com.hazelcast.cp.internal.RaftService Maven / Gradle / Ivy
Go to download
Show more of this group Show more artifacts with this name
Show all versions of hazelcast-all Show documentation
Show all versions of hazelcast-all Show documentation
Kubernetes Service Discovery for Hazelcast Discovery SPI
/*
* Copyright (c) 2008-2020, Hazelcast, Inc. All Rights Reserved.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package com.hazelcast.cp.internal;
import com.hazelcast.cluster.Member;
import com.hazelcast.cluster.impl.MemberImpl;
import com.hazelcast.config.cp.CPSubsystemConfig;
import com.hazelcast.config.cp.RaftAlgorithmConfig;
import com.hazelcast.cp.CPGroup;
import com.hazelcast.cp.CPGroupId;
import com.hazelcast.cp.CPMember;
import com.hazelcast.cp.exception.CPGroupDestroyedException;
import com.hazelcast.cp.internal.datastructures.spi.RaftManagedService;
import com.hazelcast.cp.internal.datastructures.spi.RaftRemoteService;
import com.hazelcast.cp.internal.exception.CannotRemoveCPMemberException;
import com.hazelcast.cp.internal.operation.ResetCPMemberOp;
import com.hazelcast.cp.internal.operation.unsafe.UnsafeStateReplicationOp;
import com.hazelcast.cp.internal.persistence.CPPersistenceService;
import com.hazelcast.cp.internal.raft.SnapshotAwareService;
import com.hazelcast.cp.internal.raft.impl.RaftEndpoint;
import com.hazelcast.cp.internal.raft.impl.RaftIntegration;
import com.hazelcast.cp.internal.raft.impl.RaftNode;
import com.hazelcast.cp.internal.raft.impl.RaftNodeImpl;
import com.hazelcast.cp.internal.raft.impl.RaftNodeStatus;
import com.hazelcast.cp.internal.raft.impl.dto.AppendFailureResponse;
import com.hazelcast.cp.internal.raft.impl.dto.AppendRequest;
import com.hazelcast.cp.internal.raft.impl.dto.AppendSuccessResponse;
import com.hazelcast.cp.internal.raft.impl.dto.InstallSnapshot;
import com.hazelcast.cp.internal.raft.impl.dto.PreVoteRequest;
import com.hazelcast.cp.internal.raft.impl.dto.PreVoteResponse;
import com.hazelcast.cp.internal.raft.impl.dto.TriggerLeaderElection;
import com.hazelcast.cp.internal.raft.impl.dto.VoteRequest;
import com.hazelcast.cp.internal.raft.impl.dto.VoteResponse;
import com.hazelcast.cp.internal.raft.impl.log.RaftLog;
import com.hazelcast.cp.internal.raft.impl.persistence.LogFileStructure;
import com.hazelcast.cp.internal.raft.impl.persistence.RaftStateStore;
import com.hazelcast.cp.internal.raft.impl.persistence.RestoredRaftState;
import com.hazelcast.cp.internal.raft.impl.state.RaftState;
import com.hazelcast.cp.internal.raftop.GetInitialRaftGroupMembersIfCurrentGroupMemberOp;
import com.hazelcast.cp.internal.raftop.metadata.AddCPMemberOp;
import com.hazelcast.cp.internal.raftop.metadata.ForceDestroyRaftGroupOp;
import com.hazelcast.cp.internal.raftop.metadata.GetActiveCPMembersOp;
import com.hazelcast.cp.internal.raftop.metadata.GetActiveRaftGroupByNameOp;
import com.hazelcast.cp.internal.raftop.metadata.GetActiveRaftGroupIdsOp;
import com.hazelcast.cp.internal.raftop.metadata.GetRaftGroupIdsOp;
import com.hazelcast.cp.internal.raftop.metadata.GetRaftGroupOp;
import com.hazelcast.cp.internal.raftop.metadata.RaftServicePreJoinOp;
import com.hazelcast.cp.internal.raftop.metadata.RemoveCPMemberOp;
import com.hazelcast.internal.cluster.ClusterService;
import com.hazelcast.internal.diagnostics.MetricsPlugin;
import com.hazelcast.internal.metrics.DynamicMetricsProvider;
import com.hazelcast.internal.metrics.MetricDescriptor;
import com.hazelcast.internal.metrics.MetricsCollectionContext;
import com.hazelcast.internal.metrics.MetricsRegistry;
import com.hazelcast.internal.metrics.Probe;
import com.hazelcast.internal.metrics.ProbeLevel;
import com.hazelcast.internal.partition.MigrationAwareService;
import com.hazelcast.internal.partition.MigrationEndpoint;
import com.hazelcast.internal.partition.PartitionMigrationEvent;
import com.hazelcast.internal.partition.PartitionReplicationEvent;
import com.hazelcast.internal.serialization.Data;
import com.hazelcast.internal.services.GracefulShutdownAwareService;
import com.hazelcast.internal.services.ManagedService;
import com.hazelcast.internal.services.MembershipAwareService;
import com.hazelcast.internal.services.MembershipServiceEvent;
import com.hazelcast.internal.services.PreJoinAwareService;
import com.hazelcast.internal.util.Clock;
import com.hazelcast.internal.util.ExceptionUtil;
import com.hazelcast.internal.util.executor.ManagedExecutorService;
import com.hazelcast.logging.ILogger;
import com.hazelcast.spi.exception.PartitionMigratingException;
import com.hazelcast.spi.exception.ResponseAlreadySentException;
import com.hazelcast.spi.impl.InternalCompletableFuture;
import com.hazelcast.spi.impl.NodeEngine;
import com.hazelcast.spi.impl.NodeEngineImpl;
import com.hazelcast.spi.impl.executionservice.ExecutionService;
import com.hazelcast.spi.impl.operationexecutor.impl.PartitionOperationThread;
import com.hazelcast.spi.impl.operationservice.Operation;
import com.hazelcast.spi.impl.operationservice.impl.OperationServiceImpl;
import com.hazelcast.spi.impl.servicemanager.ServiceInfo;
import java.util.ArrayList;
import java.util.Collection;
import java.util.Collections;
import java.util.List;
import java.util.Map.Entry;
import java.util.Properties;
import java.util.Set;
import java.util.UUID;
import java.util.concurrent.ConcurrentHashMap;
import java.util.concurrent.ConcurrentMap;
import java.util.concurrent.ExecutionException;
import java.util.concurrent.Future;
import java.util.concurrent.TimeUnit;
import java.util.concurrent.atomic.AtomicInteger;
import java.util.concurrent.locks.ReadWriteLock;
import java.util.concurrent.locks.ReentrantReadWriteLock;
import java.util.function.BiConsumer;
