/**
* 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.IOException;
import java.io.IOError;
import java.lang.management.ManagementFactory;
import java.lang.reflect.Constructor;
import java.util.*;
import java.util.concurrent.ExecutorService;
import java.util.concurrent.LinkedBlockingQueue;
import java.util.concurrent.TimeUnit;
import java.net.InetAddress;
import javax.management.*;
import org.apache.cassandra.concurrent.*;
import org.apache.cassandra.config.DatabaseDescriptor;
import org.apache.cassandra.db.*;
import org.apache.cassandra.dht.*;
import org.apache.cassandra.gms.*;
import org.apache.cassandra.locator.*;
import org.apache.cassandra.net.*;
import org.apache.cassandra.service.AntiEntropyService.TreeRequestVerbHandler;
import org.apache.cassandra.utils.FileUtils;
import org.apache.cassandra.utils.LogUtil;
import org.apache.cassandra.utils.FBUtilities;
import org.apache.cassandra.io.SSTableReader;
import org.apache.cassandra.io.Streaming;
import org.apache.cassandra.io.StreamRequestVerbHandler;
import org.apache.log4j.Logger;
import org.apache.log4j.Level;
import org.apache.commons.lang.StringUtils;
import com.google.common.base.Function;
import com.google.common.collect.Multimap;
import com.google.common.collect.HashMultimap;
/*
* 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 final class StorageService implements IEndPointStateChangeSubscriber, StorageServiceMBean
{
private static Logger logger_ = Logger.getLogger(StorageService.class);
public final static String MOVE_STATE = "MOVE";
// this must be a char that cannot be present in any token
public final static char Delimiter = ',';
public final static String STATE_BOOTSTRAPPING = "BOOT";
public final static String STATE_NORMAL = "NORMAL";
public final static String STATE_LEAVING = "LEAVING";
public final static String STATE_LEFT = "LEFT";
public final static String REMOVE_TOKEN = "remove";
public final static String LEFT_NORMALLY = "left";
/* All verb handler identifiers */
public final static String mutationVerbHandler_ = "ROW-MUTATION-VERB-HANDLER";
public final static String binaryVerbHandler_ = "BINARY-VERB-HANDLER";
public final static String readRepairVerbHandler_ = "READ-REPAIR-VERB-HANDLER";
public final static String readVerbHandler_ = "ROW-READ-VERB-HANDLER";
public final static String streamInitiateVerbHandler_ = "BOOTSTRAP-INITIATE-VERB-HANDLER";
public final static String streamInitiateDoneVerbHandler_ = "BOOTSTRAP-INITIATE-DONE-VERB-HANDLER";
public final static String streamFinishedVerbHandler_ = "BOOTSTRAP-TERMINATE-VERB-HANDLER";
public final static String dataFileVerbHandler_ = "DATA-FILE-VERB-HANDLER";
public final static String streamRequestVerbHandler_ = "BS-METADATA-VERB-HANDLER";
public final static String rangeVerbHandler_ = "RANGE-VERB-HANDLER";
public final static String rangeSliceVerbHandler_ = "RANGE-SLICE-VERB-HANDLER";
public final static String bootstrapTokenVerbHandler_ = "SPLITS-VERB-HANDLER";
private static IPartitioner partitioner_ = DatabaseDescriptor.getPartitioner();
private static volatile StorageService instance_;
public static IPartitioner getPartitioner() {
return partitioner_;
}
public Collection<Range> getLocalRanges()
{
return getRangesForEndPoint(FBUtilities.getLocalAddress());
}
public Range getLocalPrimaryRange()
{
return getPrimaryRangeForEndPoint(FBUtilities.getLocalAddress());
}
/*
* Factory method that gets an instance of the StorageService
* class.
*/
public static StorageService instance()
{
if (instance_ == null)
{
synchronized (StorageService.class)
{
if (instance_ == null)
{
try
{
instance_ = new StorageService();
}
catch (Throwable th)
{
logger_.error(LogUtil.throwableToString(th));
System.exit(1);
}
}
}
}
return instance_;
}
/*
* This is the endpoint snitch which depends on the network architecture. We
* need to keep this information for each endpoint so that we make decisions
* while doing things like replication etc.
*
*/
private IEndPointSnitch endPointSnitch_;
/* This abstraction maintains the token/endpoint metadata information */
private TokenMetadata tokenMetadata_ = new TokenMetadata();
private SystemTable.StorageMetadata storageMetadata_;
/* This thread pool does consistency checks when the client doesn't care about consistency */
private ExecutorService consistencyManager_ = new DebuggableThreadPoolExecutor(DatabaseDescriptor.getConsistencyThreads(),
DatabaseDescriptor.getConsistencyThreads(),
Integer.MAX_VALUE,
TimeUnit.SECONDS,
new LinkedBlockingQueue<Runnable>(),
new NamedThreadFactory("CONSISTENCY-MANAGER"));
/* We use this interface to determine where replicas need to be placed */
private AbstractReplicationStrategy replicationStrategy_;
/* Are we starting this node in bootstrap mode? */
private boolean isBootstrapMode;
private Set<InetAddress> bootstrapSet;
/* when intialized as a client, we shouldn't write to the system table. */
private boolean isClientMode;
public synchronized void addBootstrapSource(InetAddress s)
{
if (logger_.isDebugEnabled())
logger_.debug("Added " + s + " as a bootstrap source");
bootstrapSet.add(s);
}
public synchronized void removeBootstrapSource(InetAddress s)
{
bootstrapSet.remove(s);
if (logger_.isDebugEnabled())
logger_.debug("Removed " + s + " as a bootstrap source; remaining is [" + StringUtils.join(bootstrapSet, ", ") + "]");
if (bootstrapSet.isEmpty())
{
finishBootstrapping();
}
}
private void finishBootstrapping()
{
isBootstrapMode = false;
SystemTable.setBootstrapped(true);
setToken(getLocalToken());
Gossiper.instance().addApplicationState(MOVE_STATE, new ApplicationState(STATE_NORMAL + Delimiter + partitioner_.getTokenFactory().toString(getLocalToken())));
logger_.info("Bootstrap/move completed! Now serving reads.");
}
/** This method updates the local token on disk */
public void setToken(Token token)
{
if (logger_.isDebugEnabled())
logger_.debug("Setting token to " + token);
SystemTable.updateToken(token);
tokenMetadata_.updateNormalToken(token, FBUtilities.getLocalAddress());
}
public StorageService()
{
