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

   1 /*
   2  * Copyright (c) 2014 Brocade Communications 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.actor.Status.Failure;
  12 import akka.serialization.Serialization;
  13 import com.google.common.annotations.VisibleForTesting;
  14 import com.google.common.base.Preconditions;
  15 import com.google.common.primitives.UnsignedLong;
  16 import com.google.common.util.concurrent.FutureCallback;
  17 import java.util.ArrayDeque;
  18 import java.util.ArrayList;
  19 import java.util.Collection;
  20 import java.util.Collections;
  21 import java.util.Deque;
  22 import java.util.HashMap;
  23 import java.util.Iterator;
  24 import java.util.LinkedList;
  25 import java.util.Map;
  26 import java.util.concurrent.ExecutionException;
  27 import java.util.concurrent.TimeoutException;
  28 import javax.annotation.Nonnull;
  29 import org.opendaylight.controller.cluster.access.concepts.TransactionIdentifier;
  30 import org.opendaylight.controller.cluster.datastore.messages.AbortTransactionReply;
  31 import org.opendaylight.controller.cluster.datastore.messages.BatchedModifications;
  32 import org.opendaylight.controller.cluster.datastore.messages.BatchedModificationsReply;
  33 import org.opendaylight.controller.cluster.datastore.messages.CanCommitTransaction;
  34 import org.opendaylight.controller.cluster.datastore.messages.CanCommitTransactionReply;
  35 import org.opendaylight.controller.cluster.datastore.messages.CommitTransaction;
  36 import org.opendaylight.controller.cluster.datastore.messages.CommitTransactionReply;
  37 import org.opendaylight.controller.cluster.datastore.messages.ForwardedReadyTransaction;
  38 import org.opendaylight.controller.cluster.datastore.messages.ReadyLocalTransaction;
  39 import org.opendaylight.controller.cluster.datastore.messages.ReadyTransactionReply;
  40 import org.opendaylight.controller.cluster.datastore.messages.VersionedExternalizableMessage;
  41 import org.opendaylight.controller.cluster.datastore.utils.AbstractBatchedModificationsCursor;
  42 import org.opendaylight.yangtools.concepts.Identifier;
  43 import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeCandidate;
  44 import org.slf4j.Logger;
  45
  46 /**
  47  * Coordinates commits for a shard ensuring only one concurrent 3-phase commit.
  48  *
  49  * @author Thomas Pantelis
  50  */
  51 final class ShardCommitCoordinator {
  52
  53     // Interface hook for unit tests to replace or decorate the ShardDataTreeCohorts.
  54     @VisibleForTesting
  55     public interface CohortDecorator {
  56         ShardDataTreeCohort decorate(Identifier transactionID, ShardDataTreeCohort actual);
  57     }
  58
  59     private final Map<Identifier, CohortEntry> cohortCache = new HashMap<>();
  60
  61     private final ShardDataTree dataTree;
  62
  63     private final Logger log;
  64
  65     private final String name;
  66
  67     // This is a hook for unit tests to replace or decorate the ShardDataTreeCohorts.
  68     @VisibleForTesting
  69     private CohortDecorator cohortDecorator;
  70
  71     private ReadyTransactionReply readyTransactionReply;
  72
  73     ShardCommitCoordinator(final ShardDataTree dataTree, final Logger log, final String name) {
  74         this.log = log;
  75         this.name = name;
  76         this.dataTree = Preconditions.checkNotNull(dataTree);
  77     }
  78
  79     int getCohortCacheSize() {
  80         return cohortCache.size();
  81     }
  82
  83     private String persistenceId() {
  84         return dataTree.logContext();
  85     }
  86
  87     private ReadyTransactionReply readyTransactionReply(final ActorRef cohort) {
  88         if (readyTransactionReply == null) {
  89             readyTransactionReply = new ReadyTransactionReply(Serialization.serializedActorPath(cohort));
  90         }
  91
  92         return readyTransactionReply;
  93     }
  94
  95     /**
  96      * This method is called to ready a transaction that was prepared by ShardTransaction actor. It caches
  97      * the prepared cohort entry for the given transactions ID in preparation for the subsequent 3-phase commit.
