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=7ca79fc2349add0e9c19236a945aaed53676b45d;hp=225194466046583c72c8efd8b3a963418eff1266;hb=43fb391bf873b252383a8d736b2651b04da8d40d;hpb=13ad69bc47bd5fa3f1df973861a5250ded0f59a7
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 2251944660..7ca79fc234 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
@@ -12,24 +12,46 @@ import akka.actor.ActorRef;
import akka.actor.ActorSelection;
import akka.actor.Cancellable;
import akka.actor.Props;
+import akka.actor.Status.Failure;
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.base.Ticker;
+import com.google.common.collect.ImmutableList;
+import com.google.common.collect.Range;
import java.io.IOException;
+import java.util.Arrays;
import java.util.Collection;
import java.util.Collections;
+import java.util.HashMap;
import java.util.Map;
-import java.util.concurrent.ExecutionException;
import java.util.concurrent.TimeUnit;
import javax.annotation.Nonnull;
+import javax.annotation.Nullable;
+import org.opendaylight.controller.cluster.access.ABIVersion;
+import org.opendaylight.controller.cluster.access.commands.ConnectClientRequest;
+import org.opendaylight.controller.cluster.access.commands.ConnectClientSuccess;
+import org.opendaylight.controller.cluster.access.commands.LocalHistoryRequest;
+import org.opendaylight.controller.cluster.access.commands.NotLeaderException;
+import org.opendaylight.controller.cluster.access.commands.TransactionRequest;
+import org.opendaylight.controller.cluster.access.concepts.ClientIdentifier;
+import org.opendaylight.controller.cluster.access.concepts.FrontendIdentifier;
+import org.opendaylight.controller.cluster.access.concepts.LocalHistoryIdentifier;
+import org.opendaylight.controller.cluster.access.concepts.Request;
+import org.opendaylight.controller.cluster.access.concepts.RequestEnvelope;
+import org.opendaylight.controller.cluster.access.concepts.RequestException;
+import org.opendaylight.controller.cluster.access.concepts.RequestSuccess;
+import org.opendaylight.controller.cluster.access.concepts.RetiredGenerationException;
+import org.opendaylight.controller.cluster.access.concepts.RuntimeRequestException;
+import org.opendaylight.controller.cluster.access.concepts.TransactionIdentifier;
+import org.opendaylight.controller.cluster.access.concepts.UnsupportedRequestException;
import org.opendaylight.controller.cluster.common.actor.CommonConfig;
import org.opendaylight.controller.cluster.common.actor.MessageTracker;
import org.opendaylight.controller.cluster.common.actor.MessageTracker.Error;
import org.opendaylight.controller.cluster.common.actor.MeteringBehavior;
import org.opendaylight.controller.cluster.datastore.exceptions.NoShardLeaderException;
import org.opendaylight.controller.cluster.datastore.identifiers.ShardIdentifier;
-import org.opendaylight.controller.cluster.datastore.identifiers.ShardTransactionIdentifier;
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;
@@ -38,21 +60,19 @@ import org.opendaylight.controller.cluster.datastore.messages.BatchedModificatio
import org.opendaylight.controller.cluster.datastore.messages.CanCommitTransaction;
import org.opendaylight.controller.cluster.datastore.messages.CloseTransactionChain;
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.OnDemandShardState;
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;
import org.opendaylight.controller.cluster.datastore.messages.UpdateSchemaContext;
-import org.opendaylight.controller.cluster.datastore.modification.Modification;
-import org.opendaylight.controller.cluster.datastore.modification.MutableCompositeModification;
+import org.opendaylight.controller.cluster.datastore.persisted.DatastoreSnapshot;
+import org.opendaylight.controller.cluster.datastore.persisted.DatastoreSnapshot.ShardSnapshot;
import org.opendaylight.controller.cluster.datastore.utils.Dispatchers;
import org.opendaylight.controller.cluster.notifications.LeaderStateChanged;
import org.opendaylight.controller.cluster.notifications.RegisterRoleChangeListener;
@@ -62,25 +82,23 @@ 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.client.messages.OnDemandRaftState;
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.controller.cluster.raft.protobuff.client.messages.Payload;
import org.opendaylight.yangtools.concepts.Identifier;
-import org.opendaylight.yangtools.util.StringIdentifier;
-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.TipProducingDataTree;
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;
/**
- * A Shard represents a portion of the logical data tree
+ * A Shard represents a portion of the logical data tree.
