alluxio.master.block

Class DefaultBlockMaster

java.lang.Object

alluxio.master.AbstractMaster

alluxio.master.CoreMaster

alluxio.master.block.DefaultBlockMaster

All Implemented Interfaces:

BlockMaster, ContainerIdGenerable, Checkpointed, Journaled, JournalEntryIterable, Master, Server<Boolean>

@NotThreadSafe
public class DefaultBlockMaster
extends CoreMaster
implements BlockMaster

This block master manages the metadata for all the blocks and block workers in Alluxio.

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
class	DefaultBlockMaster.LostWorkerDetectionHeartbeatExecutor Lost worker periodic check.
static class	DefaultBlockMaster.Metrics Class that contains metrics related to BlockMaster.
class	DefaultBlockMaster.WorkerRegisterStreamGCExecutor Periodically checks the open worker register streams.

Field Summary

Fields inherited from class alluxio.master.CoreMaster

mBackupManager, mJournalSystem, mPort, mPrimarySelector, mSafeModeManager, mStartTimeMs

Fields inherited from class alluxio.master.AbstractMaster

mClock, mJournal, mMasterContext

Constructor Summary

Constructors

Constructor and Description
DefaultBlockMaster(MetricsMaster metricsMaster, CoreMasterContext masterContext, java.time.Clock clock, ExecutorServiceFactory executorServiceFactory) Creates a new instance of DefaultBlockMaster.

Method Summary

Modifier and Type	Method and Description
void	close() Closes the server.
void	commitBlock(long workerId, long usedBytesOnTier, String tierAlias, String mediumType, long blockId, long length) Marks a block as committed on a specific worker.
void	commitBlockInUFS(long blockId, long length, JournalContext journalContext) Marks a block as committed, but without a worker location.
void	decommissionWorker(long workerId) Decommission a worker.
void	forgetAllWorkers() Forces all workers to be lost.
BlockInfo	getBlockInfo(long blockId)
List<BlockInfo>	getBlockInfoList(List<Long> blockIds) Retrieves information for the given list of block ids.
BlockMetaStore	getBlockMetaStore()
long	getBlockReplicaCount()
long	getCapacityBytes()
CheckpointName	getCheckpointName()
java.time.Clock	getClock() Returns the BlockMaster's clock so other components can align with the BlockMaster's time.
int	getDecommissionedWorkerCount()
Set<Class<? extends Server>>	getDependencies()
StorageTierAssoc	getGlobalStorageTierAssoc()
long	getJournaledNextContainerId() Returns the internal JournaledNextContainerId.
CloseableIterator<Journal.JournalEntry>	getJournalEntryIterator()
int	getLostBlocksCount()
Iterator<Long>	getLostBlocksIterator() Returns an Iterator over the lost blocks.
int	getLostWorkerCount()
List<WorkerInfo>	getLostWorkersInfoList()
String	getName()
long	getNewContainerId()
Map<ServiceType,GrpcService>	getServices()
Map<ServiceType,GrpcService>	getStandbyServices()
Map<String,Long>	getTotalBytesOnTiers()
long	getUniqueBlockCount()
long	getUsedBytes()
Map<String,Long>	getUsedBytesOnTiers()
MasterWorkerInfo	getWorker(long workerId) Returns the internal MasterWorkerInfo object to the caller.
Set<WorkerNetAddress>	getWorkerAddresses()
int	getWorkerCount()
long	getWorkerId(WorkerNetAddress workerNetAddress) Returns a worker id for the given worker, creating one if the worker is new.
List<WorkerInfo>	getWorkerInfoList()
List<WorkerLostStorageInfo>	getWorkerLostStorage()
List<WorkerInfo>	getWorkerReport(GetWorkerReportOptions options) Gets the worker information list for report CLI.
boolean	hasRegisterLease(long workerId) Verifies if the worker currently holds a RegisterLease.
boolean	isBlockLost(long blockId)
void	notifyWorkerId(long workerId, WorkerNetAddress workerNetAddress) Notify the worker id to a master.
boolean	processJournalEntry(Journal.JournalEntry entry) Attempts to apply a journal entry.
protected MasterWorkerInfo	recordWorkerRegistration(long workerId) Re-register a lost worker or complete registration after getting a worker id.
void	registerLostWorkerFoundListener(java.util.function.Consumer<Address> function) Registers callback functions to use when lost workers become alive.
void	registerNewWorkerConfListener(java.util.function.BiConsumer<Address,List<ConfigProperty>> function) Registers callback functions to use when workers register with configuration.
void	registerWorkerDeleteListener(java.util.function.Consumer<Address> function) Registers callback functions to use when detecting lost workers.
void	registerWorkerLostListener(java.util.function.Consumer<Address> function) Registers callback functions to use when detecting lost workers.
void	releaseRegisterLease(long workerId) Releases the RegisterLease for the specified worker.
void	removeBlocks(Collection<Long> blockIds, boolean delete) Removes blocks from workers.
void	removeDecommissionedWorker(long workerId) Removes all associated metadata about the decommissioned worker from block master.
void	reportLostBlocks(List<Long> blockIds) Reports the ids of the blocks lost on workers.
void	resetState() Resets the object's journaled state.
void	start(Boolean isLeader) Starts the Alluxio server.
void	stop() Stops the Alluxio server.
Optional<RegisterLease>	tryAcquireRegisterLease(GetRegisterLeasePRequest request) Try to acquire a RegisterLease for the worker.
void	validateBlocks(java.util.function.Function<Long,Boolean> validator, boolean repair) Validates the integrity of blocks with respect to the validator.
Command	workerHeartbeat(long workerId, Map<String,Long> capacityBytesOnTiers, Map<String,Long> usedBytesOnTiers, List<Long> removedBlockIds, Map<Block.BlockLocation,List<Long>> addedBlocks, Map<String,StorageList> lostStorage, List<Metric> metrics) Updates metadata when a worker periodically heartbeats with the master.
void	workerRegister(long workerId, List<String> storageTiers, Map<String,Long> totalBytesOnTiers, Map<String,Long> usedBytesOnTiers, Map<Block.BlockLocation,List<Long>> currentBlocksOnLocation, Map<String,StorageList> lostStorage, RegisterWorkerPOptions options) Updates metadata when a worker registers with the master.
protected void	workerRegisterBatch(WorkerRegisterContext context, RegisterWorkerPRequest chunk)
void	workerRegisterFinish(WorkerRegisterContext context) Completes the worker registration stream.
protected void	workerRegisterStart(WorkerRegisterContext context, RegisterWorkerPRequest chunk)
void	workerRegisterStream(WorkerRegisterContext context, RegisterWorkerPRequest chunk, boolean isFirstMsg) Handles messages in a register stream.

