X-Git-Url: https://git.opendaylight.org/gerrit/gitweb?p=controller.git;a=blobdiff_plain;f=opendaylight%2Fmd-sal%2Fsal-distributed-datastore%2Fsrc%2Fmain%2Fjava%2Forg%2Fopendaylight%2Fcontroller%2Fcluster%2Fdatastore%2FShard.java;h=dabad41aff97b81ba83dcdf1626dca26e987a1b1;hp=e62e918e5e7eff53b8f7e00dbe1e8300d270b59f;hb=9402730d96f9633d2165de99fad65768d8537a9a;hpb=dea515c8870769408b9bea29f555d6b71ff43211 diff --git a/opendaylight/md-sal/sal-distributed-datastore/src/main/java/org/opendaylight/controller/cluster/datastore/Shard.java b/opendaylight/md-sal/sal-distributed-datastore/src/main/java/org/opendaylight/controller/cluster/datastore/Shard.java index e62e918e5e..dabad41aff 100644 --- a/opendaylight/md-sal/sal-distributed-datastore/src/main/java/org/opendaylight/controller/cluster/datastore/Shard.java +++ b/opendaylight/md-sal/sal-distributed-datastore/src/main/java/org/opendaylight/controller/cluster/datastore/Shard.java @@ -18,9 +18,6 @@ import akka.serialization.Serialization; import com.google.common.annotations.VisibleForTesting; import com.google.common.base.Optional; import com.google.common.base.Preconditions; -import com.google.common.util.concurrent.FutureCallback; -import com.google.common.util.concurrent.Futures; -import com.google.common.util.concurrent.ListenableFuture; import java.io.IOException; import java.util.HashMap; import java.util.Map; @@ -35,7 +32,6 @@ import org.opendaylight.controller.cluster.datastore.identifiers.ShardTransactio import org.opendaylight.controller.cluster.datastore.jmx.mbeans.shard.ShardMBeanFactory; import org.opendaylight.controller.cluster.datastore.jmx.mbeans.shard.ShardStats; import org.opendaylight.controller.cluster.datastore.messages.AbortTransaction; -import org.opendaylight.controller.cluster.datastore.messages.AbortTransactionReply; import org.opendaylight.controller.cluster.datastore.messages.ActorInitialized; import org.opendaylight.controller.cluster.datastore.messages.BatchedModifications; import org.opendaylight.controller.cluster.datastore.messages.CanCommitTransaction; @@ -45,7 +41,9 @@ import org.opendaylight.controller.cluster.datastore.messages.CommitTransactionR import org.opendaylight.controller.cluster.datastore.messages.CreateTransaction; import org.opendaylight.controller.cluster.datastore.messages.CreateTransactionReply; import org.opendaylight.controller.cluster.datastore.messages.ForwardedReadyTransaction; +import org.opendaylight.controller.cluster.datastore.messages.GetShardDataTree; import org.opendaylight.controller.cluster.datastore.messages.PeerAddressResolved; +import org.opendaylight.controller.cluster.datastore.messages.ReadyLocalTransaction; import org.opendaylight.controller.cluster.datastore.messages.RegisterChangeListener; import org.opendaylight.controller.cluster.datastore.messages.RegisterDataTreeChangeListener; import org.opendaylight.controller.cluster.datastore.messages.ShardLeaderStateChanged; @@ -61,6 +59,7 @@ import org.opendaylight.controller.cluster.notifications.RoleChangeNotifier; import org.opendaylight.controller.cluster.raft.RaftActor; import org.opendaylight.controller.cluster.raft.RaftActorRecoveryCohort; import org.opendaylight.controller.cluster.raft.RaftActorSnapshotCohort; +import org.opendaylight.controller.cluster.raft.RaftState; import org.opendaylight.controller.cluster.raft.base.messages.FollowerInitialSyncUpStatus; import org.opendaylight.controller.cluster.raft.messages.AppendEntriesReply; import org.opendaylight.controller.cluster.raft.protobuff.client.messages.CompositeModificationByteStringPayload; @@ -81,7 +80,10 @@ import scala.concurrent.duration.FiniteDuration; */ public class Shard extends RaftActor { - private static final Object TX_COMMIT_TIMEOUT_CHECK_MESSAGE = "txCommitTimeoutCheck"; + protected static final Object TX_COMMIT_TIMEOUT_CHECK_MESSAGE = "txCommitTimeoutCheck"; + + @VisibleForTesting + static final Object GET_SHARD_MBEAN_MESSAGE = "getShardMBeanMessage"; @VisibleForTesting static final String DEFAULT_NAME = "default"; @@ -115,7 +117,8 @@ public class Shard extends RaftActor { protected Shard(final ShardIdentifier name, final Map peerAddresses, final DatastoreContext datastoreContext, final SchemaContext schemaContext) { - super(name.toString(), new HashMap<>(peerAddresses), Optional.of(datastoreContext.getShardRaftConfig())); + super(name.toString(), new HashMap<>(peerAddresses), Optional.of(datastoreContext.getShardRaftConfig()), + DataStoreVersions.CURRENT_VERSION); this.name = name.toString(); this.datastoreContext = datastoreContext; @@ -128,14 +131,14 @@ public class Shard extends RaftActor { shardMBean = ShardMBeanFactory.getShardStatsMBean(name.toString(), datastoreContext.getDataStoreMXBeanType()); - shardMBean.setShardActor(getSelf()); + shardMBean.setShard(this); if (isMetricsCaptureEnabled()) { getContext().become(new MeteringBehavior(this)); } commitCoordinator = new ShardCommitCoordinator(store, - TimeUnit.SECONDS.convert(5, TimeUnit.MINUTES), + datastoreContext.getShardCommitQueueExpiryTimeoutInMillis(), datastoreContext.getShardTransactionCommitQueueCapacity(), self(), LOG, this.name); setTransactionCommitTimeout(); @@ -151,21 +154,17 @@ public class Shard extends RaftActor { Dispatchers.DispatcherType.Transaction), self(), getContext(), shardMBean); snapshotCohort = new ShardSnapshotCohort(transactionActorFactory, store, LOG, this.name); + + } private void setTransactionCommitTimeout() { transactionCommitTimeout = TimeUnit.MILLISECONDS.convert( - datastoreContext.getShardTransactionCommitTimeoutInSeconds(), TimeUnit.SECONDS); + datastoreContext.getShardTransactionCommitTimeoutInSeconds(), TimeUnit.SECONDS) / 2; } - public static Props props(final ShardIdentifier name, - final Map peerAddresses, - final DatastoreContext datastoreContext, final SchemaContext schemaContext) { - Preconditions.checkNotNull(name, "name should not be null"); - Preconditions.checkNotNull(peerAddresses, "peerAddresses should not be null"); - Preconditions.checkNotNull(datastoreContext, "dataStoreContext should not be null"); - Preconditions.checkNotNull(schemaContext, "schemaContext should not be null"); - + public static Props props(final ShardIdentifier name, final Map peerAddresses, + final DatastoreContext datastoreContext, final SchemaContext schemaContext) { return Props.create(new ShardCreator(name, peerAddresses, datastoreContext, schemaContext)); } @@ -217,7 +216,7 @@ public class Shard extends RaftActor { if(context.error().isPresent()){ LOG.trace("{} : AppendEntriesReply failed to arrive at the expected interval {}", persistenceId(), - context.error()); + context.error()); } try { @@ -228,6 +227,8 @@ public class Shard extends RaftActor { } else if (message instanceof ForwardedReadyTransaction) { commitCoordinator.handleForwardedReadyTransaction((ForwardedReadyTransaction) message, getSender(), this); + } else if (message instanceof ReadyLocalTransaction) { + handleReadyLocalTransaction((ReadyLocalTransaction)message); } else if (CanCommitTransaction.SERIALIZABLE_CLASS.isInstance(message)) { handleCanCommitTransaction(CanCommitTransaction.fromSerializable(message)); } else if (CommitTransaction.SERIALIZABLE_CLASS.isInstance(message)) { @@ -237,9 +238,9 @@ public class Shard extends RaftActor { } else if (CloseTransactionChain.SERIALIZABLE_CLASS.isInstance(message)) { closeTransactionChain(CloseTransactionChain.fromSerializable(message)); } else if (message instanceof