X-Git-Url: https://git.opendaylight.org/gerrit/gitweb?a=blobdiff_plain;f=opendaylight%2Fmd-sal%2Fsal-distributed-datastore%2Fsrc%2Fmain%2Fjava%2Forg%2Fopendaylight%2Fcontroller%2Fcluster%2Fdatastore%2FShardManager.java;h=724c8d2c03dc787549f511f60ac3e289ea763978;hb=9b013985b871adb08173dd8e90d5cf2af82fa5a1;hp=eb39a34dc0dc508b3543e5d3f37976c749672bfe;hpb=b34452ce75563e360ae1d02a9f2aa6223d6208c3;p=controller.git diff --git a/opendaylight/md-sal/sal-distributed-datastore/src/main/java/org/opendaylight/controller/cluster/datastore/ShardManager.java b/opendaylight/md-sal/sal-distributed-datastore/src/main/java/org/opendaylight/controller/cluster/datastore/ShardManager.java index eb39a34dc0..724c8d2c03 100644 --- a/opendaylight/md-sal/sal-distributed-datastore/src/main/java/org/opendaylight/controller/cluster/datastore/ShardManager.java +++ b/opendaylight/md-sal/sal-distributed-datastore/src/main/java/org/opendaylight/controller/cluster/datastore/ShardManager.java @@ -8,18 +8,22 @@ package org.opendaylight.controller.cluster.datastore; +import static akka.pattern.Patterns.ask; import akka.actor.ActorPath; import akka.actor.ActorRef; import akka.actor.Address; import akka.actor.Cancellable; import akka.actor.OneForOneStrategy; +import akka.actor.PoisonPill; import akka.actor.Props; import akka.actor.SupervisorStrategy; import akka.cluster.ClusterEvent; +import akka.dispatch.OnComplete; import akka.japi.Creator; import akka.japi.Function; import akka.persistence.RecoveryCompleted; import akka.serialization.Serialization; +import akka.util.Timeout; import com.google.common.annotations.VisibleForTesting; import com.google.common.base.Objects; import com.google.common.base.Optional; @@ -47,6 +51,7 @@ import org.opendaylight.controller.cluster.datastore.identifiers.ShardIdentifier import org.opendaylight.controller.cluster.datastore.jmx.mbeans.shardmanager.ShardManagerInfo; import org.opendaylight.controller.cluster.datastore.jmx.mbeans.shardmanager.ShardManagerInfoMBean; import org.opendaylight.controller.cluster.datastore.messages.ActorInitialized; +import org.opendaylight.controller.cluster.datastore.messages.AddShardReplica; import org.opendaylight.controller.cluster.datastore.messages.CreateShard; import org.opendaylight.controller.cluster.datastore.messages.CreateShardReply; import org.opendaylight.controller.cluster.datastore.messages.FindLocalShard; @@ -59,11 +64,10 @@ import org.opendaylight.controller.cluster.datastore.messages.PeerDown; import org.opendaylight.controller.cluster.datastore.messages.PeerUp; import org.opendaylight.controller.cluster.datastore.messages.RemoteFindPrimary; import org.opendaylight.controller.cluster.datastore.messages.RemotePrimaryShardFound; +import org.opendaylight.controller.cluster.datastore.messages.RemoveShardReplica; import org.opendaylight.controller.cluster.datastore.messages.ShardLeaderStateChanged; import org.opendaylight.controller.cluster.datastore.messages.SwitchShardBehavior; import org.opendaylight.controller.cluster.datastore.messages.UpdateSchemaContext; -import org.opendaylight.controller.cluster.datastore.messages.AddShardReplica; -import org.opendaylight.controller.cluster.datastore.messages.RemoveShardReplica; import org.opendaylight.controller.cluster.datastore.utils.Dispatchers; import org.opendaylight.controller.cluster.datastore.utils.PrimaryShardInfoFutureCache; import org.opendaylight.controller.cluster.notifications.RegisterRoleChangeListener; @@ -71,10 +75,15 @@ import org.opendaylight.controller.cluster.notifications.RoleChangeNotification; import org.opendaylight.controller.cluster.raft.RaftState; import org.opendaylight.controller.cluster.raft.base.messages.FollowerInitialSyncUpStatus; import org.opendaylight.controller.cluster.raft.base.messages.SwitchBehavior; +import org.opendaylight.controller.cluster.raft.messages.AddServer; +import org.opendaylight.controller.cluster.raft.messages.AddServerReply; +import org.opendaylight.controller.cluster.raft.messages.ServerChangeStatus; +import org.opendaylight.controller.cluster.raft.policy.DisableElectionsRaftPolicy; import org.opendaylight.yangtools.yang.data.api.schema.tree.DataTree; import org.opendaylight.yangtools.yang.model.api.SchemaContext; import org.slf4j.Logger; import org.slf4j.LoggerFactory; +import scala.concurrent.Future; import scala.concurrent.duration.Duration; import scala.concurrent.duration.FiniteDuration; @@ -465,11 +474,11 @@ public class ShardManager extends AbstractUntypedPersistentActorWithMetering { getSender().tell(messageSupplier.get(), getSelf()); } - private NoShardLeaderException createNoShardLeaderException(ShardIdentifier shardId) { + private static NoShardLeaderException createNoShardLeaderException(ShardIdentifier shardId) { return new NoShardLeaderException(null, shardId.toString()); } - private NotInitializedException createNotInitializedException(ShardIdentifier shardId) { + private static