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=5b8a50e18896c0a3e10db64c10b1d6fdffdc41d8;hp=ce3cae481e766f4ba4b7f93593e8d12a282ee294;hb=a51163ead1d60e66eeaf3691adb70b019ce60fb2;hpb=5c5c980e564d2b5f6cd26821ffd26997f59af260 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 ce3cae481e..5b8a50e188 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 @@ -12,17 +12,13 @@ import akka.actor.ActorRef; import akka.actor.ActorSelection; import akka.actor.Cancellable; import akka.actor.Props; -import akka.japi.Creator; import akka.persistence.RecoveryFailure; 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.Collections; import java.util.Map; import java.util.concurrent.TimeUnit; import javax.annotation.Nonnull; @@ -35,7 +31,6 @@ 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; @@ -44,7 +39,10 @@ import org.opendaylight.controller.cluster.datastore.messages.CommitTransaction; import org.opendaylight.controller.cluster.datastore.messages.CommitTransactionReply; import org.opendaylight.controller.cluster.datastore.messages.CreateTransaction; import org.opendaylight.controller.cluster.datastore.messages.CreateTransactionReply; +import org.opendaylight.controller.cluster.datastore.messages.DatastoreSnapshot; +import org.opendaylight.controller.cluster.datastore.messages.DatastoreSnapshot.ShardSnapshot; 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; @@ -65,12 +63,14 @@ 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.messages.ServerRemoved; import org.opendaylight.controller.cluster.raft.protobuff.client.messages.CompositeModificationByteStringPayload; import org.opendaylight.controller.cluster.raft.protobuff.client.messages.CompositeModificationPayload; import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTree; import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTreeCandidate; import org.opendaylight.yangtools.yang.data.api.schema.tree.DataValidationFailedException; import org.opendaylight.yangtools.yang.data.api.schema.tree.ModificationType; +import org.opendaylight.yangtools.yang.data.api.schema.tree.TreeType; import org.opendaylight.yangtools.yang.model.api.SchemaContext; import scala.concurrent.duration.Duration; import scala.concurrent.duration.FiniteDuration; @@ -83,7 +83,7 @@ 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"; @@ -118,19 +118,24 @@ public class Shard extends RaftActor { private final DataTreeChangeListenerSupport treeChangeSupport = new DataTreeChangeListenerSupport(this); private final DataChangeListenerSupport changeSupport = new DataChangeListenerSupport(this); - protected Shard(final ShardIdentifier name, final Map peerAddresses, - final DatastoreContext datastoreContext, final SchemaContext schemaContext) { - super(name.toString(), new HashMap<>(peerAddresses), Optional.of(datastoreContext.getShardRaftConfig()), - DataStoreVersions.CURRENT_VERSION); - this.name = name.toString(); - this.datastoreContext = datastoreContext; + private ShardSnapshot restoreFromSnapshot; + + private final ShardTransactionMessageRetrySupport messageRetrySupport; + + protected Shard(AbstractBuilder builder) { + super(builder.getId().toString(), builder.getPeerAddresses(), + Optional.of(builder.getDatastoreContext().getShardRaftConfig()), DataStoreVersions.CURRENT_VERSION); + + this.name = builder.getId().toString(); + this.datastoreContext = builder.getDatastoreContext(); + this.restoreFromSnapshot = builder.getRestoreFromSnapshot(); setPersistence(datastoreContext.isPersistent()); LOG.info("Shard created : {}, persistent : {}", name, datastoreContext.isPersistent()); - store = new ShardDataTree(schemaContext); + store = new ShardDataTree(builder.getSchemaContext(), builder.getTreeType()); shardMBean = ShardMBeanFactory.getShardStatsMBean(name.toString(), datastoreContext.getDataStoreMXBeanType()); @@ -142,7 +147,7 @@ public class Shard