opendaylight/md-sal/sal-distributed-datastore/src/main/java/org/opendaylight/controller/cluster/datastore/ShardDataTree.java

   1 /*
   2  * Copyright (c) 2015 Cisco Systems, Inc. and others.  All rights reserved.
   3  *
   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
   7  */
   8 package org.opendaylight.controller.cluster.datastore;
   9
  10 import akka.actor.ActorRef;
  11 import akka.util.Timeout;
  12 import com.google.common.annotations.VisibleForTesting;
  13 import com.google.common.base.MoreObjects;
  14 import com.google.common.base.Optional;
  15 import com.google.common.base.Preconditions;
  16 import com.google.common.base.Stopwatch;
  17 import com.google.common.base.Ticker;
  18 import com.google.common.base.Verify;
  19 import com.google.common.collect.ImmutableList;
  20 import com.google.common.collect.ImmutableMap;
  21 import com.google.common.collect.ImmutableMap.Builder;
  22 import com.google.common.collect.Iterables;
  23 import com.google.common.primitives.UnsignedLong;
  24 import edu.umd.cs.findbugs.annotations.SuppressFBWarnings;
  25 import java.io.File;
  26 import java.io.IOException;
  27 import java.util.AbstractMap.SimpleEntry;
  28 import java.util.ArrayDeque;
  29 import java.util.ArrayList;
  30 import java.util.Collection;
  31 import java.util.HashMap;
  32 import java.util.Iterator;
  33 import java.util.Map;
  34 import java.util.Map.Entry;
  35 import java.util.Queue;
  36 import java.util.concurrent.ExecutionException;
  37 import java.util.concurrent.TimeUnit;
  38 import java.util.concurrent.TimeoutException;
  39 import java.util.function.Consumer;
  40 import java.util.function.UnaryOperator;
  41 import javax.annotation.Nonnull;
  42 import javax.annotation.concurrent.NotThreadSafe;
  43 import org.opendaylight.controller.cluster.access.concepts.LocalHistoryIdentifier;
  44 import org.opendaylight.controller.cluster.access.concepts.TransactionIdentifier;
  45 import org.opendaylight.controller.cluster.datastore.DataTreeCohortActorRegistry.CohortRegistryCommand;
  46 import org.opendaylight.controller.cluster.datastore.ShardDataTreeCohort.State;
  47 import org.opendaylight.controller.cluster.datastore.persisted.CommitTransactionPayload;
  48 import org.opendaylight.controller.cluster.datastore.persisted.MetadataShardDataTreeSnapshot;
  49 import org.opendaylight.controller.cluster.datastore.persisted.ShardDataTreeSnapshot;
  50 import org.opendaylight.controller.cluster.datastore.persisted.ShardDataTreeSnapshotMetadata;
  51 import org.opendaylight.controller.cluster.datastore.utils.DataTreeModificationOutput;
  52 import org.opendaylight.controller.cluster.datastore.utils.PruningDataTreeModification;
  53 import org.opendaylight.controller.cluster.raft.protobuff.client.messages.Payload;
  54 import org.opendaylight.controller.md.sal.common.api.data.AsyncDataBroker.DataChangeScope;
  55 import org.opendaylight.controller.md.sal.common.api.data.AsyncDataChangeListener;
  56 import org.opendaylight.controller.md.sal.common.api.data.OptimisticLockFailedException;
  57 import org.opendaylight.controller.md.sal.common.api.data.TransactionCommitFailedException;
  58 import org.opendaylight.controller.md.sal.dom.api.DOMDataTreeChangeListener;
  59 import org.opendaylight.controller.md.sal.dom.store.impl.DataChangeListenerRegistration;
  60 import org.opendaylight.yangtools.concepts.Identifier;
  61 import org.opendaylight.yangtools.concepts.ListenerRegistration;
  62 import org.opendaylight.yangtools.yang.data.api.YangInstanceIdentifier;
  63 import org.opendaylight.yangtools.yang.data.api.schema.NormalizedNode;
  64 import org.opendaylight.yangtools.yang.data.api.schema.tree.ConflictingModificationAppliedException;
  65 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeCandidate;
  66 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeCandidateTip;
  67 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeCandidates;
  68 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeModification;
  69 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeSnapshot;
  70 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeTip;
  71 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataValidationFailedException;
  72 import org.opendaylight.yangtools.yang.data.api.schema.tree.ModificationType;
  73 import org.opendaylight.yangtools.yang.data.api.schema.tree.TipProducingDataTree;
  74 import org.opendaylight.yangtools.yang.data.api.schema.tree.TipProducingDataTreeTip;
  75 import org.opendaylight.yangtools.yang.data.api.schema.tree.TreeType;
  76 import org.opendaylight.yangtools.yang.data.impl.schema.tree.InMemoryDataTreeFactory;
  77 import org.opendaylight.yangtools.yang.model.api.SchemaContext;
  78 import org.slf4j.Logger;
  79 import org.slf4j.LoggerFactory;
  80 import scala.concurrent.duration.Duration;
  81
  82 /**
  83  * Internal shard state, similar to a DOMStore, but optimized for use in the actor system,
  84  * e.g. it does not expose public interfaces and assumes it is only ever called from a
  85  * single thread.
  86  *
  87  * <p>
  88  * This class is not part of the API contract and is subject to change at any time.
