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