/**
* Creates a {@link BlockInfo} form a given {@link MasterBlockInfo}, by populating worker
* locations.
*
* <p>NOTE: {@link #mWorkers} should already be locked before calling this method.
*
* @param masterBlockInfo the {@link MasterBlockInfo}
* @return a {@link BlockInfo} from a {@link MasterBlockInfo}. Populates worker locations
*/
private BlockInfo generateBlockInfo(MasterBlockInfo masterBlockInfo) {
// "Join" to get all the addresses of the workers.
List<BlockLocation> locations = new ArrayList<BlockLocation>();
List<MasterBlockLocation> blockLocations = masterBlockInfo.getBlockLocations();
// Sort the block locations by their alias ordinal in the master storage tier mapping
Collections.sort(
blockLocations,
new Comparator<MasterBlockLocation>() {
@Override
public int compare(MasterBlockLocation o1, MasterBlockLocation o2) {
return mGlobalStorageTierAssoc.getOrdinal(o1.getTierAlias())
- mGlobalStorageTierAssoc.getOrdinal(o2.getTierAlias());
}
});
for (MasterBlockLocation masterBlockLocation : blockLocations) {
MasterWorkerInfo workerInfo =
mWorkers.getFirstByField(mIdIndex, masterBlockLocation.getWorkerId());
if (workerInfo != null) {
locations.add(
new BlockLocation()
.setWorkerId(masterBlockLocation.getWorkerId())
.setWorkerAddress(workerInfo.getWorkerAddress())
.setTierAlias(masterBlockLocation.getTierAlias()));
}
}
return new BlockInfo()
.setBlockId(masterBlockInfo.getBlockId())
.setLength(masterBlockInfo.getLength())
.setLocations(locations);
}
/**
* Updates metadata when a worker periodically heartbeats with the master.
*
* @param workerId the worker id
* @param usedBytesOnTiers a mapping from tier alias to the used bytes
* @param removedBlockIds a list of block ids removed from this worker
* @param addedBlocksOnTiers a mapping from tier alias to the added blocks
* @return an optional command for the worker to execute
*/
public Command workerHeartbeat(
long workerId,
Map<String, Long> usedBytesOnTiers,
List<Long> removedBlockIds,
Map<String, List<Long>> addedBlocksOnTiers) {
synchronized (mBlocks) {
synchronized (mWorkers) {
if (!mWorkers.contains(mIdIndex, workerId)) {
LOG.warn("Could not find worker id: {} for heartbeat.", workerId);
return new Command(CommandType.Register, new ArrayList<Long>());
}
MasterWorkerInfo workerInfo = mWorkers.getFirstByField(mIdIndex, workerId);
processWorkerRemovedBlocks(workerInfo, removedBlockIds);
processWorkerAddedBlocks(workerInfo, addedBlocksOnTiers);
workerInfo.updateUsedBytes(usedBytesOnTiers);
workerInfo.updateLastUpdatedTimeMs();
List<Long> toRemoveBlocks = workerInfo.getToRemoveBlocks();
if (toRemoveBlocks.isEmpty()) {
return new Command(CommandType.Nothing, new ArrayList<Long>());
}
return new Command(CommandType.Free, toRemoveBlocks);
}
}
}
/**
* Returns a worker id for the given worker.
*
* @param workerNetAddress the worker {@link WorkerNetAddress}
* @return the worker id for this worker
*/
public long getWorkerId(WorkerNetAddress workerNetAddress) {
// TODO(gene): This NetAddress cloned in case thrift re-uses the object. Does thrift re-use it?
synchronized (mWorkers) {
if (mWorkers.contains(mAddressIndex, workerNetAddress)) {
// This worker address is already mapped to a worker id.
long oldWorkerId = mWorkers.getFirstByField(mAddressIndex, workerNetAddress).getId();
LOG.warn("The worker {} already exists as id {}.", workerNetAddress, oldWorkerId);
return oldWorkerId;
}
if (mLostWorkers.contains(mAddressIndex, workerNetAddress)) {
// this is one of the lost workers
final MasterWorkerInfo lostWorkerInfo =
mLostWorkers.getFirstByField(mAddressIndex, workerNetAddress);
final long lostWorkerId = lostWorkerInfo.getId();
LOG.warn("A lost worker {} has requested its old id {}.", workerNetAddress, lostWorkerId);
// Update the timestamp of the worker before it is considered an active worker.
lostWorkerInfo.updateLastUpdatedTimeMs();
mWorkers.add(lostWorkerInfo);
mLostWorkers.remove(lostWorkerInfo);
return lostWorkerId;
}
// Generate a new worker id.
long workerId = mNextWorkerId.getAndIncrement();
mWorkers.add(new MasterWorkerInfo(workerId, workerNetAddress));
LOG.info("getWorkerId(): WorkerNetAddress: {} id: {}", workerNetAddress, workerId);
return workerId;
}
}
/**
* Updates metadata when a worker registers with the master.
