next-download-manager

A small async download manager for Bottles Next written in Rust, built with reqwest and tokio.

It provides a simple high-level API to schedule downloads with a global concurrency limit, per-download progress and events, cooperative cancellation, retries with exponential backoff, and basic probing (HEAD) to discover remote metadata.

Key features

• Global concurrency control across all downloads.
• Per-download handle that:
  • Implements Future returning a DownloadResult.
  • Exposes a progress stream and per-download events.
  • Allows cooperative cancellation.
• Global event broadcast for monitoring all downloads.
• Retry policy with exponential backoff for retryable network errors.
• Optional HEAD probe to surface remote metadata (content-length, ETag, etc.).
• Removal of partial files on cancellation / failure.

Events & Progress

The crate emits typed Event values for lifecycle events:

• Queued, Probed, Started, Retrying, Completed, Failed, Cancelled.

You can subscribe to global events via DownloadManager::subscribe() (a broadcast::Receiver) or obtain a fallible-safe stream via DownloadManager::events() which drops lagged messages.

Per-download events are available from the Download handle via download.events().

Progress sampling is implemented with a Progress struct exposing:

• bytes_downloaded(), total_bytes()
• percent() (Option)
• instantaneous_bps, ema_bps and eta() when available

Progress is emitted on a watch channel and is sampled to avoid flooding consumers.

Cancellation

• Per-download: call download.cancel() on the returned handle or call the manager's cancel(id).
• Global: manager.cancel_all() cancels all downloads.
• Shutdown: manager.shutdown().await will cancel in-flight work and wait for worker tasks to finish.

Cancellation is cooperative: it aborts the HTTP request and removes the partial file.

Retry behavior

• Network errors are classified and retryable errors are retried with exponential backoff (base 1s, max 10s).
• The per-request retry limit defaults to 3 and is configurable via the RequestBuilder (retries()).
• Non-retryable errors (I/O, cancellation, some client errors) fail immediately.

Logging / Tracing

The crate uses tracing for structured logs. Attach a subscriber in your binary (for example tracing_subscriber::fmt::init() or set up a more advanced subscriber) to see info/debug traces for lifecycle events, queueing, retries, and errors.

License

This crate is GPL-3.0 (see LICENSE).