diff --git a/SWIPs/assets/swip-60/bps.proto b/SWIPs/assets/swip-60/bps.proto new file mode 100644 index 00000000..7384870b --- /dev/null +++ b/SWIPs/assets/swip-60/bps.proto @@ -0,0 +1,141 @@ +// Broadcast Pub/Sub (BPS) — protocol messages and types. +// Spec: SWIP-60 (../../swip-60.md). +// +// Revision 2 (2026-08-05), after review on PR #104: Connect split into +// Open/Subscribe (subscribers carry no cohort metadata), broker capacity +// removed from CohortSpec (it is broker-side policy, not a cohort parameter), +// Ping dropped (liveness/RTT are transport concerns), and every frame carries +// the full SOC (no handshake/data split). Field numbers renumbered — the +// draft has no deployed compatibility surface. +// +// Enum zero values (*_UNSPECIFIED): proto3 requires a zero value; it is +// deliberately NOT a legitimate wire value. It exists so that an unset field +// is detectable and no implementation can silently rely on a default. +// Receivers MUST reject messages carrying it. +// +// The singlehop (depth = 1) subset is concrete; multihop control-plane +// messages are reserved. Implementation groundwork: bee PR #5435 +// (hand-rolled byte framing with the same semantics). + +syntax = "proto3"; +package bps; + +option go_package = "github.com/ethersphere/bee/v2/pkg/bps/pb"; + +// --------------------------------------------------------------------------- +// Cohort genesis — the primitive decisions whose combinations are the "modes" +// --------------------------------------------------------------------------- + +// What the topic binds to (see SWIP-60: binding semantics). +enum TopicBinding { + TOPIC_BINDING_UNSPECIFIED = 0; // invalid on the wire (see header note) + ANCHOR = 1; // topic = full SOC/GSOC address; dedup on the wrapped CAC + SOC_ID = 2; // topic = SOC id; any owner with PO(addr, anchor) >= PO_MIN + OWNER = 3; // topic = SOC owner; any id with PO(addr, anchor) >= PO_MIN (MIC) + FEED_TOPIC = 4; // id = keccak256(topic ‖ index); graffiti MIC / feed streams +} + +// Who may author. +enum PublisherRegime { + PUBLISHER_REGIME_UNSPECIFIED = 0; // invalid on the wire (see header note) + EXPLICIT_SINGLE = 1; // opener is the sole publisher (live streaming) + EXPLICIT_LIST = 2; // set fixed at genesis: admin + publisher_list + // (dynamic grants/revocations: later revision) + IMPLICIT = 3; // authorship implied by the topic binding (PO constraint) + ALL = 4; // every peer publishes (gossipsub-equivalent cohort) +} + +// Fixed by the cohort's opener; immutable for the cohort's lifetime. +// NOTE: broker capacity is NOT a cohort parameter — a cohort cannot dictate a +// remote node's connection count. Each broker enforces its own per-topic +// stream limit and answers FULL when it is exhausted. +// NOTE: the proximity constraint for implicit bindings is a protocol +// constant, PO_MIN = 16 — not a cohort parameter (a proto3 unset uint32 is +// indistinguishable from 0, which would silently disable the constraint; +// and no use case varies it). +message CohortSpec { + bytes topic = 1; // 32 bytes, meaning per binding + TopicBinding binding = 2; + PublisherRegime publishers = 3; + bool history = 4; // deliver matching chunks from the local store + bytes admin = 5; // 20-byte eth address; set iff EXPLICIT_* + repeated bytes publisher_list = 6; // 20-byte eth addresses, excl. admin; + // set iff EXPLICIT_LIST + reserved 7; // was po_min — now protocol constant PO_MIN + bool closed = 8; // no audience: subscribers restricted to the publishers +} + +// --------------------------------------------------------------------------- +// Stream establishment, stream name "pubsub/1.0.0" — one stream per (peer, topic). +// The first message on a fresh stream is Open (fixes a new cohort) or +// Subscribe (joins an existing one); the broker answers with Ack. +// --------------------------------------------------------------------------- + +// Opener -> broker: the one peer that fixes the cohort. +message Open { + CohortSpec cohort = 1; + PublisherAuth auth = 2; // present