org.apache.cassandra.service.StorageService Maven / Gradle / Ivy
Go to download
Show more of this group Show more artifacts with this name
Show all versions of cassandra-all Show documentation
Show all versions of cassandra-all Show documentation
A fork of the Apache Cassandra Project ready to embed Elasticsearch.
/*
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
* regarding copyright ownership. The ASF licenses this file
* to you 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 org.apache.cassandra.service;
import java.io.*;
import java.lang.management.ManagementFactory;
import java.net.InetAddress;
import java.net.UnknownHostException;
import java.nio.ByteBuffer;
import java.util.*;
import java.util.Map.Entry;
import java.util.concurrent.*;
import java.util.concurrent.atomic.AtomicBoolean;
import java.util.concurrent.atomic.AtomicInteger;
import java.util.regex.MatchResult;
import java.util.regex.Pattern;
import java.util.stream.StreamSupport;
import javax.annotation.Nullable;
import javax.management.*;
import javax.management.openmbean.TabularData;
import javax.management.openmbean.TabularDataSupport;
import com.google.common.annotations.VisibleForTesting;
import com.google.common.base.Predicate;
import com.google.common.collect.*;
import com.google.common.util.concurrent.*;
import org.apache.commons.lang3.StringUtils;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
import org.apache.cassandra.auth.AuthKeyspace;
import org.apache.cassandra.auth.AuthMigrationListener;
import org.apache.cassandra.batchlog.BatchRemoveVerbHandler;
import org.apache.cassandra.batchlog.BatchStoreVerbHandler;
import org.apache.cassandra.batchlog.BatchlogManager;
import org.apache.cassandra.concurrent.NamedThreadFactory;
import org.apache.cassandra.concurrent.ScheduledExecutors;
import org.apache.cassandra.concurrent.Stage;
import org.apache.cassandra.concurrent.StageManager;
import org.apache.cassandra.config.CFMetaData;
import org.apache.cassandra.config.DatabaseDescriptor;
import org.apache.cassandra.config.Schema;
import org.apache.cassandra.config.SchemaConstants;
import org.apache.cassandra.config.ViewDefinition;
import org.apache.cassandra.db.*;
import org.apache.cassandra.db.commitlog.CommitLog;
import org.apache.cassandra.db.compaction.CompactionManager;
import org.apache.cassandra.db.lifecycle.LifecycleTransaction;
import org.apache.cassandra.dht.*;
import org.apache.cassandra.dht.Range;
import org.apache.cassandra.dht.Token.TokenFactory;
import org.apache.cassandra.exceptions.*;
import org.apache.cassandra.gms.*;
import org.apache.cassandra.hints.HintVerbHandler;
import org.apache.cassandra.hints.HintsService;
import org.apache.cassandra.io.sstable.SSTableLoader;
import org.apache.cassandra.io.util.FileUtils;
import org.apache.cassandra.locator.*;
import org.apache.cassandra.metrics.StorageMetrics;
import org.apache.cassandra.net.*;
import org.apache.cassandra.repair.*;
import org.apache.cassandra.repair.messages.RepairOption;
import org.apache.cassandra.schema.CompactionParams.TombstoneOption;
import org.apache.cassandra.schema.KeyspaceMetadata;
import org.apache.cassandra.schema.SchemaKeyspace;
import org.apache.cassandra.service.paxos.CommitVerbHandler;
import org.apache.cassandra.service.paxos.PrepareVerbHandler;
import org.apache.cassandra.service.paxos.ProposeVerbHandler;
import org.apache.cassandra.streaming.*;
import org.apache.cassandra.thrift.EndpointDetails;
import org.apache.cassandra.thrift.TokenRange;
import org.apache.cassandra.thrift.cassandraConstants;
import org.apache.cassandra.tracing.TraceKeyspace;
import org.apache.cassandra.transport.ProtocolVersion;
import org.apache.cassandra.utils.*;
import org.apache.cassandra.utils.logging.LoggingSupportFactory;
import org.apache.cassandra.utils.progress.ProgressEvent;
import org.apache.cassandra.utils.progress.ProgressEventType;
import org.apache.cassandra.utils.progress.jmx.JMXProgressSupport;
import org.apache.cassandra.utils.progress.jmx.LegacyJMXProgressSupport;
import static java.util.Arrays.asList;
import static java.util.stream.Collectors.toList;
import static org.apache.cassandra.index.SecondaryIndexManager.getIndexName;
import static org.apache.cassandra.index.SecondaryIndexManager.isIndexColumnFamily;
/**
* This abstraction contains the token/identifier of this node
* on the identifier space. This token gets gossiped around.
* This class will also maintain histograms of the load information
* of other nodes in the cluster.
*/
public class StorageService extends NotificationBroadcasterSupport implements IEndpointStateChangeSubscriber, StorageServiceMBean
{
private static final Logger logger = LoggerFactory.getLogger(StorageService.class);
public static final int RING_DELAY = getRingDelay(); // delay after which we assume ring has stablized
private final JMXProgressSupport progressSupport = new JMXProgressSupport(this);
/**
* @deprecated backward support to previous notification interface
* Will be removed on 4.0
*/
@Deprecated
private final LegacyJMXProgressSupport legacyProgressSupport;
private static final AtomicInteger threadCounter = new AtomicInteger(1);
private static int getRingDelay()
{
String newdelay = System.getProperty("cassandra.ring_delay_ms");
if (newdelay != null)
{
logger.info("Overriding RING_DELAY to {}ms", newdelay);
return Integer.parseInt(newdelay);
}
else
return 30 * 1000;
}
/* This abstraction maintains the token/endpoint metadata information */
private TokenMetadata tokenMetadata = new TokenMetadata();
public volatile VersionedValue.VersionedValueFactory valueFactory = new VersionedValue.VersionedValueFactory(tokenMetadata.partitioner);
private Thread drainOnShutdown = null;
private volatile boolean isShutdown = false;
private final List preShutdownHooks = new ArrayList<>();
private final List postShutdownHooks = new ArrayList<>();
public static final StorageService instance = new StorageService();
@Deprecated
public boolean isInShutdownHook()
{
return isShutdown();
}
public boolean isShutdown()
{
return isShutdown;
}
public Collection> getLocalRanges(String keyspaceName)
{
return getRangesForEndpoint(keyspaceName, FBUtilities.getBroadcastAddress());
}
public Collection> getPrimaryRanges(String keyspace)
{
return getPrimaryRangesForEndpoint(keyspace, FBUtilities.getBroadcastAddress());
}
public Collection> getPrimaryRangesWithinDC(String keyspace)
{
return getPrimaryRangeForEndpointWithinDC(keyspace, FBUtilities.getBroadcastAddress());
}
private final Set replicatingNodes = Collections.synchronizedSet(new HashSet());
private CassandraDaemon daemon;
private InetAddress removingNode;
/* Are we starting this node in bootstrap mode? */
private volatile boolean isBootstrapMode;
/* we bootstrap but do NOT join the ring unless told to do so */
private boolean isSurveyMode = Boolean.parseBoolean(System.getProperty("cassandra.write_survey", "false"));
/* true if node is rebuilding and receiving data */
private final AtomicBoolean isRebuilding = new AtomicBoolean();
private final AtomicBoolean isDecommissioning = new AtomicBoolean();
private volatile boolean initialized = false;
private volatile boolean joined = false;
private volatile boolean gossipActive = false;
private final AtomicBoolean authSetupCalled = new AtomicBoolean(false);
private volatile boolean authSetupComplete = false;
/* the probability for tracing any particular request, 0 disables tracing and 1 enables for all */
private double traceProbability = 0.0;
private static enum Mode { STARTING, NORMAL, JOINING, LEAVING, DECOMMISSIONED, MOVING, DRAINING, DRAINED }
private volatile Mode operationMode = Mode.STARTING;
/* Used for tracking drain progress */
private volatile int totalCFs, remainingCFs;
private static final AtomicInteger nextRepairCommand = new AtomicInteger();
private final List lifecycleSubscribers = new CopyOnWriteArrayList<>();
private final ObjectName jmxObjectName;
private Collection bootstrapTokens = null;
// true when keeping strict consistency while bootstrapping
private static final boolean useStrictConsistency = Boolean.parseBoolean(System.getProperty("cassandra.consistent.rangemovement", "true"));
private static final boolean allowSimultaneousMoves = Boolean.parseBoolean(System.getProperty("cassandra.consistent.simultaneousmoves.allow","false"));
private static final boolean joinRing = Boolean.parseBoolean(System.getProperty("cassandra.join_ring", "true"));
private boolean replacing;
private final StreamStateStore streamStateStore = new StreamStateStore();
/** This method updates the local token on disk */
public void setTokens(Collection tokens)
{
assert tokens != null && !tokens.isEmpty() : "Node needs at least one token.";
if (logger.isDebugEnabled())
logger.debug("Setting tokens to {}", tokens);
SystemKeyspace.updateTokens(tokens);
Collection localTokens = getLocalTokens();
setGossipTokens(localTokens);
tokenMetadata.updateNormalTokens(tokens, FBUtilities.getBroadcastAddress());
setMode(Mode.NORMAL, false);
}
public void setGossipTokens(Collection tokens)
{
List> states = new ArrayList>();
states.add(Pair.create(ApplicationState.TOKENS, valueFactory.tokens(tokens)));
states.add(Pair.create(ApplicationState.STATUS, valueFactory.normal(tokens)));
Gossiper.instance.addLocalApplicationStates(states);
}
public StorageService()
{
// use dedicated executor for sending JMX notifications
super(Executors.newSingleThreadExecutor());
MBeanServer mbs = ManagementFactory.getPlatformMBeanServer();
try
{
jmxObjectName = new ObjectName("org.apache.cassandra.db:type=StorageService");
mbs.registerMBean(this, jmxObjectName);
mbs.registerMBean(StreamManager.instance, new ObjectName(StreamManager.OBJECT_NAME));
}
catch (Exception e)
{
throw new RuntimeException(e);
}
legacyProgressSupport = new LegacyJMXProgressSupport(this, jmxObjectName);
/* register the verb handlers */
MessagingService.instance().registerVerbHandlers(MessagingService.Verb.MUTATION, new MutationVerbHandler());
MessagingService.instance().registerVerbHandlers(MessagingService.Verb.READ_REPAIR, new ReadRepairVerbHandler());
MessagingService.instance().registerVerbHandlers(MessagingService.Verb.READ, new ReadCommandVerbHandler());
MessagingService.instance().registerVerbHandlers(MessagingService.Verb.RANGE_SLICE, new RangeSliceVerbHandler());
MessagingService.instance().registerVerbHandlers(MessagingService.Verb.PAGED_RANGE, new RangeSliceVerbHandler());
MessagingService.instance().registerVerbHandlers(MessagingService.Verb.COUNTER_MUTATION, new CounterMutationVerbHandler());
MessagingService.instance().registerVerbHandlers(MessagingService.Verb.TRUNCATE, new TruncateVerbHandler());
MessagingService.instance().registerVerbHandlers(MessagingService.Verb.PAXOS_PREPARE, new PrepareVerbHandler());
MessagingService.instance().registerVerbHandlers(MessagingService.Verb.PAXOS_PROPOSE, new ProposeVerbHandler());
MessagingService.instance().registerVerbHandlers(MessagingService.Verb.PAXOS_COMMIT, new CommitVerbHandler());
MessagingService.instance().registerVerbHandlers(MessagingService.Verb.HINT, new HintVerbHandler());
// see BootStrapper for a summary of how the bootstrap verbs interact
MessagingService.instance().registerVerbHandlers(MessagingService.Verb.REPLICATION_FINISHED, new ReplicationFinishedVerbHandler());
MessagingService.instance().registerVerbHandlers(MessagingService.Verb.REQUEST_RESPONSE, new ResponseVerbHandler());
MessagingService.instance().registerVerbHandlers(MessagingService.Verb.INTERNAL_RESPONSE, new ResponseVerbHandler());
MessagingService.instance().registerVerbHandlers(MessagingService.Verb.REPAIR_MESSAGE, new RepairMessageVerbHandler());
MessagingService.instance().registerVerbHandlers(MessagingService.Verb.GOSSIP_SHUTDOWN, new GossipShutdownVerbHandler());
MessagingService.instance().registerVerbHandlers(MessagingService.Verb.GOSSIP_DIGEST_SYN, new GossipDigestSynVerbHandler());
MessagingService.instance().registerVerbHandlers(MessagingService.Verb.GOSSIP_DIGEST_ACK, new GossipDigestAckVerbHandler());
MessagingService.instance().registerVerbHandlers(MessagingService.Verb.GOSSIP_DIGEST_ACK2, new GossipDigestAck2VerbHandler());
MessagingService.instance().registerVerbHandlers(MessagingService.Verb.DEFINITIONS_UPDATE, new DefinitionsUpdateVerbHandler());
MessagingService.instance().registerVerbHandlers(MessagingService.Verb.SCHEMA_CHECK, new SchemaCheckVerbHandler());
MessagingService.instance().registerVerbHandlers(MessagingService.Verb.MIGRATION_REQUEST, new MigrationRequestVerbHandler());
MessagingService.instance().registerVerbHandlers(MessagingService.Verb.SNAPSHOT, new SnapshotVerbHandler());
MessagingService.instance().registerVerbHandlers(MessagingService.Verb.ECHO, new EchoVerbHandler());
MessagingService.instance().registerVerbHandlers(MessagingService.Verb.BATCH_STORE, new BatchStoreVerbHandler());
MessagingService.instance().registerVerbHandlers(MessagingService.Verb.BATCH_REMOVE, new BatchRemoveVerbHandler());
}
public void registerDaemon(CassandraDaemon daemon)
{
this.daemon = daemon;
}
public void register(IEndpointLifecycleSubscriber subscriber)
{
lifecycleSubscribers.add(subscriber);
}
public void unregister(IEndpointLifecycleSubscriber subscriber)
{
lifecycleSubscribers.remove(subscriber);
}
// should only be called via JMX
public void stopGossiping()
{
if (gossipActive)
{
logger.warn("Stopping gossip by operator request");
Gossiper.instance.stop();
gossipActive = false;
}
}
// should only be called via JMX
public synchronized void startGossiping()
{
if (!gossipActive)
{
checkServiceAllowedToStart("gossip");
logger.warn("Starting gossip by operator request");
Collection tokens = SystemKeyspace.getSavedTokens();
boolean validTokens = tokens != null && !tokens.isEmpty();
// shouldn't be called before these are set if we intend to join the ring/are in the process of doing so
if (joined || joinRing)
assert validTokens : "Cannot start gossiping for a node intended to join without valid tokens";
if (validTokens)
setGossipTokens(tokens);
Gossiper.instance.forceNewerGeneration();
Gossiper.instance.start((int) (System.currentTimeMillis() / 1000));
gossipActive = true;
}
}
// should only be called via JMX
public boolean isGossipRunning()
{
return Gossiper.instance.isEnabled();
}
// should only be called via JMX
public synchronized void startRPCServer()
{
checkServiceAllowedToStart("thrift");
if (daemon == null)
{
throw new IllegalStateException("No configured daemon");
}
daemon.thriftServer.start();
}
public void stopRPCServer()
{
if (daemon == null)
{
throw new IllegalStateException("No configured daemon");
}
if (daemon.thriftServer != null)
daemon.thriftServer.stop();
}
public boolean isRPCServerRunning()
{
if ((daemon == null) || (daemon.thriftServer == null))
{
return false;
}
return daemon.thriftServer.isRunning();
}
public synchronized void startNativeTransport()
{
checkServiceAllowedToStart("native transport");
if (daemon == null)
{
throw new IllegalStateException("No configured daemon");
}
try
{
daemon.startNativeTransport();
}
catch (Exception e)
{
throw new RuntimeException("Error starting native transport: " + e.getMessage());
}
}
public void stopNativeTransport()
{
if (daemon == null)
{
throw new IllegalStateException("No configured daemon");
}
daemon.stopNativeTransport();
}
public boolean isNativeTransportRunning()
{
if (daemon == null)
{
return false;
}
return daemon.isNativeTransportRunning();
}
public void stopTransports()
{
if (isGossipActive())
{
logger.error("Stopping gossiper");
stopGossiping();
}
if (isRPCServerRunning())
{
logger.error("Stopping RPC server");
stopRPCServer();
}
if (isNativeTransportRunning())
{
logger.error("Stopping native transport");
stopNativeTransport();
}
}
/**
* Set the Gossip flag RPC_READY to false and then
* shutdown the client services (thrift and CQL).
