Federated relay tier: peer flooding, per-member isolation, registry admission and relay discovery

AGENTS.md

@@ -3,7 +3,7 @@
 ## Repo shape
 - `rfc.txt` (FRX — Federated Retrieval Exchange, Draft 0.5) is the normative spec; `src/` is the Phase 1 `frxd` implementation (single crate, two binaries).
 - `frxd` is the member node (init/add/index/serve/relay/query/status); `frx` is the thin client (search/query/status). Relay and node roles are separate subcommands.
-- Commands: `cargo build`, `cargo test` (80 tests: unit in `src/`; e2e `tests/phase1.rs`; conformance `tests/conformance.rs`; aggregates + member directory `tests/aggregates.rs`; registry `tests/registry.rs`; concurrency/restart `tests/concurrency.rs`; real subprocess CLI `tests/cli.rs`; 1000-doc `tests/scale.rs`; purge-log absence `tests/purges.rs`; shared fixtures `tests/common/mod.rs`). No CI/lint config.
+- Commands: `cargo build`, `cargo test` (85 tests: unit in `src/`; e2e `tests/phase1.rs`; conformance `tests/conformance.rs`; aggregates + member directory `tests/aggregates.rs`; registry `tests/registry.rs`; federation/isolation/admission `tests/federation.rs`; concurrency/restart `tests/concurrency.rs`; real subprocess CLI `tests/cli.rs`; 1000-doc `tests/scale.rs`; purge-log absence `tests/purges.rs`; shared fixtures `tests/common/mod.rs`). No CI/lint config.
 - E2E pattern: relay + nodes in-process on ephemeral ports with tempdir corpora; use `tests/common/mod.rs` helpers (`spawn_relay*`, `query_envelope`, `poll_messages`, `register`) for new coverage. Raw relay polls return envelopes (payload under `body`), not response bodies.
 
 ## Editing the spec
@@ -21,7 +21,7 @@
 - Relay addresses transport mailboxes by `key` (unicast `to` = recipient pubkey; queues keyed by pubkey); the identifier is protocol identity only. Registry binding checks `map[key].id == from`.
 - Built: envelope/query/response, Tantivy index, aggregates, member directory. Not built: dashboard UI, directory watching, TLS, lineage/delegation. Responses travel relay-mediated unicast; transport is HTTP long-poll, not SSE.
 - Economics is out of protocol scope (I4: aggregates advise, contracts govern): no receipt, citation, pricing, or settlement fields or message types exist or may be added.
-- Relay verifies signatures and ±300s timestamp skew, carries only `query` broadcasts, holds no history (queue drained on poll), and returns 429 + Retry-After under backpressure — never silent drops. Mailbox polls require proof of key possession: `GET /v1/challenge` then a signed single-use nonce, so knowing a pubkey is not enough to drain its queue.
+- Relay verifies signatures and ±300s timestamp skew, carries only `query` broadcasts, holds no history (queues drained on poll), and requires challenge–response proof of key possession for mailbox polls. Per-member queues are isolated: a lagging member gets 429 + Retry-After with a `missed` count; publishers and other members are never stalled. Relays MAY flood to configured peers (`/v1/federation`, `--peer`+`--url`, hop-bounded, seen-set dedup without suppressing identical direct publishes) and MAY gate senders against a registry (`--registry`+`--ma-key`).
 - Responder searches only collections marked shared (I9), stays silent when nothing matches, and emits results with honest `truncated`/`more_available` and no scores (I6). BM25 order is a local implementation detail, not protocol surface.
 - Index layout: Tantivy at `<data_dir>/index`, collections manifest at `<data_dir>/collections.toml`; `exposure` (metadata|full) gates whether `content` is returned.
 - Egress checks live in the responder path (`src/node.rs` `respond`), not the relay — keep private collections unreachable there.
@@ -30,11 +30,11 @@
 - Relay backpressure is global: any member's full queue 429s every publisher until drained (visible per §3, but one lagging member can stall the firehose — revisit before scale).
 - Member authority (Draft 0.5 §6): the MA-signed registry snapshot is authoritative when configured (`[node] registry` = file path or URL, `ma_key` pinned; monotonic version — rollback and forgery close the node; file path is mtime-reloaded, URL is fetched at start + every 60s and cached to `<data_dir>/registry-cache.json`, so outage fails static). Keys carry optional validity windows (`not_before`/`not_after`); rotation = `registry add-key` then `revoke-key`.
 - `<data_dir>/members.toml` (name, pubkey, class, `previous` keys, mtime-reloaded) is a dev/local fallback used only when no registry is configured; empty directory without a registry is open bootstrap only when `dev_bootstrap = true` (RFC §6: explicit dev flag). Receivers drop content-bearing responses from enrichment-class senders (metadata-only, §6).
-- MA tooling: `frxd registry init|add|add-key|revoke-key|remove|list|show|serve` (signed `registry.json` + `ma-key.hex` in `--dir`); `frxd init --id/--registry/--ma-key`; `frxd key show|rotate`; `member add --previous <old>` for the fallback path.
+- MA tooling: `frxd registry init|add|add-key|revoke-key|remove|list|set-relays|show|serve` (signed `registry.json` + `ma-key.hex` in `--dir`); `frxd init --id/--registry/--ma-key`; `frxd key show|rotate`; `member add --previous <old>` for the fallback path. A node with no `[node] relays` discovers them from the registry snapshot (`doc.relays`).
 - Aggregate semantics are our implementation choices from a terse spec: requests are `aggregate` envelopes carrying only `period`; replies carry `sent` (broadcasts that month) / `passed` (responses consumed from that member); granularity floor is enforced as YYYY or YYYY-MM only (finer rejected), yearly rolls up months. Revisit with §10 sufficiency review.
 
 ## Known gaps (Phase 2/3, intentional — don't fake them)
-- No dashboard UI, no directory watching (new files need `reindex`), no TLS, no user-supplied URL ingestion, no node-side (bilateral) rate limiting.
+- No dashboard UI, no directory watching (new files need `reindex`), no TLS, no user-supplied URL ingestion, no node-side (bilateral) rate limiting, no SSE streaming (long-poll only), no unicast confidentiality (HPKE).
 - Receipts/settlement are out of protocol scope, not unimplemented (I4; Appendix B row 2). Lineage and delegation remain §10 open issues — unspecified without a supply stream, so not buildable as written; don't invent them silently.
 - Node query dedup is by `qid` only; replay inside the ±300s skew window remains possible (no nonce cache at nodes), relays have no directory/admission, and there is no end-to-end encryption — relays see everything in clear.
 

rfc.txt

@@ -20,7 +20,7 @@ FRX is a membership federation for retrieval. Content owners answer broadcast qu
 
 3. System Model
 
-Members exchange signed messages via dumb relays (multiply-operated transport, zero protocol authority). Queries are broadcast live to all members; relays hold no history and replay nothing. Responses are unicast from responder to querier. Transport DEFAULT: HTTPS + long-poll/SSE. Bilateral rate limits with visible backpressure are normative; application-layer rejection is silent.
+Members exchange signed messages via dumb relays (multiply-operated transport, zero protocol authority). Queries are broadcast live to all members; relays hold no history and replay nothing. Relays MAY flood queries among themselves — copy-only, hop-bounded, duplicate-suppressed — so a member publishes once and coverage scales with relays, not senders. Delivery is per-member and isolated: a lagging member is backpressured visibly and never stalls publishers or peers. Responses are unicast from responder to querier. Transport DEFAULT: HTTPS + long-poll/SSE. Bilateral rate limits with visible backpressure are normative; application-layer rejection is silent.
 