Methods inherited from class alluxio.master.AbstractMaster

createJournalContext, getExecutorService, getMasterContext

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Methods inherited from interface alluxio.master.block.BlockMaster

commitBlockInUFS

Methods inherited from interface alluxio.master.Master

createJournalContext, getMasterContext

Methods inherited from interface alluxio.master.journal.Journaled

applyAndJournal, restoreFromCheckpoint, writeToCheckpoint

Constructor Detail

DefaultBlockMaster

public DefaultBlockMaster(MetricsMaster metricsMaster,
                          CoreMasterContext masterContext,
                          java.time.Clock clock,
                          ExecutorServiceFactory executorServiceFactory)

Creates a new instance of DefaultBlockMaster. Used for tests where we manually control the clock.

Parameters:

metricsMaster - the metrics master

masterContext - the context for Alluxio master

clock - the clock to use for determining the time

executorServiceFactory - a factory for creating the executor service to use for running maintenance threads

Method Detail

getName

public String getName()

Specified by:

getName in interface Server<Boolean>

Returns:

the server's name

getServices

public Map<ServiceType,GrpcService> getServices()

Specified by:

getServices in interface Server<Boolean>

Returns:

a map from service names to gRPC serviced that serve RPCs for this server

getStandbyServices

public Map<ServiceType,GrpcService> getStandbyServices()

Specified by:

getStandbyServices in interface Master

Returns:

a map from service names to gRPC services that serve RPCs for this master, if the master is a standby master.

processJournalEntry

public boolean processJournalEntry(Journal.JournalEntry entry)

Description copied from interface: Journaled

Attempts to apply a journal entry.