*
* Note that other nodes will do this for us when
* they get the Gossip shutdown message, so even if
* we don't get time to broadcast this, it is not a problem.
*
* See {@link Gossiper#markAsShutdown(InetAddress)}
*/
private void shutdownClientServers()
{
setRpcReady(false);
stopRPCServer();
stopNativeTransport();
}
public void stopClient()
{
Gossiper.instance.unregister(this);
Gossiper.instance.stop();
MessagingService.instance().shutdown();
// give it a second so that task accepted before the MessagingService shutdown gets submitted to the stage (to avoid RejectedExecutionException)
Uninterruptibles.sleepUninterruptibly(1, TimeUnit.SECONDS);
StageManager.shutdownNow();
}
public boolean isInitialized()
{
return initialized;
}
public boolean isGossipActive()
{
return gossipActive;
}
public boolean isDaemonSetupCompleted()
{
return daemon == null
? false
: daemon.setupCompleted();
}
public void stopDaemon()
{
if (daemon == null)
throw new IllegalStateException("No configured daemon");
daemon.deactivate();
}
private synchronized UUID prepareForReplacement() throws ConfigurationException
{
if (SystemKeyspace.bootstrapComplete())
throw new RuntimeException("Cannot replace address with a node that is already bootstrapped");
if (!joinRing)
throw new ConfigurationException("Cannot set both join_ring=false and attempt to replace a node");
if (!DatabaseDescriptor.isAutoBootstrap() && !Boolean.getBoolean("cassandra.allow_unsafe_replace"))
throw new RuntimeException("Replacing a node without bootstrapping risks invalidating consistency " +
"guarantees as the expected data may not be present until repair is run. " +
"To perform this operation, please restart with " +
"-Dcassandra.allow_unsafe_replace=true");
InetAddress replaceAddress = DatabaseDescriptor.getReplaceAddress();
logger.info("Gathering node replacement information for {}", replaceAddress);
Map epStates = Gossiper.instance.doShadowRound();
// as we've completed the shadow round of gossip, we should be able to find the node we're replacing
if (epStates.get(replaceAddress) == null)
throw new RuntimeException(String.format("Cannot replace_address %s because it doesn't exist in gossip", replaceAddress));
try
{
VersionedValue tokensVersionedValue = epStates.get(replaceAddress).getApplicationState(ApplicationState.TOKENS);
if (tokensVersionedValue == null)
throw new RuntimeException(String.format("Could not find tokens for %s to replace", replaceAddress));
bootstrapTokens = TokenSerializer.deserialize(tokenMetadata.partitioner, new DataInputStream(new ByteArrayInputStream(tokensVersionedValue.toBytes())));
}
catch (IOException e)
{
throw new RuntimeException(e);
}
UUID localHostId = SystemKeyspace.getLocalHostId();
if (isReplacingSameAddress())
{
localHostId = Gossiper.instance.getHostId(replaceAddress, epStates);
SystemKeyspace.setLocalHostId(localHostId); // use the replacee's host Id as our own so we receive hints, etc
}
return localHostId;
}
private synchronized void checkForEndpointCollision(UUID localHostId, Set peers) throws ConfigurationException
{
if (Boolean.getBoolean("cassandra.allow_unsafe_join"))
{
logger.warn("Skipping endpoint collision check as cassandra.allow_unsafe_join=true");
return;
}
logger.debug("Starting shadow gossip round to check for endpoint collision");
Map epStates = Gossiper.instance.doShadowRound(peers);
if (epStates.isEmpty() && DatabaseDescriptor.getSeeds().contains(FBUtilities.getBroadcastAddress()))
logger.info("Unable to gossip with any peers but continuing anyway since node is in its own seed list");
// If bootstrapping, check whether any previously known status for the endpoint makes it unsafe to do so.
// If not bootstrapping, compare the host id for this endpoint learned from gossip (if any) with the local
// one, which was either read from system.local or generated at startup. If a learned id is present &
// doesn't match the local, then the node needs replacing
if (!Gossiper.instance.isSafeForStartup(FBUtilities.getBroadcastAddress(), localHostId, shouldBootstrap(), epStates))
{
throw new RuntimeException(String.format("A node with address %s already exists, cancelling join. " +
"Use cassandra.replace_address if you want to replace this node.",
FBUtilities.getBroadcastAddress()));
}
if (shouldBootstrap() && useStrictConsistency && !allowSimultaneousMoves())
{
for (Map.Entry entry : epStates.entrySet())
{
// ignore local node or empty status
if (entry.getKey().equals(FBUtilities.getBroadcastAddress()) || entry.getValue().getApplicationState(ApplicationState.STATUS) == null)
continue;
String[] pieces = splitValue(entry.getValue().getApplicationState(ApplicationState.STATUS));
assert (pieces.length > 0);
String state = pieces[0];
if (state.equals(VersionedValue.STATUS_BOOTSTRAPPING) || state.equals(VersionedValue.STATUS_LEAVING) || state.equals(VersionedValue.STATUS_MOVING))
throw new UnsupportedOperationException("Other bootstrapping/leaving/moving nodes detected, cannot bootstrap while cassandra.consistent.rangemovement is true");
}
}
}
private boolean allowSimultaneousMoves()
{
return allowSimultaneousMoves && DatabaseDescriptor.getNumTokens() == 1;
}
// for testing only
public void unsafeInitialize() throws ConfigurationException
{
initialized = true;
gossipActive = true;
Gossiper.instance.register(this);
Gossiper.instance.start((int) (System.currentTimeMillis() / 1000)); // needed for node-ring gathering.
Gossiper.instance.addLocalApplicationState(ApplicationState.NET_VERSION, valueFactory.networkVersion());
if (!MessagingService.instance().isListening())
MessagingService.instance().listen();
}
public void populateTokenMetadata()
{
if (Boolean.parseBoolean(System.getProperty("cassandra.load_ring_state", "true")))
{
logger.info("Populating token metadata from system tables");
Multimap loadedTokens = SystemKeyspace.loadTokens();
if (!shouldBootstrap()) // if we have not completed bootstrapping, we should not add ourselves as a normal token
loadedTokens.putAll(FBUtilities.getBroadcastAddress(), SystemKeyspace.getSavedTokens());
for (InetAddress ep : loadedTokens.keySet())
tokenMetadata.updateNormalTokens(loadedTokens.get(ep), ep);
logger.info("Token metadata: {}", tokenMetadata);
}
}
public synchronized void initServer() throws ConfigurationException
{
initServer(RING_DELAY);
}
public synchronized void initServer(int delay) throws ConfigurationException
{
logger.info("Cassandra version: {}", FBUtilities.getReleaseVersionString());
logger.info("Thrift API version: {}", cassandraConstants.VERSION);
logger.info("CQL supported versions: {} (default: {})",
StringUtils.join(ClientState.getCQLSupportedVersion(), ", "), ClientState.DEFAULT_CQL_VERSION);
logger.info("Native protocol supported versions: {} (default: {})",
StringUtils.join(ProtocolVersion.supportedVersions(), ", "), ProtocolVersion.CURRENT);
try
{
// Ensure StorageProxy is initialized on start-up; see CASSANDRA-3797.
Class.forName("org.apache.cassandra.service.StorageProxy");
// also IndexSummaryManager, which is otherwise unreferenced
Class.forName("org.apache.cassandra.io.sstable.IndexSummaryManager");
}
catch (ClassNotFoundException e)
{
throw new AssertionError(e);
}
// daemon threads, like our executors', continue to run while shutdown hooks are invoked
drainOnShutdown = NamedThreadFactory.createThread(new WrappedRunnable()
{
@Override
public void runMayThrow() throws InterruptedException, ExecutionException, IOException
{
drain(true);
if (FBUtilities.isWindows)
WindowsTimer.endTimerPeriod(DatabaseDescriptor.getWindowsTimerInterval());
LoggingSupportFactory.getLoggingSupport().onShutdown();
}
}, "StorageServiceShutdownHook");
Runtime.getRuntime().addShutdownHook(drainOnShutdown);
replacing = isReplacing();
if (!Boolean.parseBoolean(System.getProperty("cassandra.start_gossip", "true")))
{
logger.info("Not starting gossip as requested.");
// load ring state in preparation for starting gossip later
loadRingState();
initialized = true;
return;
}
prepareToJoin();
// Has to be called after the host id has potentially changed in prepareToJoin().
try
{
CacheService.instance.counterCache.loadSavedAsync().get();
}
catch (Throwable t)
{
JVMStabilityInspector.inspectThrowable(t);
logger.warn("Error loading counter cache", t);
}
if (joinRing)
{
joinTokenRing(delay);
}
else
{
Collection tokens = SystemKeyspace.getSavedTokens();
if (!tokens.isEmpty())
{
tokenMetadata.updateNormalTokens(tokens, FBUtilities.getBroadcastAddress());
// order is important here, the gossiper can fire in between adding these two states. It's ok to send TOKENS without STATUS, but *not* vice versa.
List> states = new ArrayList>();
states.add(Pair.create(ApplicationState.TOKENS, valueFactory.tokens(tokens)));
states.add(Pair.create(ApplicationState.STATUS, valueFactory.hibernate(true)));
Gossiper.instance.addLocalApplicationStates(states);
}
doAuthSetup();
logger.info("Not joining ring as requested. Use JMX (StorageService->joinRing()) to initiate ring joining");
}
initialized = true;
}
private void loadRingState()
{
if (Boolean.parseBoolean(System.getProperty("cassandra.load_ring_state", "true")))
{
logger.info("Loading persisted ring state");
Multimap loadedTokens = SystemKeyspace.loadTokens();
Map loadedHostIds = SystemKeyspace.loadHostIds();
for (InetAddress ep : loadedTokens.keySet())
{
if (ep.equals(FBUtilities.getBroadcastAddress()))
{
// entry has been mistakenly added, delete it
SystemKeyspace.removeEndpoint(ep);
}
else
{
if (loadedHostIds.containsKey(ep))
tokenMetadata.updateHostId(loadedHostIds.get(ep), ep);
Gossiper.instance.addSavedEndpoint(ep);
}
}
}
}
private boolean isReplacing()
{
if (System.getProperty("cassandra.replace_address_first_boot", null) != null && SystemKeyspace.bootstrapComplete())
{
logger.info("Replace address on first boot requested; this node is already bootstrapped");
return false;
}
return DatabaseDescriptor.getReplaceAddress() != null;
}
/**
* In the event of forceful termination we need to remove the shutdown hook to prevent hanging (OOM for instance)
*/
public void removeShutdownHook()
{
if (drainOnShutdown != null)
Runtime.getRuntime().removeShutdownHook(drainOnShutdown);
if (FBUtilities.isWindows)
WindowsTimer.endTimerPeriod(DatabaseDescriptor.getWindowsTimerInterval());
}
private boolean shouldBootstrap()
{
return DatabaseDescriptor.isAutoBootstrap() && !SystemKeyspace.bootstrapComplete() && !isSeed();
}
public static boolean isSeed()
{
return DatabaseDescriptor.getSeeds().contains(FBUtilities.getBroadcastAddress());
}
private void prepareToJoin() throws ConfigurationException
{
if (!joined)
{
Map appStates = new EnumMap<>(ApplicationState.class);
if (SystemKeyspace.wasDecommissioned())
{
if (Boolean.getBoolean("cassandra.override_decommission"))
{
logger.warn("This node was decommissioned, but overriding by operator request.");
SystemKeyspace.setBootstrapState(SystemKeyspace.BootstrapState.COMPLETED);
}
else
throw new ConfigurationException("This node was decommissioned and will not rejoin the ring unless cassandra.override_decommission=true has been set, or all existing data is removed and the node is bootstrapped again");
}
if (DatabaseDescriptor.getReplaceTokens().size() > 0 || DatabaseDescriptor.getReplaceNode() != null)
throw new RuntimeException("Replace method removed; use cassandra.replace_address instead");
if (!MessagingService.instance().isListening())
MessagingService.instance().listen();
UUID localHostId = SystemKeyspace.getLocalHostId();
if (replacing)
{
localHostId = prepareForReplacement();
appStates.put(ApplicationState.TOKENS, valueFactory.tokens(bootstrapTokens));
if (!DatabaseDescriptor.isAutoBootstrap())
{
// Will not do replace procedure, persist the tokens we're taking over locally
// so that they don't get clobbered with auto generated ones in joinTokenRing
SystemKeyspace.updateTokens(bootstrapTokens);
}
else if (isReplacingSameAddress())
{
//only go into hibernate state if replacing the same address (CASSANDRA-8523)
logger.warn("Writes will not be forwarded to this node during replacement because it has the same address as " +
"the node to be replaced ({}). If the previous node has been down for longer than max_hint_window_in_ms, " +
"repair must be run after the replacement process in order to make this node consistent.",
DatabaseDescriptor.getReplaceAddress());
appStates.put(ApplicationState.STATUS, valueFactory.hibernate(true));
}
}
else
{
checkForEndpointCollision(localHostId, SystemKeyspace.loadHostIds().keySet());
}
// have to start the gossip service before we can see any info on other nodes. this is necessary
// for bootstrap to get the load info it needs.
// (we won't be part of the storage ring though until we add a counterId to our state, below.)
// Seed the host ID-to-endpoint map with our own ID.
getTokenMetadata().updateHostId(localHostId, FBUtilities.getBroadcastAddress());
appStates.put(ApplicationState.NET_VERSION, valueFactory.networkVersion());
appStates.put(ApplicationState.HOST_ID, valueFactory.hostId(localHostId));
appStates.put(ApplicationState.RPC_ADDRESS, valueFactory.rpcaddress(FBUtilities.getBroadcastRpcAddress()));
appStates.put(ApplicationState.RELEASE_VERSION, valueFactory.releaseVersion());
// load the persisted ring state. This used to be done earlier in the init process,
// but now we always perform a shadow round when preparing to join and we have to
// clear endpoint states after doing that.
loadRingState();
logger.info("Starting up server gossip");
Gossiper.instance.register(this);
Gossiper.instance.start(SystemKeyspace.incrementAndGetGeneration(), appStates); // needed for node-ring gathering.
gossipActive = true;
// gossip snitch infos (local DC and rack)
gossipSnitchInfo();
// gossip Schema.emptyVersion forcing immediate check for schema updates (see MigrationManager#maybeScheduleSchemaPull)
Schema.instance.updateVersionAndAnnounce(); // Ensure we know our own actual Schema UUID in preparation for updates
LoadBroadcaster.instance.startBroadcasting();
HintsService.instance.startDispatch();
BatchlogManager.instance.start();
}
}
public void waitForSchema(int delay)
{
// first sleep the delay to make sure we see all our peers
for (int i = 0; i < delay; i += 1000)
{
// if we see schema, we can proceed to the next check directly
if (!Schema.instance.getVersion().equals(SchemaConstants.emptyVersion))
{
logger.debug("got schema: {}", Schema.instance.getVersion());
break;
}
Uninterruptibles.sleepUninterruptibly(1, TimeUnit.SECONDS);
}
// if our schema hasn't matched yet, wait until it has
// we do this by waiting for all in-flight migration requests and responses to complete
// (post CASSANDRA-1391 we don't expect this to be necessary very often, but it doesn't hurt to be careful)
if (!MigrationManager.isReadyForBootstrap())
{
setMode(Mode.JOINING, "waiting for schema information to complete", true);
MigrationManager.waitUntilReadyForBootstrap();
}
}
private void joinTokenRing(int delay) throws ConfigurationException
{
joined = true;
// We bootstrap if we haven't successfully bootstrapped before, as long as we are not a seed.