 STREAM
 TYPE
@@ -69,7 +69,7 @@ Protocol-silent by design (I2): ranking, ordering, presentation, relevance gatin
 
 6. Membership
 
-The MA governs identity, contract, expulsion — who, never quality. Identifiers are MA-hosted FQDNs (`<label>.frx.<ma-domain>`); no member-controlled DNS is required. Member-hosted identifiers — keys published in the member's own domain and allowlisted by the MA — are planned, not yet normative. The MA maintains a signed, versioned registry snapshot listing identifiers, class, and authorized keys with validity windows. Nodes pin the MA key; the snapshot is the sole authority for the key→identifier binding. Rotation publishes a successor key before retiring its predecessor; revocation removes a key or shortens its validity. Registry outage is fail-static: the last validated snapshot stays in force, and open bootstrap requires an explicit development flag. Admission cost is the Sybil defense. Expulsion grounds: fabrication, admission fraud, sustained abuse — never low quality. Escalation: local throttle → advisory aggregates → MA warning → delisting → expulsion. Aggregates are inadmissible as sanction evidence (I4). Conduct not observable on the wire — link handling, retention, gating — is governed by contract; the protocol neither observes nor adjudicates it. Membership classes: source members (own content) and enrichment members (derived corpora, e.g. GDELT/CC-NEWS bots — metadata-only exposure, transformation logic open and auditable).
+The MA governs identity, contract, expulsion — who, never quality. Identifiers are MA-hosted FQDNs (`<label>.frx.<ma-domain>`); no member-controlled DNS is required. Member-hosted identifiers — keys published in the member's own domain and allowlisted by the MA — are planned, not yet normative. The MA maintains a signed, versioned registry snapshot listing identifiers, class, authorized keys with validity windows, and the federation's relay endpoints. Members and queriers may discover relays from it; relays MAY verify sender admission against it, rejecting unlisted keys visibly. Nodes pin the MA key; the snapshot is the sole authority for the key→identifier binding. Rotation publishes a successor key before retiring its predecessor; revocation removes a key or shortens its validity. Registry outage is fail-static: the last validated snapshot stays in force, and open bootstrap requires an explicit development flag. Admission cost is the Sybil defense. Expulsion grounds: fabrication, admission fraud, sustained abuse — never low quality. Escalation: local throttle → advisory aggregates → MA warning → delisting → expulsion. Aggregates are inadmissible as sanction evidence (I4). Conduct not observable on the wire — link handling, retention, gating — is governed by contract; the protocol neither observes nor adjudicates it. Membership classes: source members (own content) and enrichment members (derived corpora, e.g. GDELT/CC-NEWS bots — metadata-only exposure, transformation logic open and auditable).
 
 7. Reference Implementation — frxd
 
@@ -98,7 +98,7 @@ A conforming implementation: signs all messages with a key authorized for its id
 10. Open Issues
 
 Consumer admission tier — automated/invite admission for distributed binaries without weakening the Sybil defense (MA policy, gates §7 adoption).
-Relay discovery and default-relay governance (shipped defaults are soft centralization; mitigate with multiple defaults + one-command self-host).
+Default-relay governance — registry-listed relays settle discovery, but who operates the defaults remains a soft centralization point (mitigate with multiple operators + one-command self-host).
 Delegation grant mechanism.
 Member-hosted identifiers — keys published in the member's own DNS instead of the MA registry; MA-hosted is normative until specified.
 MA anchor rollover — successor commitment and overlap for the registry signing key.

src/commands.rs

@@ -277,6 +277,7 @@ pub fn registry_init(dir: &Path) -> Result<()> {
         issued_at: now_ts(),
         ma_key: String::new(),
         members: Vec::new(),
+        relays: Vec::new(),
     };
     let signed = registry::sign_registry(doc, &ma);
     registry::save_registry(&registry_path, &signed)?;
@@ -390,11 +391,23 @@ pub fn registry_list(dir: &Path) -> Result<()> {
     Ok(())
 }
 
+pub fn registry_set_relays(dir: &Path, relays: &[String]) -> Result<()> {
+    mutate_registry(dir, |doc| {
+        doc.relays = relays.to_vec();
+        Ok(())
+    })?;
+    println!("relays: {}", relays.join(" "));
+    Ok(())
+}
+
 pub fn registry_show(dir: &Path) -> Result<()> {
     let (_, signed) = open_registry(dir)?;
     println!("ma_key {}", signed.doc.ma_key);
     println!("version {}", signed.doc.version);
     println!("issued_at {}", signed.doc.issued_at);
+    for relay in &signed.doc.relays {
+        println!("relay {relay}");
+    }
     Ok(())
 }
 

src/main.rs

@@ -70,6 +70,16 @@ enum Command {
         listen: String,
         #[arg(long, default_value_t = relay::DEFAULT_CAPACITY)]
         capacity: usize,
+        #[arg(long = "peer")]
+        peers: Vec<String>,
+        #[arg(long)]
+        url: Option<String>,
+        #[arg(long)]
+        registry: Option<String>,
+        #[arg(long)]
+        ma_key: Option<String>,
+        #[arg(long, default_value_t = 3)]
+        max_hops: usize,
     },
     Status,
     Member {
@@ -147,6 +157,10 @@ enum RegistryCommand {
         id: String,
     },
     List,
+    SetRelays {
+        #[arg(required = true)]
+        relays: Vec<String>,
+    },
     Show,
     Serve {
         #[arg(long, default_value = "127.0.0.1:7800")]
@@ -231,10 +245,26 @@ async fn main() -> Result<()> {
             println!("control API POST http://{}/v1/local/query", handle.addr);
             tokio::signal::ctrl_c().await?;
         }
-        Command::Relay { listen, capacity } => {
+        Command::Relay {
+            listen,
+            capacity,
+            peers,
+            url,
+            registry,
+            ma_key,
+            max_hops,
+        } => {
+            let options = relay::RelayOptions {
+                capacity,
+                peers,
+                url,
+                registry,
+                ma_key,
+                max_hops,
+            };
             let (listener, addr) = relay::bind(&listen).await?;
             println!("relay listening on http://{addr} (capacity {capacity})");
-            relay::run(listener, capacity).await?;
+            relay::run(listener, options).await?;
         }
         Command::Status => commands::status(&cli.config).await?,
         Command::Member { command } => match command {
@@ -271,6 +301,7 @@ async fn main() -> Result<()> {
             }
             RegistryCommand::Remove { id } => commands::registry_remove(&dir, &id)?,
             RegistryCommand::List => commands::registry_list(&dir)?,
+            RegistryCommand::SetRelays { relays } => commands::registry_set_relays(&dir, &relays)?,
             RegistryCommand::Show => commands::registry_show(&dir)?,
             RegistryCommand::Serve { listen } => commands::registry_serve(&dir, &listen).await?,
         },