MBeanServer mbs = ManagementFactory.getPlatformMBeanServer();
try
{
mbs.registerMBean(this, new ObjectName("org.apache.cassandra.service:type=StorageService"));
}
catch (Exception e)
{
throw new RuntimeException(e);
}
bootstrapSet = new HashSet<InetAddress>();
endPointSnitch_ = DatabaseDescriptor.getEndPointSnitch();
/* register the verb handlers */
MessagingService.instance().registerVerbHandlers(binaryVerbHandler_, new BinaryVerbHandler());
MessagingService.instance().registerVerbHandlers(mutationVerbHandler_, new RowMutationVerbHandler());
MessagingService.instance().registerVerbHandlers(readRepairVerbHandler_, new ReadRepairVerbHandler());
MessagingService.instance().registerVerbHandlers(readVerbHandler_, new ReadVerbHandler());
MessagingService.instance().registerVerbHandlers(dataFileVerbHandler_, new DataFileVerbHandler() );
MessagingService.instance().registerVerbHandlers(rangeVerbHandler_, new RangeVerbHandler());
MessagingService.instance().registerVerbHandlers(rangeSliceVerbHandler_, new RangeSliceVerbHandler());
// see BootStrapper for a summary of how the bootstrap verbs interact
MessagingService.instance().registerVerbHandlers(bootstrapTokenVerbHandler_, new BootStrapper.BootstrapTokenVerbHandler());
MessagingService.instance().registerVerbHandlers(streamRequestVerbHandler_, new StreamRequestVerbHandler() );
MessagingService.instance().registerVerbHandlers(streamInitiateVerbHandler_, new Streaming.StreamInitiateVerbHandler());
MessagingService.instance().registerVerbHandlers(streamInitiateDoneVerbHandler_, new Streaming.StreamInitiateDoneVerbHandler());
MessagingService.instance().registerVerbHandlers(streamFinishedVerbHandler_, new Streaming.StreamFinishedVerbHandler());
replicationStrategy_ = getReplicationStrategy(tokenMetadata_, partitioner_);
}
public static AbstractReplicationStrategy getReplicationStrategy(TokenMetadata tokenMetadata, IPartitioner partitioner)
{
AbstractReplicationStrategy replicationStrategy = null;
Class<AbstractReplicationStrategy> cls = DatabaseDescriptor.getReplicaPlacementStrategyClass();
Class [] parameterTypes = new Class[] { TokenMetadata.class, IPartitioner.class, int.class};
try
{
Constructor<AbstractReplicationStrategy> constructor = cls.getConstructor(parameterTypes);
replicationStrategy = constructor.newInstance(tokenMetadata, partitioner, DatabaseDescriptor.getReplicationFactor());
}
catch (Exception e)
{
throw new RuntimeException(e);
}
return replicationStrategy;
}
public void stopClient()
{
Gossiper.instance().unregister(this);
Gossiper.instance().stop();
MessagingService.shutdown();
}
public void initClient() throws IOException
{
isClientMode = true;
logger_.info("Starting up client gossip");
MessagingService.instance().listen(FBUtilities.getLocalAddress());
MessagingService.instance().listenUDP(FBUtilities.getLocalAddress());
SelectorManager.getSelectorManager().start();
SelectorManager.getUdpSelectorManager().start();
Gossiper.instance().register(this);
Gossiper.instance().start(FBUtilities.getLocalAddress(), (int)(System.currentTimeMillis() / 1000)); // needed for node-ring gathering.
}
public void initServer() throws IOException
{
isClientMode = false;
storageMetadata_ = SystemTable.initMetadata();
DatabaseDescriptor.createAllDirectories();
logger_.info("Starting up server gossip");
/* Listen for application messages */
MessagingService.instance().listen(FBUtilities.getLocalAddress());
/* Listen for control messages */
MessagingService.instance().listenUDP(FBUtilities.getLocalAddress());
SelectorManager.getSelectorManager().start();
SelectorManager.getUdpSelectorManager().start();
AntiEntropyService.instance();
StorageLoadBalancer.instance().startBroadcasting();
// 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 nodeId to our state, below.)
Gossiper.instance().register(this);
Gossiper.instance().start(FBUtilities.getLocalAddress(), storageMetadata_.getGeneration()); // needed for node-ring gathering.
if (DatabaseDescriptor.isAutoBootstrap()
&& !(DatabaseDescriptor.getSeeds().contains(FBUtilities.getLocalAddress()) || SystemTable.isBootstrapped()))
{
logger_.info("Starting in bootstrap mode (first, sleeping to get load information)");
StorageLoadBalancer.instance().waitForLoadInfo();
logger_.info("... got load info");
Token token = BootStrapper.getBootstrapToken(tokenMetadata_, StorageLoadBalancer.instance().getLoadInfo());
startBootstrap(token);
// don't finish startup (enabling thrift) until after bootstrap is done
while (isBootstrapMode)
{
try
{
Thread.sleep(100);
}
catch (InterruptedException e)
{
throw new AssertionError(e);
}
}
}
else
{
SystemTable.setBootstrapped(true);
Token token = storageMetadata_.getToken();
tokenMetadata_.updateNormalToken(token, FBUtilities.getLocalAddress());
Gossiper.instance().addApplicationState(MOVE_STATE, new ApplicationState(STATE_NORMAL + Delimiter + partitioner_.getTokenFactory().toString(token)));
}
assert tokenMetadata_.sortedTokens().size() > 0;
}
private void startBootstrap(Token token) throws IOException
{
isBootstrapMode = true;
SystemTable.updateToken(token); // DON'T use setToken, that makes us part of the ring locally which is incorrect until we are done bootstrapping
Gossiper.instance().addApplicationState(MOVE_STATE, new ApplicationState(STATE_BOOTSTRAPPING + Delimiter + partitioner_.getTokenFactory().toString(token)));
logger_.info("bootstrap sleeping " + Streaming.RING_DELAY);
try
{
Thread.sleep(Streaming.RING_DELAY);
}
catch (InterruptedException e)
{
throw new AssertionError(e);
}
new BootStrapper(replicationStrategy_, FBUtilities.getLocalAddress(), token, tokenMetadata_).startBootstrap(); // handles token update
}
public boolean isBootstrapMode()
{
return isBootstrapMode;
}
public TokenMetadata getTokenMetadata()
{
return tokenMetadata_;
}
public IEndPointSnitch getEndPointSnitch()
{
return endPointSnitch_;
}
/**
* This method performs the requisite operations to make
* sure that the N replicas are in sync. We do this in the
* background when we do not care much about consistency.