  89  */
  90 @NotThreadSafe
  91 public class ShardDataTree extends ShardDataTreeTransactionParent {
  92     private static final class CommitEntry {
  93         final SimpleShardDataTreeCohort cohort;
  94         long lastAccess;
  95
  96         CommitEntry(final SimpleShardDataTreeCohort cohort, final long now) {
  97             this.cohort = Preconditions.checkNotNull(cohort);
  98             lastAccess = now;
  99         }
 100     }
 101
 102     private static final Timeout COMMIT_STEP_TIMEOUT = new Timeout(Duration.create(5, TimeUnit.SECONDS));
 103     private static final Logger LOG = LoggerFactory.getLogger(ShardDataTree.class);
 104
 105     private final Map<LocalHistoryIdentifier, ShardDataTreeTransactionChain> transactionChains = new HashMap<>();
 106     private final DataTreeCohortActorRegistry cohortRegistry = new DataTreeCohortActorRegistry();
 107     private final Queue<CommitEntry> pendingTransactions = new ArrayDeque<>();
 108     private final Queue<CommitEntry> pendingCommits = new ArrayDeque<>();
 109     private final Queue<CommitEntry> pendingFinishCommits = new ArrayDeque<>();
 110     private final ShardDataTreeChangeListenerPublisher treeChangeListenerPublisher;
 111     private final ShardDataChangeListenerPublisher dataChangeListenerPublisher;
 112     private final Collection<ShardDataTreeMetadata<?>> metadata;
 113     private final TipProducingDataTree dataTree;
 114     private final String logContext;
 115     private final Shard shard;
 116     private Runnable runOnPendingTransactionsComplete;
 117
 118     /**
 119      * Optimistic {@link DataTreeCandidate} preparation. Since our DataTree implementation is a
 120      * {@link TipProducingDataTree}, each {@link DataTreeCandidate} is also a {@link DataTreeTip}, e.g. another
 121      * candidate can be prepared on top of it. They still need to be committed in sequence. Here we track the current
 122      * tip of the data tree, which is the last DataTreeCandidate we have in flight, or the DataTree itself.
 123      */
 124     private TipProducingDataTreeTip tip;
 125
 126     private SchemaContext schemaContext;
 127
 128     public ShardDataTree(final Shard shard, final SchemaContext schemaContext, final TipProducingDataTree dataTree,
 129             final ShardDataTreeChangeListenerPublisher treeChangeListenerPublisher,
 130             final ShardDataChangeListenerPublisher dataChangeListenerPublisher, final String logContext,
 131             final ShardDataTreeMetadata<?>... metadata) {
 132         this.dataTree = Preconditions.checkNotNull(dataTree);
 133         updateSchemaContext(schemaContext);
 134
 135         this.shard = Preconditions.checkNotNull(shard);
 136         this.treeChangeListenerPublisher = Preconditions.checkNotNull(treeChangeListenerPublisher);
 137         this.dataChangeListenerPublisher = Preconditions.checkNotNull(dataChangeListenerPublisher);
 138         this.logContext = Preconditions.checkNotNull(logContext);
 139         this.metadata = ImmutableList.copyOf(metadata);
 140         tip = dataTree;
 141     }
 142
 143     public ShardDataTree(final Shard shard, final SchemaContext schemaContext, final TreeType treeType,
 144             final YangInstanceIdentifier root,
 145             final ShardDataTreeChangeListenerPublisher treeChangeListenerPublisher,
 146             final ShardDataChangeListenerPublisher dataChangeListenerPublisher, final String logContext) {
 147         this(shard, schemaContext, InMemoryDataTreeFactory.getInstance().create(treeType, root),
 148                 treeChangeListenerPublisher, dataChangeListenerPublisher, logContext);
 149     }
 150
 151     @VisibleForTesting
 152     public ShardDataTree(final Shard shard, final SchemaContext schemaContext, final TreeType treeType) {
 153         this(shard, schemaContext, treeType, YangInstanceIdentifier.EMPTY,
 154                 new DefaultShardDataTreeChangeListenerPublisher(),
 155                 new DefaultShardDataChangeListenerPublisher(), "");
 156     }
 157
 158     final String logContext() {
 159         return logContext;
 160     }
 161
 162     final Ticker ticker() {
 163         return shard.ticker();
 164     }
 165
 166     public TipProducingDataTree getDataTree() {
 167         return dataTree;
 168     }
 169
 170     SchemaContext getSchemaContext() {
 171         return schemaContext;
 172     }
 173
 174     void updateSchemaContext(final SchemaContext newSchemaContext) {
 175         dataTree.setSchemaContext(newSchemaContext);
 176         this.schemaContext = Preconditions.checkNotNull(newSchemaContext);
 177     }
 178
 179     /**
 180      * Take a snapshot of current state for later recovery.
