org.apache.cassandra.locator.TokenMetadata 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
Palantir open source project
/*
* 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.locator;
import java.net.InetAddress;
import java.nio.ByteBuffer;
import java.util.*;
import java.util.concurrent.ConcurrentHashMap;
import java.util.concurrent.ConcurrentMap;
import java.util.concurrent.atomic.AtomicReference;
import java.util.concurrent.locks.ReadWriteLock;
import java.util.concurrent.locks.ReentrantReadWriteLock;
import com.google.common.collect.*;
import org.apache.commons.lang3.StringUtils;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
import org.apache.cassandra.config.DatabaseDescriptor;
import org.apache.cassandra.dht.Range;
import org.apache.cassandra.dht.Token;
import org.apache.cassandra.gms.FailureDetector;
import org.apache.cassandra.service.StorageService;
import org.apache.cassandra.utils.BiMultiValMap;
import org.apache.cassandra.utils.Pair;
import org.apache.cassandra.utils.SortedBiMultiValMap;
public class TokenMetadata
{
private static final Logger logger = LoggerFactory.getLogger(TokenMetadata.class);
/**
* Maintains token to endpoint map of every node in the cluster.
* Each Token is associated with exactly one Address, but each Address may have
* multiple tokens. Hence, the BiMultiValMap collection.
*/
private final BiMultiValMap tokenToEndpointMap;
/** Maintains endpoint to host ID map of every node in the cluster */
private final BiMap endpointToHostIdMap;
// Prior to CASSANDRA-603, we just had Map pendingRanges,
// which was added to when a node began bootstrap and removed from when it finished.
//
// This is inadequate when multiple changes are allowed simultaneously. For example,
// suppose that there is a ring of nodes A, C and E, with replication factor 3.
// Node D bootstraps between C and E, so its pending ranges will be E-A, A-C and C-D.
// Now suppose node B bootstraps between A and C at the same time. Its pending ranges
// would be C-E, E-A and A-B. Now both nodes need to be assigned pending range E-A,
// which we would be unable to represent with the old Map. The same thing happens
// even more obviously for any nodes that boot simultaneously between same two nodes.
//
// So, we made two changes:
//
// First, we changed pendingRanges to a Multimap> , because replication strategy
// and options are per-KeySpace).
//
// Second, we added the bootstrapTokens and leavingEndpoints collections, so we can
// rebuild pendingRanges from the complete information of what is going on, when
// additional changes are made mid-operation.
//
// Finally, note that recording the tokens of joining nodes in bootstrapTokens also
// means we can detect and reject the addition of multiple nodes at the same token
// before one becomes part of the ring.
private final BiMultiValMap bootstrapTokens = new BiMultiValMap<>();
private final BiMap replacementToOriginal = HashBiMap.create();
// (don't need to record Token here since it's still part of tokenToEndpointMap until it's done leaving)
private final Set leavingEndpoints = new HashSet<>();
// this is a cache of the calculation from {tokenToEndpointMap, bootstrapTokens, leavingEndpoints}
private final ConcurrentMap pendingRanges = new ConcurrentHashMap();
// nodes which are migrating to the new tokens in the ring
private final Set> movingEndpoints = new HashSet<>();
/* Use this lock for manipulating the token map */
private final ReadWriteLock lock = new ReentrantReadWriteLock(true);
private final ReadWriteLock publicLock = new ReentrantReadWriteLock(true);
private volatile ArrayList sortedTokens;
private final Topology topology;
private static final Comparator inetaddressCmp = new Comparator()
{
public int compare(InetAddress o1, InetAddress o2)
{
return ByteBuffer.wrap(o1.getAddress()).compareTo(ByteBuffer.wrap(o2.getAddress()));
}
};
// signals replication strategies that nodes have joined or left the ring and they need to recompute ownership
private volatile long ringVersion = 0;
public TokenMetadata()
{
this(SortedBiMultiValMap.create(null, inetaddressCmp),
HashBiMap.create(),
new Topology());
}
private TokenMetadata(BiMultiValMap tokenToEndpointMap, BiMap endpointsMap, Topology topology)
{
this.tokenToEndpointMap = tokenToEndpointMap;
this.topology = topology;
endpointToHostIdMap = endpointsMap;
sortedTokens = sortTokens();
}
private ArrayList sortTokens()
{
return new ArrayList<>(tokenToEndpointMap.keySet());
}
public void lock() {
publicLock.writeLock().lock();
}
public void unlock() {
publicLock.writeLock().unlock();
}
/** @return the number of nodes bootstrapping into source's primary range */
public int pendingRangeChanges(InetAddress source)
{
int n = 0;
Collection> sourceRanges = getPrimaryRangesFor(getTokens(source));
publicLock.readLock().lock();
lock.readLock().lock();
try
{
for (Token token : bootstrapTokens.keySet())
for (Range range : sourceRanges)
if (range.contains(token))
n++;
}
finally
{
lock.readLock().unlock();
publicLock.readLock().unlock();
}
return n;
}
/**
* Update token map with a single token/endpoint pair in normal state.