*/
public void doConsistencyCheck(Row row, List<InetAddress> endpoints, ReadCommand command)
{
consistencyManager_.submit(new ConsistencyManager(command.table, row, endpoints, command));
}
public Map<Range, List<String>> getRangeToEndPointMap()
{
/* All the ranges for the tokens */
List<Range> ranges = getAllRanges(tokenMetadata_.sortedTokens());
Map<Range, List<String>> map = new HashMap<Range, List<String>>();
for (Map.Entry<Range,List<InetAddress>> entry : constructRangeToEndPointMap(ranges).entrySet())
{
map.put(entry.getKey(), stringify(entry.getValue()));
}
return map;
}
/**
* 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.
*/
public Map<Range, List<InetAddress>> constructRangeToEndPointMap(List<Range> ranges)
{
Map<Range, List<InetAddress>> rangeToEndPointMap = new HashMap<Range, List<InetAddress>>();
for (Range range : ranges)
{
rangeToEndPointMap.put(range, replicationStrategy_.getNaturalEndpoints(range.right()));
}
return rangeToEndPointMap;
}
/*
* onChange only ever sees one ApplicationState piece change at a time, so we perform a kind of state machine here.
* We are concerned with two events: knowing the token associated with an enpoint, 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.
*/
public void onChange(InetAddress endpoint, String apStateName, ApplicationState apState)
{
if (!MOVE_STATE.equals(apStateName))
return;
String apStateValue = apState.getValue();
int index = apStateValue.indexOf(Delimiter);
assert (index != -1);
String moveName = apStateValue.substring(0, index);
String moveValue = apStateValue.substring(index+1);
if (moveName.equals(STATE_BOOTSTRAPPING))
handleStateBootstrap(endpoint, moveValue);
else if (moveName.equals(STATE_NORMAL))
handleStateNormal(endpoint, moveValue);
else if (moveName.equals(STATE_LEAVING))
handleStateLeaving(endpoint, moveValue);
else if (moveName.equals(STATE_LEFT))
handleStateLeft(endpoint, moveValue);
}
/**
* Handle node bootstrap
*
* @param endPoint bootstrapping node
* @param moveValue bootstrap token as string
*/
private void handleStateBootstrap(InetAddress endPoint, String moveValue)
{
Token token = getPartitioner().getTokenFactory().fromString(moveValue);
if (logger_.isDebugEnabled())
logger_.debug("Node " + endPoint + " state bootstrapping, token " + token);
// 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 " + endPoint + " state jump to bootstrap");
tokenMetadata_.removeEndpoint(endPoint);
}
tokenMetadata_.addBootstrapToken(token, endPoint);
calculatePendingRanges();
}
/**
* Handle node move to normal state. That is, node is entering token ring and participating
* in reads.
*
* @param endPoint node
* @param moveValue token as string
*/
private void handleStateNormal(InetAddress endPoint, String moveValue)
{
Token token = getPartitioner().getTokenFactory().fromString(moveValue);
if (logger_.isDebugEnabled())
logger_.debug("Node " + endPoint + " state normal, token " + token);
if (tokenMetadata_.isMember(endPoint))
logger_.info("Node " + endPoint + " state jump to normal");
tokenMetadata_.updateNormalToken(token, endPoint);
calculatePendingRanges();
if (!isClientMode)
SystemTable.updateToken(endPoint, token);
}
/**
* Handle node preparing to leave the ring
*
* @param endPoint node
* @param moveValue token as string
*/
private void handleStateLeaving(InetAddress endPoint, String moveValue)
{
Token token = getPartitioner().getTokenFactory().fromString(moveValue);
if (logger_.isDebugEnabled())
logger_.debug("Node " + endPoint + " state leaving, token " + token);
// 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 " + endPoint + " state jump to leaving");
tokenMetadata_.updateNormalToken(token, endPoint);
}
else if (!tokenMetadata_.getToken(endPoint).equals(token))
{
logger_.warn("Node " + endPoint + " 'leaving' token mismatch. Long network partition?");
tokenMetadata_.updateNormalToken(token, endPoint);
}
// at this point the endpoint is certainly a member with this token, so let's proceed
// normally
tokenMetadata_.addLeavingEndPoint(endPoint);
calculatePendingRanges();
}
/**
* Handle node leaving the ring. This can be either because the node was removed manually by
* removetoken command or because of decommission or loadbalance
*
* @param endPoint If reason for leaving is decommission or loadbalance (LEFT_NORMALLY),
* endPoint is the leaving node. If reason manual removetoken (REMOVE_TOKEN), endPoint
* parameter is ignored and the operation is based on the token inside moveValue.
