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         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     private PruningDataTreeModification wrapWithPruning(final DataTreeModification delegate) {
 219         return new PruningDataTreeModification(delegate, dataTree, schemaContext);
 220     }
 221
 222     private static DataTreeModification unwrap(final DataTreeModification modification) {
 223         if (modification instanceof PruningDataTreeModification) {
 224             return ((PruningDataTreeModification)modification).delegate();
 225         }
 226         return modification;
 227     }
 228
 229     /**
 230      * Apply a snapshot coming from recovery. This method does not assume the SchemaContexts match and performs data
 231      * pruning in an attempt to adjust the state to our current SchemaContext.
 232      *
 233      * @param snapshot Snapshot that needs to be applied
 234      * @throws DataValidationFailedException when the snapshot fails to apply
 235      */
 236     void applyRecoverySnapshot(final @Nonnull ShardDataTreeSnapshot snapshot) throws DataValidationFailedException {
 237         applySnapshot(snapshot, this::wrapWithPruning);
 238     }
 239
 240
 241     /**
 242      * Apply a snapshot coming from the leader. This method assumes the leader and follower SchemaContexts match and
 243      * does not perform any pruning.
 244      *
 245      * @param snapshot Snapshot that needs to be applied
 246      * @throws DataValidationFailedException when the snapshot fails to apply
 247      */
 248     void applySnapshot(final @Nonnull ShardDataTreeSnapshot snapshot) throws DataValidationFailedException {
 249         applySnapshot(snapshot, UnaryOperator.identity());
 250     }
 251
 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 = ((CommitTransactionPayload) payload).getCandidate();
 284             applyRecoveryCandidate(e.getValue());
 285             allMetadataCommittedTransaction(e.getKey());
 286         } else if (payload instanceof DataTreeCandidatePayload) {
 287             applyRecoveryCandidate(((DataTreeCandidatePayload) payload).getCandidate());
 288         } else {
 289             LOG.warn("{}: ignoring unhandled payload {}", logContext, payload);
 290         }
 291     }
 292
 293     private void applyReplicatedCandidate(final Identifier identifier, final DataTreeCandidate foreign)
 294             throws DataValidationFailedException {
 295         LOG.debug("{}: Applying foreign transaction {}", logContext, identifier);
 296
 297         final DataTreeModification mod = dataTree.takeSnapshot().newModification();
 298         DataTreeCandidates.applyToModification(mod, foreign);
 299         mod.ready();
 300
 301         LOG.trace("{}: Applying foreign modification {}", logContext, mod);
 302         dataTree.validate(mod);
 303         final DataTreeCandidate candidate = dataTree.prepare(mod);
 304         dataTree.commit(candidate);
 305
 306         notifyListeners(candidate);
 307     }
 308
 309     /**
 310      * Apply a payload coming from the leader, which could actually be us. This method assumes the leader and follower
 311      * SchemaContexts match and does not perform any pruning.
 312      *
 313      * @param identifier Payload identifier as returned from RaftActor
 314      * @param payload Payload
 315      * @throws IOException when the snapshot fails to deserialize
 316      * @throws DataValidationFailedException when the snapshot fails to apply
 317      */
 318     void applyReplicatedPayload(final Identifier identifier, final Payload payload) throws IOException,
 319             DataValidationFailedException {
 320         /*
 321          * 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
 322          * if we are the leader and it has originated with us.
 323          *
 324          * The identifier will only ever be non-null when we were the leader which achieved consensus. Unfortunately,
 325          * though, this may not be the case anymore, as we are being called some time afterwards and we may not be
 326          * acting in that capacity anymore.
 327          *
 328          * In any case, we know that this is an entry coming from replication, hence we can be sure we will not observe
 329          * pre-Boron state -- which limits the number of options here.
