This is a pre-production deployment of Warehouse. Changes made here affect the production instance of PyPI (
Help us improve Python packaging - Donate today!

Zstandard bindings for Python

Project Description

This project provides Python bindings for interfacing with the
`Zstandard <>`_ compression library. A C extension
and CFFI interface are provided.

The primary goal of the project is to provide a rich interface to the
underlying C API through a Pythonic interface while not sacrificing
performance. This means exposing most of the features and flexibility
of the C API while not sacrificing usability or safety that Python provides.

The canonical home for this project is

| |ci-status| |win-ci-status|

State of Project

The project is officially in beta state. The author is reasonably satisfied
that functionality works as advertised. **There will be some backwards
incompatible changes before 1.0, probably in the 0.9 release.** This may
involve renaming the main module from *zstd* to *zstandard* and renaming
various types and methods. Pin the package version to prevent unwanted
breakage when this change occurs!

This project is vendored and distributed with Mercurial 4.1, where it is
used in a production capacity.

There is continuous integration for Python versions 2.6, 2.7, and 3.3+
on Linux x86_x64 and Windows x86 and x86_64. The author is reasonably
confident the extension is stable and works as advertised on these

The CFFI bindings are mostly feature complete. Where a feature is implemented
in CFFI, unit tests run against both C extension and CFFI implementation to
ensure behavior parity.

Expected Changes

The author is reasonably confident in the current state of what's
implemented on the ``ZstdCompressor`` and ``ZstdDecompressor`` types.
Those APIs likely won't change significantly. Some low-level behavior
(such as naming and types expected by arguments) may change.

There will likely be arguments added to control the input and output
buffer sizes (currently, certain operations read and write in chunk
sizes using zstd's preferred defaults).

There should be an API that accepts an object that conforms to the buffer
interface and returns an iterator over compressed or decompressed output.

There should be an API that exposes an ``io.RawIOBase`` interface to
compressor and decompressor streams, like how ``gzip.GzipFile`` from
the standard library works (issue 13).

The author is on the fence as to whether to support the extremely
low level compression and decompression APIs. It could be useful to
support compression without the framing headers. But the author doesn't
believe it a high priority at this time.

There will likely be a refactoring of the module names. Currently,
``zstd`` is a C extension and ``zstd_cffi`` is the CFFI interface.
This means that all code for the C extension must be implemented in
C. ``zstd`` may be converted to a Python module so code can be reused
between CFFI and C and so not all code in the C extension has to be C.


This extension is designed to run with Python 2.6, 2.7, 3.3, 3.4, 3.5, and
3.6 on common platforms (Linux, Windows, and OS X). Only x86_64 is
currently well-tested as an architecture.


This package is uploaded to PyPI at
So, to install this package::

$ pip install zstandard

Binary wheels are made available for some platforms. If you need to
install from a source distribution, all you should need is a working C
compiler and the Python development headers/libraries. On many Linux
distributions, you can install a ``python-dev`` or ``python-devel``
package to provide these dependencies.

Packages are also uploaded to Anaconda Cloud at See that URL for how to install
this package with ``conda``.


Very crude and non-scientific benchmarking (most benchmarks fall in this
category because proper benchmarking is hard) show that the Python bindings
perform within 10% of the native C implementation.

The following table compares the performance of compressing and decompressing
a 1.1 GB tar file comprised of the files in a Firefox source checkout. Values
obtained with the ``zstd`` program are on the left. The remaining columns detail
performance of various compression APIs in the Python bindings.

| Level | Native | Simple | Stream In | Stream Out |
| | Comp / Decomp | Comp / Decomp | Comp / Decomp | Comp |
| 1 | 490 / 1338 MB/s | 458 / 1266 MB/s | 407 / 1156 MB/s | 405 MB/s |
| 2 | 412 / 1288 MB/s | 381 / 1203 MB/s | 345 / 1128 MB/s | 349 MB/s |
| 3 | 342 / 1312 MB/s | 319 / 1182 MB/s | 285 / 1165 MB/s | 287 MB/s |
| 11 | 64 / 1506 MB/s | 66 / 1436 MB/s | 56 / 1342 MB/s | 57 MB/s |

Again, these are very unscientific. But it shows that Python is capable of
compressing at several hundred MB/s and decompressing at over 1 GB/s.

Comparison to Other Python Bindings
=================================== is an alternate Python binding to
Zstandard. At the time this was written, the latest release of that
package (1.1.2) only exposed the simple APIs for compression and decompression.
This package exposes much more of the zstd API, including streaming and
dictionary compression. This package also has CFFI support.

Bundling of Zstandard Source Code

The source repository for this project contains a vendored copy of the
Zstandard source code. This is done for a few reasons.

First, Zstandard is relatively new and not yet widely available as a system
package. Providing a copy of the source code enables the Python C extension
to be compiled without requiring the user to obtain the Zstandard source code

Second, Zstandard has both a stable *public* API and an *experimental* API.
The *experimental* API is actually quite useful (contains functionality for
training dictionaries for example), so it is something we wish to expose to
Python. However, the *experimental* API is only available via static linking.
Furthermore, the *experimental* API can change at any time. So, control over
the exact version of the Zstandard library linked against is important to
ensure known behavior.

Instructions for Building and Testing

Once you have the source code, the extension can be built via

$ python build_ext

We recommend testing with ``nose``::

$ nosetests

A Tox configuration is present to test against multiple Python versions::

$ tox

Tests use the ``hypothesis`` Python package to perform fuzzing. If you
don't have it, those tests won't run. Since the fuzzing tests take longer
to execute than normal tests, you'll need to opt in to running them by
setting the ``ZSTD_SLOW_TESTS`` environment variable. This is set
automatically when using ``tox``.

