Package org.apache.hadoop.hbase.util

Class FSTableDescriptors

java.lang.Object

org.apache.hadoop.hbase.util.FSTableDescriptors

All Implemented Interfaces:

Closeable, AutoCloseable, TableDescriptors

@Private
public class FSTableDescriptors
extends Object
implements TableDescriptors

Implementation of TableDescriptors that reads descriptors from the passed filesystem. It expects descriptors to be in a file in the TABLEINFO_DIR subdir of the table's directory in FS. Can be read-only -- i.e. does not modify the filesystem or can be read and write.

Also has utility for keeping up the table descriptors tableinfo file. The table schema file is kept in the TABLEINFO_DIR subdir of the table directory in the filesystem. It has a TABLEINFO_FILE_PREFIX and then a suffix that is the edit sequenceid: e.g. .tableinfo.0000000003. This sequenceid is always increasing. It starts at zero. The table schema file with the highest sequenceid has the most recent schema edit. Usually there is one file only, the most recent but there may be short periods where there are more than one file. Old files are eventually cleaned. Presumption is that there will not be lots of concurrent clients making table schema edits. If so, the below needs a bit of a reworking and perhaps some supporting api in hdfs.

Nested Class Summary

Nested Classes

Modifier and Type	Class	Description
(package private) static final class	FSTableDescriptors.SequenceIdAndFileLength

Field Summary

Fields

Modifier and Type	Field	Description
private final Map<TableName,TableDescriptor>	cache
(package private) long	cachehits
private ThreadPoolExecutor	executor
private final org.apache.hadoop.fs.FileSystem	fs
private final boolean	fsreadonly
private boolean	fsvisited
(package private) long	invocations
private static final org.slf4j.Logger	LOG
private final org.apache.hadoop.fs.Path	rootdir
private boolean	tableDescriptorParallelLoadEnable
static final String	TABLEINFO_DIR
(package private) static final String	TABLEINFO_FILE_PREFIX	The file name prefix used to store HTD in HDFS
(package private) static final Comparator<org.apache.hadoop.fs.FileStatus>	TABLEINFO_FILESTATUS_COMPARATOR	Compare FileStatus instances by Path.getName().
private static final org.apache.hadoop.fs.PathFilter	TABLEINFO_PATHFILTER
private final boolean	usecache
(package private) static final int	WIDTH_OF_SEQUENCE_ID	Width of the sequenceid that is a suffix on a tableinfo file.

Constructor Summary

Constructors

Constructor	Description
FSTableDescriptors(org.apache.hadoop.conf.Configuration conf)	Construct a FSTableDescriptors instance using the hbase root dir of the given conf and the filesystem where that root dir lives.
FSTableDescriptors(org.apache.hadoop.fs.FileSystem fs, org.apache.hadoop.fs.Path rootdir)
FSTableDescriptors(org.apache.hadoop.fs.FileSystem fs, org.apache.hadoop.fs.Path rootdir, boolean fsreadonly, boolean usecache)
FSTableDescriptors(org.apache.hadoop.fs.FileSystem fs, org.apache.hadoop.fs.Path rootdir, boolean fsreadonly, boolean usecache, int tableDescriptorParallelLoadThreads)