 330          */
 331         if (payload instanceof CommitTransactionPayload) {
 332             if (identifier == null) {
 333                 final Entry<TransactionIdentifier, DataTreeCandidate> e = ((CommitTransactionPayload) payload).getCandidate();
 334                 applyReplicatedCandidate(e.getKey(), e.getValue());
 335                 allMetadataCommittedTransaction(e.getKey());
 336             } else {
 337                 Verify.verify(identifier instanceof TransactionIdentifier);
 338                 payloadReplicationComplete((TransactionIdentifier) identifier);
 339             }
 340         } else {
 341             LOG.debug("{}: ignoring unhandled identifier {} payload {}", logContext, identifier, payload);
 342         }
 343     }
 344
 345     private void payloadReplicationComplete(final TransactionIdentifier txId) {
 346         final CommitEntry current = pendingTransactions.peek();
 347         if (current == null) {
 348             LOG.warn("{}: No outstanding transactions, ignoring consensus on transaction {}", logContext, txId);
 349             return;
 350         }
 351
 352         if (!current.cohort.getIdentifier().equals(txId)) {
 353             LOG.warn("{}: Head of queue is {}, ignoring consensus on transaction {}", logContext,
 354                 current.cohort.getIdentifier(), txId);
 355             return;
 356         }
 357
 358         finishCommit(current.cohort);
 359     }
 360
 361     private void allMetadataCommittedTransaction(final TransactionIdentifier txId) {
 362         for (ShardDataTreeMetadata<?> m : metadata) {
 363             m.onTransactionCommitted(txId);
 364         }
 365     }
 366
 367     private ShardDataTreeTransactionChain ensureTransactionChain(final LocalHistoryIdentifier localHistoryIdentifier) {
 368         ShardDataTreeTransactionChain chain = transactionChains.get(localHistoryIdentifier);
 369         if (chain == null) {
 370             chain = new ShardDataTreeTransactionChain(localHistoryIdentifier, this);
 371             transactionChains.put(localHistoryIdentifier, chain);
 372         }
 373
 374         return chain;
 375     }
 376
 377     ReadOnlyShardDataTreeTransaction newReadOnlyTransaction(final TransactionIdentifier txId) {
 378         if (txId.getHistoryId().getHistoryId() == 0) {
 379             return new ReadOnlyShardDataTreeTransaction(txId, dataTree.takeSnapshot());
 380         }
 381
 382         return ensureTransactionChain(txId.getHistoryId()).newReadOnlyTransaction(txId);
 383     }
 384
 385     ReadWriteShardDataTreeTransaction newReadWriteTransaction(final TransactionIdentifier txId) {
 386         if (txId.getHistoryId().getHistoryId() == 0) {
 387             return new ReadWriteShardDataTreeTransaction(ShardDataTree.this, txId, dataTree.takeSnapshot()
 388                     .newModification());
 389         }
 390
 391         return ensureTransactionChain(txId.getHistoryId()).newReadWriteTransaction(txId);
 392     }
 393
 394     public void notifyListeners(final DataTreeCandidate candidate) {
 395         treeChangeListenerPublisher.publishChanges(candidate, logContext);
 396         dataChangeListenerPublisher.publishChanges(candidate, logContext);
 397     }
 398
 399     void notifyOfInitialData(final DataChangeListenerRegistration<AsyncDataChangeListener<YangInstanceIdentifier,
 400             NormalizedNode<?, ?>>> listenerReg, final Optional<DataTreeCandidate> currentState) {
 401         if (currentState.isPresent()) {
 402             ShardDataChangeListenerPublisher localPublisher = dataChangeListenerPublisher.newInstance();
 403             localPublisher.registerDataChangeListener(listenerReg.getPath(), listenerReg.getInstance(),
 404                     listenerReg.getScope());
 405             localPublisher.publishChanges(currentState.get(), logContext);
 406         }