+ *
*
* Our Shard uses InMemoryDataTree as it's internal representation and delegates all requests it
- *
*/
public class Shard extends RaftActor {
@@ -103,6 +121,17 @@ public class Shard extends RaftActor {
// FIXME: shard names should be encapsulated in their own class and this should be exposed as a constant.
public static final String DEFAULT_NAME = "default";
+ private static final Collection SUPPORTED_ABIVERSIONS;
+
+ static {
+ final ABIVersion[] values = ABIVersion.values();
+ final ABIVersion[] real = Arrays.copyOfRange(values, 1, values.length - 1);
+ SUPPORTED_ABIVERSIONS = ImmutableList.copyOf(real).reverse();
+ }
+
+ // FIXME: make this a dynamic property based on mailbox size and maximum number of clients
+ private static final int CLIENT_MAX_MESSAGES = 1000;
+
// The state of this Shard
private final ShardDataTree store;
@@ -135,7 +164,10 @@ public class Shard extends RaftActor {
private final ShardTransactionMessageRetrySupport messageRetrySupport;
- protected Shard(AbstractBuilder builder) {
+ private final FrontendMetadata frontendMetadata = new FrontendMetadata();
+ private final Map knownFrontends = new HashMap<>();
+
+ protected Shard(final AbstractBuilder builder) {
super(builder.getId().toString(), builder.getPeerAddresses(),
Optional.of(builder.getDatastoreContext().getShardRaftConfig()), DataStoreVersions.CURRENT_VERSION);
@@ -147,35 +179,41 @@ public class Shard extends RaftActor {
LOG.info("Shard created : {}, persistent : {}", name, datastoreContext.isPersistent());
- store = new ShardDataTree(builder.getSchemaContext(), builder.getTreeType(),
- new ShardDataTreeChangeListenerPublisherActorProxy(getContext(), name + "-DTCL-publisher"),
- new ShardDataChangeListenerPublisherActorProxy(getContext(), name + "-DCL-publisher"), name);
+ ShardDataTreeChangeListenerPublisherActorProxy treeChangeListenerPublisher =
+ new ShardDataTreeChangeListenerPublisherActorProxy(getContext(), name + "-DTCL-publisher");
+ ShardDataChangeListenerPublisherActorProxy dataChangeListenerPublisher =
+ new ShardDataChangeListenerPublisherActorProxy(getContext(), name + "-DCL-publisher");
+ if (builder.getDataTree() != null) {
+ store = new ShardDataTree(this, builder.getSchemaContext(), builder.getDataTree(),
+ treeChangeListenerPublisher, dataChangeListenerPublisher, name);
+ } else {
+ store = new ShardDataTree(this, builder.getSchemaContext(), builder.getTreeType(),
+ builder.getDatastoreContext().getStoreRoot(), treeChangeListenerPublisher,
+ dataChangeListenerPublisher, name);
+ }
- shardMBean = ShardMBeanFactory.getShardStatsMBean(name.toString(),
- datastoreContext.getDataStoreMXBeanType());
- shardMBean.setShard(this);
+ shardMBean = ShardMBeanFactory.getShardStatsMBean(name, datastoreContext.getDataStoreMXBeanType(), this);
if (isMetricsCaptureEnabled()) {
getContext().become(new MeteringBehavior(this));
}
- commitCoordinator = new ShardCommitCoordinator(store,
- datastoreContext.getShardCommitQueueExpiryTimeoutInMillis(),
- datastoreContext.getShardTransactionCommitQueueCapacity(), LOG, this.name);
+ commitCoordinator = new ShardCommitCoordinator(store, LOG, this.name);
setTransactionCommitTimeout();
// create a notifier actor for each cluster member
- roleChangeNotifier = createRoleChangeNotifier(name.toString());
+ roleChangeNotifier = createRoleChangeNotifier(name);
appendEntriesReplyTracker = new MessageTracker(AppendEntriesReply.class,
getRaftActorContext().getConfigParams().getIsolatedCheckIntervalInMillis());
transactionActorFactory = new ShardTransactionActorFactory(store, datastoreContext,
- new Dispatchers(context().system().dispatchers()).getDispatcherPath(
- Dispatchers.DispatcherType.Transaction), self(), getContext(), shardMBean);
+ new Dispatchers(context().system().dispatchers()).getDispatcherPath(Dispatchers.DispatcherType.Transaction),
+ self(), getContext(), shardMBean, builder.getId().getShardName());
- snapshotCohort = new ShardSnapshotCohort(transactionActorFactory, store, LOG, this.name);
+ snapshotCohort = ShardSnapshotCohort.create(getContext(), builder.getId().getMemberName(), store, LOG,
+ this.name);
messageRetrySupport = new ShardTransactionMessageRetrySupport(this);
}
@@ -185,7 +223,7 @@ public class Shard extends RaftActor {
datastoreContext.getShardTransactionCommitTimeoutInSeconds(), TimeUnit.SECONDS) / 2;
}
- private Optional createRoleChangeNotifier(String shardId) {
+ private Optional createRoleChangeNotifier(final String shardId) {
ActorRef shardRoleChangeNotifier = this.getContext().actorOf(
RoleChangeNotifier.getProps(shardId), shardId + "-notifier");
return Optional.of(shardRoleChangeNotifier);