*/
public void updateNormalToken(Token token, InetAddress endpoint)
{
updateNormalTokens(Collections.singleton(token), endpoint);
}
public void updateNormalTokens(Collection tokens, InetAddress endpoint)
{
Multimap endpointTokens = HashMultimap.create();
for (Token token : tokens)
endpointTokens.put(endpoint, token);
updateNormalTokens(endpointTokens);
}
/**
* Update token map with a set of token/endpoint pairs in normal state.
*
* Prefer this whenever there are multiple pairs to update, as each update (whether a single or multiple)
* is expensive (CASSANDRA-3831).
*/
public void updateNormalTokens(Multimap endpointTokens)
{
if (endpointTokens.isEmpty())
return;
publicLock.readLock().lock();
lock.writeLock().lock();
try
{
boolean shouldSortTokens = false;
for (InetAddress endpoint : endpointTokens.keySet())
{
Collection tokens = endpointTokens.get(endpoint);
assert tokens != null && !tokens.isEmpty();
bootstrapTokens.removeValue(endpoint);
tokenToEndpointMap.removeValue(endpoint);
topology.addEndpoint(endpoint);
leavingEndpoints.remove(endpoint);
replacementToOriginal.remove(endpoint);
removeFromMoving(endpoint); // also removing this endpoint from moving
for (Token token : tokens)
{
InetAddress prev = tokenToEndpointMap.put(token, endpoint);
if (!endpoint.equals(prev))
{
if (prev != null)
logger.warn("Token {} changing ownership from {} to {}", token, prev, endpoint);
shouldSortTokens = true;
}
}
}
if (shouldSortTokens)
sortedTokens = sortTokens();
}
finally
{
lock.writeLock().unlock();
publicLock.readLock().unlock();
}
}
/**
* Store an end-point to host ID mapping. Each ID must be unique, and
* cannot be changed after the fact.
*/
public void updateHostId(UUID hostId, InetAddress endpoint)
{
assert hostId != null;
assert endpoint != null;
publicLock.readLock().lock();
lock.writeLock().lock();
try
{
InetAddress storedEp = endpointToHostIdMap.inverse().get(hostId);
if (storedEp != null)
{
if (!storedEp.equals(endpoint) && (FailureDetector.instance.isAlive(storedEp)))
{
throw new RuntimeException(String.format("Host ID collision between active endpoint %s and %s (id=%s)",
storedEp,
endpoint,
hostId));
}
}
UUID storedId = endpointToHostIdMap.get(endpoint);
if ((storedId != null) && (!storedId.equals(hostId)))
logger.warn("Changing {}'s host ID from {} to {}", endpoint, storedId, hostId);
endpointToHostIdMap.forcePut(endpoint, hostId);
}
finally
{
lock.writeLock().unlock();
publicLock.readLock().unlock();
}
}
/** Return the unique host ID for an end-point. */
public UUID getHostId(InetAddress endpoint)
{
publicLock.readLock().lock();
lock.readLock().lock();
try
{
return endpointToHostIdMap.get(endpoint);
}
finally
{
lock.readLock().unlock();
publicLock.readLock().unlock();
}
}
/** Return the end-point for a unique host ID */
public InetAddress getEndpointForHostId(UUID hostId)
{
publicLock.readLock().lock();
lock.readLock().lock();
try
{
return endpointToHostIdMap.inverse().get(hostId);
}
finally
{
lock.readLock().unlock();
publicLock.readLock().unlock();
}
}
/** @return a copy of the endpoint-to-id map for read-only operations */
public Map getEndpointToHostIdMapForReading()
{
publicLock.readLock().lock();
lock.readLock().lock();
try
{
Map readMap = new HashMap<>();
readMap.putAll(endpointToHostIdMap);
return readMap;
}
finally
{
lock.readLock().unlock();
publicLock.readLock().unlock();
}
}
@Deprecated
public void addBootstrapToken(Token token, InetAddress endpoint)
{
addBootstrapTokens(Collections.singleton(token), endpoint);
}
public void addBootstrapTokens(Collection tokens, InetAddress endpoint)
{
addBootstrapTokens(tokens, endpoint, null);
}
private void addBootstrapTokens(Collection tokens, InetAddress endpoint, InetAddress original)
{
assert tokens != null && !tokens.isEmpty();
assert endpoint != null;
publicLock.readLock().lock();
lock.writeLock().lock();
try
{
InetAddress oldEndpoint;
for (Token token : tokens)
{