 407     }
 408
 409     void notifyOfInitialData(final YangInstanceIdentifier path, final DOMDataTreeChangeListener listener,
 410             final Optional<DataTreeCandidate> currentState) {
 411         if (currentState.isPresent()) {
 412             ShardDataTreeChangeListenerPublisher localPublisher = treeChangeListenerPublisher.newInstance();
 413             localPublisher.registerTreeChangeListener(path, listener);
 414             localPublisher.publishChanges(currentState.get(), logContext);
 415         }
 416     }
 417
 418     void closeAllTransactionChains() {
 419         for (ShardDataTreeTransactionChain chain : transactionChains.values()) {
 420             chain.close();
 421         }
 422
 423         transactionChains.clear();
 424     }
 425
 426     void closeTransactionChain(final LocalHistoryIdentifier transactionChainId) {
 427         final ShardDataTreeTransactionChain chain = transactionChains.remove(transactionChainId);
 428         if (chain != null) {
 429             chain.close();
 430         } else {
 431             LOG.debug("{}: Closing non-existent transaction chain {}", logContext, transactionChainId);
 432         }
 433     }
 434
 435     Entry<DataChangeListenerRegistration<AsyncDataChangeListener<YangInstanceIdentifier, NormalizedNode<?, ?>>>,
 436             Optional<DataTreeCandidate>> registerChangeListener(final YangInstanceIdentifier path,
 437                     final AsyncDataChangeListener<YangInstanceIdentifier, NormalizedNode<?, ?>> listener,
 438                     final DataChangeScope scope) {
 439         final DataChangeListenerRegistration<AsyncDataChangeListener<YangInstanceIdentifier, NormalizedNode<?, ?>>> reg =
 440                 dataChangeListenerPublisher.registerDataChangeListener(path, listener, scope);
 441
 442         return new SimpleEntry<>(reg, readCurrentData());
 443     }
 444
 445     private Optional<DataTreeCandidate> readCurrentData() {
 446         final Optional<NormalizedNode<?, ?>> currentState = dataTree.takeSnapshot().readNode(YangInstanceIdentifier.EMPTY);
 447         return currentState.isPresent() ? Optional.of(DataTreeCandidates.fromNormalizedNode(
 448             YangInstanceIdentifier.EMPTY, currentState.get())) : Optional.<DataTreeCandidate>absent();
 449     }
 450
 451     public Entry<ListenerRegistration<DOMDataTreeChangeListener>, Optional<DataTreeCandidate>> registerTreeChangeListener(
 452             final YangInstanceIdentifier path, final DOMDataTreeChangeListener listener) {
 453         final ListenerRegistration<DOMDataTreeChangeListener> reg = treeChangeListenerPublisher.registerTreeChangeListener(
 454                 path, listener);
 455
 456         return new SimpleEntry<>(reg, readCurrentData());
 457     }
 458
 459     int getQueueSize() {
 460         return pendingTransactions.size();
 461     }
 462
 463     @Override
 464     void abortTransaction(final AbstractShardDataTreeTransaction<?> transaction) {
 465         // Intentional no-op
 466     }
 467
 468     @Override
 469     ShardDataTreeCohort finishTransaction(final ReadWriteShardDataTreeTransaction transaction) {
 470         final DataTreeModification snapshot = transaction.getSnapshot();
 471         snapshot.ready();
 472
 473         return createReadyCohort(transaction.getId(), snapshot);
 474     }
 475
 476     public Optional<NormalizedNode<?, ?>> readNode(final YangInstanceIdentifier path) {
 477         return dataTree.takeSnapshot().readNode(path);
 478     }
 479
 480     public DataTreeSnapshot takeSnapshot() {
 481         return dataTree.takeSnapshot();
 482     }
 483
 484     public DataTreeModification newModification() {
 485         return dataTree.takeSnapshot().newModification();
 486     }
 487
 488     /**
 489      * @deprecated This method violates DataTree containment and will be removed.