  98      *
  99      * @param ready the ForwardedReadyTransaction message to process
 100      * @param sender the sender of the message
 101      * @param shard the transaction's shard actor
 102      */
 103     void handleForwardedReadyTransaction(final ForwardedReadyTransaction ready, final ActorRef sender,
 104             final Shard shard) {
 105         log.debug("{}: Readying transaction {}, client version {}", name,
 106                 ready.getTransactionId(), ready.getTxnClientVersion());
 107
 108         final ShardDataTreeCohort cohort = ready.getTransaction().ready();
 109         final CohortEntry cohortEntry = CohortEntry.createReady(cohort, ready.getTxnClientVersion());
 110         cohortCache.put(cohortEntry.getTransactionId(), cohortEntry);
 111
 112         if (ready.isDoImmediateCommit()) {
 113             cohortEntry.setDoImmediateCommit(true);
 114             cohortEntry.setReplySender(sender);
 115             cohortEntry.setShard(shard);
 116             handleCanCommit(cohortEntry);
 117         } else {
 118             // The caller does not want immediate commit - the 3-phase commit will be coordinated by the
 119             // front-end so send back a ReadyTransactionReply with our actor path.
 120             sender.tell(readyTransactionReply(shard.self()), shard.self());
 121         }
 122     }
 123
 124     /**
 125      * This method handles a BatchedModifications message for a transaction being prepared directly on the
 126      * Shard actor instead of via a ShardTransaction actor. If there's no currently cached
 127      * DOMStoreWriteTransaction, one is created. The batched modifications are applied to the write Tx. If
 128      * the BatchedModifications is ready to commit then a DOMStoreThreePhaseCommitCohort is created.
 129      *
 130      * @param batched the BatchedModifications message to process
 131      * @param sender the sender of the message
 132      */
 133     void handleBatchedModifications(final BatchedModifications batched, final ActorRef sender, final Shard shard) {
 134         CohortEntry cohortEntry = cohortCache.get(batched.getTransactionId());
 135         if (cohortEntry == null) {
 136             cohortEntry = CohortEntry.createOpen(dataTree.newReadWriteTransaction(batched.getTransactionId()),
 137                 batched.getVersion());
 138             cohortCache.put(cohortEntry.getTransactionId(), cohortEntry);
 139         }
 140
 141         if (log.isDebugEnabled()) {
 142             log.debug("{}: Applying {} batched modifications for Tx {}", name,
 143                     batched.getModifications().size(), batched.getTransactionId());
 144         }
 145
 146         cohortEntry.applyModifications(batched.getModifications());
 147
 148         if (batched.isReady()) {
 149             if (cohortEntry.getLastBatchedModificationsException() != null) {
 150                 cohortCache.remove(cohortEntry.getTransactionId());
 151                 throw cohortEntry.getLastBatchedModificationsException();
 152             }
 153
 154             if (cohortEntry.getTotalBatchedModificationsReceived() != batched.getTotalMessagesSent()) {
 155                 cohortCache.remove(cohortEntry.getTransactionId());
 156                 throw new IllegalStateException(String.format(
 157                         "The total number of batched messages received %d does not match the number sent %d",
 158                         cohortEntry.getTotalBatchedModificationsReceived(), batched.getTotalMessagesSent()));
 159             }
 160
 161             if (log.isDebugEnabled()) {
 162                 log.debug("{}: Readying Tx {}, client version {}", name,
 163                         batched.getTransactionId(), batched.getVersion());
 164             }
 165
 166             cohortEntry.setDoImmediateCommit(batched.isDoCommitOnReady());
 167             cohortEntry.ready(cohortDecorator);
 168
 169             if (batched.isDoCommitOnReady()) {
 170                 cohortEntry.setReplySender(sender);
 171                 cohortEntry.setShard(shard);
 172                 handleCanCommit(cohortEntry);
 173             } else {
 174                 sender.tell(readyTransactionReply(shard.self()), shard.self());
 175             }
 176         } else {
 177             sender.tell(new BatchedModificationsReply(batched.getModifications().size()), shard.self());
 178         }
 179     }
 180
 181     /**
 182      * This method handles {@link ReadyLocalTransaction} message. All transaction modifications have
 183      * been prepared beforehand by the sender and we just need to drive them through into the
 184      * dataTree.
