PyEClib

This library provides a simple Python interface for implementing erasure codes and is known to work with Python v2.6, 2.7 and 3.x.

To obtain the best possible performance, the library utilizes liberasurecode, which is a C based erasure code library. Please let us know if you have any issues building or installing (email: [email protected] or [email protected]).

PyECLib supports a variety of Erasure Coding backends including the standard Reed Soloman implementations provided by Jerasure [1], liberasurecode [3] and Intel ISA-L [4]. It also provides support for a flat XOR-based encoder and decoder (part of liberasurecode) - a class of HD Combination Codes based on "Flat XOR-based erasure codes in storage systems: Constructions, efficient recovery, and tradeoffs" in IEEE MSST 2010[2]). These codes are well-suited to archival use-cases, have a simple construction and require a minimum number of participating disks during single-disk reconstruction (think XOR-based LRC code).

Examples of using PyECLib are provided in the "tools" directory:

Command-line encoder:

tools/pyeclib_encode.py

Command-line decoder:

tools/pyeclib_decode.py

Utility to determine what is needed to reconstruct missing fragments:

tools/pyeclib_fragments_needed.py

A configuration utility to help compare available EC schemes in terms of performance and redundancy:

tools/pyeclib_conf_tool.py

PyEClib initialization:

ec_driver = ECDriver(k=<num_encoded_data_fragments>,
                     m=<num_encoded_parity_fragments>,
                     ec_type=<ec_scheme>))

Supported ec_type values:

• liberasurecode_rs_vand => Vandermonde Reed-Solomon encoding, software-only backend implemented by liberasurecode [3]
• jerasure_rs_vand => Vandermonde Reed-Solomon encoding, based on Jerasure [1]
• jerasure_rs_cauchy => Cauchy Reed-Solomon encoding (Jerasure variant), based on Jerasure [1]
• flat_xor_hd_3, flat_xor_hd_4 => Flat-XOR based HD combination codes, liberasurecode [3]
• isa_l_rs_vand => Intel Storage Acceleration Library (ISA-L) - SIMD accelerated Erasure Coding backends [4]
• isa_l_rs_cauchy => Cauchy Reed-Solomon encoding (ISA-L variant) [4]
• shss => NTT Lab Japan's Erasure Coding Library [5]
• libphazr => Phazr.IO's erasure code library with built-in privacy [6]

The Python API supports the following functions:

• EC Encode

  Encode N bytes of a data object into k (data) + m (parity) fragments:

  def encode(self, data_bytes)
  
  input:   data_bytes - input data object (bytes)
  returns: list of fragments (bytes)
  throws:
    ECBackendInstanceNotAvailable - if the backend library cannot be found
    ECBackendNotSupported - if the backend is not supported by PyECLib (see ec_types above)
    ECInvalidParameter - if invalid parameters were provided
    ECOutOfMemory - if the process has run out of memory
    ECDriverError - if an unknown error occurs
  

• EC Decode

  Decode between k and k+m fragments into original object:

  def decode(self, fragment_payloads)
  
  input:   list of fragment_payloads (bytes)
  returns: decoded object (bytes)
  throws:
    ECBackendInstanceNotAvailable - if the backend library cannot be found
    ECBackendNotSupported - if the backend is not supported by PyECLib (see ec_types above)
    ECInvalidParameter - if invalid parameters were provided
    ECOutOfMemory - if the process has run out of memory
    ECInsufficientFragments - if an insufficient set of fragments has been provided (e.g. not enough)
    ECInvalidFragmentMetadata - if the fragment headers appear to be corrupted
    ECDriverError - if an unknown error occurs
  

Note: bytes is a synonym to str in Python 2.6, 2.7. In Python 3.x, bytes and str types are non-interchangeable and care needs to be taken when handling input to and output from the encode() and decode() routines.

• EC Reconstruct

  Reconstruct "missing_fragment_indexes" using "available_fragment_payloads":

  def reconstruct(self, available_fragment_payloads, missing_fragment_indexes)
  
  input: available_fragment_payloads - list of fragment payloads
  input: missing_fragment_indexes - list of indexes to reconstruct
  output: list of reconstructed fragments corresponding to missing_fragment_indexes
  throws:
    ECBackendInstanceNotAvailable - if the backend library cannot be found
    ECBackendNotSupported - if the backend is not supported by PyECLib (see ec_types above)
    ECInvalidParameter - if invalid parameters were provided
    ECOutOfMemory - if the process has run out of memory
    ECInsufficientFragments - if an insufficient set of fragments has been provided (e.g. not enough)
    ECInvalidFragmentMetadata - if the fragment headers appear to be corrupted
    ECDriverError - if an unknown error occurs
  

• Minimum parity fragments needed for durability gurantees:

  def min_parity_fragments_needed(self)
  
  NOTE: Currently hard-coded to 1, so this can only be trusted for MDS codes, such as
        Reed-Solomon.
  
  output: minimum number of additional fragments needed to be synchronously written to tolerate
          the loss of any one fragment (similar guarantees to 2 out of 3 with 3x replication)
  throws:
    ECBackendInstanceNotAvailable - if the backend library cannot be found
    ECBackendNotSupported - if the backend is not supported by PyECLib (see ec_types above)
    ECInvalidParameter - if invalid parameters were provided
    ECOutOfMemory - if the process has run out of memory
    ECDriverError - if an unknown error occurs
  