@@ -199,7 +237,7 @@ public class Shard extends RaftActor {
messageRetrySupport.close();
- if(txCommitTimeoutCheckSchedule != null) {
+ if (txCommitTimeoutCheckSchedule != null) {
txCommitTimeoutCheckSchedule.cancel();
}
@@ -219,16 +257,36 @@ public class Shard extends RaftActor {
}
}
+ @SuppressWarnings("checkstyle:IllegalCatch")
@Override
protected void handleNonRaftCommand(final Object message) {
- try (final MessageTracker.Context context = appendEntriesReplyTracker.received(message)) {
+ try (MessageTracker.Context context = appendEntriesReplyTracker.received(message)) {
final Optional maybeError = context.error();
if (maybeError.isPresent()) {
LOG.trace("{} : AppendEntriesReply failed to arrive at the expected interval {}", persistenceId(),
maybeError.get());
}
- if (CreateTransaction.isSerializedType(message)) {
+ if (message instanceof RequestEnvelope) {
+ final long now = ticker().read();
+ final RequestEnvelope envelope = (RequestEnvelope)message;
+
+ try {
+ final RequestSuccess success = handleRequest(envelope, now);
+ if (success != null) {
+ envelope.sendSuccess(success, ticker().read() - now);
+ }
+ } catch (RequestException e) {
+ LOG.debug("{}: request {} failed", persistenceId(), envelope, e);
+ envelope.sendFailure(e, ticker().read() - now);
+ } catch (Exception e) {
+ LOG.debug("{}: request {} caused failure", persistenceId(), envelope, e);
+ envelope.sendFailure(new RuntimeRequestException("Request failed to process", e),
+ ticker().read() - now);
+ }
+ } else if (message instanceof ConnectClientRequest) {
+ handleConnectClient((ConnectClientRequest)message);
+ } else if (CreateTransaction.isSerializedType(message)) {
handleCreateTransaction(message);
} else if (message instanceof BatchedModifications) {
handleBatchedModifications((BatchedModifications)message);
@@ -252,27 +310,27 @@ public class Shard extends RaftActor {
updateSchemaContext((UpdateSchemaContext) message);
} else if (message instanceof PeerAddressResolved) {
PeerAddressResolved resolved = (PeerAddressResolved) message;
- setPeerAddress(resolved.getPeerId().toString(),
- resolved.getPeerAddress());
+ setPeerAddress(resolved.getPeerId(), resolved.getPeerAddress());
} else if (TX_COMMIT_TIMEOUT_CHECK_MESSAGE.equals(message)) {
+ store.checkForExpiredTransactions(transactionCommitTimeout);
commitCoordinator.checkForExpiredTransactions(transactionCommitTimeout, this);
- } else if(message instanceof DatastoreContext) {
+ } else if (message instanceof DatastoreContext) {
onDatastoreContext((DatastoreContext)message);
- } else if(message instanceof RegisterRoleChangeListener){
+ } else if (message instanceof RegisterRoleChangeListener) {
roleChangeNotifier.get().forward(message, context());
} else if (message instanceof FollowerInitialSyncUpStatus) {
shardMBean.setFollowerInitialSyncStatus(((FollowerInitialSyncUpStatus) message).isInitialSyncDone());
context().parent().tell(message, self());
- } else if(GET_SHARD_MBEAN_MESSAGE.equals(message)){
+ } else if (GET_SHARD_MBEAN_MESSAGE.equals(message)) {
sender().tell(getShardMBean(), self());
- } else if(message instanceof GetShardDataTree) {
+ } else if (message instanceof GetShardDataTree) {
sender().tell(store.getDataTree(), self());
- } else if(message instanceof ServerRemoved){
+ } else if (message instanceof ServerRemoved) {
context().parent().forward(message, context());
- } else if(ShardTransactionMessageRetrySupport.TIMER_MESSAGE_CLASS.isInstance(message)) {
+ } else if (ShardTransactionMessageRetrySupport.TIMER_MESSAGE_CLASS.isInstance(message)) {
messageRetrySupport.onTimerMessage(message);
} else if (message instanceof DataTreeCohortActorRegistry.CohortRegistryCommand) {
- commitCoordinator.processCohortRegistryCommand(getSender(),
+ store.processCohortRegistryCommand(getSender(),
(DataTreeCohortActorRegistry.CohortRegistryCommand) message);
} else {
super.handleNonRaftCommand(message);
@@ -280,12 +338,93 @@ public class Shard extends RaftActor {
}
}
+ // Acquire our frontend tracking handle and verify generation matches
+ private LeaderFrontendState getFrontend(final ClientIdentifier clientId) throws RequestException {
+ final LeaderFrontendState existing = knownFrontends.get(clientId.getFrontendId());
+ if (existing != null) {
+ final int cmp = Long.compareUnsigned(existing.getIdentifier().getGeneration(), clientId.getGeneration());
+ if (cmp == 0) {
+ return existing;
+ }
+ if (cmp > 0) {
+ LOG.debug("{}: rejecting request from outdated client {}", persistenceId(), clientId);
+ throw new RetiredGenerationException(existing.getIdentifier().getGeneration());
+ }
+
+ LOG.info("{}: retiring state {}, outdated by request from client {}", persistenceId(), existing, clientId);
