org.apache.hadoop.hbase.tool

Class LoadIncrementalHFiles

java.lang.Object

org.apache.hadoop.conf.Configured

org.apache.hadoop.hbase.tool.LoadIncrementalHFiles

All Implemented Interfaces:

org.apache.hadoop.conf.Configurable, org.apache.hadoop.util.Tool

Direct Known Subclasses:

LoadIncrementalHFiles

@InterfaceAudience.Public
public class LoadIncrementalHFiles
extends org.apache.hadoop.conf.Configured
implements org.apache.hadoop.util.Tool

Tool to load the output of HFileOutputFormat into an existing table.

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
static class	LoadIncrementalHFiles.LoadQueueItem Represents an HFile waiting to be loaded.

Field Summary

Fields

Modifier and Type	Field and Description
static String	ALWAYS_COPY_FILES
static String	CREATE_TABLE_CONF_KEY
static String	IGNORE_UNMATCHED_CF_CONF_KEY
static String	MAX_FILES_PER_REGION_PER_FAMILY
static String	NAME

Constructor Summary

Constructors

Constructor and Description
LoadIncrementalHFiles(org.apache.hadoop.conf.Configuration conf)

Method Summary

Modifier and Type	Method and Description
protected ClientServiceCallable<byte[]>	buildClientServiceCallable(Connection conn, TableName tableName, byte[] first, Collection<LoadIncrementalHFiles.LoadQueueItem> lqis, boolean copyFile)
protected void	bulkLoadPhase(Table table, Connection conn, ExecutorService pool, Deque<LoadIncrementalHFiles.LoadQueueItem> queue, org.apache.hbase.thirdparty.com.google.common.collect.Multimap<ByteBuffer,LoadIncrementalHFiles.LoadQueueItem> regionGroups, boolean copyFile, Map<LoadIncrementalHFiles.LoadQueueItem,ByteBuffer> item2RegionMap) This takes the LQI's grouped by likely regions and attempts to bulk load them.
Map<LoadIncrementalHFiles.LoadQueueItem,ByteBuffer>	doBulkLoad(Map<byte[],List<org.apache.hadoop.fs.Path>> map, Admin admin, Table table, RegionLocator regionLocator, boolean silence, boolean copyFile) Perform a bulk load of the given directory into the given pre-existing table.
Map<LoadIncrementalHFiles.LoadQueueItem,ByteBuffer>	doBulkLoad(org.apache.hadoop.fs.Path hfofDir, Admin admin, Table table, RegionLocator regionLocator) Perform a bulk load of the given directory into the given pre-existing table.
Map<LoadIncrementalHFiles.LoadQueueItem,ByteBuffer>	doBulkLoad(org.apache.hadoop.fs.Path hfofDir, Admin admin, Table table, RegionLocator regionLocator, boolean silence, boolean copyFile) Perform a bulk load of the given directory into the given pre-existing table.
protected Pair<List<LoadIncrementalHFiles.LoadQueueItem>,String>	groupOrSplit(org.apache.hbase.thirdparty.com.google.common.collect.Multimap<ByteBuffer,LoadIncrementalHFiles.LoadQueueItem> regionGroups, LoadIncrementalHFiles.LoadQueueItem item, Table table, Pair<byte[][],byte[][]> startEndKeys) Attempt to assign the given load queue item into its target region group.
static byte[][]	inferBoundaries(SortedMap<byte[],Integer> bdryMap) Infers region boundaries for a new table.
void	loadHFileQueue(Table table, Connection conn, Deque<LoadIncrementalHFiles.LoadQueueItem> queue, Pair<byte[][],byte[][]> startEndKeys) Used by the replication sink to load the hfiles from the source cluster.
void	loadHFileQueue(Table table, Connection conn, Deque<LoadIncrementalHFiles.LoadQueueItem> queue, Pair<byte[][],byte[][]> startEndKeys, boolean copyFile) Used by the replication sink to load the hfiles from the source cluster.
static void	main(String[] args)
void	prepareHFileQueue(Map<byte[],List<org.apache.hadoop.fs.Path>> map, Table table, Deque<LoadIncrementalHFiles.LoadQueueItem> queue, boolean silence) Prepare a collection of LoadIncrementalHFiles.LoadQueueItem from list of source hfiles contained in the passed directory and validates whether the prepared queue has all the valid table column families in it.
void	prepareHFileQueue(org.apache.hadoop.fs.Path hfilesDir, Table table, Deque<LoadIncrementalHFiles.LoadQueueItem> queue, boolean validateHFile) Prepare a collection of LoadIncrementalHFiles.LoadQueueItem from list of source hfiles contained in the passed directory and validates whether the prepared queue has all the valid table column families in it.
void	prepareHFileQueue(org.apache.hadoop.fs.Path hfilesDir, Table table, Deque<LoadIncrementalHFiles.LoadQueueItem> queue, boolean validateHFile, boolean silence) Prepare a collection of LoadIncrementalHFiles.LoadQueueItem from list of source hfiles contained in the passed directory and validates whether the prepared queue has all the valid table column families in it.
Map<LoadIncrementalHFiles.LoadQueueItem,ByteBuffer>	run(Map<byte[],List<org.apache.hadoop.fs.Path>> family2Files, TableName tableName) Perform bulk load on the given table.
int	run(String[] args)
Map<LoadIncrementalHFiles.LoadQueueItem,ByteBuffer>	run(String hfofDir, TableName tableName) Perform bulk load on the given table.
void	setBulkToken(String stagingDir) Called from replication sink, where it manages bulkToken(staging directory) by itself.
protected List<LoadIncrementalHFiles.LoadQueueItem>	tryAtomicRegionLoad(ClientServiceCallable<byte[]> serviceCallable, TableName tableName, byte[] first, Collection<LoadIncrementalHFiles.LoadQueueItem> lqis) Attempts to do an atomic load of many hfiles into a region.

