0.8.0 - 2026-11-12
A network release. Trackers now see every network the host has, not just the primary address, and the DHT only keeps nodes that answer us. The download speed readout follows the bytes that actually arrive, and two resource leaks are closed: uTP handshakes that never finished, and a storage crash on a truncated file.
Added
- BEP 7 multi-homed tracker announce. A host with more than one network — a
second NIC, tethering, IPv4 plus IPv6 on different interfaces —
now announces to each HTTP and UDP tracker from every eligible local address,
each on a socket bound to that address, so the tracker lists and serves peers
for each of them. Before, only one primary IPv4 and one primary IPv6 address
announced. Eligible means the first usable address of each interface that is
up, not loopback, point-to-point (a VPN or Tailscale tunnel binds and
connects anywhere, but its datagrams leave from the default interface),
link-local, ULA/site-local or IPv4-mapped, at most four per
family (
config :elixir_torrent, :tracker, max_sources_per_family: 4;multihomed_announce: falserestores the old behaviour). A host with one address per family announces exactly as before. - The added announces run concurrently (
max_concurrent_source_announces: 4), and one that has not answered 15 s (source_grace_ms) after another source succeeded is abandoned, so a source the tracker cannot reach never holds back the peers the others found. Each distinct source address carries thestarted/completed/stoppedevent once; tier failover (BEP 12) and retry/cooldown (BEP 31) still see one merged answer per tracker. - Every announce logs one
[tracker_announce] source_resultdebug line per source: tracker, event, destination, family, role (primary/secondary), bound source and result. DHT.routing_stats/0returns the DHT routing table size per family and the inbound-candidate counters below; the same numbers are logged every minute as a[dht] routing_statsdebug line.
Changed
- BEP 5: a DHT node seen only as the source of an inbound query no longer enters the routing table on that query alone. Its query is answered as before (the answer goes out first, the check never delays it); then it is pinged at the address the query came from, and only a valid answer to that ping inserts it, through the usual BEP 42 check and ping-before-evict. A node already in the table at that address refreshes as before. A node id claimed from a new address is checked the same way, and its silence no longer counts against the node that really holds that id. The check is bounded: at most 64 waiting candidates, 2 per source IP, 30 s each, 16 new pings per second, and never more than a quarter of the DHT's in-flight query slots.
Performance
- A seeding torrent dials gently: one batch of at most 8 endpoints at the normal cadence, never to a peer already known to be a seed. Before, the seed's zero download speed counted as starving, so every seeding torrent kept three batches and about 40 endpoints in flight on a 1 s tick for hours (14 seeding torrents: dial tasks 39 to 3, uTP connections 132 to 6, processes 1280 to 830).
Fixed
- The download speed readout counted verified pieces, so with 1 MiB pieces a 64 KB/s torrent read anywhere from 12 to 157 KB/s, and one slower than a piece per window read 0 while bytes were arriving. It now measures received bytes over a 20 s sliding window.
- Every downloading torrent counted as slow in the connection manager (its speed check compared a float in bytes per millisecond against an integer), so discovery escalated below 50 peers and the snub path ran above 12 peers at any speed. It now compares bytes per second against the 32 KiB/s threshold.
- A read past the end of a truncated file crashed the piece process that owns the file handles, and every other read in flight with it (five crashes in one instant on a live node). The read now fails cleanly and the upload simply does not answer that request; the peer asks again or moves on.
- An inbound uTP handshake whose peer sent only the SYN stayed in
syn_recvforever, with no owner and nothing to time it out (63 of them, the oldest 7.7 h old, on a live node). It is now closed after 30 s of silence. - Point-to-point interfaces (VPN or Tailscale tunnels) are no longer used as extra BEP 7 announce sources: their datagrams leave through the default interface, so every added announce timed out.
ssl_verify_funis updated to 1.1.9 (EEF-CVE-2026-117666), and hackney, quic, h2 and webtransport to their current releases (clears fivehex.auditadvisories).- BEP 15
connection_ids are cached per local address as well as local port. Two announce sockets bound to different addresses can get the same ephemeral port, and trackers tie the id to the client's address. - Peers a tracker returns at any of our IPv4 addresses on our listen port are recognised as ourselves, not only the primary one.