 181      *
 182      * @return A state snapshot
 183      */
 184     @Nonnull ShardDataTreeSnapshot takeStateSnapshot() {
 185         final NormalizedNode<?, ?> rootNode = dataTree.takeSnapshot().readNode(YangInstanceIdentifier.EMPTY).get();
 186         final Builder<Class<? extends ShardDataTreeSnapshotMetadata<?>>, ShardDataTreeSnapshotMetadata<?>> metaBuilder =
 187                 ImmutableMap.builder();
 188
 189         for (ShardDataTreeMetadata<?> m : metadata) {
 190             final ShardDataTreeSnapshotMetadata<?> meta = m.toSnapshot();
 191             if (meta != null) {
 192                 metaBuilder.put(meta.getType(), meta);
 193             }
 194         }
 195
 196         return new MetadataShardDataTreeSnapshot(rootNode, metaBuilder.build());
 197     }
 198
 199     private boolean anyPendingTransactions() {
 200         return !pendingTransactions.isEmpty() || !pendingCommits.isEmpty() || !pendingFinishCommits.isEmpty();
 201     }
 202
 203     private void applySnapshot(@Nonnull final ShardDataTreeSnapshot snapshot,
 204             final UnaryOperator<DataTreeModification> wrapper) throws DataValidationFailedException {
 205         final Stopwatch elapsed = Stopwatch.createStarted();
 206
 207         if (anyPendingTransactions()) {
 208             LOG.warn("{}: applying state snapshot with pending transactions", logContext);
 209         }
 210
 211         final Map<Class<? extends ShardDataTreeSnapshotMetadata<?>>, ShardDataTreeSnapshotMetadata<?>> snapshotMeta;
 212         if (snapshot instanceof MetadataShardDataTreeSnapshot) {
 213             snapshotMeta = ((MetadataShardDataTreeSnapshot) snapshot).getMetadata();
 214         } else {
 215             snapshotMeta = ImmutableMap.of();
 216         }
 217
 218         for (ShardDataTreeMetadata<?> m : metadata) {
 219             final ShardDataTreeSnapshotMetadata<?> s = snapshotMeta.get(m.getSupportedType());
 220             if (s != null) {
 221                 m.applySnapshot(s);
 222             } else {
 223                 m.reset();
 224             }
 225         }
 226
 227         final DataTreeModification mod = wrapper.apply(dataTree.takeSnapshot().newModification());
 228         // delete everything first
 229         mod.delete(YangInstanceIdentifier.EMPTY);
 230
 231         final java.util.Optional<NormalizedNode<?, ?>> maybeNode = snapshot.getRootNode();
 232         if (maybeNode.isPresent()) {
 233             // Add everything from the remote node back
 234             mod.write(YangInstanceIdentifier.EMPTY, maybeNode.get());
 235         }
 236         mod.ready();
 237
 238         final DataTreeModification unwrapped = unwrap(mod);
 239         dataTree.validate(unwrapped);
 240         DataTreeCandidateTip candidate = dataTree.prepare(unwrapped);
 241         dataTree.commit(candidate);
 242         notifyListeners(candidate);
 243
 244         LOG.debug("{}: state snapshot applied in %s", logContext, elapsed);
 245     }
 246
 247     /**
 248      * Apply a snapshot coming from the leader. This method assumes the leader and follower SchemaContexts match and
 249      * does not perform any pruning.
 250      *
 251      * @param snapshot Snapshot that needs to be applied
 252      * @throws DataValidationFailedException when the snapshot fails to apply
 253      */
 254     void applySnapshot(@Nonnull final ShardDataTreeSnapshot snapshot) throws DataValidationFailedException {
 255         applySnapshot(snapshot, UnaryOperator.identity());
 256     }
 257
 258     private PruningDataTreeModification wrapWithPruning(final DataTreeModification delegate) {
 259         return new PruningDataTreeModification(delegate, dataTree, schemaContext);
 260     }
 261
 262     private static DataTreeModification unwrap(final DataTreeModification modification) {
 263         if (modification instanceof PruningDataTreeModification) {
 264             return ((PruningDataTreeModification)modification).delegate();
 265         }
 266         return modification;
 267     }
 268
 269     /**
 270      * Apply a snapshot coming from recovery. This method does not assume the SchemaContexts match and performs data
 271      * pruning in an attempt to adjust the state to our current SchemaContext.
 272      *
 273      * @param snapshot Snapshot that needs to be applied
 274      * @throws DataValidationFailedException when the snapshot fails to apply
 275      */
 276     void applyRecoverySnapshot(final @Nonnull ShardDataTreeSnapshot snapshot) throws DataValidationFailedException {
 277         applySnapshot(snapshot, this::wrapWithPruning);
 278     }
 279
 280     @SuppressWarnings("checkstyle:IllegalCatch")
 281     private void applyRecoveryCandidate(final DataTreeCandidate candidate) throws DataValidationFailedException {
 282         final PruningDataTreeModification mod = wrapWithPruning(dataTree.takeSnapshot().newModification());
 283         DataTreeCandidates.applyToModification(mod, candidate);
 284         mod.ready();
 285
 286         final DataTreeModification unwrapped = mod.delegate();
 287         LOG.trace("{}: Applying recovery modification {}", logContext, unwrapped);
 288
 289         try {
 290             dataTree.validate(unwrapped);
 291             dataTree.commit(dataTree.prepare(unwrapped));
 292         } catch (Exception e) {
 293             File file = new File(System.getProperty("karaf.data", "."),
 294                     "failed-recovery-payload-" + logContext + ".out");
 295             DataTreeModificationOutput.toFile(file, unwrapped);
 296             throw new IllegalStateException(String.format(
 297                     "%s: Failed to apply recovery payload. Modification data was written to file %s",
 298                     logContext, file), e);
 299         }
 300     }
 301
 302     /**
 303      * Apply a payload coming from recovery. This method does not assume the SchemaContexts match and performs data
 304      * pruning in an attempt to adjust the state to our current SchemaContext.
 305      *
 306      * @param payload Payload
 307      * @throws IOException when the snapshot fails to deserialize
 308      * @throws DataValidationFailedException when the snapshot fails to apply
 309      */
 310     void applyRecoveryPayload(final @Nonnull Payload payload) throws IOException, DataValidationFailedException {
 311         if (payload instanceof CommitTransactionPayload) {
 312             final Entry<TransactionIdentifier, DataTreeCandidate> e =
 313                     ((CommitTransactionPayload) payload).getCandidate();
 314             applyRecoveryCandidate(e.getValue());
 315             allMetadataCommittedTransaction(e.getKey());
 316         } else if (payload instanceof DataTreeCandidatePayload) {
 317             applyRecoveryCandidate(((DataTreeCandidatePayload) payload).getCandidate());
 318         } else {
 319             LOG.debug("{}: ignoring unhandled payload {}", logContext, payload);
 320         }
 321     }
 322
 323     private void applyReplicatedCandidate(final Identifier identifier, final DataTreeCandidate foreign)
 324             throws DataValidationFailedException {
 325         LOG.debug("{}: Applying foreign transaction {}", logContext, identifier);
 326
 327         final DataTreeModification mod = dataTree.takeSnapshot().newModification();
 328         DataTreeCandidates.applyToModification(mod, foreign);
 329         mod.ready();
 330
 331         LOG.trace("{}: Applying foreign modification {}", logContext, mod);
 332         dataTree.validate(mod);
 333         final DataTreeCandidate candidate = dataTree.prepare(mod);
 334         dataTree.commit(candidate);
 335
 336         notifyListeners(candidate);
 337     }
 338
 339     /**
 340      * Apply a payload coming from the leader, which could actually be us. This method assumes the leader and follower
 341      * SchemaContexts match and does not perform any pruning.