// If we are a seed, or if the user manually sets auto_bootstrap to false,
// we'll skip streaming data from other nodes and jump directly into the ring.
//
// The seed check allows us to skip the RING_DELAY sleep for the single-node cluster case,
// which is useful for both new users and testing.
//
// We attempted to replace this with a schema-presence check, but you need a meaningful sleep
// to get schema info from gossip which defeats the purpose. See CASSANDRA-4427 for the gory details.
Set current = new HashSet<>();
if (logger.isDebugEnabled())
{
logger.debug("Bootstrap variables: {} {} {} {}",
DatabaseDescriptor.isAutoBootstrap(),
SystemKeyspace.bootstrapInProgress(),
SystemKeyspace.bootstrapComplete(),
DatabaseDescriptor.getSeeds().contains(FBUtilities.getBroadcastAddress()));
}
if (DatabaseDescriptor.isAutoBootstrap() && !SystemKeyspace.bootstrapComplete() && DatabaseDescriptor.getSeeds().contains(FBUtilities.getBroadcastAddress()))
{
logger.info("This node will not auto bootstrap because it is configured to be a seed node.");
}
boolean dataAvailable = true; // make this to false when bootstrap streaming failed
boolean bootstrap = shouldBootstrap();
if (bootstrap)
{
if (SystemKeyspace.bootstrapInProgress())
logger.warn("Detected previous bootstrap failure; retrying");
else
SystemKeyspace.setBootstrapState(SystemKeyspace.BootstrapState.IN_PROGRESS);
setMode(Mode.JOINING, "waiting for ring information", true);
waitForSchema(delay);
setMode(Mode.JOINING, "schema complete, ready to bootstrap", true);
if (this.daemon != null)
this.daemon.beforeBootstrap();
setMode(Mode.JOINING, "waiting for pending range calculation", true);
PendingRangeCalculatorService.instance.blockUntilFinished();
setMode(Mode.JOINING, "calculation complete, ready to bootstrap", true);
logger.debug("... got ring + schema info");
if (useStrictConsistency && !allowSimultaneousMoves() &&
(
tokenMetadata.getBootstrapTokens().valueSet().size() > 0 ||
tokenMetadata.getLeavingEndpoints().size() > 0 ||
tokenMetadata.getMovingEndpoints().size() > 0
))
{
String bootstrapTokens = StringUtils.join(tokenMetadata.getBootstrapTokens().valueSet(), ',');
String leavingTokens = StringUtils.join(tokenMetadata.getLeavingEndpoints(), ',');
String movingTokens = StringUtils.join(tokenMetadata.getMovingEndpoints().stream().map(e -> e.right).toArray(), ',');
throw new UnsupportedOperationException(String.format("Other bootstrapping/leaving/moving nodes detected, cannot bootstrap while cassandra.consistent.rangemovement is true. Nodes detected, bootstrapping: %s; leaving: %s; moving: %s;", bootstrapTokens, leavingTokens, movingTokens));
}
// get bootstrap tokens
if (!replacing)
{
if (tokenMetadata.isMember(FBUtilities.getBroadcastAddress()))
{
String s = "This node is already a member of the token ring; bootstrap aborted. (If replacing a dead node, remove the old one from the ring first.)";
throw new UnsupportedOperationException(s);
}
setMode(Mode.JOINING, "getting bootstrap token", true);
bootstrapTokens = BootStrapper.getBootstrapTokens(tokenMetadata, FBUtilities.getBroadcastAddress(), delay);
}
else
{
if (!isReplacingSameAddress())
{
try
{
// Sleep additionally to make sure that the server actually is not alive
// and giving it more time to gossip if alive.
Thread.sleep(LoadBroadcaster.BROADCAST_INTERVAL);
}
catch (InterruptedException e)
{
throw new AssertionError(e);
}
// check for operator errors...
for (Token token : bootstrapTokens)
{
InetAddress existing = tokenMetadata.getEndpoint(token);
if (existing != null)
{
long nanoDelay = delay * 1000000L;
if (Gossiper.instance.getEndpointStateForEndpoint(existing).getUpdateTimestamp() > (System.nanoTime() - nanoDelay))
throw new UnsupportedOperationException("Cannot replace a live node... ");
current.add(existing);
}
else
{
throw new UnsupportedOperationException("Cannot replace token " + token + " which does not exist!");
}
}
}
else
{
try
{
Thread.sleep(RING_DELAY);
}
catch (InterruptedException e)
{
throw new AssertionError(e);
}
}
setMode(Mode.JOINING, "Replacing a node with token(s): " + bootstrapTokens, true);
}
dataAvailable = bootstrap(bootstrapTokens);
}
else
{
bootstrapTokens = SystemKeyspace.getSavedTokens();
if (bootstrapTokens.isEmpty())
{
bootstrapTokens = BootStrapper.getBootstrapTokens(tokenMetadata, FBUtilities.getBroadcastAddress(), delay);
}
else
{
if (bootstrapTokens.size() != DatabaseDescriptor.getNumTokens())
throw new ConfigurationException("Cannot change the number of tokens from " + bootstrapTokens.size() + " to " + DatabaseDescriptor.getNumTokens());
else
logger.info("Using saved tokens {}", bootstrapTokens);
}
if (this.daemon != null)
this.daemon.ringReady();
}
// if we don't have system_traces keyspace at this point, then create it manually
maybeAddOrUpdateKeyspace(TraceKeyspace.metadata());
maybeAddOrUpdateKeyspace(SystemDistributedKeyspace.metadata());
if (!isSurveyMode)
{
if (dataAvailable)
{
finishJoiningRing(bootstrap, bootstrapTokens);
// remove the existing info about the replaced node.
if (!current.isEmpty())
{
for (InetAddress existing : current)
Gossiper.instance.replacedEndpoint(existing);
}
}
else
{
logger.warn("Some data streaming failed. Use nodetool to check bootstrap state and resume. For more, see `nodetool help bootstrap`. {}", SystemKeyspace.getBootstrapState());
}
}
else
{
logger.info("Startup complete, but write survey mode is active, not becoming an active ring member. Use JMX (StorageService->joinRing()) to finalize ring joining.");
}
}
public static boolean isReplacingSameAddress()
{
InetAddress replaceAddress = DatabaseDescriptor.getReplaceAddress();
return replaceAddress != null && replaceAddress.equals(FBUtilities.getBroadcastAddress());
}
public void gossipSnitchInfo()
{
IEndpointSnitch snitch = DatabaseDescriptor.getEndpointSnitch();
String dc = snitch.getDatacenter(FBUtilities.getBroadcastAddress());
String rack = snitch.getRack(FBUtilities.getBroadcastAddress());
Gossiper.instance.addLocalApplicationState(ApplicationState.DC, StorageService.instance.valueFactory.datacenter(dc));
Gossiper.instance.addLocalApplicationState(ApplicationState.RACK, StorageService.instance.valueFactory.rack(rack));
}
public void joinRing() throws IOException
{
SystemKeyspace.BootstrapState state = SystemKeyspace.getBootstrapState();
joinRing(state.equals(SystemKeyspace.BootstrapState.IN_PROGRESS));
}
private synchronized void joinRing(boolean resumedBootstrap) throws IOException
{
if (!joined)
{
logger.info("Joining ring by operator request");
try
{
joinTokenRing(0);
}
catch (ConfigurationException e)
{
throw new IOException(e.getMessage());
}
}
else if (isSurveyMode)
{
logger.info("Leaving write survey mode and joining ring at operator request");
finishJoiningRing(resumedBootstrap, SystemKeyspace.getSavedTokens());
isSurveyMode = false;
}
}
private void executePreJoinTasks(boolean bootstrap)
{
StreamSupport.stream(ColumnFamilyStore.all().spliterator(), false)
.filter(cfs -> Schema.instance.getUserKeyspaces().contains(cfs.keyspace.getName()))
.forEach(cfs -> cfs.indexManager.executePreJoinTasksBlocking(bootstrap));
}
private void finishJoiningRing(boolean didBootstrap, Collection tokens)
{
// start participating in the ring.
setMode(Mode.JOINING, "Finish joining ring", true);
SystemKeyspace.setBootstrapState(SystemKeyspace.BootstrapState.COMPLETED);
executePreJoinTasks(didBootstrap);
setTokens(tokens);
assert tokenMetadata.sortedTokens().size() > 0;
doAuthSetup();
}
private void doAuthSetup()
{
if (!authSetupCalled.getAndSet(true))
{
maybeAddOrUpdateKeyspace(AuthKeyspace.metadata());
DatabaseDescriptor.getRoleManager().setup();
DatabaseDescriptor.getAuthenticator().setup();
DatabaseDescriptor.getAuthorizer().setup();
MigrationManager.instance.register(new AuthMigrationListener());
authSetupComplete = true;
}
}
public boolean isAuthSetupComplete()
{
return authSetupComplete;
}
private void maybeAddKeyspace(KeyspaceMetadata ksm)
{
try
{
/*
* We use timestamp of 0, intentionally, so that varying timestamps wouldn't cause schema mismatches on
* newly added nodes.
*
* Having the initial/default timestamp as 0 also allows users to make and persist changes to replication
* of our replicated system keyspaces.
*
* In case that we need to make incompatible changes to those kesypaces/tables, we'd need to bump the timestamp
* on per-keyspace/per-table basis. So far we've never needed to.
*/
MigrationManager.announceNewKeyspace(ksm, 0, false);
}
catch (AlreadyExistsException e)
{
logger.debug("Attempted to create new keyspace {}, but it already exists", ksm.name);
}
}
/**
* Ensure the schema of a pseudo-system keyspace (a distributed system keyspace: traces, auth and the so-called distributedKeyspace),
* is up to date with what we expected (creating it if it doesn't exist and updating tables that may have been upgraded).
*/
private void maybeAddOrUpdateKeyspace(KeyspaceMetadata expected)
{
// Note that want to deal with the keyspace and its table a bit differently: for the keyspace definition
// itself, we want to create it if it doesn't exist yet, but if it does exist, we don't want to modify it,
// because user can modify the definition to change the replication factor (#6016) and we don't want to
// override it. For the tables however, we have to deal with the fact that new version can add new columns
// (#8162 being an example), so even if the table definition exists, we still need to force the "current"
// version of the schema, the one the node will be expecting.