0.7.0 - 2026-10-18
A magnet release. When a magnet's metadata arrived, the download threw away every peer the metadata phase had found and started from nothing, and penalised the peer it had just been talking to. It now starts from that pool. Discovery and the uTP transport also stop wasting slots: one tracker's rule no longer throttles the others, and a uTP connection whose dial was abandoned no longer lives forever.
Added
- A magnet's download starts from the peers its metadata phase found: the ones it
was connected to first, then
x.pepeers, then the rest of the discovered pool, IPv6 first (up to 200). Before, it started with no peers and waited for its first tracker answer. A private torrent (BEP 27) still carries over onlyx.pepeers.
Fixed
- Stopping a magnet's metadata phase no longer counts its own connections as churn. Every peer it had just connected to was blocked from re-dialling for five minutes, exactly the peers that had proven reachable.
- One tracker's
min intervalno longer throttles every tracker of the torrent. A single tracker answeringmin interval: 1800held a torrent with one peer to one announce round per 30 minutes across all its trackers; the value now applies to that tracker only, as BEP 3 defines it. - A tracker's own failure reason (for example a private tracker asking for a re-downloaded torrent) is logged as a warning once per tracker and reason, not on every retry.
- A uTP connection whose dialler died before handing it on (a dial abandoned at its timeout, a failed handshake) is closed with it. Before, an idle connected link with nothing in flight was never timed out and stayed for the rest of the session.
Changed
- The HTTP transport chain moves to hackney 4.8.5 (h2 0.12.4, quic 2.1.1, webtransport 0.4.8).
0.6.8 - 2026-10-02
A pipeline release. A log audit of a six-magnet session found that the swarm was never the limit: the engine kept its fastest peer idle at every piece boundary, let silent peers hold slots and blocks for minutes, and made every magnet wait about a minute before its first announce. Most fixes are scheduling and bookkeeping, and several of them replace a guess in the log with evidence.
Fixed
- A magnet no longer waits ~55 s before its first tracker announce. The BEP 3
event=stoppedgoodbye for the metadata-fetch phase ran synchronously, in waves of 8 trackers each waiting for its slowest dead one, before the download could start. It is now a supervised background task (bounded per tracker, sent exactly once instead of twice) andmetadata_okto first announce takes milliseconds. - The first announce of a new torrent no longer returns
nil. It fired before the torrent's model row existed,Announceread the missing reply as{:unexpected_reply, nil}, and every tracker of the torrent was benched for five minutes at once. It is now a classified:model_not_readythat retries after 5 s and logs at debug. - A bare
:badargfrom an HTTP announce is classified. Binding our source address per family (BEP 7) against a host that only has an A record madegen_tcpraise:badarg; the reason is now{:bind_family_mismatch, family}or{:connect_badarg, family}, the other family's real answer is no longer masked, gzip bodies are decoded, and a non-dictionary or out-of-range reply becomes{:bad_response, _}instead of crashing the request task. - A piece that fails SHA-1 verification now carries evidence. Every accepted
block is recorded with its sender and CRC, so a failure reports whether the
bytes on disk match what the peer sent (
verdict=received) or not (verdict=local, our bug). A peer is blamed only when it was the single contributor, never on a multi-source piece. Endgame no longer overwrites neighbouring blocks with a duplicate that is not exactly our sub-piece. - A piece whose last file edge fell exactly one byte before the piece end lost
that file from its slice list, so the read returned
:errorand the caller crashed with a badmatch (FileHandle.files_for_index/4). - A silent uTP peer is dropped after 6 consecutive retransmission timeouts or 60 s of silence, instead of holding its slot for 3-5 minutes while the RTO backed off to 60 s. Any inbound packet resets the counter, so a lossy but alive link survives.
Changed
- Each peer's request window follows its measured download rate (about 3 s of data, between 4 and 64 blocks) instead of a fixed 64. One peer can no longer take a whole piece while others could share it. After a request timeout the peer holds a single request until it delivers, and its blocks go to whoever asks next.
