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