KeyspaceMetadata defined = Schema.instance.getKSMetaData(expected.name);
// If the keyspace doesn't exist, create it
if (defined == null)
{
maybeAddKeyspace(expected);
defined = Schema.instance.getKSMetaData(expected.name);
}
// While the keyspace exists, it might miss table or have outdated one
// There is also the potential for a race, as schema migrations add the bare
// keyspace into Schema.instance before adding its tables, so double check that
// all the expected tables are present
for (CFMetaData expectedTable : expected.tables)
{
CFMetaData definedTable = defined.tables.get(expectedTable.cfName).orElse(null);
if (definedTable == null || !definedTable.equals(expectedTable))
MigrationManager.forceAnnounceNewColumnFamily(expectedTable);
}
}
public boolean isJoined()
{
return tokenMetadata.isMember(FBUtilities.getBroadcastAddress()) && !isSurveyMode;
}
public void rebuild(String sourceDc)
{
rebuild(sourceDc, null, null, null);
}
public void rebuild(String sourceDc, String keyspace, String tokens, String specificSources)
{
// check ongoing rebuild
if (!isRebuilding.compareAndSet(false, true))
{
throw new IllegalStateException("Node is still rebuilding. Check nodetool netstats.");
}
// check the arguments
if (keyspace == null && tokens != null)
{
throw new IllegalArgumentException("Cannot specify tokens without keyspace.");
}
logger.info("rebuild from dc: {}, {}, {}", sourceDc == null ? "(any dc)" : sourceDc,
keyspace == null ? "(All keyspaces)" : keyspace,
tokens == null ? "(All tokens)" : tokens);
try
{
RangeStreamer streamer = new RangeStreamer(tokenMetadata,
null,
FBUtilities.getBroadcastAddress(),
"Rebuild",
useStrictConsistency && !replacing,
DatabaseDescriptor.getEndpointSnitch(),
streamStateStore,
false);
streamer.addSourceFilter(new RangeStreamer.FailureDetectorSourceFilter(FailureDetector.instance));
if (sourceDc != null)
streamer.addSourceFilter(new RangeStreamer.SingleDatacenterFilter(DatabaseDescriptor.getEndpointSnitch(), sourceDc));
if (keyspace == null)
{
for (String keyspaceName : Schema.instance.getNonLocalStrategyKeyspaces())
streamer.addRanges(keyspaceName, getLocalRanges(keyspaceName));
}
else if (tokens == null)
{
streamer.addRanges(keyspace, getLocalRanges(keyspace));
}
else
{
Token.TokenFactory factory = getTokenFactory();
List> ranges = new ArrayList<>();
Pattern rangePattern = Pattern.compile("\\(\\s*(-?\\w+)\\s*,\\s*(-?\\w+)\\s*\\]");
try (Scanner tokenScanner = new Scanner(tokens))
{
while (tokenScanner.findInLine(rangePattern) != null)
{
MatchResult range = tokenScanner.match();
Token startToken = factory.fromString(range.group(1));
Token endToken = factory.fromString(range.group(2));
logger.info("adding range: ({},{}]", startToken, endToken);
ranges.add(new Range<>(startToken, endToken));
}
if (tokenScanner.hasNext())
throw new IllegalArgumentException("Unexpected string: " + tokenScanner.next());
}
// Ensure all specified ranges are actually ranges owned by this host
Collection> localRanges = getLocalRanges(keyspace);
for (Range specifiedRange : ranges)
{
boolean foundParentRange = false;
for (Range localRange : localRanges)
{
if (localRange.contains(specifiedRange))
{
foundParentRange = true;
break;
}
}
if (!foundParentRange)
{
throw new IllegalArgumentException(String.format("The specified range %s is not a range that is owned by this node. Please ensure that all token ranges specified to be rebuilt belong to this node.", specifiedRange.toString()));
}
}
if (specificSources != null)
{
String[] stringHosts = specificSources.split(",");
Set sources = new HashSet<>(stringHosts.length);
for (String stringHost : stringHosts)
{
try
{
InetAddress endpoint = InetAddress.getByName(stringHost);
if (FBUtilities.getBroadcastAddress().equals(endpoint))
{
throw new IllegalArgumentException("This host was specified as a source for rebuilding. Sources for a rebuild can only be other nodes in the cluster.");
}
sources.add(endpoint);
}
catch (UnknownHostException ex)
{
throw new IllegalArgumentException("Unknown host specified " + stringHost, ex);
}
}
streamer.addSourceFilter(new RangeStreamer.WhitelistedSourcesFilter(sources));
}
streamer.addRanges(keyspace, ranges);
}
StreamResultFuture resultFuture = streamer.fetchAsync();
// wait for result
resultFuture.get();
}
catch (InterruptedException e)
{
throw new RuntimeException("Interrupted while waiting on rebuild streaming");
}
catch (ExecutionException e)
{
// This is used exclusively through JMX, so log the full trace but only throw a simple RTE
logger.error("Error while rebuilding node", e.getCause());
throw new RuntimeException("Error while rebuilding node: " + e.getCause().getMessage());
}
finally
{
// rebuild is done (successfully or not)
isRebuilding.set(false);
}
}
public void setRpcTimeout(long value)
{
DatabaseDescriptor.setRpcTimeout(value);
logger.info("set rpc timeout to {} ms", value);
}
public long getRpcTimeout()
{
return DatabaseDescriptor.getRpcTimeout();
}
public void setReadRpcTimeout(long value)
{
DatabaseDescriptor.setReadRpcTimeout(value);
logger.info("set read rpc timeout to {} ms", value);
}
public long getReadRpcTimeout()
{
return DatabaseDescriptor.getReadRpcTimeout();
}
public void setRangeRpcTimeout(long value)
{
DatabaseDescriptor.setRangeRpcTimeout(value);
logger.info("set range rpc timeout to {} ms", value);
}
public long getRangeRpcTimeout()
{
return DatabaseDescriptor.getRangeRpcTimeout();
}
public void setWriteRpcTimeout(long value)
{
DatabaseDescriptor.setWriteRpcTimeout(value);
logger.info("set write rpc timeout to {} ms", value);
}
public long getWriteRpcTimeout()
{
return DatabaseDescriptor.getWriteRpcTimeout();
}
public void setCounterWriteRpcTimeout(long value)
{
DatabaseDescriptor.setCounterWriteRpcTimeout(value);
logger.info("set counter write rpc timeout to {} ms", value);
}
public long getCounterWriteRpcTimeout()
{
return DatabaseDescriptor.getCounterWriteRpcTimeout();
}
public void setCasContentionTimeout(long value)
{
DatabaseDescriptor.setCasContentionTimeout(value);
logger.info("set cas contention rpc timeout to {} ms", value);
}
public long getCasContentionTimeout()
{
return DatabaseDescriptor.getCasContentionTimeout();
}
public void setTruncateRpcTimeout(long value)
{
DatabaseDescriptor.setTruncateRpcTimeout(value);
logger.info("set truncate rpc timeout to {} ms", value);
}
public long getTruncateRpcTimeout()
{
return DatabaseDescriptor.getTruncateRpcTimeout();
}
public void setStreamingSocketTimeout(int value)
{
DatabaseDescriptor.setStreamingSocketTimeout(value);
logger.info("set streaming socket timeout to {} ms", value);
}
public int getStreamingSocketTimeout()
{
return DatabaseDescriptor.getStreamingSocketTimeout();
}
public void setStreamThroughputMbPerSec(int value)
{
DatabaseDescriptor.setStreamThroughputOutboundMegabitsPerSec(value);
logger.info("setstreamthroughput: throttle set to {}", value);
}
public int getStreamThroughputMbPerSec()
{
return DatabaseDescriptor.getStreamThroughputOutboundMegabitsPerSec();
}
public void setInterDCStreamThroughputMbPerSec(int value)
{
DatabaseDescriptor.setInterDCStreamThroughputOutboundMegabitsPerSec(value);
logger.info("setinterdcstreamthroughput: throttle set to {}", value);
}
public int getInterDCStreamThroughputMbPerSec()
{
return DatabaseDescriptor.getInterDCStreamThroughputOutboundMegabitsPerSec();
}
public int getCompactionThroughputMbPerSec()
{
return DatabaseDescriptor.getCompactionThroughputMbPerSec();
}
public void setCompactionThroughputMbPerSec(int value)
{
DatabaseDescriptor.setCompactionThroughputMbPerSec(value);
CompactionManager.instance.setRate(value);
}
public int getConcurrentCompactors()
{
return DatabaseDescriptor.getConcurrentCompactors();
}
public void setConcurrentCompactors(int value)
{
if (value <= 0)
throw new IllegalArgumentException("Number of concurrent compactors should be greater than 0.");
DatabaseDescriptor.setConcurrentCompactors(value);
CompactionManager.instance.setConcurrentCompactors(value);
}
public boolean isIncrementalBackupsEnabled()
{
return DatabaseDescriptor.isIncrementalBackupsEnabled();
}
public void setIncrementalBackupsEnabled(boolean value)
{
DatabaseDescriptor.setIncrementalBackupsEnabled(value);
}
private void setMode(Mode m, boolean log)
{
setMode(m, null, log);
}
private void setMode(Mode m, String msg, boolean log)
{
operationMode = m;
String logMsg = msg == null ? m.toString() : String.format("%s: %s", m, msg);
if (log)
logger.info(logMsg);
else
logger.debug(logMsg);
}
/**
* Bootstrap node by fetching data from other nodes.
* If node is bootstrapping as a new node, then this also announces bootstrapping to the cluster.
*
* This blocks until streaming is done.
*
* @param tokens bootstrapping tokens
* @return true if bootstrap succeeds.
*/
private boolean bootstrap(final Collection tokens)
{
isBootstrapMode = true;
SystemKeyspace.updateTokens(tokens); // DON'T use setToken, that makes us part of the ring locally which is incorrect until we are done bootstrapping
if (!replacing || !isReplacingSameAddress())
{
// if not an existing token then bootstrap
List> states = new ArrayList<>();
states.add(Pair.create(ApplicationState.TOKENS, valueFactory.tokens(tokens)));
states.add(Pair.create(ApplicationState.STATUS, replacing?
valueFactory.bootReplacing(DatabaseDescriptor.getReplaceAddress()) :
valueFactory.bootstrapping(tokens)));
Gossiper.instance.addLocalApplicationStates(states);
setMode(Mode.JOINING, "sleeping " + RING_DELAY + " ms for pending range setup", true);
Uninterruptibles.sleepUninterruptibly(RING_DELAY, TimeUnit.MILLISECONDS);
}
else
{
// Dont set any state for the node which is bootstrapping the existing token...
tokenMetadata.updateNormalTokens(tokens, FBUtilities.getBroadcastAddress());
SystemKeyspace.removeEndpoint(DatabaseDescriptor.getReplaceAddress());
}
if (!Gossiper.instance.seenAnySeed())
throw new IllegalStateException("Unable to contact any seeds!");
if (Boolean.getBoolean("cassandra.reset_bootstrap_progress"))
{
logger.info("Resetting bootstrap progress to start fresh");
SystemKeyspace.resetAvailableRanges();
}
// Force disk boundary invalidation now that local tokens are set
invalidateDiskBoundaries();
setMode(Mode.JOINING, "Starting to bootstrap...", true);
BootStrapper bootstrapper = new BootStrapper(FBUtilities.getBroadcastAddress(), tokens, tokenMetadata);
bootstrapper.addProgressListener(progressSupport);
ListenableFuture bootstrapStream = bootstrapper.bootstrap(streamStateStore, useStrictConsistency && !replacing); // handles token update
Futures.addCallback(bootstrapStream, new FutureCallback()
{
@Override
public void onSuccess(StreamState streamState)
{
bootstrapFinished();
logger.info("Bootstrap completed! for the tokens {}", tokens);
}
@Override
public void onFailure(Throwable e)
{
logger.warn("Error during bootstrap.", e);
}
});
try
{
bootstrapStream.get();
return true;
}
catch (Throwable e)
{
logger.error("Error while waiting on bootstrap to complete. Bootstrap will have to be restarted.", e);
return false;
}
}
private void invalidateDiskBoundaries()
{
for (Keyspace keyspace : Keyspace.all())
{
for (ColumnFamilyStore cfs : keyspace.getColumnFamilyStores())
{
for (final ColumnFamilyStore store : cfs.concatWithIndexes())
{
store.invalidateDiskBoundaries();
}
}
}
}
/**
* All MVs have been created during bootstrap, so mark them as built
*/
private void markViewsAsBuilt() {
for (String keyspace : Schema.instance.getUserKeyspaces())
{
for (ViewDefinition view: Schema.instance.getKSMetaData(keyspace).views)
SystemKeyspace.finishViewBuildStatus(view.ksName, view.viewName);
}
}
/**
* Called when bootstrap did finish successfully
*/
private void bootstrapFinished() {
markViewsAsBuilt();
isBootstrapMode = false;
}
public boolean resumeBootstrap()
{
if (isBootstrapMode && SystemKeyspace.bootstrapInProgress())
{
logger.info("Resuming bootstrap...");
// get bootstrap tokens saved in system keyspace
final Collection tokens = SystemKeyspace.getSavedTokens();
// already bootstrapped ranges are filtered during bootstrap
BootStrapper bootstrapper = new BootStrapper(FBUtilities.getBroadcastAddress(), tokens, tokenMetadata);
bootstrapper.addProgressListener(progressSupport);
ListenableFuture bootstrapStream = bootstrapper.bootstrap(streamStateStore, useStrictConsistency && !replacing); // handles token update
Futures.addCallback(bootstrapStream, new FutureCallback()
{
@Override
public void onSuccess(StreamState streamState)
{
bootstrapFinished();
// start participating in the ring.
// pretend we are in survey mode so we can use joinRing() here
isSurveyMode = true;
try
{
progressSupport.progress("bootstrap", ProgressEvent.createNotification("Joining ring..."));
joinRing(true);
}
catch (IOException ignore)
{
// joinRing with survey mode does not throw IOException
}
progressSupport.progress("bootstrap", new ProgressEvent(ProgressEventType.COMPLETE, 1, 1, "Resume bootstrap complete"));
logger.info("Resume complete");
}
@Override
public void onFailure(Throwable e)
{
String message = "Error during bootstrap: ";
if (e instanceof ExecutionException && e.getCause() != null)
{
message += e.getCause().getMessage();
}
else
{
message += e.getMessage();
}
logger.error(message, e);
progressSupport.progress("bootstrap", new ProgressEvent(ProgressEventType.ERROR, 1, 1, message));
progressSupport.progress("bootstrap", new ProgressEvent(ProgressEventType.COMPLETE, 1, 1, "Resume bootstrap complete"));
}
});
return true;
}
else
{
logger.info("Resuming bootstrap is requested, but the node is already bootstrapped.");
return false;
}
}
public boolean isBootstrapMode()
{
return isBootstrapMode;
}
public TokenMetadata getTokenMetadata()
{
return tokenMetadata;
}
/**
* for a keyspace, return the ranges and corresponding listen addresses.
* @param keyspace
* @return the endpoint map
*/
public Map, List> getRangeToEndpointMap(String keyspace)
{
/* All the ranges for the tokens */
Map, List> map = new HashMap<>();
for (Map.Entry,List> entry : getRangeToAddressMap(keyspace).entrySet())
{
map.put(entry.getKey().asList(), stringify(entry.getValue()));
}
return map;
}
/**
* Return the rpc address associated with an endpoint as a string.
* @param endpoint The endpoint to get rpc address for
* @return the rpc address
*/
public String getRpcaddress(InetAddress endpoint)
{
if (endpoint.equals(FBUtilities.getBroadcastAddress()))
return FBUtilities.getBroadcastRpcAddress().getHostAddress();
else if (Gossiper.instance.getEndpointStateForEndpoint(endpoint).getApplicationState(ApplicationState.RPC_ADDRESS) == null)
return endpoint.getHostAddress();
else
return Gossiper.instance.getEndpointStateForEndpoint(endpoint).getApplicationState(ApplicationState.RPC_ADDRESS).value;
}
/**
* for a keyspace, return the ranges and corresponding RPC addresses for a given keyspace.
* @param keyspace
* @return the endpoint map
*/
public Map, List> getRangeToRpcaddressMap(String keyspace)
{
/* All the ranges for the tokens */
Map, List> map = new HashMap<>();
for (Map.Entry, List> entry : getRangeToAddressMap(keyspace).entrySet())
{
List rpcaddrs = new ArrayList<>(entry.getValue().size());
for (InetAddress endpoint: entry.getValue())
{
rpcaddrs.add(getRpcaddress(endpoint));
}
map.put(entry.getKey().asList(), rpcaddrs);
}
return map;
}
public Map, List> getPendingRangeToEndpointMap(String keyspace)
{
// some people just want to get a visual representation of things. Allow null and set it to the first
// non-system keyspace.
if (keyspace == null)
keyspace = Schema.instance.getNonLocalStrategyKeyspaces().get(0);
Map, List> map = new HashMap<>();
for (Map.Entry, Collection> entry : tokenMetadata.getPendingRangesMM(keyspace).asMap().entrySet())
{
List l = new ArrayList<>(entry.getValue());
map.put(entry.getKey().asList(), stringify(l));
}
return map;
}
public Map, List> getRangeToAddressMap(String keyspace)
{
return getRangeToAddressMap(keyspace, tokenMetadata.sortedTokens());
}
public Map, List> getRangeToAddressMapInLocalDC(String keyspace)
{
Predicate isLocalDC = new Predicate()
{
public boolean apply(InetAddress address)
{
return isLocalDC(address);
}
};
Map, List> origMap = getRangeToAddressMap(keyspace, getTokensInLocalDC());
Map, List> filteredMap = Maps.newHashMap();
for (Map.Entry, List> entry : origMap.entrySet())
{
List endpointsInLocalDC = Lists.newArrayList(Collections2.filter(entry.getValue(), isLocalDC));
filteredMap.put(entry.getKey(), endpointsInLocalDC);
}
return filteredMap;
}
private List getTokensInLocalDC()
{
List filteredTokens = Lists.newArrayList();
for (Token token : tokenMetadata.sortedTokens())
{
InetAddress endpoint = tokenMetadata.getEndpoint(token);
if (isLocalDC(endpoint))
filteredTokens.add(token);
}
return filteredTokens;
}
private boolean isLocalDC(InetAddress targetHost)
{
String remoteDC = DatabaseDescriptor.getEndpointSnitch().getDatacenter(targetHost);
String localDC = DatabaseDescriptor.getEndpointSnitch().getDatacenter(FBUtilities.getBroadcastAddress());
return remoteDC.equals(localDC);
}
private Map, List> getRangeToAddressMap(String keyspace, List sortedTokens)
{
// some people just want to get a visual representation of things. Allow null and set it to the first
// non-system keyspace.
if (keyspace == null)
keyspace = Schema.instance.getNonLocalStrategyKeyspaces().get(0);
List> ranges = getAllRanges(sortedTokens);
return constructRangeToEndpointMap(keyspace, ranges);
}
/**
* The same as {@code describeRing(String)} but converts TokenRange to the String for JMX compatibility
*
* @param keyspace The keyspace to fetch information about
*
* @return a List of TokenRange(s) converted to String for the given keyspace
*/
public List describeRingJMX(String keyspace) throws IOException
{
List tokenRanges;
try
{
tokenRanges = describeRing(keyspace);
}
catch (InvalidRequestException e)
{
throw new IOException(e.getMessage());
}
List result = new ArrayList<>(tokenRanges.size());
for (TokenRange tokenRange : tokenRanges)
result.add(tokenRange.toString());
return result;
}
/**
* The TokenRange for a given keyspace.
