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  *
  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 schemaContext) {
 150         dataTree.setSchemaContext(schemaContext);
 151         this.schemaContext = Preconditions.checkNotNull(schemaContext);
 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(final @Nonnull 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         dataTree.commit(dataTree.prepare(unwrapped));
 212         LOG.debug("{}: state snapshot applied in %s", logContext, elapsed);
 213     }
 214
 215     private PruningDataTreeModification wrapWithPruning(final DataTreeModification delegate) {
 216         return new PruningDataTreeModification(delegate, dataTree, schemaContext);
 217     }
 218
 219     private static DataTreeModification unwrap(final DataTreeModification modification) {
 220         if (modification instanceof PruningDataTreeModification) {
 221             return ((PruningDataTreeModification)modification).delegate();
 222         }
 223         return modification;
 224     }
 225
 226     /**
 227      * Apply a snapshot coming from recovery. This method does not assume the SchemaContexts match and performs data
 228      * pruning in an attempt to adjust the state to our current SchemaContext.
 229      *
 230      * @param snapshot Snapshot that needs to be applied
 231      * @throws DataValidationFailedException when the snapshot fails to apply
 232      */
 233     void applyRecoverySnapshot(final @Nonnull ShardDataTreeSnapshot snapshot) throws DataValidationFailedException {
 234         applySnapshot(snapshot, this::wrapWithPruning);
 235     }
 236
 237
 238     /**
 239      * Apply a snapshot coming from the leader. This method assumes the leader and follower SchemaContexts match and
 240      * does not perform any pruning.
 241      *
 242      * @param snapshot Snapshot that needs to be applied
 243      * @throws DataValidationFailedException when the snapshot fails to apply
 244      */
 245     void applySnapshot(final @Nonnull ShardDataTreeSnapshot snapshot) throws DataValidationFailedException {
 246         applySnapshot(snapshot, UnaryOperator.identity());
 247     }
 248
 249     private void applyRecoveryCandidate(final DataTreeCandidate candidate) throws DataValidationFailedException {
 250         final PruningDataTreeModification mod = wrapWithPruning(dataTree.takeSnapshot().newModification());
 251         DataTreeCandidates.applyToModification(mod, candidate);
 252         mod.ready();
 253
 254         final DataTreeModification unwrapped = mod.delegate();
 255         LOG.trace("{}: Applying recovery modification {}", logContext, unwrapped);
 256
 257         try {
 258             dataTree.validate(unwrapped);
 259             dataTree.commit(dataTree.prepare(unwrapped));
 260         } catch (Exception e) {
 261             File file = new File(System.getProperty("karaf.data", "."),
 262                     "failed-recovery-payload-" + logContext + ".out");
 263             DataTreeModificationOutput.toFile(file, unwrapped);
 264             throw new IllegalStateException(String.format(
 265                     "%s: Failed to apply recovery payload. Modification data was written to file %s",
 266                     logContext, file), e);
 267         }
 268     }
 269
 270     /**
 271      * Apply a payload coming from recovery. This method does not assume the SchemaContexts match and performs data
 272      * pruning in an attempt to adjust the state to our current SchemaContext.
 273      *
 274      * @param payload Payload
 275      * @throws IOException when the snapshot fails to deserialize
 276      * @throws DataValidationFailedException when the snapshot fails to apply
 277      */
 278     void applyRecoveryPayload(final @Nonnull Payload payload) throws IOException, DataValidationFailedException {
 279         if (payload instanceof CommitTransactionPayload) {
 280             final Entry<TransactionIdentifier, DataTreeCandidate> e = ((CommitTransactionPayload) payload).getCandidate();
 281             applyRecoveryCandidate(e.getValue());
 282             allMetadataCommittedTransaction(e.getKey());
 283         } else if (payload instanceof DataTreeCandidatePayload) {
 284             applyRecoveryCandidate(((DataTreeCandidatePayload) payload).getCandidate());
 285         } else {
 286             LOG.warn("{}: ignoring unhandled payload {}", logContext, payload);
 287         }
 288     }
 289
 290     private void applyReplicatedCandidate(final Identifier identifier, final DataTreeCandidate foreign)
 291             throws DataValidationFailedException {
 292         LOG.debug("{}: Applying foreign transaction {}", logContext, identifier);
 293
 294         final DataTreeModification mod = dataTree.takeSnapshot().newModification();
 295         DataTreeCandidates.applyToModification(mod, foreign);
 296         mod.ready();
 297
 298         LOG.trace("{}: Applying foreign modification {}", logContext, mod);
 299         dataTree.validate(mod);
 300         final DataTreeCandidate candidate = dataTree.prepare(mod);
 301         dataTree.commit(candidate);
 302
 303         notifyListeners(candidate);
 304     }
 305
 306     /**
 307      * Apply a payload coming from the leader, which could actually be us. This method assumes the leader and follower
 308      * SchemaContexts match and does not perform any pruning.