RegisterChangeListener) { - changeSupport.onMessage((RegisterChangeListener) message, isLeader()); + changeSupport.onMessage((RegisterChangeListener) message, isLeader(), hasLeader()); } else if (message instanceof RegisterDataTreeChangeListener) { - treeChangeSupport.onMessage((RegisterDataTreeChangeListener) message, isLeader()); + treeChangeSupport.onMessage((RegisterDataTreeChangeListener) message, isLeader(), hasLeader()); } else if (message instanceof UpdateSchemaContext) { updateSchemaContext((UpdateSchemaContext) message); } else if (message instanceof PeerAddressResolved) { @@ -252,9 +253,13 @@ public class Shard extends RaftActor { onDatastoreContext((DatastoreContext)message); } else if(message instanceof RegisterRoleChangeListener){ roleChangeNotifier.get().forward(message, context()); - } else if (message instanceof FollowerInitialSyncUpStatus){ + } else if (message instanceof FollowerInitialSyncUpStatus) { shardMBean.setFollowerInitialSyncStatus(((FollowerInitialSyncUpStatus) message).isInitialSyncDone()); context().parent().tell(message, self()); + } else if(GET_SHARD_MBEAN_MESSAGE.equals(message)){ + sender().tell(getShardMBean(), self()); + } else if(message instanceof GetShardDataTree){ + sender().tell(store.getDataTree(), self()); } else { super.onReceiveCommand(message); } @@ -263,18 +268,27 @@ public class Shard extends RaftActor { } } + private boolean hasLeader() { + return getLeaderId() != null; + } + + public int getPendingTxCommitQueueSize() { + return commitCoordinator.getQueueSize(); + } + @Override protected Optional getRoleChangeNotifier() { return roleChangeNotifier; } @Override - protected LeaderStateChanged newLeaderStateChanged(String memberId, String leaderId) { + protected LeaderStateChanged newLeaderStateChanged(String memberId, String leaderId, short leaderPayloadVersion) { return new ShardLeaderStateChanged(memberId, leaderId, - isLeader() ? Optional.of(store.getDataTree()) : Optional.absent()); + isLeader() ? Optional.of(store.getDataTree()) : Optional.absent(), + leaderPayloadVersion); } - private void onDatastoreContext(DatastoreContext context) { + protected void onDatastoreContext(DatastoreContext context) { datastoreContext = context; commitCoordinator.setQueueCapacity(datastoreContext.getShardTransactionCommitQueueCapacity()); @@ -293,30 +307,31 @@ public class Shard extends RaftActor { private void handleTransactionCommitTimeoutCheck() { CohortEntry cohortEntry = commitCoordinator.getCurrentCohortEntry(); if(cohortEntry != null) { - long elapsed = System.currentTimeMillis() - cohortEntry.getLastAccessTime(); - if(elapsed > transactionCommitTimeout) { + if(cohortEntry.isExpired(transactionCommitTimeout)) { LOG.warn("{}: Current transaction {} has timed out after {} ms - aborting", persistenceId(), cohortEntry.getTransactionID(), transactionCommitTimeout); doAbortTransaction(cohortEntry.getTransactionID(), null); } } + + commitCoordinator.cleanupExpiredCohortEntries(); } private static boolean isEmptyCommit(final DataTreeCandidate candidate) { return ModificationType.UNMODIFIED.equals(candidate.getRootNode().getModificationType()); } - void continueCommit(final CohortEntry cohortEntry) throws Exception { - final DataTreeCandidate candidate = cohortEntry.getCohort().getCandidate(); + void continueCommit(final CohortEntry cohortEntry) { + final DataTreeCandidate candidate = cohortEntry.getCandidate(); // If we do not have any followers and we are not using persistence // or if cohortEntry has no modifications // we can apply modification to the state immediately if ((!hasFollowers() && !persistence().isRecoveryApplicable()) || isEmptyCommit(candidate)) { - applyModificationToState(getSender(), cohortEntry.getTransactionID(), candidate); + applyModificationToState(cohortEntry.getReplySender(), cohortEntry.getTransactionID(), candidate); } else { - Shard.this.persistData(getSender(), cohortEntry.getTransactionID(), + Shard.this.persistData(cohortEntry.getReplySender(), cohortEntry.getTransactionID(), DataTreeCandidatePayload.create(candidate)); } } @@ -331,10 +346,7 @@ public class Shard extends RaftActor { LOG.debug("{}: Finishing commit for transaction {}", persistenceId(), cohortEntry.getTransactionID()); try { - // We block on the future here so we don't have to worry about possibly accessing our - // state on a different thread outside of our dispatcher. Also, the data store - // currently uses a same thread executor anyway. - cohortEntry.getCohort().commit().get(); + cohortEntry.commit(); sender.tell(CommitTransactionReply.INSTANCE.toSerializable(), getSelf()); @@ -367,7 +379,7 @@ public class Shard extends RaftActor { cohortEntry = commitCoordinator.getAndRemoveCohortEntry(transactionID); if(cohortEntry != null) { try { - store.applyForeignCandidate(transactionID, cohortEntry.getCohort().getCandidate()); + store.applyForeignCandidate(transactionID, cohortEntry.getCandidate()); } catch (DataValidationFailedException e) { shardMBean.incrementFailedTransactionsCount(); LOG.error("{}: Failed to re-apply transaction {}", persistenceId(), transactionID, e); @@ -393,8 +405,24 @@ public class Shard extends RaftActor { commitCoordinator.handleCanCommit(canCommit.getTransactionID(), getSender(), this); } + private void noLeaderError(String errMessage, Object message) { + // TODO: rather than throwing an immediate exception, we could schedule a timer to try again to make + // it more resilient in case we're in the process of electing a new leader. + getSender().tell(new akka.actor.Status.Failure(new NoShardLeaderException(errMessage, persistenceId())), getSelf()); + } + + protected void handleBatchedModificationsLocal(BatchedModifications batched, ActorRef sender) { + try { + commitCoordinator.handleBatchedModifications(batched, sender, this); + } catch (Exception e) { + LOG.error("{}: Error handling BatchedModifications for Tx {}", persistenceId(), + batched.getTransactionID(), e); + sender.tell(new akka.actor.Status.Failure(e), getSelf()); + } + } + private void handleBatchedModifications(BatchedModifications batched) { - // This message is sent to prepare the modificationsa transaction directly on the Shard as an + // This message is sent to prepare the modifications transaction directly on the Shard as an // optimization to avoid the extra overhead of a separate ShardTransaction actor. On the last // BatchedModifications message, the caller sets the ready flag in the message indicating // modifications are complete. The reply contains the cohort actor path (this actor) for the caller @@ -407,13 +435,9 @@ public class Shard extends RaftActor { // window where it could have a stale leader during leadership transitions. // if(isLeader()) { - try { - commitCoordinator.handleBatchedModifications(batched, getSender(), this); - } catch (Exception e) { - LOG.error("{}: Error handling BatchedModifications for Tx {}", persistenceId(), - batched.getTransactionID(), e); - getSender().tell(new akka.actor.Status.Failure(e), getSelf()); - } + failIfIsolatedLeader(getSender()); + + handleBatchedModificationsLocal(batched, getSender()); } else { ActorSelection leader = getLeader(); if(leader != null) { @@ -423,53 +447,56 @@ public class Shard extends RaftActor { LOG.debug("{}: Forwarding BatchedModifications to leader {}", persistenceId(), leader); leader.forward(batched, getContext()); } else { - // TODO: rather than throwing an immediate exception, we could schedule a timer to try again to make - // it more resilient in case we're in the process of electing a new