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

   1 /*
   2  * Copyright (c) 2015 Brocade Communications Systems, Inc. and others.  All rights reserved.
   3  * Copyright (c) 2015 Cisco Systems, Inc. and others.  All rights reserved.
   4  *
   5  * This program and the accompanying materials are made available under the
   6  * terms of the Eclipse Public License v1.0 which accompanies this distribution,
   7  * and is available at http://www.eclipse.org/legal/epl-v10.html
   8  */
   9 package org.opendaylight.controller.cluster.datastore;
  10
  11 import akka.actor.ActorSelection;
  12 import akka.dispatch.OnComplete;
  13 import akka.pattern.AskTimeoutException;
  14 import akka.util.Timeout;
  15 import com.google.common.base.Preconditions;
  16 import java.util.concurrent.TimeUnit;
  17 import org.opendaylight.controller.cluster.datastore.exceptions.NoShardLeaderException;
  18 import org.opendaylight.controller.cluster.datastore.exceptions.ShardLeaderNotRespondingException;
  19 import org.opendaylight.controller.cluster.datastore.identifiers.TransactionIdentifier;
  20 import org.opendaylight.controller.cluster.datastore.messages.CreateTransaction;
  21 import org.opendaylight.controller.cluster.datastore.messages.CreateTransactionReply;
  22 import org.opendaylight.controller.cluster.datastore.messages.PrimaryShardInfo;
  23 import org.opendaylight.controller.cluster.datastore.utils.ActorContext;
  24 import org.slf4j.Logger;
  25 import org.slf4j.LoggerFactory;
  26 import scala.concurrent.Future;
  27 import scala.concurrent.duration.FiniteDuration;
  28
  29 /**
  30  * Handles creation of TransactionContext instances for remote transactions. This class creates
  31  * remote transactions, if necessary, by sending CreateTransaction messages with retries, up to a limit,
  32  * if the shard doesn't have a leader yet. This is done by scheduling a retry task after a short delay.
  33  * <p>
  34  * The end result from a completed CreateTransaction message is a TransactionContext that is
  35  * used to perform transaction operations. Transaction operations that occur before the
  36  * CreateTransaction completes are cache via a TransactionContextWrapper and executed once the
  37  * CreateTransaction completes, successfully or not.
  38  */
  39 final class RemoteTransactionContextSupport {
  40     private static final Logger LOG = LoggerFactory.getLogger(RemoteTransactionContextSupport.class);
  41
  42     private static final long CREATE_TX_TRY_INTERVAL_IN_MS = 1000;
  43     private static final long MAX_CREATE_TX_MSG_TIMEOUT_IN_MS = 5000;
  44
  45     private final TransactionProxy parent;
  46     private final String shardName;
  47
  48     /**
  49      * The target primary shard.
  50      */
  51     private volatile PrimaryShardInfo primaryShardInfo;
  52
  53     /**
  54      * The total timeout for creating a tx on the primary shard.
  55      */
  56     private volatile long totalCreateTxTimeout;
  57
  58     private final Timeout createTxMessageTimeout;
  59
  60     private final TransactionContextWrapper transactionContextWrapper;
  61
  62     RemoteTransactionContextSupport(final TransactionContextWrapper transactionContextWrapper, final TransactionProxy parent,
  63             final String shardName) {
  64         this.parent = Preconditions.checkNotNull(parent);
  65         this.shardName = shardName;
  66         this.transactionContextWrapper = transactionContextWrapper;
  67
  68         // For the total create tx timeout, use 2 times the election timeout. This should be enough time for
  69         // a leader re-election to occur if we happen to hit it in transition.
  70         totalCreateTxTimeout = parent.getActorContext().getDatastoreContext().getShardRaftConfig()
  71                 .getElectionTimeOutInterval().toMillis() * 2;
  72
  73         // We'll use the operationTimeout for the the create Tx message timeout so it can be set appropriately
  74         // for unit tests but cap it at MAX_CREATE_TX_MSG_TIMEOUT_IN_MS. The operationTimeout could be set
  75         // larger than the totalCreateTxTimeout in production which we don't want.