Method Summary

Modifier and Type	Method	Description
void	close()
private static TableDescriptorBuilder	createMetaTableDescriptorBuilder(org.apache.hadoop.conf.Configuration conf)
boolean	createTableDescriptor(TableDescriptor htd)	Create new TableDescriptor in HDFS.
boolean	createTableDescriptor(TableDescriptor htd, boolean forceCreation)	Create new TableDescriptor in HDFS.
static boolean	createTableDescriptorForTableDirectory(org.apache.hadoop.fs.FileSystem fs, org.apache.hadoop.fs.Path tableDir, TableDescriptor htd, boolean forceCreation)	Create a new TableDescriptor in HDFS in the specified table directory.
boolean	createTableDescriptorForTableDirectory(org.apache.hadoop.fs.Path tableDir, TableDescriptor htd, boolean forceCreation)	Create a new TableDescriptor in HDFS in the specified table directory.
private static void	deleteMalformedFile(org.apache.hadoop.fs.FileSystem fs, org.apache.hadoop.fs.Path file)
private static void	deleteTableDescriptorFiles(org.apache.hadoop.fs.FileSystem fs, org.apache.hadoop.fs.Path dir, int maxSequenceId)	Deletes files matching the table info file pattern within the given directory whose sequenceId is at most the given max sequenceId.
static void	deleteTableDescriptors(org.apache.hadoop.fs.FileSystem fs, org.apache.hadoop.fs.Path tableDir)
private static String	formatTableInfoSequenceId(int number)
TableDescriptor	get(TableName tableName)	Get the current table descriptor for the given table, or null if none exists.
Map<String,TableDescriptor>	getAll()	Returns a map from table name to table descriptor for all tables.
Map<String,TableDescriptor>	getByNamespace(String name)	Find descriptors by namespace.
static ColumnFamilyDescriptor	getNamespaceFamilyDescForMeta(org.apache.hadoop.conf.Configuration conf)
static ColumnFamilyDescriptor	getReplBarrierFamilyDescForMeta()
static TableDescriptor	getTableDescriptorFromFs(org.apache.hadoop.fs.FileSystem fs, org.apache.hadoop.fs.Path tableDir)	Returns the latest table descriptor for the table located at the given directory directly from the file system if it exists.
private static Optional<Pair<org.apache.hadoop.fs.FileStatus,TableDescriptor>>	getTableDescriptorFromFs(org.apache.hadoop.fs.FileSystem fs, org.apache.hadoop.fs.Path tableDir, boolean readonly)
static TableDescriptor	getTableDescriptorFromFs(org.apache.hadoop.fs.FileSystem fs, org.apache.hadoop.fs.Path hbaseRootDir, TableName tableName)	Returns the latest table descriptor for the given table directly from the file system if it exists, bypassing the local cache.
private org.apache.hadoop.fs.Path	getTableDir(TableName tableName)	Return the table directory in HDFS
static ColumnFamilyDescriptor	getTableFamilyDescForMeta(org.apache.hadoop.conf.Configuration conf)
(package private) static String	getTableInfoFileName(int sequenceId, byte[] content)	Returns Name of tableinfo file.
(package private) static FSTableDescriptors.SequenceIdAndFileLength	getTableInfoSequenceIdAndFileLength(org.apache.hadoop.fs.Path p)	Returns the current sequence id and file length or 0 if none found.
private void	internalGet(org.apache.hadoop.fs.Path dir, Map<String,TableDescriptor> tds, AtomicBoolean allvisited)
static boolean	isTableDir(org.apache.hadoop.fs.FileSystem fs, org.apache.hadoop.fs.Path tableDir)	Check whether we have a valid TableDescriptor.
protected boolean	isUsecache()
TableDescriptor	remove(TableName tablename)	Removes the table descriptor from the local cache and returns it.
static TableDescriptor	tryUpdateAndGetMetaTableDescriptor(org.apache.hadoop.conf.Configuration conf, org.apache.hadoop.fs.FileSystem fs, org.apache.hadoop.fs.Path rootdir)
static void	tryUpdateMetaTableDescriptor(org.apache.hadoop.conf.Configuration conf)
void	update(TableDescriptor td, boolean cacheOnly)	Add or update descriptor
(package private) org.apache.hadoop.fs.Path	updateTableDescriptor(TableDescriptor td)
private static org.apache.hadoop.fs.Path	writeTableDescriptor(org.apache.hadoop.fs.FileSystem fs, TableDescriptor td, org.apache.hadoop.fs.Path tableDir, org.apache.hadoop.fs.FileStatus currentDescriptorFile)	Attempts to write a new table descriptor to the given table's directory.

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.hbase.TableDescriptors

exists, update

Field Details

LOG

private static final org.slf4j.Logger LOG

fs

private final org.apache.hadoop.fs.FileSystem fs

rootdir

private final org.apache.hadoop.fs.Path rootdir

fsreadonly

private final boolean fsreadonly

usecache

private final boolean usecache

fsvisited

private volatile boolean fsvisited

tableDescriptorParallelLoadEnable

private boolean tableDescriptorParallelLoadEnable

executor

private ThreadPoolExecutor executor

cachehits

long cachehits

invocations

long invocations

TABLEINFO_FILE_PREFIX

static final String TABLEINFO_FILE_PREFIX

The file name prefix used to store HTD in HDFS

See Also:

• Constant Field Values

TABLEINFO_DIR

public static final String TABLEINFO_DIR

See Also:

• Constant Field Values

cache

private final Map<TableName,TableDescriptor> cache

TABLEINFO_FILESTATUS_COMPARATOR

static final Comparator<org.apache.hadoop.fs.FileStatus> TABLEINFO_FILESTATUS_COMPARATOR

Compare FileStatus instances by Path.getName(). Returns in reverse order.

TABLEINFO_PATHFILTER

private static final org.apache.hadoop.fs.PathFilter TABLEINFO_PATHFILTER

WIDTH_OF_SEQUENCE_ID

static final int WIDTH_OF_SEQUENCE_ID

Width of the sequenceid that is a suffix on a tableinfo file.

See Also:

• Constant Field Values

Constructor Details

FSTableDescriptors

public FSTableDescriptors(org.apache.hadoop.conf.Configuration conf)
                   throws IOException

Construct a FSTableDescriptors instance using the hbase root dir of the given conf and the filesystem where that root dir lives. This instance can do write operations (is not read only).

