A Python module for fast random access to zstd-compressed files without full decompression.

IndexedZstdFile implements Python's io.BufferedReader interface, so it works as a drop-in replacement for open() on .zst files — supporting seek(), read(), readline(), tell(), and context managers.

Under the hood it uses libzstd-seek to build a jump table of frame boundaries, enabling O(1) seeking to any position in multi-frame archives.

This project is based on indexed_bzip2 to target zstd specifically.

Zstd files can contain multiple independently compressed frames. indexed_zstd scans frame boundaries on first access and builds an in-memory jump table that maps uncompressed offsets to compressed positions. When you seek(), only the relevant frame is decompressed.

Seeking within a frame is emulated by decompressing from the frame start, so the more frames your archive has, the faster random access will be.

To create multi-frame archives use t2sz or the --stream-size option of the zstd CLI.

Pre-built wheels are available for Linux, macOS, and Windows:

pip install indexed-zstd

If no wheel is available for your platform, pip will build from source automatically. In that case you need zstd development headers and a C++17 compiler:

# Debian/Ubuntu
sudo apt install libzstd-dev

# macOS
brew install zstd

conda install -c conda-forge indexed_zstd

yay -S python-indexed-zstd

from indexed_zstd import IndexedZstdFile

with IndexedZstdFile("example.zst") as f:
    f.seek(1024)
    data = f.read(256)
    print(f.tell())       # 1280
    print(f.seekable())   # True

from indexed_zstd import IndexedZstdFile

with IndexedZstdFile("logfile.zst") as f:
    for line in f:
        if b"ERROR" in line:
            print(line.decode())

import os
from indexed_zstd import IndexedZstdFile

fd = os.open("example.zst", os.O_RDONLY)
with IndexedZstdFile(fd) as f:
    data = f.read()

from indexed_zstd import IndexedZstdFile

with IndexedZstdFile("example.zst") as f:
    print(f.size())              # uncompressed size in bytes
    print(f.number_of_frames())  # number of zstd frames
    print(f.is_multiframe())     # True if more than one frame
    print(f.block_offsets())     # {compressed_offset: uncompressed_offset, ...}

IndexedZstdFile inherits from io.BufferedReader and adds:

All standard io.BufferedReader methods are available: read(), readline(), readlines(), seek(), tell(), seekable(), readable(), fileno(), close(), etc.

The test suite requires gen_seekable (built from the bundled submodule) and covers API, error paths, round-trip, reference, and heavy-data scenarios.

# Build gen_seekable from the submodule
cmake -S indexed_zstd/libzstd-seek -B indexed_zstd/libzstd-seek/build -DBUILD_TESTS=ON
cmake --build indexed_zstd/libzstd-seek/build --target gen_seekable

# Add it to PATH
export PATH="$PWD/indexed_zstd/libzstd-seek/build/tests:$PATH"

# Run the standard test suite (111 tests)
python -m pytest tests/ -v -m "not heavy and not reference"

Additional test categories (optional):

# Run everything including reference and heavy tests
python -m pytest tests/ -v

Requires a C++17 compiler, Cython, and platform-specific zstd libraries.

# Clone with submodules (includes libzstd-seek)
git clone --recurse-submodules https://github.com/martinellimarco/indexed_zstd.git
cd indexed_zstd
pip install cython setuptools

sudo apt install libzstd-dev    # Debian/Ubuntu
pip install .

brew install zstd
pip install .

Requires Visual Studio Build Tools with the C++ workload.

python libzstd/_get_zstd.py    # downloads zstd headers and DLL
pip install .

MIT