The ``cffi`` Python package needs to be installed in order to build the CFFI
bindings. If it isn't present, the CFFI bindings won't be built.

To create a virtualenv with all development dependencies, do something
like the following::

# Python 2
$ virtualenv venv

# Python 3
$ python3 -m venv venv

$ source venv/bin/activate
$ pip install cffi hypothesis nose tox


The compiled C extension provides a ``zstd`` Python module. The CFFI
bindings provide a ``zstd_cffi`` module. Both provide an identical API
interface. The types, functions, and attributes exposed by these modules
are documented in the sections below.

.. note::

The documentation in this section makes references to various zstd
concepts and functionality. The ``Concepts`` section below explains
these concepts in more detail.


The ``ZstdCompressor`` class provides an interface for performing
compression operations.

Each instance is associated with parameters that control compression
behavior. These come from the following named arguments (all optional):

Integer compression level. Valid values are between 1 and 22.
Compression dictionary to use.

Note: When using dictionary data and ``compress()`` is called multiple
times, the ``CompressionParameters`` derived from an integer compression
``level`` and the first compressed data's size will be reused for all
subsequent operations. This may not be desirable if source data size
varies significantly.
A ``CompressionParameters`` instance (overrides the ``level`` value).
Whether a 4 byte checksum should be written with the compressed data.
Defaults to False. If True, the decompressor can verify that decompressed
data matches the original input data.
Whether the size of the uncompressed data will be written into the
header of compressed data. Defaults to False. The data will only be
written if the compressor knows the size of the input data. This is
likely not true for streaming compression.
Whether to write the dictionary ID into the compressed data.
Defaults to True. The dictionary ID is only written if a dictionary
is being used.
Enables and sets the number of threads to use for multi-threaded compression
operations. Defaults to 0, which means to use single-threaded compression.
Negative values will resolve to the number of logical CPUs in the system.
Read below for more info on multi-threaded compression. This argument only
controls thread count for operations that operate on individual pieces of
data. APIs that spawn multiple threads for working on multiple pieces of
data have their own ``threads`` argument.

Unless specified otherwise, assume that no two methods of ``ZstdCompressor``
instances can be called from multiple Python threads simultaneously. In other
words, assume instances are not thread safe unless stated otherwise.

Simple API

``compress(data)`` compresses and returns data as a one-shot operation.::

cctx = zstd.ZstdCompressor()
compressed = cctx.compress(b'data to compress')

The ``data`` argument can be any object that implements the *buffer protocol*.

Unless ``compression_params`` or ``dict_data`` are passed to the
``ZstdCompressor``, each invocation of ``compress()`` will calculate the
optimal compression parameters for the configured compression ``level`` and
input data size (some parameters are fine-tuned for small input sizes).

If a compression dictionary is being used, the compression parameters
determined from the first input's size will be reused for subsequent

There is currently a deficiency in zstd's C APIs that makes it difficult
to round trip empty inputs when ``write_content_size=True``. Attempting
this will raise a ``ValueError`` unless ``allow_empty=True`` is passed
to ``compress()``.

Streaming Input API

``write_to(fh)`` (which behaves as a context manager) allows you to *stream*
data into a compressor.::

cctx = zstd.ZstdCompressor(level=10)
with cctx.write_to(fh) as compressor:
compressor.write(b'chunk 0')
compressor.write(b'chunk 1')

The argument to ``write_to()`` must have a ``write(data)`` method. As
compressed data is available, ``write()`` will be called with the compressed
data as its argument. Many common Python types implement ``write()``, including
open file handles and ``io.BytesIO``.

``write_to()`` returns an object representing a streaming compressor instance.
It **must** be used as a context manager. That object's ``write(data)`` method
is used to feed data into the compressor.

A ``flush()`` method can be called to evict whatever data remains within the
compressor's internal state into the output object. This may result in 0 or
more ``write()`` calls to the output object.

Both ``write()`` and ``flush()`` return the number of bytes written to the
object's ``write()``. In many cases, small inputs do not accumulate enough
data to cause a write and ``write()`` will return ``0``.

If the size of the data being fed to this streaming compressor is known,
you can declare it before compression begins::

cctx = zstd.ZstdCompressor()
with cctx.write_to(fh, size=data_len) as compressor:

Declaring the size of the source data allows compression parameters to
be tuned. And if ``write_content_size`` is used, it also results in the
content size being written into the frame header of the output data.

The size of chunks being ``write()`` to the destination can be specified::

cctx = zstd.ZstdCompressor()
with cctx.write_to(fh, write_size=32768) as compressor:

To see how much memory is being used by the streaming compressor::

cctx = zstd.ZstdCompressor()
with cctx.write_to(fh) as compressor:
byte_size = compressor.memory_size()

Streaming Output API

``read_from(reader)`` provides a mechanism to stream data out of a compressor
as an iterator of data chunks.::

cctx = zstd.ZstdCompressor()
for chunk in cctx.read_from(fh):
# Do something with emitted data.

``read_from()`` accepts an object that has a ``read(size)`` method or conforms
to the buffer protocol. (``bytes`` and ``memoryview`` are 2 common types that
provide the buffer protocol.)

Uncompressed data is fetched from the source either by calling ``read(size)``
or by fetching a slice of data from the object directly (in the case where
the buffer protocol is being used). The returned iterator consists of chunks
of compressed data.

If reading from the source via ``read()``, ``read()`` will be called until
it raises or returns an empty bytes (``b''``). It is perfectly valid for
the source to deliver fewer bytes than were what requested by ``read(size)``.