* @param moveValue (REMOVE_TOKEN|LEFT_NORMALLY)<Delimiter><token>
*/
private void handleStateLeft(InetAddress endPoint, String moveValue)
{
int index = moveValue.indexOf(Delimiter);
assert (index != -1);
String typeOfState = moveValue.substring(0, index);
Token token = getPartitioner().getTokenFactory().fromString(moveValue.substring(index + 1));
// endPoint itself is leaving
if (typeOfState.equals(LEFT_NORMALLY))
{
if (logger_.isDebugEnabled())
logger_.debug("Node " + endPoint + " state left, token " + token);
// If the node is member, remove all references to it. If not, call
// removeBootstrapToken just in case it is there (very unlikely chain of events)
if (tokenMetadata_.isMember(endPoint))
{
if (!tokenMetadata_.getToken(endPoint).equals(token))
logger_.warn("Node " + endPoint + " 'left' token mismatch. Long network partition?");
tokenMetadata_.removeEndpoint(endPoint);
}
}
else
{
// if we're here, endPoint is not leaving but broadcasting remove token command
assert (typeOfState.equals(REMOVE_TOKEN));
InetAddress endPointThatLeft = tokenMetadata_.getEndPoint(token);
if (logger_.isDebugEnabled())
logger_.debug("Token " + token + " removed manually (endpoint was " + ((endPointThatLeft == null) ? "unknown" : endPointThatLeft) + ")");
if (endPointThatLeft != null)
{
restoreReplicaCount(endPointThatLeft);
tokenMetadata_.removeEndpoint(endPointThatLeft);
}
}
// remove token from bootstrap tokens just in case it is still there
tokenMetadata_.removeBootstrapToken(token);
calculatePendingRanges();
}
/**
* Calculate pending ranges according to bootsrapping and leaving nodes. Reasoning is:
*
* (1) When in doubt, it is better to write too much to a node than too little. That is, if
* there are multiple nodes moving, calculate the biggest ranges a node could have. Cleaning
* up unneeded data afterwards is better than missing writes during movement.
* (2) When a node leaves, ranges for other nodes can only grow (a node might get additional
* ranges, but it will not lose any of its current ranges as a result of a leave). Therefore
* we will first remove _all_ leaving tokens for the sake of calculation and then check what
* ranges would go where if all nodes are to leave. This way we get the biggest possible
* ranges with regard current leave operations, covering all subsets of possible final range
* values.
* (3) When a node bootstraps, ranges of other nodes can only get smaller. Without doing
* complex calculations to see if multiple bootstraps overlap, we simply base calculations
* on the same token ring used before (reflecting situation after all leave operations have
* completed). Bootstrapping nodes will be added and removed one by one to that metadata and
* checked what their ranges would be. This will give us the biggest possible ranges the
* node could have. It might be that other bootstraps make our actual final ranges smaller,
* but it does not matter as we can clean up the data afterwards.
*
* NOTE: This is heavy and ineffective operation. This will be done only once when a node
* changes state in the cluster, so it should be manageable.
*/
private void calculatePendingRanges()
{
calculatePendingRanges(tokenMetadata_, replicationStrategy_);
}
// public & static for testing purposes
public static void calculatePendingRanges(TokenMetadata tm, AbstractReplicationStrategy strategy)
{
Multimap<Range, InetAddress> pendingRanges = HashMultimap.create();
Map<Token, InetAddress> bootstrapTokens = tm.getBootstrapTokens();
Set<InetAddress> leavingEndPoints = tm.getLeavingEndPoints();
if (bootstrapTokens.isEmpty() && leavingEndPoints.isEmpty())
{
if (logger_.isDebugEnabled())
logger_.debug("No bootstrapping or leaving nodes -> empty pending ranges");
tm.setPendingRanges(pendingRanges);
return;
}
Multimap<InetAddress, Range> addressRanges = strategy.getAddressRanges();
// Copy of metadata reflecting the situation after all leave operations are finished.
TokenMetadata allLeftMetadata = tm.cloneAfterAllLeft();
// get all ranges that will be affected by leaving nodes
Set<Range> affectedRanges = new HashSet<Range>();
for (InetAddress endPoint : leavingEndPoints)
affectedRanges.addAll(addressRanges.get(endPoint));
// for each of those ranges, find what new nodes will be responsible for the range when
// all leaving nodes are gone.
for (Range range : affectedRanges)
{
List<InetAddress> currentEndPoints = strategy.getNaturalEndpoints(range.right(), tm);
List<InetAddress> newEndPoints = strategy.getNaturalEndpoints(range.right(), allLeftMetadata);
newEndPoints.removeAll(currentEndPoints);
pendingRanges.putAll(range, newEndPoints);
}
// At this stage pendingRanges has been updated according to leave operations. We can
// now finish the calculation by checking bootstrapping nodes.
// For each of the bootstrapping nodes, simply add and remove them one by one to
// allLeftMetadata and check in between what their ranges would be.
