org.apache.hadoop.hive.ql.io

Class RCFile

java.lang.Object

org.apache.hadoop.hive.ql.io.RCFile

public class RCFile
extends Object

RCFiles, short of Record Columnar File, are flat files consisting of binary key/value pairs, which shares much similarity with SequenceFile. RCFile stores columns of a table in a record columnar way. It first partitions rows horizontally into row splits. and then it vertically partitions each row split in a columnar way. RCFile first stores the meta data of a row split, as the key part of a record, and all the data of a row split as the value part. When writing, RCFile.Writer first holds records' value bytes in memory, and determines a row split if the raw bytes size of buffered records overflow a given parameterWriter.columnsBufferSize, which can be set like: conf.setInt(COLUMNS_BUFFER_SIZE_CONF_STR, 4 * 1024 * 1024) .

RCFile provides RCFile.Writer, RCFile.Reader and classes for writing, reading respectively.

RCFile stores columns of a table in a record columnar way. It first partitions rows horizontally into row splits. and then it vertically partitions each row split in a columnar way. RCFile first stores the meta data of a row split, as the key part of a record, and all the data of a row split as the value part.

RCFile compresses values in a more fine-grained manner then record level compression. However, It currently does not support compress the key part yet. The actual compression algorithm used to compress key and/or values can be specified by using the appropriate CompressionCodec.

The RCFile.Reader is used to read and explain the bytes of RCFile.

RCFile Formats

RC Header

• version - 3 bytes of magic header RCF, followed by 1 byte of actual version number (e.g. RCF1)
• compression - A boolean which specifies if compression is turned on for keys/values in this file.
• compression codec - CompressionCodec class which is used for compression of keys and/or values (if compression is enabled).
• metadata - SequenceFile.Metadata for this file.
• sync - A sync marker to denote end of the header.

RCFile Format

• Header
• Record
• Key part
  • Record length in bytes
  • Key length in bytes
  • Number_of_rows_in_this_record(vint)
  • Column_1_ondisk_length(vint)
  • Column_1_row_1_value_plain_length
  • Column_1_row_2_value_plain_length
  • ...
  • Column_2_ondisk_length(vint)
  • Column_2_row_1_value_plain_length
  • Column_2_row_2_value_plain_length
  • ...
• Value part
  • Compressed or plain data of [column_1_row_1_value, column_1_row_2_value,....]
  • Compressed or plain data of [column_2_row_1_value, column_2_row_2_value,....]

 The following is a pseudo-BNF grammar for RCFile. Comments are prefixed
 with dashes:

 rcfile ::=
   <file-header>
   <rcfile-rowgroup>+

 file-header ::=
   <file-version-header>
   <file-key-class-name>              (only exists if version is seq6)
   <file-value-class-name>            (only exists if version is seq6)
   <file-is-compressed>
   <file-is-block-compressed>         (only exists if version is seq6)
   [<file-compression-codec-class>]
   <file-header-metadata>
   <file-sync-field>

 -- The normative RCFile implementation included with Hive is actually
 -- based on a modified version of Hadoop's SequenceFile code. Some
 -- things which should have been modified were not, including the code
 -- that writes out the file version header. Consequently, RCFile and
 -- SequenceFile originally shared the same version header.  A newer
 -- release has created a unique version string.

 file-version-header ::= Byte[4] {'S', 'E', 'Q', 6}
                     |   Byte[4] {'R', 'C', 'F', 1}

 -- The name of the Java class responsible for reading the key buffer
 -- component of the rowgroup.

 file-key-class-name ::=
   Text {"org.apache.hadoop.hive.ql.io.RCFile$KeyBuffer"}

 -- The name of the Java class responsible for reading the value buffer
 -- component of the rowgroup.

 file-value-class-name ::=
   Text {"org.apache.hadoop.hive.ql.io.RCFile$ValueBuffer"}

 -- Boolean variable indicating whether or not the file uses compression
 -- for the key and column buffer sections.

 file-is-compressed ::= Byte[1]

 -- A boolean field indicating whether or not the file is block compressed.
 -- This field is *always* false. According to comments in the original
 -- RCFile implementation this field was retained for backwards
 -- compatability with the SequenceFile format.

 file-is-block-compressed ::= Byte[1] {false}

 -- The Java class name of the compression codec iff <file-is-compressed>
 -- is true. The named class must implement
 -- org.apache.hadoop.io.compress.CompressionCodec.
 -- The expected value is org.apache.hadoop.io.compress.GzipCodec.

 file-compression-codec-class ::= Text

 -- A collection of key-value pairs defining metadata values for the
 -- file. The Map is serialized using standard JDK serialization, i.e.
 -- an Int corresponding to the number of key-value pairs, followed by
 -- Text key and value pairs. The following metadata properties are
 -- mandatory for all RCFiles:
 --
 -- hive.io.rcfile.column.number: the number of columns in the RCFile

 file-header-metadata ::= Map<Text, Text>

 -- A 16 byte marker that is generated by the writer. This marker appears
 -- at regular intervals at the beginning of rowgroup-headers, and is
 -- intended to enable readers to skip over corrupted rowgroups.

 file-sync-hash ::= Byte[16]

 -- Each row group is split into three sections: a header, a set of
 -- key buffers, and a set of column buffers. The header section includes
 -- an optional sync hash, information about the size of the row group, and
 -- the total number of rows in the row group. Each key buffer
 -- consists of run-length encoding data which is used to decode
 -- the length and offsets of individual fields in the corresponding column
 -- buffer.

