/** Return all of the neighbors with whom we share the provided range. */
static Set<InetAddress> getNeighbors(String table, Range<Token> toRepair) {
StorageService ss = StorageService.instance;
Map<Range<Token>, List<InetAddress>> replicaSets = ss.getRangeToAddressMap(table);
Range<Token> rangeSuperSet = null;
for (Range<Token> range : ss.getLocalRanges(table)) {
if (range.contains(toRepair)) {
rangeSuperSet = range;
break;
} else if (range.intersects(toRepair)) {
throw new IllegalArgumentException(
"Requested range intersects a local range but is not fully contained in one; this would lead to imprecise repair");
}
}
if (rangeSuperSet == null || !replicaSets.containsKey(toRepair)) return Collections.emptySet();
Set<InetAddress> neighbors = new HashSet<InetAddress>(replicaSets.get(rangeSuperSet));
neighbors.remove(FBUtilities.getBroadcastAddress());
// Excluding all node with version <= 0.7 since they don't know how to
// create a correct merkle tree (they build it over the full range)
Iterator<InetAddress> iter = neighbors.iterator();
while (iter.hasNext()) {
InetAddress endpoint = iter.next();
if (Gossiper.instance.getVersion(endpoint) <= MessagingService.VERSION_07) {
logger.info(
"Excluding "
+ endpoint
+ " from repair because it is on version 0.7 or sooner. You should consider updating this node before running repair again.");
iter.remove();
}
}
return neighbors;
}
public void handleStreamEvent(StreamEvent event) {
if (event.eventType == StreamEvent.Type.STREAM_PREPARED) {
SessionInfo session = ((StreamEvent.SessionPreparedEvent) event).session;
sessionsByHost.put(session.peer, session);
} else if (event.eventType == StreamEvent.Type.FILE_PROGRESS) {
ProgressInfo progressInfo = ((StreamEvent.ProgressEvent) event).progress;
// update progress
Set<ProgressInfo> progresses = progressByHost.get(progressInfo.peer);
if (progresses == null) {
progresses = Sets.newSetFromMap(new ConcurrentHashMap<ProgressInfo, Boolean>());
progressByHost.put(progressInfo.peer, progresses);
}
if (progresses.contains(progressInfo)) progresses.remove(progressInfo);
progresses.add(progressInfo);
StringBuilder sb = new StringBuilder();
sb.append("\rprogress: ");
long totalProgress = 0;
long totalSize = 0;
for (Map.Entry<InetAddress, Set<ProgressInfo>> entry : progressByHost.entrySet()) {
SessionInfo session = sessionsByHost.get(entry.getKey());
long size = session.getTotalSizeToSend();
long current = 0;
int completed = 0;
for (ProgressInfo progress : entry.getValue()) {
if (progress.currentBytes == progress.totalBytes) completed++;
current += progress.currentBytes;
}
totalProgress += current;
totalSize += size;
sb.append("[").append(entry.getKey());
sb.append(" ").append(completed).append("/").append(session.getTotalFilesToSend());
sb.append(" (").append(size == 0 ? 100L : current * 100L / size).append("%)] ");
}
long time = System.nanoTime();
long deltaTime = Math.max(1L, TimeUnit.NANOSECONDS.toMillis(time - lastTime));
lastTime = time;
long deltaProgress = totalProgress - lastProgress;
lastProgress = totalProgress;
sb.append("[total: ")
.append(totalSize == 0 ? 100L : totalProgress * 100L / totalSize)
.append("% - ");
sb.append(mbPerSec(deltaProgress, deltaTime)).append("MB/s");
sb.append(" (avg: ")
.append(mbPerSec(totalProgress, TimeUnit.NANOSECONDS.toMillis(time - start)))
.append("MB/s)]");
System.out.print(sb.toString());
}
}
public void convict(InetAddress endpoint, double phi) {
if (!endpoints.contains(endpoint)) return;
// We want a higher confidence in the failure detection than usual because failing a repair
// wrongly has a high cost.