import static com.hazelcast.cluster.memberselector.MemberSelectors.NON_LOCAL_MEMBER_SELECTOR;
import static com.hazelcast.cp.CPGroup.DEFAULT_GROUP_NAME;
import static com.hazelcast.cp.CPGroup.METADATA_CP_GROUP_NAME;
import static com.hazelcast.cp.internal.RaftGroupMembershipManager.MANAGEMENT_TASK_PERIOD_IN_MILLIS;
import static com.hazelcast.cp.internal.raft.QueryPolicy.LEADER_LOCAL;
import static com.hazelcast.cp.internal.raft.QueryPolicy.LINEARIZABLE;
import static com.hazelcast.cp.internal.raft.impl.RaftNodeImpl.newRaftNode;
import static com.hazelcast.internal.config.ConfigValidator.checkCPSubsystemConfig;
import static com.hazelcast.internal.metrics.MetricDescriptorConstants.CP_DISCRIMINATOR_GROUPID;
import static com.hazelcast.internal.metrics.MetricDescriptorConstants.CP_METRIC_RAFT_SERVICE_DESTROYED_GROUP_IDS;
import static com.hazelcast.internal.metrics.MetricDescriptorConstants.CP_METRIC_RAFT_SERVICE_MISSING_MEMBERS;
import static com.hazelcast.internal.metrics.MetricDescriptorConstants.CP_METRIC_RAFT_SERVICE_NODES;
import static com.hazelcast.internal.metrics.MetricDescriptorConstants.CP_METRIC_RAFT_SERVICE_TERMINATED_RAFT_NODE_GROUP_IDS;
import static com.hazelcast.internal.metrics.MetricDescriptorConstants.CP_PREFIX_RAFT;
import static com.hazelcast.internal.metrics.MetricDescriptorConstants.CP_PREFIX_RAFT_GROUP;
import static com.hazelcast.internal.metrics.MetricDescriptorConstants.CP_PREFIX_RAFT_METADATA;
import static com.hazelcast.internal.metrics.MetricDescriptorConstants.CP_TAG_NAME;
import static com.hazelcast.internal.util.Preconditions.checkFalse;
import static com.hazelcast.internal.util.Preconditions.checkNotNull;
import static com.hazelcast.internal.util.Preconditions.checkState;
import static com.hazelcast.internal.util.Preconditions.checkTrue;
import static com.hazelcast.internal.util.UuidUtil.newUnsecureUUID;
import static com.hazelcast.spi.impl.InternalCompletableFuture.newCompletedFuture;
import static com.hazelcast.spi.impl.executionservice.ExecutionService.SYSTEM_EXECUTOR;
import static java.util.Collections.newSetFromMap;
import static java.util.concurrent.TimeUnit.MILLISECONDS;
import static java.util.concurrent.TimeUnit.SECONDS;
/**
* Contains {@link RaftNode} instances that run the Raft consensus algorithm
* for the created CP groups. Also implements CP Subsystem management methods.
*/
@SuppressWarnings({"checkstyle:methodcount", "checkstyle:classfanoutcomplexity", "checkstyle:classdataabstractioncoupling"})
public class RaftService implements ManagedService, SnapshotAwareService, GracefulShutdownAwareService,
MembershipAwareService, PreJoinAwareService, RaftNodeLifecycleAwareService,
MigrationAwareService, DynamicMetricsProvider {
public static final String SERVICE_NAME = "hz:core:raft";
public static final String CP_SUBSYSTEM_EXECUTOR = "hz:cpSubsystem";
static final String CP_SUBSYSTEM_MANAGEMENT_EXECUTOR = "hz:cpSubsystemManagement";
private static final long REMOVE_MISSING_MEMBER_TASK_PERIOD_SECONDS = 1;
private static final int AWAIT_DISCOVERY_STEP_MILLIS = 10;
private final ReadWriteLock nodeLock = new ReentrantReadWriteLock();
@Probe(name = CP_METRIC_RAFT_SERVICE_NODES)
private final ConcurrentMap nodes = new ConcurrentHashMap<>();
private final ConcurrentMap nodeMetrics = new ConcurrentHashMap<>();
private final NodeEngineImpl nodeEngine;
private final ILogger logger;
@Probe(name = CP_METRIC_RAFT_SERVICE_DESTROYED_GROUP_IDS)
private final Set destroyedGroupIds = newSetFromMap(new ConcurrentHashMap<>());
@Probe(name = CP_METRIC_RAFT_SERVICE_TERMINATED_RAFT_NODE_GROUP_IDS)
private final Set terminatedRaftNodeGroupIds = newSetFromMap(new ConcurrentHashMap<>());
private final CPSubsystemConfig config;
private final RaftInvocationManager invocationManager;
private final MetadataRaftGroupManager metadataGroupManager;
@Probe(name = CP_METRIC_RAFT_SERVICE_MISSING_MEMBERS)
private final ConcurrentMap missingMembers = new ConcurrentHashMap<>();
private final int metricsPeriod;
private final boolean cpSubsystemEnabled;
private final UnsafeModePartitionState[] unsafeModeStates;
public RaftService(NodeEngine nodeEngine) {
this.nodeEngine = (NodeEngineImpl) nodeEngine;
this.logger = nodeEngine.getLogger(getClass());
CPSubsystemConfig cpSubsystemConfig = nodeEngine.getConfig().getCPSubsystemConfig();
this.config = cpSubsystemConfig != null ? new CPSubsystemConfig(cpSubsystemConfig) : new CPSubsystemConfig();
checkCPSubsystemConfig(config);
this.cpSubsystemEnabled = config.getCPMemberCount() > 0;
this.invocationManager = new RaftInvocationManager(nodeEngine, this);
this.metadataGroupManager = new MetadataRaftGroupManager(this.nodeEngine, this, config);
if (cpSubsystemEnabled) {
this.unsafeModeStates = null;
} else {
this.unsafeModeStates = new UnsafeModePartitionState[nodeEngine.getPartitionService().getPartitionCount()];
for (int i = 0; i < unsafeModeStates.length; i++) {
unsafeModeStates[i] = new UnsafeModePartitionState();
}
}
MetricsRegistry metricsRegistry = this.nodeEngine.getMetricsRegistry();
metricsRegistry.registerStaticMetrics(this, CP_PREFIX_RAFT);
metricsRegistry.registerStaticMetrics(metadataGroupManager, CP_PREFIX_RAFT_METADATA);
metricsRegistry.registerDynamicMetricsProvider(this);
this.metricsPeriod = nodeEngine.getProperties().getInteger(MetricsPlugin.PERIOD_SECONDS);
}
@Override
public void init(NodeEngine nodeEngine, Properties properties) {
if (!metadataGroupManager.init()) {
return;
}
if (config.getMissingCPMemberAutoRemovalSeconds() > 0) {
ExecutionService executionService = nodeEngine.getExecutionService();
executionService.scheduleWithRepetition(CP_SUBSYSTEM_MANAGEMENT_EXECUTOR, new AutoRemoveMissingCPMemberTask(),
REMOVE_MISSING_MEMBER_TASK_PERIOD_SECONDS, REMOVE_MISSING_MEMBER_TASK_PERIOD_SECONDS, SECONDS);