iff the opener publishes (explicit regimes) +} + +// Joiner -> broker: names the topic — nothing more. Subscribers carry no +// cohort metadata; auth is present iff the joiner publishes (publishers +// connect directly to the broker). +message Subscribe { + bytes topic = 1; // 32 bytes + PublisherAuth auth = 2; // present iff publisher +} + +message PublisherAuth { + bytes owner = 1; // 20-byte eth address of the SOC owner key + bytes id = 2; // 32-byte SOC id, when the binding fixes it +} + +// Broker -> peer, answering Open or Subscribe. The echoed CohortSpec lets a +// subscriber verify every message end-to-end against the topic binding. +message Ack { + Status status = 1; + CohortSpec cohort = 2; // set iff status == OK +} + +enum Status { + STATUS_UNSPECIFIED = 0; // invalid on the wire (see header note) + OK = 1; + FULL = 2; // broker at its per-topic capacity; + // a singlehop broker refuses — nothing else + UNKNOWN_TOPIC = 3; // Subscribe for a topic the broker does not serve + REJECTED = 4; // e.g. publisher not on the list, invalid auth, + // non-publisher Subscribe on a closed cohort +} + +// --------------------------------------------------------------------------- +// Messages — SOC-only is a protocol feature +// --------------------------------------------------------------------------- + +// A full single-owner chunk in transit. Every frame is self-contained: no +// per-stream handshake state, and no format change if the stream model +// evolves (e.g. topic-muxed streams later). +message Soc { + bytes id = 1; // 32 bytes + bytes owner = 2; // 20 bytes (recoverable from signature; explicit for cheap filtering) + bytes signature = 3; // 65 bytes + bytes span = 4; // 8 bytes LE + bytes payload = 5; // wrapped-CAC data, <= 4096 bytes +} + +// Publisher -> broker. +message Publish { + Soc soc = 1; +} + +// Broker -> subscriber. +message Broadcast { + oneof frame { + Soc soc = 1; + // 2–15 reserved: multihop control plane (Beacon, Reparent, Expect, + // DcutrSignal, SwapProposal) — named to fix intent, not final. + } +} + +// Keepalive / RTT: none at the BPS level. Liveness is the transport's job +// (libp2p), and latency metrics for reorganisation policies (SWATCH) are +// sourced there as well. diff --git a/SWIPs/swip-60.md b/SWIPs/swip-60.md new file mode 100644 index 00000000..f6f3f836 --- /dev/null +++ b/SWIPs/swip-60.md @@ -0,0 +1,344 @@ +--- +SWIP: 60 +title: BPS singlehop — brokered broadcast pub/sub, base protocol +author: Viktor Trón (@zelig), Viktor Tóth (@nugaon) +discussions-to: https://discord.gg/Q6BvSkCv +status: Draft +type: Standards Track (Networking) +created: 2026-08-03 +--- + + + +- **Business line**: real-time topic streams for dApps without storing chunks or polling — + enough on its own for small closed collaboration cohorts (collaborative remix editing, a + strudel livecoding session, multiparty games) and basic single-publisher limited-audience + live streaming. +- **Dev line**: implement one libp2p protocol (`pubsub/1.0.0`, messages in + [bps.proto](assets/swip-60/bps.proto)) plus a WebSocket bridge on the Bee API; done when + a broker, publishers and subscribers interoperate per the conformance section. Groundwork + exists in bee [#5435](https://github.com/ethersphere/bee/pull/5435). +- Bandwidth-incentive integration is a separate SWIP (bps-bw-incentives). +- Broker discovery integration is from a separate SWIP (bps-broker-discovery, building on + [SWIP-59 MEX](https://github.com/ethersphere/SWIPs/pull/103)). + +## Simple Summary + +A real-time messaging protocol: WebSocket clients publish and subscribe to topic streams +through Bee nodes. One full node per topic acts as **broker**, re-broadcasting each message +over direct, long-lived p2p streams to a capacity-bounded set of connected peers. Messages are +single-owner chunks, so every subscriber verifies authorship end-to-end; the broker can +withhold, never forge. + +## Motivation + +Swarm's event primitives (GSOC, PSS) require full-node operation; light clients can only +poll storage. BPS singlehop is the smallest protocol that fixes this: one broker, direct +streams, authenticated messages, an explicit capacity bound. Everything larger — multihop +trees, adaptive reorganisation, incentives, discovery — is layered on top by later SWIPs +without changing the semantics defined here. + +## Specification + +### The contract + +Per topic-cohort: + +- messages come from **publishers, and publishers only**; +- they arrive at **all subscribers**. + +### Cohort genesis: the parameters + +A cohort is fully described by a `CohortSpec` ([bps.proto](assets/swip-60/bps.proto)), +fixed the moment the first peer contacts a BPS-speaking full node with a topic. There is +no mode enum; **modes are combinations of these parameters**. + +| parameter | values | meaning | +|---|---|---| +| `topic` | 32 bytes | interpreted per `binding` | +| `binding` | `ANCHOR` / `SOC_ID` / `OWNER` / `FEED_TOPIC` | what the topic binds to; fixes which SOCs qualify as messages and the dedup rule | +| `publishers` | `EXPLICIT_SINGLE` / `EXPLICIT_LIST` / `IMPLICIT` / `ALL` | who may author | +| `admin` + `publisher_list` | eth addresses | set iff explicit publishers; with `EXPLICIT_LIST` the full publisher set is **fixed at genesis** (dynamic grants/revocations are deferred to a later revision) | +| `history` | bool | deliver matching chunks already in the local store (mechanism in bps-history; a singlehop broker MAY refuse) | +| `closed` | bool | no audience: subscribers are restricted to the publisher set (all and only publishers subscribe) | + +The proximity constraint for implicit bindings is a **protocol constant**, not a cohort +parameter: `PO_MIN = 16`. (Making it a parameter invited proto3's unset-equals-0 +footgun — an omitted value silently disabling the constraint — and no use case varies +it.) + +Broker **capacity is deliberately not a cohort parameter**: a cohort cannot dictate a +remote node's connection count. Each broker enforces its own per-topic stream limit and +answers `FULL` when it is exhausted. + +Binding semantics (dedup rule in parentheses): + +- **`ANCHOR`** — topic = full SOC/GSOC address; all messages share one address (dedup on + the wrapped CAC — the guard against unsolicited republication of old SOCs, sound only + under an application-level requirement: payloads are distinct, i.e. the application + includes some index in the payload). +- **`SOC_ID`** — topic = SOC id; any owner with `PO(socAddr(id, owner), anchor) ≥ PO_MIN` + qualifies (dedup on chunk address). +- **`OWNER`** — topic = `keccak256(owner)`; any id under the same PO constraint — MIC + semantics (dedup on chunk address). The broker never inverts the hash: it recovers + the owner from the SOC signature and checks `keccak256(owner) == topic`; the topic + doubles as the PO anchor. +- **`FEED_TOPIC`** — id = `keccak256(topic ‖ index)`; feed-update streams, graffiti MIC + (dedup on chunk address). + +Under **explicit publisher regimes**, legitimacy is list membership, not proximity — +the PO constraint does not apply — and where dedup is on the wrapped CAC (`ANCHOR`), +the SOC id does no protocol work: it is **unconstrained**, and publishers MAY use it as +a plain sequence number. The full sequential construction — signed as a feed update, +carried as a bare index, making missed updates detectable and recoverable — is +**self-indexed feeds, [SWIP-65](https://github.com/ethersphere/SWIPs/pull/106)**. + +### Roles and capacity + +- **Broker**: the first full node contacted; root of the (here, depth = 1) multicast tree. + Enforces its own per-topic capacity. **At capacity it MUST answer `Open`/`Subscribe` + with a refusal** (`FULL`); referral to another attachment point is reserved for + bps-multihop — a singlehop-only broker simply refuses. +- **Opener**: the one peer that fixes the `CohortSpec` (`Open`); with explicit publisher + regimes the opener publishes. +- **Publisher**: sends and receives. MUST be directly connected to the broker; direct + connection is necessary, not sufficient — with explicit publishers, the genesis list + decides. +- **Subscriber**: receives only; joins by naming the topic (`Subscribe`) and carries no + cohort metadata — the broker echoes the `CohortSpec` back so every message can be + verified end-to-end. Does not exist in `closed` cohorts. + +### Information flow + +```mermaid +sequenceDiagram + autonumber + participant PD as publisher dApp + participant PN as publisher's bee node