extends RaftActor { commitCoordinator = new ShardCommitCoordinator(store, datastoreContext.getShardCommitQueueExpiryTimeoutInMillis(), - datastoreContext.getShardTransactionCommitQueueCapacity(), self(), LOG, this.name); + datastoreContext.getShardTransactionCommitQueueCapacity(), LOG, this.name); setTransactionCommitTimeout(); @@ -158,7 +163,7 @@ public class Shard extends RaftActor { snapshotCohort = new ShardSnapshotCohort(transactionActorFactory, store, LOG, this.name); - + messageRetrySupport = new ShardTransactionMessageRetrySupport(this); } private void setTransactionCommitTimeout() { @@ -166,17 +171,6 @@ public class Shard extends RaftActor { 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"); - - return Props.create(new ShardCreator(name, peerAddresses, datastoreContext, schemaContext)); - } - private Optional createRoleChangeNotifier(String shardId) { ActorRef shardRoleChangeNotifier = this.getContext().actorOf( RoleChangeNotifier.getProps(shardId), shardId + "-notifier"); @@ -189,6 +183,8 @@ public class Shard extends RaftActor { super.postStop(); + messageRetrySupport.close(); + if(txCommitTimeoutCheckSchedule != null) { txCommitTimeoutCheckSchedule.cancel(); } @@ -225,7 +221,7 @@ 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 { @@ -234,8 +230,7 @@ public class Shard extends RaftActor { } else if (BatchedModifications.class.isInstance(message)) { handleBatchedModifications((BatchedModifications)message); } else if (message instanceof ForwardedReadyTransaction) { - commitCoordinator.handleForwardedReadyTransaction((ForwardedReadyTransaction) message, - getSender(), this); + handleForwardedReadyTransaction((ForwardedReadyTransaction) message); } else if (message instanceof ReadyLocalTransaction) { handleReadyLocalTransaction((ReadyLocalTransaction)message); } else if (CanCommitTransaction.SERIALIZABLE_CLASS.isInstance(message)) { @@ -247,9 +242,9 @@ 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) { @@ -267,6 +262,12 @@ public class Shard extends RaftActor { 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 if(message instanceof ServerRemoved){ + context().parent().forward(message, context()); + } else if(ShardTransactionMessageRetrySupport.TIMER_MESSAGE_CLASS.isInstance(message)) { + messageRetrySupport.onTimerMessage(message); } else { super.onReceiveCommand(message); } @@ -275,6 +276,10 @@ public class Shard extends RaftActor { } } + private boolean hasLeader() { + return getLeaderId() != null; + } + public int getPendingTxCommitQueueSize() { return commitCoordinator.getQueueSize(); } @@ -291,7 +296,7 @@ public class Shard extends RaftActor { leaderPayloadVersion); } - private void onDatastoreContext(DatastoreContext context) { + protected void onDatastoreContext(DatastoreContext context) { datastoreContext = context; commitCoordinator.setQueueCapacity(datastoreContext.getShardTransactionCommitQueueCapacity()); @@ -325,8 +330,8 @@ public class Shard extends RaftActor { 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 @@ -335,7 +340,7 @@ public class Shard extends RaftActor { applyModificationToState(cohortEntry.getReplySender(), cohortEntry.getTransactionID(), candidate); } else { Shard.this.persistData(cohortEntry.getReplySender(), cohortEntry.getTransactionID(), - DataTreeCandidatePayload.create(candidate)); + DataTreeCandidatePayload.create(candidate)); } } @@ -349,10 +354,7 @@ 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()); @@ -385,7 +387,7 @@ 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); @@ -411,10 +413,14 @@ 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) { @@ -433,13 +439,7 @@ public class Shard extends RaftActor { if(isLeader()) { failIfIsolatedLeader(getSender()); - 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()); - } + handleBatchedModificationsLocal(batched, getSender()); } else { ActorSelection