Specified by:

processJournalEntry in interface Journaled

Parameters:

entry - the entry to apply

Returns:

whether the entry type is supported by this journaled object

resetState

public void resetState()

Description copied from interface: Journaled

Resets the object's journaled state.

Specified by:

resetState in interface Journaled

getCheckpointName

public CheckpointName getCheckpointName()

Specified by:

getCheckpointName in interface Checkpointed

Returns:

a name for this checkpointed class

getJournalEntryIterator

public CloseableIterator<Journal.JournalEntry> getJournalEntryIterator()

Specified by:

getJournalEntryIterator in interface JournalEntryIterable

Returns:

an CloseableIterator that iterates all the journal entries

start

public void start(Boolean isLeader)
           throws IOException

Description copied from interface: Server

Starts the Alluxio server.

Specified by:

start in interface Server<Boolean>

Overrides:

start in class AbstractMaster

Parameters:

isLeader - the start options

Throws:

IOException

stop

public void stop()
          throws IOException

Description copied from interface: Server

Stops the Alluxio server. Here, anything created or started in #start(T) should be cleaned up and shutdown.

Specified by:

stop in interface Server<Boolean>

Overrides:

stop in class AbstractMaster

Throws:

IOException

close

public void close()
           throws IOException

Description copied from interface: Server

Closes the server.

Specified by:

close in interface Server<Boolean>

Overrides:

close in class AbstractMaster

Throws:

IOException

getWorkerCount

public int getWorkerCount()

Specified by:

getWorkerCount in interface BlockMaster

Returns:

the number of live workers

getLostWorkerCount

public int getLostWorkerCount()

Specified by:

getLostWorkerCount in interface BlockMaster

Returns:

the number of lost workers

getDecommissionedWorkerCount

public int getDecommissionedWorkerCount()

Specified by:

getDecommissionedWorkerCount in interface BlockMaster

Returns:

the number of decommissioned workers

getCapacityBytes

public long getCapacityBytes()

Specified by:

getCapacityBytes in interface BlockMaster

Returns:

the total capacity (in bytes) on all tiers, on all workers of Alluxio

getUniqueBlockCount

public long getUniqueBlockCount()

Specified by:

getUniqueBlockCount in interface BlockMaster

Returns:

the unique block count on all workers of Alluxio

getBlockReplicaCount

public long getBlockReplicaCount()

Specified by:

getBlockReplicaCount in interface BlockMaster

Returns:

the replica block count on all workers of Alluxio

getGlobalStorageTierAssoc

public StorageTierAssoc getGlobalStorageTierAssoc()

Specified by:

getGlobalStorageTierAssoc in interface BlockMaster

Returns:

the global storage tier mapping

getUsedBytes

public long getUsedBytes()

Specified by:

getUsedBytes in interface BlockMaster

Returns:

the total used bytes on all tiers, on all workers of Alluxio

getWorkerInfoList

public List<WorkerInfo> getWorkerInfoList()
                                   throws UnavailableException

Specified by:

getWorkerInfoList in interface BlockMaster

Returns:

a list of WorkerInfo objects representing the live workers in Alluxio

Throws:

UnavailableException

getLostWorkersInfoList

public List<WorkerInfo> getLostWorkersInfoList()
                                        throws UnavailableException

Specified by:

getLostWorkersInfoList in interface BlockMaster

Returns:

a list of WorkerInfos of lost workers

Throws:

UnavailableException

removeDecommissionedWorker

public void removeDecommissionedWorker(long workerId)
                                throws NotFoundException

Description copied from interface: BlockMaster

Removes all associated metadata about the decommissioned worker from block master. The worker to free must have been decommissioned.

Specified by:

removeDecommissionedWorker in interface BlockMaster

Parameters:

workerId - the workerId of target worker

Throws:

NotFoundException

getWorkerAddresses

public Set<WorkerNetAddress> getWorkerAddresses()
                                         throws UnavailableException

Specified by:

getWorkerAddresses in interface BlockMaster

Returns:

a set of live worker addresses

Throws:

UnavailableException

getWorkerReport

public List<WorkerInfo> getWorkerReport(GetWorkerReportOptions options)
                                 throws UnavailableException,
                                        InvalidArgumentException

Description copied from interface: BlockMaster

Gets the worker information list for report CLI.