Methods inherited from class org.apache.hadoop.conf.Configured

getConf, setConf

Methods inherited from class java.lang.Object

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

Methods inherited from interface org.apache.hadoop.conf.Configurable

getConf, setConf

Field Detail

NAME

public static final String NAME

See Also:

Constant Field Values

MAX_FILES_PER_REGION_PER_FAMILY

public static final String MAX_FILES_PER_REGION_PER_FAMILY

See Also:

Constant Field Values

CREATE_TABLE_CONF_KEY

public static final String CREATE_TABLE_CONF_KEY

See Also:

Constant Field Values

IGNORE_UNMATCHED_CF_CONF_KEY

public static final String IGNORE_UNMATCHED_CF_CONF_KEY

See Also:

Constant Field Values

ALWAYS_COPY_FILES

public static final String ALWAYS_COPY_FILES

See Also:

Constant Field Values

Constructor Detail

LoadIncrementalHFiles

public LoadIncrementalHFiles(org.apache.hadoop.conf.Configuration conf)

Method Detail

prepareHFileQueue

public void prepareHFileQueue(org.apache.hadoop.fs.Path hfilesDir,
                              Table table,
                              Deque<LoadIncrementalHFiles.LoadQueueItem> queue,
                              boolean validateHFile)
                       throws IOException

Prepare a collection of LoadIncrementalHFiles.LoadQueueItem from list of source hfiles contained in the passed directory and validates whether the prepared queue has all the valid table column families in it.

Parameters:

hfilesDir - directory containing list of hfiles to be loaded into the table

table - table to which hfiles should be loaded

queue - queue which needs to be loaded into the table

validateHFile - if true hfiles will be validated for its format

Throws:

IOException - If any I/O or network error occurred

prepareHFileQueue

public void prepareHFileQueue(org.apache.hadoop.fs.Path hfilesDir,
                              Table table,
                              Deque<LoadIncrementalHFiles.LoadQueueItem> queue,
                              boolean validateHFile,
                              boolean silence)
                       throws IOException

Prepare a collection of LoadIncrementalHFiles.LoadQueueItem from list of source hfiles contained in the passed directory and validates whether the prepared queue has all the valid table column families in it.

Parameters:

hfilesDir - directory containing list of hfiles to be loaded into the table

table - table to which hfiles should be loaded

queue - queue which needs to be loaded into the table

validateHFile - if true hfiles will be validated for its format

silence - true to ignore unmatched column families

Throws:

IOException - If any I/O or network error occurred

prepareHFileQueue

public void prepareHFileQueue(Map<byte[],List<org.apache.hadoop.fs.Path>> map,
                              Table table,
                              Deque<LoadIncrementalHFiles.LoadQueueItem> queue,
                              boolean silence)
                       throws IOException

Prepare a collection of LoadIncrementalHFiles.LoadQueueItem from list of source hfiles contained in the passed directory and validates whether the prepared queue has all the valid table column families in it.