- A peer that exhausts its pinned piece re-pins to another active piece with unclaimed blocks in the same step and keeps its requests in flight, so the pipeline stays full across piece boundaries. Before, it waited for a new piece to start or for the 2 s reconcile tick, and a fast peer sat idle most of the time.
- A trickling pin (a few blocks delivered, nothing for 30 s) is released, and a silent seed that holds our requests is snubbed once at least four other peers are delivering. The only working source is never evicted.
0.6.7 - 2026-09-14
A measurement release. Every fix here is a case where the engine was doing something wrong and the instruments said it was fine — a tracker retried forever because its failure had no reason attached, healthy peers marked unreachable because a local error was counted as a network one, and a download speed of 0 B/s reported on a torrent moving at 100 KB/s.
Fixed
- A tracker whose hostname no longer resolves is dropped for the session instead
of being announced to forever. Hackney raises
:badargout of its connect path for a name with no A and no AAAA record rather than returning:nxdomain, so the HTTP announce returned a reason with noretry_inand got the default retry interval —tracker.openbittorrent.com, defunct, was re-announced six times an hour across four torrents. The UDP side never had this problem because it resolves up front and answersretry_in: "never". The:badargcatch now resolves the host itself and only writes a tracker off when DNS genuinely has nothing; other sources of:badargkeep the old opaque reason, because a permanent disable is too destructive to apply on a guess. - Announce failures name the torrent and the announce URL. The line read
request failure reason: <term>with neither, which made the most actionable message in the log useless: a tracker's own bencoded failure reason asks the user to re-add a torrent it never identified. Two classification gaps went with it — HTTPoison surfaces hackney's connect timeout as{:timeout, {:gen_statem, :call, [pid, :connect, 8000]}}, which the connect-timeout case never matched (42 of 89 warnings in a 21-minute window), and HTTP 4xx/5xx was not classified at all even though a 403 or 521 from a public announce-list entry is a dead tracker that BEP 12 already fails over. :add_peer_failedno longer counts as a peer being unreachable. It is emitted only after TCP connect and the full BEP 3 handshake succeed — we already hold the peer's id — and means the supervisor could not start, which is a local fault. It was depressing the measured per-family dial yield that drives the address-family throttle and writingDialBackoffrows toward the hard-fail threshold: one torrent held 61 of its 62 known endpoints sticky-blocked while connected to 4 peers. The catch-all also hid the outcome behind it — peer supervisors register under{peer_id, hash}, so a farm handing one id out from many IPs loses the race atstart_child, which is:already_connected.- The download speed readout no longer reports 0 B/s on a moving torrent.
Torrent.Modeldifferenceddownloadedover its 5 s tick, but that counter advances only when a whole piece verifies, so the sample was quantized to piece size: at 55 KB/s with 1 MiB pieces three ticks in four read exactly 0.0. Two live torrents reported 0 B/s while gaining a combined 196 KB/s, and the ETA derived from it was:infinitythroughout. The rate is now averaged over a 60 s window, held while a window containing progress matures, and clamped bypiece_length / elapsedonly when nothing has arrived at all. A torrent with nothing completed for 10 minutes reports 0. This is an improvement, not a cure: the source is still piece-granular, so a swarm delivering in bursts still oscillates, and a torrent slower than one piece per window reads 0 until its first piece lands. - A packed
valuesstring in aget_peersresponse is no longer mistaken for an IPv6 peer. 18 is a multiple of 6, so a string carrying three compact IPv4 peers has exactly the size of one compact IPv6 peer, and the decoder tested the IPv6 stride first — inventing an IPv6 address out of three IPv4 ones. A phantom endpoint is not harmless: it consumes a dial slot, can never connect, and its failure is recorded against the IPv6 family, which is what drives the per-family dial throttle — so the guess teaches the engine that IPv6 does not work. BEP 32 runs the IPv6 DHT as a separate DHT, which makes the transport the authority: the family the response arrived on now decides, and length remains the tie-breaker only where that family's own unit does not divide the string. - A disk error while serving a BEP 52 hash request returns an error instead of
raising.