leader. - getSender().tell(new akka.actor.Status.Failure(new NoShardLeaderException(String.format( - "Could not find the leader for shard %s. This typically happens" + - " when the system is coming up or recovering and a leader is being elected. Try again" + - " later.", persistenceId()))), getSelf()); + noLeaderError("Could not commit transaction " + batched.getTransactionID(), batched); } } } - private void handleAbortTransaction(final AbortTransaction abort) { - doAbortTransaction(abort.getTransactionID(), getSender()); - } - - void doAbortTransaction(final String transactionID, final ActorRef sender) { - final CohortEntry cohortEntry = commitCoordinator.getCohortEntryIfCurrent(transactionID); - if(cohortEntry != null) { - LOG.debug("{}: Aborting transaction {}", persistenceId(), transactionID); + private boolean failIfIsolatedLeader(ActorRef sender) { + if(isIsolatedLeader()) { + sender.tell(new akka.actor.Status.Failure(new NoShardLeaderException(String.format( + "Shard %s was the leader but has lost contact with all of its followers. Either all" + + " other follower nodes are down or this node is isolated by a network partition.", + persistenceId()))), getSelf()); + return true; + } - // We don't remove the cached cohort entry here (ie pass false) in case the Tx was - // aborted during replication in which case we may still commit locally if replication - // succeeds. - commitCoordinator.currentTransactionComplete(transactionID, false); + return false; + } - final ListenableFuture future = cohortEntry.getCohort().abort(); - final ActorRef self = getSelf(); + protected boolean isIsolatedLeader() { + return getRaftState() == RaftState.IsolatedLeader; + } - Futures.addCallback(future, new FutureCallback() { - @Override - public void onSuccess(final Void v) { - shardMBean.incrementAbortTransactionsCount(); + private void handleReadyLocalTransaction(final ReadyLocalTransaction message) { + if (isLeader()) { + failIfIsolatedLeader(getSender()); - if(sender != null) { - sender.tell(AbortTransactionReply.INSTANCE.toSerializable(), self); - } - } + try { + commitCoordinator.handleReadyLocalTransaction(message, getSender(), this); + } catch (Exception e) { + LOG.error("{}: Error handling ReadyLocalTransaction for Tx {}", persistenceId(), + message.getTransactionID(), e); + getSender().tell(new akka.actor.Status.Failure(e), getSelf()); + } + } else { + ActorSelection leader = getLeader(); + if (leader != null) { + LOG.debug("{}: Forwarding ReadyLocalTransaction to leader {}", persistenceId(), leader); + message.setRemoteVersion(getCurrentBehavior().getLeaderPayloadVersion()); + leader.forward(message, getContext()); + } else { + noLeaderError("Could not commit transaction " + message.getTransactionID(), message); + } + } + } - @Override - public void onFailure(final Throwable t) { - LOG.error("{}: An exception happened during abort", persistenceId(), t); + private void handleAbortTransaction(final AbortTransaction abort) { + doAbortTransaction(abort.getTransactionID(), getSender()); + } - if(sender != null) { - sender.tell(new akka.actor.Status.Failure(t), self); - } - } - }); - } + void doAbortTransaction(final String transactionID, final ActorRef sender) { + commitCoordinator.handleAbort(transactionID, sender, this); } private void handleCreateTransaction(final Object message) { @@ -478,10 +505,8 @@ public class Shard extends RaftActor { } else if (getLeader() != null) { getLeader().forward(message, getContext()); } else { - getSender().tell(new akka.actor.Status.Failure(new NoShardLeaderException(String.format( - "Could not find leader for shard %s so transaction cannot be created. This typically happens" + - " when the system is coming up or recovering and a leader is being elected. Try