Package org.apache.hadoop.hbase.io.hfile

Class HFileWriterImpl

java.lang.Object

org.apache.hadoop.hbase.io.hfile.HFileWriterImpl

All Implemented Interfaces:

Closeable, AutoCloseable, HFile.Writer, CellSink, ShipperListener

@Private
public class HFileWriterImpl
extends Object
implements HFile.Writer

Common functionality needed by all versions of HFile writers.

Field Summary

Fields

Modifier and Type	Field	Description
private List<HFileBlock.BlockWritable>	additionalLoadOnOpenData	Additional data items to be written to the "load-on-open" section.
protected final HFileDataBlockEncoder	blockEncoder	The data block encoding which will be used.
protected HFileBlock.Writer	blockWriter	block writer
protected final CacheConfig	cacheConf	Cache configuration for caching data on write.
protected final boolean	closeOutputStream	True if we opened the outputStream (and so will close it).
private HFileBlockIndex.BlockIndexWriter	dataBlockIndexWriter
private final int	encodedBlockSizeLimit	Block size limit after encoding, used to unify encoded block Cache entry size
protected long	entryCount	Total # of key/value entries, i.e.
protected HFileInfo	fileInfo	A "file info" block: a key-value map of file-wide metadata.
protected Cell	firstCellInBlock	First cell in a block.
private long	firstDataBlockOffset	The offset of the first data block or -1 if the file is empty.
protected final HFileContext	hFileContext
protected final HFileIndexBlockEncoder	indexBlockEncoder
private List<InlineBlockWriter>	inlineBlockWriters	Inline block writers for multi-level block index and compound Blooms.
static final int	KEY_VALUE_VER_WITH_MEMSTORE	Version for KeyValue which includes memstore timestamp
static final byte[]	KEY_VALUE_VERSION	KeyValue version in FileInfo
protected byte[]	keyOfBiggestCell	Key of the biggest cell.
protected Cell	lastCell	The Cell previously appended.
private Cell	lastCellOfPreviousBlock	The last(stop) Cell of the previous data block.
protected long	lastDataBlockOffset	The offset of the last data block or 0 if the file is empty.
protected long	lenOfBiggestCell	Len of the biggest cell.
private static final org.slf4j.Logger	LOG
protected long	maxMemstoreTS
private int	maxTagsLength
private HFileBlockIndex.BlockIndexWriter	metaBlockIndexWriter
protected List<org.apache.hadoop.io.Writable>	metaData	Writables representing meta block data.
protected List<byte[]>	metaNames	Meta block names.
protected final String	name	Name for this object used when logging or in toString.
protected org.apache.hadoop.fs.FSDataOutputStream	outputStream	FileSystem stream to write into.
protected final org.apache.hadoop.fs.Path	path	May be null if we were passed a stream.
protected long	totalKeyLength	Used for calculating the average key length.
protected long	totalUncompressedBytes	Total uncompressed bytes, maybe calculate a compression ratio later.
protected long	totalValueLength	Used for calculating the average value length.
static final String	UNIFIED_ENCODED_BLOCKSIZE_RATIO	if this feature is enabled, preCalculate encoded data size before real encoding happens
private static final long	UNSET

Fields inherited from interface org.apache.hadoop.hbase.io.hfile.HFile.Writer

MAX_MEMSTORE_TS_KEY

Constructor Summary

Constructors

Constructor	Description
HFileWriterImpl(org.apache.hadoop.conf.Configuration conf, CacheConfig cacheConf, org.apache.hadoop.fs.Path path, org.apache.hadoop.fs.FSDataOutputStream outputStream, HFileContext fileContext)