 309      *
 310      * @param identifier Payload identifier as returned from RaftActor
 311      * @param payload Payload
 312      * @throws IOException when the snapshot fails to deserialize
 313      * @throws DataValidationFailedException when the snapshot fails to apply
 314      */
 315     void applyReplicatedPayload(final Identifier identifier, final Payload payload) throws IOException,
 316             DataValidationFailedException {
 317         /*
 318          * 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
 319          * if we are the leader and it has originated with us.
 320          *
 321          * The identifier will only ever be non-null when we were the leader which achieved consensus. Unfortunately,
 322          * though, this may not be the case anymore, as we are being called some time afterwards and we may not be
 323          * acting in that capacity anymore.
 324          *
 325          * In any case, we know that this is an entry coming from replication, hence we can be sure we will not observe
 326          * pre-Boron state -- which limits the number of options here.
 327          */
 328         if (payload instanceof CommitTransactionPayload) {
 329             if (identifier == null) {
 330                 final Entry<TransactionIdentifier, DataTreeCandidate> e = ((CommitTransactionPayload) payload).getCandidate();
 331                 applyReplicatedCandidate(e.getKey(), e.getValue());
 332                 allMetadataCommittedTransaction(e.getKey());
 333             } else {
 334                 Verify.verify(identifier instanceof TransactionIdentifier);
 335                 payloadReplicationComplete((TransactionIdentifier) identifier);
 336             }
 337         } else {
 338             LOG.warn("{}: ignoring unhandled identifier {} payload {}", logContext, identifier, payload);
 339         }
 340     }
 341
 342     private void payloadReplicationComplete(final TransactionIdentifier txId) {
 343         final CommitEntry current = pendingTransactions.peek();
 344         if (current == null) {
 345             LOG.warn("{}: No outstanding transactions, ignoring consensus on transaction {}", logContext, txId);
 346             return;
 347         }
 348
 349         if (!current.cohort.getIdentifier().equals(txId)) {
 350             LOG.warn("{}: Head of queue is {}, ignoring consensus on transaction {}", logContext,
 351                 current.cohort.getIdentifier(), txId);
 352             return;
 353         }
 354
 355         finishCommit(current.cohort);
 356     }
 357
 358     private void allMetadataCommittedTransaction(final TransactionIdentifier txId) {
 359         for (ShardDataTreeMetadata<?> m : metadata) {
 360             m.onTransactionCommitted(txId);
 361         }
 362     }
 363
 364     private ShardDataTreeTransactionChain ensureTransactionChain(final LocalHistoryIdentifier localHistoryIdentifier) {
 365         ShardDataTreeTransactionChain chain = transactionChains.get(localHistoryIdentifier);
 366         if (chain == null) {
 367             chain = new ShardDataTreeTransactionChain(localHistoryIdentifier, this);
 368             transactionChains.put(localHistoryIdentifier, chain);
 369         }
 370
 371         return chain;
 372     }
 373
 374     ReadOnlyShardDataTreeTransaction newReadOnlyTransaction(final TransactionIdentifier txId) {
 375         if (txId.getHistoryId().getHistoryId() == 0) {
 376             return new ReadOnlyShardDataTreeTransaction(txId, dataTree.takeSnapshot());
 377         }
 378
 379         return ensureTransactionChain(txId.getHistoryId()).newReadOnlyTransaction(txId);
 380     }
 381
 382     ReadWriteShardDataTreeTransaction newReadWriteTransaction(final TransactionIdentifier txId) {
 383         if (txId.getHistoryId().getHistoryId() == 0) {
 384             return new ReadWriteShardDataTreeTransaction(ShardDataTree.this, txId, dataTree.takeSnapshot()
 385                     .newModification());
 386         }
 387
 388         return ensureTransactionChain(txId.getHistoryId()).newReadWriteTransaction(txId);
 389     }
 390
 391     public void notifyListeners(final DataTreeCandidate candidate) {
 392         treeChangeListenerPublisher.publishChanges(candidate, logContext);
 393         dataChangeListenerPublisher.publishChanges(candidate, logContext);
 394     }
 395
 396     void notifyOfInitialData(final DataChangeListenerRegistration<AsyncDataChangeListener<YangInstanceIdentifier,