+ existing.retire();
+ knownFrontends.remove(clientId.getFrontendId());
+ } else {
+ LOG.debug("{}: client {} is not yet known", persistenceId(), clientId);
+ }
+
+ final LeaderFrontendState ret = new LeaderFrontendState(persistenceId(), clientId, store);
+ knownFrontends.put(clientId.getFrontendId(), ret);
+ LOG.debug("{}: created state {} for client {}", persistenceId(), ret, clientId);
+ return ret;
+ }
+
+ private static @Nonnull ABIVersion selectVersion(final ConnectClientRequest message) {
+ final Range clientRange = Range.closed(message.getMinVersion(), message.getMaxVersion());
+ for (ABIVersion v : SUPPORTED_ABIVERSIONS) {
+ if (clientRange.contains(v)) {
+ return v;
+ }
+ }
+
+ throw new IllegalArgumentException(String.format(
+ "No common version between backend versions %s and client versions %s", SUPPORTED_ABIVERSIONS,
+ clientRange));
+ }
+
+ @SuppressWarnings("checkstyle:IllegalCatch")
+ private void handleConnectClient(final ConnectClientRequest message) {
+ try {
+ if (!isLeader() || !isLeaderActive()) {
+ LOG.debug("{}: not currently leader, rejecting request {}", persistenceId(), message);
+ throw new NotLeaderException(getSelf());
+ }
+
+ final ABIVersion selectedVersion = selectVersion(message);
+ final LeaderFrontendState frontend = getFrontend(message.getTarget());
+ frontend.reconnect();
+ message.getReplyTo().tell(new ConnectClientSuccess(message.getTarget(), message.getSequence(), getSelf(),
+ ImmutableList.of(), store.getDataTree(), CLIENT_MAX_MESSAGES).toVersion(selectedVersion),
+ ActorRef.noSender());
+ } catch (RequestException | RuntimeException e) {
+ message.getReplyTo().tell(new Failure(e), ActorRef.noSender());
+ }
+ }
+
+ private @Nullable RequestSuccess handleRequest(final RequestEnvelope envelope, final long now)
+ throws RequestException {
+ // We are not the leader, hence we want to fail-fast.
+ if (!isLeader() || !isLeaderActive()) {
+ LOG.debug("{}: not currently leader, rejecting request {}", persistenceId(), envelope);
+ throw new NotLeaderException(getSelf());
+ }
+
+ final Request request = envelope.getMessage();
+ if (request instanceof TransactionRequest) {
+ final TransactionRequest txReq = (TransactionRequest)request;
+ final ClientIdentifier clientId = txReq.getTarget().getHistoryId().getClientId();
+ return getFrontend(clientId).handleTransactionRequest(txReq, envelope, now);
+ } else if (request instanceof LocalHistoryRequest) {
+ final LocalHistoryRequest lhReq = (LocalHistoryRequest)request;
+ final ClientIdentifier clientId = lhReq.getTarget().getClientId();
+ return getFrontend(clientId).handleLocalHistoryRequest(lhReq, envelope, now);
+ } else {
+ LOG.debug("{}: rejecting unsupported request {}", persistenceId(), request);
+ throw new UnsupportedRequestException(request);
+ }
+ }
+
private boolean hasLeader() {
return getLeaderId() != null;
}
public int getPendingTxCommitQueueSize() {
- return commitCoordinator.getQueueSize();
+ return store.getQueueSize();
}
public int getCohortCacheSize() {
@@ -298,55 +437,41 @@ public class Shard extends RaftActor {
}
@Override
- protected LeaderStateChanged newLeaderStateChanged(String memberId, String leaderId, short leaderPayloadVersion) {
+ protected LeaderStateChanged newLeaderStateChanged(final String memberId, final String leaderId,
+ final short leaderPayloadVersion) {
return isLeader() ? new ShardLeaderStateChanged(memberId, leaderId, store.getDataTree(), leaderPayloadVersion)
: new ShardLeaderStateChanged(memberId, leaderId, leaderPayloadVersion);
}
- protected void onDatastoreContext(DatastoreContext context) {
+ protected void onDatastoreContext(final DatastoreContext context) {
datastoreContext = context;
- commitCoordinator.setQueueCapacity(datastoreContext.getShardTransactionCommitQueueCapacity());
-
setTransactionCommitTimeout();
- if(datastoreContext.isPersistent() && !persistence().isRecoveryApplicable()) {
- setPersistence(true);
- } else if(!datastoreContext.isPersistent() && persistence().isRecoveryApplicable()) {
- setPersistence(false);
- }
+ setPersistence(datastoreContext.isPersistent());
updateConfigParams(datastoreContext.getShardRaftConfig());
}
- private static boolean isEmptyCommit(final DataTreeCandidate candidate) {
- return ModificationType.UNMODIFIED.equals(candidate.getRootNode().getModificationType());
- }
-
- 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(cohortEntry.getReplySender(), cohortEntry.getTransactionID(), candidate);
+ // applyState() will be invoked once consensus is reached on the payload
+ void persistPayload(final Identifier id, final Payload payload, final boolean batchHint) {