src/node.rs

@@ -23,7 +23,7 @@ use crate::message::{
     AggregateBody, Envelope, QueryBody, ResponseBody, ResponseItem, TYPE_AGGREGATE, TYPE_QUERY,
     TYPE_RESPONSE, build_response, timestamp_is_fresh,
 };
-use crate::registry::{SignedRegistry, authorized_keys, load_registry, verify_registry};
+use crate::registry::Watcher;
 
 #[derive(Default)]
 struct Aggregates {
@@ -37,8 +37,7 @@ pub struct Node {
     index: LocalIndex,
     members: RwLock<Vec<Member>>,
     members_mtime: Mutex<Option<SystemTime>>,
-    registry: Mutex<Option<SignedRegistry>>,
-    registry_mtime: Mutex<Option<SystemTime>>,
+    registry: Option<Arc<Watcher>>,
     aggregates: Mutex<Aggregates>,
     seen: Mutex<HashSet<String>>,
     pending: Mutex<HashMap<String, Vec<(String, ResponseBody)>>>,
@@ -122,14 +121,28 @@ impl Node {
             .timeout(Duration::from_secs(15))
             .build()
             .context("building http client")?;
-        let node = Arc::new(Self {
+        let registry = match (
+            config.node.registry.as_deref(),
+            config.node.ma_key.as_deref(),
+        ) {
+            (Some(source), Some(ma_key)) => Some(Watcher::new(
+                source,
+                ma_key,
+                Some(config.registry_cache_path()),
+            )),
+            (Some(_), None) => return Err(anyhow!("registry configured without ma_key")),
+            _ => None,
+        };
+        if let Some(watcher) = &registry {
+            watcher.load_initial();
+        }
+        Ok(Arc::new(Self {
             config,
             key,
             index,
             members: RwLock::new(members),
             members_mtime: Mutex::new(members_mtime),
-            registry: Mutex::new(None),
-            registry_mtime: Mutex::new(None),
+            registry,
             aggregates: Mutex::new(Aggregates::default()),
             seen: Mutex::new(HashSet::new()),
             pending: Mutex::new(HashMap::new()),
@@ -137,90 +150,7 @@ impl Node {
             client,
             sent: AtomicU64::new(0),
             received: AtomicU64::new(0),
-        });
-        node.load_initial_registry();
-        Ok(node)
-    }
-
-    fn load_initial_registry(&self) {
-        let Some(source) = &self.config.node.registry else {
-            return;
-        };
-        let result = if source.starts_with("http://") || source.starts_with("https://") {
-            load_registry(&self.config.registry_cache_path())
-        } else {
-            let path = std::path::PathBuf::from(source);
-            let loaded = load_registry(&path);
-            if loaded.is_ok() {
-                let mtime = fs::metadata(&path)
-                    .and_then(|metadata| metadata.modified())
-                    .ok();
-                *self.registry_mtime.lock().expect("registry mtime lock") = mtime;
-            }
-            loaded
-        };
-        if let Ok(signed) = result {
-            if let Err(error) = self.verify_and_apply_registry(signed) {
-                eprintln!("cached registry rejected: {error}");
-            }
-        }
-    }
-
-    fn verify_and_apply_registry(&self, signed: SignedRegistry) -> Result<()> {
-        let ma_key = self
-            .config
-            .node
-            .ma_key
-            .as_deref()
-            .ok_or_else(|| anyhow!("registry configured without ma_key"))?;
-        verify_registry(&signed, ma_key)?;
-        {
-            let current = self.registry.lock().expect("registry lock");
-            if let Some(existing) = current.as_ref() {
-                if signed.doc.version <= existing.doc.version {
-                    return Err(anyhow!("registry version rollback rejected"));
-                }
-            }
-        }
-        *self.registry.lock().expect("registry lock") = Some(signed);
-        Ok(())
-    }
-
-    fn refresh_registry(&self) {
-        let Some(source) = &self.config.node.registry else {
-            return;
-        };
-        if source.starts_with("http://") || source.starts_with("https://") {
-            return;
-        }
-        let path = std::path::PathBuf::from(source);
-        let mtime = fs::metadata(&path)
-            .and_then(|metadata| metadata.modified())
-            .ok();
-        {
-            let last = self.registry_mtime.lock().expect("registry mtime lock");
-            if *last == mtime {
-                return;
-            }
-        }
-        let Ok(signed) = load_registry(&path) else {
-            return;
-        };
-        *self.registry_mtime.lock().expect("registry mtime lock") = mtime;
-        if let Err(error) = self.verify_and_apply_registry(signed) {
-            eprintln!("registry update rejected: {error}");
-        }
-    }
-
-    async fn fetch_registry(&self) -> Result<()> {
-        let Some(url) = &self.config.node.registry else {
-            return Ok(());
-        };
-        let response = self.client.get(url).send().await?;
-        let signed: SignedRegistry = response.json().await?;
-        self.verify_and_apply_registry(signed.clone())?;
-        crate::registry::save_registry(&self.config.registry_cache_path(), &signed)?;
-        Ok(())
+        }))
     }
 