  76         long operationTimeout = parent.getActorContext().getOperationTimeout().duration().toMillis();
  77         createTxMessageTimeout = new Timeout(Math.min(operationTimeout, MAX_CREATE_TX_MSG_TIMEOUT_IN_MS),
  78                 TimeUnit.MILLISECONDS);
  79     }
  80
  81     String getShardName() {
  82         return shardName;
  83     }
  84
  85     private TransactionType getTransactionType() {
  86         return parent.getType();
  87     }
  88
  89     private ActorContext getActorContext() {
  90         return parent.getActorContext();
  91     }
  92
  93     private TransactionIdentifier getIdentifier() {
  94         return parent.getIdentifier();
  95     }
  96
  97     /**
  98      * Sets the target primary shard and initiates a CreateTransaction try.
  99      */
 100     void setPrimaryShard(PrimaryShardInfo primaryShardInfo) {
 101         this.primaryShardInfo = primaryShardInfo;
 102
 103         if (getTransactionType() == TransactionType.WRITE_ONLY  &&
 104                 getActorContext().getDatastoreContext().isWriteOnlyTransactionOptimizationsEnabled()) {
 105             ActorSelection primaryShard = primaryShardInfo.getPrimaryShardActor();
 106
 107             LOG.debug("Tx {} Primary shard {} found - creating WRITE_ONLY transaction context",
 108                 getIdentifier(), primaryShard);
 109
 110             // For write-only Tx's we prepare the transaction modifications directly on the shard actor
 111             // to avoid the overhead of creating a separate transaction actor.
 112             transactionContextWrapper.executePriorTransactionOperations(createValidTransactionContext(
 113                     primaryShard, primaryShard.path().toString(), primaryShardInfo.getPrimaryShardVersion()));
 114         } else {
 115             tryCreateTransaction();
 116         }
 117     }
 118
 119     /**
 120      * Performs a CreateTransaction try async.
 121      */
 122     private void tryCreateTransaction() {
 123         if(LOG.isDebugEnabled()) {
 124             LOG.debug("Tx {} Primary shard {} found - trying create transaction", getIdentifier(),
 125                     primaryShardInfo.getPrimaryShardActor());
 126         }
 127
 128         Object serializedCreateMessage = new CreateTransaction(getIdentifier().toString(),
 129                 getTransactionType().ordinal(), getIdentifier().getChainId(),
 130                     primaryShardInfo.getPrimaryShardVersion()).toSerializable();
 131
 132         Future<Object> createTxFuture = getActorContext().executeOperationAsync(
 133                 primaryShardInfo.getPrimaryShardActor(), serializedCreateMessage, createTxMessageTimeout);
 134
 135         createTxFuture.onComplete(new OnComplete<Object>() {
 136             @Override
 137             public void onComplete(Throwable failure, Object response) {
 138                 onCreateTransactionComplete(failure, response);
 139             }
 140         }, getActorContext().getClientDispatcher());
 141     }
 142
 143     private void tryFindPrimaryShard() {
 144         LOG.debug("Tx {} Retrying findPrimaryShardAsync for shard {}", getIdentifier(), shardName);
 145
 146         this.primaryShardInfo = null;
 147         Future<PrimaryShardInfo> findPrimaryFuture = getActorContext().findPrimaryShardAsync(shardName);
 148         findPrimaryFuture.onComplete(new OnComplete<PrimaryShardInfo>() {
 149             @Override
 150             public void onComplete(final Throwable failure, final PrimaryShardInfo primaryShardInfo) {
 151                 onFindPrimaryShardComplete(failure, primaryShardInfo);
 152             }
 153         }, getActorContext().getClientDispatcher());
 154     }
 155
 156     private void onFindPrimaryShardComplete(final Throwable failure, final PrimaryShardInfo primaryShardInfo) {
 157         if (failure == null) {
 158             this.primaryShardInfo = primaryShardInfo;
 159             tryCreateTransaction();
 160         } else {
 161             LOG.debug("Tx {}: Find primary for shard {} failed", getIdentifier(), shardName, failure);
 162
 163             onCreateTransactionComplete(failure, null);
 164         }
 165     }
 166
 167     private void onCreateTransactionComplete(Throwable failure, Object response) {
 168         // An AskTimeoutException will occur if the local shard forwards to an unavailable remote leader or
 169         // the cached remote leader actor is no longer available.
 170         boolean retryCreateTransaction = primaryShardInfo != null &&
 171                 (failure instanceof NoShardLeaderException || failure instanceof AskTimeoutException);
 172         if(retryCreateTransaction) {
 173             // Schedule a retry unless we're out of retries. Note: totalCreateTxTimeout is volatile as it may
 174             // be written by different threads however not concurrently, therefore decrementing it
 175             // non-atomically here is ok.