Like ``write_to()``, ``read_from()`` also accepts a ``size`` argument
declaring the size of the input stream::

cctx = zstd.ZstdCompressor()
for chunk in cctx.read_from(fh, size=some_int):

You can also control the size that data is ``read()`` from the source and
the ideal size of output chunks::

cctx = zstd.ZstdCompressor()
for chunk in cctx.read_from(fh, read_size=16384, write_size=8192):

Unlike ``write_to()``, ``read_from()`` does not give direct control over the
sizes of chunks fed into the compressor. Instead, chunk sizes will be whatever
the object being read from delivers. These will often be of a uniform size.

Stream Copying API

``copy_stream(ifh, ofh)`` can be used to copy data between 2 streams while
compressing it.::

cctx = zstd.ZstdCompressor()
cctx.copy_stream(ifh, ofh)

For example, say you wish to compress a file::

cctx = zstd.ZstdCompressor()
with open(input_path, 'rb') as ifh, open(output_path, 'wb') as ofh:
cctx.copy_stream(ifh, ofh)

It is also possible to declare the size of the source stream::

cctx = zstd.ZstdCompressor()
cctx.copy_stream(ifh, ofh, size=len_of_input)

You can also specify how large the chunks that are ``read()`` and ``write()``
from and to the streams::

cctx = zstd.ZstdCompressor()
cctx.copy_stream(ifh, ofh, read_size=32768, write_size=16384)

The stream copier returns a 2-tuple of bytes read and written::

cctx = zstd.ZstdCompressor()
read_count, write_count = cctx.copy_stream(ifh, ofh)

Compressor API

``compressobj()`` returns an object that exposes ``compress(data)`` and
``flush()`` methods. Each returns compressed data or an empty bytes.

The purpose of ``compressobj()`` is to provide an API-compatible interface
with ``zlib.compressobj`` and ``bz2.BZ2Compressor``. This allows callers to
swap in different compressor objects while using the same API.

``flush()`` accepts an optional argument indicating how to end the stream.
``zstd.COMPRESSOBJ_FLUSH_FINISH`` (the default) ends the compression stream.
Once this type of flush is performed, ``compress()`` and ``flush()`` can
no longer be called. This type of flush **must** be called to end the
compression context. If not called, returned data may be incomplete.

A ``zstd.COMPRESSOBJ_FLUSH_BLOCK`` argument to ``flush()`` will flush a
zstd block. Flushes of this type can be performed multiple times. The next
call to ``compress()`` will begin a new zstd block.

Here is how this API should be used::

cctx = zstd.ZstdCompressor()
cobj = cctx.compressobj()
data = cobj.compress(b'raw input 0')
data = cobj.compress(b'raw input 1')
data = cobj.flush()

Or to flush blocks::

cobj = cctx.compressobj()
data = cobj.compress(b'chunk in first block')
data = cobj.flush(zstd.COMPRESSOBJ_FLUSH_BLOCK)
data = cobj.compress(b'chunk in second block')
data = cobj.flush()

For best performance results, keep input chunks under 256KB. This avoids
extra allocations for a large output object.

It is possible to declare the input size of the data that will be fed into
the compressor::

cctx = zstd.ZstdCompressor()
cobj = cctx.compressobj(size=6)
data = cobj.compress(b'foobar')
data = cobj.flush()

Batch Compression API

(Experimental. Not yet supported in CFFI bindings.)

``multi_compress_to_buffer(data, [threads=0])`` performs compression of multiple
inputs as a single operation.

Data to be compressed can be passed as a ``BufferWithSegmentsCollection``, a
``BufferWithSegments``, or a list containing byte like objects. Each element of
the container will be compressed individually using the configured parameters
on the ``ZstdCompressor`` instance.

The ``threads`` argument controls how many threads to use for compression. The
default is ``0`` which means to use a single thread. Negative values use the
number of logical CPUs in the machine.

The function returns a ``BufferWithSegmentsCollection``. This type represents
N discrete memory allocations, eaching holding 1 or more compressed frames.

Output data is written to shared memory buffers. This means that unlike
regular Python objects, a reference to *any* object within the collection
keeps the shared buffer and therefore memory backing it alive. This can have
undesirable effects on process memory usage.

The API and behavior of this function is experimental and will likely change.
Known deficiencies include:

* If asked to use multiple threads, it will always spawn that many threads,
even if the input is too small to use them. It should automatically lower
the thread count when the extra threads would just add overhead.
* The buffer allocation strategy is fixed. There is room to make it dynamic,
perhaps even to allow one output buffer per input, facilitating a variation
of the API to return a list without the adverse effects of shared memory


The ``ZstdDecompressor`` class provides an interface for performing

Each instance is associated with parameters that control decompression. These
come from the following named arguments (all optional):

Compression dictionary to use.

The interface of this class is very similar to ``ZstdCompressor`` (by design).

Unless specified otherwise, assume that no two methods of ``ZstdDecompressor``
instances can be called from multiple Python threads simultaneously. In other
words, assume instances are not thread safe unless stated otherwise.

Simple API

``decompress(data)`` can be used to decompress an entire compressed zstd
frame in a single operation.::

dctx = zstd.ZstdDecompressor()
decompressed = dctx.decompress(data)

By default, ``decompress(data)`` will only work on data written with the content
size encoded in its header. This can be achieved by creating a
``ZstdCompressor`` with ``write_content_size=True``. If compressed data without
an embedded content size is seen, ``zstd.ZstdError`` will be raised.

