ORC File Format

Optimized Row Columnar(ORC)文件格式提供了存储Hive数据的高效方法。它的设计是为了克服其他Hive文件格式的限制。使用ORC文件可以提高Hive在读取、写入和处理数据时的性能。

与RCFile格式相比，ORC文件格式有很多优点:

• 单个文件作为每个任务的输出，它减少了NameNode的负载。
• Hive类型支持，包括datetime、decimal和复杂类型(struct、list、map和union)
• 在文件中存储轻量级索引
  • 跳过不通过谓词过滤的行组
  • 寻找一个给定的行
• 基于数据类型的块模式压缩
  • 整数列的运行长度编码
  • 对字符串列的字典编码
• 使用单独的记录阅读器并发读取相同的文件
• 能够不扫描标记分割文件
• 限制读写所需的内存
• 使用协议缓冲区存储的元数据，允许添加和删除字段

文件结构

ORC文件包含称为stripes的行数据组，以及文件页脚中的辅助信息。在文件的末尾，一个postscript 保存了压缩参数和压缩页脚的大小。

默认的脚本大小为250MB。大的stripes可以大块的高效的从HDFS上读取。

文件页脚包含文件中的stripes 列表、每个stripes 的行数和每列的数据类型。它还包含列级的聚合计数、最小值、最大值和总和。

该图说明了ORC文件结构:

Stripe Structure

As shown in the diagram, each stripe in an ORC file holds index data, row data, and a stripe footer.

The stripe footer contains a directory of stream locations. Row data is used in table scans.

Index data includes min and max values for each column and the row positions within each column. (A bit field or bloom filter could also be included.) Row index entries provide offsets that enable seeking to the right compression block and byte within a decompressed block.  Note that ORC indexes are used only for the selection of stripes and row groups and not for answering queries.

Having relatively frequent row index entries enables row-skipping within a stripe for rapid reads, despite large stripe sizes. By default every 10,000 rows can be skipped.

With the ability to skip large sets of rows based on filter predicates, you can sort a table on its secondary keys to achieve a big reduction in execution time. For example, if the primary partition is transaction date, the table can be sorted on state, zip code, and last name. Then looking for records in one state will skip the records of all other states.

A complete specification of the format is given in the ORC specification.

HiveQL Syntax

File formats are specified at the table (or partition) level. You can specify the ORC file format with HiveQL statements such as these:

• CREATE TABLE ... STORED AS ORC
• ALTER TABLE ... [PARTITION partition_spec] SET FILEFORMAT ORC
• SET hive.default.fileformat=Orc

The parameters are all placed in the TBLPROPERTIES (see Create Table). They are:

For example, creating an ORC stored table without compression:

create table Addresses (name string,street string,city string,state string,zip int
) stored as orc tblproperties ("orc.compress"="NONE");

Serialization and Compression

The serialization of column data in an ORC file depends on whether the data type is integer or string.

Integer Column Serialization

Integer columns are serialized in two streams.

1. present bit stream: is the value non-null?
2. data stream: a stream of integers

Integer data is serialized in a way that takes advantage of the common distribution of numbers:

• Integers are encoded using a variable-width encoding that has fewer bytes for small integers.
• Repeated values are run-length encoded.
• Values that differ by a constant in the range (-128 to 127) are run-length encoded.

The variable-width encoding is based on Google's protocol buffers and uses the high bit to represent whether this byte is not the last and the lower 7 bits to encode data. To encode negative numbers, a zigzag encoding is used where 0, -1, 1, -2, and 2 map into 0, 1, 2, 3, 4, and 5 respectively.

Each set of numbers is encoded this way:

• If the first byte (b0) is negative:
  • -b0 variable-length integers follow.
• If the first byte (b0) is positive:
  • it represents b0 + 3 repeated integers
  • the second byte (-128 to +127) is added between each repetition
  • 1 variable-length integer.

In run-length encoding, the first byte specifies run length and whether the values are literals or duplicates. Duplicates can step by -128 to +128. Run-length encoding uses protobuf style variable-length integers.

String Column Serialization

Serialization of string columns uses a dictionary to form unique column values. The dictionary is sorted to speed up predicate filtering and improve compression ratios.

String columns are serialized in four streams.

1. present bit stream: is the value non-null?
2. dictionary data: the bytes for the strings
3. dictionary length: the length of each entry
4. row data: the row values

Both the dictionary length and the row values are run-length encoded streams of integers.

Compression

Streams are compressed using a codec, which is specified as a table property for all streams in that table. To optimize memory use, compression is done incrementally as each block is produced. Compressed blocks can be jumped over without first having to be decompressed for scanning. Positions in the stream are represented by a block start location and an offset into the block.

The codec can be Snappy, Zlib, or none.

ORC File Dump Utility

The ORC file dump utility analyzes ORC files.  To invoke it, use this command:

Specifying -d in the command will cause it to dump the ORC file data rather than the metadata (Hive 1.1.0 and later).

Specifying --rowindex with a comma separated list of column ids will cause it to print row indexes for the specified columns, where 0 is the top level struct containing all of the columns and 1 is the first column id (Hive 1.1.0 and later).

Specifying -t in the command will print the timezone id of the writer.

Specifying -j in the command will print the ORC file metadata in JSON format. To pretty print the JSON metadata, add -p to the command.

Specifying --recover in the command will recover a corrupted ORC file generated by Hive streaming.

Specifying --skip-dump along with --recover will perform recovery without dumping metadata.

Specifying --backup-path with a new-path will let the recovery tool move corrupted files to the specified backup path (default: /tmp).

<location-of-orc-file> is the URI of the ORC file.

<location-of-orc-file-or-directory> is the URI of the ORC file or directory. From Hive 1.3.0 onward, this URI can be a directory containing ORC files.

ORC Configuration Parameters

The ORC configuration parameters are described in Hive Configuration Properties – ORC File Format.

ORC Format Specification

The ORC specification has moved to ORC project

原文地址http://www.bieryun.com/2447.html