if (phi < 2 * DatabaseDescriptor.getPhiConvictThreshold()) return;
// Though unlikely, it is possible to arrive here multiple time and we
// want to avoid print an error message twice
if (!isFailed.compareAndSet(false, true)) return;
failedNode(endpoint);
}
@Override
public void init(String keyspace) {
Iterator<InetAddress> hostiter = hosts.iterator();
while (hostiter.hasNext()) {
try {
// Query endpoint to ranges map and schemas from thrift
InetAddress host = hostiter.next();
Cassandra.Client client =
createThriftClient(
host.getHostAddress(), rpcPort, this.user, this.passwd, this.transportFactory);
setPartitioner(client.describe_partitioner());
Token.TokenFactory tkFactory = getPartitioner().getTokenFactory();
for (TokenRange tr : client.describe_ring(keyspace)) {
Range<Token> range =
new Range<>(
tkFactory.fromString(tr.start_token),
tkFactory.fromString(tr.end_token),
getPartitioner());
for (String ep : tr.endpoints) {
addRangeForEndpoint(range, InetAddress.getByName(ep));
}
}
String query =
String.format(
"SELECT * FROM %s.%s WHERE keyspace_name = '%s'",
Keyspace.SYSTEM_KS, SystemKeyspace.SCHEMA_COLUMNFAMILIES_CF, keyspace);
CqlResult result =
client.execute_cql3_query(
ByteBufferUtil.bytes(query), Compression.NONE, ConsistencyLevel.ONE);
for (CqlRow row : result.rows) {
CFMetaData metadata = CFMetaData.fromThriftCqlRow(row);
knownCfs.put(metadata.cfName, metadata);
}
break;
} catch (Exception e) {
if (!hostiter.hasNext())
throw new RuntimeException("Could not retrieve endpoint ranges: ", e);
}
}
}
// we don't care about the return value but care about it throwing exception
public void runMayThrow() throws Exception {
logger.info(
String.format(
"[repair #%s] new session: will sync %s on range %s for %s.%s",
getName(), repairedNodes(), range, tablename, Arrays.toString(cfnames)));
if (endpoints.isEmpty()) {
differencingDone.signalAll();
logger.info(
String.format(
"[repair #%s] No neighbors to repair with on range %s: session completed",
getName(), range));
return;
}
// Checking all nodes are live
for (InetAddress endpoint : endpoints) {
if (!FailureDetector.instance.isAlive(endpoint)) {
differencingDone.signalAll();
logger.info(
String.format(
"[repair #%s] Cannot proceed on repair because a neighbor (%s) is dead: session failed",
getName(), endpoint));
return;
}
if (Gossiper.instance.getVersion(endpoint) < MessagingService.VERSION_11 && isSequential) {
logger.info(
String.format(
"[repair #%s] Cannot repair using snapshots as node %s is pre-1.1",
getName(), endpoint));
return;
}
}
AntiEntropyService.instance.sessions.put(getName(), this);
Gossiper.instance.register(this);
FailureDetector.instance.registerFailureDetectionEventListener(this);
try {
// Create and queue a RepairJob for each column family
for (String cfname : cfnames) {
RepairJob job = new RepairJob(cfname);
jobs.offer(job);
activeJobs.put(cfname, job);
}
jobs.peek().sendTreeRequests();
// block whatever thread started this session until all requests have been returned:
// if this thread dies, the session will still complete in the background
completed.await();
if (exception == null) {
logger.info(String.format("[repair #%s] session completed successfully", getName()));
} else {
logger.error(
String.format("[repair #%s] session completed with the following error", getName()),
exception);
throw exception;
}
} catch (InterruptedException e) {
throw new RuntimeException("Interrupted while waiting for repair.");
} finally {
// mark this session as terminated
terminate();
FailureDetector.instance.unregisterFailureDetectionEventListener(this);
Gossiper.instance.unregister(this);
AntiEntropyService.instance.sessions.remove(getName());
}
}
public int completed(R request) {
requests.remove(request);
return requests.size();
}
public void add(R request) {
requests.add(request);
}