admin_ekaii/orthrus
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions
Fork maintenu de Hydra — cross-seed multi-tracker + bench.db borné, suivi upstream release-by-release. AGPL-3.0.
248 commits 5 branches 0 tags 30 MiB
  • Go 61.3%
  • Rust 16.9%
  • JavaScript 13.8%
  • CSS 3.5%
  • HTML 3.1%
  • Other 1.3%
Find a file
Exact
Claude 29d877bcd8
Some checks failed
docker / build (amd64, linux/amd64, ubuntu-latest) (push) Failing after 22s
Details
docker / build (arm64, linux/arm64, ubuntu-24.04-arm) (push) Has been cancelled
Details
docker / merge (push) Has been cancelled
Details
docs(fork): document the full patch set, v2 hardening, inject-lane fixes, tiering pointer, backlog status 
FORK.md/PATCHES only covered the original two patches; eight more fork
commits had landed undocumented. Now:
- FORK.md: sections for the announce-engine hardening pass (9fa1c8c), the
  inject-lane correctness fixes (07106af, e9a85ee, 59a76bc, d39eb3a), the
  bench.db production note (per_torrent_snapshots plateau), and a pointer to
  the tiering-v35 branch + docs/TIERING.md.
- PATCHES: all 8 main commits + the 5 tiering-v35 commits (marked), with
  verified per-commit file lists.
- README banner: mentions the hardening + tiering branch.
- upstream-sync.sh: PATCHED_FILES list matches what our patches actually own.
- docs/CROSS-SEED-BACKLOG.md: dated Status section with an item-by-item
  shipped/open table, live corrections (c411 proxy for the torznab <link>
  HTML trap, /api URL, queuedUP resume timer), open-question resolutions.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-08-23 01:01:46 +02:00
.github feat(windows): update Hydra from the notification area 2026-08-14 19:15:35 +02:00
cmd feat(xseed): harden the multi-tracker announce engine for scale + stability 2026-08-15 20:18:04 +02:00
configs fix(bench): bound race_snapshots growth with a configurable [bench] section 2026-08-15 15:01:02 +02:00
deploy/xseed fix(xseed): c411 torznab url must end in /api for cross-seed (both /api and /api/torznab serve it) 2026-08-15 20:31:44 +02:00
docs docs(fork): document the full patch set, v2 hardening, inject-lane fixes, tiering pointer, backlog status 2026-08-23 01:01:46 +02:00
internal fix(announce): real bytes-left on immediate announce + hoard dispatch by ownership 2026-08-17 03:39:27 +02:00
packaging feat(windows): update Hydra from the notification area 2026-08-14 19:15:35 +02:00
proto Hydra — high-scale BitTorrent engine (initial public release) 2026-07-28 14:10:12 +02:00
scripts docs(fork): document the full patch set, v2 hardening, inject-lane fixes, tiering pointer, backlog status 2026-08-23 01:01:46 +02:00
third_party fix(windows): patch librqbit-dht for SIO_UDP_CONNRESET + tolerate os error 10054 in reader (DHT binds tokio UdpSocket directly, bypassing the dualstack fork) 2026-07-31 11:32:26 +02:00
typhon-engine fix(release): build the engine static against musl so the tarballs run anywhere 2026-08-13 13:45:26 +02:00
vendor Hydra — high-scale BitTorrent engine (initial public release) 2026-07-28 14:10:12 +02:00
web feat(hoard): force a download from the list, and count the filters against the filtering 2026-08-14 16:21:56 +02:00
.dockerignore feat(ui): column visibility, changelog tab, update badge 2026-07-29 22:40:37 +02:00
.gitignore chore(gitignore): actually ignore manual backup files 2026-08-10 18:12:19 +02:00
CHANGELOG.md feat(windows): update Hydra from the notification area 2026-08-14 19:15:35 +02:00
docker-compose.yml fix(engine): flush resume data on shutdown, and make the budget configurable 2026-08-13 11:30:07 +02:00
Dockerfile build: keep the Go build cache between image builds 2026-08-11 08:44:38 +02:00
embed.go feat: zero-setup start (find/generate config next to the binary) 2026-07-30 23:12:46 +02:00
entrypoint.sh feat(docker): PUID/PGID support so Hydra stops writing files as root [linux] 2026-08-07 18:41:43 +02:00
FORK.md docs(fork): document the full patch set, v2 hardening, inject-lane fixes, tiering pointer, backlog status 2026-08-23 01:01:46 +02:00
go.mod Hydra — high-scale BitTorrent engine (initial public release) 2026-07-28 14:10:12 +02:00
go.sum Hydra — high-scale BitTorrent engine (initial public release) 2026-07-28 14:10:12 +02:00
LICENSE Hydra — high-scale BitTorrent engine (initial public release) 2026-07-28 14:10:12 +02:00
PATCHES docs(fork): document the full patch set, v2 hardening, inject-lane fixes, tiering pointer, backlog status 2026-08-23 01:01:46 +02:00
README.md docs(fork): document the full patch set, v2 hardening, inject-lane fixes, tiering pointer, backlog status 2026-08-23 01:01:46 +02:00
SECURITY.md ci: prune container images from a checked plan after every release 2026-08-10 22:37:18 +02:00