leader = getLeader(); if(leader != null) { @@ -449,13 +449,14 @@ public class Shard extends RaftActor { LOG.debug("{}: Forwarding BatchedModifications to leader {}", persistenceId(), leader); leader.forward(batched, getContext()); } else { - noLeaderError("Could not commit transaction " + batched.getTransactionID(), batched); + messageRetrySupport.addMessageToRetry(batched, getSender(), + "Could not commit transaction " + batched.getTransactionID()); } } } private boolean failIfIsolatedLeader(ActorRef sender) { - if(getRaftState() == RaftState.IsolatedLeader) { + 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.", @@ -466,6 +467,10 @@ public class Shard extends RaftActor { return false; } + protected boolean isIsolatedLeader() { + return getRaftState() == RaftState.IsolatedLeader; + } + private void handleReadyLocalTransaction(final ReadyLocalTransaction message) { if (isLeader()) { failIfIsolatedLeader(getSender()); @@ -484,7 +489,29 @@ public class Shard extends RaftActor { message.setRemoteVersion(getCurrentBehavior().getLeaderPayloadVersion()); leader.forward(message, getContext()); } else { - noLeaderError("Could not commit transaction " + message.getTransactionID(), message); + messageRetrySupport.addMessageToRetry(message, getSender(), + "Could not commit transaction " + message.getTransactionID()); + } + } + } + + private void handleForwardedReadyTransaction(ForwardedReadyTransaction forwardedReady) { + if (isLeader()) { + failIfIsolatedLeader(getSender()); + + commitCoordinator.handleForwardedReadyTransaction(forwardedReady, getSender(), this); + } else { + ActorSelection leader = getLeader(); + if (leader != null) { + LOG.debug("{}: Forwarding ForwardedReadyTransaction to leader {}", persistenceId(), leader); + + ReadyLocalTransaction readyLocal = new ReadyLocalTransaction(forwardedReady.getTransactionID(), + forwardedReady.getTransaction().getSnapshot(), forwardedReady.isDoImmediateCommit()); + readyLocal.setRemoteVersion(getCurrentBehavior().getLeaderPayloadVersion()); + leader.forward(readyLocal, getContext()); + } else { + messageRetrySupport.addMessageToRetry(forwardedReady, getSender(), + "Could not commit transaction " + forwardedReady.getTransactionID()); } } } @@ -494,38 +521,7 @@ public class Shard extends RaftActor { } void doAbortTransaction(final String transactionID, final ActorRef sender) { - final CohortEntry cohortEntry = commitCoordinator.getCohortEntryIfCurrent(transactionID); - if(cohortEntry != null) { - LOG.debug("{}: Aborting transaction {}", persistenceId(), transactionID); - - // 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); - - final ListenableFuture future = cohortEntry.getCohort().abort(); - final ActorRef self = getSelf(); - - Futures.addCallback(future, new FutureCallback() { - @Override - public void onSuccess(final Void v) { - shardMBean.incrementAbortTransactionsCount(); - - if(sender != null) { - sender.tell(AbortTransactionReply.INSTANCE.toSerializable(), self); - } - } - - @Override - public void onFailure(final Throwable t) { - LOG.error("{}: An exception happened during abort", persistenceId(), t); - - if(sender != null) { - sender.tell(new akka.actor.Status.Failure(t), self); - } - } - }); - } + commitCoordinator.handleAbort(transactionID, sender, this); } private void handleCreateTransaction(final Object message) { @@ -621,11 +617,14 @@ public class Shard extends RaftActor { @Override @Nonnull protected RaftActorRecoveryCohort getRaftActorRecoveryCohort() { - return new ShardRecoveryCoordinator(store, store.getSchemaContext(), persistenceId(), LOG); + return new ShardRecoveryCoordinator(store, store.getSchemaContext(), + restoreFromSnapshot != null ? restoreFromSnapshot.getSnapshot() : null, persistenceId(), LOG); } @Override protected void onRecoveryComplete() { + restoreFromSnapshot = null; + //notify shard