oldEndpoint = bootstrapTokens.get(token);
if (oldEndpoint != null && !oldEndpoint.equals(endpoint))
throw new RuntimeException("Bootstrap Token collision between " + oldEndpoint + " and " + endpoint + " (token " + token);
oldEndpoint = tokenToEndpointMap.get(token);
if (oldEndpoint != null && !oldEndpoint.equals(endpoint) && !oldEndpoint.equals(original))
throw new RuntimeException("Bootstrap Token collision between " + oldEndpoint + " and " + endpoint + " (token " + token);
}
bootstrapTokens.removeValue(endpoint);
for (Token token : tokens)
bootstrapTokens.put(token, endpoint);
}
finally
{
lock.writeLock().unlock();
publicLock.readLock().unlock();
}
}
public void addReplaceTokens(Collection replacingTokens, InetAddress newNode, InetAddress oldNode)
{
assert replacingTokens != null && !replacingTokens.isEmpty();
assert newNode != null && oldNode != null;
publicLock.readLock().lock();
lock.writeLock().lock();
try
{
Collection oldNodeTokens = tokenToEndpointMap.inverse().get(oldNode);
if (!replacingTokens.containsAll(oldNodeTokens) || !oldNodeTokens.containsAll(replacingTokens))
{
throw new RuntimeException(String.format("Node %s is trying to replace node %s with tokens %s with a " +
"different set of tokens %s.", newNode, oldNode, oldNodeTokens,
replacingTokens));
}
logger.debug("Replacing {} with {}", newNode, oldNode);
replacementToOriginal.put(newNode, oldNode);
addBootstrapTokens(replacingTokens, newNode, oldNode);
}
finally
{
lock.writeLock().unlock();
publicLock.readLock().unlock();
}
}
public Optional getReplacementNode(InetAddress endpoint)
{
return Optional.ofNullable(replacementToOriginal.inverse().get(endpoint));
}
public Optional getReplacingNode(InetAddress endpoint)
{
return Optional.ofNullable((replacementToOriginal.get(endpoint)));
}
public void removeBootstrapTokens(Collection tokens)
{
assert tokens != null && !tokens.isEmpty();
publicLock.readLock().lock();
lock.writeLock().lock();
try
{
for (Token token : tokens)
bootstrapTokens.remove(token);
}
finally
{
lock.writeLock().unlock();
publicLock.readLock().unlock();
}
}
public void addLeavingEndpoint(InetAddress endpoint)
{
assert endpoint != null;
publicLock.readLock().lock();
lock.writeLock().lock();
try
{
leavingEndpoints.add(endpoint);
}
finally
{
lock.writeLock().unlock();
publicLock.readLock().unlock();
}
}
/**
* Add a new moving endpoint
* @param token token which is node moving to
* @param endpoint address of the moving node
*/
public void addMovingEndpoint(Token token, InetAddress endpoint)
{
assert endpoint != null;
publicLock.readLock().lock();
lock.writeLock().lock();
try
{
movingEndpoints.add(Pair.create(token, endpoint));
}
finally
{
lock.writeLock().unlock();
publicLock.readLock().unlock();
}
}
public void removeEndpoint(InetAddress endpoint)
{
assert endpoint != null;
publicLock.readLock().lock();
lock.writeLock().lock();
try
{
bootstrapTokens.removeValue(endpoint);
tokenToEndpointMap.removeValue(endpoint);
topology.removeEndpoint(endpoint);
leavingEndpoints.remove(endpoint);
if (replacementToOriginal.remove(endpoint) != null)
{
logger.debug("Node {} failed during replace.", endpoint);
}
endpointToHostIdMap.remove(endpoint);
sortedTokens = sortTokens();
invalidateCachedRings();
}
finally
{
lock.writeLock().unlock();
publicLock.readLock().unlock();
}
}
/**
* This is called when the snitch properties for this endpoint are updated, see CASSANDRA-10238.
*/
public void updateTopology(InetAddress endpoint)
{
assert endpoint != null;
publicLock.readLock().lock();
lock.writeLock().lock();
try
{
logger.info("Updating topology for {}", endpoint);
topology.updateEndpoint(endpoint);
invalidateCachedRings();
}
finally
{
lock.writeLock().unlock();
publicLock.readLock().unlock();
}
}
/**
* This is called when the snitch properties for many endpoints are updated, it will update
* the topology mappings of any endpoints whose snitch has changed, see CASSANDRA-10238.