     fn refresh_members(&self) {
@@ -311,23 +241,31 @@ impl Node {
             .with_context(|| format!("binding {}", node.config.node.listen))?;
         let addr = listener.local_addr()?;
         let mut tasks = Vec::new();
-        if let Some(source) = &node.config.node.registry {
-            if source.starts_with("http://") || source.starts_with("https://") {
-                if let Err(error) = node.fetch_registry().await {
+        if let Some(watcher) = node.registry.clone() {
+            if watcher.is_url() {
+                if let Err(error) = watcher.fetch(&node.client).await {
                     eprintln!("initial registry fetch failed: {error}");
                 }
-                let node_for_registry = node.clone();
+                let client = node.client.clone();
                 tasks.push(tokio::spawn(async move {
                     loop {
-                        if let Err(error) = node_for_registry.fetch_registry().await {
+                        tokio::time::sleep(Duration::from_secs(60)).await;
+                        if let Err(error) = watcher.fetch(&client).await {
                             eprintln!("registry fetch failed: {error}");
                         }
-                        tokio::time::sleep(Duration::from_secs(60)).await;
                     }
                 }));
             }
         }
-        for relay in node.config.node.relays.clone() {
+        let relays = if node.config.node.relays.is_empty() {
+            node.registry
+                .as_ref()
+                .map(|watcher| watcher.relays())
+                .unwrap_or_default()
+        } else {
+            node.config.node.relays.clone()
+        };
+        for relay in relays {
             tasks.push(tokio::spawn(poll_relay(node.clone(), relay)));
         }
         let app = router(node.clone());
@@ -456,18 +394,11 @@ impl Node {
             return;
         }
         self.refresh_members();
-        self.refresh_registry();
-        let (class, listed) = if self.config.node.registry.is_some() {
-            let registry = self.registry.lock().expect("registry lock");
-            let authorized = registry
-                .as_ref()
-                .map(|signed| authorized_keys(signed, now_ts()));
-            match authorized {
-                Some(map) => match map.get(&envelope.key) {
-                    Some((id, class)) if id == &envelope.from => (Some(class.clone()), true),
+        let (class, listed) = if let Some(watcher) = &self.registry {
+            watcher.refresh_if_changed();
+            match watcher.authorized(&envelope.key, now_ts()) {
+                Some((id, class)) if id == envelope.from => (Some(class), true),
                 _ => (None, false),
-                },
-                None => (None, false),
             }
         } else {
             let members = self.members.read().expect("members lock");
@@ -761,12 +692,7 @@ async fn local_query(
 
 async fn local_status(State(node): State<Arc<Node>>) -> Response {
     let aggregates = node.aggregate_for(&current_period(), None);
-    let registry_version = node
-        .registry
-        .lock()
-        .expect("registry lock")
-        .as_ref()
-        .map(|signed| signed.doc.version);
+    let registry_version = node.registry.as_ref().and_then(|watcher| watcher.version());
     Json(json!({
         "name": node.config.node.name,
         "id": node.config.node.id,

src/registry.rs

@@ -1,6 +1,8 @@
 use std::collections::HashMap;
 use std::fs;
-use std::path::Path;
+use std::path::{Path, PathBuf};
+use std::sync::{Arc, Mutex};
+use std::time::SystemTime;
 
 use anyhow::{Context, Result, anyhow};
 use serde::{Deserialize, Serialize};
@@ -35,6 +37,8 @@ pub struct RegistryDoc {
     pub ma_key: String,
     #[serde(default)]
     pub members: Vec<RegistryMember>,
+    #[serde(default)]
+    pub relays: Vec<String>,
 }
 
 #[derive(Debug, Clone, Serialize, Deserialize)]
@@ -75,6 +79,127 @@ pub fn save_registry(path: &Path, signed: &SignedRegistry) -> Result<()> {
     Ok(())
 }
 
+pub struct Watcher {
+    source: String,
+    ma_key: String,
+    cache_path: Option<PathBuf>,
+    current: Mutex<Option<SignedRegistry>>,
+    mtime: Mutex<Option<SystemTime>>,
+}
+
+impl Watcher {
+    pub fn new(source: &str, ma_key: &str, cache_path: Option<PathBuf>) -> Arc<Self> {
+        Arc::new(Self {
+            source: source.to_string(),
+            ma_key: ma_key.to_string(),
+            cache_path,
+            current: Mutex::new(None),
+            mtime: Mutex::new(None),
+        })
+    }
+
+    pub fn is_url(&self) -> bool {
+        self.source.starts_with("http://") || self.source.starts_with("https://")
+    }
+
+    pub fn load_initial(&self) {
+        let loaded = if self.is_url() {
+            match &self.cache_path {
+                Some(cache) => load_registry(cache),
+                None => Err(anyhow!("no cache path for URL registry")),
+            }
+        } else {
+            let path = PathBuf::from(&self.source);
+            let loaded = load_registry(&path);
+            if loaded.is_ok() {
+                *self.mtime.lock().expect("registry mtime lock") = fs::metadata(&path)
+                    .and_then(|metadata| metadata.modified())
+                    .ok();
+            }
+            loaded
+        };
+        if let Ok(signed) = loaded {
+            if let Err(error) = self.apply(signed) {
+                eprintln!("cached registry rejected: {error}");
+            }
+        }
+    }
+
+    pub fn refresh_if_changed(&self) {
+        if self.is_url() {
+            return;
+        }
+        let path = PathBuf::from(&self.source);
+        let mtime = fs::metadata(&path)
+            .and_then(|metadata| metadata.modified())
+            .ok();
+        {
+            let last = self.mtime.lock().expect("registry mtime lock");
+            if *last == mtime {
+                return;
+            }
+        }
+        let Ok(signed) = load_registry(&path) else {
+            return;
+        };
+        *self.mtime.lock().expect("registry mtime lock") = mtime;
+        if let Err(error) = self.apply(signed) {
+            eprintln!("registry update rejected: {error}");
+        }
+    }
+
+    pub async fn fetch(&self, client: &reqwest::Client) -> Result<()> {
+        if !self.is_url() {
+            return Ok(());
+        }
+        let response = client.get(&self.source).send().await?;
+        let signed: SignedRegistry = response.json().await?;
+        self.apply(signed.clone())?;
+        if let Some(cache) = &self.cache_path {
+            save_registry(cache, &signed)?;
+        }
+        Ok(())
+    }
+
+    fn apply(&self, signed: SignedRegistry) -> Result<()> {
+        verify_registry(&signed, &self.ma_key)?;
+        {
+            let current = self.current.lock().expect("registry lock");
+            if let Some(existing) = current.as_ref() {
+                if signed.doc.version <= existing.doc.version {
+                    return Err(anyhow!("registry version rollback rejected"));
+                }
+            }
+        }
+        *self.current.lock().expect("registry lock") = Some(signed);
+        Ok(())
+    }
+
+    pub fn authorized(&self, key: &str, now: u64) -> Option<(String, String)> {
+        let current = self.current.lock().expect("registry lock");
+        current
+            .as_ref()
+            .and_then(|signed| authorized_keys(signed, now).remove(key))
+    }
+
+    pub fn version(&self) -> Option<u64> {
+        self.current
+            .lock()
+            .expect("registry lock")
+            .as_ref()
+            .map(|signed| signed.doc.version)
+    }
+
+    pub fn relays(&self) -> Vec<String> {
+        self.current
+            .lock()
+            .expect("registry lock")
+            .as_ref()
+            .map(|signed| signed.doc.relays.clone())
+            .unwrap_or_default()
+    }
+}
+
 pub fn authorized_keys(signed: &SignedRegistry, now: u64) -> HashMap<String, (String, String)> {
     let mut authorized = HashMap::new();
     for member in &signed.doc.members {
@@ -117,6 +242,7 @@ mod tests {
                 issued_at: now_ts(),
                 ma_key: String::new(),
                 members,
+                relays: Vec::new(),
             },
             key,
         )

src/relay.rs

@@ -3,7 +3,7 @@ use std::net::SocketAddr;
 use std::sync::{Arc, Mutex};
 use std::time::{Duration, Instant};
 