Specified by:

getWorkerReport in interface BlockMaster

Parameters:

options - the GetWorkerReportOptions defines the info range

Returns:

a list of WorkerInfo objects representing the workers in Alluxio

Throws:

UnavailableException

InvalidArgumentException

getWorkerLostStorage

public List<WorkerLostStorageInfo> getWorkerLostStorage()

Specified by:

getWorkerLostStorage in interface BlockMaster

Returns:

a list of worker lost storage information

removeBlocks

public void removeBlocks(Collection<Long> blockIds,
                         boolean delete)
                  throws UnavailableException

Description copied from interface: BlockMaster

Removes blocks from workers.

Specified by:

removeBlocks in interface BlockMaster

Parameters:

blockIds - a list of block ids to remove from Alluxio space

delete - whether to delete blocks' metadata in Master

Throws:

UnavailableException

decommissionWorker

public void decommissionWorker(long workerId)
                        throws Exception

Description copied from interface: BlockMaster

Decommission a worker.

Specified by:

decommissionWorker in interface BlockMaster

Parameters:

workerId - the WorkerInfo of worker to be decommissioned

Throws:

Exception

validateBlocks

public void validateBlocks(java.util.function.Function<Long,Boolean> validator,
                           boolean repair)
                    throws UnavailableException

Description copied from interface: BlockMaster

Validates the integrity of blocks with respect to the validator. A warning will be printed if blocks are invalid.

Specified by:

validateBlocks in interface BlockMaster

Parameters:

validator - a function returns true if the given block id is valid

repair - if true, deletes the invalid blocks

Throws:

UnavailableException - if the invalid blocks cannot be deleted

getJournaledNextContainerId

public long getJournaledNextContainerId()

Description copied from interface: BlockMaster

Returns the internal JournaledNextContainerId.

Specified by:

getJournaledNextContainerId in interface BlockMaster

Returns:

mJournaledNextContainerId

getNewContainerId

public long getNewContainerId()
                       throws UnavailableException

Specified by:

getNewContainerId in interface ContainerIdGenerable

Returns:

a new block container id

Throws:

UnavailableException

commitBlock

public void commitBlock(long workerId,
                        long usedBytesOnTier,
                        String tierAlias,
                        String mediumType,
                        long blockId,
                        long length)
                 throws NotFoundException,
                        UnavailableException

Description copied from interface: BlockMaster

Marks a block as committed on a specific worker.

Specified by:

commitBlock in interface BlockMaster

Parameters:

workerId - the worker id committing the block

usedBytesOnTier - the updated used bytes on the tier of the worker

tierAlias - the alias of the storage tier where the worker is committing the block to

mediumType - the medium type where the worker is committing the block to

blockId - the committing block id

length - the length of the block

Throws:

NotFoundException - if the workerId is not active

UnavailableException

commitBlockInUFS

public void commitBlockInUFS(long blockId,
                             long length,
                             JournalContext journalContext)

Description copied from interface: BlockMaster

Marks a block as committed, but without a worker location. This means the block is only in ufs. Append any created journal entries to the included context.

Specified by:

commitBlockInUFS in interface BlockMaster

Parameters:

blockId - the id of the block to commit

length - the length of the block

journalContext - the journal context

getBlockInfo

public BlockInfo getBlockInfo(long blockId)
                       throws BlockInfoException,
                              UnavailableException

Specified by:

getBlockInfo in interface BlockMaster

Parameters:

blockId - the block id to get information for

Returns:

the BlockInfo for the given block id

Throws:

BlockInfoException - if the block info is not found

UnavailableException

getBlockInfoList

public List<BlockInfo> getBlockInfoList(List<Long> blockIds)
                                 throws UnavailableException

Description copied from interface: BlockMaster

Retrieves information for the given list of block ids.

Specified by:

getBlockInfoList in interface BlockMaster

Parameters:

blockIds - A list of block ids to retrieve the information for

Returns:

A list of BlockInfo objects corresponding to the input list of block ids. The list is in the same order as the input list

Throws:

UnavailableException

getTotalBytesOnTiers

public Map<String,Long> getTotalBytesOnTiers()

Specified by:

getTotalBytesOnTiers in interface BlockMaster

Returns:

the total bytes on each storage tier

getUsedBytesOnTiers

public Map<String,Long> getUsedBytesOnTiers()

Specified by:

getUsedBytesOnTiers in interface BlockMaster

Returns:

the used bytes on each storage tier

recordWorkerRegistration

@Nullable
protected MasterWorkerInfo recordWorkerRegistration(long workerId)

Re-register a lost worker or complete registration after getting a worker id. This method requires no locking on MasterWorkerInfo because it is only reading final fields.