*/
public void updateTopology()
{
publicLock.readLock().lock();
lock.writeLock().lock();
try
{
logger.info("Updating topology for all endpoints that have changed");
topology.updateEndpoints();
invalidateCachedRings();
}
finally
{
lock.writeLock().unlock();
publicLock.readLock().unlock();
}
}
/**
* Remove pair of token/address from moving endpoints
* @param endpoint address of the moving node
*/
public void removeFromMoving(InetAddress endpoint)
{
assert endpoint != null;
publicLock.readLock().lock();
lock.writeLock().lock();
try
{
for (Pair pair : movingEndpoints)
{
if (pair.right.equals(endpoint))
{
movingEndpoints.remove(pair);
break;
}
}
invalidateCachedRings();
}
finally
{
lock.writeLock().unlock();
publicLock.readLock().unlock();
}
}
public Collection getTokens(InetAddress endpoint)
{
assert endpoint != null;
assert isMember(endpoint); // don't want to return nulls
publicLock.readLock().lock();
lock.readLock().lock();
try
{
return new ArrayList<>(tokenToEndpointMap.inverse().get(endpoint));
}
finally
{
lock.readLock().unlock();
publicLock.readLock().unlock();
}
}
@Deprecated
public Token getToken(InetAddress endpoint)
{
return getTokens(endpoint).iterator().next();
}
public boolean isMember(InetAddress endpoint)
{
assert endpoint != null;
publicLock.readLock().lock();
lock.readLock().lock();
try
{
return tokenToEndpointMap.inverse().containsKey(endpoint);
}
finally
{
lock.readLock().unlock();
publicLock.readLock().unlock();
}
}
public boolean isLeaving(InetAddress endpoint)
{
assert endpoint != null;
publicLock.readLock().lock();
lock.readLock().lock();
try
{
return leavingEndpoints.contains(endpoint);
}
finally
{
lock.readLock().unlock();
publicLock.readLock().unlock();
}
}
public boolean isMoving(InetAddress endpoint)
{
assert endpoint != null;
publicLock.readLock().lock();
lock.readLock().lock();
try
{
for (Pair pair : movingEndpoints)
{
if (pair.right.equals(endpoint))
return true;
}
return false;
}
finally
{
lock.readLock().unlock();
publicLock.readLock().unlock();
}
}
private final AtomicReference cachedTokenMap = new AtomicReference<>();
/**
* Create a copy of TokenMetadata with only tokenToEndpointMap. That is, pending ranges,
* bootstrap tokens and leaving endpoints are not included in the copy.
*/
public TokenMetadata cloneOnlyTokenMap()
{
publicLock.readLock().lock();
lock.readLock().lock();
try
{
return new TokenMetadata(SortedBiMultiValMap.create(tokenToEndpointMap, null, inetaddressCmp),
HashBiMap.create(endpointToHostIdMap),
new Topology(topology));
}
finally
{
lock.readLock().unlock();
publicLock.readLock().unlock();
}
}
/**
* Return a cached TokenMetadata with only tokenToEndpointMap, i.e., the same as cloneOnlyTokenMap but
* uses a cached copy that is invalided when the ring changes, so in the common case
* no extra locking is required.
*
* Callers must *NOT* mutate the returned metadata object.
*/
public TokenMetadata cachedOnlyTokenMap()
{
publicLock.readLock().lock();
try {
TokenMetadata tm = cachedTokenMap.get();
if (tm != null)
return tm;
// synchronize to prevent thundering herd (CASSANDRA-6345)
synchronized (this)
{
if ((tm = cachedTokenMap.get()) != null)
return tm;
tm = cloneOnlyTokenMap();
cachedTokenMap.set(tm);
return tm;
}
} finally {
publicLock.readLock().unlock();
}
}
/**
* Create a copy of TokenMetadata with tokenToEndpointMap reflecting situation after all
* current leave operations have finished.
*
* @return new token metadata
*/
public TokenMetadata cloneAfterAllLeft()
{
publicLock.readLock().lock();
lock.readLock().lock();
try
{
return removeEndpoints(cloneOnlyTokenMap(), leavingEndpoints);
}
finally
{
lock.readLock().unlock();
publicLock.readLock().unlock();
}
}
private static TokenMetadata removeEndpoints(TokenMetadata allLeftMetadata, Set leavingEndpoints)
{
for (InetAddress endpoint : leavingEndpoints)
allLeftMetadata.removeEndpoint(endpoint);
return allLeftMetadata;
}
/**
* Create a copy of TokenMetadata with tokenToEndpointMap reflecting situation after all
* current leave, and move operations have finished.