NotInitializedException createNotInitializedException(ShardIdentifier shardId) { return new NotInitializedException(String.format( "Found primary shard %s but it's not initialized yet. Please try again later", shardId)); } @@ -746,41 +755,174 @@ public class ShardManager extends AbstractUntypedPersistentActorWithMetering { return mBean; } - private void onAddShardReplica (AddShardReplica shardReplicaMsg) { - String shardName = shardReplicaMsg.getShardName(); + private DatastoreContext getInitShardDataStoreContext() { + return (DatastoreContext.newBuilderFrom(datastoreContext) + .customRaftPolicyImplementation(DisableElectionsRaftPolicy.class.getName()) + .build()); + } - // verify the local shard replica is already available in the controller node + private void checkLocalShardExists(final String shardName, final ActorRef sender) { if (localShards.containsKey(shardName)) { - LOG.debug ("Local shard {} already available in the controller node", shardName); - getSender().tell(new akka.actor.Status.Failure( - new IllegalArgumentException(String.format("Local shard %s already exists", - shardName))), getSelf()); - return; + String msg = String.format("Local shard %s already exists", shardName); + LOG.debug ("{}: {}", persistenceId(), msg); + sender.tell(new akka.actor.Status.Failure(new IllegalArgumentException(msg)), getSelf()); } + } + + private void onAddShardReplica (AddShardReplica shardReplicaMsg) { + final String shardName = shardReplicaMsg.getShardName(); + + // verify the local shard replica is already available in the controller node + LOG.debug ("onAddShardReplica: {}", shardReplicaMsg); + + checkLocalShardExists(shardName, getSender()); + // verify the shard with the specified name is present in the cluster configuration if (!(this.configuration.isShardConfigured(shardName))) { - LOG.debug ("No module configuration exists for shard {}", shardName); - getSender().tell(new akka.actor.Status.Failure(new IllegalArgumentException( - String.format("No module configuration exists for shard %s", - shardName))), getSelf()); + String msg = String.format("No module configuration exists for shard %s", shardName); + LOG.debug ("{}: {}", persistenceId(), msg); + getSender().tell(new akka.actor.Status.Failure(new IllegalArgumentException(msg)), getSelf()); return; } // Create the localShard - getSender().tell(new akka.actor.Status.Success(true), getSelf()); + if (schemaContext == null) { + String msg = String.format( + "No SchemaContext is available in order to create a local shard instance for %s", shardName); + LOG.debug ("{}: {}", persistenceId(), msg); + getSender().tell(new akka.actor.Status.Failure(new IllegalStateException(msg)), getSelf()); + return; + } + + Map peerAddresses = getPeerAddresses(shardName); + if (peerAddresses.isEmpty()) { + String msg = String.format("Cannot add replica for shard %s because no peer is available", shardName); + LOG.debug ("{}: {}", persistenceId(), msg); + getSender().tell(new akka.actor.Status.Failure(new IllegalStateException(msg)), getSelf()); + return; + } + + Timeout findPrimaryTimeout = new Timeout(datastoreContext.getShardInitializationTimeout().duration().$times(2)); + + final ActorRef sender = getSender(); + Future futureObj = ask(getSelf(), new RemoteFindPrimary(shardName, true), findPrimaryTimeout); + futureObj.onComplete(new OnComplete() { + @Override + public void onComplete(Throwable failure, Object response) { + if (failure != null) { + LOG.debug ("{}: Received failure from FindPrimary for shard {}", persistenceId(), shardName, failure); + sender.tell(new akka.actor.Status.Failure(new RuntimeException( + String.format("Failed to find leader for shard %s", shardName), failure)), + getSelf()); + } else { + if (!(response instanceof RemotePrimaryShardFound)) { + String msg = String.format("Failed to find leader for shard %s: received response: %s", + shardName, response); + LOG.debug ("{}: {}", persistenceId(), msg); + sender.tell(new akka.actor.Status.Failure(new RuntimeException(msg)), getSelf()); + return; + } + + RemotePrimaryShardFound message = (RemotePrimaryShardFound)response; + addShard (shardName, message, sender); + } + } + }, new Dispatchers(context().system().dispatchers()).getDispatcher(Dispatchers.DispatcherType.Client)); + } + + private void addShard(final String shardName, final RemotePrimaryShardFound response, final ActorRef sender) { + checkLocalShardExists(shardName, sender); + + ShardIdentifier shardId = getShardIdentifier(cluster.getCurrentMemberName(), shardName); + String localShardAddress = peerAddressResolver.getShardActorAddress(shardName, cluster.getCurrentMemberName()); + final ShardInformation shardInfo = new ShardInformation(shardName, shardId, + getPeerAddresses(shardName), getInitShardDataStoreContext(), + new DefaultShardPropsCreator(), peerAddressResolver); + localShards.put(shardName, shardInfo); + shardInfo.setActor(newShardActor(schemaContext, shardInfo)); + + //inform ShardLeader to add this shard as a replica by sending an AddServer message + LOG.debug ("{}: Sending AddServer message to peer {} for shard {}", persistenceId(), + response.getPrimaryPath(), shardId); + + Timeout addServerTimeout = new Timeout(datastoreContext + .getShardLeaderElectionTimeout().duration().