}
MetricsRegistry metricsRegistry = this.nodeEngine.getMetricsRegistry();
metricsRegistry.scheduleAtFixedRate(new PublishNodeMetricsTask(), metricsPeriod, SECONDS, ProbeLevel.INFO);
}
@Override
public void reset() {
}
@Override
public void shutdown(boolean terminate) {
if (getCPPersistenceService().isEnabled()) {
List futures = new ArrayList<>(nodes.size());
for (RaftNode raftNode : nodes.values()) {
futures.add(raftNode.forceSetTerminatedStatus());
}
for (Future future : futures) {
try {
future.get();
} catch (Exception e) {
logger.severe("Error while terminating RaftNode", e);
}
}
}
}
@Override
public MetadataRaftGroupSnapshot takeSnapshot(CPGroupId groupId, long commitIndex) {
return metadataGroupManager.takeSnapshot(groupId, commitIndex);
}
@Override
public void restoreSnapshot(CPGroupId groupId, long commitIndex, MetadataRaftGroupSnapshot snapshot) {
metadataGroupManager.restoreSnapshot(groupId, commitIndex, snapshot);
}
public InternalCompletableFuture> getAllCPGroupIds() {
return invocationManager.query(getMetadataGroupId(), new GetRaftGroupIdsOp(), LINEARIZABLE);
}
public InternalCompletableFuture> getCPGroupIds() {
return invocationManager.query(getMetadataGroupId(), new GetActiveRaftGroupIdsOp(), LINEARIZABLE);
}
public InternalCompletableFuture getCPGroup(CPGroupId groupId) {
return invocationManager.query(getMetadataGroupId(), new GetRaftGroupOp(groupId), LINEARIZABLE);
}
public InternalCompletableFuture getCPGroup(String name) {
return invocationManager.query(getMetadataGroupId(), new GetActiveRaftGroupByNameOp(name), LINEARIZABLE);
}
InternalCompletableFuture resetCPSubsystem() {
checkState(cpSubsystemEnabled, "CP Subsystem is not enabled!");
InternalCompletableFuture future = newCompletableFuture();
ClusterService clusterService = nodeEngine.getClusterService();
Collection members = clusterService.getMembers(NON_LOCAL_MEMBER_SELECTOR);
if (!clusterService.isMaster()) {
return complete(future, new IllegalStateException("Only master can reset CP Subsystem!"));
}
if (config.getCPMemberCount() > members.size() + 1) {
return complete(future, new IllegalStateException("Not enough cluster members to reset CP Subsystem! "
+ "Required: " + config.getCPMemberCount() + ", available: " + (members.size() + 1)));
}
BiConsumer callback = new BiConsumer() {
final AtomicInteger latch = new AtomicInteger(members.size());
volatile Throwable failure;
@Override
public void accept(Void aVoid, Throwable throwable) {
if (throwable == null) {
if (latch.decrementAndGet() == 0) {
if (failure == null) {
future.complete(null);
} else {
complete(future, failure);
}
}
} else {
failure = throwable;
if (latch.decrementAndGet() == 0) {
complete(future, throwable);
}
}
}
};
long seed = newSeed();
logger.warning("Resetting CP Subsystem with groupId seed: " + seed);
resetLocal(seed);
OperationServiceImpl operationService = nodeEngine.getOperationService();
for (Member member : members) {
Operation op = new ResetCPMemberOp(seed);
operationService.invokeOnTarget(SERVICE_NAME, op, member.getAddress()).whenCompleteAsync(callback);
}
return future;
}
private long newSeed() {
long currentSeed = metadataGroupManager.getGroupIdSeed();
long seed = Clock.currentTimeMillis();
while (seed <= currentSeed) {
seed++;
}
return seed;
}
public void resetLocal(long seed) {
if (seed == 0L) {
throw new IllegalArgumentException("Seed cannot be zero!");
}
if (seed == metadataGroupManager.getGroupIdSeed()) {
// we have already seen this seed
logger.severe("Ignoring reset request. Current groupId seed is already equal to " + seed);
return;
}
nodeLock.writeLock().lock();
try {
// we should clear the current raft state before resetting the metadata manager
resetLocalRaftState();
getCPPersistenceService().reset();
metadataGroupManager.restart(seed);
logger.info("Local CP state is reset with groupId seed: " + seed);
} finally {
nodeLock.writeLock().unlock();
}
}
private void resetLocalRaftState() {
// node.forceSetTerminatedStatus() will trigger RaftNodeLifecycleAwareService.onRaftGroupDestroyed()
// which will attempt to acquire the read lock on nodeLock. In order to prevent it, we first
// add group ids into destroyedGroupIds to short-cut RaftNodeLifecycleAwareService.onRaftGroupDestroyed()
List futures = new ArrayList<>(nodes.size());
destroyedGroupIds.addAll(nodes.keySet());
for (RaftNode node : nodes.values()) {
InternalCompletableFuture f = node.forceSetTerminatedStatus();
futures.add(f);
}
for (InternalCompletableFuture future : futures) {
try {
future.get();
} catch (Exception e) {
logger.warning(e);
}
}
nodes.clear();
for (ServiceInfo serviceInfo : nodeEngine.getServiceInfos(RaftRemoteService.class)) {
if (serviceInfo.getService() instanceof RaftManagedService) {
((RaftManagedService) serviceInfo.getService()).onCPSubsystemRestart();
}
}
nodeMetrics.clear();
missingMembers.clear();
invocationManager.reset();
}
public InternalCompletableFuture promoteToCPMember() {
InternalCompletableFuture future = newCompletableFuture();
if (!metadataGroupManager.isDiscoveryCompleted()) {
return complete(future, new IllegalStateException("CP Subsystem discovery is not completed yet!"));
}
if (nodeEngine.getLocalMember().isLiteMember()) {
return complete(future, new IllegalStateException("Lite members cannot be promoted to CP member!"));
}
if (getLocalCPMember() != null) {
future.complete(null);
return future;
}
MemberImpl localMember = nodeEngine.getLocalMember();
// Local member may be recovered during restart, for instance via Hot Restart,
// but Raft state cannot be recovered back.
// That's why we generate a new UUID while promoting a member to CP.
// This new UUID generation can be removed when Hot Restart allows to recover Raft state.