 490      */
 491     @VisibleForTesting
 492     @Deprecated
 493     public DataTreeCandidate commit(final DataTreeModification modification) throws DataValidationFailedException {
 494         modification.ready();
 495         dataTree.validate(modification);
 496         DataTreeCandidate candidate = dataTree.prepare(modification);
 497         dataTree.commit(candidate);
 498         return candidate;
 499     }
 500
 501     public Collection<ShardDataTreeCohort> getAndClearPendingTransactions() {
 502         Collection<ShardDataTreeCohort> ret = new ArrayList<>(pendingTransactions.size());
 503         for(CommitEntry entry: pendingTransactions) {
 504             ret.add(entry.cohort);
 505         }
 506
 507         pendingTransactions.clear();
 508         return ret;
 509     }
 510
 511     private void processNextTransaction() {
 512         while (!pendingTransactions.isEmpty()) {
 513             final CommitEntry entry = pendingTransactions.peek();
 514             final SimpleShardDataTreeCohort cohort = entry.cohort;
 515             final DataTreeModification modification = cohort.getDataTreeModification();
 516
 517             if(cohort.getState() != State.CAN_COMMIT_PENDING) {
 518                 break;
 519             }
 520
 521             LOG.debug("{}: Validating transaction {}", logContext, cohort.getIdentifier());
 522             Exception cause;
 523             try {
 524                 dataTree.validate(modification);
 525                 LOG.debug("{}: Transaction {} validated", logContext, cohort.getIdentifier());
 526                 cohort.successfulCanCommit();
 527                 entry.lastAccess = shard.ticker().read();
 528                 return;
 529             } catch (ConflictingModificationAppliedException e) {
 530                 LOG.warn("{}: Store Tx {}: Conflicting modification for path {}.", logContext, cohort.getIdentifier(),
 531                     e.getPath());
 532                 cause = new OptimisticLockFailedException("Optimistic lock failed.", e);
 533             } catch (DataValidationFailedException e) {
 534                 LOG.warn("{}: Store Tx {}: Data validation failed for path {}.", logContext, cohort.getIdentifier(),
 535                     e.getPath(), e);
 536
 537                 // For debugging purposes, allow dumping of the modification. Coupled with the above
 538                 // precondition log, it should allow us to understand what went on.
 539                 LOG.debug("{}: Store Tx {}: modifications: {} tree: {}", cohort.getIdentifier(), modification, dataTree);
 540                 cause = new TransactionCommitFailedException("Data did not pass validation.", e);
 541             } catch (Exception e) {
 542                 LOG.warn("{}: Unexpected failure in validation phase", logContext, e);
 543                 cause = e;
 544             }
 545
 546             // Failure path: propagate the failure, remove the transaction from the queue and loop to the next one
 547             pendingTransactions.poll().cohort.failedCanCommit(cause);
 548         }
 549
 550         maybeRunOperationOnPendingTransactionsComplete();
 551     }
 552
 553     void startCanCommit(final SimpleShardDataTreeCohort cohort) {
 554         final SimpleShardDataTreeCohort current = pendingTransactions.peek().cohort;
 555         if (!cohort.equals(current)) {
 556             LOG.debug("{}: Transaction {} scheduled for canCommit step", logContext, cohort.getIdentifier());
 557             return;
 558         }
 559
 560         processNextTransaction();
 561     }
 562
 563     private void failPreCommit(final Exception cause) {
 564         shard.getShardMBean().incrementFailedTransactionsCount();
 565         pendingTransactions.poll().cohort.failedPreCommit(cause);
 566         processNextTransaction();
 567     }
 568
 569     void startPreCommit(final SimpleShardDataTreeCohort cohort) {
 570         final CommitEntry entry = pendingTransactions.peek();
 571         Preconditions.checkState(entry != null, "Attempted to pre-commit of %s when no transactions pending", cohort);
 572
 573         final SimpleShardDataTreeCohort current = entry.cohort;
 574         Verify.verify(cohort.equals(current), "Attempted to pre-commit %s while %s is pending", cohort, current);
 575         final DataTreeCandidateTip candidate;
 576         try {
 577             candidate = dataTree.prepare(cohort.getDataTreeModification());
 578         } catch (Exception e) {
 579             failPreCommit(e);
 580             return;
 581         }
 582
 583         try {
 584             cohort.userPreCommit(candidate);
 585         } catch (ExecutionException | TimeoutException e) {
 586             failPreCommit(e);
 587             return;
 588         }
 589
 590         entry.lastAccess = shard.ticker().read();
 591         cohort.successfulPreCommit(candidate);
 592     }
 593
 594     private void failCommit(final Exception cause) {
 595         shard.getShardMBean().incrementFailedTransactionsCount();
 596         pendingTransactions.poll().cohort.failedCommit(cause);
 597         processNextTransaction();
 598     }
 599
 600     private void finishCommit(final SimpleShardDataTreeCohort cohort) {