-use anyhow::Result;
+use anyhow::{Result, anyhow};
 use axum::extract::{Query, State};
 use axum::http::{StatusCode, header};
 use axum::response::{IntoResponse, Response};
@@ -15,13 +15,40 @@ use tokio::net::TcpListener;
 
 use crate::crypto::{now_ts, poll_signing_bytes, random_nonce, verify_signature};
 use crate::message::{Envelope, TYPE_AGGREGATE, TYPE_QUERY, TYPE_RESPONSE, timestamp_is_fresh};
+use crate::registry::Watcher;
 
 pub const DEFAULT_CAPACITY: usize = 256;
 const CHALLENGE_TTL: Duration = Duration::from_secs(120);
+const SEEN_TTL: Duration = Duration::from_secs(30);
+
+#[derive(Clone, Debug)]
+pub struct RelayOptions {
+    pub capacity: usize,
+    pub peers: Vec<String>,
+    pub url: Option<String>,
+    pub registry: Option<String>,
+    pub ma_key: Option<String>,
+    pub max_hops: usize,
+}
+
+impl Default for RelayOptions {
+    fn default() -> Self {
+        Self {
+            capacity: DEFAULT_CAPACITY,
+            peers: Vec::new(),
+            url: None,
+            registry: None,
+            ma_key: None,
+            max_hops: 3,
+        }
+    }
+}
 
 #[derive(Default)]
 struct MemberQueue {
     items: VecDeque<(u64, Envelope)>,
+    lagging: bool,
+    missed: u64,
 }
 
 struct Challenge {
@@ -32,57 +59,136 @@ struct Challenge {
 struct Inner {
     members: HashMap<String, MemberQueue>,
     challenges: HashMap<String, Challenge>,
+    seen: HashMap<String, Instant>,
     seq: u64,
 }
 
 pub struct Relay {
     inner: Mutex<Inner>,
-    capacity: usize,
+    options: RelayOptions,
+    registry: Option<Arc<Watcher>>,
+    client: reqwest::Client,
 }
 
 impl Relay {
-    pub fn new(capacity: usize) -> Arc<Self> {
-        Arc::new(Self {
+    pub fn new(options: RelayOptions) -> Result<Arc<Self>> {
+        let registry = match (options.registry.as_deref(), options.ma_key.as_deref()) {
+            (Some(source), Some(ma_key)) => Some(Watcher::new(source, ma_key, None)),
+            (Some(_), None) => return Err(anyhow!("registry configured without ma_key")),
+            _ => None,
+        };
+        if !options.peers.is_empty() && options.url.is_none() {
+            return Err(anyhow!("--url is required when --peer is set"));
+        }
+        let relay = Arc::new(Self {
             inner: Mutex::new(Inner {
                 members: HashMap::new(),
                 challenges: HashMap::new(),
+                seen: HashMap::new(),
                 seq: 0,
             }),
-            capacity,
-        })
+            options,
+            registry,
+            client: reqwest::Client::builder()
+                .timeout(Duration::from_secs(5))
+                .build()?,
+        });
+        if let Some(watcher) = &relay.registry {
+            watcher.load_initial();
+        }
+        Ok(relay)
     }
 
-    fn push(&self, targets: Option<&str>, envelope: Envelope) -> Result<usize, StatusCode> {
-        let mut inner = self.inner.lock().expect("relay lock");
-        if inner
-            .members
-            .values()
-            .any(|q| q.items.len() >= self.capacity)
-        {
-            return Err(StatusCode::TOO_MANY_REQUESTS);
+    fn admit(&self, envelope: &Envelope) -> bool {
+        match &self.registry {
+            Some(watcher) => {
+                watcher.refresh_if_changed();
+                watcher.authorized(&envelope.key, now_ts()).is_some()
             }
+            None => true,
+        }
+    }
+
+    fn mark_seen(&self, signature: &str) -> bool {
+        let mut inner = self.inner.lock().expect("relay lock");
+        let now = Instant::now();
+        inner
+            .seen
+            .retain(|_, at| now.duration_since(*at) < SEEN_TTL);
+        if inner.seen.contains_key(signature) {
+            return false;
+        }
+        inner.seen.insert(signature.to_string(), now);
+        true
+    }
+
+    fn push_local(&self, target: Option<&str>, envelope: Envelope) -> Result<usize, StatusCode> {
+        let mut inner = self.inner.lock().expect("relay lock");
         inner.seq += 1;
         let seq = inner.seq;
-        match targets {
+        match target {
             Some(member) => {
                 let Some(queue) = inner.members.get_mut(member) else {
                     return Err(StatusCode::NOT_FOUND);
                 };
+                if queue.items.len() >= self.options.capacity {
+                    queue.lagging = true;
+                    queue.missed += 1;
+                    return Err(StatusCode::TOO_MANY_REQUESTS);
+                }
                 queue.items.push_back((seq, envelope));
                 Ok(1)
             }
             None => {
-                let from = envelope.key.clone();
-                inner.members.entry(from).or_default();
+                let publisher = envelope.key.clone();
+                inner.members.entry(publisher).or_default();
                 let mut delivered = 0;
                 for queue in inner.members.values_mut() {
+                    if queue.items.len() >= self.options.capacity {
+                        queue.lagging = true;
+                        queue.missed += 1;
+                    } else {
                         queue.items.push_back((seq, envelope.clone()));
                         delivered += 1;
                     }
+                }
                 Ok(delivered)
             }
         }
     }
+
+    fn forward(&self, envelope: Envelope, origin: Option<&str>, hops: usize) {
+        if self.options.peers.is_empty() {
+            return;
+        }
+        let Some(url) = self.options.url.clone() else {
+            return;
+        };
+        let peers: Vec<String> = self
+            .options
+            .peers
+            .iter()
+            .filter(|peer| Some(peer.as_str()) != origin)
+            .cloned()
+            .collect();
+        if peers.is_empty() {
+            return;
+        }
+        let client = self.client.clone();
+        tokio::spawn(async move {
+            for peer in peers {
+                let target = format!(
+                    "{}/v1/federation?origin={}&hops={}",
+                    peer.trim_end_matches('/'),
+                    url,
+                    hops
+                );
+                if let Err(error) = client.post(&target).json(&envelope).send().await {
+                    eprintln!("federation forward to {peer} failed: {error}");
+                }
+            }
+        });
+    }
 }
 
 pub fn router(relay: Arc<Relay>) -> Router {
@@ -90,13 +196,30 @@ pub fn router(relay: Arc<Relay>) -> Router {
         .route("/health", get(health))
         .route("/v1/challenge", get(challenge))
         .route("/v1/publish", post(publish))
+        .route("/v1/federation", post(federation))
         .route("/v1/unicast", post(unicast))
         .route("/v1/poll", get(poll))
         .with_state(relay)
 }
 