CPMemberInfo member = new CPMemberInfo(newUnsecureUUID(), localMember.getAddress());
logger.info("Adding new CP member: " + member);
invocationManager.invoke(getMetadataGroupId(), new AddCPMemberOp(member))
.whenCompleteAsync((response, t) -> {
if (t == null) {
metadataGroupManager.initPromotedCPMember(member);
future.complete(null);
} else {
complete(future, t);
}
});
return future;
}
private InternalCompletableFuture newCompletableFuture() {
ManagedExecutorService executor = nodeEngine.getExecutionService().getExecutor(SYSTEM_EXECUTOR);
return InternalCompletableFuture.withExecutor(executor);
}
public InternalCompletableFuture removeCPMember(UUID cpMemberUuid) {
ClusterService clusterService = nodeEngine.getClusterService();
InternalCompletableFuture future = newCompletableFuture();
BiConsumer removeMemberCallback = (response, t) -> {
if (t == null) {
future.complete(null);
} else {
if (t instanceof CannotRemoveCPMemberException) {
t = new IllegalStateException(t.getMessage());
}
complete(future, t);
}
};
invocationManager.>invoke(getMetadataGroupId(), new GetActiveCPMembersOp())
.whenCompleteAsync((cpMembers, t) -> {
if (t == null) {
CPMemberInfo cpMemberToRemove = null;
for (CPMember cpMember : cpMembers) {
if (cpMember.getUuid().equals(cpMemberUuid)) {
cpMemberToRemove = (CPMemberInfo) cpMember;
break;
}
}
if (cpMemberToRemove == null) {
complete(future, new IllegalArgumentException("No CPMember found with uuid: " + cpMemberUuid));
return;
} else {
Member member = clusterService.getMember(cpMemberToRemove.getAddress());
if (member != null) {
logger.warning("Only unreachable/crashed CP members should be removed. " + member + " is alive but "
+ cpMemberToRemove + " with the same address is being removed.");
}
}
invokeTriggerRemoveMember(cpMemberToRemove).whenCompleteAsync(removeMemberCallback);
} else {
complete(future, t);
}
});
return future;
}
/**
* this method is idempotent
*/
public InternalCompletableFuture forceDestroyCPGroup(String groupName) {
return invocationManager.invoke(getMetadataGroupId(), new ForceDestroyRaftGroupOp(groupName));
}
public InternalCompletableFuture> getCPMembers() {
return invocationManager.query(getMetadataGroupId(), new GetActiveCPMembersOp(), LINEARIZABLE);
}
public boolean isDiscoveryCompleted() {
return metadataGroupManager.isDiscoveryCompleted();
}
public boolean awaitUntilDiscoveryCompleted(long timeout, TimeUnit timeUnit) throws InterruptedException {
long timeoutMillis = timeUnit.toMillis(timeout);
while (timeoutMillis > 0 && !metadataGroupManager.isDiscoveryCompleted()) {
long sleepMillis = Math.min(AWAIT_DISCOVERY_STEP_MILLIS, timeoutMillis);
Thread.sleep(sleepMillis);
timeoutMillis -= sleepMillis;
}
return metadataGroupManager.isDiscoveryCompleted();
}
@Override
public boolean onShutdown(long timeout, TimeUnit unit) {
CPMemberInfo localMember = getLocalCPMember();
if (localMember == null) {
return true;
}
if (getCPPersistenceService().isEnabled()) {
// When persistence is enabled, we do not remove this member from CP Subsystem.
// Because it is supposed to recover by restoring disk data.
return true;
}
logger.fine("Triggering remove member procedure for " + localMember);
if (ensureCPMemberRemoved(localMember, unit.toNanos(timeout))) {
return true;
}
logger.fine("Remove member procedure NOT completed for " + localMember + " in " + unit.toMillis(timeout) + " ms.");
return false;
}
private boolean ensureCPMemberRemoved(CPMemberInfo member, long remainingTimeNanos) {
while (remainingTimeNanos > 0) {
long start = System.nanoTime();
try {
if (metadataGroupManager.getActiveMembers().size() == 1) {
logger.warning("I am one of the last 2 CP members...");
return true;
}
invokeTriggerRemoveMember(member).get();
logger.fine(member + " is marked as being removed.");
break;
} catch (ExecutionException e) {
if (!(e.getCause() instanceof CannotRemoveCPMemberException)) {
throw ExceptionUtil.rethrow(e);
}
remainingTimeNanos -= (System.nanoTime() - start);
if (remainingTimeNanos <= 0) {
throw new IllegalStateException(e.getMessage());
}
try {
Thread.sleep(MANAGEMENT_TASK_PERIOD_IN_MILLIS);
} catch (InterruptedException e2) {
Thread.currentThread().interrupt();
return false;
}
} catch (Exception e) {
throw ExceptionUtil.rethrow(e);
}
}
return true;
}
@Override
public Operation getPreJoinOperation() {
if (!cpSubsystemEnabled) {
return null;
}
boolean master = nodeEngine.getClusterService().isMaster();
boolean discoveryCompleted = metadataGroupManager.isDiscoveryCompleted();
RaftGroupId metadataGroupId = metadataGroupManager.getMetadataGroupId();
return master ? new RaftServicePreJoinOp(discoveryCompleted, metadataGroupId) : null;
}
@Override
public void memberAdded(MembershipServiceEvent event) {
metadataGroupManager.broadcastActiveCPMembers();
updateMissingMembers();
}
@Override
public void memberRemoved(MembershipServiceEvent event) {
updateMissingMembers();
}
void updateMissingMembers() {
if (config.getMissingCPMemberAutoRemovalSeconds() == 0 || !metadataGroupManager.isDiscoveryCompleted()
|| (!isStartCompleted() && getCPPersistenceService().getCPMetadataStore().containsLocalMemberFile())) {
return;
}
// since only the Metadata CP group members keep the active CP member list,
// they will be the ones that keep track of missing CP members.
Collection activeMembers = metadataGroupManager.getActiveMembers();
missingMembers.keySet().retainAll(activeMembers);
ClusterService clusterService = nodeEngine.getClusterService();
for (CPMemberInfo cpMember : activeMembers) {
if (clusterService.getMember(cpMember.getAddress()) == null) {
if (missingMembers.putIfAbsent(cpMember, Clock.currentTimeMillis()) == null) {
logger.warning(cpMember + " is not present in the cluster. It will be auto-removed after "
+ config.getMissingCPMemberAutoRemovalSeconds() + " seconds.");
}
} else if (missingMembers.remove(cpMember) != null) {
logger.info(cpMember + " is removed from the missing members list as it is in the cluster.");
}
}
}
Collection getMissingMembers() {
return Collections.unmodifiableSet(missingMembers.keySet());
}
public MetadataRaftGroupManager getMetadataGroupManager() {
return metadataGroupManager;
}
public RaftInvocationManager getInvocationManager() {
return invocationManager;
}
public void handlePreVoteRequest(CPGroupId groupId, PreVoteRequest request, RaftEndpoint target) {
RaftNode node = getOrInitRaftNodeIfTargetLocalCPMember(groupId, request, target);
if (node != null) {
node.handlePreVoteRequest(request);
