watchftp 0.0.1

Simple, modern and fast FTP watching

watchftp

Async, polling-based watcher for FTP/SFTP endpoints with optional Redis-backed persistence and change streams.

Overview

watchftp scans remote trees using protocol-native metadata (MLSD for FTP, AsyncSSH for SFTP), fingerprints directories, and emits batched change events (create/modify/delete/heuristic move). The engine is fully async, tuned for large trees (250k+ files), and can persist its state plus event stream in Redis for multi-consumer durability.

Key capabilities:

• Async core: aioftp + AsyncSSH connectors with bounded concurrency.
• Pattern-aware traversal: glob/regex includes/excludes compiled via wcmatch fallback.
• Directory fingerprints: Merkle-style digests to skip cold subtrees and detect changes efficiently.
• Redis integration: Optional persistence and Streams; tenants are fully namespaced (wf:{tenant}:{server}:...). Defaults to redis://127.0.0.1:6379/0 when Redis streaming is enabled without a URL so local testing “just works”.
• Config via Pydantic: Load settings from environment variables using pydantic-settings, including scale presets, watch-root limits (with -1 meaning unlimited), and security toggles.

Getting Started

Requirements

• Python 3.12+
• FTP/SFTP endpoints to monitor
• Optional Redis (local development uses redis://127.0.0.1:6379/0)

Install with uv (recommended) or pip:

uv pip install -e .
# or
pip install -e .

Quick start

import asyncio
from watchftp import WatchConfig, WatchPath, Watcher, tune_for_scale

cfg = WatchConfig(
    tenant_id="demo-tenant",
    protocol="sftp",
    host="sftp.internal",
    username="svc",
    private_key="~/.ssh/id_ed25519",
    watch_paths=[WatchPath(root="/incoming", include=["**/*.csv"], exclude=["**/tmp/**"])]
)
tune_for_scale(cfg, "medium")

async def main():
    async with Watcher(cfg) as watcher:
        async for batch in watcher.watch():
            for event in batch:
                print(event.type, event.path)

asyncio.run(main())

Enable Redis streams and persistence by setting cfg.event_bus = "redis_stream" (or WATCHFTP_EVENT_BUS=redis_stream). If no redis_url is provided, the watcher will connect to redis://127.0.0.1:6379/0 automatically for local development.

Documentation

We publish docs with MkDocs Material. To preview locally:

uv pip install mkdocs mkdocs-material
mkdocs serve

Changes are automatically built and deployed to GitHub Pages whenever pyproject.toml's version is bumped on main (see .github/workflows/docs.yml).

Configuration

The main knobs live on WatchConfig (Pydantic model):

• tenant_id: required for namespacing Redis state.
• watch_paths: list of WatchPath roots with include/exclude patterns (glob or regex).
• max_watch_roots: cap per process (-1 for unlimited).
• Concurrency knobs (poll_interval, max_concurrency, per-protocol limits).
• Security toggles (ftps_verify_cert, sftp_known_hosts, allowed ciphers).
• Persistence settings (redis_url, redis_ttl_seconds, event_bus).

Environment variables are loaded via WatcherSettings with the prefix WATCHFTP_ (e.g., WATCHFTP_TENANT, WATCHFTP_REDIS_URL). Nested values use __ (double underscore) separators.

• configure_logging(level="INFO", structured=True): optional helper to emit JSON logs via logging. Call once in your app (from watchftp import configure_logging).
• ensure_metrics_server(port=9464): starts the Prometheus HTTP endpoint (uses prometheus-client).

Example:

from watchftp import configure_logging, ensure_metrics_server

configure_logging("INFO")
ensure_metrics_server()

Project Roadmap

Testing

• Unit tests (Redis + helpers): python -m unittest

• Integration tests (FTP/SFTP against pyftpdlib + AsyncSSH) require Python 3.12+ and can be run with pytest -m integration. They start ephemeral servers locally.

• M0 (current): Config scaffolding, scheduler skeleton, Redis index + stream sink, unit tests.

• M1: Real FTP/SFTP traversal, diff engine polish, adaptive scheduler knobs.

• M2: Advanced Redis-backed recovery, rename heuristics, metrics exposure, docs/examples.

See docs/project.md for the full PRD, architecture, and sizing matrix.