Parameters:

map - map of family to List of hfiles

table - table to which hfiles should be loaded

queue - queue which needs to be loaded into the table

silence - true to ignore unmatched column families

Throws:

IOException - If any I/O or network error occurred

doBulkLoad

public Map<LoadIncrementalHFiles.LoadQueueItem,ByteBuffer> doBulkLoad(org.apache.hadoop.fs.Path hfofDir,
                                                                      Admin admin,
                                                                      Table table,
                                                                      RegionLocator regionLocator)
                                                               throws TableNotFoundException,
                                                                      IOException

Perform a bulk load of the given directory into the given pre-existing table. This method is not threadsafe.

Parameters:

hfofDir - the directory that was provided as the output path of a job using HFileOutputFormat

admin - the Admin

table - the table to load into

regionLocator - region locator

Throws:

TableNotFoundException - if table does not yet exist

IOException

doBulkLoad

public Map<LoadIncrementalHFiles.LoadQueueItem,ByteBuffer> doBulkLoad(Map<byte[],List<org.apache.hadoop.fs.Path>> map,
                                                                      Admin admin,
                                                                      Table table,
                                                                      RegionLocator regionLocator,
                                                                      boolean silence,
                                                                      boolean copyFile)
                                                               throws TableNotFoundException,
                                                                      IOException

Perform a bulk load of the given directory into the given pre-existing table. This method is not threadsafe.

Parameters:

map - map of family to List of hfiles

admin - the Admin

table - the table to load into

regionLocator - region locator

silence - true to ignore unmatched column families

copyFile - always copy hfiles if true

Throws:

TableNotFoundException - if table does not yet exist

IOException

doBulkLoad

public Map<LoadIncrementalHFiles.LoadQueueItem,ByteBuffer> doBulkLoad(org.apache.hadoop.fs.Path hfofDir,
                                                                      Admin admin,
                                                                      Table table,
                                                                      RegionLocator regionLocator,
                                                                      boolean silence,
                                                                      boolean copyFile)
                                                               throws TableNotFoundException,
                                                                      IOException

Perform a bulk load of the given directory into the given pre-existing table. This method is not threadsafe.

Parameters:

hfofDir - the directory that was provided as the output path of a job using HFileOutputFormat

admin - the Admin

table - the table to load into

regionLocator - region locator

silence - true to ignore unmatched column families

copyFile - always copy hfiles if true

Throws:

TableNotFoundException - if table does not yet exist

IOException

loadHFileQueue

public void loadHFileQueue(Table table,
                           Connection conn,
                           Deque<LoadIncrementalHFiles.LoadQueueItem> queue,
                           Pair<byte[][],byte[][]> startEndKeys)
                    throws IOException

Used by the replication sink to load the hfiles from the source cluster. It does the following,

1. LoadIncrementalHFiles#groupOrSplitPhase(Table, ExecutorService, Deque, Pair)}
2. LoadIncrementalHFiles#bulkLoadPhase(Table, Connection, ExecutorService, Deque, Multimap)

Parameters:

table - Table to which these hfiles should be loaded to

conn - Connection to use

queue - LoadIncrementalHFiles.LoadQueueItem has hfiles yet to be loaded

startEndKeys - starting and ending row keys of the region

Throws:

IOException

loadHFileQueue

public void loadHFileQueue(Table table,
                           Connection conn,
                           Deque<LoadIncrementalHFiles.LoadQueueItem> queue,
                           Pair<byte[][],byte[][]> startEndKeys,
                           boolean copyFile)
                    throws IOException

Used by the replication sink to load the hfiles from the source cluster. It does the following,

1. LoadIncrementalHFiles#groupOrSplitPhase(Table, ExecutorService, Deque, Pair)}
2. LoadIncrementalHFiles#bulkLoadPhase(Table, Connection, ExecutorService, Deque, Multimap)

Parameters:

table - Table to which these hfiles should be loaded to

conn - Connection to use

queue - LoadIncrementalHFiles.LoadQueueItem has hfiles yet to be loaded

startEndKeys - starting and ending row keys of the region

Throws:

IOException

bulkLoadPhase

protected void bulkLoadPhase(Table table,
                             Connection conn,
                             ExecutorService pool,
                             Deque<LoadIncrementalHFiles.LoadQueueItem> queue,
                             org.apache.hbase.thirdparty.com.google.common.collect.Multimap<ByteBuffer,LoadIncrementalHFiles.LoadQueueItem> regionGroups,
                             boolean copyFile,
                             Map<LoadIncrementalHFiles.LoadQueueItem,ByteBuffer> item2RegionMap)
                      throws IOException

This takes the LQI's grouped by likely regions and attempts to bulk load them. Any failures are re-queued for another pass with the groupOrSplitPhase.

protected for testing.