}
}
public void handlePreVoteResponse(CPGroupId groupId, PreVoteResponse response, RaftEndpoint target) {
RaftNode node = getOrInitRaftNodeIfTargetLocalCPMember(groupId, response, target);
if (node != null) {
node.handlePreVoteResponse(response);
}
}
public void handleVoteRequest(CPGroupId groupId, VoteRequest request, RaftEndpoint target) {
RaftNode node = getOrInitRaftNodeIfTargetLocalCPMember(groupId, request, target);
if (node != null) {
node.handleVoteRequest(request);
}
}
public void handleVoteResponse(CPGroupId groupId, VoteResponse response, RaftEndpoint target) {
RaftNode node = getOrInitRaftNodeIfTargetLocalCPMember(groupId, response, target);
if (node != null) {
node.handleVoteResponse(response);
}
}
public void handleAppendEntries(CPGroupId groupId, AppendRequest request, RaftEndpoint target) {
RaftNode node = getOrInitRaftNodeIfTargetLocalCPMember(groupId, request, target);
if (node != null) {
node.handleAppendRequest(request);
}
}
public void handleAppendResponse(CPGroupId groupId, AppendSuccessResponse response, RaftEndpoint target) {
RaftNode node = getOrInitRaftNodeIfTargetLocalCPMember(groupId, response, target);
if (node != null) {
node.handleAppendResponse(response);
}
}
public void handleAppendResponse(CPGroupId groupId, AppendFailureResponse response, RaftEndpoint target) {
RaftNode node = getOrInitRaftNodeIfTargetLocalCPMember(groupId, response, target);
if (node != null) {
node.handleAppendResponse(response);
}
}
public void handleSnapshot(CPGroupId groupId, InstallSnapshot request, RaftEndpoint target) {
RaftNode node = getOrInitRaftNodeIfTargetLocalCPMember(groupId, request, target);
if (node != null) {
node.handleInstallSnapshot(request);
}
}
public void handleTriggerLeaderElection(CPGroupId groupId, TriggerLeaderElection request, RaftEndpoint target) {
RaftNode node = getOrInitRaftNodeIfTargetLocalCPMember(groupId, request, target);
if (node != null) {
node.handleTriggerLeaderElection(request);
}
}
public Collection getAllRaftNodes() {
return new ArrayList<>(nodes.values());
}
public RaftNode getRaftNode(CPGroupId groupId) {
return nodes.get(groupId);
}
public RaftNode getOrInitRaftNode(CPGroupId groupId) {
RaftNode node = nodes.get(groupId);
if (node == null && isStartCompleted() && isDiscoveryCompleted() && !destroyedGroupIds.contains(groupId)
&& !terminatedRaftNodeGroupIds.contains(groupId)) {
logger.fine("RaftNode[" + groupId + "] does not exist. Asking to the METADATA CP group...");
nodeEngine.getExecutionService().execute(CP_SUBSYSTEM_EXECUTOR, new InitializeRaftNodeTask(groupId));
}
return node;
}
private RaftNode getOrInitRaftNodeIfTargetLocalCPMember(CPGroupId groupId, Object message, RaftEndpoint target) {
RaftNode node = getOrInitRaftNode(groupId);
if (node == null) {
if (logger.isFineEnabled()) {
logger.warning("RaftNode[" + groupId + "] does not exist to handle: " + message);
}
return null;
}
if (!target.equals(node.getLocalMember())) {
if (logger.isFineEnabled()) {
logger.warning("Won't handle " + message + ". We are not the expected target: " + target + ", local endpoint: "
+ node.getLocalMember());
}
return null;
}
return node;
}
boolean isStartCompleted() {
return nodeEngine.getNode().getNodeExtension().isStartCompleted();
}
public boolean isRaftGroupDestroyed(CPGroupId groupId) {
return destroyedGroupIds.contains(groupId);
}
public CPSubsystemConfig getConfig() {
return config;
}
public CPMemberInfo getLocalCPMember() {
return metadataGroupManager.getLocalCPMember();
}
public RaftEndpoint getLocalCPEndpoint() {
CPMemberInfo localCPMember = getLocalCPMember();
return localCPMember != null ? localCPMember.toRaftEndpoint() : null;
}
public void createRaftNode(CPGroupId groupId, Collection members) {
createRaftNode(groupId, members, getLocalCPEndpoint());
}
void createRaftNode(CPGroupId groupId, Collection members, RaftEndpoint localCPMember) {
/*
* WARNING:
* This method is acquiring a lock.
* Make sure that you don't call this method from a partition thread.
*/
assert !(Thread.currentThread() instanceof PartitionOperationThread)
: "Cannot create RaftNode of " + groupId + " in a partition thread!";
if (nodes.containsKey(groupId) || !isStartCompleted() || !hasSameSeed(groupId)) {
return;
}
if (getLocalCPMember() == null) {
logger.warning("Not creating Raft node for " + groupId + " because local CP member is not initialized yet.");
return;
}
nodeLock.readLock().lock();
try {
if (destroyedGroupIds.contains(groupId)) {
logger.warning("Not creating RaftNode[" + groupId + "] since the CP group is already destroyed.");
return;
} else if (terminatedRaftNodeGroupIds.contains(groupId)) {
if (!nodeEngine.isRunning()) {
logger.fine("Not creating RaftNode[" + groupId + "] since the local CP member is already terminated.");
return;
}
}
int partitionId = getCPGroupPartitionId(groupId);
RaftIntegration integration = new NodeEngineRaftIntegration(nodeEngine, groupId, localCPMember, partitionId);
RaftAlgorithmConfig raftAlgorithmConfig = config.getRaftAlgorithmConfig();
CPPersistenceService persistenceService = getCPPersistenceService();
RaftStateStore stateStore = persistenceService.createRaftStateStore((RaftGroupId) groupId, null);
RaftNodeImpl node = newRaftNode(groupId, localCPMember, members, raftAlgorithmConfig, integration, stateStore);
registerNodeMetrics(groupId);
if (nodes.putIfAbsent(groupId, node) == null) {
if (destroyedGroupIds.contains(groupId)) {
deregisterNodeMetrics(groupId);
nodes.remove(groupId, node);
logger.warning("Not creating RaftNode[" + groupId + "] since the CP group is already destroyed.");
return;
}
node.start();
logger.info("RaftNode[" + groupId + "] is created with " + members);
}
} finally {
nodeLock.readLock().unlock();
}
}
CPPersistenceService getCPPersistenceService() {
return nodeEngine.getNode().getNodeExtension().getCPPersistenceService();
}
public RaftNodeImpl restoreRaftNode(RaftGroupId groupId, RestoredRaftState restoredState, LogFileStructure logFileStructure) {
int partitionId = getCPGroupPartitionId(groupId);
RaftIntegration integration = new NodeEngineRaftIntegration(nodeEngine, groupId, restoredState.localEndpoint(),
partitionId);
RaftAlgorithmConfig raftAlgorithmConfig = config.getRaftAlgorithmConfig();
RaftStateStore stateStore = getCPPersistenceService().createRaftStateStore(groupId, logFileStructure);
RaftNodeImpl node = RaftNodeImpl.restoreRaftNode(groupId, restoredState, raftAlgorithmConfig, integration, stateStore);
// no need to lock here...