manager getContext().parent().tell(new ActorInitialized(), getSelf()); @@ -693,8 +692,9 @@ 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) { @@ -711,6 +711,10 @@ public class Shard extends RaftActor { @Override protected void onLeaderChanged(String oldLeader, String newLeader) { shardMBean.incrementLeadershipChangeCount(); + + if(hasLeader()) { + messageRetrySupport.retryMessages(); + } } @Override @@ -723,37 +727,124 @@ public class Shard extends RaftActor { return commitCoordinator; } + public DatastoreContext getDatastoreContext() { + return datastoreContext; + } - private static class ShardCreator implements Creator { + @VisibleForTesting + public ShardDataTree getDataStore() { + return store; + } + + @VisibleForTesting + ShardStats getShardMBean() { + return shardMBean; + } + + public static Builder builder() { + return new Builder(); + } - private static final long serialVersionUID = 1L; + public static abstract class AbstractBuilder, S extends Shard> { + private final Class shardClass; + private ShardIdentifier id; + private Map peerAddresses = Collections.emptyMap(); + private DatastoreContext datastoreContext; + private SchemaContext schemaContext; + private DatastoreSnapshot.ShardSnapshot restoreFromSnapshot; + private volatile boolean sealed; - final ShardIdentifier name; - final Map peerAddresses; - final DatastoreContext datastoreContext; - final SchemaContext schemaContext; + protected AbstractBuilder(Class shardClass) { + this.shardClass = shardClass; + } - ShardCreator(final ShardIdentifier name, final Map peerAddresses, - final DatastoreContext datastoreContext, final SchemaContext schemaContext) { - this.name = name; + protected void checkSealed() { + Preconditions.checkState(!sealed, "Builder isalready sealed - further modifications are not allowed"); + } + + @SuppressWarnings("unchecked") + private T self() { + return (T) this; + } + + public T id(ShardIdentifier id) { + checkSealed(); + this.id = id; + return self(); + } + + public T peerAddresses(Map peerAddresses) { + checkSealed(); this.peerAddresses = peerAddresses; + return self(); + } + + public T datastoreContext(DatastoreContext datastoreContext) { + checkSealed(); this.datastoreContext = datastoreContext; + return self(); + } + + public T schemaContext(SchemaContext schemaContext) { + checkSealed(); this.schemaContext = schemaContext; + return self(); } - @Override - public Shard create() throws Exception { - return new Shard(name, peerAddresses, datastoreContext, schemaContext); + public T restoreFromSnapshot(DatastoreSnapshot.ShardSnapshot restoreFromSnapshot) { + checkSealed(); + this.restoreFromSnapshot = restoreFromSnapshot; + return self(); } - } - @VisibleForTesting - public ShardDataTree getDataStore() { - return store; + public ShardIdentifier getId() { + return id; + } + + public Map getPeerAddresses() { + return peerAddresses; + } + + public DatastoreContext getDatastoreContext() { + return datastoreContext; + } + + public SchemaContext getSchemaContext() { + return schemaContext; + } + + public DatastoreSnapshot.ShardSnapshot getRestoreFromSnapshot() { + return restoreFromSnapshot; + } + + public TreeType getTreeType() { + switch (datastoreContext.getLogicalStoreType()) { + case CONFIGURATION: + return TreeType.CONFIGURATION; + case OPERATIONAL: + return TreeType.OPERATIONAL; + } + + throw new IllegalStateException("Unhandled logical store type " + datastoreContext.getLogicalStoreType()); + } + + protected void verify() { + Preconditions.checkNotNull(id, "id 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 Props props() { + sealed = true; + verify(); + return Props.create(shardClass, this); + } } - @VisibleForTesting - ShardStats getShardMBean() { - return shardMBean; + public static class Builder extends AbstractBuilder { + private Builder() { + super(Shard.class); + } } }