*
* @param workerId the worker id of the worker registering
* @param storageTiers a list of storage tier aliases in order of their position in the worker's
* hierarchy
* @param totalBytesOnTiers a mapping from storage tier alias to total bytes
* @param usedBytesOnTiers a mapping from storage tier alias to the used byes
* @param currentBlocksOnTiers a mapping from storage tier alias to a list of blocks
* @throws NoWorkerException if workerId cannot be found
*/
public void workerRegister(
long workerId,
List<String> storageTiers,
Map<String, Long> totalBytesOnTiers,
Map<String, Long> usedBytesOnTiers,
Map<String, List<Long>> currentBlocksOnTiers)
throws NoWorkerException {
synchronized (mBlocks) {
synchronized (mWorkers) {
if (!mWorkers.contains(mIdIndex, workerId)) {
throw new NoWorkerException("Could not find worker id: " + workerId + " to register.");
}
MasterWorkerInfo workerInfo = mWorkers.getFirstByField(mIdIndex, workerId);
workerInfo.updateLastUpdatedTimeMs();
// Gather all blocks on this worker.
HashSet<Long> blocks = new HashSet<Long>();
for (List<Long> blockIds : currentBlocksOnTiers.values()) {
blocks.addAll(blockIds);
}
// Detect any lost blocks on this worker.
Set<Long> removedBlocks =
workerInfo.register(
mGlobalStorageTierAssoc, storageTiers, totalBytesOnTiers, usedBytesOnTiers, blocks);
processWorkerRemovedBlocks(workerInfo, removedBlocks);
processWorkerAddedBlocks(workerInfo, currentBlocksOnTiers);
LOG.info("registerWorker(): {}", workerInfo);
}
}
}
/**
* Marks a block as committed on a specific worker.
*
* @param workerId the worker id committing the block
* @param usedBytesOnTier the updated used bytes on the tier of the worker
* @param tierAlias the alias of the storage tier where the worker is committing the block to
* @param blockId the committing block id
* @param length the length of the block
*/
public void commitBlock(
long workerId, long usedBytesOnTier, String tierAlias, long blockId, long length) {
LOG.debug(
"Commit block from worker: {}",
FormatUtils.parametersToString(workerId, usedBytesOnTier, blockId, length));
synchronized (mBlocks) {
synchronized (mWorkers) {
MasterWorkerInfo workerInfo = mWorkers.getFirstByField(mIdIndex, workerId);
workerInfo.addBlock(blockId);
workerInfo.updateUsedBytes(tierAlias, usedBytesOnTier);
workerInfo.updateLastUpdatedTimeMs();
MasterBlockInfo masterBlockInfo = mBlocks.get(blockId);
if (masterBlockInfo == null) {
masterBlockInfo = new MasterBlockInfo(blockId, length);
mBlocks.put(blockId, masterBlockInfo);
BlockInfoEntry blockInfo =
BlockInfoEntry.newBuilder()
.setBlockId(masterBlockInfo.getBlockId())
.setLength(masterBlockInfo.getLength())
.build();
writeJournalEntry(JournalEntry.newBuilder().setBlockInfo(blockInfo).build());
flushJournal();
}
masterBlockInfo.addWorker(workerId, tierAlias);
mLostBlocks.remove(blockId);
}
}
}
/**
* Gets info about the lost workers.
*
* @return a set of worker info
*/
public Set<WorkerInfo> getLostWorkersInfo() {
synchronized (mWorkers) {
Set<WorkerInfo> ret = new HashSet<WorkerInfo>(mLostWorkers.size());
for (MasterWorkerInfo worker : mLostWorkers) {
ret.add(worker.generateClientWorkerInfo());
}
return ret;
}
}
/** @return the total used bytes on all tiers, on all workers of Tachyon */
public long getUsedBytes() {
long ret = 0;
synchronized (mWorkers) {
for (MasterWorkerInfo worker : mWorkers) {
ret += worker.getUsedBytes();
}
}
return ret;
}
/** @return a list of {@link WorkerInfo} objects representing the workers in Tachyon */
public List<WorkerInfo> getWorkerInfoList() {
synchronized (mWorkers) {
List<WorkerInfo> workerInfoList = new ArrayList<WorkerInfo>(mWorkers.size());
for (MasterWorkerInfo masterWorkerInfo : mWorkers) {
workerInfoList.add(masterWorkerInfo.generateClientWorkerInfo());
}
return workerInfoList;
}
}
/** @return the used bytes on each storage tier */
public Map<String, Long> getUsedBytesOnTiers() {
Map<String, Long> ret = new HashMap<String, Long>();
synchronized (mWorkers) {
for (MasterWorkerInfo worker : mWorkers) {
for (Map.Entry<String, Long> entry : worker.getUsedBytesOnTiers().entrySet()) {
Long used = ret.get(entry.getKey());
ret.put(entry.getKey(), (used == null ? 0L : used) + entry.getValue());
}
}
}
return ret;
}
/**
* Updates the worker and block metadata for blocks added to a worker.