If the compressed data doesn't have its content size embedded within it,
decompression can be attempted by specifying the ``max_output_size``

dctx = zstd.ZstdDecompressor()
uncompressed = dctx.decompress(data, max_output_size=1048576)

Ideally, ``max_output_size`` will be identical to the decompressed output

If ``max_output_size`` is too small to hold the decompressed data,
``zstd.ZstdError`` will be raised.

If ``max_output_size`` is larger than the decompressed data, the allocated
output buffer will be resized to only use the space required.

Please note that an allocation of the requested ``max_output_size`` will be
performed every time the method is called. Setting to a very large value could
result in a lot of work for the memory allocator and may result in
``MemoryError`` being raised if the allocation fails.

If the exact size of decompressed data is unknown, it is **strongly**
recommended to use a streaming API.

Streaming Input API

``write_to(fh)`` can be used to incrementally send compressed data to a

dctx = zstd.ZstdDecompressor()
with dctx.write_to(fh) as decompressor:

This behaves similarly to ``zstd.ZstdCompressor``: compressed data is written to
the decompressor by calling ``write(data)`` and decompressed output is written
to the output object by calling its ``write(data)`` method.

Calls to ``write()`` will return the number of bytes written to the output
object. Not all inputs will result in bytes being written, so return values
of ``0`` are possible.

The size of chunks being ``write()`` to the destination can be specified::

dctx = zstd.ZstdDecompressor()
with dctx.write_to(fh, write_size=16384) as decompressor:

You can see how much memory is being used by the decompressor::

dctx = zstd.ZstdDecompressor()
with dctx.write_to(fh) as decompressor:
byte_size = decompressor.memory_size()

Streaming Output API

``read_from(fh)`` provides a mechanism to stream decompressed data out of a
compressed source as an iterator of data chunks.::

dctx = zstd.ZstdDecompressor()
for chunk in dctx.read_from(fh):
# Do something with original data.

``read_from()`` accepts a) an object with a ``read(size)`` method that will
return compressed bytes b) an object conforming to the buffer protocol that
can expose its data as a contiguous range of bytes. The ``bytes`` and
``memoryview`` types expose this buffer protocol.

``read_from()`` returns an iterator whose elements are chunks of the
decompressed data.

The size of requested ``read()`` from the source can be specified::

dctx = zstd.ZstdDecompressor()
for chunk in dctx.read_from(fh, read_size=16384):

It is also possible to skip leading bytes in the input data::

dctx = zstd.ZstdDecompressor()
for chunk in dctx.read_from(fh, skip_bytes=1):

Skipping leading bytes is useful if the source data contains extra
*header* data but you want to avoid the overhead of making a buffer copy
or allocating a new ``memoryview`` object in order to decompress the data.

Similarly to ``ZstdCompressor.read_from()``, the consumer of the iterator
controls when data is decompressed. If the iterator isn't consumed,
decompression is put on hold.

When ``read_from()`` is passed an object conforming to the buffer protocol,
the behavior may seem similar to what occurs when the simple decompression
API is used. However, this API works when the decompressed size is unknown.
Furthermore, if feeding large inputs, the decompressor will work in chunks
instead of performing a single operation.

Stream Copying API

``copy_stream(ifh, ofh)`` can be used to copy data across 2 streams while
performing decompression.::

dctx = zstd.ZstdDecompressor()
dctx.copy_stream(ifh, ofh)

e.g. to decompress a file to another file::

dctx = zstd.ZstdDecompressor()
with open(input_path, 'rb') as ifh, open(output_path, 'wb') as ofh:
dctx.copy_stream(ifh, ofh)

The size of chunks being ``read()`` and ``write()`` from and to the streams
can be specified::

dctx = zstd.ZstdDecompressor()
dctx.copy_stream(ifh, ofh, read_size=8192, write_size=16384)

Decompressor API

``decompressobj()`` returns an object that exposes a ``decompress(data)``
methods. Compressed data chunks are fed into ``decompress(data)`` and
uncompressed output (or an empty bytes) is returned. Output from subsequent
calls needs to be concatenated to reassemble the full decompressed byte

The purpose of ``decompressobj()`` is to provide an API-compatible interface
with ``zlib.decompressobj`` and ``bz2.BZ2Decompressor``. This allows callers
to swap in different decompressor objects while using the same API.

Each object is single use: once an input frame is decoded, ``decompress()``
can no longer be called.

Here is how this API should be used::

dctx = zstd.ZstdDeompressor()
dobj = cctx.decompressobj()
data = dobj.decompress(compressed_chunk_0)
data = dobj.decompress(compressed_chunk_1)

Batch Decompression API

(Experimental. Not yet supported in CFFI bindings.)

``multi_decompress_to_buffer()`` performs decompression of multiple
frames as a single operation and returns a ``BufferWithSegmentsCollection``
containing decompressed data for all inputs.

Compressed frames can be passed to the function as a ``BufferWithSegments``,
a ``BufferWithSegmentsCollection``, or as a list containing objects that
conform to the buffer protocol. For best performance, pass a
``BufferWithSegmentsCollection`` or a ``BufferWithSegments``, as
minimal input validation will be done for that type. If calling from
Python (as opposed to C), constructing one of these instances may add
overhead cancelling out the performance overhead of validation for list

The decompressed size of each frame must be discoverable. It can either be
embedded within the zstd frame (``write_content_size=True`` argument to
``ZstdCompressor``) or passed in via the ``decompressed_sizes`` argument.

The ``decompressed_sizes`` argument is an object conforming to the buffer
protocol which holds an array of 64-bit unsigned integers in the machine's
native format defining the decompressed sizes of each frame. If this argument
is passed, it avoids having to scan each frame for its decompressed size.
This frame scanning can add noticeable overhead in some scenarios.