 397             NormalizedNode<?, ?>>> listenerReg, final Optional<DataTreeCandidate> currentState) {
 398         if (currentState.isPresent()) {
 399             ShardDataChangeListenerPublisher localPublisher = dataChangeListenerPublisher.newInstance();
 400             localPublisher.registerDataChangeListener(listenerReg.getPath(), listenerReg.getInstance(),
 401                     listenerReg.getScope());
 402             localPublisher.publishChanges(currentState.get(), logContext);
 403         }
 404     }
 405
 406     void notifyOfInitialData(final YangInstanceIdentifier path, final DOMDataTreeChangeListener listener,
 407             final Optional<DataTreeCandidate> currentState) {
 408         if (currentState.isPresent()) {
 409             ShardDataTreeChangeListenerPublisher localPublisher = treeChangeListenerPublisher.newInstance();
 410             localPublisher.registerTreeChangeListener(path, listener);
 411             localPublisher.publishChanges(currentState.get(), logContext);
 412         }
 413     }
 414
 415     void closeAllTransactionChains() {
 416         for (ShardDataTreeTransactionChain chain : transactionChains.values()) {
 417             chain.close();
 418         }
 419
 420         transactionChains.clear();
 421     }
 422
 423     void closeTransactionChain(final LocalHistoryIdentifier transactionChainId) {
 424         final ShardDataTreeTransactionChain chain = transactionChains.remove(transactionChainId);
 425         if (chain != null) {
 426             chain.close();
 427         } else {
 428             LOG.debug("{}: Closing non-existent transaction chain {}", logContext, transactionChainId);
 429         }
 430     }
 431
 432     Entry<DataChangeListenerRegistration<AsyncDataChangeListener<YangInstanceIdentifier, NormalizedNode<?, ?>>>,
 433             Optional<DataTreeCandidate>> registerChangeListener(final YangInstanceIdentifier path,
 434                     final AsyncDataChangeListener<YangInstanceIdentifier, NormalizedNode<?, ?>> listener,
 435                     final DataChangeScope scope) {
 436         final DataChangeListenerRegistration<AsyncDataChangeListener<YangInstanceIdentifier, NormalizedNode<?, ?>>> reg =
 437                 dataChangeListenerPublisher.registerDataChangeListener(path, listener, scope);
 438
 439         return new SimpleEntry<>(reg, readCurrentData());
 440     }
 441
 442     private Optional<DataTreeCandidate> readCurrentData() {
 443         final Optional<NormalizedNode<?, ?>> currentState = dataTree.takeSnapshot().readNode(YangInstanceIdentifier.EMPTY);
 444         return currentState.isPresent() ? Optional.of(DataTreeCandidates.fromNormalizedNode(
 445             YangInstanceIdentifier.EMPTY, currentState.get())) : Optional.<DataTreeCandidate>absent();
 446     }
 447
 448     public Entry<ListenerRegistration<DOMDataTreeChangeListener>, Optional<DataTreeCandidate>> registerTreeChangeListener(
 449             final YangInstanceIdentifier path, final DOMDataTreeChangeListener listener) {
 450         final ListenerRegistration<DOMDataTreeChangeListener> reg = treeChangeListenerPublisher.registerTreeChangeListener(
 451                 path, listener);
 452
 453         return new SimpleEntry<>(reg, readCurrentData());
 454     }
 455
 456     int getQueueSize() {
 457         return pendingTransactions.size();
 458     }
 459
 460     @Override
 461     void abortTransaction(final AbstractShardDataTreeTransaction<?> transaction) {
 462         // Intentional no-op
 463     }
 464
 465     @Override
 466     ShardDataTreeCohort finishTransaction(final ReadWriteShardDataTreeTransaction transaction) {
 467         final DataTreeModification snapshot = transaction.getSnapshot();
 468         snapshot.ready();
 469
 470         return createReadyCohort(transaction.getId(), snapshot);
 471     }
 472
 473     public Optional<NormalizedNode<?, ?>> readNode(final YangInstanceIdentifier path) {
 474         return dataTree.takeSnapshot().readNode(path);
 475     }
 476
 477     public DataTreeSnapshot takeSnapshot() {
 478         return dataTree.takeSnapshot();
 479     }
 480
 481     public DataTreeModification newModification() {
 482         return dataTree.takeSnapshot().newModification();
 483     }
 484
 485     /**
 486      * @deprecated This method violates DataTree containment and will be removed.