*
* @return new token metadata
*/
public TokenMetadata cloneAfterAllSettled()
{
publicLock.readLock().lock();
lock.readLock().lock();
try
{
TokenMetadata metadata = cloneOnlyTokenMap();
for (InetAddress endpoint : leavingEndpoints)
metadata.removeEndpoint(endpoint);
for (Pair pair : movingEndpoints)
metadata.updateNormalToken(pair.left, pair.right);
return metadata;
}
finally
{
lock.readLock().unlock();
publicLock.readLock().unlock();
}
}
public InetAddress getEndpoint(Token token)
{
publicLock.readLock().lock();
lock.readLock().lock();
try
{
return tokenToEndpointMap.get(token);
}
finally
{
lock.readLock().unlock();
publicLock.readLock().unlock();
}
}
public Collection> getPrimaryRangesFor(Collection tokens)
{
Collection> ranges = new ArrayList<>(tokens.size());
for (Token right : tokens)
ranges.add(new Range<>(getPredecessor(right), right));
return ranges;
}
@Deprecated
public Range getPrimaryRangeFor(Token right)
{
return getPrimaryRangesFor(Arrays.asList(right)).iterator().next();
}
public ArrayList sortedTokens()
{
return sortedTokens;
}
public Multimap, InetAddress> getPendingRangesMM(String keyspaceName)
{
Multimap, InetAddress> map = HashMultimap.create();
PendingRangeMaps pendingRangeMaps = this.pendingRanges.get(keyspaceName);
if (pendingRangeMaps != null)
{
for (Map.Entry, List> entry : pendingRangeMaps)
{
Range range = entry.getKey();
for (InetAddress address : entry.getValue())
{
map.put(range, address);
}
}
}
return map;
}
/** a mutable map may be returned but caller should not modify it */
public PendingRangeMaps getPendingRanges(String keyspaceName)
{
return this.pendingRanges.get(keyspaceName);
}
public List> getPendingRanges(String keyspaceName, InetAddress endpoint)
{
List> ranges = new ArrayList<>();
for (Map.Entry, InetAddress> entry : getPendingRangesMM(keyspaceName).entries())
{
if (entry.getValue().equals(endpoint))
{
ranges.add(entry.getKey());
}
}
return ranges;
}
/**
* 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.
*/
public void calculatePendingRanges(AbstractReplicationStrategy strategy, String keyspaceName)
{
// avoid race between both branches - do not use a lock here as this will block any other unrelated operations!
synchronized (pendingRanges)
{
if (bootstrapTokens.isEmpty() && leavingEndpoints.isEmpty() && movingEndpoints.isEmpty())
{
if (logger.isTraceEnabled())
logger.trace("No bootstrapping, leaving or moving nodes -> empty pending ranges for {}", keyspaceName);
pendingRanges.put(keyspaceName, new PendingRangeMaps());
}
else
{
if (logger.isDebugEnabled())
logger.debug("Starting pending range calculation for {}", keyspaceName);
long startedAt = System.currentTimeMillis();
// create clone of current state
BiMultiValMap bootstrapTokens = new BiMultiValMap<>();
Set leavingEndpoints = new HashSet<>();
Set> movingEndpoints = new HashSet<>();
TokenMetadata metadata;
publicLock.readLock().lock();
lock.readLock().lock();
try
{
bootstrapTokens.putAll(this.bootstrapTokens);
leavingEndpoints.addAll(this.leavingEndpoints);
movingEndpoints.addAll(this.movingEndpoints);
metadata = this.cloneOnlyTokenMap();
}
finally
{
lock.readLock().unlock();
publicLock.readLock().unlock();
}
pendingRanges.put(keyspaceName, calculatePendingRanges(strategy, metadata, bootstrapTokens,
leavingEndpoints, movingEndpoints));
long took = System.currentTimeMillis() - startedAt;
if (logger.isDebugEnabled())
logger.debug("Pending range calculation for {} completed (took: {}ms)", keyspaceName, took);
if (logger.isTraceEnabled())
logger.trace("Calculated pending ranges for {}:\n{}", keyspaceName, (pendingRanges.isEmpty() ? "" : printPendingRanges()));
}
}
}
/**
* @see TokenMetadata#calculatePendingRanges(AbstractReplicationStrategy, String)
*/
private static PendingRangeMaps calculatePendingRanges(AbstractReplicationStrategy strategy,
TokenMetadata metadata,
BiMultiValMap bootstrapTokens,
Set leavingEndpoints,
Set> movingEndpoints)
{
PendingRangeMaps newPendingRanges = new PendingRangeMaps();
Multimap> addressRanges = strategy.getAddressRanges(metadata);
// Copy of metadata reflecting the situation after all leave operations are finished.
TokenMetadata allLeftMetadata = removeEndpoints(metadata.cloneOnlyTokenMap(), leavingEndpoints);
// get all ranges that will be affected by leaving nodes
Set> affectedRanges = new HashSet>();
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)
{
Set currentEndpoints = ImmutableSet.copyOf(strategy.calculateNaturalEndpoints(range.right, metadata));
Set newEndpoints = ImmutableSet.copyOf(strategy.calculateNaturalEndpoints(range.right, allLeftMetadata));
for (InetAddress address : Sets.difference(newEndpoints, currentEndpoints))
{
newPendingRanges.addPendingRange(range, address);
}
}
// At this stage newPendingRanges has been updated according to leave operations. We can
// now continue 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.