The ``threads`` argument controls the number of threads to use to perform
decompression operations. The default (``0``) or the value ``1`` means to
use a single thread. Negative values use the number of logical CPUs in the

.. note::

It is possible to pass a ``mmap.mmap()`` instance into this function by
wrapping it with a ``BufferWithSegments`` instance (which will define the
offsets of frames within the memory mapped region).

This function is logically equivalent to performing ``dctx.decompress()``
on each input frame and returning the result.

This function exists to perform decompression on multiple frames as fast
as possible by having as little overhead as possible. Since decompression is
performed as a single operation and since the decompressed output is stored in
a single buffer, extra memory allocations, Python objects, and Python function
calls are avoided. This is ideal for scenarios where callers need to access
decompressed data for multiple frames.

Currently, the implementation always spawns multiple threads when requested,
even if the amount of work to do is small. In the future, it will be smarter
about avoiding threads and their associated overhead when the amount of
work to do is small.

Content-Only Dictionary Chain Decompression

``decompress_content_dict_chain(frames)`` performs decompression of a list of
zstd frames produced using chained *content-only* dictionary compression. Such
a list of frames is produced by compressing discrete inputs where each
non-initial input is compressed with a *content-only* dictionary consisting
of the content of the previous input.

For example, say you have the following inputs::

inputs = [b'input 1', b'input 2', b'input 3']

The zstd frame chain consists of:

1. ``b'input 1'`` compressed in standalone/discrete mode
2. ``b'input 2'`` compressed using ``b'input 1'`` as a *content-only* dictionary
3. ``b'input 3'`` compressed using ``b'input 2'`` as a *content-only* dictionary

Each zstd frame **must** have the content size written.

The following Python code can be used to produce a *content-only dictionary

def make_chain(inputs):
frames = []

# First frame is compressed in standalone/discrete mode.
zctx = zstd.ZstdCompressor(write_content_size=True)

# Subsequent frames use the previous fulltext as a content-only dictionary
for i, raw in enumerate(inputs[1:]):
dict_data = zstd.ZstdCompressionDict(inputs[i])
zctx = zstd.ZstdCompressor(write_content_size=True, dict_data=dict_data)

return frames

``decompress_content_dict_chain()`` returns the uncompressed data of the last
element in the input chain.

It is possible to implement *content-only dictionary chain* decompression
on top of other Python APIs. However, this function will likely be significantly
faster, especially for long input chains, as it avoids the overhead of
instantiating and passing around intermediate objects between C and Python.

Multi-Threaded Compression

``ZstdCompressor`` accepts a ``threads`` argument that controls the number
of threads to use for compression. The way this works is that input is split
into segments and each segment is fed into a worker pool for compression. Once
a segment is compressed, it is flushed/appended to the output.

The segment size for multi-threaded compression is chosen from the window size
of the compressor. This is derived from the ``window_log`` attribute of a
``CompressionParameters`` instance. By default, segment sizes are in the 1+MB

If multi-threaded compression is requested and the input is smaller than the
configured segment size, only a single compression thread will be used. If the
input is smaller than the segment size multiplied by the thread pool size or
if data cannot be delivered to the compressor fast enough, not all requested
compressor threads may be active simultaneously.

Compared to non-multi-threaded compression, multi-threaded compression has
higher per-operation overhead. This includes extra memory operations,
thread creation, lock acquisition, etc.

Due to the nature of multi-threaded compression using *N* compression
*states*, the output from multi-threaded compression will likely be larger
than non-multi-threaded compression. The difference is usually small. But
there is a CPU/wall time versus size trade off that may warrant investigation.

Output from multi-threaded compression does not require any special handling
on the decompression side. In other words, any zstd decompressor should be able
to consume data produced with multi-threaded compression.

Dictionary Creation and Management

Compression dictionaries are represented as the ``ZstdCompressionDict`` type.

Instances can be constructed from bytes::

dict_data = zstd.ZstdCompressionDict(data)

It is possible to construct a dictionary from *any* data. Unless the
data begins with a magic header, the dictionary will be treated as
*content-only*. *Content-only* dictionaries allow compression operations
that follow to reference raw data within the content. For one use of
*content-only* dictionaries, see

More interestingly, instances can be created by *training* on sample data::

dict_data = zstd.train_dictionary(size, samples)

This takes a list of bytes instances and creates and returns a

You can see how many bytes are in the dictionary by calling ``len()``::

dict_data = zstd.train_dictionary(size, samples)
dict_size = len(dict_data) # will not be larger than ``size``

Once you have a dictionary, you can pass it to the objects performing
compression and decompression::

dict_data = zstd.train_dictionary(16384, samples)

cctx = zstd.ZstdCompressor(dict_data=dict_data)
for source_data in input_data:
compressed = cctx.compress(source_data)
# Do something with compressed data.

dctx = zstd.ZstdDecompressor(dict_data=dict_data)
for compressed_data in input_data:
buffer = io.BytesIO()
with dctx.write_to(buffer) as decompressor:
# Do something with raw data in ``buffer``.

Dictionaries have unique integer IDs. You can retrieve this ID via::

dict_id = zstd.dictionary_id(dict_data)

You can obtain the raw data in the dict (useful for persisting and constructing
a ``ZstdCompressionDict`` later) via ``as_bytes()``::

dict_data = zstd.train_dictionary(size, samples)
raw_data = dict_data.as_bytes()

The following named arguments to ``train_dictionary`` can also be used
to further control dictionary generation.