Throws:

IOException

buildClientServiceCallable

protected ClientServiceCallable<byte[]> buildClientServiceCallable(Connection conn,
                                                                   TableName tableName,
                                                                   byte[] first,
                                                                   Collection<LoadIncrementalHFiles.LoadQueueItem> lqis,
                                                                   boolean copyFile)

groupOrSplit

protected Pair<List<LoadIncrementalHFiles.LoadQueueItem>,String> groupOrSplit(org.apache.hbase.thirdparty.com.google.common.collect.Multimap<ByteBuffer,LoadIncrementalHFiles.LoadQueueItem> regionGroups,
                                                                              LoadIncrementalHFiles.LoadQueueItem item,
                                                                              Table table,
                                                                              Pair<byte[][],byte[][]> startEndKeys)
                                                                       throws IOException

Attempt to assign the given load queue item into its target region group. If the hfile boundary no longer fits into a region, physically splits the hfile such that the new bottom half will fit and returns the list of LQI's corresponding to the resultant hfiles.

protected for testing

Throws:

IOException - if an IO failure is encountered

tryAtomicRegionLoad

protected List<LoadIncrementalHFiles.LoadQueueItem> tryAtomicRegionLoad(ClientServiceCallable<byte[]> serviceCallable,
                                                                        TableName tableName,
                                                                        byte[] first,
                                                                        Collection<LoadIncrementalHFiles.LoadQueueItem> lqis)
                                                                 throws IOException

Attempts to do an atomic load of many hfiles into a region. If it fails, it returns a list of hfiles that need to be retried. If it is successful it will return an empty list.

NOTE: To maintain row atomicity guarantees, region server callable should succeed atomically and fails atomically.

Protected for testing.

Returns:

empty list if success, list of items to retry on recoverable failure

Throws:

IOException

run

public Map<LoadIncrementalHFiles.LoadQueueItem,ByteBuffer> run(String hfofDir,
                                                               TableName tableName)
                                                        throws IOException

Perform bulk load on the given table.

Parameters:

hfofDir - the directory that was provided as the output path of a job using HFileOutputFormat

tableName - the table to load into

Throws:

IOException

run

public Map<LoadIncrementalHFiles.LoadQueueItem,ByteBuffer> run(Map<byte[],List<org.apache.hadoop.fs.Path>> family2Files,
                                                               TableName tableName)
                                                        throws IOException

Perform bulk load on the given table.

Parameters:

family2Files - map of family to List of hfiles

tableName - the table to load into

Throws:

IOException

run

public int run(String[] args)
        throws Exception

Specified by:

run in interface org.apache.hadoop.util.Tool

Throws:

Exception

main

public static void main(String[] args)
                 throws Exception

Throws:

Exception

setBulkToken

public void setBulkToken(String stagingDir)

Called from replication sink, where it manages bulkToken(staging directory) by itself. This is used only when SecureBulkLoadEndpoint is configured in hbase.coprocessor.region.classes property. This directory is used as a temporary directory where all files are initially copied/moved from user given directory, set all the required file permissions and then from their it is finally loaded into a table. This should be set only when, one would like to manage the staging directory by itself. Otherwise this tool will handle this by itself.

Parameters:

stagingDir - staging directory path

inferBoundaries

public static byte[][] inferBoundaries(SortedMap<byte[],Integer> bdryMap)

Infers region boundaries for a new table.

Parameter:
bdryMap is a map between keys to an integer belonging to {+1, -1}

• If a key is a start key of a file, then it maps to +1
• If a key is an end key of a file, then it maps to -1

Algo:

1. Poll on the keys in order:
   1. Keep adding the mapped values to these keys (runningSum)
   2. Each time runningSum reaches 0, add the start Key from when the runningSum had started to a boundary list.
2. Return the boundary list.