Throws:

IOException

FSTableDescriptors

public FSTableDescriptors(org.apache.hadoop.fs.FileSystem fs,
 org.apache.hadoop.fs.Path rootdir)

FSTableDescriptors

public FSTableDescriptors(org.apache.hadoop.fs.FileSystem fs,
 org.apache.hadoop.fs.Path rootdir,
 boolean fsreadonly,
 boolean usecache)

FSTableDescriptors

public FSTableDescriptors(org.apache.hadoop.fs.FileSystem fs,
 org.apache.hadoop.fs.Path rootdir,
 boolean fsreadonly,
 boolean usecache,
 int tableDescriptorParallelLoadThreads)

Method Details

tryUpdateMetaTableDescriptor

public static void tryUpdateMetaTableDescriptor(org.apache.hadoop.conf.Configuration conf)
                                         throws IOException

Throws:

IOException

tryUpdateAndGetMetaTableDescriptor

public static TableDescriptor tryUpdateAndGetMetaTableDescriptor(org.apache.hadoop.conf.Configuration conf,
 org.apache.hadoop.fs.FileSystem fs,
 org.apache.hadoop.fs.Path rootdir)
                                                          throws IOException

Throws:

IOException

getTableFamilyDescForMeta

public static ColumnFamilyDescriptor getTableFamilyDescForMeta(org.apache.hadoop.conf.Configuration conf)

getReplBarrierFamilyDescForMeta

public static ColumnFamilyDescriptor getReplBarrierFamilyDescForMeta()

getNamespaceFamilyDescForMeta

public static ColumnFamilyDescriptor getNamespaceFamilyDescForMeta(org.apache.hadoop.conf.Configuration conf)

createMetaTableDescriptorBuilder

private static TableDescriptorBuilder createMetaTableDescriptorBuilder(org.apache.hadoop.conf.Configuration conf)
                                                                throws IOException

Throws:

IOException

isUsecache

protected boolean isUsecache()

get

@Nullable
public TableDescriptor get(TableName tableName)

Get the current table descriptor for the given table, or null if none exists.

Uses a local cache of the descriptor but still checks the filesystem on each call if fsvisited is not true, i.e, we haven't done a full scan yet, to see if a newer file has been created since the cached one was read.

Specified by:

get in interface TableDescriptors

getAll

public Map<String,TableDescriptor> getAll()
                                   throws IOException

Returns a map from table name to table descriptor for all tables.

Specified by:

getAll in interface TableDescriptors

Returns:

Map of all descriptors.

Throws:

IOException

internalGet

private void internalGet(org.apache.hadoop.fs.Path dir,
 Map<String,TableDescriptor> tds,
 AtomicBoolean allvisited)

getByNamespace

public Map<String,TableDescriptor> getByNamespace(String name)
                                           throws IOException

Find descriptors by namespace.

Specified by:

getByNamespace in interface TableDescriptors

Returns:

Map of all descriptors.

Throws:

IOException

See Also:

• get(org.apache.hadoop.hbase.TableName)

update

public void update(TableDescriptor td,
 boolean cacheOnly)
            throws IOException

Description copied from interface: TableDescriptors

Add or update descriptor

Specified by:

update in interface TableDescriptors

Parameters:

td - Descriptor to set into TableDescriptors

cacheOnly - only add the given htd to cache, without updating the storage. For example, when creating table, we will write the descriptor to fs when creating the fs layout, so we do not need to update the fs again.

Throws:

IOException

updateTableDescriptor

org.apache.hadoop.fs.Path updateTableDescriptor(TableDescriptor td)
                                         throws IOException

Throws:

IOException

remove

public TableDescriptor remove(TableName tablename)
                       throws IOException

Removes the table descriptor from the local cache and returns it. If not in read only mode, it also deletes the entire table directory(!) from the FileSystem.

Specified by:

remove in interface TableDescriptors

Throws:

IOException

isTableDir

public static boolean isTableDir(org.apache.hadoop.fs.FileSystem fs,
 org.apache.hadoop.fs.Path tableDir)
                          throws IOException

Check whether we have a valid TableDescriptor.

Throws:

IOException

getTableDir

private org.apache.hadoop.fs.Path getTableDir(TableName tableName)

Return the table directory in HDFS

formatTableInfoSequenceId

private static String formatTableInfoSequenceId(int number)

Parameters:

number - Number to use as suffix.

Returns:

Returns zero-prefixed decimal version of passed number (Does absolute in case number is negative).

close

public void close()
           throws IOException

Specified by:

close in interface AutoCloseable

Specified by:

close in interface Closeable

Specified by:

close in interface TableDescriptors

Throws:

IOException

getTableInfoSequenceIdAndFileLength

static FSTableDescriptors.SequenceIdAndFileLength getTableInfoSequenceIdAndFileLength(org.apache.hadoop.fs.Path p)

Returns the current sequence id and file length or 0 if none found.