registerNodeMetrics(groupId);
RaftNode prev = nodes.putIfAbsent(groupId, node);
checkState(prev == null, "Could not restore " + groupId + " because its Raft node already exists!");
node.start();
logger.info("RaftNode[" + groupId + "] is restored.");
return node;
}
@Override
public void provideDynamicMetrics(MetricDescriptor descriptor,
MetricsCollectionContext context) {
MetricDescriptor rootDescriptor = descriptor.withPrefix(CP_PREFIX_RAFT_GROUP);
for (Entry entry : nodeMetrics.entrySet()) {
CPGroupId groupId = entry.getKey();
MetricDescriptor groupDescriptor = rootDescriptor
.copy()
.withDiscriminator(CP_DISCRIMINATOR_GROUPID, String.valueOf(groupId.getId()))
.withTag(CP_TAG_NAME, groupId.getName());
context.collect(groupDescriptor, entry.getValue());
}
}
private void registerNodeMetrics(CPGroupId groupId) {
nodeMetrics.putIfAbsent(groupId, new RaftNodeMetrics());
}
private void deregisterNodeMetrics(CPGroupId groupId) {
nodeMetrics.remove(groupId);
}
private boolean hasSameSeed(CPGroupId groupId) {
return getMetadataGroupId().getSeed() == ((RaftGroupId) groupId).getSeed();
}
public boolean updateInvocationManagerMembers(long groupIdSeed, long membersCommitIndex,
Collection members) {
return invocationManager.getRaftInvocationContext().setMembers(groupIdSeed, membersCommitIndex, members);
}
public void terminateRaftNode(CPGroupId groupId, boolean groupDestroyed) {
if (destroyedGroupIds.contains(groupId) || !hasSameSeed(groupId)) {
return;
}
assert !(Thread.currentThread() instanceof PartitionOperationThread)
: "Cannot terminate RaftNode of " + groupId + " in a partition thread!";
nodeLock.readLock().lock();
try {
if (destroyedGroupIds.contains(groupId)) {
return;
}
if (groupDestroyed) {
destroyedGroupIds.add(groupId);
}
terminatedRaftNodeGroupIds.add(groupId);
RaftNode node = nodes.get(groupId);
CPPersistenceService persistenceService = getCPPersistenceService();
if (node != null) {
destroyRaftNode(node, groupDestroyed);
logger.info("RaftNode[" + groupId + "] is destroyed.");
} else if (groupDestroyed && persistenceService.isEnabled()) {
persistenceService.removeRaftStateStore((RaftGroupId) groupId);
logger.info("RaftStateStore of RaftNode[" + groupId + "] is deleted.");
}
} finally {
nodeLock.readLock().unlock();
}
}
public void stepDownRaftNode(CPGroupId groupId) {
if (terminatedRaftNodeGroupIds.contains(groupId) || !hasSameSeed(groupId)) {
return;
}
assert !(Thread.currentThread() instanceof PartitionOperationThread)
: "Cannot step down RaftNode of " + groupId + " in a partition thread!";
nodeLock.readLock().lock();
try {
if (terminatedRaftNodeGroupIds.contains(groupId)) {
return;
}
CPPersistenceService persistenceService = getCPPersistenceService();
RaftNode node = nodes.get(groupId);
if (node != null && node.getStatus() == RaftNodeStatus.STEPPED_DOWN) {
terminatedRaftNodeGroupIds.add(groupId);
destroyRaftNode(node, true);
logger.fine("RaftNode[" + groupId + "] has stepped down.");
} else if (node == null && persistenceService.isEnabled()) {
persistenceService.removeRaftStateStore((RaftGroupId) groupId);
logger.info("RaftStateStore of RaftNode[" + groupId + "] is deleted.");
}
} finally {
nodeLock.readLock().unlock();
}
}
private void destroyRaftNode(RaftNode node, boolean removeRaftStateStore) {
RaftGroupId groupId = (RaftGroupId) node.getGroupId();
node.forceSetTerminatedStatus().whenCompleteAsync((v, t) -> {
nodes.remove(groupId, node);
deregisterNodeMetrics(groupId);
CPPersistenceService persistenceService = getCPPersistenceService();
try {
if (removeRaftStateStore && persistenceService.isEnabled()) {
persistenceService.removeRaftStateStore(groupId);
logger.info("RaftStateStore of RaftNode[" + groupId + "] is deleted.");
}
} catch (Exception e) {
logger.severe("Deletion of RaftStateStore of RaftNode[" + groupId + "] failed.", e);
}
});
}
public RaftGroupId createRaftGroupForProxy(String name) {
String groupName = getGroupNameForProxy(name);
if (cpSubsystemEnabled) {
try {
CPGroupSummary group = getGroupSummaryForProxy(groupName).joinInternal();
if (group != null) {
return (RaftGroupId) group.id();
}
return invocationManager.createRaftGroup(groupName).get();
} catch (InterruptedException e) {
Thread.currentThread().interrupt();
throw new IllegalStateException("Could not create CP group: " + groupName);
} catch (ExecutionException e) {
throw new IllegalStateException("Could not create CP group: " + groupName);
}
} else {
return createPartitionBasedRaftGroupId(name, groupName);
}
}
private RaftGroupId createPartitionBasedRaftGroupId(String name, String groupName) {
if (DEFAULT_GROUP_NAME.equals(groupName)) {
// In unsafe mode, if there's no group specified,
// we will use proxy name as group name.
groupName = name;
}
Data key = nodeEngine.getSerializationService().toData(groupName);
int partitionId = nodeEngine.getPartitionService().getPartitionId(key);
return new RaftGroupId(groupName, 0, partitionId);
}
public InternalCompletableFuture createRaftGroupForProxyAsync(String name) {
String groupName = getGroupNameForProxy(name);
if (cpSubsystemEnabled) {
InternalCompletableFuture future = newCompletableFuture();
InternalCompletableFuture groupIdFuture = getGroupSummaryForProxy(groupName);
groupIdFuture.whenCompleteAsync((response, throwable) -> {
if (throwable == null) {
if (response != null) {
future.complete(response.id());
} else {
invocationManager.createRaftGroup(groupName).whenCompleteAsync((r, t) -> {
complete(future, r, t);
});
}
} else {
complete(future, throwable);
}
});
return future;
} else {
return newCompletedFuture(createPartitionBasedRaftGroupId(name, groupName));
}
}
private InternalCompletableFuture getGroupSummaryForProxy(String groupName) {
return invocationManager.query(getMetadataGroupId(), new GetActiveRaftGroupByNameOp(groupName), LINEARIZABLE);
}
private InternalCompletableFuture invokeTriggerRemoveMember(CPMemberInfo member) {
return invocationManager.invoke(getMetadataGroupId(), new RemoveCPMemberOp(member));
}
private static InternalCompletableFuture complete(InternalCompletableFuture future, Throwable t) {
future.completeExceptionally(t);
return future;
}
private static void complete(InternalCompletableFuture future,
T value,
Throwable t) {
if (t == null) {
future.complete(value);
} else {
future.completeExceptionally(t);
}
}
public static String withoutDefaultGroupName(String name) {
name = name.trim();
int i = name.indexOf("@");
if (i == -1) {
return name;
}
checkTrue(name.indexOf("@", i + 1) == -1, "Custom group name must be specified at most once");
String groupName = name.substring(i + 1).trim();
if (groupName.equalsIgnoreCase(DEFAULT_GROUP_NAME)) {
return name.substring(0, i);
}
return name;
}
public static String getGroupNameForProxy(String name) {
name = name.trim();
int i = name.indexOf("@");
if (i == -1) {
return DEFAULT_GROUP_NAME;
}
checkTrue(i < (name.length() - 1), "Custom CP group name cannot be empty string");
checkTrue(name.indexOf("@", i + 1) == -1, "Custom group name must be specified at most once");
String groupName = name.substring(i + 1).trim();