(WS bridge) + participant B as broker
(root, full node) + participant SN as subscriber's bee node
(WS bridge + mux) + participant SD as subscriber dApp(s) + + PN->>B: Open(CohortSpec, auth) + Note over PN,B: opener fixes the cohort; publisher ⇒
direct connection to broker + B-->>PN: Ack(OK) + SN->>B: Subscribe(topic) + B-->>SN: Ack(OK, CohortSpec) + Note over B,SN: echoed spec ⇒ subscriber verifies
every message end-to-end + + PD->>PN: WS: payload + PN->>B: Publish(SOC) + B->>B: validate: SOC sig ⊨ topic binding
(+ dedup per binding) + + par fan-out to every subscriber stream + B->>SN: Broadcast(SOC) — every frame self-contained + SN->>SN: mux: one p2p stream → N WS sessions + SN->>SD: WS: payload + and publisher's own subscription (if subscriber too) + B->>PN: Broadcast(SOC) + PN->>PD: WS: payload + end +``` + +The broadcast is **end-to-end authenticated**: every subscriber re-verifies the SOC +signature against the topic binding regardless of path. + +### Wire protocol + +Messages are defined in [bps.proto](assets/swip-60/bps.proto). Framing notes: + +- Transport: libp2p stream `pubsub/1.0.0`, one stream per (peer, topic), + protobuf-over-libp2p as bee protocols elsewhere. The first message on a fresh stream is + `Open` (fixes a new cohort) or `Subscribe` (joins one — topic only, no cohort + metadata); the broker answers with `Ack`, echoing the `CohortSpec` to subscribers. +- **`Open` is idempotent**: naming an already-open topic with an **identical** spec is + equivalent to `Subscribe`; with a mismatched spec it is answered `REJECTED`. + Implicit-publisher cohorts rely on this — the first subscriber is the opener, so a + client need not know whether it is first. +- **Stream model rationale**: per-topic streams give per-cohort flow control, teardown + and role typing, and match bee's protocol idiom. Because every frame carries the full + SOC (self-contained, no per-stream handshake state), a later move to topic-muxed + streams requires no format change. +- Every `Broadcast` frame carries the **full SOC** (id, owner, signature, span, payload); + there is no handshake/data frame split. +- No BPS-level keepalive or RTT probing: liveness is the transport's job, and latency + metrics for reorganisation policies are sourced there too. +- Broker validation on `Publish`: SOC signature verifies against the topic binding, PO + constraint holds where applicable, sender is a legitimate publisher, message is not a + duplicate per the binding's dedup rule. Invalid ⇒ drop; repeated invalid ⇒ disconnect + (blocklisting policy). + +### API (WebSocket bridge) + +One endpoint pair on the Bee API. Endpoint shape follows bee +[#5435](https://github.com/ethersphere/bee/pull/5435), generalised from its single +hardcoded mode to the full parameter space; serialization conventions follow the SOC +subscription family — GSOC/MIC/MOC (bee +[#5486](https://github.com/ethersphere/bee/pull/5486), +[#5497](https://github.com/ethersphere/bee/pull/5497)) — whose `/mic/subscribe/{owner}` +and `/moc/subscribe/{id}` endpoints are the storage-fed counterparts of the `OWNER` and +`SOC_ID` bindings, so a dApp switches between stored and live feeds without +reformatting. All p2p framing is transparent to WS clients; one p2p stream is muxed to +N local WS sessions per topic. + +**`GET /pubsub/{topic}`** — upgrades to a WebSocket session on the topic. `{topic}` is +the 32-byte topic hex-encoded, or an arbitrary string hashed to 32 bytes (mnemonic +topics). Query parameters: + +| parameter | maps to | meaning | +|---|---|---| +| `peer` | — | broker underlay multiaddr; required until