Merkle.leaf_range_response_from_disk/7documents{:error, term()}and handles a failed:file.open/2that way, but the per-leaf reads underneath it pattern-matched{:ok, block} = :file.pread/3, so an I/O error — or a file truncated between the stat that producedfile_lengthand the read — raisedMatchErrorfrom the middle of the function.HashServecatches that and answershash_reject, which is correct on the wire but reached by the wrong path; any other caller got an exception for a disk condition.:file.pread/3also answers a bare:eofrather than an error tuple, which is now distinguished from the legitimate padded-leaf case that reads no bytes at all.
Changed
- The HTTP stack behind tracker announces and BEP 19 web seeds moved up: hackney 4.7.2 → 4.7.4, h2 0.11.0 → 0.12.0, webtransport 0.4.4 → 0.4.5, quic 1.8.0 → 1.8.2.
0.6.6 - 2026-08-27
A swarm-health release. Under CGNAT, where a torrent runs on a handful of peers, three separate mechanisms were removing peers we could not afford to lose or parking them on work they were not allowed to do — torrents sat at 99% for hours with unchoked idle peers and unclaimed blocks side by side.
Fixed
- Peers are no longer banned for correctly answering a request we withdrew.
A cancel is not atomic: BEP 6 obliges the peer to answer every request exactly
once, so a block cancelled, choked or repinned away still arrives one RTT
later. Both the piece and the reject landed in the "not outstanding" branch,
which disconnected the peer and blacklisted its ID across every torrent — 46
bans from
handle_pieceand 29 fromhandle_rejectin five minutes, hitting the Fast clients hardest precisely because they are the ones obliged to reject. Withdrawn blocks are now remembered in a bounded per-block count (one answer per request, so arrival order stops mattering) evicted by generation rather than cleared wholesale. An answer matching neither set is wasteful, not malicious: it ends the connection at 512 blocks without banning. - Fast-extension messages sent without the extension advertised, and frames that
stall half-way, no longer ban the peer. 71 of 98 disconnects in one
three-minute window were mainstream qBittorrent and Transmission builds.
have_all/have_nonenow go through the normal bitfield handler — a seeder recorded as having nothing is a peer we can never request from — and the advisory messages are logged and ignored. A truncated frame is a congested path, so the connection still drops, but the ID is not blacklisted. - Peer bans expire. The blacklist was a
MapSetthat only grew, so one bad frame excluded a peer for the whole session across every torrent: 352 IDs in ten minutes while eight of nine torrents could not exceed three connections. Bans now last 30 minutes with a cap and a background sweep, and each records the rule that fired. - The dial layer no longer writes off a candidate pool a CGNAT host cannot
refill. A failure row was cleared only by
mark_productive/3, which asks whether an endpoint was useful rather than whether it is reachable — a leecher we keep choked never delivers bytes and so carried its failure history forever; registration now clears it, inbound peers included. Retention was also refreshed on every failure, so a row re-dialled once per 30 minutes never aged out and reached fail counts above 100; escalation now counts failures within a 10-minute streak window. Sticky blocks became last-resort undermin_countpressure (productive, then soft before sticky, then v6 before v4, then fewest failures) instead of refusing resurrection absolutely — 1151 of 1162 active blocks did, leaving a 50-endpoint request returning 3-8. - Endgame now applies to pieces that were already in flight when the torrent
crossed the threshold. A worker read the mode once at
State.download/3, so exactly the pieces endgame exists for ran without redundancy: one 1 MiB piece held a torrent at 99.939% for over an hour with all 27 remaining blocks in flight to a single peer that had logged 297 request timeouts on it, while three unchoked peers holding the piece had nothing they were allowed to ask for.:reconcile_pumpupgrades active workers level-triggered; the transition is idempotent and one-way and re-queues in-flight subpieces so endgame adds sources for a block rather than taking it from the peer already fetching it. - A pin is released from an unchoked peer that delivers nothing. The staleness
test required
choke_me, but a choked peer holds no requests at all — the harmful case was the one it skipped, an unchoked peer sitting on a full 64-request pipeline nobody else may touch. Two such peers re-requested their 64 blocks 821 and 622 times in five minutes while two pieces with every block unclaimed had no peer. Zero bytes now releases the pin either way, on a longer threshold when unchoked (60 s, more than a block timeout) than when choked. - A peer is no longer repinned off the piece it is fetching. "Drained" ignored who had claimed the blocks, so a peer that had claimed the rest of its piece made it look finished, was moved away, and the move cancelled the very requests that drained it — oscillating at the 2 s tick, 17 wire requests per block received. Draining now also requires no in-flight requests from that peer. Measured over seven minutes live: requests per block 17:1 → 1.1:1, re-request factor 5.66× → 1.08×, swarm 28 → 73 peers, 1.79 → 2.47 MB/s.