 342      *
 343      * @param identifier Payload identifier as returned from RaftActor
 344      * @param payload Payload
 345      * @throws IOException when the snapshot fails to deserialize
 346      * @throws DataValidationFailedException when the snapshot fails to apply
 347      */
 348     void applyReplicatedPayload(final Identifier identifier, final Payload payload) throws IOException,
 349             DataValidationFailedException {
 350         /*
 351          * 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
 352          * if we are the leader and it has originated with us.
 353          *
 354          * The identifier will only ever be non-null when we were the leader which achieved consensus. Unfortunately,
 355          * though, this may not be the case anymore, as we are being called some time afterwards and we may not be
 356          * acting in that capacity anymore.
 357          *
 358          * In any case, we know that this is an entry coming from replication, hence we can be sure we will not observe
 359          * pre-Boron state -- which limits the number of options here.
 360          */
 361         if (payload instanceof CommitTransactionPayload) {
 362             if (identifier == null) {
 363                 final Entry<TransactionIdentifier, DataTreeCandidate> e =
 364                         ((CommitTransactionPayload) payload).getCandidate();
 365                 applyReplicatedCandidate(e.getKey(), e.getValue());
 366                 allMetadataCommittedTransaction(e.getKey());
 367             } else {
 368                 Verify.verify(identifier instanceof TransactionIdentifier);
 369                 payloadReplicationComplete((TransactionIdentifier) identifier);
 370             }
 371         } else {
 372             LOG.warn("{}: ignoring unhandled identifier {} payload {}", logContext, identifier, payload);
 373         }
 374     }
 375
 376     private void payloadReplicationComplete(final TransactionIdentifier txId) {
 377         final CommitEntry current = pendingFinishCommits.peek();
 378         if (current == null) {
 379             LOG.warn("{}: No outstanding transactions, ignoring consensus on transaction {}", logContext, txId);
 380             return;
 381         }
 382
 383         if (!current.cohort.getIdentifier().equals(txId)) {
 384             LOG.debug("{}: Head of pendingFinishCommits queue is {}, ignoring consensus on transaction {}", logContext,
 385                 current.cohort.getIdentifier(), txId);
 386             return;
 387         }
 388
 389         finishCommit(current.cohort);
 390     }
 391
 392     private void allMetadataCommittedTransaction(final TransactionIdentifier txId) {
 393         for (ShardDataTreeMetadata<?> m : metadata) {
 394             m.onTransactionCommitted(txId);
 395         }
 396     }
 397
 398     ShardDataTreeTransactionChain ensureTransactionChain(final LocalHistoryIdentifier localHistoryIdentifier) {
 399         ShardDataTreeTransactionChain chain = transactionChains.get(localHistoryIdentifier);
 400         if (chain == null) {
 401             chain = new ShardDataTreeTransactionChain(localHistoryIdentifier, this);
 402             transactionChains.put(localHistoryIdentifier, chain);
 403         }
 404
 405         return chain;
 406     }
 407
 408     ReadOnlyShardDataTreeTransaction newReadOnlyTransaction(final TransactionIdentifier txId) {
 409         if (txId.getHistoryId().getHistoryId() == 0) {
 410             return new ReadOnlyShardDataTreeTransaction(txId, dataTree.takeSnapshot());
 411         }
 412
 413         return ensureTransactionChain(txId.getHistoryId()).newReadOnlyTransaction(txId);
 414     }
 415
 416     ReadWriteShardDataTreeTransaction newReadWriteTransaction(final TransactionIdentifier txId) {
 417         if (txId.getHistoryId().getHistoryId() == 0) {
 418             return new ReadWriteShardDataTreeTransaction(ShardDataTree.this, txId, dataTree.takeSnapshot()
 419                     .newModification());
 420         }
 421
 422         return ensureTransactionChain(txId.getHistoryId()).newReadWriteTransaction(txId);
 423     }
 424
 425     @VisibleForTesting
 426     public void notifyListeners(final DataTreeCandidate candidate) {
 427         treeChangeListenerPublisher.publishChanges(candidate, logContext);
 428         dataChangeListenerPublisher.publishChanges(candidate, logContext);
 429     }
 430
 431     void notifyOfInitialData(final DataChangeListenerRegistration<AsyncDataChangeListener<YangInstanceIdentifier,
 432             NormalizedNode<?, ?>>> listenerReg, final Optional<DataTreeCandidate> currentState) {
 433         if (currentState.isPresent()) {
 434             ShardDataChangeListenerPublisher localPublisher = dataChangeListenerPublisher.newInstance();
 435             localPublisher.registerDataChangeListener(listenerReg.getPath(), listenerReg.getInstance(),
 436                     listenerReg.getScope());
 437             localPublisher.publishChanges(currentState.get(), logContext);
 438         }
 439     }
 440
 441     void notifyOfInitialData(final YangInstanceIdentifier path, final DOMDataTreeChangeListener listener,
 442             final Optional<DataTreeCandidate> currentState) {
 443         if (currentState.isPresent()) {
 444             ShardDataTreeChangeListenerPublisher localPublisher = treeChangeListenerPublisher.newInstance();
 445             localPublisher.registerTreeChangeListener(path, listener);
 446             localPublisher.publishChanges(currentState.get(), logContext);
 447         }
 448     }
 449
 450     void closeAllTransactionChains() {
 451         for (ShardDataTreeTransactionChain chain : transactionChains.values()) {
 452             chain.close();
 453         }
 454
 455         transactionChains.clear();
 456     }
 457
 458     void closeTransactionChain(final LocalHistoryIdentifier transactionChainId) {
 459         final ShardDataTreeTransactionChain chain = transactionChains.remove(transactionChainId);