Parameters:

p - Path to a .tableinfo file.

getTableInfoFileName

static String getTableInfoFileName(int sequenceId,
 byte[] content)

Returns Name of tableinfo file.

getTableDescriptorFromFs

public static TableDescriptor getTableDescriptorFromFs(org.apache.hadoop.fs.FileSystem fs,
 org.apache.hadoop.fs.Path hbaseRootDir,
 TableName tableName)
                                                throws IOException

Returns the latest table descriptor for the given table directly from the file system if it exists, bypassing the local cache. Returns null if it's not found.

Throws:

IOException

getTableDescriptorFromFs

public static TableDescriptor getTableDescriptorFromFs(org.apache.hadoop.fs.FileSystem fs,
 org.apache.hadoop.fs.Path tableDir)
                                                throws IOException

Returns the latest table descriptor for the table located at the given directory directly from the file system if it exists.

Throws:

IOException

deleteMalformedFile

private static void deleteMalformedFile(org.apache.hadoop.fs.FileSystem fs,
 org.apache.hadoop.fs.Path file)
                                 throws IOException

Throws:

IOException

getTableDescriptorFromFs

private static Optional<Pair<org.apache.hadoop.fs.FileStatus,TableDescriptor>> getTableDescriptorFromFs(org.apache.hadoop.fs.FileSystem fs,
 org.apache.hadoop.fs.Path tableDir,
 boolean readonly)
                                                                                                 throws IOException

Throws:

IOException

deleteTableDescriptors

public static void deleteTableDescriptors(org.apache.hadoop.fs.FileSystem fs,
 org.apache.hadoop.fs.Path tableDir)
                                   throws IOException

Throws:

IOException

deleteTableDescriptorFiles

private static void deleteTableDescriptorFiles(org.apache.hadoop.fs.FileSystem fs,
 org.apache.hadoop.fs.Path dir,
 int maxSequenceId)
                                        throws IOException

Deletes files matching the table info file pattern within the given directory whose sequenceId is at most the given max sequenceId.

Throws:

IOException

writeTableDescriptor

private static org.apache.hadoop.fs.Path writeTableDescriptor(org.apache.hadoop.fs.FileSystem fs,
 TableDescriptor td,
 org.apache.hadoop.fs.Path tableDir,
 org.apache.hadoop.fs.FileStatus currentDescriptorFile)
                                                       throws IOException

Attempts to write a new table descriptor to the given table's directory. It begins at the currentSequenceId + 1 and tries 10 times to find a new sequence number not already in use.

Removes the current descriptor file if passed in.

Returns:

Descriptor file or null if we failed write.

Throws:

IOException

createTableDescriptor

public boolean createTableDescriptor(TableDescriptor htd)
                              throws IOException

Create new TableDescriptor in HDFS. Happens when we are creating table. Used by tests.

Returns:

True if we successfully created file.

Throws:

IOException

createTableDescriptor

public boolean createTableDescriptor(TableDescriptor htd,
 boolean forceCreation)
                              throws IOException

Create new TableDescriptor in HDFS. Happens when we are creating table. If forceCreation is true then even if previous table descriptor is present it will be overwritten

Returns:

True if we successfully created file.

Throws:

IOException

createTableDescriptorForTableDirectory

public boolean createTableDescriptorForTableDirectory(org.apache.hadoop.fs.Path tableDir,
 TableDescriptor htd,
 boolean forceCreation)
                                               throws IOException

Create a new TableDescriptor in HDFS in the specified table directory. Happens when we create a new table during cluster start or in Clone and Create Table Procedures. Checks readOnly flag passed on construction.

Parameters:

tableDir - table directory under which we should write the file

htd - description of the table to write

forceCreation - if true,then even if previous table descriptor is present it will be overwritten

Returns:

true if the we successfully created the file, false if the file already exists and we weren't forcing the descriptor creation.

Throws:

IOException - if a filesystem error occurs

createTableDescriptorForTableDirectory

public static boolean createTableDescriptorForTableDirectory(org.apache.hadoop.fs.FileSystem fs,
 org.apache.hadoop.fs.Path tableDir,
 TableDescriptor htd,
 boolean forceCreation)
                                                      throws IOException

Create a new TableDescriptor in HDFS in the specified table directory. Happens when we create a new table snapshoting. Does not enforce read-only. That is for caller to determine.

Parameters:

fs - Filesystem to use.

tableDir - table directory under which we should write the file

htd - description of the table to write

forceCreation - if true,then even if previous table descriptor is present it will be overwritten

Returns:

true if the we successfully created the file, false if the file already exists and we weren't forcing the descriptor creation.

Throws:

IOException - if a filesystem error occurs