checkTrue(groupName.length() > 0, "Custom CP group name cannot be empty string");
checkFalse(groupName.equalsIgnoreCase(METADATA_CP_GROUP_NAME), "CP data structures cannot run on the METADATA CP group!");
return groupName.equalsIgnoreCase(DEFAULT_GROUP_NAME) ? DEFAULT_GROUP_NAME : groupName;
}
public static String getObjectNameForProxy(String name) {
int i = name.indexOf("@");
if (i == -1) {
return name;
}
checkTrue(i < (name.length() - 1), "Object name cannot be empty string");
checkTrue(name.indexOf("@", i + 1) == -1,
"Custom CP group name must be specified at most once");
String objectName = name.substring(0, i).trim();
checkTrue(objectName.length() > 0, "Object name cannot be empty string");
return objectName;
}
public RaftGroupId getMetadataGroupId() {
return metadataGroupManager.getMetadataGroupId();
}
public boolean isCpSubsystemEnabled() {
return cpSubsystemEnabled;
}
@SuppressWarnings({"checkstyle:npathcomplexity", "checkstyle:cyclomaticcomplexity"})
public void handleActiveCPMembers(RaftGroupId receivedMetadataGroupId, long membersCommitIndex,
Collection members) {
if (!metadataGroupManager.isDiscoveryCompleted()) {
if (logger.isFineEnabled()) {
logger.fine("Ignoring received active CP members: " + members + " since discovery is in progress.");
}
return;
}
checkNotNull(members);
checkTrue(members.size() > 0, "Active CP members list cannot be empty");
if (members.size() == 1) {
logger.fine("There is one active CP member left: " + members);
return;
}
CPMemberInfo localMember = getLocalCPMember();
members = replaceLocalMemberIfAddressChanged(membersCommitIndex, members, localMember);
if (updateInvocationManagerMembers(receivedMetadataGroupId.getSeed(), membersCommitIndex, members)) {
if (logger.isFineEnabled()) {
logger.fine("Handled new active CP members list: " + members + ", members commit index: " + membersCommitIndex
+ ", METADATA group id seed: " + receivedMetadataGroupId.getSeed());
}
}
RaftGroupId metadataGroupId = getMetadataGroupId();
if (receivedMetadataGroupId.getSeed() < metadataGroupId.getSeed() || metadataGroupId.equals(receivedMetadataGroupId)) {
return;
}
if (!isStartCompleted()) {
if (!metadataGroupId.equals(receivedMetadataGroupId)) {
logger.severe("Restored METADATA groupId: " + metadataGroupId + " is different than received METADATA groupId: "
+ receivedMetadataGroupId + ". There must have been a CP Subsystem reset while this member was down...");
}
return;
}
if (getRaftNode(receivedMetadataGroupId) != null) {
if (logger.isFineEnabled()) {
logger.fine(localMember + " is already part of METADATA group but received active CP members!");
}
return;
}
if (!receivedMetadataGroupId.equals(metadataGroupId) && getRaftNode(metadataGroupId) != null) {
logger.warning(localMember + " was part of " + metadataGroupId + ", but received active CP members for "
+ receivedMetadataGroupId + ".");
return;
}
metadataGroupManager.handleMetadataGroupId(receivedMetadataGroupId);
}
@SuppressWarnings({"checkstyle:npathcomplexity", "checkstyle:cyclomaticcomplexity"})
private Collection replaceLocalMemberIfAddressChanged(long membersCommitIndex, Collection members,
CPMemberInfo localMember) {
if (localMember != null && !members.contains(localMember)) {
// If I am present in the received CP member list with another address, I replace my local member.
// In addition, I will remove any other member that has my address.
CPMemberInfo otherMember = null;
CPMemberInfo staleLocalMember = null;
for (CPMemberInfo m : members) {
if (m.getAddress().equals(localMember.getAddress()) && !m.getUuid().equals(localMember.getUuid())) {
otherMember = m;
} else if (!m.getAddress().equals(localMember.getAddress()) && m.getUuid().equals(localMember.getUuid())) {
staleLocalMember = m;
}
}
if (otherMember != null || staleLocalMember != null) {
members = new ArrayList<>(members);
members.remove(otherMember);
members.remove(staleLocalMember);
if (logger.isFineEnabled()) {
// prints null if there is no other member with the same address but it is ok in a debug log...
logger.fine("Removing other member: " + otherMember + " in received CP members list: " + members
+ " and commit index: " + membersCommitIndex);
}
}
if (staleLocalMember != null) {
members.add(localMember);
if (logger.isFineEnabled()) {
logger.fine("Replacing stale local member: " + staleLocalMember + " with: " + localMember
+ " in received CP members list: " + members + " and commit index: " + membersCommitIndex);
}
} else if (nodeEngine.getNode().isRunning()) {
boolean missingAutoRemovalEnabled = config.getMissingCPMemberAutoRemovalSeconds() > 0;
logger.severe("Local " + localMember + " is not part of received active CP members: " + members
+ ". It seems local member is removed from CP Subsystem. "
+ "Auto removal of missing members is " + (missingAutoRemovalEnabled ? "enabled." : "disabled."));
}
}
return members;
}
@Override
public void onRaftNodeTerminated(CPGroupId groupId) {
nodeEngine.getExecutionService().execute(CP_SUBSYSTEM_EXECUTOR, () -> terminateRaftNode(groupId, false));
}
@Override
public void onRaftNodeSteppedDown(CPGroupId groupId) {
nodeEngine.getExecutionService().execute(CP_SUBSYSTEM_EXECUTOR, () -> stepDownRaftNode(groupId));
}
public Collection getLeadedGroups() {
Collection groupIds = new ArrayList<>();
RaftEndpoint localEndpoint = getLocalCPEndpoint();
for (RaftNode raftNode : nodes.values()) {
if (CPGroup.METADATA_CP_GROUP_NAME.equals(raftNode.getGroupId().getName())) {
// Ignore metadata group
continue;
}
RaftEndpoint leader = raftNode.getLeader();
if (leader != null && leader.equals(localEndpoint)) {
groupIds.add(raftNode.getGroupId());
}
}
return groupIds;
}
public InternalCompletableFuture transferLeadership(CPGroupId groupId, CPMemberInfo destination) {
RaftNode raftNode = getRaftNode(groupId);
if (raftNode == null) {
throw new IllegalStateException("RaftNode does not exist for group: " + groupId);
}
return raftNode.transferLeadership(destination.toRaftEndpoint());
}
public int getCPGroupPartitionId(CPGroupId groupId) {
int partitionCount = nodeEngine.getPartitionService().getPartitionCount();
return getCPGroupPartitionId(groupId, partitionCount);
}
public static int getCPGroupPartitionId(CPGroupId groupId, int partitionCount) {
assert groupId.getId() >= 0 : "Invalid groupId: " + groupId;
return (int) (groupId.getId() % partitionCount);
}
public long nextUnsafeModeCommitIndex(CPGroupId groupId) {
assert !cpSubsystemEnabled;
int partitionId = getCPGroupPartitionId(groupId);
UnsafeModePartitionState unsafeModeState = unsafeModeStates[partitionId];
return unsafeModeState.nextCommitIndex();
}
public void registerUnsafeWaitingOperation(CPGroupId groupId, long commitIndex, Operation op) {
assert !cpSubsystemEnabled;
int partitionId = getCPGroupPartitionId(groupId);
UnsafeModePartitionState unsafeModeState = unsafeModeStates[partitionId];
if (!unsafeModeState.registerWaitingOp(commitIndex, op)) {
throw new IllegalArgumentException("Cannot register " + op + " with index " + commitIndex);
}
}
/**
* Completes all futures registered with {@code indices}
* in the CP group associated with {@code groupId}.