-pub async fn run(listener: TcpListener, capacity: usize) -> Result<()> {
-    let relay = Relay::new(capacity);
+pub async fn run(listener: TcpListener, options: RelayOptions) -> Result<()> {
+    let relay = Relay::new(options)?;
+    if let Some(watcher) = relay.registry.clone() {
+        if watcher.is_url() {
+            let client = relay.client.clone();
+            if let Err(error) = watcher.fetch(&client).await {
+                eprintln!("initial registry fetch failed: {error}");
+            }
+            tokio::spawn(async move {
+                loop {
+                    tokio::time::sleep(Duration::from_secs(60)).await;
+                    if let Err(error) = watcher.fetch(&client).await {
+                        eprintln!("registry fetch failed: {error}");
+                    }
+                }
+            });
+        }
+    }
     axum::serve(listener, router(relay)).await?;
     Ok(())
 }
@@ -152,14 +275,65 @@ async fn publish(State(relay): State<Arc<Relay>>, Json(envelope): Json<Envelope>
     if envelope.msg_type != TYPE_QUERY {
         return bad_request("relay carries broadcast queries only");
     }
-    match relay.push(None, envelope) {
-        Ok(delivered) => (
+    if !relay.admit(&envelope) {
+        return bad_request("sender not admitted");
+    }
+    relay.mark_seen(&envelope.sig);
+    let delivered = relay.push_local(None, envelope.clone()).unwrap_or(0);
+    relay.forward(envelope, None, 1);
+    (
         StatusCode::ACCEPTED,
         Json(json!({ "delivered": delivered })),
     )
-            .into_response(),
-        Err(status) => backpressure(status),
+        .into_response()
 }
+
+#[derive(Deserialize)]
+struct FederationParams {
+    origin: Option<String>,
+    hops: Option<usize>,
+}
+
+async fn federation(
+    State(relay): State<Arc<Relay>>,
+    Query(params): Query<FederationParams>,
+    Json(envelope): Json<Envelope>,
+) -> Response {
+    let Some(origin) = params.origin.clone() else {
+        return bad_request("origin required");
+    };
+    if !relay.options.peers.iter().any(|peer| peer == &origin) {
+        return bad_request("unknown peer origin");
+    }
+    if envelope.verify().is_err() {
+        return bad_request("invalid signature");
+    }
+    if !timestamp_is_fresh(envelope.ts, now_ts()) {
+        return bad_request("stale timestamp");
+    }
+    if envelope.msg_type != TYPE_QUERY {
+        return bad_request("relay carries broadcast queries only");
+    }
+    if !relay.admit(&envelope) {
+        return bad_request("sender not admitted");
+    }
+    if !relay.mark_seen(&envelope.sig) {
+        return (
+            StatusCode::ACCEPTED,
+            Json(json!({ "delivered": 0, "duplicate": true })),
+        )
+            .into_response();
+    }
+    let delivered = relay.push_local(None, envelope.clone()).unwrap_or(0);
+    let hops = params.hops.unwrap_or(1);
+    if hops < relay.options.max_hops {
+        relay.forward(envelope, Some(&origin), hops + 1);
+    }
+    (
+        StatusCode::ACCEPTED,
+        Json(json!({ "delivered": delivered })),
+    )
+        .into_response()
 }
 
 #[derive(Deserialize)]
@@ -181,14 +355,17 @@ async fn unicast(
     if envelope.msg_type != TYPE_RESPONSE && envelope.msg_type != TYPE_AGGREGATE {
         return bad_request("unicast carries responses and aggregates only");
     }
-    match relay.push(Some(&params.to), envelope) {
+    if !relay.admit(&envelope) {
+        return bad_request("sender not admitted");
+    }
+    match relay.push_local(Some(&params.to), envelope) {
         Ok(_) => (StatusCode::OK, Json(json!({ "delivered": true }))).into_response(),
         Err(StatusCode::NOT_FOUND) => (
             StatusCode::NOT_FOUND,
             Json(json!({ "error": "member not connected" })),
         )
             .into_response(),
-        Err(status) => backpressure(status),
+        Err(status) => backpressure(status, 0),
     }
 }
 
@@ -225,15 +402,29 @@ async fn poll(State(relay): State<Arc<Relay>>, Query(params): Query<PollParams>)
     let timeout = Duration::from_millis(params.timeout_ms.unwrap_or(25_000).min(60_000));
     let deadline = Instant::now() + timeout;
     loop {
-        let batch: Vec<Envelope> = {
+        let (batch, lagged): (Vec<Envelope>, Option<u64>) = {
             let mut inner = relay.inner.lock().expect("relay lock");
             let queue = inner.members.entry(params.member.clone()).or_default();
+            if queue.lagging {
+                let missed = queue.missed;
+                queue.lagging = false;
+                queue.missed = 0;
+                queue.items.clear();
+                (Vec::new(), Some(missed))
+            } else {
+                (
                     queue
                         .items
                         .drain(..)
                         .map(|(_, envelope)| envelope)
-                .collect()
+                        .collect(),
+                    None,
+                )
+            }
         };
+        if let Some(missed) = lagged {
+            return backpressure(StatusCode::TOO_MANY_REQUESTS, missed);
+        }
         if !batch.is_empty() {
             return (StatusCode::OK, Json(json!({ "messages": batch }))).into_response();
         }
@@ -252,11 +443,11 @@ fn unauthorized(message: &str) -> Response {
     (StatusCode::UNAUTHORIZED, Json(json!({ "error": message }))).into_response()
 }
 