+ boolean canSkipPayload = !hasFollowers() && !persistence().isRecoveryApplicable();
+ if (canSkipPayload) {
+ applyState(self(), id, payload);
} else {
- persistData(cohortEntry.getReplySender(), cohortEntry.getTransactionID(),
- DataTreeCandidatePayload.create(candidate));
+ // We are faking the sender
+ persistData(self(), id, payload, batchHint);
}
}
private void handleCommitTransaction(final CommitTransaction commit) {
if (isLeader()) {
- if(!commitCoordinator.handleCommit(new StringIdentifier(commit.getTransactionID()), getSender(), this)) {
- shardMBean.incrementFailedTransactionsCount();
- }
+ commitCoordinator.handleCommit(commit.getTransactionId(), getSender(), this);
} else {
ActorSelection leader = getLeader();
if (leader == null) {
messageRetrySupport.addMessageToRetry(commit, getSender(),
- "Could not commit transaction " + commit.getTransactionID());
+ "Could not commit transaction " + commit.getTransactionId());
} else {
LOG.debug("{}: Forwarding CommitTransaction to leader {}", persistenceId(), leader);
leader.forward(commit, getContext());
@@ -354,96 +479,16 @@ public class Shard extends RaftActor {
}
}
- private void finishCommit(@Nonnull final ActorRef sender, @Nonnull final Identifier transactionID,
- @Nonnull final CohortEntry cohortEntry) {
- LOG.debug("{}: Finishing commit for transaction {}", persistenceId(), cohortEntry.getTransactionID());
-
- try {
- try {
- cohortEntry.commit();
- } catch(ExecutionException e) {
- // We may get a "store tree and candidate base differ" IllegalStateException from commit under
- // certain edge case scenarios so we'll try to re-apply the candidate from scratch as a last
- // resort. Eg, we're a follower and a tx payload is replicated but the leader goes down before
- // applying it to the state. We then become the leader and a second tx is pre-committed and
- // replicated. When consensus occurs, this will cause the first tx to be applied as a foreign
- // candidate via applyState prior to the second tx. Since the second tx has already been
- // pre-committed, when it gets here to commit it will get an IllegalStateException.
-
- // FIXME - this is not an ideal way to handle this scenario. This is temporary - a cleaner
- // solution will be forthcoming.
- if(e.getCause() instanceof IllegalStateException) {
- LOG.debug("{}: commit failed for transaction {} - retrying as foreign candidate", persistenceId(),
- transactionID, e);
- store.applyForeignCandidate(transactionID, cohortEntry.getCandidate());
- } else {
- throw e;
- }
- }
-
- sender.tell(CommitTransactionReply.instance(cohortEntry.getClientVersion()).toSerializable(), getSelf());
-
- shardMBean.incrementCommittedTransactionCount();
- shardMBean.setLastCommittedTransactionTime(System.currentTimeMillis());
-
- } catch (Exception e) {
- sender.tell(new akka.actor.Status.Failure(e), getSelf());
-
- LOG.error("{}, An exception occurred while committing transaction {}", persistenceId(),
- transactionID, e);
- shardMBean.incrementFailedTransactionsCount();
- } finally {
- commitCoordinator.currentTransactionComplete(transactionID, true);
- }
- }
-
- private void finishCommit(@Nonnull final ActorRef sender, final @Nonnull Identifier transactionID) {
- // With persistence enabled, this method is called via applyState by the leader strategy
- // after the commit has been replicated to a majority of the followers.
-
- CohortEntry cohortEntry = commitCoordinator.getCohortEntryIfCurrent(transactionID);
- if (cohortEntry == null) {
- // The transaction is no longer the current commit. This can happen if the transaction
- // was aborted prior, most likely due to timeout in the front-end. We need to finish
- // committing the transaction though since it was successfully persisted and replicated
- // however we can't use the original cohort b/c it was already preCommitted and may
- // conflict with the current commit or may have been aborted so we commit with a new
- // transaction.
- cohortEntry = commitCoordinator.getAndRemoveCohortEntry(transactionID);
- if(cohortEntry != null) {
- try {
- store.applyForeignCandidate(transactionID, cohortEntry.getCandidate());
- } catch (DataValidationFailedException e) {
- shardMBean.incrementFailedTransactionsCount();
- LOG.error("{}: Failed to re-apply transaction {}", persistenceId(), transactionID, e);
- }
-
- sender.tell(CommitTransactionReply.instance(cohortEntry.getClientVersion()).toSerializable(),
- getSelf());
- } else {
- // This really shouldn't happen - it likely means that persistence or replication
- // took so long to complete such that the cohort entry was expired from the cache.