- Two endgame defects that parked torrents just short of completion: the
drained-pin probe branched on the torrent's mode rather than the worker's,
holding a pin for work that worker could never hand out, and
endgame_preferred_index/2destructured the peer key backwards ({_hash, peer_id}againstPeer.make_key/2's{id, hash}), hashing the torrent hash and so funnelling the entire swarm onto one index. - Outside endgame, a peer may leave a piece whose blocks are all claimed.
piece_has_waiting?/2counts blocks in flight to other peers, which a normal-mode worker can never hand out — 27 of 37 peers were pinned to 4 claimed pieces while 8 pieces with 49-64 free blocks had no peer at all. - A piece worker whose holders have all disconnected is released. The abort
check also required the swarm to be empty, so on any torrent with peers such a
worker held one of the
@max_parallel_piecesslots forever — 7 of 12 slots live, capping a torrent with 26 unchoked peers at 5 pieces in flight. - A piece failing its SHA-1 check now blames the peer that supplied it. The
worker previously just re-requested every block and could pick the same peer
again; one torrent sat at 99.84% for hours re-downloading one index. Blocks
now remember their source, a peer stops being asked for an index it has
corrupted, and is dropped after
@max_hash_failuresdistinct ruined pieces. - Writes to a peer that has stopped reading are bounded. A TCP socket defaults
to
send_timeout: :infinity, so a sender blocked inside:prim_inet.send/4never returned and its mailbox only grew — one held 20863 messages. Accepted and dialled sockets now take a 30 s send timeout and close on it. - The upload delivery task no longer crashes when a peer cannot take a block.
Only
:noprocwas tolerated, so a peer shutting down mid-call produced one crash report per in-flight block — 333 in fifteen minutes. BEP 3 permits simply not answering a request, so this is now a cancellation with one debug line; any other exit still crashes. - One pending pump wake per torrent. Every trigger — the 2 s reconcile tick,
each peer handoff, every
requests_are_dealtclosure — started its own self-rescheduling{:next_piece}chain, and they accumulated: 99 → ~2600 discovery dial cycles per minute over twelve minutes with no change in swarm size, until the tracker answered 403. - A peer disconnect logs why it ended.
Peer.Endpointsmonitors a supervisor withauto_shutdown: :any_significant, which exits with a bare:shutdownwhatever the child's reason was, so every disconnect readreason=:shutdown— useless for the one question worth asking, whether the peer left or we dropped it. A protocol error now also logs the rejected wire message, and the piece-bounds check logs the block alongside the torrent's geometry. - The background DHT metadata-lookup dedup table has a permanent owner. It was
created by whichever
Magnet.Fetcherran first and died with that torrent, so every later background task crashed withArgumentErroron insert; a supervised GenServer now owns it.
Performance
- Dial batches overlap instead of serialising on their slowest endpoint. One endpoint can hold a slot for the whole connect + handshake budget (measured successes at 16 s, 42 s and 42 s, worst case near 55 s), and until it resolved the manager would start nothing new for that torrent. That hurt worst where it mattered most: a starved torrent with an all-IPv4 queue is capped to a four-endpoint probe batch, so it made four attempts per minute against a ~1% CGNAT success rate. Batches are now bounded by endpoints in flight (40) and concurrent batches (3), with in-flight endpoints excluded from selection so two batches cannot dial the same peer.
