Stream-based deduplication for repeating sequences
uniqseq identifies and removes repeated multi-record patterns from streaming data. Unlike traditional line-by-line deduplication tools, it detects when sequences of records repeat, where a record can be a line, a byte sequence, or any delimiter-separated unit.
Works with text streams (line-delimited, null-delimited, etc.) and binary streams (byte-delimited with any delimiter), processes data in a single pass, and maintains bounded memory usage.
# Input with repeated 3-line sequence
$ cat app.log
Starting process...
Loading config
Connecting to DB
Starting process...
Loading config
Connecting to DB
Done
# Remove duplicates (specify window size to match pattern length)
$ uniqseq --window-size 3 app.log
Starting process...
Loading config
Connecting to DB
Done- Sequence detection - Identifies repeating multi-record patterns
- Flexible delimiters - Text with any delimiter or byte streams
- Streaming architecture - Single-pass processing with real-time output
- Memory efficient - Bounded memory usage for unlimited input
- Pattern filtering - Selectively deduplicate with regex patterns
- Content transformation - Match on normalized content while preserving original output
- Python API & CLI - Use as a command-line tool or import as a library
- Sequence libraries - Save and reuse pattern libraries across sessions
brew tap jeffreyurban/uniqseq && brew install uniqseqHomebrew manages the Python dependency and provides easy updates via brew upgrade.
pipx install uniqseqpipx installs in an isolated environment with global CLI access. Works on macOS, Linux, and Windows. Update with pipx upgrade uniqseq.
pip install uniqseqUse pip if you want to use uniqseq as a library in your Python projects.
# Development installation
git clone https://github.com/JeffreyUrban/uniqseq
cd uniqseq
pip install -e ".[dev]"Requirements: Python 3.9+
# Basic usage (deduplicate 10-line sequences by default)
uniqseq app.log > clean.log
# Adjust window size for your data
uniqseq --window-size 3 build.log # 3-line patterns
uniqseq --window-size 5 errors.log # 5-line patterns
# Stream processing
tail -f app.log | uniqseq --window-size 5
# Ignore timestamps when comparing
uniqseq --skip-chars 24 timestamped.log
# Only deduplicate ERROR lines
uniqseq --track "^ERROR" app.log
# See what was removed
uniqseq --annotate app.logfrom uniqseq import UniqSeq
# Initialize with configuration
deduplicator = UniqSeq(
window_size=3,
skip_chars=0,
max_history=100000
)
# Process stream
with open("app.log") as infile, open("clean.log", "w") as outfile:
for line in infile:
deduplicator.process_line(line.rstrip("\n"), outfile)
deduplicator.flush(outfile)- Log processing - Clean repeated error traces, stack traces, debug output
- Build systems - Deduplicate compiler warnings, test failures
- Terminal sessions - Clean up verbose CLI output (from
scriptcommand) - Monitoring & alerting - Reduce noise from repeated alert patterns
- Data pipelines - Filter redundant multi-line records in ETL workflows
- Binary analysis - Deduplicate repeated byte sequences in memory dumps, network captures
uniqseq uses a sliding window with hash-based pattern detection:
- Buffering - Maintains a sliding window of N records
- Hashing - Computes a hash for each window position
- History tracking - Records which window patterns have been seen
- Sequence tracking - Tracks known multi-window sequences
- Matching - Compares current windows against history and known sequences
- Transformation - Optionally normalizes content for matching while preserving original data in output
Output is produced with minimal delay. When a window doesn't match any known pattern, the oldest buffered record is immediately emitted.
Read the full documentation at uniqseq.readthedocs.io
Key sections:
- Getting Started - Installation and quick start guide
- Use Cases - Real-world examples across different domains
- Guides - Window size selection, performance tips, common patterns
- Reference - Complete CLI and Python API documentation
# Clone repository
git clone https://github.com/JeffreyUrban/uniqseq.git
cd uniqseq
# Install development dependencies
pip install -e ".[dev]"
# Run tests
pytest
# Run with coverage
pytest --cov=uniqseq --cov-report=html- Time complexity: O(n) - linear with input size
- Space complexity: O(h + u×w) where h=history depth, u=known sequences, w=window size
- Throughput: Approximately constant records per second
- Memory: Bounded by configurable history depth
MIT License - See LICENSE file for details