Integer selectivity level. Default is 9. Larger values yield more data in
Integer compression level. Default is 6.
Integer dictionary ID for the produced dictionary. Default is 0, which
means to use a random value.
Controls writing of informational messages to ``stderr``. ``0`` (the
default) means to write nothing. ``1`` writes errors. ``2`` writes
progression info. ``3`` writes more details. And ``4`` writes all info.

Cover Dictionaries

An alternate dictionary training mechanism named *cover* is also available.
More details about this training mechanism are available in the paper
*Effective Construction of Relative Lempel-Ziv Dictionaries* (authors:
Liao, Petri, Moffat, Wirth).

To use this mechanism, use ``zstd.train_cover_dictionary()`` instead of
``zstd.train_dictionary()``. The function behaves nearly the same except
its arguments are different and the returned dictionary will contain ``k``
and ``d`` attributes reflecting the parameters to the cover algorithm.

.. note::

The ``k`` and ``d`` attributes are only populated on dictionary
instances created by this function. If a ``ZstdCompressionDict`` is
constructed from raw bytes data, the ``k`` and ``d`` attributes will
be ``0``.

The segment and dmer size parameters to the cover algorithm can either be
specified manually or you can ask ``train_cover_dictionary()`` to try
multiple values and pick the best one, where *best* means the smallest
compressed data size.

In manual mode, the ``k`` and ``d`` arguments must be specified or a
``ZstdError`` will be raised.

In automatic mode (triggered by specifying ``optimize=True``), ``k``
and ``d`` are optional. If a value isn't specified, then default values for
both are tested. The ``steps`` argument can control the number of steps
through ``k`` values. The ``level`` argument defines the compression level
that will be used when testing the compressed size. And ``threads`` can
specify the number of threads to use for concurrent operation.

This function takes the following arguments:

Target size in bytes of the dictionary to generate.
A list of bytes holding samples the dictionary will be trained from.
Parameter to cover algorithm defining the segment size. A reasonable range
is [16, 2048+].
Parameter to cover algorithm defining the dmer size. A reasonable range is
[6, 16]. ``d`` must be less than or equal to ``k``.
Integer dictionary ID for the produced dictionary. Default is 0, which uses
a random value.
When true, test dictionary generation with multiple parameters.
Integer target compression level when testing compression with
``optimize=True``. Default is 1.
Number of steps through ``k`` values to perform when ``optimize=True``.
Default is 32.
Number of threads to use when ``optimize=True``. Default is 0, which means
to use a single thread. A negative value can be specified to use as many
threads as there are detected logical CPUs.
Controls writing of informational messages to ``stderr``. See the
documentation for ``train_dictionary()`` for more.

Explicit Compression Parameters

Zstandard's integer compression levels along with the input size and dictionary
size are converted into a data structure defining multiple parameters to tune
behavior of the compression algorithm. It is possible to use define this
data structure explicitly to have lower-level control over compression behavior.

The ``zstd.CompressionParameters`` type represents this data structure.
You can see how Zstandard converts compression levels to this data structure
by calling ``zstd.get_compression_parameters()``. e.g.::

params = zstd.get_compression_parameters(5)

This function also accepts the uncompressed data size and dictionary size
to adjust parameters::

params = zstd.get_compression_parameters(3, source_size=len(data), dict_size=len(dict_data))

You can also construct compression parameters from their low-level components::

params = zstd.CompressionParameters(20, 6, 12, 5, 4, 10, zstd.STRATEGY_FAST)

You can then configure a compressor to use the custom parameters::

cctx = zstd.ZstdCompressor(compression_params=params)

The members/attributes of ``CompressionParameters`` instances are as follows::

* window_log
* chain_log
* hash_log
* search_log
* search_length
* target_length
* strategy

This is the order the arguments are passed to the constructor if not using
named arguments.

You'll need to read the Zstandard documentation for what these parameters

Frame Inspection

Data emitted from zstd compression is encapsulated in a *frame*. This frame
begins with a 4 byte *magic number* header followed by 2 to 14 bytes describing
the frame in more detail. For more info, see

``zstd.get_frame_parameters(data)`` parses a zstd *frame* header from a bytes
instance and return a ``FrameParameters`` object describing the frame.

Depending on which fields are present in the frame and their values, the
length of the frame parameters varies. If insufficient bytes are passed
in to fully parse the frame parameters, ``ZstdError`` is raised. To ensure
frame parameters can be parsed, pass in at least 18 bytes.

``FrameParameters`` instances have the following attributes:

Integer size of original, uncompressed content. This will be ``0`` if the
original content size isn't written to the frame (controlled with the
``write_content_size`` argument to ``ZstdCompressor``) or if the input
content size was ``0``.

Integer size of maximum back-reference distance in compressed data.

Integer of dictionary ID used for compression. ``0`` if no dictionary
ID was used or if the dictionary ID was ``0``.

Bool indicating whether a 4 byte content checksum is stored at the end
of the frame.

Misc Functionality


Given a ``CompressionParameters`` struct, estimate the memory size required
to perform compression.


Estimate the memory size requirements for a decompressor instance.