 rcfile-rowgroup ::=
   <rowgroup-header>
   <rowgroup-key-data>
   <rowgroup-column-buffers>

 rowgroup-header ::=
   [<rowgroup-sync-marker>, <rowgroup-sync-hash>]
   <rowgroup-record-length>
   <rowgroup-key-length>
   <rowgroup-compressed-key-length>

 -- rowgroup-key-data is compressed if the column data is compressed.
 rowgroup-key-data ::=
   <rowgroup-num-rows>
   <rowgroup-key-buffers>

 -- An integer (always -1) signaling the beginning of a sync-hash
 -- field.

 rowgroup-sync-marker ::= Int

 -- A 16 byte sync field. This must match the <file-sync-hash> value read
 -- in the file header.

 rowgroup-sync-hash ::= Byte[16]

 -- The record-length is the sum of the number of bytes used to store
 -- the key and column parts, i.e. it is the total length of the current
 -- rowgroup.

 rowgroup-record-length ::= Int

 -- Total length in bytes of the rowgroup's key sections.

 rowgroup-key-length ::= Int

 -- Total compressed length in bytes of the rowgroup's key sections.

 rowgroup-compressed-key-length ::= Int

 -- Number of rows in the current rowgroup.

 rowgroup-num-rows ::= VInt

 -- One or more column key buffers corresponding to each column
 -- in the RCFile.

 rowgroup-key-buffers ::= <rowgroup-key-buffer>+

 -- Data in each column buffer is stored using a run-length
 -- encoding scheme that is intended to reduce the cost of
 -- repeated column field values. This mechanism is described
 -- in more detail in the following entries.

 rowgroup-key-buffer ::=
   <column-buffer-length>
   <column-buffer-uncompressed-length>
   <column-key-buffer-length>
   <column-key-buffer>

 -- The serialized length on disk of the corresponding column buffer.

 column-buffer-length ::= VInt

 -- The uncompressed length of the corresponding column buffer. This
 -- is equivalent to column-buffer-length if the RCFile is not compressed.

 column-buffer-uncompressed-length ::= VInt

 -- The length in bytes of the current column key buffer

 column-key-buffer-length ::= VInt

 -- The column-key-buffer contains a sequence of serialized VInt values
 -- corresponding to the byte lengths of the serialized column fields
 -- in the corresponding rowgroup-column-buffer. For example, consider
 -- an integer column that contains the consecutive values 1, 2, 3, 44.
 -- The RCFile format stores these values as strings in the column buffer,
 -- e.g. "12344". The length of each column field is recorded in
 -- the column-key-buffer as a sequence of VInts: 1,1,1,2. However,
 -- if the same length occurs repeatedly, then we replace repeated
 -- run lengths with the complement (i.e. negative) of the number of
 -- repetitions, so 1,1,1,2 becomes 1,~2,2.

 column-key-buffer ::= Byte[column-key-buffer-length]

 rowgroup-column-buffers ::= <rowgroup-value-buffer>+

 -- RCFile stores all column data as strings regardless of the
 -- underlying column type. The strings are neither length-prefixed or
 -- null-terminated, and decoding them into individual fields requires
 -- the use of the run-length information contained in the corresponding
 -- column-key-buffer.

 rowgroup-column-buffer ::= Byte[column-buffer-length]

 Byte ::= An eight-bit byte

 VInt ::= Variable length integer. The high-order bit of each byte
 indicates whether more bytes remain to be read. The low-order seven
 bits are appended as increasingly more significant bits in the
 resulting integer value.

 Int ::= A four-byte integer in big-endian format.

 Text ::= VInt, Chars (Length prefixed UTF-8 characters)
 
 

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
static class	RCFile.KeyBuffer KeyBuffer is the key of each record in RCFile.
static class	RCFile.Reader Read KeyBuffer/ValueBuffer pairs from a RCFile.
static class	RCFile.ValueBuffer ValueBuffer is the value of each record in RCFile.
static class	RCFile.Writer Write KeyBuffer/ValueBuffer pairs to a RCFile.

Field Summary

Fields

Modifier and Type	Field and Description
static String	BLOCK_MISSING_MESSAGE
static String	COLUMN_NUMBER_CONF_STR
static String	COLUMN_NUMBER_METADATA_STR
static String	RECORD_INTERVAL_CONF_STR
static int	SYNC_INTERVAL The number of bytes between sync points.
static String	TOLERATE_CORRUPTIONS_CONF_STR

Constructor Summary

Constructors

Constructor and Description
RCFile()

Method Summary

Methods

Modifier and Type	Method and Description
static org.apache.hadoop.io.SequenceFile.Metadata	createMetadata(org.apache.hadoop.io.Text... values) Create a metadata object with alternating key-value pairs.

Methods inherited from class java.lang.Object

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

Field Detail

COLUMN_NUMBER_METADATA_STR

public static final String COLUMN_NUMBER_METADATA_STR

See Also:

Constant Field Values

RECORD_INTERVAL_CONF_STR

public static final String RECORD_INTERVAL_CONF_STR

COLUMN_NUMBER_CONF_STR

public static final String COLUMN_NUMBER_CONF_STR

TOLERATE_CORRUPTIONS_CONF_STR

public static final String TOLERATE_CORRUPTIONS_CONF_STR

BLOCK_MISSING_MESSAGE

public static final String BLOCK_MISSING_MESSAGE

See Also:

Constant Field Values

SYNC_INTERVAL

public static final int SYNC_INTERVAL

The number of bytes between sync points.

See Also:

Constant Field Values

Constructor Detail

RCFile

public RCFile()

Method Detail

createMetadata

public static org.apache.hadoop.io.SequenceFile.Metadata createMetadata(org.apache.hadoop.io.Text... values)

Create a metadata object with alternating key-value pairs. Eg. metadata(key1, value1, key2, value2)