- IllegalStateException ex = new IllegalStateException(
- String.format("%s: Could not finish committing transaction %s - no CohortEntry found",
- persistenceId(), transactionID));
- LOG.error(ex.getMessage());
- sender.tell(new akka.actor.Status.Failure(ex), getSelf());
- }
- } else {
- finishCommit(sender, transactionID, cohortEntry);
- }
- }
-
private void handleCanCommitTransaction(final CanCommitTransaction canCommit) {
- LOG.debug("{}: Can committing transaction {}", persistenceId(), canCommit.getTransactionID());
+ LOG.debug("{}: Can committing transaction {}", persistenceId(), canCommit.getTransactionId());
if (isLeader()) {
- commitCoordinator.handleCanCommit(new StringIdentifier(canCommit.getTransactionID()), getSender(), this);
+ commitCoordinator.handleCanCommit(canCommit.getTransactionId(), getSender(), this);
} else {
ActorSelection leader = getLeader();
if (leader == null) {
messageRetrySupport.addMessageToRetry(canCommit, getSender(),
- "Could not canCommit transaction " + canCommit.getTransactionID());
+ "Could not canCommit transaction " + canCommit.getTransactionId());
} else {
LOG.debug("{}: Forwarding CanCommitTransaction to leader {}", persistenceId(), leader);
leader.forward(canCommit, getContext());
@@ -451,17 +496,18 @@ public class Shard extends RaftActor {
}
}
- protected void handleBatchedModificationsLocal(BatchedModifications batched, ActorRef sender) {
+ @SuppressWarnings("checkstyle:IllegalCatch")
+ protected void handleBatchedModificationsLocal(final BatchedModifications batched, final ActorRef sender) {
try {
- commitCoordinator.handleBatchedModifications(batched, sender, this, store.getSchemaContext());
+ 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());
+ batched.getTransactionId(), e);
+ sender.tell(new Failure(e), getSelf());
}
}
- private void handleBatchedModifications(BatchedModifications batched) {
+ private void handleBatchedModifications(final BatchedModifications batched) {
// 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
@@ -481,30 +527,31 @@ public class Shard extends RaftActor {
ActorSelection leader = getLeader();
if (!isLeaderActive || leader == null) {
messageRetrySupport.addMessageToRetry(batched, getSender(),
- "Could not commit transaction " + batched.getTransactionID());
+ "Could not commit transaction " + batched.getTransactionId());
} else {
// If this is not the first batch and leadership changed in between batched messages,
// we need to reconstruct previous BatchedModifications from the transaction
// DataTreeModification, honoring the max batched modification count, and forward all the
// previous BatchedModifications to the new leader.
- Collection newModifications = commitCoordinator.createForwardedBatchedModifications(
- batched, datastoreContext.getShardBatchedModificationCount());
+ Collection newModifications = commitCoordinator
+ .createForwardedBatchedModifications(batched,
+ datastoreContext.getShardBatchedModificationCount());
LOG.debug("{}: Forwarding {} BatchedModifications to leader {}", persistenceId(),
newModifications.size(), leader);
- for(BatchedModifications bm: newModifications) {
+ for (BatchedModifications bm : newModifications) {
leader.forward(bm, getContext());
}
}
}
}
- 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.",
+ private boolean failIfIsolatedLeader(final ActorRef sender) {
+ if (isIsolatedLeader()) {
+ sender.tell(new 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;
}
@@ -516,23 +563,24 @@ public class Shard extends RaftActor {
return getRaftState() == RaftState.IsolatedLeader;
}
+ @SuppressWarnings("checkstyle:IllegalCatch")
private void handleReadyLocalTransaction(final ReadyLocalTransaction message) {
- LOG.debug("{}: handleReadyLocalTransaction for {}", persistenceId(), message.getTransactionID());
+ LOG.debug("{}: handleReadyLocalTransaction for {}", persistenceId(), message.getTransactionId());
boolean isLeaderActive = isLeaderActive();
if (isLeader() && isLeaderActive) {
try {
- commitCoordinator.handleReadyLocalTransaction(message, getSender(), this, store.getSchemaContext());
+ 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());
+ message.getTransactionId(), e);
+ getSender().tell(new Failure(e), getSelf());
}
} else {
ActorSelection leader = getLeader();
if (!isLeaderActive || leader == null) {
messageRetrySupport.addMessageToRetry(message, getSender(),
- "Could not commit transaction " + message.getTransactionID());
+ "Could not commit transaction " + message.getTransactionId());
} else {
LOG.debug("{}: Forwarding ReadyLocalTransaction to leader {}", persistenceId(), leader);
message.setRemoteVersion(getCurrentBehavior().getLeaderPayloadVersion());
@@ -541,22 +589,21 @@ public class Shard extends RaftActor {
}
}
- private void handleForwardedReadyTransaction(ForwardedReadyTransaction forwardedReady) {
- LOG.debug("{}: handleForwardedReadyTransaction for {}", persistenceId(), forwardedReady.getTransactionID());
+ private void handleForwardedReadyTransaction(final ForwardedReadyTransaction forwardedReady) {
+ LOG.debug("{}: handleForwardedReadyTransaction for {}", persistenceId(), forwardedReady.getTransactionId());
boolean isLeaderActive = isLeaderActive();