*
* @param keyspace The keyspace to fetch information about
*
* @return a List of TokenRange(s) for the given keyspace
*
* @throws InvalidRequestException if there is no ring information available about keyspace
*/
public List describeRing(String keyspace) throws InvalidRequestException
{
return describeRing(keyspace, false);
}
/**
* The same as {@code describeRing(String)} but considers only the part of the ring formed by nodes in the local DC.
*/
public List describeLocalRing(String keyspace) throws InvalidRequestException
{
return describeRing(keyspace, true);
}
private List describeRing(String keyspace, boolean includeOnlyLocalDC) throws InvalidRequestException
{
if (!Schema.instance.getKeyspaces().contains(keyspace))
throw new InvalidRequestException("No such keyspace: " + keyspace);
if (keyspace == null || Keyspace.open(keyspace).getReplicationStrategy() instanceof LocalStrategy)
throw new InvalidRequestException("There is no ring for the keyspace: " + keyspace);
List ranges = new ArrayList<>();
Token.TokenFactory tf = getTokenFactory();
Map, List> rangeToAddressMap =
includeOnlyLocalDC
? getRangeToAddressMapInLocalDC(keyspace)
: getRangeToAddressMap(keyspace);
for (Map.Entry, List> entry : rangeToAddressMap.entrySet())
{
Range range = entry.getKey();
List addresses = entry.getValue();
List endpoints = new ArrayList<>(addresses.size());
List rpc_endpoints = new ArrayList<>(addresses.size());
List epDetails = new ArrayList<>(addresses.size());
for (InetAddress endpoint : addresses)
{
EndpointDetails details = new EndpointDetails();
details.host = endpoint.getHostAddress();
details.datacenter = DatabaseDescriptor.getEndpointSnitch().getDatacenter(endpoint);
details.rack = DatabaseDescriptor.getEndpointSnitch().getRack(endpoint);
endpoints.add(details.host);
rpc_endpoints.add(getRpcaddress(endpoint));
epDetails.add(details);
}
TokenRange tr = new TokenRange(tf.toString(range.left.getToken()), tf.toString(range.right.getToken()), endpoints)
.setEndpoint_details(epDetails)
.setRpc_endpoints(rpc_endpoints);
ranges.add(tr);
}
return ranges;
}
public Map getTokenToEndpointMap()
{
Map mapInetAddress = tokenMetadata.getNormalAndBootstrappingTokenToEndpointMap();
// in order to preserve tokens in ascending order, we use LinkedHashMap here
Map mapString = new LinkedHashMap<>(mapInetAddress.size());
List tokens = new ArrayList<>(mapInetAddress.keySet());
Collections.sort(tokens);
for (Token token : tokens)
{
mapString.put(token.toString(), mapInetAddress.get(token).getHostAddress());
}
return mapString;
}
public String getLocalHostId()
{
return getTokenMetadata().getHostId(FBUtilities.getBroadcastAddress()).toString();
}
public UUID getLocalHostUUID()
{
return getTokenMetadata().getHostId(FBUtilities.getBroadcastAddress());
}
public Map getHostIdMap()
{
return getEndpointToHostId();
}
public Map getEndpointToHostId()
{
Map mapOut = new HashMap<>();
for (Map.Entry entry : getTokenMetadata().getEndpointToHostIdMapForReading().entrySet())
mapOut.put(entry.getKey().getHostAddress(), entry.getValue().toString());
return mapOut;
}
public UUID getHostId(InetAddress endpoint)
{
return getTokenMetadata().getHostId(endpoint);
}
public Map getHostIdToEndpoint()
{
Map mapOut = new HashMap<>();
for (Map.Entry entry : getTokenMetadata().getEndpointToHostIdMapForReading().entrySet())
mapOut.put(entry.getValue().toString(), entry.getKey().getHostAddress());
return mapOut;
}
/**
* Construct the range to endpoint mapping based on the true view
* of the world.
* @param ranges
* @return mapping of ranges to the replicas responsible for them.
*/
private Map, List> constructRangeToEndpointMap(String keyspace, List> ranges)
{
Map, List> rangeToEndpointMap = new HashMap<>(ranges.size());
for (Range range : ranges)
{
rangeToEndpointMap.put(range, Keyspace.open(keyspace).getReplicationStrategy().getNaturalEndpoints(range.right));
}
return rangeToEndpointMap;
}
public void beforeChange(InetAddress endpoint, EndpointState currentState, ApplicationState newStateKey, VersionedValue newValue)
{
// no-op
}
/*
* Handle the reception of a new particular ApplicationState for a particular endpoint. Note that the value of the
* ApplicationState has not necessarily "changed" since the last known value, if we already received the same update
* from somewhere else.
*
* onChange only ever sees one ApplicationState piece change at a time (even if many ApplicationState updates were
* received at the same time), so we perform a kind of state machine here. We are concerned with two events: knowing
* the token associated with an endpoint, and knowing its operation mode. Nodes can start in either bootstrap or
* normal mode, and from bootstrap mode can change mode to normal. A node in bootstrap mode needs to have
* pendingranges set in TokenMetadata; a node in normal mode should instead be part of the token ring.
*
* Normal progression of ApplicationState.STATUS values for a node should be like this:
* STATUS_BOOTSTRAPPING,token
* if bootstrapping. stays this way until all files are received.
* STATUS_NORMAL,token
* ready to serve reads and writes.
* STATUS_LEAVING,token
* get ready to leave the cluster as part of a decommission
* STATUS_LEFT,token
* set after decommission is completed.
*
* Other STATUS values that may be seen (possibly anywhere in the normal progression):
* STATUS_MOVING,newtoken
* set if node is currently moving to a new token in the ring
* REMOVING_TOKEN,deadtoken
* set if the node is dead and is being removed by its REMOVAL_COORDINATOR
* REMOVED_TOKEN,deadtoken
* set if the node is dead and has been removed by its REMOVAL_COORDINATOR
*
* Note: Any time a node state changes from STATUS_NORMAL, it will not be visible to new nodes. So it follows that
* you should never bootstrap a new node during a removenode, decommission or move.
*/
public void onChange(InetAddress endpoint, ApplicationState state, VersionedValue value)
{
if (state == ApplicationState.STATUS)
{
String[] pieces = splitValue(value);
assert (pieces.length > 0);
String moveName = pieces[0];
switch (moveName)
{
case VersionedValue.STATUS_BOOTSTRAPPING_REPLACE:
handleStateBootreplacing(endpoint, pieces);
break;
case VersionedValue.STATUS_BOOTSTRAPPING:
handleStateBootstrap(endpoint);
break;
case VersionedValue.STATUS_NORMAL:
handleStateNormal(endpoint, VersionedValue.STATUS_NORMAL);
break;
case VersionedValue.SHUTDOWN:
handleStateNormal(endpoint, VersionedValue.SHUTDOWN);
break;
case VersionedValue.REMOVING_TOKEN:
case VersionedValue.REMOVED_TOKEN:
handleStateRemoving(endpoint, pieces);
break;
case VersionedValue.STATUS_LEAVING:
handleStateLeaving(endpoint);
break;
case VersionedValue.STATUS_LEFT:
handleStateLeft(endpoint, pieces);
break;
case VersionedValue.STATUS_MOVING:
handleStateMoving(endpoint, pieces);
break;
}
}
else
{
EndpointState epState = Gossiper.instance.getEndpointStateForEndpoint(endpoint);
if (epState == null || Gossiper.instance.isDeadState(epState))
{
logger.debug("Ignoring state change for dead or unknown endpoint: {}", endpoint);
return;
}
if (getTokenMetadata().isMember(endpoint))
{
final ExecutorService executor = StageManager.getStage(Stage.MUTATION);
switch (state)
{
case RELEASE_VERSION:
SystemKeyspace.updatePeerInfo(endpoint, "release_version", value.value, executor);
break;
case DC:
updateTopology(endpoint);
SystemKeyspace.updatePeerInfo(endpoint, "data_center", value.value, executor);
break;
case RACK:
updateTopology(endpoint);
SystemKeyspace.updatePeerInfo(endpoint, "rack", value.value, executor);
break;
case RPC_ADDRESS:
try
{
SystemKeyspace.updatePeerInfo(endpoint, "rpc_address", InetAddress.getByName(value.value), executor);
}
catch (UnknownHostException e)
{
throw new RuntimeException(e);
}
break;
case SCHEMA:
SystemKeyspace.updatePeerInfo(endpoint, "schema_version", UUID.fromString(value.value), executor);
MigrationManager.instance.scheduleSchemaPull(endpoint, epState);
break;
case HOST_ID:
SystemKeyspace.updatePeerInfo(endpoint, "host_id", UUID.fromString(value.value), executor);
break;
case RPC_READY:
notifyRpcChange(endpoint, epState.isRpcReady());
break;
case NET_VERSION:
updateNetVersion(endpoint, value);
break;
}
}
}
}
private static String[] splitValue(VersionedValue value)
{
return value.value.split(VersionedValue.DELIMITER_STR, -1);
}
private void updateNetVersion(InetAddress endpoint, VersionedValue value)
{
try
{
MessagingService.instance().setVersion(endpoint, Integer.parseInt(value.value));
}
catch (NumberFormatException e)
{
throw new AssertionError("Got invalid value for NET_VERSION application state: " + value.value);
}
}
public void updateTopology(InetAddress endpoint)
{
if (getTokenMetadata().isMember(endpoint))
{
getTokenMetadata().updateTopology(endpoint);
}
}
public void updateTopology()
{
getTokenMetadata().updateTopology();
}
private void updatePeerInfo(InetAddress endpoint)
{
EndpointState epState = Gossiper.instance.getEndpointStateForEndpoint(endpoint);
final ExecutorService executor = StageManager.getStage(Stage.MUTATION);
for (Map.Entry entry : epState.states())
{
switch (entry.getKey())
{
case RELEASE_VERSION:
SystemKeyspace.updatePeerInfo(endpoint, "release_version", entry.getValue().value, executor);
break;
case DC:
SystemKeyspace.updatePeerInfo(endpoint, "data_center", entry.getValue().value, executor);
break;
case RACK:
SystemKeyspace.updatePeerInfo(endpoint, "rack", entry.getValue().value, executor);
break;
case RPC_ADDRESS:
try
{
SystemKeyspace.updatePeerInfo(endpoint, "rpc_address", InetAddress.getByName(entry.getValue().value), executor);
}
catch (UnknownHostException e)
{
throw new RuntimeException(e);
}
break;
case SCHEMA:
SystemKeyspace.updatePeerInfo(endpoint, "schema_version", UUID.fromString(entry.getValue().value), executor);
break;
case HOST_ID:
SystemKeyspace.updatePeerInfo(endpoint, "host_id", UUID.fromString(entry.getValue().value), executor);
break;
}
}
}
private void notifyRpcChange(InetAddress endpoint, boolean ready)
{
if (ready)
notifyUp(endpoint);
else
notifyDown(endpoint);
}
private void notifyUp(InetAddress endpoint)
{
if (!isRpcReady(endpoint) || !Gossiper.instance.isAlive(endpoint))
return;
for (IEndpointLifecycleSubscriber subscriber : lifecycleSubscribers)
subscriber.onUp(endpoint);
}
private void notifyDown(InetAddress endpoint)
{
for (IEndpointLifecycleSubscriber subscriber : lifecycleSubscribers)
subscriber.onDown(endpoint);
}
private void notifyJoined(InetAddress endpoint)
{
if (!isStatus(endpoint, VersionedValue.STATUS_NORMAL))
return;
for (IEndpointLifecycleSubscriber subscriber : lifecycleSubscribers)
subscriber.onJoinCluster(endpoint);
}
private void notifyMoved(InetAddress endpoint)
{
for (IEndpointLifecycleSubscriber subscriber : lifecycleSubscribers)
subscriber.onMove(endpoint);
}
private void notifyLeft(InetAddress endpoint)
{
for (IEndpointLifecycleSubscriber subscriber : lifecycleSubscribers)
subscriber.onLeaveCluster(endpoint);
}
private boolean isStatus(InetAddress endpoint, String status)
{
EndpointState state = Gossiper.instance.getEndpointStateForEndpoint(endpoint);
return state != null && state.getStatus().equals(status);
}
public boolean isRpcReady(InetAddress endpoint)
{
if (MessagingService.instance().getVersion(endpoint) < MessagingService.VERSION_22)
return true;
EndpointState state = Gossiper.instance.getEndpointStateForEndpoint(endpoint);
return state != null && state.isRpcReady();
}
/**
* Set the RPC status. Because when draining a node we need to set the RPC
* status to not ready, and drain is called by the shutdown hook, it may be that value is false
* and there is no local endpoint state. In this case it's OK to just do nothing. Therefore,
* we assert that the local endpoint state is not null only when value is true.
*
* @param value - true indicates that RPC is ready, false indicates the opposite.
*/
public void setRpcReady(boolean value)
{
EndpointState state = Gossiper.instance.getEndpointStateForEndpoint(FBUtilities.getBroadcastAddress());
// if value is false we're OK with a null state, if it is true we are not.
assert !value || state != null;
if (state != null)
Gossiper.instance.addLocalApplicationState(ApplicationState.RPC_READY, valueFactory.rpcReady(value));
}
private Collection getTokensFor(InetAddress endpoint)
{
try
{
EndpointState state = Gossiper.instance.getEndpointStateForEndpoint(endpoint);
if (state == null)
return Collections.emptyList();
VersionedValue versionedValue = state.getApplicationState(ApplicationState.TOKENS);
if (versionedValue == null)
return Collections.emptyList();
return TokenSerializer.deserialize(tokenMetadata.partitioner, new DataInputStream(new ByteArrayInputStream(versionedValue.toBytes())));
}
catch (IOException e)
{
throw new RuntimeException(e);
}
}
/**
* Handle node bootstrap
*
* @param endpoint bootstrapping node
*/
private void handleStateBootstrap(InetAddress endpoint)
{
Collection tokens;
// explicitly check for TOKENS, because a bootstrapping node might be bootstrapping in legacy mode; that is, not using vnodes and no token specified
tokens = getTokensFor(endpoint);
if (logger.isDebugEnabled())
logger.debug("Node {} state bootstrapping, token {}", endpoint, tokens);
// if this node is present in token metadata, either we have missed intermediate states
// or the node had crashed. Print warning if needed, clear obsolete stuff and
// continue.
if (tokenMetadata.isMember(endpoint))
{
// If isLeaving is false, we have missed both LEAVING and LEFT. However, if
// isLeaving is true, we have only missed LEFT. Waiting time between completing
// leave operation and rebootstrapping is relatively short, so the latter is quite
// common (not enough time for gossip to spread). Therefore we report only the
// former in the log.
if (!tokenMetadata.isLeaving(endpoint))
logger.info("Node {} state jump to bootstrap", endpoint);
tokenMetadata.removeEndpoint(endpoint);
}
tokenMetadata.addBootstrapTokens(tokens, endpoint);
PendingRangeCalculatorService.instance.update();
tokenMetadata.updateHostId(Gossiper.instance.getHostId(endpoint), endpoint);
}
private void handleStateBootreplacing(InetAddress newNode, String[] pieces)
{
InetAddress oldNode;
try
{
oldNode = InetAddress.getByName(pieces[1]);
}
catch (Exception e)
{
logger.error("Node {} tried to replace malformed endpoint {}.", newNode, pieces[1], e);
return;
}
if (FailureDetector.instance.isAlive(oldNode))
{
throw new RuntimeException(String.format("Node %s is trying to replace alive node %s.", newNode, oldNode));
}
Optional replacingNode = tokenMetadata.getReplacingNode(newNode);
if (replacingNode.isPresent() && !replacingNode.get().equals(oldNode))
{
throw new RuntimeException(String.format("Node %s is already replacing %s but is trying to replace %s.",
newNode, replacingNode.get(), oldNode));
}
Collection tokens = getTokensFor(newNode);
if (logger.isDebugEnabled())
logger.debug("Node {} is replacing {}, tokens {}", newNode, oldNode, tokens);
tokenMetadata.addReplaceTokens(tokens, newNode, oldNode);
PendingRangeCalculatorService.instance.update();
tokenMetadata.updateHostId(Gossiper.instance.getHostId(newNode), newNode);
}
/**
* Handle node move to normal state. That is, node is entering token ring and participating
* in reads.