 487      */
 488     @VisibleForTesting
 489     @Deprecated
 490     public DataTreeCandidate commit(final DataTreeModification modification) throws DataValidationFailedException {
 491         modification.ready();
 492         dataTree.validate(modification);
 493         DataTreeCandidate candidate = dataTree.prepare(modification);
 494         dataTree.commit(candidate);
 495         return candidate;
 496     }
 497
 498     public Collection<ShardDataTreeCohort> getAndClearPendingTransactions() {
 499         Collection<ShardDataTreeCohort> ret = new ArrayList<>(pendingTransactions.size());
 500         for(CommitEntry entry: pendingTransactions) {
 501             ret.add(entry.cohort);
 502         }
 503
 504         pendingTransactions.clear();
 505         return ret;
 506     }
 507
 508     private void processNextTransaction() {
 509         while (!pendingTransactions.isEmpty()) {
 510             final CommitEntry entry = pendingTransactions.peek();
 511             final SimpleShardDataTreeCohort cohort = entry.cohort;
 512             final DataTreeModification modification = cohort.getDataTreeModification();
 513
 514             if(cohort.getState() != State.CAN_COMMIT_PENDING) {
 515                 break;
 516             }
 517
 518             LOG.debug("{}: Validating transaction {}", logContext, cohort.getIdentifier());
 519             Exception cause;
 520             try {
 521                 dataTree.validate(modification);
 522                 LOG.debug("{}: Transaction {} validated", logContext, cohort.getIdentifier());
 523                 cohort.successfulCanCommit();
 524                 entry.lastAccess = shard.ticker().read();
 525                 return;
 526             } catch (ConflictingModificationAppliedException e) {
 527                 LOG.warn("{}: Store Tx {}: Conflicting modification for path {}.", logContext, cohort.getIdentifier(),
 528                     e.getPath());
 529                 cause = new OptimisticLockFailedException("Optimistic lock failed.", e);
 530             } catch (DataValidationFailedException e) {
 531                 LOG.warn("{}: Store Tx {}: Data validation failed for path {}.", logContext, cohort.getIdentifier(),
 532                     e.getPath(), e);
 533
 534                 // For debugging purposes, allow dumping of the modification. Coupled with the above
 535                 // precondition log, it should allow us to understand what went on.
 536                 LOG.debug("{}: Store Tx {}: modifications: {} tree: {}", cohort.getIdentifier(), modification, dataTree);
 537                 cause = new TransactionCommitFailedException("Data did not pass validation.", e);
 538             } catch (Exception e) {
 539                 LOG.warn("{}: Unexpected failure in validation phase", logContext, e);
 540                 cause = e;
 541             }
 542
 543             // Failure path: propagate the failure, remove the transaction from the queue and loop to the next one
 544             pendingTransactions.poll().cohort.failedCanCommit(cause);
 545         }
 546
 547         maybeRunOperationOnPendingTransactionsComplete();
 548     }
 549
 550     void startCanCommit(final SimpleShardDataTreeCohort cohort) {
 551         final SimpleShardDataTreeCohort current = pendingTransactions.peek().cohort;
 552         if (!cohort.equals(current)) {
 553             LOG.debug("{}: Transaction {} scheduled for canCommit step", logContext, cohort.getIdentifier());
 554             return;
 555         }
 556
 557         processNextTransaction();
 558     }
 559
 560     private void failPreCommit(final Exception cause) {
 561         shard.getShardMBean().incrementFailedTransactionsCount();
 562         pendingTransactions.poll().cohort.failedPreCommit(cause);
 563         processNextTransaction();
 564     }
 565
 566     void startPreCommit(final SimpleShardDataTreeCohort cohort) {
 567         final CommitEntry entry = pendingTransactions.peek();
 568         Preconditions.checkState(entry != null, "Attempted to pre-commit of %s when no transactions pending", cohort);
 569
 570         final SimpleShardDataTreeCohort current = entry.cohort;
 571         Verify.verify(cohort.equals(current), "Attempted to pre-commit %s while %s is pending", cohort, current);
 572         final DataTreeCandidateTip candidate;
 573         try {
 574             candidate = dataTree.prepare(cohort.getDataTreeModification());
 575         } catch (Exception e) {
 576             failPreCommit(e);
 577             return;
 578         }
 579
 580         try {
 581             cohort.userPreCommit(candidate);
 582         } catch (ExecutionException | TimeoutException e) {
 583             failPreCommit(e);
 584             return;
 585         }
 586
 587         entry.lastAccess = shard.ticker().read();
 588         cohort.successfulPreCommit(candidate);
 589     }
 590
 591     private void failCommit(final Exception cause) {
 592         shard.getShardMBean().incrementFailedTransactionsCount();
 593         pendingTransactions.poll().cohort.failedCommit(cause);
 594         processNextTransaction();
 595     }
 596
 597     private void finishCommit(final SimpleShardDataTreeCohort cohort) {