if (isLeader() && isLeaderActive) {
- commitCoordinator.handleForwardedReadyTransaction(forwardedReady, getSender(), this,
- store.getSchemaContext());
+ commitCoordinator.handleForwardedReadyTransaction(forwardedReady, getSender(), this);
} else {
ActorSelection leader = getLeader();
if (!isLeaderActive || leader == null) {
messageRetrySupport.addMessageToRetry(forwardedReady, getSender(),
- "Could not commit transaction " + forwardedReady.getTransactionID());
+ "Could not commit transaction " + forwardedReady.getTransactionId());
} else {
LOG.debug("{}: Forwarding ForwardedReadyTransaction to leader {}", persistenceId(), leader);
- ReadyLocalTransaction readyLocal = new ReadyLocalTransaction(forwardedReady.getTransactionID(),
+ ReadyLocalTransaction readyLocal = new ReadyLocalTransaction(forwardedReady.getTransactionId(),
forwardedReady.getTransaction().getSnapshot(), forwardedReady.isDoImmediateCommit());
readyLocal.setRemoteVersion(getCurrentBehavior().getLeaderPayloadVersion());
leader.forward(readyLocal, getContext());
@@ -565,11 +612,11 @@ public class Shard extends RaftActor {
}
private void handleAbortTransaction(final AbortTransaction abort) {
- doAbortTransaction(abort.getTransactionID(), getSender());
+ doAbortTransaction(abort.getTransactionId(), getSender());
}
- void doAbortTransaction(final String transactionID, final ActorRef sender) {
- commitCoordinator.handleAbort(new StringIdentifier(transactionID), sender, this);
+ void doAbortTransaction(final Identifier transactionID, final ActorRef sender) {
+ commitCoordinator.handleAbort(transactionID, sender, this);
}
private void handleCreateTransaction(final Object message) {
@@ -578,50 +625,39 @@ public class Shard extends RaftActor {
} else if (getLeader() != null) {
getLeader().forward(message, getContext());
} else {
- getSender().tell(new akka.actor.Status.Failure(new NoShardLeaderException(
+ getSender().tell(new Failure(new NoShardLeaderException(
"Could not create a shard transaction", persistenceId())), getSelf());
}
}
private void closeTransactionChain(final CloseTransactionChain closeTransactionChain) {
- store.closeTransactionChain(closeTransactionChain.getTransactionChainId());
+ final LocalHistoryIdentifier id = closeTransactionChain.getIdentifier();
+ store.closeTransactionChain(id, null);
+ store.purgeTransactionChain(id, null);
}
- private void createTransaction(CreateTransaction createTransaction) {
+ @SuppressWarnings("checkstyle:IllegalCatch")
+ private void createTransaction(final CreateTransaction createTransaction) {
try {
- if(TransactionType.fromInt(createTransaction.getTransactionType()) != TransactionType.READ_ONLY &&
- failIfIsolatedLeader(getSender())) {
+ if (TransactionType.fromInt(createTransaction.getTransactionType()) != TransactionType.READ_ONLY
+ && failIfIsolatedLeader(getSender())) {
return;
}
ActorRef transactionActor = createTransaction(createTransaction.getTransactionType(),
- createTransaction.getTransactionId(), createTransaction.getTransactionChainId());
+ createTransaction.getTransactionId());
getSender().tell(new CreateTransactionReply(Serialization.serializedActorPath(transactionActor),
createTransaction.getTransactionId(), createTransaction.getVersion()).toSerializable(), getSelf());
} catch (Exception e) {
- getSender().tell(new akka.actor.Status.Failure(e), getSelf());
+ getSender().tell(new Failure(e), getSelf());
}
}
- private ActorRef createTransaction(int transactionType, String remoteTransactionId, String transactionChainId) {
- ShardTransactionIdentifier transactionId = new ShardTransactionIdentifier(remoteTransactionId);
+ private ActorRef createTransaction(final int transactionType, final TransactionIdentifier transactionId) {
LOG.debug("{}: Creating transaction : {} ", persistenceId(), transactionId);
return transactionActorFactory.newShardTransaction(TransactionType.fromInt(transactionType),
- transactionId, transactionChainId);
- }
-
- private void commitWithNewTransaction(final Modification modification) {
- ReadWriteShardDataTreeTransaction tx = store.newReadWriteTransaction(modification.toString(), null);
- modification.apply(tx.getSnapshot());
- try {
- snapshotCohort.syncCommitTransaction(tx);
- shardMBean.incrementCommittedTransactionCount();
- shardMBean.setLastCommittedTransactionTime(System.currentTimeMillis());
- } catch (Exception e) {
- shardMBean.incrementFailedTransactionsCount();
- LOG.error("{}: Failed to commit", persistenceId(), e);
- }
+ transactionId);
}
private void updateSchemaContext(final UpdateSchemaContext message) {
@@ -647,8 +683,8 @@ public class Shard extends RaftActor {
@Override
@Nonnull
protected RaftActorRecoveryCohort getRaftActorRecoveryCohort() {
- return new ShardRecoveryCoordinator(store, store.getSchemaContext(),
- restoreFromSnapshot != null ? restoreFromSnapshot.getSnapshot() : null, persistenceId(), LOG);
+ return new ShardRecoveryCoordinator(store,
+ restoreFromSnapshot != null ? restoreFromSnapshot.getSnapshot() : null, persistenceId(), LOG);
}
@Override
@@ -659,7 +695,7 @@ public class Shard extends RaftActor {
getContext().parent().tell(new ActorInitialized(), getSelf());
// Being paranoid here - this method should only be called once but just in case...