 460         if (chain != null) {
 461             chain.close();
 462         } else {
 463             LOG.debug("{}: Closing non-existent transaction chain {}", logContext, transactionChainId);
 464         }
 465     }
 466
 467     Entry<DataChangeListenerRegistration<AsyncDataChangeListener<YangInstanceIdentifier, NormalizedNode<?, ?>>>,
 468             Optional<DataTreeCandidate>> registerChangeListener(final YangInstanceIdentifier path,
 469                     final AsyncDataChangeListener<YangInstanceIdentifier, NormalizedNode<?, ?>> listener,
 470                     final DataChangeScope scope) {
 471         DataChangeListenerRegistration<AsyncDataChangeListener<YangInstanceIdentifier, NormalizedNode<?, ?>>> reg =
 472                 dataChangeListenerPublisher.registerDataChangeListener(path, listener, scope);
 473
 474         return new SimpleEntry<>(reg, readCurrentData());
 475     }
 476
 477     private Optional<DataTreeCandidate> readCurrentData() {
 478         final Optional<NormalizedNode<?, ?>> currentState =
 479                 dataTree.takeSnapshot().readNode(YangInstanceIdentifier.EMPTY);
 480         return currentState.isPresent() ? Optional.of(DataTreeCandidates.fromNormalizedNode(
 481             YangInstanceIdentifier.EMPTY, currentState.get())) : Optional.<DataTreeCandidate>absent();
 482     }
 483
 484     public Entry<ListenerRegistration<DOMDataTreeChangeListener>, Optional<DataTreeCandidate>>
 485             registerTreeChangeListener(final YangInstanceIdentifier path, final DOMDataTreeChangeListener listener) {
 486         final ListenerRegistration<DOMDataTreeChangeListener> reg =
 487                 treeChangeListenerPublisher.registerTreeChangeListener(path, listener);
 488
 489         return new SimpleEntry<>(reg, readCurrentData());
 490     }
 491
 492     int getQueueSize() {
 493         return pendingTransactions.size() + pendingCommits.size() + pendingFinishCommits.size();
 494     }
 495
 496     @Override
 497     void abortTransaction(final AbstractShardDataTreeTransaction<?> transaction) {
 498         // Intentional no-op
 499     }
 500
 501     @Override
 502     ShardDataTreeCohort finishTransaction(final ReadWriteShardDataTreeTransaction transaction) {
 503         final DataTreeModification snapshot = transaction.getSnapshot();
 504         snapshot.ready();
 505
 506         return createReadyCohort(transaction.getIdentifier(), snapshot);
 507     }
 508
 509     public Optional<NormalizedNode<?, ?>> readNode(final YangInstanceIdentifier path) {
 510         return dataTree.takeSnapshot().readNode(path);
 511     }
 512
 513     DataTreeSnapshot takeSnapshot() {
 514         return dataTree.takeSnapshot();
 515     }
 516
 517     @VisibleForTesting
 518     public DataTreeModification newModification() {
 519         return dataTree.takeSnapshot().newModification();
 520     }
 521
 522     /**
 523      * Commits a modification.
 524      *
 525      * @deprecated This method violates DataTree containment and will be removed.
 526      */
 527     @VisibleForTesting
 528     @Deprecated
 529     public DataTreeCandidate commit(final DataTreeModification modification) throws DataValidationFailedException {
 530         // Direct modification commit is a utility, which cannot be used while we have transactions in-flight
 531         Preconditions.checkState(tip == dataTree, "Cannot modify data tree while transacgitons are pending");
 532
 533         modification.ready();
 534         dataTree.validate(modification);
 535         DataTreeCandidate candidate = dataTree.prepare(modification);
 536         dataTree.commit(candidate);
 537         return candidate;
 538     }
 539
 540     public Collection<ShardDataTreeCohort> getAndClearPendingTransactions() {
 541         Collection<ShardDataTreeCohort> ret = new ArrayList<>(getQueueSize());
 542
 543         for (CommitEntry entry: pendingFinishCommits) {
 544             ret.add(entry.cohort);
 545         }
 546
 547         for (CommitEntry entry: pendingCommits) {
 548             ret.add(entry.cohort);
 549         }
 550
 551         for (CommitEntry entry: pendingTransactions) {
 552             ret.add(entry.cohort);
 553         }
 554
 555         pendingFinishCommits.clear();
 556         pendingCommits.clear();
 557         pendingTransactions.clear();
 558         tip = dataTree;
 559         return ret;
 560     }
 561
 562     @SuppressWarnings("checkstyle:IllegalCatch")
 563     private void processNextPendingTransaction() {
 564         processNextPending(pendingTransactions, State.CAN_COMMIT_PENDING, entry -> {
 565             final SimpleShardDataTreeCohort cohort = entry.cohort;
 566             final DataTreeModification modification = cohort.getDataTreeModification();
 567
 568             LOG.debug("{}: Validating transaction {}", logContext, cohort.getIdentifier());
 569             Exception cause;
 570             try {
 571                 tip.validate(modification);
 572                 LOG.debug("{}: Transaction {} validated", logContext, cohort.getIdentifier());
 573                 cohort.successfulCanCommit();
 574                 entry.lastAccess = shard.ticker().read();
 575                 return;
 576             } catch (ConflictingModificationAppliedException e) {
 577                 LOG.warn("{}: Store Tx {}: Conflicting modification for path {}.", logContext, cohort.getIdentifier(),
 578                     e.getPath());
 579                 cause = new OptimisticLockFailedException("Optimistic lock failed.", e);
 580             } catch (DataValidationFailedException e) {
 581                 LOG.warn("{}: Store Tx {}: Data validation failed for path {}.", logContext, cohort.getIdentifier(),
 582                     e.getPath(), e);
 583
 584                 // For debugging purposes, allow dumping of the modification. Coupled with the above
 585                 // precondition log, it should allow us to understand what went on.