*
* @param endpoint node
*/
private void handleStateNormal(final InetAddress endpoint, final String status)
{
Collection tokens = getTokensFor(endpoint);
Set tokensToUpdateInMetadata = new HashSet<>();
Set tokensToUpdateInSystemKeyspace = new HashSet<>();
Set endpointsToRemove = new HashSet<>();
if (logger.isDebugEnabled())
logger.debug("Node {} state {}, token {}", endpoint, status, tokens);
if (tokenMetadata.isMember(endpoint))
logger.info("Node {} state jump to {}", endpoint, status);
if (tokens.isEmpty() && status.equals(VersionedValue.STATUS_NORMAL))
logger.error("Node {} is in state normal but it has no tokens, state: {}",
endpoint,
Gossiper.instance.getEndpointStateForEndpoint(endpoint));
Optional replacingNode = tokenMetadata.getReplacingNode(endpoint);
if (replacingNode.isPresent())
{
assert !endpoint.equals(replacingNode.get()) : "Pending replacement endpoint with same address is not supported";
logger.info("Node {} will complete replacement of {} for tokens {}", endpoint, replacingNode.get(), tokens);
if (FailureDetector.instance.isAlive(replacingNode.get()))
{
logger.error("Node {} cannot complete replacement of alive node {}.", endpoint, replacingNode.get());
return;
}
endpointsToRemove.add(replacingNode.get());
}
Optional replacementNode = tokenMetadata.getReplacementNode(endpoint);
if (replacementNode.isPresent())
{
logger.warn("Node {} is currently being replaced by node {}.", endpoint, replacementNode.get());
}
updatePeerInfo(endpoint);
// Order Matters, TM.updateHostID() should be called before TM.updateNormalToken(), (see CASSANDRA-4300).
UUID hostId = Gossiper.instance.getHostId(endpoint);
InetAddress existing = tokenMetadata.getEndpointForHostId(hostId);
if (replacing && isReplacingSameAddress() && Gossiper.instance.getEndpointStateForEndpoint(DatabaseDescriptor.getReplaceAddress()) != null
&& (hostId.equals(Gossiper.instance.getHostId(DatabaseDescriptor.getReplaceAddress()))))
logger.warn("Not updating token metadata for {} because I am replacing it", endpoint);
else
{
if (existing != null && !existing.equals(endpoint))
{
if (existing.equals(FBUtilities.getBroadcastAddress()))
{
logger.warn("Not updating host ID {} for {} because it's mine", hostId, endpoint);
tokenMetadata.removeEndpoint(endpoint);
endpointsToRemove.add(endpoint);
}
else if (Gossiper.instance.compareEndpointStartup(endpoint, existing) > 0)
{
logger.warn("Host ID collision for {} between {} and {}; {} is the new owner", hostId, existing, endpoint, endpoint);
tokenMetadata.removeEndpoint(existing);
endpointsToRemove.add(existing);
tokenMetadata.updateHostId(hostId, endpoint);
}
else
{
logger.warn("Host ID collision for {} between {} and {}; ignored {}", hostId, existing, endpoint, endpoint);
tokenMetadata.removeEndpoint(endpoint);
endpointsToRemove.add(endpoint);
}
}
else
tokenMetadata.updateHostId(hostId, endpoint);
}
for (final Token token : tokens)
{
// we don't want to update if this node is responsible for the token and it has a later startup time than endpoint.
InetAddress currentOwner = tokenMetadata.getEndpoint(token);
if (currentOwner == null)
{
logger.debug("New node {} at token {}", endpoint, token);
tokensToUpdateInMetadata.add(token);
tokensToUpdateInSystemKeyspace.add(token);
}
else if (endpoint.equals(currentOwner))
{
// set state back to normal, since the node may have tried to leave, but failed and is now back up
tokensToUpdateInMetadata.add(token);
tokensToUpdateInSystemKeyspace.add(token);
}
else if (Gossiper.instance.compareEndpointStartup(endpoint, currentOwner) > 0)
{
tokensToUpdateInMetadata.add(token);
tokensToUpdateInSystemKeyspace.add(token);
// currentOwner is no longer current, endpoint is. Keep track of these moves, because when
// a host no longer has any tokens, we'll want to remove it.
Multimap epToTokenCopy = getTokenMetadata().getEndpointToTokenMapForReading();
epToTokenCopy.get(currentOwner).remove(token);
if (epToTokenCopy.get(currentOwner).size() < 1)
endpointsToRemove.add(currentOwner);
logger.info("Nodes {} and {} have the same token {}. {} is the new owner",
endpoint,
currentOwner,
token,
endpoint);
}
else
{
logger.info("Nodes {} and {} have the same token {}. Ignoring {}",
endpoint,
currentOwner,
token,
endpoint);
}
}
// capture because updateNormalTokens clears moving and member status
boolean isMember = tokenMetadata.isMember(endpoint);
boolean isMoving = tokenMetadata.isMoving(endpoint);
tokenMetadata.updateNormalTokens(tokensToUpdateInMetadata, endpoint);
for (InetAddress ep : endpointsToRemove)
{
removeEndpoint(ep);
if (replacing && DatabaseDescriptor.getReplaceAddress().equals(ep))
Gossiper.instance.replacementQuarantine(ep); // quarantine locally longer than normally; see CASSANDRA-8260
}
if (!tokensToUpdateInSystemKeyspace.isEmpty())
SystemKeyspace.updateTokens(endpoint, tokensToUpdateInSystemKeyspace, StageManager.getStage(Stage.MUTATION));
if (isMoving || operationMode == Mode.MOVING)
{
tokenMetadata.removeFromMoving(endpoint);
notifyMoved(endpoint);
}
else if (!isMember) // prior to this, the node was not a member
{
notifyJoined(endpoint);
}
PendingRangeCalculatorService.instance.update();
}
/**
* Handle node preparing to leave the ring
*
* @param endpoint node
*/
private void handleStateLeaving(InetAddress endpoint)
{
Collection tokens = getTokensFor(endpoint);
if (logger.isDebugEnabled())
logger.debug("Node {} state leaving, tokens {}", endpoint, tokens);
// If the node is previously unknown or tokens do not match, update tokenmetadata to
// have this node as 'normal' (it must have been using this token before the
// leave). This way we'll get pending ranges right.
if (!tokenMetadata.isMember(endpoint))
{
logger.info("Node {} state jump to leaving", endpoint);
tokenMetadata.updateNormalTokens(tokens, endpoint);
}
else if (!tokenMetadata.getTokens(endpoint).containsAll(tokens))
{
logger.warn("Node {} 'leaving' token mismatch. Long network partition?", endpoint);
tokenMetadata.updateNormalTokens(tokens, endpoint);
}
// at this point the endpoint is certainly a member with this token, so let's proceed
// normally
tokenMetadata.addLeavingEndpoint(endpoint);
PendingRangeCalculatorService.instance.update();
}
/**
* Handle node leaving the ring. This will happen when a node is decommissioned
*
* @param endpoint If reason for leaving is decommission, endpoint is the leaving node.
* @param pieces STATE_LEFT,token
*/
private void handleStateLeft(InetAddress endpoint, String[] pieces)
{
assert pieces.length >= 2;
Collection tokens = getTokensFor(endpoint);
if (logger.isDebugEnabled())
logger.debug("Node {} state left, tokens {}", endpoint, tokens);
excise(tokens, endpoint, extractExpireTime(pieces));
}
/**
* Handle node moving inside the ring.
*
* @param endpoint moving endpoint address
* @param pieces STATE_MOVING, token
*/
private void handleStateMoving(InetAddress endpoint, String[] pieces)
{
assert pieces.length >= 2;
Token token = getTokenFactory().fromString(pieces[1]);
if (logger.isDebugEnabled())
logger.debug("Node {} state moving, new token {}", endpoint, token);
tokenMetadata.addMovingEndpoint(token, endpoint);
PendingRangeCalculatorService.instance.update();
}
/**
* Handle notification that a node being actively removed from the ring via 'removenode'
*
* @param endpoint node
* @param pieces either REMOVED_TOKEN (node is gone) or REMOVING_TOKEN (replicas need to be restored)
*/
private void handleStateRemoving(InetAddress endpoint, String[] pieces)
{
assert (pieces.length > 0);
if (endpoint.equals(FBUtilities.getBroadcastAddress()))
{
logger.info("Received removenode gossip about myself. Is this node rejoining after an explicit removenode?");
try
{
drain();
}
catch (Exception e)
{
throw new RuntimeException(e);
}
return;
}
if (tokenMetadata.isMember(endpoint))
{
String state = pieces[0];
Collection removeTokens = tokenMetadata.getTokens(endpoint);
if (VersionedValue.REMOVED_TOKEN.equals(state))
{
excise(removeTokens, endpoint, extractExpireTime(pieces));
}
else if (VersionedValue.REMOVING_TOKEN.equals(state))
{
if (logger.isDebugEnabled())
logger.debug("Tokens {} removed manually (endpoint was {})", removeTokens, endpoint);
// Note that the endpoint is being removed
tokenMetadata.addLeavingEndpoint(endpoint);
PendingRangeCalculatorService.instance.update();
// find the endpoint coordinating this removal that we need to notify when we're done
String[] coordinator = splitValue(Gossiper.instance.getEndpointStateForEndpoint(endpoint).getApplicationState(ApplicationState.REMOVAL_COORDINATOR));
UUID hostId = UUID.fromString(coordinator[1]);
// grab any data we are now responsible for and notify responsible node
restoreReplicaCount(endpoint, tokenMetadata.getEndpointForHostId(hostId));
}
}
else // now that the gossiper has told us about this nonexistent member, notify the gossiper to remove it
{
if (VersionedValue.REMOVED_TOKEN.equals(pieces[0]))
addExpireTimeIfFound(endpoint, extractExpireTime(pieces));
removeEndpoint(endpoint);
}
}
private void excise(Collection tokens, InetAddress endpoint)
{
logger.info("Removing tokens {} for {}", tokens, endpoint);
UUID hostId = tokenMetadata.getHostId(endpoint);
if (hostId != null && tokenMetadata.isMember(endpoint))
{
// enough time for writes to expire and MessagingService timeout reporter callback to fire, which is where
// hints are mostly written from - using getMinRpcTimeout() / 2 for the interval.
long delay = DatabaseDescriptor.getMinRpcTimeout() + DatabaseDescriptor.getWriteRpcTimeout();
ScheduledExecutors.optionalTasks.schedule(() -> HintsService.instance.excise(hostId), delay, TimeUnit.MILLISECONDS);
}
removeEndpoint(endpoint);
tokenMetadata.removeEndpoint(endpoint);
if (!tokens.isEmpty())
tokenMetadata.removeBootstrapTokens(tokens);
notifyLeft(endpoint);
PendingRangeCalculatorService.instance.update();
}
private void excise(Collection tokens, InetAddress endpoint, long expireTime)
{
addExpireTimeIfFound(endpoint, expireTime);
excise(tokens, endpoint);
}
/** unlike excise we just need this endpoint gone without going through any notifications **/
private void removeEndpoint(InetAddress endpoint)
{
Gossiper.instance.removeEndpoint(endpoint);
SystemKeyspace.removeEndpoint(endpoint);
}
protected void addExpireTimeIfFound(InetAddress endpoint, long expireTime)
{
if (expireTime != 0L)
{
Gossiper.instance.addExpireTimeForEndpoint(endpoint, expireTime);
}
}
protected long extractExpireTime(String[] pieces)
{
return Long.parseLong(pieces[2]);
}
/**
* Finds living endpoints responsible for the given ranges
*
* @param keyspaceName the keyspace ranges belong to
* @param ranges the ranges to find sources for
* @return multimap of addresses to ranges the address is responsible for
*/
private Multimap> getNewSourceRanges(String keyspaceName, Set> ranges)
{
InetAddress myAddress = FBUtilities.getBroadcastAddress();
Multimap, InetAddress> rangeAddresses = Keyspace.open(keyspaceName).getReplicationStrategy().getRangeAddresses(tokenMetadata.cloneOnlyTokenMap());
Multimap> sourceRanges = HashMultimap.create();
IFailureDetector failureDetector = FailureDetector.instance;
// find alive sources for our new ranges
for (Range range : ranges)
{
Collection possibleRanges = rangeAddresses.get(range);
IEndpointSnitch snitch = DatabaseDescriptor.getEndpointSnitch();
List sources = snitch.getSortedListByProximity(myAddress, possibleRanges);
assert (!sources.contains(myAddress));
for (InetAddress source : sources)
{
if (failureDetector.isAlive(source))
{
sourceRanges.put(source, range);
break;
}
}
}
return sourceRanges;
}
/**
* Sends a notification to a node indicating we have finished replicating data.
*
* @param remote node to send notification to
*/
private void sendReplicationNotification(InetAddress remote)
{
// notify the remote token
MessageOut msg = new MessageOut(MessagingService.Verb.REPLICATION_FINISHED);
IFailureDetector failureDetector = FailureDetector.instance;
if (logger.isDebugEnabled())
logger.debug("Notifying {} of replication completion\n", remote);
while (failureDetector.isAlive(remote))
{
AsyncOneResponse iar = MessagingService.instance().sendRR(msg, remote);
try
{
iar.get(DatabaseDescriptor.getRpcTimeout(), TimeUnit.MILLISECONDS);
return; // done
}
catch(TimeoutException e)
{
// try again
}
}
}
/**
* Called when an endpoint is removed from the ring. This function checks
* whether this node becomes responsible for new ranges as a
* consequence and streams data if needed.
