compression.zstd — Compression compatible with the Zstandard format

Added in version 3.14.

Source code: Lib/compression/zstd/__init__.py


This module provides classes and functions for compressing and decompressing data using the Zstandard (or zstd) compression algorithm. The zstd manual describes Zstandard as “a fast lossless compression algorithm, targeting real-time compression scenarios at zlib-level and better compression ratios.” Also included is a file interface that supports reading and writing the contents of .zst files created by the zstd utility, as well as raw zstd compressed streams.

The compression.zstd module contains:

Exceptions

exception compression.zstd.ZstdError

This exception is raised when an error occurs during compression or decompression, or while initializing the (de)compressor state.

Reading and writing compressed files

compression.zstd.open(file, /, mode='rb', *, level=None, options=None, zstd_dict=None, encoding=None, errors=None, newline=None)

Open a Zstandard-compressed file in binary or text mode, returning a file object.

The file argument can be either a file name (given as a str, bytes or path-like object), in which case the named file is opened, or it can be an existing file object to read from or write to.

The mode argument can be either 'rb' for reading (default), 'wb' for overwriting, 'ab' for appending, or 'xb' for exclusive creation. These can equivalently be given as 'r', 'w', 'a', and 'x' respectively. You may also open in text mode with 'rt', 'wt', 'at', and 'xt' respectively.

When reading, the options argument can be a dictionary providing advanced decompression parameters; see DecompressionParameter for detailed information about supported parameters. The zstd_dict argument is a ZstdDict instance to be used during decompression. When reading, if the level argument is not None, a TypeError will be raised.

When writing, the options argument can be a dictionary providing advanced decompression parameters; see CompressionParameter for detailed information about supported parameters. The level argument is the compression level to use when writing compressed data. Only one of level or options may be non-None. The zstd_dict argument is a ZstdDict instance to be used during compression.

In binary mode, this function is equivalent to the ZstdFile constructor: ZstdFile(file, mode, ...). In this case, the encoding, errors, and newline parameters must not be provided.

In text mode, a ZstdFile object is created, and wrapped in an io.TextIOWrapper instance with the specified encoding, error handling behavior, and line endings.

class compression.zstd.ZstdFile(file, /, mode='rb', *, level=None, options=None, zstd_dict=None)

Open a Zstandard-compressed file in binary mode.

A ZstdFile can wrap an already-open file object, or operate directly on a named file. The file argument specifies either the file object to wrap, or the name of the file to open (as a str, bytes or path-like object). If wrapping an existing file object, the wrapped file will not be closed when the ZstdFile is closed.

The mode argument can be either 'rb' for reading (default), 'wb' for overwriting, 'xb' for exclusive creation, or 'ab' for appending. These can equivalently be given as 'r', 'w', 'x' and 'a' respectively.

If file is a file object (rather than an actual file name), a mode of 'w' does not truncate the file, and is instead equivalent to 'a'.

When reading, the options argument can be a dictionary providing advanced decompression parameters; see DecompressionParameter for detailed information about supported parameters. The zstd_dict argument is a ZstdDict instance to be used during decompression. When reading, if the level argument is not None, a TypeError will be raised.

When writing, the options argument can be a dictionary providing advanced decompression parameters; see CompressionParameter for detailed information about supported parameters. The level argument is the compression level to use when writing compressed data. Only one of level or options may be passed. The zstd_dict argument is a ZstdDict instance to be used during compression.

ZstdFile supports all the members specified by io.BufferedIOBase, except for detach() and truncate(). Iteration and the with statement are supported.

The following method and attributes are also provided:

peek(size=-1)

Return buffered data without advancing the file position. At least one byte of data will be returned, unless EOF has been reached. The exact number of bytes returned is unspecified (the size argument is ignored).

Note

While calling peek() does not change the file position of the ZstdFile, it may change the position of the underlying file object (for example, if the ZstdFile was constructed by passing a file object for file).

mode

'rb' for reading and 'wb' for writing.

name

The name of the Zstandard file. Equivalent to the name attribute of the underlying file object.

Compressing and decompressing data in memory

compression.zstd.compress(data, level=None, options=None, zstd_dict=None)

Compress data (a bytes-like object), returning the compressed data as a bytes object.

The level argument is an integer controlling the level of compression. level is an alternative to setting CompressionParameter.compression_level in options. Use bounds() on compression_level to get the values that can be passed for level. If advanced compression options are needed, the level argument must be omitted and in the options dictionary the CompressionParameter.compression_level parameter should be set.