*
* <p>NOTE: {@link #mBlocks} should already be locked before calling this method.
*
* @param workerInfo The worker metadata object
* @param addedBlockIds A mapping from storage tier alias to a list of block ids added
*/
private void processWorkerAddedBlocks(
MasterWorkerInfo workerInfo, Map<String, List<Long>> addedBlockIds) {
for (Map.Entry<String, List<Long>> entry : addedBlockIds.entrySet()) {
for (long blockId : entry.getValue()) {
MasterBlockInfo masterBlockInfo = mBlocks.get(blockId);
if (masterBlockInfo != null) {
workerInfo.addBlock(blockId);
masterBlockInfo.addWorker(workerInfo.getId(), entry.getKey());
mLostBlocks.remove(blockId);
} else {
LOG.warn("Failed to register workerId: {} to blockId: {}", workerInfo.getId(), blockId);
}
}
}
}
/**
* Updates the worker and block metadata for blocks removed from a worker.
*
* <p>NOTE: {@link #mBlocks} should already be locked before calling this method.
*
* @param workerInfo The worker metadata object
* @param removedBlockIds A list of block ids removed from the worker
*/
private void processWorkerRemovedBlocks(
MasterWorkerInfo workerInfo, Collection<Long> removedBlockIds) {
for (long removedBlockId : removedBlockIds) {
MasterBlockInfo masterBlockInfo = mBlocks.get(removedBlockId);
if (masterBlockInfo == null) {
LOG.warn(
"Worker {} removed block {} but block does not exist.",
workerInfo.getId(),
removedBlockId);
// Continue to remove the remaining blocks.
continue;
}
LOG.info("Block {} is removed on worker {}.", removedBlockId, workerInfo.getId());
workerInfo.removeBlock(masterBlockInfo.getBlockId());
masterBlockInfo.removeWorker(workerInfo.getId());
if (masterBlockInfo.getNumLocations() == 0) {
mLostBlocks.add(removedBlockId);
}
}
}
/**
* Removes blocks from workers.
*
* @param blockIds a list of block ids to remove from Tachyon space
*/
public void removeBlocks(List<Long> blockIds) {
synchronized (mBlocks) {
synchronized (mWorkers) {
for (long blockId : blockIds) {
MasterBlockInfo masterBlockInfo = mBlocks.get(blockId);
if (masterBlockInfo == null) {
continue;
}
for (long workerId : new ArrayList<Long>(masterBlockInfo.getWorkers())) {
masterBlockInfo.removeWorker(workerId);
MasterWorkerInfo worker = mWorkers.getFirstByField(mIdIndex, workerId);
if (worker != null) {
worker.updateToRemovedBlock(true, blockId);
}
}
mLostBlocks.remove(blockId);
}
}
}
}
@Override
public void heartbeat() {
LOG.debug("System status checking.");
TachyonConf conf = MasterContext.getConf();
int masterWorkerTimeoutMs = conf.getInt(Constants.MASTER_WORKER_TIMEOUT_MS);
synchronized (mBlocks) {
synchronized (mWorkers) {
Iterator<MasterWorkerInfo> iter = mWorkers.iterator();
while (iter.hasNext()) {
MasterWorkerInfo worker = iter.next();
final long lastUpdate = CommonUtils.getCurrentMs() - worker.getLastUpdatedTimeMs();
if (lastUpdate > masterWorkerTimeoutMs) {
LOG.error("The worker {} got timed out!", worker);
mLostWorkers.add(worker);
iter.remove();
processLostWorker(worker);
}
}
}
}
}
// Synchronized on mBlocks by the caller
private void processLostWorker(MasterWorkerInfo latest) {
final Set<Long> lostBlocks = latest.getBlocks();
processWorkerRemovedBlocks(latest, lostBlocks);
}
@Override
public Object getFieldValue(MasterWorkerInfo o) {
return o.getWorkerAddress();
}