*
* This is rather ineffective, but it does not matter so much
* since this is called very seldom
*
* @param endpoint the node that left
*/
private void restoreReplicaCount(InetAddress endpoint, final InetAddress notifyEndpoint)
{
Multimap>>> rangesToFetch = HashMultimap.create();
InetAddress myAddress = FBUtilities.getBroadcastAddress();
for (String keyspaceName : Schema.instance.getNonLocalStrategyKeyspaces())
{
Multimap, InetAddress> changedRanges = getChangedRangesForLeaving(keyspaceName, endpoint);
Set> myNewRanges = new HashSet<>();
for (Map.Entry, InetAddress> entry : changedRanges.entries())
{
if (entry.getValue().equals(myAddress))
myNewRanges.add(entry.getKey());
}
Multimap> sourceRanges = getNewSourceRanges(keyspaceName, myNewRanges);
for (Map.Entry>> entry : sourceRanges.asMap().entrySet())
{
rangesToFetch.put(keyspaceName, entry);
}
}
StreamPlan stream = new StreamPlan("Restore replica count");
for (String keyspaceName : rangesToFetch.keySet())
{
for (Map.Entry>> entry : rangesToFetch.get(keyspaceName))
{
InetAddress source = entry.getKey();
InetAddress preferred = SystemKeyspace.getPreferredIP(source);
Collection> ranges = entry.getValue();
if (logger.isDebugEnabled())
logger.debug("Requesting from {} ranges {}", source, StringUtils.join(ranges, ", "));
stream.requestRanges(source, preferred, keyspaceName, ranges);
}
}
StreamResultFuture future = stream.execute();
Futures.addCallback(future, new FutureCallback()
{
public void onSuccess(StreamState finalState)
{
sendReplicationNotification(notifyEndpoint);
}
public void onFailure(Throwable t)
{
logger.warn("Streaming to restore replica count failed", t);
// We still want to send the notification
sendReplicationNotification(notifyEndpoint);
}
});
}
// needs to be modified to accept either a keyspace or ARS.
private Multimap, InetAddress> getChangedRangesForLeaving(String keyspaceName, InetAddress endpoint)
{
// First get all ranges the leaving endpoint is responsible for
Collection> ranges = getRangesForEndpoint(keyspaceName, endpoint);
if (logger.isDebugEnabled())
logger.debug("Node {} ranges [{}]", endpoint, StringUtils.join(ranges, ", "));
Map, List> currentReplicaEndpoints = new HashMap<>(ranges.size());
// Find (for each range) all nodes that store replicas for these ranges as well
TokenMetadata metadata = tokenMetadata.cloneOnlyTokenMap(); // don't do this in the loop! #7758
for (Range range : ranges)
currentReplicaEndpoints.put(range, Keyspace.open(keyspaceName).getReplicationStrategy().calculateNaturalEndpoints(range.right, metadata));
TokenMetadata temp = tokenMetadata.cloneAfterAllLeft();
// endpoint might or might not be 'leaving'. If it was not leaving (that is, removenode
// command was used), it is still present in temp and must be removed.
if (temp.isMember(endpoint))
temp.removeEndpoint(endpoint);
Multimap, InetAddress> changedRanges = HashMultimap.create();
// Go through the ranges and for each range check who will be
// storing replicas for these ranges when the leaving endpoint
// is gone. Whoever is present in newReplicaEndpoints list, but
// not in the currentReplicaEndpoints list, will be needing the
// range.
for (Range range : ranges)
{
Collection newReplicaEndpoints = Keyspace.open(keyspaceName).getReplicationStrategy().calculateNaturalEndpoints(range.right, temp);
newReplicaEndpoints.removeAll(currentReplicaEndpoints.get(range));
if (logger.isDebugEnabled())
if (newReplicaEndpoints.isEmpty())
logger.debug("Range {} already in all replicas", range);
else
logger.debug("Range {} will be responsibility of {}", range, StringUtils.join(newReplicaEndpoints, ", "));
changedRanges.putAll(range, newReplicaEndpoints);
}
return changedRanges;
}
public void onJoin(InetAddress endpoint, EndpointState epState)
{
for (Map.Entry entry : epState.states())
{
onChange(endpoint, entry.getKey(), entry.getValue());
}
MigrationManager.instance.scheduleSchemaPull(endpoint, epState);
}
public void onAlive(InetAddress endpoint, EndpointState state)
{
MigrationManager.instance.scheduleSchemaPull(endpoint, state);
if (tokenMetadata.isMember(endpoint))
notifyUp(endpoint);
}
public void onRemove(InetAddress endpoint)
{
tokenMetadata.removeEndpoint(endpoint);
PendingRangeCalculatorService.instance.update();
}
public void onDead(InetAddress endpoint, EndpointState state)
{
MessagingService.instance().convict(endpoint);
notifyDown(endpoint);
}
public void onRestart(InetAddress endpoint, EndpointState state)
{
// If we have restarted before the node was even marked down, we need to reset the connection pool
if (state.isAlive())
onDead(endpoint, state);
// Then, the node may have been upgraded and changed its messaging protocol version. If so, we
// want to update that before we mark the node live again to avoid problems like CASSANDRA-11128.
VersionedValue netVersion = state.getApplicationState(ApplicationState.NET_VERSION);
if (netVersion != null)
updateNetVersion(endpoint, netVersion);
}
public String getLoadString()
{
return FileUtils.stringifyFileSize(StorageMetrics.load.getCount());
}
public Map getLoadMap()
{
Map map = new HashMap<>();
for (Map.Entry entry : LoadBroadcaster.instance.getLoadInfo().entrySet())
{
map.put(entry.getKey().getHostAddress(), FileUtils.stringifyFileSize(entry.getValue()));
}
// gossiper doesn't see its own updates, so we need to special-case the local node
map.put(FBUtilities.getBroadcastAddress().getHostAddress(), getLoadString());
return map;
}
// TODO
public final void deliverHints(String host)
{
throw new UnsupportedOperationException();
}
public Collection getLocalTokens()
{
Collection tokens = SystemKeyspace.getSavedTokens();
assert tokens != null && !tokens.isEmpty(); // should not be called before initServer sets this
return tokens;
}
@Nullable
public InetAddress getEndpointForHostId(UUID hostId)
{
return tokenMetadata.getEndpointForHostId(hostId);
}
@Nullable
public UUID getHostIdForEndpoint(InetAddress address)
{
return tokenMetadata.getHostId(address);
}
/* These methods belong to the MBean interface */
public List getTokens()
{
return getTokens(FBUtilities.getBroadcastAddress());
}
public List getTokens(String endpoint) throws UnknownHostException
{
return getTokens(InetAddress.getByName(endpoint));
}
private List getTokens(InetAddress endpoint)
{
List strTokens = new ArrayList<>();
for (Token tok : getTokenMetadata().getTokens(endpoint))
strTokens.add(tok.toString());
return strTokens;
}
public String getReleaseVersion()
{
return FBUtilities.getReleaseVersionString();
}
public String getSchemaVersion()
{
return Schema.instance.getVersion().toString();
}
public List getLeavingNodes()
{
return stringify(tokenMetadata.getLeavingEndpoints());
}
public List getMovingNodes()
{
List endpoints = new ArrayList<>();
for (Pair node : tokenMetadata.getMovingEndpoints())
{
endpoints.add(node.right.getHostAddress());
}
return endpoints;
}
public List getJoiningNodes()
{
return stringify(tokenMetadata.getBootstrapTokens().valueSet());
}
public List getLiveNodes()
{
return stringify(Gossiper.instance.getLiveMembers());
}
public Set getLiveRingMembers()
{
return getLiveRingMembers(false);
}
public Set getLiveRingMembers(boolean excludeDeadStates)
{
Set ret = new HashSet<>();
for (InetAddress ep : Gossiper.instance.getLiveMembers())
{
if (excludeDeadStates)
{
EndpointState epState = Gossiper.instance.getEndpointStateForEndpoint(ep);
if (epState == null || Gossiper.instance.isDeadState(epState))
continue;
}
if (tokenMetadata.isMember(ep))
ret.add(ep);
}
return ret;
}
public List getUnreachableNodes()
{
return stringify(Gossiper.instance.getUnreachableMembers());
}
public String[] getAllDataFileLocations()
{
String[] locations = DatabaseDescriptor.getAllDataFileLocations();
for (int i = 0; i < locations.length; i++)
locations[i] = FileUtils.getCanonicalPath(locations[i]);
return locations;
}
public String getCommitLogLocation()
{
return FileUtils.getCanonicalPath(DatabaseDescriptor.getCommitLogLocation());
}
public String getSavedCachesLocation()
{
return FileUtils.getCanonicalPath(DatabaseDescriptor.getSavedCachesLocation());
}
private List stringify(Iterable endpoints)
{
List stringEndpoints = new ArrayList<>();
for (InetAddress ep : endpoints)
{
stringEndpoints.add(ep.getHostAddress());
}
return stringEndpoints;
}
public int getCurrentGenerationNumber()
{
return Gossiper.instance.getCurrentGenerationNumber(FBUtilities.getBroadcastAddress());
}
public int forceKeyspaceCleanup(String keyspaceName, String... tables) throws IOException, ExecutionException, InterruptedException
{
return forceKeyspaceCleanup(0, keyspaceName, tables);
}
public int forceKeyspaceCleanup(int jobs, String keyspaceName, String... tables) throws IOException, ExecutionException, InterruptedException
{
if (SchemaConstants.isLocalSystemKeyspace(keyspaceName))
throw new RuntimeException("Cleanup of the system keyspace is neither necessary nor wise");
CompactionManager.AllSSTableOpStatus status = CompactionManager.AllSSTableOpStatus.SUCCESSFUL;
for (ColumnFamilyStore cfStore : getValidColumnFamilies(false, false, keyspaceName, tables))
{
CompactionManager.AllSSTableOpStatus oneStatus = cfStore.forceCleanup(jobs);
if (oneStatus != CompactionManager.AllSSTableOpStatus.SUCCESSFUL)
status = oneStatus;
}
return status.statusCode;
}
public int scrub(boolean disableSnapshot, boolean skipCorrupted, String keyspaceName, String... tables) throws IOException, ExecutionException, InterruptedException
{
return scrub(disableSnapshot, skipCorrupted, true, 0, keyspaceName, tables);
}
public int scrub(boolean disableSnapshot, boolean skipCorrupted, boolean checkData, String keyspaceName, String... tables) throws IOException, ExecutionException, InterruptedException
{
return scrub(disableSnapshot, skipCorrupted, checkData, 0, keyspaceName, tables);
}
public int scrub(boolean disableSnapshot, boolean skipCorrupted, boolean checkData, int jobs, String keyspaceName, String... tables) throws IOException, ExecutionException, InterruptedException
{
return scrub(disableSnapshot, skipCorrupted, checkData, false, jobs, keyspaceName, tables);
}
public int scrub(boolean disableSnapshot, boolean skipCorrupted, boolean checkData, boolean reinsertOverflowedTTL, int jobs, String keyspaceName, String... tables) throws IOException, ExecutionException, InterruptedException
{
CompactionManager.AllSSTableOpStatus status = CompactionManager.AllSSTableOpStatus.SUCCESSFUL;
for (ColumnFamilyStore cfStore : getValidColumnFamilies(true, false, keyspaceName, tables))
{
CompactionManager.AllSSTableOpStatus oneStatus = cfStore.scrub(disableSnapshot, skipCorrupted, reinsertOverflowedTTL, checkData, jobs);
if (oneStatus != CompactionManager.AllSSTableOpStatus.SUCCESSFUL)
status = oneStatus;
}
return status.statusCode;
}
public int verify(boolean extendedVerify, String keyspaceName, String... tableNames) throws IOException, ExecutionException, InterruptedException
{
CompactionManager.AllSSTableOpStatus status = CompactionManager.AllSSTableOpStatus.SUCCESSFUL;
for (ColumnFamilyStore cfStore : getValidColumnFamilies(false, false, keyspaceName, tableNames))
{
CompactionManager.AllSSTableOpStatus oneStatus = cfStore.verify(extendedVerify);
if (oneStatus != CompactionManager.AllSSTableOpStatus.SUCCESSFUL)
status = oneStatus;
}
return status.statusCode;
}
public int upgradeSSTables(String keyspaceName, boolean excludeCurrentVersion, String... tableNames) throws IOException, ExecutionException, InterruptedException
{
return upgradeSSTables(keyspaceName, excludeCurrentVersion, 0, tableNames);
}
public int upgradeSSTables(String keyspaceName, boolean excludeCurrentVersion, int jobs, String... tableNames) throws IOException, ExecutionException, InterruptedException
{
CompactionManager.AllSSTableOpStatus status = CompactionManager.AllSSTableOpStatus.SUCCESSFUL;
for (ColumnFamilyStore cfStore : getValidColumnFamilies(true, true, keyspaceName, tableNames))
{
CompactionManager.AllSSTableOpStatus oneStatus = cfStore.sstablesRewrite(excludeCurrentVersion, jobs);
if (oneStatus != CompactionManager.AllSSTableOpStatus.SUCCESSFUL)
status = oneStatus;
}
return status.statusCode;
}
public void forceKeyspaceCompaction(boolean splitOutput, String keyspaceName, String... tableNames) throws IOException, ExecutionException, InterruptedException
{
for (ColumnFamilyStore cfStore : getValidColumnFamilies(true, false, keyspaceName, tableNames))
{
cfStore.forceMajorCompaction(splitOutput);
}
}
public int relocateSSTables(String keyspaceName, String ... columnFamilies) throws IOException, ExecutionException, InterruptedException
{
return relocateSSTables(0, keyspaceName, columnFamilies);
}
public int relocateSSTables(int jobs, String keyspaceName, String ... columnFamilies) throws IOException, ExecutionException, InterruptedException
{
CompactionManager.AllSSTableOpStatus status = CompactionManager.AllSSTableOpStatus.SUCCESSFUL;
for (ColumnFamilyStore cfs : getValidColumnFamilies(false, false, keyspaceName, columnFamilies))
{
CompactionManager.AllSSTableOpStatus oneStatus = cfs.relocateSSTables(jobs);
if (oneStatus != CompactionManager.AllSSTableOpStatus.SUCCESSFUL)
status = oneStatus;
}
return status.statusCode;
}
public int garbageCollect(String tombstoneOptionString, int jobs, String keyspaceName, String ... columnFamilies) throws IOException, ExecutionException, InterruptedException
{
TombstoneOption tombstoneOption = TombstoneOption.valueOf(tombstoneOptionString);
CompactionManager.AllSSTableOpStatus status = CompactionManager.AllSSTableOpStatus.SUCCESSFUL;
for (ColumnFamilyStore cfs : getValidColumnFamilies(false, false, keyspaceName, columnFamilies))
{
CompactionManager.AllSSTableOpStatus oneStatus = cfs.garbageCollect(tombstoneOption, jobs);
if (oneStatus != CompactionManager.AllSSTableOpStatus.SUCCESSFUL)
status = oneStatus;
}
return status.statusCode;
}
/**
* Takes the snapshot of a multiple column family from different keyspaces. A snapshot name must be specified.
*
* @param tag
* the tag given to the snapshot; may not be null or empty
* @param options
* Map of options (skipFlush is the only supported option for now)
* @param entities
* list of keyspaces / tables in the form of empty | ks1 ks2 ... | ks1.cf1,ks2.cf2,...
*/
@Override
public void takeSnapshot(String tag, Map options, String... entities) throws IOException
{
boolean skipFlush = Boolean.parseBoolean(options.getOrDefault("skipFlush", "false"));
if (entities != null && entities.length > 0 && entities[0].contains("."))
{
takeMultipleTableSnapshot(tag, skipFlush, entities);
}
else
{
takeSnapshot(tag, skipFlush, entities);
}
}
/**
* Takes the snapshot of a specific table. A snapshot name must be
* specified.
*
* @param keyspaceName
* the keyspace which holds the specified table
* @param tableName
* the table to snapshot
* @param tag
* the tag given to the snapshot; may not be null or empty
*/
public void takeTableSnapshot(String keyspaceName, String tableName, String tag)
throws IOException
{
takeMultipleTableSnapshot(tag, false, keyspaceName + "." + tableName);
}
public void forceKeyspaceCompactionForTokenRange(String keyspaceName, String startToken, String endToken, String... tableNames) throws IOException, ExecutionException, InterruptedException
{
Collection> tokenRanges = createRepairRangeFrom(startToken, endToken);
for (ColumnFamilyStore cfStore : getValidColumnFamilies(true, false, keyspaceName, tableNames))
{
cfStore.forceCompactionForTokenRange(tokenRanges);
}
}
/**
* Takes the snapshot for the given keyspaces. A snapshot name must be specified.