The options argument is a Python dictionary containing advanced compression parameters. The valid keys and values for compression parameters are documented as part of the CompressionParameter documentation.

The zstd_dict argument is an instance of ZstdDict containing trained data to improve compression efficiency. The function train_dict() can be used to generate a Zstandard dictionary.

compression.zstd.decompress(data, zstd_dict=None, options=None)

Decompress data (a bytes-like object), returning the uncompressed data as a bytes object.

The options argument is a Python dictionary containing advanced decompression parameters. The valid keys and values for compression parameters are documented as part of the DecompressionParameter documentation.

The zstd_dict argument is an instance of ZstdDict containing trained data used during compression. This must be the same Zstandard dictionary used during compression.

If data is the concatenation of multiple distinct compressed frames, decompress all of these frames, and return the concatenation of the results.

class compression.zstd.ZstdCompressor(level=None, options=None, zstd_dict=None)

Create a compressor object, which can be used to compress data incrementally.

For a more convenient way of compressing a single chunk of data, see the module-level function compress().

The level argument is an integer controlling the level of compression. level is an alternative to setting CompressionParameter.compression_level in options. Use bounds() on compression_level to get the values that can be passed for level. If advanced compression options are needed, the level argument must be omitted and in the options dictionary the CompressionParameter.compression_level parameter should be set.

The options argument is a Python dictionary containing advanced compression parameters. The valid keys and values for compression parameters are documented as part of the CompressionParameter documentation.

The zstd_dict argument is an optional instance of ZstdDict containing trained data to improve compression efficiency. The function train_dict() can be used to generate a Zstandard dictionary.

compress(data, mode=ZstdCompressor.CONTINUE)

Compress data (a bytes-like object), returning a bytes object with compressed data if possible, or otherwise an empty bytes object. Some of data may be buffered internally, for use in later calls to compress() and flush(). The returned data should be concatenated with the output of any previous calls to compress().

The mode argument is a ZstdCompressor attribute, either CONTINUE, FLUSH_BLOCK, or FLUSH_FRAME.

When all data has been provided to the compressor, call the flush() method to finish the compression process. If compress() is called with mode set to FLUSH_FRAME, flush() should not be called, as it would write out a new empty frame.

flush(mode=ZstdCompressor.FLUSH_FRAME)

Finish the compression process, returning a bytes object containing any data stored in the compressor’s internal buffers.

The mode argument is a ZstdCompressor attribute, either FLUSH_BLOCK, or FLUSH_FRAME.

CONTINUE

Collect more data for compression, which may or may not generate output immediately. This mode optimizes the compression ratio by maximizing the amount of data per block and frame.

FLUSH_BLOCK

Complete and write a block to the data stream. The data returned so far can be immediately decompressed. Past data can still be referenced in future blocks generated by calls to compress(), improving compression.

FLUSH_FRAME

Complete and write out a frame. Future data provided to compress() will be written into a new frame and cannot reference past data.

class compression.zstd.ZstdDecompressor(zstd_dict=None, options=None)

Create a decompressor object, which can be used to decompress data incrementally.

For a more convenient way of decompressing an entire compressed stream at once, see the module-level function decompress().

The options argument is a Python dictionary containing advanced decompression parameters. The valid keys and values for compression parameters are documented as part of the DecompressionParameter documentation.

The zstd_dict argument is an instance of ZstdDict containing trained data used during compression. This must be the same Zstandard dictionary used during compression.

Note

This class does not transparently handle inputs containing multiple compressed frames, unlike the decompress() function and ZstdFile class. To decompress a multi-frame input, you should use decompress(), ZstdFile if working with a file object, or multiple ZstdDecompressor instances.

decompress(data, max_length=-1)

Decompress data (a bytes-like object), returning uncompressed data as bytes. Some of data may be buffered internally, for use in later calls to decompress(). The returned data should be concatenated with the output of any previous calls to decompress().

If max_length is non-negative, the method returns at most max_length bytes of decompressed data. If this limit is reached and further output can be produced, the needs_input attribute will be set to False. In this case, the next call to decompress() may provide data as b'' to obtain more of the output.

If all of the input data was decompressed and returned (either because this was less than max_length bytes, or because max_length was negative), the needs_input attribute will be set to True.

