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