeModification());
1289 cohort.setNewCandidate(candidate);
1291 } catch (RuntimeException | DataValidationFailedException e) {
1292 LOG.debug("{}: Failed to reprepare queued transaction {}", logContext, cohort.getIdentifier(), e);
1293 cohort.reportFailure(e);
1299 void setRunOnPendingTransactionsComplete(final Runnable operation) {
1300 runOnPendingTransactionsComplete = operation;
1301 maybeRunOperationOnPendingTransactionsComplete();
1304 private void maybeRunOperationOnPendingTransactionsComplete() {
1305 if (runOnPendingTransactionsComplete != null && !anyPendingTransactions()) {
1306 LOG.debug("{}: Pending transactions complete - running operation {}", logContext,
1307 runOnPendingTransactionsComplete);
1309 runOnPendingTransactionsComplete.run();
1310 runOnPendingTransactionsComplete = null;
1314 ShardStats getStats() {
1315 return shard.getShardMBean();
1318 Iterator<SimpleShardDataTreeCohort> cohortIterator() {
1319 return Iterables.transform(Iterables.concat(pendingFinishCommits, pendingCommits, pendingTransactions),
1320 e -> e.cohort).iterator();
1323 void removeTransactionChain(final LocalHistoryIdentifier id) {
1324 if (transactionChains.remove(id) != null) {
1325 LOG.debug("{}: Removed transaction chain {}", logContext, id);