Attempting to decompress data after the end of a frame will raise a ZstdError. Any data found after the end of the frame is ignored and saved in the unused_data attribute.

eof

True if the end-of-stream marker has been reached.

unused_data

Data found after the end of the compressed stream.

Before the end of the stream is reached, this will be b''.

needs_input

False if the decompress() method can provide more decompressed data before requiring new compressed input.

Zstandard dictionaries

compression.zstd.train_dict(samples, dict_size)

Train a Zstandard dictionary, returning a ZstdDict instance. Zstandard dictionaries enable more efficient compression of smaller sizes of data, which is traditionally difficult to compress due to less repetition. If you are compressing multiple similar groups of data (such as similar files), Zstandard dictionaries can improve compression ratios and speed significantly.

The samples argument (an iterable of bytes objects), is the population of samples used to train the Zstandard dictionary.

The dict_size argument, an integer, is the maximum size (in bytes) the Zstandard dictionary should be. The Zstandard documentation suggests an absolute maximum of no more than 100 KB, but the maximum can often be smaller depending on the data. Larger dictionaries generally slow down compression, but improve compression ratios. Smaller dictionaries lead to faster compression, but reduce the compression ratio.

compression.zstd.finalize_dict(zstd_dict, /, samples, dict_size, level)

An advanced function for converting a “raw content” Zstandard dictionary into a regular Zstandard dictionary. “Raw content” dictionaries are a sequence of bytes that do not need to follow the structure of a normal Zstandard dictionary.

The zstd_dict argument is a ZstdDict instance with the dict_content containing the raw dictionary contents.

The samples argument (an iterable of bytes objects), contains sample data for generating the Zstandard dictionary.

The dict_size argument, an integer, is the maximum size (in bytes) the Zstandard dictionary should be. See train_dict() for suggestions on the maximum dictionary size.

The level argument (an integer) is the compression level expected to be passed to the compressors using this dictionary. The dictionary information varies for each compression level, so tuning for the proper compression level can make compression more efficient.

class compression.zstd.ZstdDict(dict_content, /, *, is_raw=False)

A wrapper around Zstandard dictionaries. Dictionaries can be used to improve the compression of many small chunks of data. Use train_dict() if you need to train a new dictionary from sample data.

The dict_content argument (a bytes-like object), is the already trained dictionary information.

The is_raw argument, a boolean, is an advanced parameter controlling the meaning of dict_content. True means dict_content is a “raw content” dictionary, without any format restrictions. False means dict_content is an ordinary Zstandard dictionary, created from Zstandard functions, for example, train_dict() or the external zstd CLI.

When passing a ZstdDict to a function, the as_digested_dict and as_undigested_dict attributes can control how the dictionary is loaded by passing them as the zstd_dict argument, for example, compress(data, zstd_dict=zd.as_digested_dict). Digesting a dictionary is a costly operation that occurs when loading a Zstandard dictionary. When making multiple calls to compression or decompression, passing a digested dictionary will reduce the overhead of loading the dictionary.

Difference for compression

Digested dictionary

Undigested dictionary

Advanced parameters of the compressor which may be overridden by the dictionary’s parameters

window_log, hash_log, chain_log, search_log, min_match, target_length, strategy, enable_long_distance_matching, ldm_hash_log, ldm_min_match, ldm_bucket_size_log, ldm_hash_rate_log, and some non-public parameters.

None

ZstdDict internally caches the dictionary

Yes. It’s faster when loading a digested dictionary again with the same compression level.

No. If you wish to load an undigested dictionary multiple times, consider reusing a compressor object.

If passing a ZstdDict without any attribute, an undigested dictionary is passed by default when compressing and a digested dictionary is generated if necessary and passed by default when decompressing.

dict_content

The content of the Zstandard dictionary, a bytes object. It’s the same as the dict_content argument in the __init__ method. It can be used with other programs, such as the zstd CLI program.

dict_id

Identifier of the Zstandard dictionary, a non-negative int value.

Non-zero means the dictionary is ordinary, created by Zstandard functions and following the Zstandard format.

0 means a “raw content” dictionary, free of any format restriction, used for advanced users.

Note

The meaning of 0 for ZstdDict.dict_id is different from the dictionary_id attribute to the get_frame_info() function.

as_digested_dict

Load as a digested dictionary.

as_undigested_dict

Load as an undigested dictionary.

Advanced parameter control

class compression.zstd.CompressionParameter

An IntEnum containing the advanced compression parameter keys that can be used when compressing data.