 185      *
 186      * @param message the ReadyLocalTransaction message to process
 187      * @param sender the sender of the message
 188      * @param shard the transaction's shard actor
 189      */
 190     void handleReadyLocalTransaction(final ReadyLocalTransaction message, final ActorRef sender, final Shard shard) {
 191         final ShardDataTreeCohort cohort = dataTree.createReadyCohort(message.getTransactionId(),
 192             message.getModification());
 193         final CohortEntry cohortEntry = CohortEntry.createReady(cohort, DataStoreVersions.CURRENT_VERSION);
 194         cohortCache.put(cohortEntry.getTransactionId(), cohortEntry);
 195         cohortEntry.setDoImmediateCommit(message.isDoCommitOnReady());
 196
 197         log.debug("{}: Applying local modifications for Tx {}", name, message.getTransactionId());
 198
 199         if (message.isDoCommitOnReady()) {
 200             cohortEntry.setReplySender(sender);
 201             cohortEntry.setShard(shard);
 202             handleCanCommit(cohortEntry);
 203         } else {
 204             sender.tell(readyTransactionReply(shard.self()), shard.self());
 205         }
 206     }
 207
 208     Collection<BatchedModifications> createForwardedBatchedModifications(final BatchedModifications from,
 209             final int maxModificationsPerBatch) {
 210         CohortEntry cohortEntry = cohortCache.remove(from.getTransactionId());
 211         if (cohortEntry == null || cohortEntry.getTransaction() == null) {
 212             return Collections.singletonList(from);
 213         }
 214
 215         cohortEntry.applyModifications(from.getModifications());
 216
 217         final LinkedList<BatchedModifications> newModifications = new LinkedList<>();
 218         cohortEntry.getTransaction().getSnapshot().applyToCursor(new AbstractBatchedModificationsCursor() {
 219             @Override
 220             protected BatchedModifications getModifications() {
 221                 if (newModifications.isEmpty()
 222                         || newModifications.getLast().getModifications().size() >= maxModificationsPerBatch) {
 223                     newModifications.add(new BatchedModifications(from.getTransactionId(), from.getVersion()));
 224                 }
 225
 226                 return newModifications.getLast();
 227             }
 228         });
 229
 230         BatchedModifications last = newModifications.getLast();
 231         last.setDoCommitOnReady(from.isDoCommitOnReady());
 232         last.setReady(from.isReady());
 233         last.setTotalMessagesSent(newModifications.size());
 234         return newModifications;
 235     }
 236
 237     private void handleCanCommit(final CohortEntry cohortEntry) {
 238         cohortEntry.canCommit(new FutureCallback<Void>() {
 239             @Override
 240             public void onSuccess(final Void result) {
 241                 log.debug("{}: canCommit for {}: success", name, cohortEntry.getTransactionId());
 242
 243                 if (cohortEntry.isDoImmediateCommit()) {
 244                     doCommit(cohortEntry);
 245                 } else {
 246                     cohortEntry.getReplySender().tell(
 247                         CanCommitTransactionReply.yes(cohortEntry.getClientVersion()).toSerializable(),
 248                         cohortEntry.getShard().self());
 249                 }
 250             }
 251
 252             @Override
 253             public void onFailure(final Throwable failure) {
 254                 log.debug("{}: An exception occurred during canCommit for {}: {}", name,
 255                         cohortEntry.getTransactionId(), failure);
 256
 257                 cohortCache.remove(cohortEntry.getTransactionId());
 258                 cohortEntry.getReplySender().tell(new Failure(failure), cohortEntry.getShard().self());
 259             }
 260         });
 261     }
 262
 263     /**
 264      * This method handles the canCommit phase for a transaction.
 265      *
 266      * @param transactionID the ID of the transaction to canCommit
 267      * @param sender the actor to which to send the response
 268      * @param shard the transaction's shard actor
 269      */
 270     void handleCanCommit(final Identifier transactionID, final ActorRef sender, final Shard shard) {
 271         // Lookup the cohort entry that was cached previously (or should have been) by
 272         // transactionReady (via the ForwardedReadyTransaction message).
 273         final CohortEntry cohortEntry = cohortCache.get(transactionID);
 274         if (cohortEntry == null) {
 275             // Either canCommit was invoked before ready (shouldn't happen) or a long time passed
 276             // between canCommit and ready and the entry was expired from the cache or it was aborted.