 601         final TransactionIdentifier txId = cohort.getIdentifier();
 602         final DataTreeCandidate candidate = cohort.getCandidate();
 603
 604         LOG.debug("{}: Resuming commit of transaction {}", logContext, txId);
 605
 606         try {
 607             dataTree.commit(candidate);
 608         } catch (Exception e) {
 609             LOG.error("{}: Failed to commit transaction {}", logContext, txId, e);
 610             failCommit(e);
 611             return;
 612         }
 613
 614         shard.getShardMBean().incrementCommittedTransactionCount();
 615         shard.getShardMBean().setLastCommittedTransactionTime(System.currentTimeMillis());
 616
 617         // FIXME: propagate journal index
 618         pendingTransactions.poll().cohort.successfulCommit(UnsignedLong.ZERO);
 619
 620         LOG.trace("{}: Transaction {} committed, proceeding to notify", logContext, txId);
 621         notifyListeners(candidate);
 622
 623         processNextTransaction();
 624     }
 625
 626     void startCommit(final SimpleShardDataTreeCohort cohort, final DataTreeCandidate candidate) {
 627         final CommitEntry entry = pendingTransactions.peek();
 628         Preconditions.checkState(entry != null, "Attempted to start commit of %s when no transactions pending", cohort);
 629
 630         final SimpleShardDataTreeCohort current = entry.cohort;
 631         Verify.verify(cohort.equals(current), "Attempted to commit %s while %s is pending", cohort, current);
 632
 633         if (shard.canSkipPayload() || candidate.getRootNode().getModificationType() == ModificationType.UNMODIFIED) {
 634             LOG.debug("{}: No replication required, proceeding to finish commit", logContext);
 635             finishCommit(cohort);
 636             return;
 637         }
 638
 639         final TransactionIdentifier txId = cohort.getIdentifier();
 640         final Payload payload;
 641         try {
 642             payload = CommitTransactionPayload.create(txId, candidate);
 643         } catch (IOException e) {
 644             LOG.error("{}: Failed to encode transaction {} candidate {}", logContext, txId, candidate, e);
 645             pendingTransactions.poll().cohort.failedCommit(e);
 646             return;
 647         }
 648
 649         // Once completed, we will continue via payloadReplicationComplete
 650         entry.lastAccess = shard.ticker().read();
 651         shard.persistPayload(txId, payload);
 652         LOG.debug("{}: Transaction {} submitted to persistence", logContext, txId);
 653     }
 654
 655     void processCohortRegistryCommand(final ActorRef sender, final CohortRegistryCommand message) {
 656         cohortRegistry.process(sender, message);
 657     }
 658
 659     ShardDataTreeCohort createReadyCohort(final TransactionIdentifier txId,
 660             final DataTreeModification modification) {
 661         SimpleShardDataTreeCohort cohort = new SimpleShardDataTreeCohort(this, modification, txId,
 662                 cohortRegistry.createCohort(schemaContext, txId, COMMIT_STEP_TIMEOUT));
 663         pendingTransactions.add(new CommitEntry(cohort, shard.ticker().read()));
 664         return cohort;
 665     }
 666
 667     void checkForExpiredTransactions(final long transactionCommitTimeoutMillis) {
 668         final long timeout = TimeUnit.MILLISECONDS.toNanos(transactionCommitTimeoutMillis);
 669         final long now = shard.ticker().read();
 670         final CommitEntry currentTx = pendingTransactions.peek();
 671         if (currentTx != null && currentTx.lastAccess + timeout < now) {
 672             LOG.warn("{}: Current transaction {} has timed out after {} ms in state {}", logContext,
 673                     currentTx.cohort.getIdentifier(), transactionCommitTimeoutMillis, currentTx.cohort.getState());
 674             boolean processNext = true;
 675             switch (currentTx.cohort.getState()) {
 676                 case CAN_COMMIT_PENDING:
 677                     pendingTransactions.poll().cohort.failedCanCommit(new TimeoutException());
 678                     break;
 679                 case CAN_COMMIT_COMPLETE:
 680                     pendingTransactions.poll().cohort.reportFailure(new TimeoutException());
 681                     break;
 682                 case PRE_COMMIT_PENDING:
 683                     pendingTransactions.poll().cohort.failedPreCommit(new TimeoutException());
 684                     break;
 685                 case PRE_COMMIT_COMPLETE:
 686                     // FIXME: this is a legacy behavior problem. Three-phase commit protocol specifies that after we
 687                     //        are ready we should commit the transaction, not abort it. Our current software stack does
 688                     //        not allow us to do that consistently, because we persist at the time of commit, hence
 689                     //        we can end up in a state where we have pre-committed a transaction, then a leader failover
 690                     //        occurred ... the new leader does not see the pre-committed transaction and does not have
 691                     //        a running timer. To fix this we really need two persistence events.