for (Map.Entry<Token, InetAddress> entry : bootstrapTokens.entrySet())
{
InetAddress endPoint = entry.getValue();
allLeftMetadata.updateNormalToken(entry.getKey(), endPoint);
for (Range range : strategy.getAddressRanges(allLeftMetadata).get(endPoint))
pendingRanges.put(range, endPoint);
allLeftMetadata.removeEndpoint(endPoint);
}
tm.setPendingRanges(pendingRanges);
if (logger_.isDebugEnabled())
logger_.debug("Pending ranges:\n" + (pendingRanges.isEmpty() ? "<empty>" : tm.printPendingRanges()));
}
/**
* Called when endPoint is removed from the ring without proper
* STATE_LEAVING -> STATE_LEFT sequence. 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 node that has left
*/
private void restoreReplicaCount(InetAddress endPoint)
{
InetAddress myAddress = FBUtilities.getLocalAddress();
// get all ranges that change ownership (that is, a node needs
// to take responsibility for new range)
Multimap<Range, InetAddress> changedRanges = getChangedRangesForLeaving(endPoint);
// check if any of these ranges are coming our way
Set<Range> myNewRanges = new HashSet<Range>();
for (Map.Entry<Range, InetAddress> entry : changedRanges.entries())
{
if (entry.getValue().equals(myAddress))
myNewRanges.add(entry.getKey());
}
if (!myNewRanges.isEmpty())
{
if (logger_.isDebugEnabled())
logger_.debug(endPoint + " was removed, my added ranges: " + StringUtils.join(myNewRanges, ", "));
Multimap<Range, InetAddress> rangeAddresses = replicationStrategy_.getRangeAddresses(tokenMetadata_);
Multimap<InetAddress, Range> sourceRanges = HashMultimap.create();
IFailureDetector failureDetector = FailureDetector.instance();
// find alive sources for our new ranges
for (Range myNewRange : myNewRanges)
{
List<InetAddress> sources = DatabaseDescriptor.getEndPointSnitch().getSortedListByProximity(myAddress, rangeAddresses.get(myNewRange));
assert (!sources.contains(myAddress));
for (InetAddress source : sources)
{
if (source.equals(endPoint))
continue;
if (failureDetector.isAlive(source))
{
sourceRanges.put(source, myNewRange);
break;
}
}
}
// Finally we have a list of addresses and ranges to
// stream. Proceed to stream
for (Map.Entry<InetAddress, Collection<Range>> entry : sourceRanges.asMap().entrySet())
{
if (logger_.isDebugEnabled())
logger_.debug("Requesting from " + entry.getKey() + " ranges " + StringUtils.join(entry.getValue(), ", "));
Streaming.requestRanges(entry.getKey(), entry.getValue());
}
}
}
private Multimap<Range, InetAddress> getChangedRangesForLeaving(InetAddress endpoint)
{
// First get all ranges the leaving endpoint is responsible for
Collection<Range> ranges = getRangesForEndPoint(endpoint);
if (logger_.isDebugEnabled())
logger_.debug("Node " + endpoint + " ranges [" + StringUtils.join(ranges, ", ") + "]");
Map<Range, ArrayList<InetAddress>> currentReplicaEndpoints = new HashMap<Range, ArrayList<InetAddress>>();
// Find (for each range) all nodes that store replicas for these ranges as well
for (Range range : ranges)
currentReplicaEndpoints.put(range, replicationStrategy_.getNaturalEndpoints(range.right(), tokenMetadata_));
TokenMetadata temp = tokenMetadata_.cloneAfterAllLeft();
// endpoint might or might not be 'leaving'. If it was not leaving (that is, removetoken
// command was used), it is still present in temp and must be removed.
if (temp.isMember(endpoint))
temp.removeEndpoint(endpoint);
Multimap<Range, 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 newReplicaEndpoins list, but
// not in the currentReplicaEndpoins list, will be needing the
// range.
for (Range range : ranges)
{
ArrayList<InetAddress> newReplicaEndpoints = replicationStrategy_.getNaturalEndpoints(range.right(), temp);
newReplicaEndpoints.removeAll(currentReplicaEndpoints.get(range));
if (logger_.isDebugEnabled())
if (newReplicaEndpoints.isEmpty())
logger_.debug("Range " + range + " already in all replicas");
else
logger_.debug("Range " + range + " will be responsibility of " + StringUtils.join(newReplicaEndpoints, ", "));
changedRanges.putAll(range, newReplicaEndpoints);
}
return changedRanges;
}
public void onJoin(InetAddress endpoint, EndPointState epState)
{
for (Map.Entry<String,ApplicationState> entry : epState.getSortedApplicationStates())
{
onChange(endpoint, entry.getKey(), entry.getValue());
}
}
public void onAlive(InetAddress endpoint, EndPointState state)
{
if (!isClientMode)
deliverHints(endpoint);
}
public void onDead(InetAddress endpoint, EndPointState state) {}
/** raw load value */
public double getLoad()
{
double bytes = 0;
for (String tableName : Table.getAllTableNames())
{
Table table;
try
{
table = Table.open(tableName);
}
catch (IOException e)
{
throw new IOError(e);
}
for (String cfName : table.getColumnFamilies())
{
ColumnFamilyStore cfs = table.getColumnFamilyStore(cfName);
for (SSTableReader sstable : cfs.getSSTables())
{
bytes += sstable.bytesOnDisk();
}
}
}
return bytes;
}
public String getLoadString()
{
return FileUtils.stringifyFileSize(getLoad());
}
public Map<String, String> getLoadMap()
{
Map<String, String> map = new HashMap<String, String>();
for (Map.Entry<InetAddress,Double> entry : StorageLoadBalancer.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.getLocalAddress().getHostAddress(), getLoadString());
return map;
}
/**
* Deliver hints to the specified node when it has crashed
* and come back up/ marked as alive after a network partition
*/
public final void deliverHints(InetAddress endpoint)
{
HintedHandOffManager.instance().deliverHints(endpoint);
}
public Token getLocalToken()
{
return storageMetadata_.getToken();
}
/* This methods belong to the MBean interface */
public String getToken()
{
return getLocalToken().toString();
}
public Set<String> getLiveNodes()
{
return stringify(Gossiper.instance().getLiveMembers());
}
public Set<String> getUnreachableNodes()
{
return stringify(Gossiper.instance().getUnreachableMembers());
}
private Set<String> stringify(Set<InetAddress> endPoints)
{
Set<String> stringEndPoints = new HashSet<String>();
for (InetAddress ep : endPoints)
{
stringEndPoints.add(ep.getHostAddress());
}
return stringEndPoints;
}
private List<String> stringify(List<InetAddress> endPoints)
{
List<String> stringEndPoints = new ArrayList<String>();
for (InetAddress ep : endPoints)
{
stringEndPoints.add(ep.getHostAddress());
}
return stringEndPoints;
}
public int getCurrentGenerationNumber()
{
return Gossiper.instance().getCurrentGenerationNumber(FBUtilities.getLocalAddress());
}
public void forceTableCleanup() throws IOException
{
List<String> tables = DatabaseDescriptor.getTables();
for (String tName : tables)
{
if (tName.equals(Table.SYSTEM_TABLE))
continue;
Table table = Table.open(tName);
table.forceCleanup();
}
}
public void forceTableCompaction() throws IOException
{
List<String> tables = DatabaseDescriptor.getTables();
for ( String tName : tables )
{
Table table = Table.open(tName);
table.forceCompaction();
}
}
/**
* Takes the snapshot for a given table.