 277             IllegalStateException ex = new IllegalStateException(
 278                     String.format("%s: Cannot canCommit transaction %s - no cohort entry found", name, transactionID));
 279             log.error(ex.getMessage());
 280             sender.tell(new Failure(ex), shard.self());
 281             return;
 282         }
 283
 284         cohortEntry.setReplySender(sender);
 285         cohortEntry.setShard(shard);
 286
 287         handleCanCommit(cohortEntry);
 288     }
 289
 290     private void doCommit(final CohortEntry cohortEntry) {
 291         log.debug("{}: Committing transaction {}", name, cohortEntry.getTransactionId());
 292
 293         // We perform the preCommit phase here atomically with the commit phase. This is an
 294         // optimization to eliminate the overhead of an extra preCommit message. We lose front-end
 295         // coordination of preCommit across shards in case of failure but preCommit should not
 296         // normally fail since we ensure only one concurrent 3-phase commit.
 297         cohortEntry.preCommit(new FutureCallback<DataTreeCandidate>() {
 298             @Override
 299             public void onSuccess(final DataTreeCandidate candidate) {
 300                 finishCommit(cohortEntry.getReplySender(), cohortEntry);
 301             }
 302
 303             @Override
 304             public void onFailure(final Throwable failure) {
 305                 log.error("{} An exception occurred while preCommitting transaction {}", name,
 306                         cohortEntry.getTransactionId(), failure);
 307
 308                 cohortCache.remove(cohortEntry.getTransactionId());
 309                 cohortEntry.getReplySender().tell(new Failure(failure), cohortEntry.getShard().self());
 310             }
 311         });
 312     }
 313
 314     private void finishCommit(@Nonnull final ActorRef sender, @Nonnull final CohortEntry cohortEntry) {
 315         log.debug("{}: Finishing commit for transaction {}", persistenceId(), cohortEntry.getTransactionId());
 316
 317         cohortEntry.commit(new FutureCallback<UnsignedLong>() {
 318             @Override
 319             public void onSuccess(final UnsignedLong result) {
 320                 final TransactionIdentifier txId = cohortEntry.getTransactionId();
 321                 log.debug("{}: Transaction {} committed as {}, sending response to {}", persistenceId(), txId, result,
 322                     sender);
 323
 324                 cohortCache.remove(cohortEntry.getTransactionId());
 325                 sender.tell(CommitTransactionReply.instance(cohortEntry.getClientVersion()).toSerializable(),
 326                     cohortEntry.getShard().self());
 327             }
 328
 329             @Override
 330             public void onFailure(final Throwable failure) {
 331                 log.error("{}, An exception occurred while committing transaction {}", persistenceId(),
 332                         cohortEntry.getTransactionId(), failure);
 333
 334                 cohortCache.remove(cohortEntry.getTransactionId());
 335                 sender.tell(new Failure(failure), cohortEntry.getShard().self());
 336             }
 337         });
 338     }
 339
 340     /**
 341      * This method handles the preCommit and commit phases for a transaction.
 342      *
 343      * @param transactionID the ID of the transaction to commit
 344      * @param sender the actor to which to send the response
 345      * @param shard the transaction's shard actor
 346      */
 347     void handleCommit(final Identifier transactionID, final ActorRef sender, final Shard shard) {
 348         final CohortEntry cohortEntry = cohortCache.get(transactionID);
 349         if (cohortEntry == null) {
 350             // Either a long time passed between canCommit and commit and the entry was expired from the cache
 351             // or it was aborted.