The following module constants/attributes are exposed:

This module attribute exposes a 3-tuple of the Zstandard version. e.g.
``(1, 0, 0)``
Integer max compression level accepted by compression functions
Recommended chunk size to feed to compressor functions
Recommended chunk size for compression output
Recommended chunk size to feed into decompresor functions
Recommended chunk size for decompression output

bytes containing header of the Zstandard frame
Frame header as an integer

Minimum value for compression parameter
Maximum value for compression parameter
Minimum value for compression parameter
Maximum value for compression parameter
Minimum value for compression parameter
Maximum value for compression parameter
Minimum value for compression parameter
Maximum value for compression parameter
Minimum value for compression parameter
Maximum value for compression parameter
Minimum value for compression parameter
Maximum value for compression parameter
Compression strategy
Compression strategy
Compression strategy
Compression strategy
Compression strategy
Compression strategy
Compression strategy

Performance Considerations

The ``ZstdCompressor`` and ``ZstdDecompressor`` types maintain state to a
persistent compression or decompression *context*. Reusing a ``ZstdCompressor``
or ``ZstdDecompressor`` instance for multiple operations is faster than
instantiating a new ``ZstdCompressor`` or ``ZstdDecompressor`` for each
operation. The differences are magnified as the size of data decreases. For
example, the difference between *context* reuse and non-reuse for 100,000
100 byte inputs will be significant (possiby over 10x faster to reuse contexts)
whereas 10 1,000,000 byte inputs will be more similar in speed (because the
time spent doing compression dwarfs time spent creating new *contexts*).

Buffer Types

The API exposes a handful of custom types for interfacing with memory buffers.
The primary goal of these types is to facilitate efficient multi-object

The essential idea is to have a single memory allocation provide backing
storage for multiple logical objects. This has 2 main advantages: fewer
allocations and optimal memory access patterns. This avoids having to allocate
a Python object for each logical object and furthermore ensures that access of
data for objects can be sequential (read: fast) in memory.


The ``BufferWithSegments`` type represents a memory buffer containing N
discrete items of known lengths (segments). It is essentially a fixed size
memory address and an array of 2-tuples of ``(offset, length)`` 64-bit
unsigned native endian integers defining the byte offset and length of each
segment within the buffer.

Instances behave like containers.

``len()`` returns the number of segments within the instance.

``o[index]`` or ``__getitem__`` obtains a ``BufferSegment`` representing an
individual segment within the backing buffer. That returned object references
(not copies) memory. This means that iterating all objects doesn't copy
data within the buffer.

The ``.size`` attribute contains the total size in bytes of the backing

Instances conform to the buffer protocol. So a reference to the backing bytes
can be obtained via ``memoryview(o)``. A *copy* of the backing bytes can also
be obtained via ``.tobytes()``.

The ``.segments`` attribute exposes the array of ``(offset, length)`` for
segments within the buffer. It is a ``BufferSegments`` type.


The ``BufferSegment`` type represents a segment within a ``BufferWithSegments``.
It is essentially a reference to N bytes within a ``BufferWithSegments``.

``len()`` returns the length of the segment in bytes.

``.offset`` contains the byte offset of this segment within its parent
``BufferWithSegments`` instance.

The object conforms to the buffer protocol. ``.tobytes()`` can be called to
obtain a ``bytes`` instance with a copy of the backing bytes.


This type represents an array of ``(offset, length)`` integers defining segments
within a ``BufferWithSegments``.

The array members are 64-bit unsigned integers using host/native bit order.

Instances conform to the buffer protocol.


The ``BufferWithSegmentsCollection`` type represents a virtual spanning view
of multiple ``BufferWithSegments`` instances.

Instances are constructed from 1 or more ``BufferWithSegments`` instances. The
resulting object behaves like an ordered sequence whose members are the
segments within each ``BufferWithSegments``.

``len()`` returns the number of segments within all ``BufferWithSegments``

``o[index]`` and ``__getitem__(index)`` return the ``BufferSegment`` at
that offset as if all ``BufferWithSegments`` instances were a single

If the object is composed of 2 ``BufferWithSegments`` instances with the
first having 2 segments and the second have 3 segments, then ``b[0]``
and ``b[1]`` access segments in the first object and ``b[2]``, ``b[3]``,
and ``b[4]`` access segments from the second.

Choosing an API

There are multiple APIs for performing compression and decompression. This is
because different applications have different needs and the library wants to
facilitate optimal use in as many use cases as possible.

From a high-level, APIs are divided into *one-shot* and *streaming*. See
the ``Concepts`` section for a description of how these are different at
the C layer.

The *one-shot* APIs are useful for small data, where the input or output
size is known. (The size can come from a buffer length, file size, or
stored in the zstd frame header.) A limitation of the *one-shot* APIs is that
input and output must fit in memory simultaneously. For say a 4 GB input,
this is often not feasible.

The *one-shot* APIs also perform all work as a single operation. So, if you
feed it large input, it could take a long time for the function to return.

The streaming APIs do not have the limitations of the simple API. But the
price you pay for this flexibility is that they are more complex than a
single function call.

The streaming APIs put the caller in control of compression and decompression
behavior by allowing them to directly control either the input or output side
of the operation.

With the *streaming input*, *compressor*, and *decompressor* APIs, the caller
has full control over the input to the compression or decompression stream.
They can directly choose when new data is operated on.

With the *streaming ouput* APIs, the caller has full control over the output
of the compression or decompression stream. It can choose when to receive
new data.

When using the *streaming* APIs that operate on file-like or stream objects,
it is important to consider what happens in that object when I/O is requested.
There is potential for long pauses as data is read or written from the
underlying stream (say from interacting with a filesystem or network). This
could add considerable overhead.


It is important to have a basic understanding of how Zstandard works in order
to optimally use this library. In addition, there are some low-level Python
concepts that are worth explaining to aid understanding. This section aims to
provide that knowledge.

Zstandard Frames and Compression Format

Compressed zstandard data almost always exists within a container called a
*frame*. (For the technically curious, see the
`specification <>_.)

The frame contains a header and optional trailer. The header contains a
magic number to self-identify as a zstd frame and a description of the
compressed data that follows.