$times(4)); + Future futureObj = ask(getContext().actorSelection(response.getPrimaryPath()), + new AddServer(shardId.toString(), localShardAddress, true), addServerTimeout); + + futureObj.onComplete(new OnComplete() { + @Override + public void onComplete(Throwable failure, Object addServerResponse) { + if (failure != null) { + LOG.debug ("{}: AddServer request to {} for {} failed", persistenceId(), + response.getPrimaryPath(), shardName, failure); + + // Remove the shard + localShards.remove(shardName); + if (shardInfo.getActor() != null) { + shardInfo.getActor().tell(PoisonPill.getInstance(), getSelf()); + } + + sender.tell(new akka.actor.Status.Failure(new RuntimeException( + String.format("AddServer request to leader %s for shard %s failed", + response.getPrimaryPath(), shardName), failure)), getSelf()); + } else { + AddServerReply reply = (AddServerReply)addServerResponse; + onAddServerReply(shardName, shardInfo, reply, sender, response.getPrimaryPath()); + } + } + }, new Dispatchers(context().system().dispatchers()). + getDispatcher(Dispatchers.DispatcherType.Client)); return; } + private void onAddServerReply (String shardName, ShardInformation shardInfo, + AddServerReply replyMsg, ActorRef sender, String leaderPath) { + LOG.debug ("{}: Received {} for shard {} from leader {}", persistenceId(), replyMsg, shardName, leaderPath); + + if (replyMsg.getStatus() == ServerChangeStatus.OK) { + LOG.debug ("{}: Leader shard successfully added the replica shard {}", persistenceId(), shardName); + + // Make the local shard voting capable + shardInfo.setDatastoreContext(datastoreContext, getSelf()); + + mBean.addLocalShard(shardInfo.getShardId().toString()); + sender.tell(new akka.actor.Status.Success(true), getSelf()); + } else { + LOG.warn ("{}: Leader failed to add shard replica {} with status {} - removing the local shard", + persistenceId(), shardName, replyMsg.getStatus()); + + //remove the local replica created + localShards.remove(shardName); + if (shardInfo.getActor() != null) { + shardInfo.getActor().tell(PoisonPill.getInstance(), getSelf()); + } + switch (replyMsg.getStatus()) { + case TIMEOUT: + sender.tell(new akka.actor.Status.Failure(new RuntimeException( + String.format("The shard leader %s timed out trying to replicate the initial data to the new shard %s. Possible causes - there was a problem replicating the data or shard leadership changed while replicating the shard data", + leaderPath, shardName))), getSelf()); + break; + case NO_LEADER: + sender.tell(new akka.actor.Status.Failure(new RuntimeException(String.format( + "There is no shard leader available for shard %s", shardName))), getSelf()); + break; + default : + sender.tell(new akka.actor.Status.Failure(new RuntimeException(String.format( + "AddServer request to leader %s for shard %s failed with status %s", + leaderPath, shardName, replyMsg.getStatus()))), getSelf()); + } + } + } + private void onRemoveShardReplica (RemoveShardReplica shardReplicaMsg) { String shardName = shardReplicaMsg.getShardName(); - boolean deleteStatus = false; // verify the local shard replica is available in the controller node if (!localShards.containsKey(shardName)) { - LOG.debug ("Local shard replica {} is not available in the controller node", shardName); - getSender().tell(new akka.actor.Status.Failure( - new IllegalArgumentException(String.format("Local shard %s not available", - shardName))), getSelf()); + String msg = String.format("Local shard %s does not", shardName); + LOG.debug ("{}: {}", persistenceId(), msg); + getSender().tell(new akka.actor.Status.Failure(new IllegalArgumentException(msg)), getSelf()); return; } // call RemoveShard for the shardName @@ -808,7 +950,7 @@ public class ShardManager extends AbstractUntypedPersistentActorWithMetering { private String leaderId; private short leaderVersion; - private final DatastoreContext datastoreContext; + private DatastoreContext datastoreContext; private final ShardPropsCreator shardPropsCreator; private final ShardPeerAddressResolver addressResolver; @@ -993,6 +1135,16 @@ public class ShardManager extends AbstractUntypedPersistentActorWithMetering { void setLeaderVersion(short leaderVersion) { this.leaderVersion = leaderVersion; } + + void setDatastoreContext(DatastoreContext datastoreContext, ActorRef sender) { + this.datastoreContext = datastoreContext; + //notify the datastoreContextchange + LOG.debug ("Notifying RaftPolicy change via datastoreContextChange for {}", + this.shardName); + if (actor != null) { + actor.tell(this.datastoreContext, sender); + } + } } private static class ShardManagerCreator implements Creator {