• Fragments needed for EC Reconstruct

  Return the indexes of fragments needed to reconstruct "missing_fragment_indexes":

  def fragments_needed(self, missing_fragment_indexes)
  
  input: list of missing_fragment_indexes
  output: list of fragments needed to reconstruct fragments listed in missing_fragment_indexes
  throws:
    ECBackendInstanceNotAvailable - if the backend library cannot be found
    ECBackendNotSupported - if the backend is not supported by PyECLib (see ec_types above)
    ECInvalidParameter - if invalid parameters were provided
    ECOutOfMemory - if the process has run out of memory
    ECDriverError - if an unknown error occurs
  

• Get EC Metadata

  Return an opaque header known by the underlying library or a formatted header (Python dict):

  def get_metadata(self, fragment, formatted = 0)
  
  input: raw fragment payload
  input: boolean specifying if returned header is opaque buffer or formatted string
  output: fragment header (opaque or formatted)
  throws:
    ECBackendInstanceNotAvailable - if the backend library cannot be found
    ECBackendNotSupported - if the backend is not supported by PyECLib (see ec_types above)
    ECInvalidParameter - if invalid parameters were provided
    ECOutOfMemory - if the process has run out of memory
    ECDriverError - if an unknown error occurs
  

• Verify EC Stripe Consistency

  Use opaque buffers from get_metadata() to verify a the consistency of a stripe:

  def verify_stripe_metadata(self, fragment_metadata_list)
  
  intput: list of opaque fragment headers
  output: formatted string containing the 'status' (0 is success) and 'reason' if verification fails
  throws:
    ECBackendInstanceNotAvailable - if the backend library cannot be found
    ECBackendNotSupported - if the backend is not supported by PyECLib (see ec_types above)
    ECInvalidParameter - if invalid parameters were provided
    ECOutOfMemory - if the process has run out of memory
    ECDriverError - if an unknown error occurs
  

• Get EC Segment Info

  Return a dict with the keys - segment_size, last_segment_size, fragment_size, last_fragment_size and num_segments:

  def get_segment_info(self, data_len, segment_size)
  
  input: total data_len of the object to store
  input: target segment size used to segment the object into multiple EC stripes
  output: a dict with keys - segment_size, last_segment_size, fragment_size, last_fragment_size and num_segments
  throws:
    ECBackendInstanceNotAvailable - if the backend library cannot be found
    ECBackendNotSupported - if the backend is not supported by PyECLib (see ec_types above)
    ECInvalidParameter - if invalid parameters were provided
    ECOutOfMemory - if the process has run out of memory
    ECDriverError - if an unknown error occurs
  

• Get EC Segment Info given a list of ranges, data length and segment size:

  def get_segment_info_byterange(self, ranges, data_len, segment_size)
  
  input: byte ranges
  input: total data_len of the object to store
  input: target segment size used to segment the object into multiple EC stripes
  output: (see below)
  throws:
    ECBackendInstanceNotAvailable - if the backend library cannot be found
    ECBackendNotSupported - if the backend is not supported by PyECLib (see ec_types above)
    ECInvalidParameter - if invalid parameters were provided
    ECOutOfMemory - if the process has run out of memory
    ECDriverError - if an unknown error occurs
  

  Assume a range request is given for an object with segment size 3K and a 1 MB file:

  Ranges = (0, 1), (1, 12), (10, 1000), (0, segment_size-1),
           (1, segment_size+1), (segment_size-1, 2*segment_size)
  

  This will return a map keyed on the ranges, where there is a recipe given for each range:

  {
   (0, 1): {0: (0, 1)},
   (10, 1000): {0: (10, 1000)},
   (1, 12): {0: (1, 12)},
   (0, 3071): {0: (0, 3071)},
   (3071, 6144): {0: (3071, 3071), 1: (0, 3071), 2: (0, 0)},
   (1, 3073): {0: (1, 3071), 1: (0,0)}
  }
  

Quick Start

Install pre-requisites:

* Python 2.6, 2.7 or 3.x (including development packages), argparse, setuptools
* liberasurecode v1.2.0 or greater [3]
* Erasure code backend libraries, gf-complete and Jerasure [1],[2], ISA-L [4] etc

An example for ubuntu to install dependency packages:

$ sudo apt-get install build-essential python-dev python-pip liberasurecode-dev
$ sudo pip install -U bindep -r test-requirements.txt

If you want to confirm all dependency packages installed successfully, try:

$ sudo bindep -f bindep.txt

Note: currently liberasurecode-dev/liberasurecode-devel in package repo is older than v1.2.0

Install PyECLib:

$ sudo python setup.py install

Run test suite included:

$ ./.unittests

If all of this works, then you should be good to go. If not, send us an email!

If the test suite fails because it cannot find any of the shared libraries, then you probably need to add /usr/local/lib to the path searched when loading libraries. The best way to do this (on Linux) is to add '/usr/local/lib' to:

/etc/ld.so.conf

and then make sure to run:

$ sudo ldconfig

References

[1] Jerasure, C library that supports erasure coding in storage applications, http://jerasure.org

[2] Greenan, Kevin M et al, "Flat XOR-based erasure codes in storage systems", http://www.kaymgee.com/Kevin_Greenan/Publications_files/greenan-msst10.pdf

[3] liberasurecode, C API abstraction layer for erasure coding backends, https://github.com/openstack/liberasurecode

[4] Intel(R) Storage Acceleration Library (Open Source Version), https://01.org/intel%C2%AE-storage-acceleration-library-open-source-version

[5] Kota Tsuyuzaki <[email protected]>, "NTT SHSS Erasure Coding backend"

[6] Jim Cheung <[email protected]>, "Phazr.IO libphazr erasure code backend with built-in privacy"