broker discovery exists (bps-broker-discovery) — early deployments configure it | +| `binding`, `publishers`, `admin`, `publisher-list`, `closed`, `history` | `CohortSpec` | **presence of cohort parameters makes the session the opener**: the node sends `Open` with the assembled spec; absence makes it a joiner: the node sends `Subscribe(topic)` and learns the spec from the `Ack` echo | +| `owner` (+ `id` where the binding does not fix it) | `PublisherAuth` | **presence makes the session a publisher** (read–write); absence, a subscriber (read-only) | + +Headers: + +- `swarm-keep-alive` (seconds, default 60): ping period of the **local WS link only** — + not to be confused with the p2p layer, which has no keepalive. +- `swarm-soc-fields` (per bee [#5497](https://github.com/ethersphere/bee/pull/5497)): + comma-separated SOC fields serialized per outbound message — `address`, + `recoveredPubKey`, `identifier`, `signature`, `wrappedAddress`, `span`, `payload`; + default `payload`. This is how dApps on implicit-binding streams (`OWNER`, `SOC_ID`, + feed) attribute messages — no BPS-specific frame format. +- `swarm-cache-wrapped-chunk` (per bee + [#5497](https://github.com/ethersphere/bee/pull/5497)): when true, the wrapped chunk + of every incoming message is stored in the local cache, resolvable through the bytes + endpoint — for streams whose messages reference content larger than one chunk. + +**`GET /pubsub/`** — lists the node's active topics: topic address, cohort parameters, +own role (broker / subscriber), connected peers. + +**Signing — the key-holding rule.** Message signing is the dApp's business: **the node +never holds publisher keys**. Inbound (publisher → node): `sig ‖ span ‖ payload`, +signed client-side (bee-js). Where the binding does not fix the SOC id, the frame is +prefixed with it — for feed bindings the prefix is the bare index, the signed id being +the feed id `keccak256(topic ‖ index)` (self-indexed feeds, +[SWIP-65](https://github.com/ethersphere/SWIPs/pull/106)); +under explicit regimes with `ANCHOR` binding the id does no work and there is no +prefix. The node assembles the SOC, validates it exactly as a broker would, and +publishes. End-to-end verification against the `Ack`-echoed `CohortSpec` is performed +by the local node — node and dApp are one trust domain. + +**Worked API calls — the jam cohort** (see Configurations below). Seat A opens — cohort +parameters present ⇒ `Open`, `owner` present ⇒ read–write: + +``` +wss://node:1633/pubsub/jam-tuesday?peer= + &binding=anchor&publishers=list&closed=true + &admin=0xA…&publisher-list=0xB…,0xC…,0xD…&owner=0xA… +``` + +Seats B–D join — no cohort parameters ⇒ `Subscribe`, spec learned from the `Ack` echo: + +``` +wss://node:1633/pubsub/jam-tuesday?peer=&owner=0xB… +``` + +The join URL minus `owner` is the complete out-of-band invite (topic mnemonic + broker) +until broker discovery exists. A fifth peer's `Subscribe` gets `REJECTED`. A live MIC — +all SOCs of one owner, the light-client twin of `/mic/subscribe/{owner}` — is the +implicit case: first subscriber opens with +`?binding=owner&publishers=implicit` (idempotent `Open`), topic = `keccak256(owner)`, +read-only, `swarm-soc-fields: identifier,payload`. + +### Configurations (worked examples) + +Modes are rows over the parameters; two normative examples: + +**The 4-seat jam cohort** — collaborative remix editing, a strudel livecoding session, a +multiparty game. + +``` +binding: ANCHOR (topic = mnemonic anchor) publishers: EXPLICIT_LIST (admin + 3) +closed: true (all and only publishers subscribe) history: false +``` + +Every seat sends and receives; there is no audience; the genesis list **is** the seat +bound — a fifth peer's `Subscribe` gets `REJECTED`. + +**Basic live streaming** — single publisher, open audience: + +``` +binding: FEED_TOPIC (sequential index) publishers: EXPLICIT_SINGLE +closed: false history: false +``` + +### The modes — enumerated as combinations of dimension choices + +Known use cases attach here; each mode is nothing