 586                 LOG.debug("{}: Store Tx {}: modifications: {} tree: {}", cohort.getIdentifier(), modification,
 587                         dataTree);
 588                 cause = new TransactionCommitFailedException("Data did not pass validation.", e);
 589             } catch (Exception e) {
 590                 LOG.warn("{}: Unexpected failure in validation phase", logContext, e);
 591                 cause = e;
 592             }
 593
 594             // Failure path: propagate the failure, remove the transaction from the queue and loop to the next one
 595             pendingTransactions.poll().cohort.failedCanCommit(cause);
 596         });
 597     }
 598
 599     private void processNextPending() {
 600         processNextPendingFinishCommit();
 601         processNextPendingCommit();
 602         processNextPendingTransaction();
 603     }
 604
 605     private void processNextPending(Queue<CommitEntry> queue, State allowedState, Consumer<CommitEntry> processor) {
 606         while (!queue.isEmpty()) {
 607             final CommitEntry entry = queue.peek();
 608             final SimpleShardDataTreeCohort cohort = entry.cohort;
 609
 610             if (cohort.isFailed()) {
 611                 LOG.debug("{}: Removing failed transaction {}", logContext, cohort.getIdentifier());
 612                 queue.remove();
 613                 continue;
 614             }
 615
 616             if (cohort.getState() == allowedState) {
 617                 processor.accept(entry);
 618             }
 619
 620             break;
 621         }
 622
 623         maybeRunOperationOnPendingTransactionsComplete();
 624     }
 625
 626     private void processNextPendingCommit() {
 627         processNextPending(pendingCommits, State.COMMIT_PENDING,
 628             entry -> startCommit(entry.cohort, entry.cohort.getCandidate()));
 629     }
 630
 631     private void processNextPendingFinishCommit() {
 632         processNextPending(pendingFinishCommits, State.FINISH_COMMIT_PENDING,
 633             entry -> payloadReplicationComplete(entry.cohort.getIdentifier()));
 634     }
 635
 636     private boolean peekNextPendingCommit() {
 637         final CommitEntry first = pendingCommits.peek();
 638         return first != null && first.cohort.getState() == State.COMMIT_PENDING;
 639     }
 640
 641     void startCanCommit(final SimpleShardDataTreeCohort cohort) {
 642         final SimpleShardDataTreeCohort current = pendingTransactions.peek().cohort;
 643         if (!cohort.equals(current)) {
 644             LOG.debug("{}: Transaction {} scheduled for canCommit step", logContext, cohort.getIdentifier());
 645             return;
 646         }
 647
 648         processNextPendingTransaction();
 649     }
 650
 651     private void failPreCommit(final Exception cause) {
 652         shard.getShardMBean().incrementFailedTransactionsCount();
 653         pendingTransactions.poll().cohort.failedPreCommit(cause);
 654         processNextPendingTransaction();
 655     }
 656
 657     @SuppressWarnings("checkstyle:IllegalCatch")
 658     void startPreCommit(final SimpleShardDataTreeCohort cohort) {
 659         final CommitEntry entry = pendingTransactions.peek();
 660         Preconditions.checkState(entry != null, "Attempted to pre-commit of %s when no transactions pending", cohort);
 661
 662         final SimpleShardDataTreeCohort current = entry.cohort;
 663         Verify.verify(cohort.equals(current), "Attempted to pre-commit %s while %s is pending", cohort, current);
 664
 665         LOG.debug("{}: Preparing transaction {}", logContext, current.getIdentifier());
 666
 667         final DataTreeCandidateTip candidate;
 668         try {
 669             candidate = tip.prepare(cohort.getDataTreeModification());
 670             cohort.userPreCommit(candidate);
 671         } catch (ExecutionException | TimeoutException | RuntimeException e) {
 672             failPreCommit(e);
 673             return;
 674         }
 675
 676         // Set the tip of the data tree.
 677         tip = Verify.verifyNotNull(candidate);
 678
 679         entry.lastAccess = shard.ticker().read();
 680
 681         pendingTransactions.remove();
 682         pendingCommits.add(entry);
 683
 684         LOG.debug("{}: Transaction {} prepared", logContext, current.getIdentifier());
 685
 686         cohort.successfulPreCommit(candidate);
 687
 688         processNextPendingTransaction();
 689     }
 690
 691     private void failCommit(final Exception cause) {
 692         shard.getShardMBean().incrementFailedTransactionsCount();
 693         pendingFinishCommits.poll().cohort.failedCommit(cause);
 694         processNextPending();
 695     }
 696
 697     @SuppressWarnings("checkstyle:IllegalCatch")
 698     private void finishCommit(final SimpleShardDataTreeCohort cohort) {
 699         final TransactionIdentifier txId = cohort.getIdentifier();
 700         final DataTreeCandidate candidate = cohort.getCandidate();
 701
 702         LOG.debug("{}: Resuming commit of transaction {}", logContext, txId);
 703
 704         if (tip == candidate) {
 705             // All pending candidates have been committed, reset the tip to the data tree.