0.6.5 - 2026-08-18
Added
- A built-in RC4 stream cipher (
Peer.MSE.RC4), selected automatically when the linked libcrypto does not expose RC4. OpenSSL 3 moved RC4 into the optionallegacyprovider and the Windows ERTS ships no legacy module, so:cryptoomits:rc4there — and because the whole MSE handshake is RC4-encrypted, including the exchange that selects a plaintext stream, such a node could not speak MSE at all and every peer dial failed.config :elixir_torrent, :mse_rc4accepts:auto(default),:crypto,:pureor:disabled.
Fixed
- Tracker announces now choose their source address by asking the kernel which
local address routes to the destination, instead of taking one from
Acceptor.all_global_ips/0. Windows enforces the strong host model and refuses a bind whose address does not belong to the routing interface, so with a VPN, a second NIC or tethering every announce failed with:eaddrnotavail. BEP 7 is unchanged where a routable global address exists; an announce refused for the bind is retried once unbound. Magnet.Fetcher.finalize_piece_attempt/1no longer raisesFunctionClauseErroron a non-retryable piece attempt, which previously took down the entire magnet fetch instead of that one attempt.:econnresetand:econnaborted— how a departing peer is reported on Windows — are now treated as retryable.- The DHT IPv4 socket logged
bind=<primary ip>while actually being bound to0.0.0.0.
Changed
- The test suite now runs on Windows. It previously failed before the first test
on a
/dev/nullpath, and then onlocalhostresolution: the suite forbids DNS on the wire, Unix answerslocalhostfrom/etc/hosts, and Windows ships that line commented out. Three platform assumptions in tests and three TOCTOU races were fixed alongside.
0.6.4 - 2026-08-12
Added
- Test coverage for paths that previously only ran against a live network or a hostile peer: address classification and the NAT-PMP codec, the
Peer.Controller.Statewire state machine including BEP 16 super-seed bookkeeping, the BEP 3 choke algorithm and swarm teardown, DHT (BEP 5) degraded-mode callbacks, ut_pex (BEP 11/40) ordering, tracker announce edge cases, BEP 9 metadata fetch, STUN parsing, dial backoff sweeps, and BEP 52 v2 padding-gap reads/writes.
Changed
Acceptor.compute_all_global_ips/1is now a public arity-1 function taking the:inet.getifaddrs()result, so the rules deciding which addresses may be advertised to trackers, DHT and PEX can be evaluated against a supplied interface list rather than the host's own.Magnet.Fetcher's BEP 5 retry windows (dht_retry_delays_ms/0,dht_deep_retry_delay_ms/0) are read through the existing fetcher config override instead of compile-time attributes, so the propagation waits are adjustable without changing the retry shape. Defaults unchanged.
0.6.3 - 2026-08-11
Fixed
- Release pipeline:
cosign sign-blobnow writes a--bundlefile instead of the deprecated--output-signature/--output-certificateflags, which current cosign silently ignores in favor of its new Sigstore bundle format (0.6.2's release signing failed withcreate bundle file: open : no such file or directoryas a result). The bundle is named*.tar.sigstore.jsonso OpenSSF Scorecard's Signed-Releases check recognizes it. No runtime/library changes.
0.6.2 - 2026-08-11
Fixed
- Release pipeline: publishing to Hex.pm is now idempotent (skips cleanly if the tag's version is already on Hex.pm instead of failing), and cosign's release-signing step retries on transient OIDC-token fetch failures instead of falling through to an interactive device-code flow that stalls CI. No runtime/library changes.