Parameters:

workerId - the worker id to register

getWorkerId

public long getWorkerId(WorkerNetAddress workerNetAddress)

Description copied from interface: BlockMaster

Returns a worker id for the given worker, creating one if the worker is new.

Specified by:

getWorkerId in interface BlockMaster

Parameters:

workerNetAddress - the worker WorkerNetAddress

Returns:

the worker id for this worker

notifyWorkerId

public void notifyWorkerId(long workerId,
                           WorkerNetAddress workerNetAddress)

Description copied from interface: BlockMaster

Notify the worker id to a master.

Specified by:

notifyWorkerId in interface BlockMaster

Parameters:

workerId - the worker id

workerNetAddress - the worker address

tryAcquireRegisterLease

public Optional<RegisterLease> tryAcquireRegisterLease(GetRegisterLeasePRequest request)

Description copied from interface: BlockMaster

Try to acquire a RegisterLease for the worker. If the lease is not granted, this will return empty immediately rather than blocking.

Specified by:

tryAcquireRegisterLease in interface BlockMaster

Parameters:

request - the request with all information for the master to make a decision with

Returns:

if empty, that means the lease is not granted

hasRegisterLease

public boolean hasRegisterLease(long workerId)

Description copied from interface: BlockMaster

Verifies if the worker currently holds a RegisterLease.

Specified by:

hasRegisterLease in interface BlockMaster

Parameters:

workerId - the worker ID

Returns:

whether a lease is found

releaseRegisterLease

public void releaseRegisterLease(long workerId)

Description copied from interface: BlockMaster

Releases the RegisterLease for the specified worker. If the worker currently does not hold a lease, return without throwing an error. The lease may have been recycled already.

Specified by:

releaseRegisterLease in interface BlockMaster

Parameters:

workerId - the worker ID

workerRegister

public void workerRegister(long workerId,
                           List<String> storageTiers,
                           Map<String,Long> totalBytesOnTiers,
                           Map<String,Long> usedBytesOnTiers,
                           Map<Block.BlockLocation,List<Long>> currentBlocksOnLocation,
                           Map<String,StorageList> lostStorage,
                           RegisterWorkerPOptions options)
                    throws NotFoundException

Description copied from interface: BlockMaster

Updates metadata when a worker registers with the master.

Specified by:

workerRegister in interface BlockMaster

Parameters:

workerId - the worker id of the worker registering

storageTiers - a list of storage tier aliases in order of their position in the worker's hierarchy

totalBytesOnTiers - a mapping from storage tier alias to total bytes

usedBytesOnTiers - a mapping from storage tier alias to the used byes

currentBlocksOnLocation - a mapping from storage tier alias to a list of blocks

lostStorage - a mapping from storage tier alias to a list of lost storage paths

options - the options that may contain worker configuration

Throws:

NotFoundException - if workerId cannot be found

getWorker

public MasterWorkerInfo getWorker(long workerId)
                           throws NotFoundException

Description copied from interface: BlockMaster

Returns the internal MasterWorkerInfo object to the caller. This is specifically for the tests and the WorkerRegisterContext. Note that this operations on the object requires locking. See the javadoc of MasterWorkerInfo for how the locking should be carefully done. When in doubt, do not use this API. Find other methods in this class that exposes necessary information.

Specified by:

getWorker in interface BlockMaster

Parameters:

workerId - the worker ID

Returns:

the MasterWorkerInfo for the worker

Throws:

NotFoundException

workerRegisterStream

public void workerRegisterStream(WorkerRegisterContext context,
                                 RegisterWorkerPRequest chunk,
                                 boolean isFirstMsg)

Description copied from interface: BlockMaster

Handles messages in a register stream.