 706             tip = dataTree;
 707         }
 708
 709         try {
 710             dataTree.commit(candidate);
 711         } catch (Exception e) {
 712             LOG.error("{}: Failed to commit transaction {}", logContext, txId, e);
 713             failCommit(e);
 714             return;
 715         }
 716
 717         shard.getShardMBean().incrementCommittedTransactionCount();
 718         shard.getShardMBean().setLastCommittedTransactionTime(System.currentTimeMillis());
 719
 720         // FIXME: propagate journal index
 721         pendingFinishCommits.poll().cohort.successfulCommit(UnsignedLong.ZERO);
 722
 723         LOG.trace("{}: Transaction {} committed, proceeding to notify", logContext, txId);
 724         notifyListeners(candidate);
 725
 726         processNextPending();
 727     }
 728
 729     void startCommit(final SimpleShardDataTreeCohort cohort, final DataTreeCandidate candidate) {
 730         final CommitEntry entry = pendingCommits.peek();
 731         Preconditions.checkState(entry != null, "Attempted to start commit of %s when no transactions pending", cohort);
 732
 733         final SimpleShardDataTreeCohort current = entry.cohort;
 734         if (!cohort.equals(current)) {
 735             LOG.debug("{}: Transaction {} scheduled for commit step", logContext, cohort.getIdentifier());
 736             return;
 737         }
 738
 739         LOG.debug("{}: Starting commit for transaction {}", logContext, current.getIdentifier());
 740
 741         final TransactionIdentifier txId = cohort.getIdentifier();
 742         if (shard.canSkipPayload() || candidate.getRootNode().getModificationType() == ModificationType.UNMODIFIED) {
 743             LOG.debug("{}: No replication required, proceeding to finish commit", logContext);
 744             pendingCommits.remove();
 745             pendingFinishCommits.add(entry);
 746             cohort.finishCommitPending();
 747             payloadReplicationComplete(txId);
 748             return;
 749         }
 750
 751         final Payload payload;
 752         try {
 753             payload = CommitTransactionPayload.create(txId, candidate);
 754         } catch (IOException e) {
 755             LOG.error("{}: Failed to encode transaction {} candidate {}", logContext, txId, candidate, e);
 756             pendingCommits.poll().cohort.failedCommit(e);
 757             processNextPending();
 758             return;
 759         }
 760
 761         // We process next transactions pending canCommit before we call persistPayload to possibly progress subsequent
 762         // transactions to the COMMIT_PENDING state so the payloads can be batched for replication. This is done for
 763         // single-shard transactions that immediately transition from canCommit to preCommit to commit. Note that
 764         // if the next pending transaction is progressed to COMMIT_PENDING and this method (startCommit) is called,
 765         // the next transaction will not attempt to replicate b/c the current transaction is still at the head of the
 766         // pendingCommits queue.
 767         processNextPendingTransaction();
 768
 769         // After processing next pending transactions, we can now remove the current transaction from pendingCommits.
 770         // Note this must be done before the call to peekNextPendingCommit below so we check the next transaction
 771         // in order to properly determine the batchHint flag for the call to persistPayload.
 772         pendingCommits.remove();
 773         pendingFinishCommits.add(entry);
 774
 775         // See if the next transaction is pending commit (ie in the COMMIT_PENDING state) so it can be batched with
 776         // this transaction for replication.
 777         boolean replicationBatchHint = peekNextPendingCommit();
 778
 779         // Once completed, we will continue via payloadReplicationComplete
 780         shard.persistPayload(txId, payload, replicationBatchHint);
 781
 782         entry.lastAccess = shard.ticker().read();
 783
 784         LOG.debug("{}: Transaction {} submitted to persistence", logContext, txId);
 785
 786         // Process the next transaction pending commit, if any. If there is one it will be batched with this
 787         // transaction for replication.
 788         processNextPendingCommit();
 789     }
 790
 791     void processCohortRegistryCommand(final ActorRef sender, final CohortRegistryCommand message) {
 792         cohortRegistry.process(sender, message);
 793     }
 794
 795     @Override
 796     ShardDataTreeCohort createReadyCohort(final TransactionIdentifier txId,
 797             final DataTreeModification modification) {
 798         SimpleShardDataTreeCohort cohort = new SimpleShardDataTreeCohort(this, modification, txId,
 799                 cohortRegistry.createCohort(schemaContext, txId, COMMIT_STEP_TIMEOUT));
 800         pendingTransactions.add(new CommitEntry(cohort, shard.ticker().read()));
 801         return cohort;
 802     }
 803
 804     @SuppressFBWarnings(value = "DB_DUPLICATE_SWITCH_CLAUSES", justification = "See inline comments below.")
 805     void checkForExpiredTransactions(final long transactionCommitTimeoutMillis) {
 806         final long timeout = TimeUnit.MILLISECONDS.toNanos(transactionCommitTimeoutMillis);
 807         final long now = shard.ticker().read();
 808
 809         final Queue<CommitEntry> currentQueue = !pendingFinishCommits.isEmpty() ? pendingFinishCommits :
 810             !pendingCommits.isEmpty() ? pendingCommits : pendingTransactions;
 811         final CommitEntry currentTx = currentQueue.peek();
 812         if (currentTx != null && currentTx.lastAccess + timeout < now) {
 813             LOG.warn("{}: Current transaction {} has timed out after {} ms in state {}", logContext,
 814                     currentTx.cohort.getIdentifier(), transactionCommitTimeoutMillis, currentTx.cohort.getState());
 815             boolean processNext = true;
 816             switch (currentTx.cohort.getState()) {
 817                 case CAN_COMMIT_PENDING:
 818                     currentQueue.remove().cohort.failedCanCommit(new TimeoutException());
 819                     break;
 820                 case CAN_COMMIT_COMPLETE:
 821                     // The suppression of the FindBugs "DB_DUPLICATE_SWITCH_CLAUSES" warning pertains to this clause
 822                     // whose code is duplicated with PRE_COMMIT_COMPLETE. The clauses aren't combined in case the code
 823                     // in PRE_COMMIT_COMPLETE is changed.
 824                     currentQueue.remove().cohort.reportFailure(new TimeoutException());
 825                     break;
 826                 case PRE_COMMIT_PENDING:
 827                     currentQueue.remove().cohort.failedPreCommit(new TimeoutException());
 828                     break;
 829                 case PRE_COMMIT_COMPLETE:
 830                     // FIXME: this is a legacy behavior problem. Three-phase commit protocol specifies that after we
 831                     //        are ready we should commit the transaction, not abort it. Our current software stack does
 832                     //        not allow us to do that consistently, because we persist at the time of commit, hence
 833                     //        we can end up in a state where we have pre-committed a transaction, then a leader failover
 834                     //        occurred ... the new leader does not see the pre-committed transaction and does not have
 835                     //        a running timer. To fix this we really need two persistence events.