Method Summary

Modifier and Type	Method	Description
private void	addBloomFilter(BloomFilterWriter bfw, BlockType blockType)
void	addDeleteFamilyBloomFilter(BloomFilterWriter bfw)	Store delete family Bloom filter in the file, which is only supported in HFile V2.
void	addGeneralBloomFilter(BloomFilterWriter bfw)	Store general Bloom filter in the file.
void	addInlineBlockWriter(InlineBlockWriter ibw)	Adds an inline block writer such as a multi-level block index writer or a compound Bloom filter writer.
void	append(Cell cell)	Add key/value to file.
void	appendFileInfo(byte[] k, byte[] v)	Add to the file info.
void	appendMetaBlock(String metaBlockName, org.apache.hadoop.io.Writable content)	Add a meta block to the end of the file.
void	beforeShipped()	The action that needs to be performed before Shipper.shipped() is performed
protected void	checkBlockBoundary()	At a block boundary, write all the inline blocks and opens new block.
protected boolean	checkKey(Cell cell)	Checks that the given Cell's key does not violate the key order.
protected void	checkValue(byte[] value, int offset, int length)	Checks the given value for validity.
void	close()
static Compression.Algorithm	compressionByName(String algoName)
protected static org.apache.hadoop.fs.FSDataOutputStream	createOutputStream(org.apache.hadoop.conf.Configuration conf, org.apache.hadoop.fs.FileSystem fs, org.apache.hadoop.fs.Path path, InetSocketAddress[] favoredNodes)	A helper method to create HFile output streams in constructors
private void	doCacheOnWrite(long offset)	Caches the last written HFile block.
private void	finishBlock()	Clean up the data block that is currently being written.
protected void	finishClose(FixedFileTrailer trailer)
protected void	finishFileInfo()
protected void	finishInit(org.apache.hadoop.conf.Configuration conf)	Additional initialization steps
HFileContext	getFileContext()	Return the file context for the HFile this writer belongs to
Cell	getLastCell()
private String	getLexicalErrorMessage(Cell cell)
protected int	getMajorVersion()
static Cell	getMidpoint(CellComparator comparator, Cell left, Cell right)	Try to return a Cell that falls between left and right but that is shorter; i.e.
private static byte[]	getMinimumMidpointArray(byte[] leftArray, int leftOffset, int leftLength, byte[] rightArray, int rightOffset, int rightLength)	Try to get a byte array that falls between left and right as short as possible with lexicographical order;
private static byte[]	getMinimumMidpointArray(ByteBuffer left, int leftOffset, int leftLength, ByteBuffer right, int rightOffset, int rightLength)	Try to create a new byte array that falls between left and right as short as possible with lexicographical order.
protected int	getMinorVersion()
org.apache.hadoop.fs.Path	getPath()	Returns Path or null if we were passed a stream rather than a Path.
long	getPos()
protected void	newBlock()	Ready a new block for writing.
String	toString()
protected final void	writeFileInfo(FixedFileTrailer trailer, DataOutputStream out)	Sets the file info offset in the trailer, finishes up populating fields in the file info, and writes the file info into the given data output.
private void	writeInlineBlocks(boolean closing)	Gives inline block writers an opportunity to contribute blocks.

Methods inherited from class java.lang.Object

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

Field Details

LOG

private static final org.slf4j.Logger LOG

UNSET

private static final long UNSET

See Also:

• Constant Field Values

UNIFIED_ENCODED_BLOCKSIZE_RATIO

public static final String UNIFIED_ENCODED_BLOCKSIZE_RATIO

if this feature is enabled, preCalculate encoded data size before real encoding happens

See Also:

• Constant Field Values

encodedBlockSizeLimit

private final int encodedBlockSizeLimit

Block size limit after encoding, used to unify encoded block Cache entry size

lastCell

protected Cell lastCell

The Cell previously appended. Becomes the last cell in the file.

outputStream

protected org.apache.hadoop.fs.FSDataOutputStream outputStream

FileSystem stream to write into.

closeOutputStream

protected final boolean closeOutputStream

True if we opened the outputStream (and so will close it).

fileInfo

protected HFileInfo fileInfo

A "file info" block: a key-value map of file-wide metadata.

entryCount

protected long entryCount

Total # of key/value entries, i.e. how many times add() was called.

totalKeyLength

protected long totalKeyLength

Used for calculating the average key length.

totalValueLength

protected long totalValueLength

Used for calculating the average value length.

lenOfBiggestCell

protected long lenOfBiggestCell

Len of the biggest cell.

keyOfBiggestCell

protected byte[] keyOfBiggestCell

Key of the biggest cell.

totalUncompressedBytes

protected long totalUncompressedBytes

Total uncompressed bytes, maybe calculate a compression ratio later.

metaNames

protected List<byte[]> metaNames

Meta block names.

metaData

protected List<org.apache.hadoop.io.Writable> metaData

Writables representing meta block data.

firstCellInBlock

protected Cell firstCellInBlock

First cell in a block. This reference should be short-lived since we write hfiles in a burst.

path

protected final org.apache.hadoop.fs.Path path

May be null if we were passed a stream.

cacheConf

protected final CacheConfig cacheConf

Cache configuration for caching data on write.

