Skip to main content

ORC Format

Description​

ORC (Optimized Row Columnar) is a columnar storage format designed for Hadoop workloads. It provides efficient compression and fast columnar access, making it ideal for analytical queries. ORC files store data in a column-oriented format with built-in compression and indexing.

File Extensions​

  • .orc - ORC files

Implementation Details​

Reading​

The ORC implementation:

  • Uses pyorc library for reading
  • Supports schema inference from file
  • Converts ORC records to Python dictionaries
  • Efficient columnar reading

Writing​

Writing support:

  • Requires schema or keys specification
  • Supports compression (default: level 5)
  • Writes ORC files with schema
  • Efficient columnar writing

Key Features​

  • Columnar storage: Efficient for analytical queries
  • Compression: Built-in compression support
  • Schema support: Requires schema definition for writing
  • Totals support: Can count total rows
  • Type preservation: Maintains data types

Usage​

from iterable import open_iterable

# Basic reading
with open_iterable('data.orc') as source:
    for row in source:
        print(row)

# Writing with schema
with open_iterable('output.orc', mode='w', iterableargs={
    'keys': ['id', 'name', 'age'],
    'compression': 5  # Compression level (0-9)
}) as dest:
    dest.write({'id': '1', 'name': 'John', 'age': '30'})

Parameters​

  • keys (list[str]): Column names (required for writing if schema not provided)
  • schema (list[str]): ORC schema definition (optional, overrides keys)
  • compression (int): Compression level 0-9 (default: 5)

Limitations​

  1. pyorc dependency: Requires pyorc package
  2. Schema required: Must specify schema or keys for writing
  3. Flat data only: Designed for tabular/flat data structures
  4. Binary format: Not human-readable
  5. Hadoop ecosystem: Primarily designed for Hadoop environments

Compression Support​

ORC has built-in compression (separate from file-level compression):

  • Compression levels 0-9
  • Default compression level: 5

ORC files can also be compressed with file-level codecs:

  • GZip (.orc.gz)
  • BZip2 (.orc.bz2)
  • LZMA (.orc.xz)
  • LZ4 (.orc.lz4)
  • ZIP (.orc.zip)
  • Brotli (.orc.br)
  • ZStandard (.orc.zst)

Use Cases​

  • Hadoop ecosystems: Data storage in Hadoop
  • Data warehousing: Analytical data storage
  • ETL pipelines: Intermediate format for data transformation
  • Big data: Large-scale data processing

Error Handling​

  • Missing dependency: optional libraries raise ImportError with an install hint (pip install 'iterabledata[<extra>]' when an extra exists).
  • Write mode: read-only formats raise WriteNotSupportedError or ValueError when opened with mode="w".
  • Bad or unsupported input: may raise ValueError, OSError, or library-specific errors.
  • See Troubleshooting for decoding, detection, and engine issues.
  • Parquet - Similar columnar format
  • Arrow - Another columnar format
  • Delta Lake - Transactional layer over Parquet