0.6.1 - 2026-08-11
Changed
- Runtime pin moved to Elixir 1.20.3 + OTP 29.0.5 (ERTS 17.0.5)
- Hackney 4.7.2, H2 0.11.0, QUIC 1.8.0, WebTransport 0.4.4 (HTTP stack behind HTTPoison, used by HTTP trackers, scrape, and BEP 19 web seeds)
0.6.0 - 2026-08-04
Added
- BEP 14 (LSD): Local Service Discovery via UDP multicast (
BT-SEARCH) - BEP 19 (WebSeed): HTTP
Rangeseeding fromurl-listviaTorrent.WebSeed - BEP 48: periodic tracker scrape; skip confirmed dead-swarm announce URLs
- BEP 52 (phases 1–2/6): detect v1/hybrid/v2, parse peers' v2 handshake capability; accept hybrid torrents/magnets; reject pure-v2 cleanly
- Lazy piece storage: on-demand piece read/write/verify workers with idle terminate
- Peer.ConnectionManager: outbound dial orchestration with per-family throttle
- Mid-download session checkpoint (bitfield + counters every 30s)
- Magnet
x.pepeers kept as live dial candidates after metadata fetch - Developer quality gate: warnings-as-errors compile, Dialyzer, and all enabled Credo checks at every priority via
mix quality
Changed
Magnet.Connectionsources its BEP 3 choke/unchoke/interested wire ids fromPeer.Constinstead of a local copy- CI publish workflow renamed to
build-and-publish, gated to runmix hex.publishon semver tag pushes
Fixed
- NAT-PMP and UPnP passive UDP receive handling now decodes the datagram payload instead of the source-address tuple
- Global 2 MiB recv-buffer ceiling now bounds every wire frame's declared length, not only the six previously-capped ids
- Per-connection stall watchdog disconnects peers that trickle a declared frame length in forever instead of completing it
- Declared bitfield/piece frame lengths capped ahead of buffering, closing a memory-amplification gap alongside the existing LTEP/BEP 52 caps
- Tracker
startedis now resent on the first announce of every session (e.g. after a cold restart), not only once per info hash ever
0.5.1 - 2026-07-22
Changed
- README: single-row shields.io badges at the top for release, Hex, HexDocs, changelog, source, Web UI, and macOS downloads
0.5.0 - 2026-07-22
Added
- BEP 9/10 (magnet + ut_metadata): magnet URI parsing,
download_magnet/1, LTEP extension protocol, metadata fetch/serve, magnet bootstrap and fetcher supervisors - BEP 11 (ut_pex): peer exchange encode/decode, ingest, and broadcast over LTEP
- BEP 32 (IPv6 DHT): separate IPv4/IPv6 routing tables, compact node/peer codecs, dual-stack KRPC and lookup
- BEP 42: secure DHT node ID generation from the local IPv4 address
- BEP 55 (ut_holepunch): rendezvous/connect/forward hole-punch messages over LTEP;
Peer.Holepunch.Storefor punch state - DHT (BEP 5): routing-table persistence across restarts, dual-stack socket wiring, runtime node-id path fix
- NAT traversal: NAT-PMP and UPnP IGD port-mapping clients, STUN mapping-behaviour detection,
NAT.PortMapperorchestrator - MSE/PE: Message Stream Encryption handshake and RC4 transport layer over
Peer.Transport - uTP (BEP 29): zombie-connection hardening and BEP 29 off-spec bug fixes
- Peer discovery/dial:
Peer.Endpoints, sticky dial backoff with churn and hard-fail escalation,Peer.DialStatsper address-family outcomes - Acceptor: global IPv4/IPv6 address helpers,
IpCacheperiodicgetifaddrssnapshot via:persistent_term, dual-stack listen logging; LTEP handshakes advertise listen port and global addresses - README: BEP 9, 10, 11, 32, 42, and 55 rows in Supported BEPs table;
download_magnet/1in public API table
Changed
- Peer I/O: MSE/PE encryption integrated into the peer transport and sender receive path
- Peer dial: immediate-drop churn backoff, endpoint PID lookup via
Peer.Endpoints.get_pid/3
Fixed
- DHT node ID persistence file path resolved at runtime instead of compile time
Notes
- Started at runtime (via
ElixirTorrentApplication): TCP acceptor (withIpCache),PeerDiscovery, magnet fetcher/bootstrap, andPeer.Holepunch.Store - Implemented and compiled, not started in supervision: mainline DHT server, uTP dispatcher,
NAT.PortMapper,Peer.Endpoints,Peer.DialBackoff, andPeer.DialStats— wiring deferred to a future release - Default peer acquisition remains TCP + tracker (plus magnet metadata bootstrap); DHT announce, uTP fallback dials, NAT mapping, and full hole-punch initiation require the dormant services above