- if(txCommitTimeoutCheckSchedule == null) {
+ if (txCommitTimeoutCheckSchedule == null) {
// Schedule a message to be periodically sent to check if the current in-progress
// transaction should be expired and aborted.
FiniteDuration period = Duration.create(transactionCommitTimeout / 3, TimeUnit.MILLISECONDS);
@@ -671,45 +707,14 @@ public class Shard extends RaftActor {
@Override
protected void applyState(final ActorRef clientActor, final Identifier identifier, final Object data) {
- if (data instanceof DataTreeCandidatePayload) {
- if (clientActor == null) {
- // No clientActor indicates a replica coming from the leader
- try {
- store.applyForeignCandidate(identifier, ((DataTreeCandidatePayload)data).getCandidate());
- } catch (DataValidationFailedException | IOException e) {
- LOG.error("{}: Error applying replica {}", persistenceId(), identifier, e);
- }
- } else {
- // Replication consensus reached, proceed to commit
- finishCommit(clientActor, identifier);
+ if (data instanceof Payload) {
+ try {
+ store.applyReplicatedPayload(identifier, (Payload)data);
+ } catch (DataValidationFailedException | IOException e) {
+ LOG.error("{}: Error applying replica {}", persistenceId(), identifier, e);
}
- } else if (data instanceof CompositeModificationPayload) {
- Object modification = ((CompositeModificationPayload) data).getModification();
-
- applyModificationToState(clientActor, identifier, modification);
- } else if(data instanceof CompositeModificationByteStringPayload ){
- Object modification = ((CompositeModificationByteStringPayload) data).getModification();
-
- applyModificationToState(clientActor, identifier, modification);
} else {
- LOG.error("{}: Unknown state received {} Class loader = {} CompositeNodeMod.ClassLoader = {}",
- persistenceId(), data, data.getClass().getClassLoader(),
- CompositeModificationPayload.class.getClassLoader());
- }
- }
-
- private void applyModificationToState(ActorRef clientActor, Identifier identifier, Object modification) {
- if(modification == null) {
- LOG.error(
- "{}: modification is null - this is very unexpected, clientActor = {}, identifier = {}",
- persistenceId(), identifier, clientActor != null ? clientActor.path().toString() : null);
- } else if(clientActor == null) {
- // There's no clientActor to which to send a commit reply so we must be applying
- // replicated state from the leader.
- commitWithNewTransaction(MutableCompositeModification.fromSerializable(modification));
- } else {
- // This must be the OK to commit after replication consensus.
- finishCommit(clientActor, identifier);
+ LOG.error("{}: Unknown state for {} received {}", persistenceId(), identifier, data);
}
}
@@ -722,7 +727,7 @@ public class Shard extends RaftActor {
// If this actor is no longer the leader close all the transaction chains
if (!isLeader) {
- if(LOG.isDebugEnabled()) {
+ if (LOG.isDebugEnabled()) {
LOG.debug(
"{}: onStateChanged: Closing all transaction chains because shard {} is no longer the leader",
persistenceId(), getId());
@@ -731,48 +736,63 @@ public class Shard extends RaftActor {
store.closeAllTransactionChains();
}
- if(hasLeader && !isIsolatedLeader()) {
+ if (hasLeader && !isIsolatedLeader()) {
messageRetrySupport.retryMessages();
}
}
@Override
- protected void onLeaderChanged(String oldLeader, String newLeader) {
+ protected void onLeaderChanged(final String oldLeader, final String newLeader) {
shardMBean.incrementLeadershipChangeCount();
boolean hasLeader = hasLeader();
- if(hasLeader && !isLeader()) {
+ if (hasLeader && !isLeader()) {
// Another leader was elected. If we were the previous leader and had pending transactions, convert
// them to transaction messages and send to the new leader.
ActorSelection leader = getLeader();
- if(leader != null) {
- Collection