The bounds() method can be used on any attribute to get the valid values for that parameter.

Parameters are optional; any omitted parameter will have it’s value selected automatically.

Example getting the lower and upper bound of compression_level:

lower, upper = CompressionParameter.compression_level.bounds()

Example setting the window_log to the maximum size:

_lower, upper = CompressionParameter.window_log.bounds()
options = {CompressionParameter.window_log: upper}
compress(b'venezuelan beaver cheese', options=options)
bounds()

Return the tuple of int bounds, (lower, upper), of a compression parameter. This method should be called on the attribute you wish to retrieve the bounds of. For example, to get the valid values for compression_level, one may check the result of CompressionParameter.compression_level.bounds().

Both the lower and upper bounds are inclusive.

compression_level

A high-level means of setting other compression parameters that affect the speed and ratio of compressing data. Setting the level to zero uses COMPRESSION_LEVEL_DEFAULT.

window_log

Maximum allowed back-reference distance the compressor can use when compressing data, expressed as power of two, 1 << window_log bytes. This parameter greatly influences the memory usage of compression. Higher values require more memory but gain better compression values.

A value of zero causes the value to be selected automatically.

hash_log

Size of the initial probe table, as a power of two. The resulting memory usage is 1 << (hash_log+2) bytes. Larger tables improve compression ratio of strategies <= dfast, and improve compression speed of strategies > dfast.

A value of zero causes the value to be selected automatically.

chain_log

Size of the multi-probe search table, as a power of two. The resulting memory usage is 1 << (chain_log+2) bytes. Larger tables result in better and slower compression. This parameter has no effect for the fast strategy. It’s still useful when using dfast strategy, in which case it defines a secondary probe table.

A value of zero causes the value to be selected automatically.

search_log

Number of search attempts, as a power of two. More attempts result in better and slower compression. This parameter is useless for fast and dfast strategies.

A value of zero causes the value to be selected automatically.

min_match

Minimum size of searched matches. Larger values increase compression and decompression speed, but decrease ratio. Note that Zstandard can still find matches of smaller size, it just tweaks its search algorithm to look for this size and larger. For all strategies < btopt, the effective minimum is 4; for all strategies > fast, the effective maximum is 6.

A value of zero causes the value to be selected automatically.

target_length

The impact of this field depends on the selected Strategy.

For strategies btopt, btultra and btultra2, the value is the length of a match considered “good enough” to stop searching. Larger values make compression ratios better, but compresses slower.

For strategy fast, it is the distance between match sampling. Larger values make compression faster, but with a worse compression ratio.

A value of zero causes the value to be selected automatically.

strategy

The higher the value of selected strategy, the more complex the compression technique used by zstd, resulting in higher compression ratios but slower compression.

See also

Strategy

enable_long_distance_matching

Long distance matching can be used to improve compression for large inputs by finding large matches at greater distances. It increases memory usage and window size.

True or 1 enable long distance matching while False or 0 disable it.

Enabling this parameter increases default window_log to 128 MiB except when expressly set to a different value. This setting is enabled by default if window_log >= 128 MiB and the compression strategy >= btopt (compression level 16+).

ldm_hash_log

Size of the table for long distance matching, as a power of two. Larger values increase memory usage and compression ratio, but decrease compression speed.

A value of zero causes the value to be selected automatically.

ldm_min_match

Minimum match size for long distance matcher. Larger or too small values can often decrease the compression ratio.

A value of zero causes the value to be selected automatically.

ldm_bucket_size_log

Log size of each bucket in the long distance matcher hash table for collision resolution. Larger values improve collision resolution but decrease compression speed.

A value of zero causes the value to be selected automatically.

ldm_hash_rate_log

Frequency of inserting/looking up entries into the long distance matcher hash table. Larger values improve compression speed. Deviating far from the default value will likely result in a compression ratio decrease.

A value of zero causes the value to be selected automatically.

content_size_flag

Write the size of the data to be compressed into the Zstandard frame header when known prior to compressing.

This flag only takes effect under the following two scenarios:

All other compression calls may not write the size information into the frame header.

True or 1 enable the content size flag while False or 0 disable it.

checksum_flag

A four-byte checksum using XXHash64 of the uncompressed content is written at the end of each frame. Zstandard’s decompression code verifies the checksum. If there is a mismatch a ZstdError exception is raised.

True or 1 enable checksum generation while False or 0 disable it.

dict_id_flag

When compressing with a ZstdDict, the dictionary’s ID is written into the frame header.