 836                     //
 837                     //        The first one, done at pre-commit time will hold the transaction payload. When consensus
 838                     //        is reached, we exit the pre-commit phase and start the pre-commit timer. Followers do not
 839                     //        apply the state in this event.
 840                     //
 841                     //        The second one, done at commit (or abort) time holds only the transaction identifier and
 842                     //        signals to followers that the state should (or should not) be applied.
 843                     //
 844                     //        In order to make the pre-commit timer working across failovers, though, we need
 845                     //        a per-shard cluster-wide monotonic time, so a follower becoming the leader can accurately
 846                     //        restart the timer.
 847                     currentQueue.remove().cohort.reportFailure(new TimeoutException());
 848                     break;
 849                 case COMMIT_PENDING:
 850                     LOG.warn("{}: Transaction {} is still committing, cannot abort", logContext,
 851                         currentTx.cohort.getIdentifier());
 852                     currentTx.lastAccess = now;
 853                     processNext = false;
 854                     return;
 855                 case ABORTED:
 856                 case COMMITTED:
 857                 case FAILED:
 858                 case READY:
 859                 default:
 860                     currentQueue.remove();
 861             }
 862
 863             if (processNext) {
 864                 processNextPending();
 865             }
 866         }
 867     }
 868
 869     boolean startAbort(final SimpleShardDataTreeCohort cohort) {
 870         final Iterator<CommitEntry> it = Iterables.concat(pendingFinishCommits, pendingCommits,
 871                 pendingTransactions).iterator();
 872         if (!it.hasNext()) {
 873             LOG.debug("{}: no open transaction while attempting to abort {}", logContext, cohort.getIdentifier());
 874             return true;
 875         }
 876
 877         // First entry is special, as it may already be committing
 878         final CommitEntry first = it.next();
 879         if (cohort.equals(first.cohort)) {
 880             if (cohort.getState() != State.COMMIT_PENDING) {
 881                 LOG.debug("{}: aborting head of queue {} in state {}", logContext, cohort.getIdentifier(),
 882                     cohort.getIdentifier());
 883
 884                 it.remove();
 885                 if (cohort.getCandidate() != null) {
 886                     rebaseTransactions(it, dataTree);
 887                 }
 888
 889                 processNextPending();
 890                 return true;
 891             }
 892
 893             LOG.warn("{}: transaction {} is committing, skipping abort", logContext, cohort.getIdentifier());
 894             return false;
 895         }
 896
 897         TipProducingDataTreeTip newTip = MoreObjects.firstNonNull(first.cohort.getCandidate(), dataTree);
 898         while (it.hasNext()) {
 899             final CommitEntry e = it.next();
 900             if (cohort.equals(e.cohort)) {
 901                 LOG.debug("{}: aborting queued transaction {}", logContext, cohort.getIdentifier());
 902
 903                 it.remove();
 904                 if (cohort.getCandidate() != null) {
 905                     rebaseTransactions(it, newTip);
 906                 }
 907
 908                 return true;
 909             } else {
 910                 newTip = MoreObjects.firstNonNull(e.cohort.getCandidate(), newTip);
 911             }
 912         }
 913
 914         LOG.debug("{}: aborted transaction {} not found in the queue", logContext, cohort.getIdentifier());
 915         return true;
 916     }
 917
 918     @SuppressWarnings("checkstyle:IllegalCatch")
 919     private void rebaseTransactions(Iterator<CommitEntry> iter, @Nonnull TipProducingDataTreeTip newTip) {
 920         tip = Preconditions.checkNotNull(newTip);
 921         while (iter.hasNext()) {
 922             final SimpleShardDataTreeCohort cohort = iter.next().cohort;
 923             if (cohort.getState() == State.CAN_COMMIT_COMPLETE) {
 924                 LOG.debug("{}: Revalidating queued transaction {}", logContext, cohort.getIdentifier());
 925
 926                 try {
 927                     tip.validate(cohort.getDataTreeModification());
 928                 } catch (DataValidationFailedException | RuntimeException e) {
 929                     LOG.debug("{}: Failed to revalidate queued transaction {}", logContext, cohort.getIdentifier(), e);
 930                     cohort.reportFailure(e);
 931                 }
 932             } else if (cohort.getState() == State.PRE_COMMIT_COMPLETE) {
 933                 LOG.debug("{}: Repreparing queued transaction {}", logContext, cohort.getIdentifier());
 934
 935                 try {
 936                     tip.validate(cohort.getDataTreeModification());
 937                     DataTreeCandidateTip candidate = tip.prepare(cohort.getDataTreeModification());
 938                     cohort.userPreCommit(candidate);
 939
 940                     cohort.setNewCandidate(candidate);
 941                     tip = candidate;
 942                 } catch (ExecutionException | TimeoutException | RuntimeException | DataValidationFailedException e) {
 943                     LOG.debug("{}: Failed to reprepare queued transaction {}", logContext, cohort.getIdentifier(), e);
 944                     cohort.reportFailure(e);
 945                 }
 946             }
 947         }
 948     }
 949
 950     void setRunOnPendingTransactionsComplete(final Runnable operation) {
 951         runOnPendingTransactionsComplete = operation;
 952         maybeRunOperationOnPendingTransactionsComplete();
 953     }
 954
 955     private void maybeRunOperationOnPendingTransactionsComplete() {
 956         if (runOnPendingTransactionsComplete != null && !anyPendingTransactions()) {
 957             LOG.debug("{}: Pending transactions complete - running operation {}", logContext,
 958                     runOnPendingTransactionsComplete);
 959
 960             runOnPendingTransactionsComplete.run();
 961             runOnPendingTransactionsComplete = null;
 962         }
 963     }
 964 }