*
* @param tableName the name of the table.
* @param tag the tag given to the snapshot (null is permissible)
*/
public void takeSnapshot(String tableName, String tag) throws IOException
{
Table tableInstance = getValidTable(tableName);
tableInstance.snapshot(tag);
}
private Table getValidTable(String tableName) throws IOException
{
if (DatabaseDescriptor.getTable(tableName) == null)
{
throw new IOException("Table " + tableName + "does not exist");
}
return Table.open(tableName);
}
/**
* Takes a snapshot for every table.
*
* @param tag the tag given to the snapshot (null is permissible)
*/
public void takeAllSnapshot(String tag) throws IOException
{
for (String tableName: DatabaseDescriptor.getTables())
{
Table tableInstance = Table.open(tableName);
tableInstance.snapshot(tag);
}
}
/**
* Remove all the existing snapshots.
*/
public void clearSnapshot() throws IOException
{
for (String tableName: DatabaseDescriptor.getTables())
{
Table tableInstance = Table.open(tableName);
tableInstance.clearSnapshot();
}
if (logger_.isDebugEnabled())
logger_.debug("Cleared out all snapshot directories");
}
public Iterable<ColumnFamilyStore> getValidColumnFamilies(String tableName, String... columnFamilies) throws IOException
{
Table table = getValidTable(tableName);
Set<ColumnFamilyStore> valid = new HashSet<ColumnFamilyStore>();
for (String cfName : columnFamilies.length == 0 ? table.getColumnFamilies() : Arrays.asList(columnFamilies))
{
ColumnFamilyStore cfStore = table.getColumnFamilyStore(cfName);
if (cfStore == null)
{
// this means there was a cf passed in that is not recognized in the keyspace. report it and continue.
logger_.warn(String.format("Invalid column family specified: %s. Proceeding with others.", cfName));
continue;
}
valid.add(cfStore);
}
return valid;
}
/**
* Flush all memtables for a table and column families.
* @param tableName
* @param columnFamilies
* @throws IOException
*/
public void forceTableFlush(final String tableName, final String... columnFamilies) throws IOException
{
for (ColumnFamilyStore cfStore : getValidColumnFamilies(tableName, columnFamilies))
{
logger_.debug("Forcing binary flush on keyspace " + tableName + ", CF " + cfStore.getColumnFamilyName());
cfStore.forceFlushBinary();
logger_.debug("Forcing flush on keyspace " + tableName + ", CF " + cfStore.getColumnFamilyName());
cfStore.forceFlush();
}
}
/**
* Trigger proactive repair for a table and column families.
* @param tableName
* @param columnFamilies
* @throws IOException
*/
public void forceTableRepair(final String tableName, final String... columnFamilies) throws IOException
{
// request that all relevant endpoints generate trees
final MessagingService ms = MessagingService.instance();
final List<InetAddress> endpoints = getNaturalEndpoints(getLocalToken());
for (ColumnFamilyStore cfStore : getValidColumnFamilies(tableName, columnFamilies))
{
Message request = TreeRequestVerbHandler.makeVerb(tableName, cfStore.getColumnFamilyName());
for (InetAddress endpoint : endpoints)
ms.sendOneWay(request, endpoint);
}
}
/* End of MBean interface methods */
/**
* This method returns the predecessor of the endpoint ep on the identifier
* space.
*/
InetAddress getPredecessor(InetAddress ep)
{
Token token = tokenMetadata_.getToken(ep);
return tokenMetadata_.getEndPoint(tokenMetadata_.getPredecessor(token));
}
/*
* This method returns the successor of the endpoint ep on the identifier
* space.
*/
public InetAddress getSuccessor(InetAddress ep)
{
Token token = tokenMetadata_.getToken(ep);
return tokenMetadata_.getEndPoint(tokenMetadata_.getSuccessor(token));
}
/**
* Get the primary range for the specified endpoint.
* @param ep endpoint we are interested in.
* @return range for the specified endpoint.
*/
public Range getPrimaryRangeForEndPoint(InetAddress ep)
{
return tokenMetadata_.getPrimaryRangeFor(tokenMetadata_.getToken(ep));
}
/**
* Get all ranges an endpoint is responsible for.
* @param ep endpoint we are interested in.
* @return ranges for the specified endpoint.
*/
Collection<Range> getRangesForEndPoint(InetAddress ep)
{
return replicationStrategy_.getAddressRanges().get(ep);
}
/**
* Get all ranges that span the ring given a set
* of tokens. All ranges are in sorted order of
* ranges.
* @return ranges in sorted order
*/
public List<Range> getAllRanges(List<Token> sortedTokens)
{
if (logger_.isDebugEnabled())
logger_.debug("computing ranges for " + StringUtils.join(sortedTokens, ", "));
List<Range> ranges = new ArrayList<Range>();
int size = sortedTokens.size();
for (int i = 1; i < size; ++i)
{
Range range = new Range(sortedTokens.get(i - 1), sortedTokens.get(i));
ranges.add(range);
}
Range range = new Range(sortedTokens.get(size - 1), sortedTokens.get(0));
ranges.add(range);
return ranges;
}
/**
* This method returns the endpoint that is responsible for storing the
* specified key.