-fn backpressure(status: StatusCode) -> Response {
+fn backpressure(status: StatusCode, missed: u64) -> Response {
     (
         status,
         [(header::RETRY_AFTER, "1")],
-        Json(json!({ "error": "transport backpressure" })),
+        Json(json!({ "error": "lagging", "missed": missed })),
     )
         .into_response()
 }

tests/common/mod.rs

@@ -49,7 +49,11 @@ pub async fn spawn_relay() -> String {
 pub async fn spawn_relay_with_capacity(capacity: usize) -> String {
     let (listener, addr) = relay::bind("127.0.0.1:0").await.unwrap();
     tokio::spawn(async move {
-        let _ = relay::run(listener, capacity).await;
+        let options = relay::RelayOptions {
+            capacity,
+            ..Default::default()
+        };
+        let _ = relay::run(listener, options).await;
     });
     format!("http://{addr}")
 }

tests/conformance.rs

@@ -478,38 +478,53 @@ async fn stale_envelopes_are_rejected() {
 }
 
 #[tokio::test(flavor = "multi_thread", worker_threads = 4)]
-async fn backpressure_is_visible_and_recoverable() {
+async fn lagging_member_is_isolated_and_reported() {
     let relay_url = spawn_relay_with_capacity(1).await;
     let http = client();
-    let member = Keypair::generate();
+    let active = Keypair::generate();
+    let slow = Keypair::generate();
     let publisher = Keypair::generate();
-    register(&http, &relay_url, &member).await;
+    register(&http, &relay_url, &active).await;
+    register(&http, &relay_url, &slow).await;
 
     let first = query_envelope(&publisher, &publisher.public_hex(), "one", 5);
     assert_eq!(
         publish(&http, &relay_url, &first).await.status(),
         reqwest::StatusCode::ACCEPTED
     );
+    assert_eq!(
+        poll_messages(&http, &relay_url, &active, 300).await.len(),
+        1
+    );
+
     let second = query_envelope(&publisher, &publisher.public_hex(), "two", 5);
-    let response = publish(&http, &relay_url, &second).await;
+    assert_eq!(
+        publish(&http, &relay_url, &second).await.status(),
+        reqwest::StatusCode::ACCEPTED,
+        "a lagging subscriber must not stall publishers"
+    );
+    assert_eq!(
+        poll_messages(&http, &relay_url, &active, 300).await.len(),
+        1
+    );
+
+    let response = poll(&http, &relay_url, &slow, 300).await;
     assert_eq!(response.status(), reqwest::StatusCode::TOO_MANY_REQUESTS);
     assert_eq!(
         response
             .headers()
             .get("retry-after")
-            .and_then(|v| v.to_str().ok()),
+            .and_then(|value| value.to_str().ok()),
         Some("1")
     );
-
-    assert_eq!(
-        poll_messages(&http, &relay_url, &member, 300).await.len(),
-        1
+    let payload: Value = response.json().await.unwrap();
+    assert!(
+        payload.get("missed").and_then(Value::as_u64).unwrap_or(0) >= 1,
+        "lagging member must be told how much it missed: {payload}"
     );
     assert_eq!(
-        poll_messages(&http, &relay_url, &publisher, 300)
-            .await
-            .len(),
-        1
+        poll(&http, &relay_url, &slow, 50).await.status(),
+        reqwest::StatusCode::NO_CONTENT
     );
 
     let third = query_envelope(&publisher, &publisher.public_hex(), "three", 5);
@@ -517,6 +532,30 @@ async fn backpressure_is_visible_and_recoverable() {
         publish(&http, &relay_url, &third).await.status(),
         reqwest::StatusCode::ACCEPTED
     );
+    assert_eq!(poll_messages(&http, &relay_url, &slow, 300).await.len(), 1);
+
+    let fourth = query_envelope(&publisher, &publisher.public_hex(), "four", 5);
+    assert_eq!(
+        publish(&http, &relay_url, &fourth).await.status(),
+        reqwest::StatusCode::ACCEPTED
+    );
+    let fifth = query_envelope(&publisher, &publisher.public_hex(), "five", 5);
+    assert_eq!(
+        publish(&http, &relay_url, &fifth).await.status(),
+        reqwest::StatusCode::ACCEPTED
+    );
+    let direct = test_envelope(
+        &publisher,
+        TYPE_RESPONSE,
+        json!({"qid": "q", "results": [], "truncated": false, "more_available": 0, "cursor": null}),
+    );
+    assert_eq!(
+        unicast(&http, &relay_url, &slow.public_hex(), &direct)
+            .await
+            .status(),
+        reqwest::StatusCode::TOO_MANY_REQUESTS,
+        "unicast to a lagging member must be visibly rejected"
+    );
 }
 
 #[tokio::test(flavor = "multi_thread", worker_threads = 4)]