*
* @return {@code true} if the CP group exists, {@code false} otherwise.
*/
public boolean completeFutures(CPGroupId groupId, Collection indices, Object result) {
if (cpSubsystemEnabled) {
RaftNodeImpl raftNode = (RaftNodeImpl) getRaftNode(groupId);
if (raftNode == null) {
return false;
}
for (Long index : indices) {
raftNode.completeFuture(index, result);
}
} else {
int partitionId = getCPGroupPartitionId(groupId);
UnsafeModePartitionState unsafeModeState = unsafeModeStates[partitionId];
for (Long index : indices) {
Operation op = unsafeModeState.removeWaitingOp(index);
sendOperationResponse(op, result);
}
}
return true;
}
/**
* Completes all futures registered with {@code indices}
* in the CP group associated with {@code groupId}.
*
* @return {@code true} if the CP group exists, {@code false} otherwise.
*/
public boolean completeFutures(CPGroupId groupId, Collection> results) {
if (cpSubsystemEnabled) {
RaftNodeImpl raftNode = (RaftNodeImpl) getRaftNode(groupId);
if (raftNode == null) {
return false;
}
for (Entry result : results) {
raftNode.completeFuture(result.getKey(), result.getValue());
}
} else {
int partitionId = getCPGroupPartitionId(groupId);
UnsafeModePartitionState unsafeModeState = unsafeModeStates[partitionId];
for (Entry result : results) {
Operation op = unsafeModeState.removeWaitingOp(result.getKey());
sendOperationResponse(op, result.getValue());
}
}
return true;
}
private void sendOperationResponse(Operation op, Object result) {
if (op != null) {
try {
op.sendResponse(result);
} catch (ResponseAlreadySentException e) {
op.logError(e);
}
}
}
@Override
public Operation prepareReplicationOperation(PartitionReplicationEvent event) {
if (cpSubsystemEnabled) {
return null;
}
if (event.getReplicaIndex() > getBackupCount()) {
return null;
}
UnsafeModePartitionState state = unsafeModeStates[event.getPartitionId()];
return state.commitIndex() == 0 ? null : new UnsafeStateReplicationOp(state);
}
@Override
public void beforeMigration(PartitionMigrationEvent event) {
}
@Override
public void commitMigration(PartitionMigrationEvent event) {
if (cpSubsystemEnabled) {
return;
}
if (event.getMigrationEndpoint() == MigrationEndpoint.SOURCE) {
UnsafeModePartitionState state = unsafeModeStates[event.getPartitionId()];
if (event.getCurrentReplicaIndex() == 0) {
// Waiting operations are registered only on primary.
Object ex = new PartitionMigratingException("Partition " + event.getPartitionId() + " is migrating!");
for (Operation op : state.getWaitingOps()) {
op.sendResponse(ex);
}
}
int thresholdReplicaIndex = event.getNewReplicaIndex();
if (thresholdReplicaIndex == -1 || thresholdReplicaIndex > getBackupCount()) {
state.reset();
}
}
}
@Override
public void rollbackMigration(PartitionMigrationEvent event) {
if (cpSubsystemEnabled) {
return;
}
if (event.getMigrationEndpoint() == MigrationEndpoint.DESTINATION) {
int thresholdReplicaIndex = event.getCurrentReplicaIndex();
if (thresholdReplicaIndex == -1 || thresholdReplicaIndex > getBackupCount()) {
unsafeModeStates[event.getPartitionId()].reset();
}
}
}
private int getBackupCount() {
return 1;
}
public void applyUnsafeModeState(int partitionId, UnsafeModePartitionState state) {
assert !cpSubsystemEnabled;
unsafeModeStates[partitionId].apply(state);
}
private class InitializeRaftNodeTask implements Runnable {
private final CPGroupId groupId;
InitializeRaftNodeTask(CPGroupId groupId) {
this.groupId = groupId;
}
@Override
public void run() {
queryInitialMembersFromMetadataRaftGroup();
}
private void queryInitialMembersFromMetadataRaftGroup() {
RaftOp op = new GetRaftGroupOp(groupId);
InternalCompletableFuture f = invocationManager.query(getMetadataGroupId(), op, LEADER_LOCAL);
f.whenCompleteAsync((group, throwable) -> {
if (throwable == null) {
if (group != null) {
if (group.members().contains(getLocalCPMember())) {
createRaftNode(groupId, group.initialMembers());
} else {
// I can be the member that is just added to the raft group...
queryInitialMembersFromTargetRaftGroup();
}
} else if (logger.isFineEnabled()) {
logger.fine("Cannot get initial members of " + groupId + " from the METADATA CP group");
}
} else {
if (throwable instanceof CPGroupDestroyedException) {
CPGroupId destroyedGroupId = ((CPGroupDestroyedException) throwable).getGroupId();
terminateRaftNode(destroyedGroupId, true);
}
if (logger.isFineEnabled()) {
logger.fine("Cannot get initial members of " + groupId + " from the METADATA CP group", throwable);
}
}
});
}
void queryInitialMembersFromTargetRaftGroup() {
RaftEndpoint localEndpoint = getLocalCPEndpoint();
if (localEndpoint == null) {
return;
}
RaftOp op = new GetInitialRaftGroupMembersIfCurrentGroupMemberOp(localEndpoint);
InternalCompletableFuture> f = invocationManager.query(groupId, op, LEADER_LOCAL);
f.whenCompleteAsync((initialMembers, t) -> {
if (t == null) {
createRaftNode(groupId, initialMembers);
} else {
if (logger.isFineEnabled()) {
logger.fine("Cannot get initial members of " + groupId + " from the CP group itself", t);
}
}
});
}
}
private class AutoRemoveMissingCPMemberTask implements Runnable {
@Override
public void run() {
try {
if (!metadataGroupManager.isMetadataGroupLeader() || metadataGroupManager.getMembershipChangeSchedule() != null) {
return;
}
for (Entry e : missingMembers.entrySet()) {
long missingTimeSeconds = MILLISECONDS.toSeconds(System.currentTimeMillis() - e.getValue());
if (missingTimeSeconds >= config.getMissingCPMemberAutoRemovalSeconds()) {
CPMemberInfo missingMember = e.getKey();
logger.warning("Removing " + missingMember + " since it is absent for " + missingTimeSeconds
+ " seconds.");
removeCPMember(missingMember.getUuid()).get();
logger.info("Auto-removal of " + missingMember + " is successful.");
return;
}
}
} catch (Exception e) {
logger.severe("RemoveMissingMembersTask failed", e);
}
}
}
private class PublishNodeMetricsTask implements Runnable {
@Override
public void run() {
for (RaftNode node : nodes.values()) {
final RaftNodeImpl raftNode = (RaftNodeImpl) node;
final RaftNodeMetrics metrics = nodeMetrics.get(node.getGroupId());
assert metrics != null;
raftNode.execute(() -> {
RaftState state = raftNode.state();
RaftLog log = state.log();
metrics.update(state.term(), state.commitIndex(), state.lastApplied(),
log.lastLogOrSnapshotTerm(), log.snapshotIndex(),
log.lastLogOrSnapshotIndex(), log.availableCapacity());
});
}
}
}
}