 598         final TransactionIdentifier txId = cohort.getIdentifier();
 599         final DataTreeCandidate candidate = cohort.getCandidate();
 600
 601         LOG.debug("{}: Resuming commit of transaction {}", logContext, txId);
 602
 603         try {
 604             dataTree.commit(candidate);
 605         } catch (Exception e) {
 606             LOG.error("{}: Failed to commit transaction {}", logContext, txId, e);
 607             failCommit(e);
 608             return;
 609         }
 610
 611         shard.getShardMBean().incrementCommittedTransactionCount();
 612         shard.getShardMBean().setLastCommittedTransactionTime(System.currentTimeMillis());
 613
 614         // FIXME: propagate journal index
 615         pendingTransactions.poll().cohort.successfulCommit(UnsignedLong.ZERO);
 616
 617         LOG.trace("{}: Transaction {} committed, proceeding to notify", logContext, txId);
 618         notifyListeners(candidate);
 619
 620         processNextTransaction();
 621     }
 622
 623     void startCommit(final SimpleShardDataTreeCohort cohort, final DataTreeCandidate candidate) {
 624         final CommitEntry entry = pendingTransactions.peek();
 625         Preconditions.checkState(entry != null, "Attempted to start commit of %s when no transactions pending", cohort);
 626
 627         final SimpleShardDataTreeCohort current = entry.cohort;
 628         Verify.verify(cohort.equals(current), "Attempted to commit %s while %s is pending", cohort, current);
 629
 630         if (shard.canSkipPayload() || candidate.getRootNode().getModificationType() == ModificationType.UNMODIFIED) {
 631             LOG.debug("{}: No replication required, proceeding to finish commit", logContext);
 632             finishCommit(cohort);
 633             return;
 634         }
 635
 636         final TransactionIdentifier txId = cohort.getIdentifier();
 637         final Payload payload;
 638         try {
 639             payload = CommitTransactionPayload.create(txId, candidate);
 640         } catch (IOException e) {
 641             LOG.error("{}: Failed to encode transaction {} candidate {}", logContext, txId, candidate, e);
 642             pendingTransactions.poll().cohort.failedCommit(e);
 643             return;
 644         }
 645
 646         // Once completed, we will continue via payloadReplicationComplete
 647         entry.lastAccess = shard.ticker().read();
 648         shard.persistPayload(txId, payload);
 649         LOG.debug("{}: Transaction {} submitted to persistence", logContext, txId);
 650     }
 651
 652     void processCohortRegistryCommand(final ActorRef sender, final CohortRegistryCommand message) {
 653         cohortRegistry.process(sender, message);
 654     }
 655
 656     ShardDataTreeCohort createReadyCohort(final TransactionIdentifier txId,
 657             final DataTreeModification modification) {
 658         SimpleShardDataTreeCohort cohort = new SimpleShardDataTreeCohort(this, modification, txId,
 659                 cohortRegistry.createCohort(schemaContext, txId, COMMIT_STEP_TIMEOUT));
 660         pendingTransactions.add(new CommitEntry(cohort, shard.ticker().read()));
 661         return cohort;
 662     }
 663
 664     void checkForExpiredTransactions(final long transactionCommitTimeoutMillis) {
 665         final long timeout = TimeUnit.MILLISECONDS.toNanos(transactionCommitTimeoutMillis);
 666         final long now = shard.ticker().read();
 667         final CommitEntry currentTx = pendingTransactions.peek();
 668         if (currentTx != null && currentTx.lastAccess + timeout < now) {
 669             LOG.warn("{}: Current transaction {} has timed out after {} ms in state {}", logContext,
 670                     currentTx.cohort.getIdentifier(), transactionCommitTimeoutMillis, currentTx.cohort.getState());
 671             boolean processNext = true;
 672             switch (currentTx.cohort.getState()) {
 673                 case CAN_COMMIT_PENDING:
 674                     pendingTransactions.poll().cohort.failedCanCommit(new TimeoutException());
 675                     break;
 676                 case CAN_COMMIT_COMPLETE:
 677                     pendingTransactions.poll().cohort.reportFailure(new TimeoutException());
 678                     break;
 679                 case PRE_COMMIT_PENDING:
 680                     pendingTransactions.poll().cohort.failedPreCommit(new TimeoutException());
 681                     break;
 682                 case PRE_COMMIT_COMPLETE:
 683                     // FIXME: this is a legacy behavior problem. Three-phase commit protocol specifies that after we
 684                     //        are ready we should commit the transaction, not abort it. Our current software stack does
 685                     //        not allow us to do that consistently, because we persist at the time of commit, hence
 686                     //        we can end up in a state where we have pre-committed a transaction, then a leader failover
 687                     //        occurred ... the new leader does not see the pre-committed transaction and does not have
 688                     //        a running timer. To fix this we really need two persistence events.