Specified by:

workerRegisterStream in interface BlockMaster

Parameters:

context - the stream context that contains the worker information

chunk - the message in a stream

isFirstMsg - whether the message is the 1st in a stream

workerRegisterStart

protected void workerRegisterStart(WorkerRegisterContext context,
                                   RegisterWorkerPRequest chunk)

workerRegisterBatch

protected void workerRegisterBatch(WorkerRegisterContext context,
                                   RegisterWorkerPRequest chunk)

workerRegisterFinish

public void workerRegisterFinish(WorkerRegisterContext context)

Description copied from interface: BlockMaster

Completes the worker registration stream.

Specified by:

workerRegisterFinish in interface BlockMaster

Parameters:

context - the stream context to be closed

workerHeartbeat

public Command workerHeartbeat(long workerId,
                               Map<String,Long> capacityBytesOnTiers,
                               Map<String,Long> usedBytesOnTiers,
                               List<Long> removedBlockIds,
                               Map<Block.BlockLocation,List<Long>> addedBlocks,
                               Map<String,StorageList> lostStorage,
                               List<Metric> metrics)

Description copied from interface: BlockMaster

Updates metadata when a worker periodically heartbeats with the master.

Specified by:

workerHeartbeat in interface BlockMaster

Parameters:

workerId - the worker id

capacityBytesOnTiers - a mapping from tier alias to the capacity bytes

usedBytesOnTiers - a mapping from tier alias to the used bytes

removedBlockIds - a list of block ids removed from this worker

addedBlocks - a mapping from tier alias to the added blocks

lostStorage - a mapping from tier alias to lost storage paths

metrics - worker metrics

Returns:

an optional command for the worker to execute

getClock

public java.time.Clock getClock()

Description copied from interface: BlockMaster

Returns the BlockMaster's clock so other components can align with the BlockMaster's time.

Specified by:

getClock in interface BlockMaster

Returns:

the current clock

isBlockLost

public boolean isBlockLost(long blockId)

Specified by:

isBlockLost in interface BlockMaster

Parameters:

blockId - the block ID

Returns:

whether the block is considered lost in Alluxio

getLostBlocksIterator

public Iterator<Long> getLostBlocksIterator()

Description copied from interface: BlockMaster

Returns an Iterator over the lost blocks. Note that the iterator should not be shared across threads.

Specified by:

getLostBlocksIterator in interface BlockMaster

Returns:

an Iterator

getLostBlocksCount

public int getLostBlocksCount()

Specified by:

getLostBlocksCount in interface BlockMaster

Returns:

the number of lost blocks in Alluxio

reportLostBlocks

public void reportLostBlocks(List<Long> blockIds)

Description copied from interface: BlockMaster

Reports the ids of the blocks lost on workers.

Specified by:

reportLostBlocks in interface BlockMaster

Parameters:

blockIds - the ids of the lost blocks

getDependencies

public Set<Class<? extends Server>> getDependencies()

Specified by:

getDependencies in interface Server<Boolean>

Overrides:

getDependencies in class AbstractMaster

Returns:

the server dependencies

forgetAllWorkers

public void forgetAllWorkers()

Forces all workers to be lost. This should only be used for testing.

registerLostWorkerFoundListener

public void registerLostWorkerFoundListener(java.util.function.Consumer<Address> function)

Description copied from interface: BlockMaster

Registers callback functions to use when lost workers become alive.

Specified by:

registerLostWorkerFoundListener in interface BlockMaster

Parameters:

function - the function to register

registerWorkerLostListener

public void registerWorkerLostListener(java.util.function.Consumer<Address> function)

Description copied from interface: BlockMaster

Registers callback functions to use when detecting lost workers.

Specified by:

registerWorkerLostListener in interface BlockMaster

Parameters:

function - the function to register

registerWorkerDeleteListener

public void registerWorkerDeleteListener(java.util.function.Consumer<Address> function)

Description copied from interface: BlockMaster

Registers callback functions to use when detecting lost workers.

Specified by:

registerWorkerDeleteListener in interface BlockMaster

Parameters:

function - the function to register

registerNewWorkerConfListener

public void registerNewWorkerConfListener(java.util.function.BiConsumer<Address,List<ConfigProperty>> function)

Description copied from interface: BlockMaster

Registers callback functions to use when workers register with configuration.

Specified by:

registerNewWorkerConfListener in interface BlockMaster

Parameters:

function - the function to register

getBlockMetaStore

public BlockMetaStore getBlockMetaStore()

Returns:

the block meta store