*
* @param tag the tag given to the snapshot; may not be null or empty
* @param keyspaceNames the names of the keyspaces to snapshot; empty means "all."
*/
public void takeSnapshot(String tag, String... keyspaceNames) throws IOException
{
takeSnapshot(tag, false, keyspaceNames);
}
/**
* Takes the snapshot of a multiple column family from different keyspaces. A snapshot name must be specified.
*
* @param tag
* the tag given to the snapshot; may not be null or empty
* @param tableList
* list of tables from different keyspace in the form of ks1.cf1 ks2.cf2
*/
public void takeMultipleTableSnapshot(String tag, String... tableList)
throws IOException
{
takeMultipleTableSnapshot(tag, false, tableList);
}
/**
* Takes the snapshot for the given keyspaces. A snapshot name must be specified.
*
* @param tag the tag given to the snapshot; may not be null or empty
* @param skipFlush Skip blocking flush of memtable
* @param keyspaceNames the names of the keyspaces to snapshot; empty means "all."
*/
private void takeSnapshot(String tag, boolean skipFlush, String... keyspaceNames) throws IOException
{
if (operationMode == Mode.JOINING)
throw new IOException("Cannot snapshot until bootstrap completes");
if (tag == null || tag.equals(""))
throw new IOException("You must supply a snapshot name.");
Iterable keyspaces;
if (keyspaceNames.length == 0)
{
keyspaces = Keyspace.all();
}
else
{
ArrayList t = new ArrayList<>(keyspaceNames.length);
for (String keyspaceName : keyspaceNames)
t.add(getValidKeyspace(keyspaceName));
keyspaces = t;
}
// Do a check to see if this snapshot exists before we actually snapshot
for (Keyspace keyspace : keyspaces)
if (keyspace.snapshotExists(tag))
throw new IOException("Snapshot " + tag + " already exists.");
for (Keyspace keyspace : keyspaces)
keyspace.snapshot(tag, null, skipFlush);
}
/**
* Takes the snapshot of a multiple column family from different keyspaces. A snapshot name must be specified.
*
*
* @param tag
* the tag given to the snapshot; may not be null or empty
* @param skipFlush
* Skip blocking flush of memtable
* @param tableList
* list of tables from different keyspace in the form of ks1.cf1 ks2.cf2
*/
private void takeMultipleTableSnapshot(String tag, boolean skipFlush, String... tableList)
throws IOException
{
Map> keyspaceColumnfamily = new HashMap>();
for (String table : tableList)
{
String splittedString[] = StringUtils.split(table, '.');
if (splittedString.length == 2)
{
String keyspaceName = splittedString[0];
String tableName = splittedString[1];
if (keyspaceName == null)
throw new IOException("You must supply a keyspace name");
if (operationMode.equals(Mode.JOINING))
throw new IOException("Cannot snapshot until bootstrap completes");
if (tableName == null)
throw new IOException("You must supply a table name");
if (tag == null || tag.equals(""))
throw new IOException("You must supply a snapshot name.");
Keyspace keyspace = getValidKeyspace(keyspaceName);
ColumnFamilyStore columnFamilyStore = keyspace.getColumnFamilyStore(tableName);
// As there can be multiple column family from same keyspace check if snapshot exist for that specific
// columnfamily and not for whole keyspace
if (columnFamilyStore.snapshotExists(tag))
throw new IOException("Snapshot " + tag + " already exists.");
if (!keyspaceColumnfamily.containsKey(keyspace))
{
keyspaceColumnfamily.put(keyspace, new ArrayList());
}
// Add Keyspace columnfamily to map in order to support atomicity for snapshot process.
// So no snapshot should happen if any one of the above conditions fail for any keyspace or columnfamily
keyspaceColumnfamily.get(keyspace).add(tableName);
}
else
{
throw new IllegalArgumentException(
"Cannot take a snapshot on secondary index or invalid column family name. You must supply a column family name in the form of keyspace.columnfamily");
}
}
for (Entry> entry : keyspaceColumnfamily.entrySet())
{
for (String table : entry.getValue())
entry.getKey().snapshot(tag, table, skipFlush);
}
}
private Keyspace getValidKeyspace(String keyspaceName) throws IOException
{
if (!Schema.instance.getKeyspaces().contains(keyspaceName))
{
throw new IOException("Keyspace " + keyspaceName + " does not exist");
}
return Keyspace.open(keyspaceName);
}
/**
* Remove the snapshot with the given name from the given keyspaces.
* If no tag is specified we will remove all snapshots.
*/
public void clearSnapshot(String tag, String... keyspaceNames) throws IOException
{
if(tag == null)
tag = "";
Set keyspaces = new HashSet<>();
for (String dataDir : DatabaseDescriptor.getAllDataFileLocations())
{
for(String keyspaceDir : new File(dataDir).list())
{
// Only add a ks if it has been specified as a param, assuming params were actually provided.
if (keyspaceNames.length > 0 && !Arrays.asList(keyspaceNames).contains(keyspaceDir))
continue;
keyspaces.add(keyspaceDir);
}
}
for (String keyspace : keyspaces)
Keyspace.clearSnapshot(tag, keyspace);
if (logger.isDebugEnabled())
logger.debug("Cleared out snapshot directories");
}
public Map getSnapshotDetails()
{
Map snapshotMap = new HashMap<>();
for (Keyspace keyspace : Keyspace.all())
{
if (SchemaConstants.isLocalSystemKeyspace(keyspace.getName()))
continue;
for (ColumnFamilyStore cfStore : keyspace.getColumnFamilyStores())
{
for (Map.Entry> snapshotDetail : cfStore.getSnapshotDetails().entrySet())
{
TabularDataSupport data = (TabularDataSupport)snapshotMap.get(snapshotDetail.getKey());
if (data == null)
{
data = new TabularDataSupport(SnapshotDetailsTabularData.TABULAR_TYPE);
snapshotMap.put(snapshotDetail.getKey(), data);
}
SnapshotDetailsTabularData.from(snapshotDetail.getKey(), keyspace.getName(), cfStore.getColumnFamilyName(), snapshotDetail, data);
}
}
}
return snapshotMap;
}
public long trueSnapshotsSize()
{
long total = 0;
for (Keyspace keyspace : Keyspace.all())
{
if (SchemaConstants.isLocalSystemKeyspace(keyspace.getName()))
continue;
for (ColumnFamilyStore cfStore : keyspace.getColumnFamilyStores())
{
total += cfStore.trueSnapshotsSize();
}
}
return total;
}
public void refreshSizeEstimates() throws ExecutionException
{
FBUtilities.waitOnFuture(ScheduledExecutors.optionalTasks.submit(SizeEstimatesRecorder.instance));
}
/**
* @param allowIndexes Allow index CF names to be passed in
* @param autoAddIndexes Automatically add secondary indexes if a CF has them
* @param keyspaceName keyspace
* @param cfNames CFs
* @throws java.lang.IllegalArgumentException when given CF name does not exist
*/
public Iterable getValidColumnFamilies(boolean allowIndexes, boolean autoAddIndexes, String keyspaceName, String... cfNames) throws IOException
{
Keyspace keyspace = getValidKeyspace(keyspaceName);
return keyspace.getValidColumnFamilies(allowIndexes, autoAddIndexes, cfNames);
}
/**
* Flush all memtables for a keyspace and column families.
* @param keyspaceName
* @param tableNames
* @throws IOException
*/
public void forceKeyspaceFlush(String keyspaceName, String... tableNames) throws IOException
{
for (ColumnFamilyStore cfStore : getValidColumnFamilies(true, false, keyspaceName, tableNames))
{
logger.debug("Forcing flush on keyspace {}, CF {}", keyspaceName, cfStore.name);
cfStore.forceBlockingFlush();
}
}
public int repairAsync(String keyspace, Map repairSpec)
{
RepairOption option = RepairOption.parse(repairSpec, tokenMetadata.partitioner);
// if ranges are not specified
if (option.getRanges().isEmpty())
{
if (option.isPrimaryRange())
{
// when repairing only primary range, neither dataCenters nor hosts can be set
if (option.getDataCenters().isEmpty() && option.getHosts().isEmpty())
option.getRanges().addAll(getPrimaryRanges(keyspace));
// except dataCenters only contain local DC (i.e. -local)
else if (option.isInLocalDCOnly())
option.getRanges().addAll(getPrimaryRangesWithinDC(keyspace));
else
throw new IllegalArgumentException("You need to run primary range repair on all nodes in the cluster.");
}
else
{
option.getRanges().addAll(getLocalRanges(keyspace));
}
}
return forceRepairAsync(keyspace, option, false);
}
@Deprecated
public int forceRepairAsync(String keyspace,
boolean isSequential,
Collection dataCenters,
Collection hosts,
boolean primaryRange,
boolean fullRepair,
String... tableNames)
{
return forceRepairAsync(keyspace, isSequential ? RepairParallelism.SEQUENTIAL.ordinal() : RepairParallelism.PARALLEL.ordinal(), dataCenters, hosts, primaryRange, fullRepair, tableNames);
}
@Deprecated
public int forceRepairAsync(String keyspace,
int parallelismDegree,
Collection dataCenters,
Collection hosts,
boolean primaryRange,
boolean fullRepair,
String... tableNames)
{
if (parallelismDegree < 0 || parallelismDegree > RepairParallelism.values().length - 1)
{
throw new IllegalArgumentException("Invalid parallelism degree specified: " + parallelismDegree);
}
RepairParallelism parallelism = RepairParallelism.values()[parallelismDegree];
if (FBUtilities.isWindows && parallelism != RepairParallelism.PARALLEL)
{
logger.warn("Snapshot-based repair is not yet supported on Windows. Reverting to parallel repair.");
parallelism = RepairParallelism.PARALLEL;
}
RepairOption options = new RepairOption(parallelism, primaryRange, !fullRepair, false, 1, Collections.>emptyList(), false, false);
if (dataCenters != null)
{
options.getDataCenters().addAll(dataCenters);
}
if (hosts != null)
{
options.getHosts().addAll(hosts);
}
if (primaryRange)
{
// when repairing only primary range, neither dataCenters nor hosts can be set
if (options.getDataCenters().isEmpty() && options.getHosts().isEmpty())
options.getRanges().addAll(getPrimaryRanges(keyspace));
// except dataCenters only contain local DC (i.e. -local)
else if (options.getDataCenters().size() == 1 && options.getDataCenters().contains(DatabaseDescriptor.getLocalDataCenter()))
options.getRanges().addAll(getPrimaryRangesWithinDC(keyspace));
else
throw new IllegalArgumentException("You need to run primary range repair on all nodes in the cluster.");
}
else
{
options.getRanges().addAll(getLocalRanges(keyspace));
}
if (tableNames != null)
{
for (String table : tableNames)
{
options.getColumnFamilies().add(table);
}
}
return forceRepairAsync(keyspace, options, true);
}
@Deprecated
public int forceRepairAsync(String keyspace,
boolean isSequential,
boolean isLocal,
boolean primaryRange,
boolean fullRepair,
String... tableNames)
{
Set dataCenters = null;
if (isLocal)
{
dataCenters = Sets.newHashSet(DatabaseDescriptor.getLocalDataCenter());
}
return forceRepairAsync(keyspace, isSequential, dataCenters, null, primaryRange, fullRepair, tableNames);
}
@Deprecated
public int forceRepairRangeAsync(String beginToken,
String endToken,
String keyspaceName,
boolean isSequential,
Collection dataCenters,
Collection hosts,
boolean fullRepair,
String... tableNames)
{
return forceRepairRangeAsync(beginToken, endToken, keyspaceName,
isSequential ? RepairParallelism.SEQUENTIAL.ordinal() : RepairParallelism.PARALLEL.ordinal(),
dataCenters, hosts, fullRepair, tableNames);
}
@Deprecated
public int forceRepairRangeAsync(String beginToken,
String endToken,
String keyspaceName,
int parallelismDegree,
Collection dataCenters,
Collection hosts,
boolean fullRepair,
String... tableNames)
{
if (parallelismDegree < 0 || parallelismDegree > RepairParallelism.values().length - 1)
{
throw new IllegalArgumentException("Invalid parallelism degree specified: " + parallelismDegree);
}
RepairParallelism parallelism = RepairParallelism.values()[parallelismDegree];
if (FBUtilities.isWindows && parallelism != RepairParallelism.PARALLEL)
{
logger.warn("Snapshot-based repair is not yet supported on Windows. Reverting to parallel repair.");
parallelism = RepairParallelism.PARALLEL;
}
if (!fullRepair)
logger.warn("Incremental repair can't be requested with subrange repair " +
"because each subrange repair would generate an anti-compacted table. " +
"The repair will occur but without anti-compaction.");
Collection> repairingRange = createRepairRangeFrom(beginToken, endToken);
RepairOption options = new RepairOption(parallelism, false, !fullRepair, false, 1, repairingRange, true, false);
if (dataCenters != null)
{
options.getDataCenters().addAll(dataCenters);
}
if (hosts != null)
{
options.getHosts().addAll(hosts);
}
if (tableNames != null)
{
for (String table : tableNames)
{
options.getColumnFamilies().add(table);
}
}
logger.info("starting user-requested repair of range {} for keyspace {} and column families {}",
repairingRange, keyspaceName, tableNames);
return forceRepairAsync(keyspaceName, options, true);
}
@Deprecated
public int forceRepairRangeAsync(String beginToken,
String endToken,
String keyspaceName,
boolean isSequential,
boolean isLocal,
boolean fullRepair,
String... tableNames)
{
Set dataCenters = null;
if (isLocal)
{
dataCenters = Sets.newHashSet(DatabaseDescriptor.getLocalDataCenter());
}
return forceRepairRangeAsync(beginToken, endToken, keyspaceName, isSequential, dataCenters, null, fullRepair, tableNames);
}
/**
* Create collection of ranges that match ring layout from given tokens.
*
* @param beginToken beginning token of the range
* @param endToken end token of the range
* @return collection of ranges that match ring layout in TokenMetadata
*/
@VisibleForTesting
Collection> createRepairRangeFrom(String beginToken, String endToken)
{
Token parsedBeginToken = getTokenFactory().fromString(beginToken);
Token parsedEndToken = getTokenFactory().fromString(endToken);
// Break up given range to match ring layout in TokenMetadata
ArrayList> repairingRange = new ArrayList<>();
ArrayList tokens = new ArrayList<>(tokenMetadata.sortedTokens());
if (!tokens.contains(parsedBeginToken))
{
tokens.add(parsedBeginToken);
}
if (!tokens.contains(parsedEndToken))
{
tokens.add(parsedEndToken);
}
// tokens now contain all tokens including our endpoints
Collections.sort(tokens);
int start = tokens.indexOf(parsedBeginToken), end = tokens.indexOf(parsedEndToken);
for (int i = start; i != end; i = (i+1) % tokens.size())
{
Range range = new Range<>(tokens.get(i), tokens.get((i+1) % tokens.size()));
repairingRange.add(range);
}
return repairingRange;
}
public TokenFactory getTokenFactory()
{
return tokenMetadata.partitioner.getTokenFactory();
}
public int forceRepairAsync(String keyspace, RepairOption options, boolean legacy)
{
if (options.getRanges().isEmpty() || Keyspace.open(keyspace).getReplicationStrategy().getReplicationFactor() < 2)
return 0;
int cmd = nextRepairCommand.incrementAndGet();
NamedThreadFactory.createThread(createRepairTask(cmd, keyspace, options, legacy), "Repair-Task-" + threadCounter.incrementAndGet()).start();
return cmd;
}
private FutureTask