Multimap bootstrapAddresses = bootstrapTokens.inverse();
for (InetAddress endpoint : bootstrapAddresses.keySet())
{
Collection tokens = bootstrapAddresses.get(endpoint);
allLeftMetadata.updateNormalTokens(tokens, endpoint);
for (Range range : strategy.getAddressRanges(allLeftMetadata).get(endpoint))
{
newPendingRanges.addPendingRange(range, endpoint);
}
allLeftMetadata.removeEndpoint(endpoint);
}
// At this stage newPendingRanges has been updated according to leaving and bootstrapping nodes.
// We can now finish the calculation by checking moving nodes.
// For each of the moving nodes, we do the same thing we did for bootstrapping:
// simply add and remove them one by one to allLeftMetadata and check in between what their ranges would be.
for (Pair moving : movingEndpoints)
{
//Calculate all the ranges which will could be affected. This will include the ranges before and after the move.
Set> moveAffectedRanges = new HashSet<>();
InetAddress endpoint = moving.right; // address of the moving node
//Add ranges before the move
for (Range range : strategy.getAddressRanges(allLeftMetadata).get(endpoint))
{
moveAffectedRanges.add(range);
}
allLeftMetadata.updateNormalToken(moving.left, endpoint);
//Add ranges after the move
for (Range range : strategy.getAddressRanges(allLeftMetadata).get(endpoint))
{
moveAffectedRanges.add(range);
}
for(Range range : moveAffectedRanges)
{
Set currentEndpoints = ImmutableSet.copyOf(strategy.calculateNaturalEndpoints(range.right, metadata));
Set newEndpoints = ImmutableSet.copyOf(strategy.calculateNaturalEndpoints(range.right, allLeftMetadata));
Set difference = Sets.difference(newEndpoints, currentEndpoints);
for(final InetAddress address : difference)
{
Collection> newRanges = strategy.getAddressRanges(allLeftMetadata).get(address);
Collection> oldRanges = strategy.getAddressRanges(metadata).get(address);
//We want to get rid of any ranges which the node is currently getting.
newRanges.removeAll(oldRanges);
for(Range newRange : newRanges)
{
for(Range pendingRange : newRange.subtractAll(oldRanges))
{
newPendingRanges.addPendingRange(pendingRange, address);
}
}
}
}
allLeftMetadata.removeEndpoint(endpoint);
}
return newPendingRanges;
}
public Token getPredecessor(Token token)
{
List tokens = sortedTokens();
int index = Collections.binarySearch(tokens, token);
assert index >= 0 : token + " not found in " + StringUtils.join(tokenToEndpointMap.keySet(), ", ");
return (Token) (index == 0 ? tokens.get(tokens.size() - 1) : tokens.get(index - 1));
}
public Token getSuccessor(Token token)
{
List tokens = sortedTokens();
int index = Collections.binarySearch(tokens, token);
assert index >= 0 : token + " not found in " + StringUtils.join(tokenToEndpointMap.keySet(), ", ");
return (Token) ((index == (tokens.size() - 1)) ? tokens.get(0) : tokens.get(index + 1));
}
/** @return a copy of the bootstrapping tokens map */
public BiMultiValMap getBootstrapTokens()
{
publicLock.readLock().lock();
lock.readLock().lock();
try
{
return new BiMultiValMap(bootstrapTokens);
}
finally
{
lock.readLock().unlock();
publicLock.readLock().unlock();
}
}
public Set getAllEndpoints()
{
publicLock.readLock().lock();
lock.readLock().lock();
try
{
return ImmutableSet.copyOf(endpointToHostIdMap.keySet());
}
finally
{
lock.readLock().unlock();
publicLock.readLock().unlock();
}
}
/** caller should not modify leavingEndpoints */
public Set getLeavingEndpoints()
{
publicLock.readLock().lock();
lock.readLock().lock();
try
{
return ImmutableSet.copyOf(leavingEndpoints);
}
finally
{
lock.readLock().unlock();
publicLock.readLock().unlock();
}
}
/**
* Endpoints which are migrating to the new tokens
* @return set of addresses of moving endpoints
*/
public Set> getMovingEndpoints()
{
publicLock.readLock().lock();
lock.readLock().lock();
try
{
return ImmutableSet.copyOf(movingEndpoints);
}
finally
{
lock.readLock().unlock();
publicLock.readLock().unlock();
}
}
public static int firstTokenIndex(final ArrayList ring, Token start, boolean insertMin)
{
assert ring.size() > 0;
// insert the minimum token (at index == -1) if we were asked to include it and it isn't a member of the ring
int i = Collections.binarySearch(ring, start);
if (i < 0)
{
i = (i + 1) * (-1);
if (i >= ring.size())
i = insertMin ? -1 : 0;
}
return i;
}
public static Token firstToken(final ArrayList ring, Token start)
{
return ring.get(firstTokenIndex(ring, start, false));
}
/**
* iterator over the Tokens in the given ring, starting with the token for the node owning start
* (which does not have to be a Token in the ring)
* @param includeMin True if the minimum token should be returned in the ring even if it has no owner.