 352             IllegalStateException ex = new IllegalStateException(
 353                     String.format("%s: Cannot commit transaction %s - no cohort entry found", name, transactionID));
 354             log.error(ex.getMessage());
 355             sender.tell(new Failure(ex), shard.self());
 356             return;
 357         }
 358
 359         cohortEntry.setReplySender(sender);
 360         doCommit(cohortEntry);
 361     }
 362
 363     @SuppressWarnings("checkstyle:IllegalCatch")
 364     void handleAbort(final Identifier transactionID, final ActorRef sender, final Shard shard) {
 365         CohortEntry cohortEntry = cohortCache.remove(transactionID);
 366         if (cohortEntry == null) {
 367             return;
 368         }
 369
 370         log.debug("{}: Aborting transaction {}", name, transactionID);
 371
 372         final ActorRef self = shard.getSelf();
 373         try {
 374             cohortEntry.abort();
 375
 376             shard.getShardMBean().incrementAbortTransactionsCount();
 377
 378             if (sender != null) {
 379                 sender.tell(AbortTransactionReply.instance(cohortEntry.getClientVersion()).toSerializable(), self);
 380             }
 381         } catch (InterruptedException | ExecutionException | TimeoutException e) {
 382             log.error("{}: An exception happened during abort", name, e);
 383
 384             if (sender != null) {
 385                 sender.tell(new Failure(e), self);
 386             }
 387         }
 388     }
 389
 390     void checkForExpiredTransactions(final long timeout, final Shard shard) {
 391         Iterator<CohortEntry> iter = cohortCache.values().iterator();
 392         while (iter.hasNext()) {
 393             CohortEntry cohortEntry = iter.next();
 394             if (cohortEntry.isFailed()) {
 395                 iter.remove();
 396             }
 397         }
 398     }
 399
 400     void abortPendingTransactions(final String reason, final Shard shard) {
 401         final Failure failure = new Failure(new RuntimeException(reason));
 402         Collection<ShardDataTreeCohort> pending = dataTree.getAndClearPendingTransactions();
 403
 404         log.debug("{}: Aborting {} pending queued transactions", name, pending.size());
 405
 406         for (ShardDataTreeCohort cohort : pending) {
 407             CohortEntry cohortEntry = cohortCache.remove(cohort.getIdentifier());
 408             if (cohortEntry == null) {
 409                 continue;
 410             }
 411
 412             if (cohortEntry.getReplySender() != null) {
 413                 cohortEntry.getReplySender().tell(failure, shard.self());
 414             }
 415         }
 416
 417         cohortCache.clear();
 418     }
 419
 420     Collection<?> convertPendingTransactionsToMessages(final int maxModificationsPerBatch) {
 421         final Collection<VersionedExternalizableMessage> messages = new ArrayList<>();
 422         for (ShardDataTreeCohort cohort : dataTree.getAndClearPendingTransactions()) {
 423             CohortEntry cohortEntry = cohortCache.remove(cohort.getIdentifier());
 424             if (cohortEntry == null) {
 425                 continue;
 426             }
 427
 428             final Deque<BatchedModifications> newMessages = new ArrayDeque<>();
 429             cohortEntry.getDataTreeModification().applyToCursor(new AbstractBatchedModificationsCursor() {
 430                 @Override
 431                 protected BatchedModifications getModifications() {
 432                     final BatchedModifications lastBatch = newMessages.peekLast();
 433
 434                     if (lastBatch != null && lastBatch.getModifications().size() >= maxModificationsPerBatch) {
 435                         return lastBatch;
 436                     }
 437
 438                     // Allocate a new message
 439                     final BatchedModifications ret = new BatchedModifications(cohortEntry.getTransactionId(),
 440                         cohortEntry.getClientVersion());
 441                     newMessages.add(ret);
 442                     return ret;
 443                 }
 444             });
 445
 446             final BatchedModifications last = newMessages.peekLast();
 447             if (last != null) {
 448                 final boolean immediate = cohortEntry.isDoImmediateCommit();
 449                 last.setDoCommitOnReady(immediate);
 450                 last.setReady(true);
 451                 last.setTotalMessagesSent(newMessages.size());
 452
 453                 messages.addAll(newMessages);
 454
 455                 if (!immediate) {
 456                     switch (cohort.getState()) {
 457                         case CAN_COMMIT_COMPLETE:
 458                         case CAN_COMMIT_PENDING:
 459                             messages.add(new CanCommitTransaction(cohortEntry.getTransactionId(),
 460                                 cohortEntry.getClientVersion()));
 461                             break;
 462                         case PRE_COMMIT_COMPLETE:
 463                         case PRE_COMMIT_PENDING:
 464                             messages.add(new CommitTransaction(cohortEntry.getTransactionId(),
 465                                 cohortEntry.getClientVersion()));
 466                             break;
 467                         default:
 468                             break;
 469                     }
 470                 }
 471             }
 472         }
 473
 474         return messages;
 475     }
 476
 477     @VisibleForTesting
 478     void setCohortDecorator(final CohortDecorator cohortDecorator) {
 479         this.cohortDecorator = cohortDecorator;
 480     }
 481 }