name

protected final String name

Name for this object used when logging or in toString. Is either the result of a toString on stream or else name of passed file Path.

blockEncoder

protected final HFileDataBlockEncoder blockEncoder

The data block encoding which will be used. NoOpDataBlockEncoder.INSTANCE if there is no encoding.

indexBlockEncoder

protected final HFileIndexBlockEncoder indexBlockEncoder

hFileContext

protected final HFileContext hFileContext

maxTagsLength

private int maxTagsLength

KEY_VALUE_VERSION

public static final byte[] KEY_VALUE_VERSION

KeyValue version in FileInfo

KEY_VALUE_VER_WITH_MEMSTORE

public static final int KEY_VALUE_VER_WITH_MEMSTORE

Version for KeyValue which includes memstore timestamp

See Also:

• Constant Field Values

inlineBlockWriters

private List<InlineBlockWriter> inlineBlockWriters

Inline block writers for multi-level block index and compound Blooms.

blockWriter

protected HFileBlock.Writer blockWriter

block writer

dataBlockIndexWriter

private HFileBlockIndex.BlockIndexWriter dataBlockIndexWriter

metaBlockIndexWriter

private HFileBlockIndex.BlockIndexWriter metaBlockIndexWriter

firstDataBlockOffset

private long firstDataBlockOffset

The offset of the first data block or -1 if the file is empty.

lastDataBlockOffset

protected long lastDataBlockOffset

The offset of the last data block or 0 if the file is empty.

lastCellOfPreviousBlock

private Cell lastCellOfPreviousBlock

The last(stop) Cell of the previous data block. This reference should be short-lived since we write hfiles in a burst.

additionalLoadOnOpenData

private List<HFileBlock.BlockWritable> additionalLoadOnOpenData

Additional data items to be written to the "load-on-open" section.

maxMemstoreTS

protected long maxMemstoreTS

Constructor Details

HFileWriterImpl

public HFileWriterImpl(org.apache.hadoop.conf.Configuration conf,
 CacheConfig cacheConf,
 org.apache.hadoop.fs.Path path,
 org.apache.hadoop.fs.FSDataOutputStream outputStream,
 HFileContext fileContext)

Method Details

appendFileInfo

public void appendFileInfo(byte[] k,
 byte[] v)
                    throws IOException

Add to the file info. All added key/value pairs can be obtained using HFile.Reader.getHFileInfo().

Specified by:

appendFileInfo in interface HFile.Writer

Parameters:

k - Key

v - Value

Throws:

IOException - in case the key or the value are invalid

writeFileInfo

protected final void writeFileInfo(FixedFileTrailer trailer,
 DataOutputStream out)
                            throws IOException

Sets the file info offset in the trailer, finishes up populating fields in the file info, and writes the file info into the given data output. The reason the data output is not always outputStream is that we store file info as a block in version 2.

Parameters:

trailer - fixed file trailer

out - the data output to write the file info to

Throws:

IOException

getPos

public long getPos()
            throws IOException

Throws:

IOException

checkKey

protected boolean checkKey(Cell cell)
                    throws IOException

Checks that the given Cell's key does not violate the key order.

Parameters:

cell - Cell whose key to check.

Returns:

true if the key is duplicate

Throws:

IOException - if the key or the key order is wrong

getLexicalErrorMessage

private String getLexicalErrorMessage(Cell cell)

checkValue

protected void checkValue(byte[] value,
 int offset,
 int length)
                   throws IOException

Checks the given value for validity.

Throws:

IOException

getPath

public org.apache.hadoop.fs.Path getPath()

Returns Path or null if we were passed a stream rather than a Path.

Specified by:

getPath in interface HFile.Writer

toString

public String toString()

Overrides:

toString in class Object

compressionByName

public static Compression.Algorithm compressionByName(String algoName)

createOutputStream

protected static org.apache.hadoop.fs.FSDataOutputStream createOutputStream(org.apache.hadoop.conf.Configuration conf,
 org.apache.hadoop.fs.FileSystem fs,
 org.apache.hadoop.fs.Path path,
 InetSocketAddress[] favoredNodes)
                                                                     throws IOException

A helper method to create HFile output streams in constructors

Throws:

IOException

finishInit

protected void finishInit(org.apache.hadoop.conf.Configuration conf)

Additional initialization steps

checkBlockBoundary

protected void checkBlockBoundary()
                           throws IOException

At a block boundary, write all the inline blocks and opens new block.