*/
public static Iterator ringIterator(final ArrayList ring, Token start, boolean includeMin)
{
if (ring.isEmpty())
return includeMin ? Iterators.singletonIterator(StorageService.getPartitioner().getMinimumToken())
: Collections.emptyIterator();
final boolean insertMin = includeMin && !ring.get(0).isMinimum();
final int startIndex = firstTokenIndex(ring, start, insertMin);
return new AbstractIterator()
{
int j = startIndex;
protected Token computeNext()
{
if (j < -1)
return endOfData();
try
{
// return minimum for index == -1
if (j == -1)
return StorageService.getPartitioner().getMinimumToken();
// return ring token for other indexes
return ring.get(j);
}
finally
{
j++;
if (j == ring.size())
j = insertMin ? -1 : 0;
if (j == startIndex)
// end iteration
j = -2;
}
}
};
}
/** used by tests */
public void clearUnsafe()
{
publicLock.readLock().lock();
lock.writeLock().lock();
try
{
tokenToEndpointMap.clear();
endpointToHostIdMap.clear();
bootstrapTokens.clear();
leavingEndpoints.clear();
pendingRanges.clear();
movingEndpoints.clear();
sortedTokens.clear();
topology.clear();
invalidateCachedRings();
}
finally
{
lock.writeLock().unlock();
publicLock.readLock().unlock();
}
}
public String toString()
{
StringBuilder sb = new StringBuilder();
publicLock.readLock().lock();
lock.readLock().lock();
try
{
Set eps = tokenToEndpointMap.inverse().keySet();
if (!eps.isEmpty())
{
sb.append("Normal Tokens:");
sb.append(System.getProperty("line.separator"));
for (InetAddress ep : eps)
{
sb.append(ep);
sb.append(':');
sb.append(tokenToEndpointMap.inverse().get(ep));
sb.append(System.getProperty("line.separator"));
}
}
if (!bootstrapTokens.isEmpty())
{
sb.append("Bootstrapping Tokens:" );
sb.append(System.getProperty("line.separator"));
for (Map.Entry entry : bootstrapTokens.entrySet())
{
sb.append(entry.getValue()).append(':').append(entry.getKey());
sb.append(System.getProperty("line.separator"));
}
}
if (!leavingEndpoints.isEmpty())
{
sb.append("Leaving Endpoints:");
sb.append(System.getProperty("line.separator"));
for (InetAddress ep : leavingEndpoints)
{
sb.append(ep);
sb.append(System.getProperty("line.separator"));
}
}
if (!pendingRanges.isEmpty())
{
sb.append("Pending Ranges:");
sb.append(System.getProperty("line.separator"));
sb.append(printPendingRanges());
}
}
finally
{
lock.readLock().unlock();
publicLock.readLock().unlock();
}
return sb.toString();
}
private String printPendingRanges()
{
StringBuilder sb = new StringBuilder();
for (PendingRangeMaps pendingRangeMaps : pendingRanges.values())
{
sb.append(pendingRangeMaps.printPendingRanges());
}
return sb.toString();
}
public Collection pendingEndpointsFor(Token token, String keyspaceName)
{
PendingRangeMaps pendingRangeMaps = this.pendingRanges.get(keyspaceName);
if (pendingRangeMaps == null)
return Collections.emptyList();
return pendingRangeMaps.pendingEndpointsFor(token);
}
/**
* @deprecated retained for benefit of old tests
*/
public Collection getWriteEndpoints(Token token, String keyspaceName, Collection naturalEndpoints)
{
return ImmutableList.copyOf(Iterables.concat(naturalEndpoints, pendingEndpointsFor(token, keyspaceName)));
}
/** @return an endpoint to token multimap representation of tokenToEndpointMap (a copy) */
public Multimap getEndpointToTokenMapForReading()
{
publicLock.readLock().lock();
lock.readLock().lock();
try
{
Multimap cloned = HashMultimap.create();
for (Map.Entry entry : tokenToEndpointMap.entrySet())
cloned.put(entry.getValue(), entry.getKey());
return cloned;
}
finally
{
lock.readLock().unlock();
publicLock.readLock().unlock();
}
}
/**
* @return a (stable copy, won't be modified) Token to Endpoint map for all the normal and bootstrapping nodes
* in the cluster.