True or 1 enable storing the dictionary ID while False or 0 disable it.

nb_workers

Select how many threads will be spawned to compress in parallel. When nb_workers > 0, enables multi-threaded compression, a value of 1 means “one-thread multi-threaded mode”. More workers improve speed, but also increase memory usage and slightly reduce compression ratio.

A value of zero disables multi-threading.

job_size

Size of a compression job, in bytes. This value is enforced only when nb_workers >= 1. Each compression job is completed in parallel, so this value can indirectly impact the number of active threads.

A value of zero causes the value to be selected automatically.

overlap_log

Sets how much data is reloaded from previous jobs (threads) for new jobs to be used by the look behind window during compression. This value is only used when nb_workers >= 1. Acceptable values vary from 0 to 9.

  • 0 means dynamically set the overlap amount

  • 1 means no overlap

  • 9 means use a full window size from the previous job

Each increment halves/doubles the overlap size. “8” means an overlap of window_size/2, “7” means an overlap of window_size/4, etc.

class compression.zstd.DecompressionParameter

An IntEnum containing the advanced decompression parameter keys that can be used when decompressing data. Parameters are optional; any omitted parameter will have it’s value selected automatically.

The bounds() method can be used on any attribute to get the valid values for that parameter.

Example setting the window_log_max to the maximum size:

data = compress(b'Some very long buffer of bytes...')

_lower, upper = DecompressionParameter.window_log_max.bounds()

options = {DecompressionParameter.window_log_max: upper}
decompress(data, options=options)
bounds()

Return the tuple of int bounds, (lower, upper), of a decompression parameter. This method should be called on the attribute you wish to retrieve the bounds of.

Both the lower and upper bounds are inclusive.

window_log_max

The base-two logarithm of the maximum size of the window used during decompression. This can be useful to limit the amount of memory used when decompressing data. A larger maximum window size leads to faster decompression.

A value of zero causes the value to be selected automatically.

class compression.zstd.Strategy

An IntEnum containing strategies for compression. Higher-numbered strategies correspond to more complex and slower compression.

Note

The values of attributes of Strategy are not necessarily stable across zstd versions. Only the ordering of the attributes may be relied upon. The attributes are listed below in order.

The following strategies are available:

fast
dfast
greedy
lazy
lazy2
btlazy2
btopt
btultra
btultra2

Miscellaneous

compression.zstd.get_frame_info(frame_buffer)

Retrieve a FrameInfo object containing metadata about a Zstandard frame. Frames contain metadata related to the compressed data they hold.

class compression.zstd.FrameInfo

Metadata related to a Zstandard frame.

decompressed_size

The size of the decompressed contents of the frame.

dictionary_id

An integer representing the Zstandard dictionary ID needed for decompressing the frame. 0 means the dictionary ID was not recorded in the frame header. This may mean that a Zstandard dictionary is not needed, or that the ID of a required dictionary was not recorded.

compression.zstd.COMPRESSION_LEVEL_DEFAULT

The default compression level for Zstandard: 3.

compression.zstd.zstd_version_info

Version number of the runtime zstd library as a tuple of integers (major, minor, release).

Examples

Reading in a compressed file:

from compression import zstd

with zstd.open("file.zst") as f:
    file_content = f.read()

Creating a compressed file:

from compression import zstd

data = b"Insert Data Here"
with zstd.open("file.zst", "w") as f:
    f.write(data)

Compressing data in memory:

from compression import zstd

data_in = b"Insert Data Here"
data_out = zstd.compress(data_in)

Incremental compression:

from compression import zstd

comp = zstd.ZstdCompressor()
out1 = comp.compress(b"Some data\n")
out2 = comp.compress(b"Another piece of data\n")
out3 = comp.compress(b"Even more data\n")
out4 = comp.flush()
# Concatenate all the partial results:
result = b"".join([out1, out2, out3, out4])

Writing compressed data to an already-open file:

from compression import zstd

with open("myfile", "wb") as f:
    f.write(b"This data will not be compressed\n")
    with zstd.open(f, "w") as zstf:
        zstf.write(b"This *will* be compressed\n")
    f.write(b"Not compressed\n")

Creating a compressed file using compression parameters:

from compression import zstd

options = {
   zstd.CompressionParameter.checksum_flag: 1
}
with zstd.open("file.zst", "w", options=options) as f:
    f.write(b"Mind if I squeeze in?")