 689                     //
 690                     //        The first one, done at pre-commit time will hold the transaction payload. When consensus
 691                     //        is reached, we exit the pre-commit phase and start the pre-commit timer. Followers do not
 692                     //        apply the state in this event.
 693                     //
 694                     //        The second one, done at commit (or abort) time holds only the transaction identifier and
 695                     //        signals to followers that the state should (or should not) be applied.
 696                     //
 697                     //        In order to make the pre-commit timer working across failovers, though, we need
 698                     //        a per-shard cluster-wide monotonic time, so a follower becoming the leader can accurately
 699                     //        restart the timer.
 700                     pendingTransactions.poll().cohort.reportFailure(new TimeoutException());
 701                     break;
 702                 case COMMIT_PENDING:
 703                     LOG.warn("{}: Transaction {} is still committing, cannot abort", logContext,
 704                         currentTx.cohort.getIdentifier());
 705                     currentTx.lastAccess = now;
 706                     processNext = false;
 707                     return;
 708                 case ABORTED:
 709                 case COMMITTED:
 710                 case FAILED:
 711                 case READY:
 712                 default:
 713                     pendingTransactions.poll();
 714             }
 715
 716             if (processNext) {
 717                 processNextTransaction();
 718             }
 719         }
 720     }
 721
 722     void startAbort(final SimpleShardDataTreeCohort cohort) {
 723         final Iterator<CommitEntry> it = pendingTransactions.iterator();
 724         if (!it.hasNext()) {
 725             LOG.debug("{}: no open transaction while attempting to abort {}", logContext, cohort.getIdentifier());
 726             return;
 727         }
 728
 729         // First entry is special, as it may already be committing
 730         final CommitEntry first = it.next();
 731         if (cohort.equals(first.cohort)) {
 732             if (cohort.getState() != State.COMMIT_PENDING) {
 733                 LOG.debug("{}: aborted head of queue {} in state {}", logContext, cohort.getIdentifier(),
 734                     cohort.getIdentifier());
 735                 pendingTransactions.poll();
 736                 processNextTransaction();
 737             } else {
 738                 LOG.warn("{}: transaction {} is committing, skipping abort", logContext, cohort.getIdentifier());
 739             }
 740
 741             return;
 742         }
 743
 744         while (it.hasNext()) {
 745             final CommitEntry e = it.next();
 746             if (cohort.equals(e.cohort)) {
 747                 LOG.debug("{}: aborting queued transaction {}", logContext, cohort.getIdentifier());
 748                 it.remove();
 749                 return;
 750             }
 751         }
 752
 753         LOG.debug("{}: aborted transaction {} not found in the queue", logContext, cohort.getIdentifier());
 754     }
 755
 756     void setRunOnPendingTransactionsComplete(final Runnable operation) {
 757         runOnPendingTransactionsComplete = operation;
 758         maybeRunOperationOnPendingTransactionsComplete();
 759     }
 760
 761     private void maybeRunOperationOnPendingTransactionsComplete() {
 762       if (runOnPendingTransactionsComplete != null && pendingTransactions.isEmpty()) {
 763           LOG.debug("{}: Pending transactions complete - running operation {}", logContext,
 764                   runOnPendingTransactionsComplete);
 765
 766           runOnPendingTransactionsComplete.run();
 767           runOnPendingTransactionsComplete = null;
 768       }
 769   }
 770 }