 176             if(totalCreateTxTimeout > 0) {
 177                 long scheduleInterval = CREATE_TX_TRY_INTERVAL_IN_MS;
 178                 if(failure instanceof AskTimeoutException) {
 179                     // Since we use the createTxMessageTimeout for the CreateTransaction request and it timed
 180                     // out, subtract it from the total timeout. Also since the createTxMessageTimeout period
 181                     // has already elapsed, we can immediately schedule the retry (10 ms is virtually immediate).
 182                     totalCreateTxTimeout -= createTxMessageTimeout.duration().toMillis();
 183                     scheduleInterval = 10;
 184                 }
 185
 186                 totalCreateTxTimeout -= scheduleInterval;
 187
 188                 LOG.debug("Tx {}: create tx on shard {} failed with exception \"{}\" - scheduling retry in {} ms",
 189                         getIdentifier(), shardName, failure, scheduleInterval);
 190
 191                 getActorContext().getActorSystem().scheduler().scheduleOnce(
 192                         FiniteDuration.create(scheduleInterval, TimeUnit.MILLISECONDS),
 193                         new Runnable() {
 194                             @Override
 195                             public void run() {
 196                                 tryFindPrimaryShard();
 197                             }
 198                         }, getActorContext().getClientDispatcher());
 199                 return;
 200             }
 201         }
 202
 203         createTransactionContext(failure, response);
 204     }
 205
 206     private void createTransactionContext(Throwable failure, Object response) {
 207         // Create the TransactionContext from the response or failure. Store the new
 208         // TransactionContext locally until we've completed invoking the
 209         // TransactionOperations. This avoids thread timing issues which could cause
 210         // out-of-order TransactionOperations. Eg, on a modification operation, if the
 211         // TransactionContext is non-null, then we directly call the TransactionContext.
 212         // However, at the same time, the code may be executing the cached
 213         // TransactionOperations. So to avoid thus timing, we don't publish the
 214         // TransactionContext until after we've executed all cached TransactionOperations.
 215         TransactionContext localTransactionContext;
 216         if(failure != null) {
 217             LOG.debug("Tx {} Creating NoOpTransaction because of error", getIdentifier(), failure);
 218
 219             Throwable resultingEx = failure;
 220             if(failure instanceof AskTimeoutException) {
 221                 resultingEx = new ShardLeaderNotRespondingException(String.format(
 222                         "Could not create a %s transaction on shard %s. The shard leader isn't responding.",
 223                         parent.getType(), shardName), failure);
 224             } else if(!(failure instanceof NoShardLeaderException)) {
 225                 resultingEx = new Exception(String.format(
 226                     "Error creating %s transaction on shard %s", parent.getType(), shardName), failure);
 227             }
 228
 229             localTransactionContext = new NoOpTransactionContext(resultingEx, getIdentifier());
 230         } else if (CreateTransactionReply.isSerializedType(response)) {
 231             localTransactionContext = createValidTransactionContext(
 232                     CreateTransactionReply.fromSerializable(response));
 233         } else {
 234             IllegalArgumentException exception = new IllegalArgumentException(String.format(
 235                     "Invalid reply type %s for CreateTransaction", response.getClass()));
 236
 237             localTransactionContext = new NoOpTransactionContext(exception, getIdentifier());
 238         }
 239
 240         transactionContextWrapper.executePriorTransactionOperations(localTransactionContext);
 241     }
 242
 243     private TransactionContext createValidTransactionContext(CreateTransactionReply reply) {
 244         LOG.debug("Tx {} Received {}", getIdentifier(), reply);
 245
 246         return createValidTransactionContext(getActorContext().actorSelection(reply.getTransactionPath()),
 247                 reply.getTransactionPath(), primaryShardInfo.getPrimaryShardVersion());
 248     }
 249
 250     private TransactionContext createValidTransactionContext(ActorSelection transactionActor, String transactionPath,
 251             short remoteTransactionVersion) {
 252         final TransactionContext ret = new RemoteTransactionContext(transactionContextWrapper.getIdentifier(),
 253                 transactionActor, getActorContext(), remoteTransactionVersion, transactionContextWrapper.getLimiter());
 254
 255         if(parent.getType() == TransactionType.READ_ONLY) {
 256             TransactionContextCleanup.track(this, ret);
 257         }
 258
 259         return ret;
 260     }
 261 }
 262