Among other things, the frame *optionally* contains the size of the
decompressed data the frame represents, a 32-bit checksum of the
decompressed data (to facilitate verification during decompression),
and the ID of the dictionary used to compress the data.

Storing the original content size in the frame (``write_content_size=True``
to ``ZstdCompressor``) is important for performance in some scenarios. Having
the decompressed size stored there (or storing it elsewhere) allows
decompression to perform a single memory allocation that is exactly sized to
the output. This is faster than continuously growing a memory buffer to hold

Compression and Decompression Contexts

In order to perform a compression or decompression operation with the zstd
C API, you need what's called a *context*. A context essentially holds
configuration and state for a compression or decompression operation. For
example, a compression context holds the configured compression level.

Contexts can be reused for multiple operations. Since creating and
destroying contexts is not free, there are performance advantages to
reusing contexts.

The ``ZstdCompressor`` and ``ZstdDecompressor`` types are essentially
wrappers around these contexts in the zstd C API.

One-shot And Streaming Operations

A compression or decompression operation can either be performed as a
single *one-shot* operation or as a continuous *streaming* operation.

In one-shot mode (the *simple* APIs provided by the Python interface),
**all** input is handed to the compressor or decompressor as a single buffer
and **all** output is returned as a single buffer.

In streaming mode, input is delivered to the compressor or decompressor as
a series of chunks via multiple function calls. Likewise, output is
obtained in chunks as well.

Streaming operations require an additional *stream* object to be created
to track the operation. These are logical extensions of *context*

There are advantages and disadvantages to each mode of operation. There
are scenarios where certain modes can't be used. See the
``Choosing an API`` section for more.


A compression *dictionary* is essentially data used to seed the compressor
state so it can achieve better compression. The idea is that if you are
compressing a lot of similar pieces of data (e.g. JSON documents or anything
sharing similar structure), then you can find common patterns across multiple
objects then leverage those common patterns during compression and
decompression operations to achieve better compression ratios.

Dictionary compression is generally only useful for small inputs - data no
larger than a few kilobytes. The upper bound on this range is highly dependent
on the input data and the dictionary.

Python Buffer Protocol

Many functions in the library operate on objects that implement Python's
`buffer protocol <>`_.

The *buffer protocol* is an internal implementation detail of a Python
type that allows instances of that type (objects) to be exposed as a raw
pointer (or buffer) in the C API. In other words, it allows objects to be
exposed as an array of bytes.

From the perspective of the C API, objects implementing the *buffer protocol*
all look the same: they are just a pointer to a memory address of a defined
length. This allows the C API to be largely type agnostic when accessing their
data. This allows custom types to be passed in without first converting them
to a specific type.

Many Python types implement the buffer protocol. These include ``bytes``
(``str`` on Python 2), ``bytearray``, ``array.array``, ``io.BytesIO``,
``mmap.mmap``, and ``memoryview``.

``python-zstandard`` APIs that accept objects conforming to the buffer
protocol require that the buffer is *C contiguous* and has a single
dimension (``ndim==1``). This is usually the case. An example of where it
is not is a Numpy matrix type.

Requiring Output Sizes for Non-Streaming Decompression APIs

Non-streaming decompression APIs require that either the output size is
explicitly defined (either in the zstd frame header or passed into the
function) or that a max output size is specified. This restriction is for
your safety.

The *one-shot* decompression APIs store the decompressed result in a
single buffer. This means that a buffer needs to be pre-allocated to hold
the result. If the decompressed size is not known, then there is no universal
good default size to use. Any default will fail or will be highly sub-optimal
in some scenarios (it will either be too small or will put stress on the
memory allocator to allocate a too large block).

A *helpful* API may retry decompression with buffers of increasing size.
While useful, there are obvious performance disadvantages, namely redoing
decompression N times until it works. In addition, there is a security
concern. Say the input came from highly compressible data, like 1 GB of the
same byte value. The output size could be several magnitudes larger than the
input size. An input of <100KB could decompress to >1GB. Without a bounds
restriction on the decompressed size, certain inputs could exhaust all system
memory. That's not good and is why the maximum output size is limited.

Note on Zstandard's *Experimental* API

Many of the Zstandard APIs used by this module are marked as *experimental*
within the Zstandard project. This includes a large number of useful
features, such as compression and frame parameters and parts of dictionary

It is unclear how Zstandard's C API will evolve over time, especially with
regards to this *experimental* functionality. We will try to maintain
backwards compatibility at the Python API level. However, we cannot
guarantee this for things not under our control.

Since a copy of the Zstandard source code is distributed with this
module and since we compile against it, the behavior of a specific
version of this module should be constant for all of time. So if you
pin the version of this module used in your projects (which is a Python
best practice), you should be buffered from unwanted future changes.


A lot of time has been invested into this project by the author.

If you find this project useful and would like to thank the author for
their work, consider donating some money. Any amount is appreciated.

.. image::
:alt: Donate via PayPal

.. |ci-status| image::

.. |win-ci-status| image::
:alt: Windows build status
Release History

Release History

This version
History Node


History Node


History Node


History Node


History Node


History Node


History Node


History Node


History Node


History Node


History Node


History Node


History Node


History Node


History Node


History Node


History Node


Download Files

Supported By

WebFaction WebFaction Technical Writing Elastic Elastic Search Pingdom Pingdom Monitoring Dyn Dyn DNS Sentry Sentry Error Logging CloudAMQP CloudAMQP RabbitMQ Heroku Heroku PaaS Kabu Creative Kabu Creative UX & Design Fastly Fastly CDN DigiCert DigiCert EV Certificate Rackspace Rackspace Cloud Servers DreamHost DreamHost Log Hosting