*/
public Map getNormalAndBootstrappingTokenToEndpointMap()
{
publicLock.readLock().lock();
lock.readLock().lock();
try
{
Map map = new HashMap<>(tokenToEndpointMap.size() + bootstrapTokens.size());
map.putAll(tokenToEndpointMap);
map.putAll(bootstrapTokens);
return map;
}
finally
{
lock.readLock().unlock();
publicLock.readLock().unlock();
}
}
/**
* @return the Topology map of nodes to DCs + Racks
*
* This is only allowed when a copy has been made of TokenMetadata, to avoid concurrent modifications
* when Topology methods are subsequently used by the caller.
*/
public Topology getTopology()
{
assert this != StorageService.instance.getTokenMetadata();
return topology;
}
public long getRingVersion()
{
return ringVersion;
}
public void invalidateCachedRings()
{
ringVersion++;
cachedTokenMap.set(null);
}
/**
* Tracks the assignment of racks and endpoints in each datacenter for all the "normal" endpoints
* in this TokenMetadata. This allows faster calculation of endpoints in NetworkTopologyStrategy.
*/
public static class Topology
{
/** multi-map of DC to endpoints in that DC */
private final Multimap dcEndpoints;
/** map of DC to multi-map of rack to endpoints in that rack */
private final Map> dcRacks;
/** reverse-lookup map for endpoint to current known dc/rack assignment */
private final Map> currentLocations;
Topology()
{
dcEndpoints = HashMultimap.create();
dcRacks = new HashMap<>();
currentLocations = new HashMap<>();
}
void clear()
{
dcEndpoints.clear();
dcRacks.clear();
currentLocations.clear();
}
/**
* construct deep-copy of other
*/
Topology(Topology other)
{
dcEndpoints = HashMultimap.create(other.dcEndpoints);
dcRacks = new HashMap<>();
for (String dc : other.dcRacks.keySet())
dcRacks.put(dc, HashMultimap.create(other.dcRacks.get(dc)));
currentLocations = new HashMap<>(other.currentLocations);
}
/**
* Stores current DC/rack assignment for ep
*/
void addEndpoint(InetAddress ep)
{
IEndpointSnitch snitch = DatabaseDescriptor.getEndpointSnitch();
String dc = snitch.getDatacenter(ep);
String rack = snitch.getRack(ep);
Pair current = currentLocations.get(ep);
if (current != null)
{
if (current.left.equals(dc) && current.right.equals(rack))
return;
doRemoveEndpoint(ep, current);
}
doAddEndpoint(ep, dc, rack);
}
private void doAddEndpoint(InetAddress ep, String dc, String rack)
{
dcEndpoints.put(dc, ep);
if (!dcRacks.containsKey(dc))
dcRacks.put(dc, HashMultimap.create());
dcRacks.get(dc).put(rack, ep);
currentLocations.put(ep, Pair.create(dc, rack));
}
/**
* Removes current DC/rack assignment for ep
*/
void removeEndpoint(InetAddress ep)
{
if (!currentLocations.containsKey(ep))
return;
doRemoveEndpoint(ep, currentLocations.remove(ep));
}
private void doRemoveEndpoint(InetAddress ep, Pair current)
{
dcRacks.get(current.left).remove(current.right, ep);
dcEndpoints.remove(current.left, ep);
}
void updateEndpoint(InetAddress ep)
{
IEndpointSnitch snitch = DatabaseDescriptor.getEndpointSnitch();
if (snitch == null || !currentLocations.containsKey(ep))
return;
updateEndpoint(ep, snitch);
}
void updateEndpoints()
{
IEndpointSnitch snitch = DatabaseDescriptor.getEndpointSnitch();
if (snitch == null)
return;
for (InetAddress ep : currentLocations.keySet())
updateEndpoint(ep, snitch);
}
private void updateEndpoint(InetAddress ep, IEndpointSnitch snitch)
{
Pair current = currentLocations.get(ep);
String dc = snitch.getDatacenter(ep);
String rack = snitch.getRack(ep);
if (dc.equals(current.left) && rack.equals(current.right))
return;
doRemoveEndpoint(ep, current);
doAddEndpoint(ep, dc, rack);
}
/**
* @return multi-map of DC to endpoints in that DC
*/
public Multimap getDatacenterEndpoints()
{
return dcEndpoints;
}
/**
* @return map of DC to multi-map of rack to endpoints in that rack
*/
public Map> getDatacenterRacks()
{
return dcRacks;
}
}
}