*
* @param key - key for which we need to find the endpoint
* @return value - the endpoint responsible for this key
*/
public InetAddress getPrimary(String key)
{
return getPrimary(partitioner_.getToken(key));
}
public InetAddress getPrimary(Token token)
{
InetAddress endpoint = FBUtilities.getLocalAddress();
List tokens = new ArrayList<Token>(tokenMetadata_.sortedTokens());
if (tokens.size() > 0)
{
int index = Collections.binarySearch(tokens, token);
if (index >= 0)
{
/*
* retrieve the endpoint based on the token at this index in the
* tokens list
*/
endpoint = tokenMetadata_.getEndPoint((Token) tokens.get(index));
}
else
{
index = (index + 1) * (-1);
if (index < tokens.size())
endpoint = tokenMetadata_.getEndPoint((Token) tokens.get(index));
else
endpoint = tokenMetadata_.getEndPoint((Token) tokens.get(0));
}
}
return endpoint;
}
/**
* This method determines whether the local endpoint is the
* primary for the given key.
* @param key
* @return true if the local endpoint is the primary replica.
*/
public boolean isPrimary(String key)
{
InetAddress endpoint = getPrimary(key);
return FBUtilities.getLocalAddress().equals(endpoint);
}
/**
* This method returns the N endpoints that are responsible for storing the
* specified key i.e for replication.
*
* @param key - key for which we need to find the endpoint return value -
* the endpoint responsible for this key
*/
public List<InetAddress> getNaturalEndpoints(String key)
{
return getNaturalEndpoints(partitioner_.getToken(key));
}
/**
* This method returns the N endpoints that are responsible for storing the
* specified key i.e for replication.
*
* @param token - token for which we need to find the endpoint return value -
* the endpoint responsible for this token
*/
public List<InetAddress> getNaturalEndpoints(Token token)
{
return replicationStrategy_.getNaturalEndpoints(token);
}
/**
* This method attempts to return N endpoints that are responsible for storing the
* specified key i.e for replication.
*
* @param key - key for which we need to find the endpoint return value -
* the endpoint responsible for this key
*/
public List<InetAddress> getLiveNaturalEndpoints(String key)
{
return getLiveNaturalEndpoints(partitioner_.getToken(key));
}
public List<InetAddress> getLiveNaturalEndpoints(Token token)
{
List<InetAddress> liveEps = new ArrayList<InetAddress>();
List<InetAddress> endpoints = replicationStrategy_.getNaturalEndpoints(token);
for (InetAddress endpoint : endpoints)
{
if (FailureDetector.instance().isAlive(endpoint))
liveEps.add(endpoint);
}
return liveEps;
}
/**
* This method returns the N endpoints that are responsible for storing the
* specified key i.e for replication.
*
* @param key - key for which we need to find the endpoint return value -
* the endpoint responsible for this key
*/
public Map<InetAddress, InetAddress> getHintedEndpointMap(String key, List<InetAddress> naturalEndpoints)
{
return replicationStrategy_.getHintedEndpoints(partitioner_.getToken(key), naturalEndpoints);
}
/**
* This function finds the closest live endpoint that contains a given key.
*/
public InetAddress findSuitableEndPoint(String key) throws IOException, UnavailableException
{
List<InetAddress> endpoints = getNaturalEndpoints(key);
endPointSnitch_.sortByProximity(FBUtilities.getLocalAddress(), endpoints);
for (InetAddress endpoint : endpoints)
{
if (FailureDetector.instance().isAlive(endpoint))
return endpoint;
}
throw new UnavailableException(); // no nodes that could contain key are alive
}
Map<String, String> getStringEndpointMap()
{
HashMap<String, String> map = new HashMap<String, String>();
for (Token t : tokenMetadata_.sortedTokens())
{
map.put(t.toString(), tokenMetadata_.getEndPoint(t).getHostAddress());
}
return map;
}
public void setLog4jLevel(String classQualifier, String rawLevel)
{
Level level = Level.toLevel(rawLevel);
Logger.getLogger(classQualifier).setLevel(level);
logger_.info("set log level to " + level + " for classes under '" + classQualifier + "' (if the level doesn't look like '" + rawLevel + "' then log4j couldn't parse '" + rawLevel + "')");
}
/**
* @param splits: number of ranges to break into. Minimum 2.
* @return list of Tokens (_not_ keys!) breaking up the data this node is responsible for into `splits` pieces.
* There will be 1 more token than splits requested. So for splits of 2, tokens T1 T2 T3 will be returned,
* where (T1, T2] is the first range and (T2, T3] is the second. The first token will always be the left
* Token of this node's primary range, and the last will always be the Right token of that range.
*/
public List<String> getSplits(int splits)
{
assert splits > 1;
// we use the actual Range token for the first and last brackets of the splits to ensure correctness
// (we're only operating on 1/128 of the keys remember)
Range range = getLocalPrimaryRange();
List<String> tokens = new ArrayList<String>();
tokens.add(range.left().toString());
List<DecoratedKey> decoratedKeys = SSTableReader.getIndexedDecoratedKeys();
if (decoratedKeys.size() < splits)
{
// not enough keys to generate good splits -- generate random ones instead
// (since this only happens when we don't have many keys, it doesn't really matter that the splits are poor)
for (int i = 1; i < splits; i++)
{
tokens.add(partitioner_.getRandomToken().toString());
}
}
else
{
for (int i = 1; i < splits; i++)
{
int index = i * (decoratedKeys.size() / splits);
tokens.add(decoratedKeys.get(index).token.toString());
}
}
tokens.add(range.right().toString());
return tokens;
}
/**
* Broadcast leaving status and update local tokenMetadata_ accordingly
*/
private void startLeaving()
{
Gossiper.instance().addApplicationState(MOVE_STATE, new ApplicationState(STATE_LEAVING + Delimiter + getLocalToken().toString()));
tokenMetadata_.addLeavingEndPoint(FBUtilities.getLocalAddress());
calculatePendingRanges();
}
public void decommission() throws InterruptedException
{
if (!tokenMetadata_.isMember(FBUtilities.getLocalAddress()))
throw new UnsupportedOperationException("local node is not a member of the token ring yet");
if (tokenMetadata_.cloneAfterAllLeft().sortedTokens().size() < 2)
throw new UnsupportedOperationException("no other normal nodes in the ring; decommission would be pointless");
if (tokenMetadata_.getPendingRanges(FBUtilities.getLocalAddress()).size() > 0)
throw new UnsupportedOperationException("data is currently moving to this node; unable to leave the ring");
logger_.info("DECOMMISSIONING");
startLeaving();
logger_.info("decommission sleeping " + Streaming.RING_DELAY);
Thread.sleep(Streaming.RING_DELAY);
Runnable finishLeaving = new Runnable()
{
public void run()
{
Gossiper.instance().stop();
MessagingService.shutdown();
logger_.info("DECOMMISSION FINISHED.");
// let op be responsible for killing the process
}
};
unbootstrap(finishLeaving);
}
private void leaveRing()
{
SystemTable.setBootstrapped(false);
tokenMetadata_.removeEndpoint(FBUtilities.getLocalAddress());
calculatePendingRanges();
if (logger_.isDebugEnabled())
logger_.debug("");
Gossiper.instance().addApplicationState(MOVE_STATE, new ApplicationState(STATE_LEFT + Delimiter + LEFT_NORMALLY + Delimiter + getLocalToken().toString()));
try
{
Thread.sleep(2 * Gossiper.intervalInMillis_);
}
catch (InterruptedException e)
{
throw new AssertionError(e);
}
}
private void unbootstrap(final Runnable onFinish)
{
Multimap<Range, InetAddress> rangesMM = getChangedRangesForLeaving(FBUtilities.getLocalAddress());
if (logger_.isDebugEnabled())
logger_.debug("Ranges needing transfer are [" + StringUtils.join(rangesMM.keySet(), ",") + "]");
if (rangesMM.isEmpty())
{
// nothing needs transfer, so leave immediately. this can happen when replication factor == number of nodes.