0.4.0 - 2026-07-08
Added
- BEP 29 (uTP):
Peer.Transportover TCP and uTP; uTP packet stack with LEDBAT; connection lifecycle and dispatcher; dial backoff and endpoint registry; outbound TCP-first dial with uTP fallback; inbound uTP handshake entry - BEP 5 (DHT): Mainline DHT KRPC node (ping, find_node, get_peers, announce_peer); k-bucket routing table with bootstrap routers; iterative lookup and announce token workflow; DHT and uTP share one UDP socket via
DHT.send_udp/3 - README: BEP 5 (Mainline DHT) row in Supported BEPs table
Changed
- Peer I/O:
Peer.Senderis the sole socket owner overPeer.Transport;Peer.Receiverremoved; post-handshake socket handoff usesSender.activate/1
Fixed
- Torrent progress and
downloaded/leftcounters reconciled with the on-disk bitfield - Piece pipeline: idempotent piece statistics init, stale
:processingcleanup, exclude-aware picker - In-flight piece download recovery when subpiece requests stall (
waiting: []no longer finishes early) - Incoming piece block bounds validation;
Peer.log_id/1for safe timeout logging - Download scheduler: parallel piece picking with availability checks, resume handoff before controller scheduling, per-piece interested peer assignment
Notes
- uTP (BEP 29) and DHT (BEP 5) are implemented and compiled in, but runtime activation (supervision wiring) is deferred to a future release; the default peer-acquisition/download path remains TCP + tracker-based.
0.3.0 - 2026-06-28
Download location
ElixirTorrent.download/2accepts an optional:download_dirkeyword — base directory for downloaded files (defaults toFile.cwd!/0)- Session snapshots remain under
{File.cwd!()}/.elixir_torrent/state/regardless of:download_dir remove/2withdelete_data: trueremoves files from the torrent's download directory
File layout
- Multi-file torrents whose files share no common top-level folder are written under a directory named after the torrent (sanitized
info.name) - Multi-file torrents that already use a shared root folder (e.g.
dir/a.bin,dir/b.bin) keep their original paths - Single-file torrents are written directly into the download root
Docs
- README About section with project background
0.2.0 - 2026-06-11
Session persistence
- Saved session state under
.elixir_torrent/state/{info_hash}.term(relative toFile.cwd!/0) - On
download/1, an existing session is loaded and the bitfield is verified against disk before resuming remove/2deletes the session file;delete_data: truealso removes downloaded files
Graceful shutdown API
stop_and_serialize/1— stop piece downloads, disconnect peers (BEP 3), send trackerevent=stopped, persist session, then stop the torrent processstop_all_and_serialize/0— same for every active torrentlist/0— returns info hashes for all running torrent processes
Peer disconnect
- Peers receive BEP 3 cancel / not interested / choke before TCP connections close
- Used during shutdown so peers are notified cleanly
0.1.2 - 2026-06-09
ElixirTorrent.list_files/1— file list with per-file download progressElixirTorrent.remove/2— stop a torrent; optionaldelete_data: trueremoves files from disk- Requires Elixir
~> 1.20
0.1.1 - 2026-02-22
- Published HexDocs for the public API (
ElixirTorrent,Torrents) ElixirTorrent.stats/2documented as the preferred way to read runtime stats
0.1.0 - 2026-02-22
First public release — BitTorrent client engine publishable as a Hex dependency.
Public API
ElixirTorrent.download/1— start a download from a.torrentfile on diskElixirTorrent.stats/2— poll name, speeds, and progress for a running torrent- Escript entrypoint for CLI usage
Protocol & networking
- Peer wire protocol (BEP 3) with choking, rarest-first piece selection, and endgame mode
- Fast Extension (BEP 6) —
allowed_fast, reject on choked requests - IPv6 tracker peers (
peers6) and dual-stack listen sockets - Multi-homed HTTP announce (BEP 7) over IPv4 and IPv6
- HTTP and UDP trackers (BEP 15), compact peer lists (BEP 23)
Reliability
- Improved choke recovery, piece availability tracking, and tracker announce handling