Throws:

IOException

finishBlock

private void finishBlock()
                  throws IOException

Clean up the data block that is currently being written.

Throws:

IOException

getMidpoint

public static Cell getMidpoint(CellComparator comparator,
 Cell left,
 Cell right)

Try to return a Cell that falls between left and right but that is shorter; i.e. takes up less space. This trick is used building HFile block index. Its an optimization. It does not always work. In this case we'll just return the right cell.

Returns:

A cell that sorts between left and right.

getMinimumMidpointArray

private static byte[] getMinimumMidpointArray(byte[] leftArray,
 int leftOffset,
 int leftLength,
 byte[] rightArray,
 int rightOffset,
 int rightLength)

Try to get a byte array that falls between left and right as short as possible with lexicographical order;

Returns:

Return a new array that is between left and right and minimally sized else just return null if left == right.

getMinimumMidpointArray

private static byte[] getMinimumMidpointArray(ByteBuffer left,
 int leftOffset,
 int leftLength,
 ByteBuffer right,
 int rightOffset,
 int rightLength)

Try to create a new byte array that falls between left and right as short as possible with lexicographical order.

Returns:

Return a new array that is between left and right and minimally sized else just return null if left == right.

writeInlineBlocks

private void writeInlineBlocks(boolean closing)
                        throws IOException

Gives inline block writers an opportunity to contribute blocks.

Throws:

IOException

doCacheOnWrite

private void doCacheOnWrite(long offset)

Caches the last written HFile block.

Parameters:

offset - the offset of the block we want to cache. Used to determine the cache key.

newBlock

protected void newBlock()
                 throws IOException

Ready a new block for writing.

Throws:

IOException

appendMetaBlock

public void appendMetaBlock(String metaBlockName,
 org.apache.hadoop.io.Writable content)

Add a meta block to the end of the file. Call before close(). Metadata blocks are expensive. Fill one with a bunch of serialized data rather than do a metadata block per metadata instance. If metadata is small, consider adding to file info using appendFileInfo(byte[], byte[]) name of the block will call readFields to get data later (DO NOT REUSE)

Specified by:

appendMetaBlock in interface HFile.Writer

close

public void close()
           throws IOException

Specified by:

close in interface AutoCloseable

Specified by:

close in interface Closeable

Throws:

IOException

addInlineBlockWriter

public void addInlineBlockWriter(InlineBlockWriter ibw)

Description copied from interface: HFile.Writer

Adds an inline block writer such as a multi-level block index writer or a compound Bloom filter writer.

Specified by:

addInlineBlockWriter in interface HFile.Writer

addGeneralBloomFilter

public void addGeneralBloomFilter(BloomFilterWriter bfw)

Description copied from interface: HFile.Writer

Store general Bloom filter in the file. This does not deal with Bloom filter internals but is necessary, since Bloom filters are stored differently in HFile version 1 and version 2.

Specified by:

addGeneralBloomFilter in interface HFile.Writer

addDeleteFamilyBloomFilter

public void addDeleteFamilyBloomFilter(BloomFilterWriter bfw)

Description copied from interface: HFile.Writer

Store delete family Bloom filter in the file, which is only supported in HFile V2.

Specified by:

addDeleteFamilyBloomFilter in interface HFile.Writer

addBloomFilter

private void addBloomFilter(BloomFilterWriter bfw,
 BlockType blockType)

getFileContext

public HFileContext getFileContext()

Description copied from interface: HFile.Writer

Return the file context for the HFile this writer belongs to

Specified by:

getFileContext in interface HFile.Writer

append

public void append(Cell cell)
            throws IOException

Add key/value to file. Keys must be added in an order that agrees with the Comparator passed on construction. Cell to add. Cannot be empty nor null.

Specified by:

append in interface CellSink

Parameters:

cell - the cell to be added

Throws:

IOException

beforeShipped

public void beforeShipped()
                   throws IOException

Description copied from interface: ShipperListener

The action that needs to be performed before Shipper.shipped() is performed

Specified by:

beforeShipped in interface ShipperListener

Throws:

IOException

getLastCell

public Cell getLastCell()

finishFileInfo

protected void finishFileInfo()
                       throws IOException

Throws:

IOException

getMajorVersion

protected int getMajorVersion()

getMinorVersion

protected int getMinorVersion()

finishClose

protected void finishClose(FixedFileTrailer trailer)
                    throws IOException

Throws:

IOException