README.md

Orthrus

Orthrus is a maintained fork of Hydra. It adds real multi-tracker cross-seed, an announce engine hardened for tens-of-thousands-of-torrents scale, and a bounded bench.db; the tiering-v35 branch (running in production) adds NVMe hot-storage tiering and announce-schedule persistence. It tracks upstream release-by-release. See FORK.md for the full patch set and how upstream syncs are reviewed, and docs/TIERING.md (on tiering-v35) for the tiering architecture and ops runbook. Everything below is upstream's documentation.

Hydra

A self-hosted BitTorrent daemon built for scale and seeding: a Go control plane driving a purpose-built Rust engine ("Typhon"). Hydra holds 100k+ torrents in a single instance, exposes a live web UI, a native REST API, and a qBittorrent-compatible shim so your existing *arr / autobrr / cross-seed setup just works.

Status: Hydra is being opened up from a private homelab project. It is used in production but some rough edges remain; issues and PRs welcome.

Why Hydra?

  • Two engine roles. A race session (aggressive, low-latency, for hot downloads) and a hoard session (upload-optimized, for long-term seeding), each tuned independently. You can also run any number of extra engines.
  • Scales. Tens of thousands of torrents per instance, with a push-based (SSE) UI that streams the torrent list instead of shipping giant REST blobs.
  • Flexible networking. Direct, SOCKS5 egress, a lightweight PROXY-v2 relay (change your seedbox's public IP without a VPN tunnel), or gluetun with hot listen-port rebind (no restart when your VPN's forwarded port rotates).
  • Distributed. Run engines across several machines and aggregate them behind one front; route new torrents per-category with placement/strategy and a save-path per agent.
  • Data-aware adds. Add a torrent whose data is already on disk (a re-add, a cross-seed, or a half-finished download) and Hydra hash-checks what's there instead of blindly re-downloading over it: verified pieces are kept and served, the rest is fetched.
  • Drop-in. A qBittorrent v2 API shim means autobrr, Sonarr/Radarr, cross-seed, etc. talk to Hydra unchanged.

Screenshots

Overview — live dashboard: global up/down, seeding/leeching counts, per-session throughput, all streamed over SSE.

Overview

Race — per-torrent detail with live peer speed and progress timelines for the hot download you're racing.

Race timeline

Hoard — the long-term seeding set: tens of thousands of torrents in one virtualized, push-updated table.

Hoard

Agents — run engines across several machines and manage them from one front: status, free space, and per-engine roles.

Agents

Benchmark — built-in throughput history so you can see exactly what your box sustains.

Benchmark

Give Hydra time to stop

On SIGTERM Hydra saves its store and then asks each engine to flush its resume data. Docker kills a container ten seconds after SIGTERM by default, which is not enough to get through both engines: the flush is cut short and the next start re-checks pieces that were already complete.

The bundled docker-compose.yml sets stop_grace_period: 30s for you. If you run Hydra with a plain docker run, pass the same budget yourself:

docker run --stop-timeout 30 ...      # and: docker stop -t 30 hydra

Each engine gets ten seconds of that budget, and the two are stopped one after the other. If you hold several hundred thousand torrents the sweep takes longer -- raise HYDRA_STOP_TIMEOUT (e.g. 45s) and keep the supervisor's grace period above twice that.

Documentation

Install steps, architecture, every networking mode and the edge cases live in the Wiki:

API reference: docs/API.md. Companion VPS relay: hydra-relay.

License

Hydra is licensed under the GNU Affero General Public License v3.0 (AGPL-3.0). See LICENSE. In short: you're free to run, study, modify, and share it, including self-hosting a modified version, but if you offer a modified Hydra to others over a network, you must make your source available under the same license.