tests/federation.rs

@@ -0,0 +1,251 @@
+mod common;
+
+use std::fs;
+use std::path::Path;
+use std::time::Duration;
+
+use common::{
+    client, collection, config_for, messages_of_type, poll, poll_messages, publish, query_envelope,
+    register, test_envelope,
+};
+use frxd::crypto::{Keypair, now_ts};
+use frxd::index::LocalIndex;
+use frxd::message::{EXPOSURE_FULL, TYPE_QUERY, TYPE_RESPONSE};
+use frxd::node::Node;
+use frxd::registry::{self, KeyEntry, RegistryDoc, RegistryMember, SignedRegistry};
+use frxd::relay::{self, RelayOptions};
+
+async fn spawn_relay_with(options: RelayOptions) -> String {
+    let (listener, addr) = relay::bind("127.0.0.1:0").await.unwrap();
+    tokio::spawn(async move {
+        let _ = relay::run(listener, options).await;
+    });
+    format!("http://{addr}")
+}
+
+async fn federation_mesh() -> [String; 3] {
+    let mut listeners = Vec::new();
+    let mut urls = Vec::new();
+    for _ in 0..3 {
+        let (listener, addr) = relay::bind("127.0.0.1:0").await.unwrap();
+        listeners.push(listener);
+        urls.push(format!("http://{addr}"));
+    }
+    let peers: Vec<Vec<String>> = vec![
+        vec![urls[1].clone(), urls[2].clone()],
+        vec![urls[0].clone(), urls[2].clone()],
+        vec![urls[0].clone(), urls[1].clone()],
+    ];
+    for (index, listener) in listeners.into_iter().enumerate() {
+        let options = RelayOptions {
+            url: Some(urls[index].clone()),
+            peers: peers[index].clone(),
+            ..Default::default()
+        };
+        tokio::spawn(async move {
+            let _ = relay::run(listener, options).await;
+        });
+    }
+    [urls[0].clone(), urls[1].clone(), urls[2].clone()]
+}
+
+#[tokio::test(flavor = "multi_thread", worker_threads = 4)]
+async fn mesh_floods_queries_exactly_once() {
+    let [r1, _r2, r3] = federation_mesh().await;
+    let http = client();
+    let member = Keypair::generate();
+    let publisher = Keypair::generate();
+    register(&http, &r3, &member).await;
+
+    let envelope = query_envelope(&publisher, &publisher.public_hex(), "mesh", 5);
+    assert!(publish(&http, &r1, &envelope).await.status().is_success());
+
+    let messages = poll_messages(&http, &r3, &member, 900).await;
+    assert_eq!(messages.len(), 1, "expected one flooded copy: {messages:?}");
+
+    tokio::time::sleep(Duration::from_millis(300)).await;
+    let duplicates = poll_messages(&http, &r3, &member, 200).await;
+    assert!(
+        duplicates.is_empty(),
+        "mesh delivered duplicates: {duplicates:?}"
+    );
+}
+
+#[tokio::test(flavor = "multi_thread", worker_threads = 4)]
+async fn federation_rejects_unknown_origin() {
+    let [r1, _r2, _r3] = federation_mesh().await;
+    let http = client();
+    let key = Keypair::generate();
+    let envelope = query_envelope(&key, &key.public_hex(), "spoof", 5);
+    let response = http
+        .post(format!(
+            "{r1}/v1/federation?origin=http://127.0.0.1:1&hops=1"
+        ))
+        .json(&envelope)
+        .send()
+        .await
+        .unwrap();
+    assert_eq!(response.status(), reqwest::StatusCode::BAD_REQUEST);
+}
+
+fn registry_doc(
+    ma: &Keypair,
+    members: Vec<RegistryMember>,
+    relays: Vec<String>,
+    version: u64,
+) -> SignedRegistry {
+    registry::sign_registry(
+        RegistryDoc {
+            version,
+            issued_at: now_ts(),
+            ma_key: String::new(),
+            members,
+            relays,
+        },
+        ma,
+    )
+}
+
+fn member(id: &str, key: &Keypair, class: &str) -> RegistryMember {
+    RegistryMember {
+        id: id.to_string(),
+        class: class.to_string(),
+        keys: vec![KeyEntry {
+            key: key.public_hex(),
+            not_before: 0,
+            not_after: None,
+        }],
+    }
+}
+
+#[tokio::test(flavor = "multi_thread", worker_threads = 4)]
+async fn relay_admission_gates_unlisted_keys() {
+    let root = tempfile::tempdir().unwrap();
+    let ma = Keypair::generate();
+    let alice = Keypair::generate();
+    let registry_path = root.path().join("registry.json");
+    registry::save_registry(
+        &registry_path,
+        &registry_doc(
+            &ma,
+            vec![member("alice.frx.example", &alice, "source")],
+            vec![],
+            1,
+        ),
+    )
+    .unwrap();
+
+    let relay_url = spawn_relay_with(RelayOptions {
+        registry: Some(registry_path.display().to_string()),
+        ma_key: Some(ma.public_hex()),
+        ..Default::default()
+    })
+    .await;
+    let http = client();
+
+    let accepted = query_envelope(&alice, "alice.frx.example", "rust", 5);
+    assert_eq!(
+        publish(&http, &relay_url, &accepted).await.status(),
+        reqwest::StatusCode::ACCEPTED
+    );
+
+    let stranger = Keypair::generate();
+    let denied = query_envelope(&stranger, &stranger.public_hex(), "rust", 5);
+    assert_eq!(
+        publish(&http, &relay_url, &denied).await.status(),
+        reqwest::StatusCode::BAD_REQUEST,
+        "unlisted key must not publish"
+    );
+}
+
+#[tokio::test(flavor = "multi_thread", worker_threads = 4)]
+async fn node_discovers_relays_from_registry() {
+    let root = tempfile::tempdir().unwrap();
+    let ma = Keypair::generate();
+    let alice = Keypair::generate();
+
+    let (listener, relay_addr) = relay::bind("127.0.0.1:0").await.unwrap();
+    tokio::spawn(async move {
+        let _ = relay::run(listener, RelayOptions::default()).await;
+    });
+    let relay_url = format!("http://{relay_addr}");
+
+    let docs = root.path().join("bob-docs");
+    fs::create_dir_all(&docs).unwrap();
+    fs::write(docs.join("doc.txt"), "registry discovery rust document").unwrap();
+
+    let mut bob_config = config_for(&root.path().join("bob"), "bob", "http://127.0.0.1:1");
+    bob_config.node.relays = Vec::new();
+    bob_config.node.id = Some("bob.frx.example".to_string());
+    let registry_path = root.path().join("registry.json");
+    bob_config.node.registry = Some(registry_path.display().to_string());
+    bob_config.node.ma_key = Some(ma.public_hex());
+    bob_config.node.dev_bootstrap = false;
+    registry::save_registry(
+        &registry_path,
+        &registry_doc(
+            &ma,
+            vec![
+                member("alice.frx.example", &alice, "source"),
+                member("bob.frx.example", &bob_config.load_key().unwrap(), "source"),
+            ],
+            vec![relay_url.clone()],
+            1,
+        ),
+    )
+    .unwrap();
+    {
+        let index = LocalIndex::open(&bob_config.index_dir()).unwrap();
+        index
+            .add_collection(&collection("docs", &docs, true, EXPOSURE_FULL))
+            .unwrap();
+    }
+    let _bob = Node::start(bob_config).await.unwrap();
+
+    let http = client();
+    register(&http, &relay_url, &alice).await;
+    let envelope = query_envelope(&alice, "alice.frx.example", "rust", 5);
+    assert!(
+        publish(&http, &relay_url, &envelope)
+            .await
+            .status()
+            .is_success()
+    );
+
+    let mut responses = Vec::new();
+    for _ in 0..5 {
+        let messages = poll_messages(&http, &relay_url, &alice, 200).await;
+        responses.extend(messages_of_type(&messages, TYPE_RESPONSE));
+        if !responses.is_empty() {
+            break;
+        }
+    }
+    assert_eq!(
+        responses.len(),
+        1,
+        "node did not discover its relay from the registry"
+    );
+}
+
+#[tokio::test(flavor = "multi_thread", worker_threads = 4)]
+async fn unknown_types_never_flood() {
+    let [r1, _r2, _r3] = federation_mesh().await;
+    let http = client();
+    let key = Keypair::generate();
+    let response = test_envelope(
+        &key,
+        TYPE_RESPONSE,
+        serde_json::json!({"qid": "q", "results": [], "truncated": false, "more_available": 0, "cursor": null}),
+    );
+    let result = http
+        .post(format!(
+            "{r1}/v1/federation?origin={}&hops=1",
+            "http://unknown"
+        ))
+        .json(&response)
+        .send()
+        .await
+        .unwrap();
+    assert_eq!(result.status(), reqwest::StatusCode::BAD_REQUEST);
+    let _ = TYPE_QUERY;
+}

tests/registry.rs

@@ -41,6 +41,7 @@ fn ma_registry(ma: &Keypair, members: Vec<RegistryMember>, version: u64) -> Sign
             issued_at: now_ts(),
             ma_key: String::new(),
             members,
+            relays: Vec::new(),
         },
         ma,
     )