pbz2 v0.2.2

Stream and parallel-process .bz2 files via pbzip2 (parallel bzip2).

Reads compressed files through a pbzip2 -dc subprocess for multi-core decompression — no temp files, no full decompression to disk — and falls back to the stdlib bz2 module when the pbzip2 binary is unavailable. Iterate raw lines, newline-aligned text chunks, or parsed JSONL records, or fan chunks out across a process pool for parallel parsing. Includes a CLI for quick inspection and a Python API for custom pipelines. Corrupt or truncated input raises instead of silently yielding partial data.

Project Structure

pbz2/
├── pbz2/                 # Python library
│   ├── reader.py         # Streaming readers (open_decompress, iter_*)
│   ├── parallel.py       # Process-pool chunk processing
│   └── cli.py            # Typer CLI commands
├── tests/                # Test suite
└── pyproject.toml        # Project configuration

Installation

uv add pbz2

From source:

git clone https://github.com/gitronald/pbz2.git
cd pbz2
uv sync

From a specific branch:

uv add git+https://github.com/gitronald/pbz2.git@dev

Install the pbzip2 binary for parallel decompression (optional — without it, reads fall back to single-threaded stdlib bz2):

sudo apt install pbzip2     # Debian/Ubuntu
brew install pbzip2         # macOS

Note: the parallel speedup only applies to files that were compressed with pbzip2. pbzip2 writes its output as multiple independent bzip2 streams that can be decompressed concurrently; a file compressed with standard bzip2 (or Python's bz2) is a single stream, which pbzip2 can only decompress on one core. Compress with pbzip2 data.json to get parallel decompression later.

CLI Commands

Quick inspection of .bz2 files from the shell:

# Count lines
pbz2 count data.json.bz2

# Print the first N lines
pbz2 head data.json.bz2 -n 5

Python API

Iterate

import pbz2

# Parsed JSON objects from a .json.bz2 file
for obj in pbz2.iter_jsonl("data.json.bz2"):
    ...

# Raw UTF-8 lines
for line in pbz2.iter_lines("data.txt.bz2"):
    ...

# Newline-aligned text chunks (useful for batched processing)
for chunk in pbz2.iter_chunks("data.txt.bz2"):
    ...

Parallel processing

process_parallel streams chunks of newline-terminated records through a worker pool. The worker function receives raw text chunks (so parsing happens in the worker, not the main process), and on_result runs in the main process to handle each result as it completes.

import json
import pbz2

def parse_chunk(chunk: str) -> list[dict]:
    # split on "\n" only -- str.splitlines() also breaks on U+2028/U+2029 etc.,
    # which can appear raw inside records and would shatter them
    return [json.loads(line) for line in chunk.split("\n") if line]

def save(records: list[dict]) -> None:
    ...  # write to db, file, etc.

pbz2.process_parallel(
    "data.json.bz2",
    worker_fn=parse_chunk,
    on_result=save,
    num_processes=8,
)

Reference

Common options

• num_processors — pbzip2 worker count (default: cpu_count - 1)
• bufsize_mb — OS pipe buffer between pbzip2 and Python (default: 32 MB)
• stream_buffer_mb — Python-side read chunk size (default: 4 MB)