leaveRing();
onFinish.run();
return;
}
final Set<Map.Entry<Range, InetAddress>> pending = new HashSet<Map.Entry<Range, InetAddress>>(rangesMM.entries());
for (final Map.Entry<Range, InetAddress> entry : rangesMM.entries())
{
final Range range = entry.getKey();
final InetAddress newEndpoint = entry.getValue();
final Runnable callback = new Runnable()
{
public synchronized void run()
{
pending.remove(entry);
if (pending.isEmpty())
{
leaveRing();
onFinish.run();
}
}
};
StageManager.getStage(StageManager.streamStage_).execute(new Runnable()
{
public void run()
{
// TODO each call to transferRanges re-flushes, this is potentially a lot of waste
Streaming.transferRanges(newEndpoint, Arrays.asList(range), callback);
}
});
}
}
public void move(String newToken) throws InterruptedException
{
move(partitioner_.getTokenFactory().fromString(newToken));
}
public void loadBalance() throws IOException, InterruptedException
{
move((Token)null);
}
/**
* move the node to new token or find a new token to boot to according to load
*
* @param token new token to boot to, or if null, find balanced token to boot to
*/
private void move(final Token token) throws InterruptedException
{
if (tokenMetadata_.getPendingRanges(FBUtilities.getLocalAddress()).size() > 0)
throw new UnsupportedOperationException("data is currently moving to this node; unable to leave the ring");
logger_.info("starting move. leaving token " + getLocalToken());
startLeaving();
logger_.info("move sleeping " + Streaming.RING_DELAY);
Thread.sleep(Streaming.RING_DELAY);
Runnable finishMoving = new Runnable()
{
public void run()
{
try
{
Token bootstrapToken = token;
if (bootstrapToken == null)
bootstrapToken = BootStrapper.getBalancedToken(tokenMetadata_, StorageLoadBalancer.instance().getLoadInfo());
logger_.info("re-bootstrapping to new token " + bootstrapToken);
startBootstrap(bootstrapToken);
}
catch (IOException e)
{
throw new IOError(e);
}
}
};
unbootstrap(finishMoving);
}
public void removeToken(String tokenString)
{
Token token = partitioner_.getTokenFactory().fromString(tokenString);
// Here we could refuse the operation from continuing if we
// cannot find the endpoint for this token from metadata, but
// that would prevent this command from being issued by a node
// that has never seen the failed node.
InetAddress endPoint = tokenMetadata_.getEndPoint(token);
if (endPoint != null)
{
// Let's make sure however that we're not removing a live
// token (member)
if (Gossiper.instance().getLiveMembers().contains(endPoint))
throw new UnsupportedOperationException("Node " + endPoint + " is alive and owns this token. Use decommission command to remove it from the ring");
restoreReplicaCount(endPoint);
tokenMetadata_.removeEndpoint(endPoint);
calculatePendingRanges();
}
// This is not the cleanest way as we're adding STATE_LEFT for
// a foreign token to our own EP state. Another way would be
// to add new AP state for this command, but that would again
// increase the amount of data to be gossiped in the cluster -
// not good. REMOVE_TOKEN|LEFT_NORMALLY is used to distinguish
// between removetoken command and normal state left, so it is
// not so bad.
Gossiper.instance().addApplicationState(MOVE_STATE, new ApplicationState(STATE_LEFT + Delimiter + REMOVE_TOKEN + Delimiter + token.toString()));
}
public WriteResponseHandler getWriteResponseHandler(int blockFor, int consistency_level)
{
return replicationStrategy_.getWriteResponseHandler(blockFor, consistency_level);
}
public AbstractReplicationStrategy getReplicationStrategy()
{
return replicationStrategy_;
}
public boolean isClientMode()
{
return isClientMode;
}
// Never ever do this at home. Used by tests.
AbstractReplicationStrategy setReplicationStrategyUnsafe(AbstractReplicationStrategy newStrategy)
{
AbstractReplicationStrategy oldStrategy = replicationStrategy_;
replicationStrategy_ = newStrategy;
return oldStrategy;
}
// Never ever do this at home. Used by tests.
IPartitioner setPartitionerUnsafe(IPartitioner newPartitioner)
{
IPartitioner oldPartitioner = partitioner_;
partitioner_ = newPartitioner;
return oldPartitioner;
}
}