 692                     //
 693                     //        The first one, done at pre-commit time will hold the transaction payload. When consensus
 694                     //        is reached, we exit the pre-commit phase and start the pre-commit timer. Followers do not
 695                     //        apply the state in this event.
 696                     //
 697                     //        The second one, done at commit (or abort) time holds only the transaction identifier and
 698                     //        signals to followers that the state should (or should not) be applied.
 699                     //
 700                     //        In order to make the pre-commit timer working across failovers, though, we need
 701                     //        a per-shard cluster-wide monotonic time, so a follower becoming the leader can accurately
 702                     //        restart the timer.
 703                     pendingTransactions.poll().cohort.reportFailure(new TimeoutException());
 704                     break;
 705                 case COMMIT_PENDING:
 706                     LOG.warn("{}: Transaction {} is still committing, cannot abort", logContext,
 707                         currentTx.cohort.getIdentifier());
 708                     currentTx.lastAccess = now;
 709                     processNext = false;
 710                     return;
 711                 case ABORTED:
 712                 case COMMITTED:
 713                 case FAILED:
 714                 case READY:
 715                 default:
 716                     pendingTransactions.poll();
 717             }
 718
 719             if (processNext) {
 720                 processNextTransaction();
 721             }
 722         }
 723     }
 724
 725     void startAbort(final SimpleShardDataTreeCohort cohort) {
 726         final Iterator<CommitEntry> it = pendingTransactions.iterator();
 727         if (!it.hasNext()) {
 728             LOG.debug("{}: no open transaction while attempting to abort {}", logContext, cohort.getIdentifier());
 729             return;
 730         }
 731
 732         // First entry is special, as it may already be committing
 733         final CommitEntry first = it.next();
 734         if (cohort.equals(first.cohort)) {
 735             if (cohort.getState() != State.COMMIT_PENDING) {
 736                 LOG.debug("{}: aborted head of queue {} in state {}", logContext, cohort.getIdentifier(),
 737                     cohort.getIdentifier());
 738                 pendingTransactions.poll();
 739                 processNextTransaction();
 740             } else {
 741                 LOG.warn("{}: transaction {} is committing, skipping abort", logContext, cohort.getIdentifier());
 742             }
 743
 744             return;
 745         }
 746
 747         while (it.hasNext()) {
 748             final CommitEntry e = it.next();
 749             if (cohort.equals(e.cohort)) {
 750                 LOG.debug("{}: aborting queued transaction {}", logContext, cohort.getIdentifier());
 751                 it.remove();
 752                 return;
 753             }
 754         }
 755
 756         LOG.debug("{}: aborted transaction {} not found in the queue", logContext, cohort.getIdentifier());
 757     }
 758
 759     void setRunOnPendingTransactionsComplete(final Runnable operation) {
 760         runOnPendingTransactionsComplete = operation;
 761         maybeRunOperationOnPendingTransactionsComplete();
 762     }
 763
 764     private void maybeRunOperationOnPendingTransactionsComplete() {
 765       if (runOnPendingTransactionsComplete != null && pendingTransactions.isEmpty()) {
 766           LOG.debug("{}: Pending transactions complete - running operation {}", logContext,
 767                   runOnPendingTransactionsComplete);
 768
 769           runOnPendingTransactionsComplete.run();
 770           runOnPendingTransactionsComplete = null;
 771       }
 772   }
 773 }