again" + - " later.", persistenceId()))), getSelf()); + getSender().tell(new akka.actor.Status.Failure(new NoShardLeaderException( + "Could not create a shard transaction", persistenceId())), getSelf()); } } @@ -493,12 +518,17 @@ public class Shard extends RaftActor { ShardTransactionIdentifier transactionId, String transactionChainId, short clientVersion ) { - return transactionActorFactory.newShardTransaction(TransactionProxy.TransactionType.fromInt(transactionType), + return transactionActorFactory.newShardTransaction(TransactionType.fromInt(transactionType), transactionId, transactionChainId, clientVersion); } private void createTransaction(CreateTransaction createTransaction) { try { + if(TransactionType.fromInt(createTransaction.getTransactionType()) != TransactionType.READ_ONLY && + failIfIsolatedLeader(getSender())) { + return; + } + ActorRef transactionActor = createTransaction(createTransaction.getTransactionType(), createTransaction.getTransactionId(), createTransaction.getTransactionChainId(), createTransaction.getVersion()); @@ -554,14 +584,15 @@ public class Shard extends RaftActor { } @Override - protected RaftActorSnapshotCohort getRaftActorSnapshotCohort() { + @VisibleForTesting + public RaftActorSnapshotCohort getRaftActorSnapshotCohort() { return snapshotCohort; } @Override @Nonnull protected RaftActorRecoveryCohort getRaftActorRecoveryCohort() { - return new ShardRecoveryCoordinator(store, persistenceId(), LOG); + return new ShardRecoveryCoordinator(store, store.getSchemaContext(), persistenceId(), LOG); } @Override @@ -633,8 +664,9 @@ public class Shard extends RaftActor { @Override protected void onStateChanged() { boolean isLeader = isLeader(); - changeSupport.onLeadershipChange(isLeader); - treeChangeSupport.onLeadershipChange(isLeader); + boolean hasLeader = hasLeader(); + changeSupport.onLeadershipChange(isLeader, hasLeader); + treeChangeSupport.onLeadershipChange(isLeader, hasLeader); // If this actor is no longer the leader close all the transaction chains if (!isLeader) { @@ -663,22 +695,33 @@ public class Shard extends RaftActor { return commitCoordinator; } + public DatastoreContext getDatastoreContext() { + return datastoreContext; + } - private static class ShardCreator implements Creator { - + protected abstract static class AbstractShardCreator implements Creator { private static final long serialVersionUID = 1L; - final ShardIdentifier name; - final Map peerAddresses; - final DatastoreContext datastoreContext; - final SchemaContext schemaContext; + protected final ShardIdentifier name; + protected final Map peerAddresses; + protected final DatastoreContext datastoreContext; + protected final SchemaContext schemaContext; + + protected AbstractShardCreator(final ShardIdentifier name, final Map peerAddresses, + final DatastoreContext datastoreContext, final SchemaContext schemaContext) { + this.name = Preconditions.checkNotNull(name, "name should not be null"); + this.peerAddresses = Preconditions.checkNotNull(peerAddresses, "peerAddresses should not be null"); + this.datastoreContext = Preconditions.checkNotNull(datastoreContext, "dataStoreContext should not be null"); + this.schemaContext = Preconditions.checkNotNull(schemaContext, "schemaContext should not be null"); + } + } + + private static class ShardCreator extends AbstractShardCreator { + private static final long serialVersionUID = 1L; ShardCreator(final ShardIdentifier name, final Map peerAddresses, final DatastoreContext datastoreContext, final SchemaContext schemaContext) { - this.name = name; - this.peerAddresses = peerAddresses; - this.datastoreContext = datastoreContext; - this.schemaContext = schemaContext; + super(name, peerAddresses, datastoreContext, schemaContext); } @Override