more than a row — a combination of +publisher/subscriber info, topic match type, and history. (`+/−` = both configurations +meaningful.) + +| # of pubs | pubs implicit? | subscribers | topic / anchor match | history | use case | +|---|---|---|---|---|---| +| 1 | — | all | feed topic, index sequential | — | live video streaming | +| any | — | all | feed topic, index sequential | — | live videoconference | +| — | + | all | feed topic | +/— | tags, adverts; private co-authoring | +| all | — | all | topic a mere mnemonic of the cohort | +/— | gossip cohort for multi-party / group chat | +| any | + | all | anchor (ephemeral GSOC) | +/— | anythread comments / troll-box | +| any | + | all | ID = `keccak256(topic ‖ index)` | +/— | following one or more feeds | +| — | + | all | feed special, mined index | +/— | following graffiti soc | + +The audience is bounded by the broker's capacity; scaling past it is bps-multihop's +business. + +Rows requiring implicit publishers or history are specified in bps-implicit-publisher and +bps-history respectively. + +## Rationale: why not gossipsub + +libp2p ships gossipsub, a battle-tested mesh multicast. BPS builds its own protocol +because gossipsub's core mechanisms — flooding to a random mesh, IHAVE/IWANT +pull-recovery — are exactly what an incentivised network rejects: **no node wants to pay +for a message it did not ask for.** That one economic fact dissolves gossipsub's +machinery: metered edges mean no redundant paths and no transport-level duplicates; a +cohort's `CohortSpec` scopes every session; authentication is structural (SOC-signed +against the topic binding), so brokers and relays forward without being trusted — an +intermediate can withhold, never forge; and withholding is a liveness fault recoverable +by re-pointing or relocating the topic. Multihop forwarding (bps-multihop) adds capacity +without reintroducing flooding: every edge still pays upstream, every node still receives +only its topic's stream. + +## Out of scope (deliberately) + +Multihop relaying and referral (bps-multihop), reorganisation policies (SWATCH, SPORE — +policy SWIPs over this protocol's events and actions, no new frames), bandwidth incentives +(bps-bw-incentives), broker discovery (SWIP-59 MEX; early deployments hardcode brokers), +history delivery mechanism (bps-history), implicit-publisher event sourcing +(bps-implicit-publisher), and **dynamic publisher-list changes** — grants/revocations +after genesis are deferred to a later revision; the `EXPLICIT_LIST` set is fixed at +`Open`. + +## Conformance (definition of done) + +An implementation is conformant when: + +1. a broker enforces its per-topic capacity, publisher legitimacy, per-binding validation + and dedup; +2. a subscriber re-verifies every message end-to-end (against the `Ack`-echoed + `CohortSpec`) and detects (only) liveness faults; +3. the two worked configurations above interoperate across independent implementations + against the frames in [bps.proto](assets/swip-60/bps.proto); +4. a `FULL` refusal is issued at capacity — and nothing else is (no referral); +5. the WS bridge round-trips both worked configurations end to end — open, publish, + subscribe — with all signing on the client side (the node holds no publisher keys). + +## Backwards compatibility + +New protocol; no existing behaviour changes. Reserved `Broadcast` frame fields hold the +multihop control plane, so bps-multihop extends without a version bump; self-contained +frames mean a change of stream model needs no format change either. + +## References + +Wire: [bps.proto](assets/swip-60/bps.proto) · origin: +[PR #93](https://github.com/ethersphere/SWIPs/pull/93) "Add: pubsub" · broker discovery: +[SWIP-59 MEX, PR #103](https://github.com/ethersphere/SWIPs/pull/103) · implementation: +bee [#5435](https://github.com/ethersphere/bee/pull/5435), bee-js +[#1151](https://github.com/ethersphere/bee-js/pull/1151) + +## Copyright + +Copyright and related rights waived via [CC0](https://creativecommons.org/publicdomain/zero/1.0/).