@scp-dev/broker 0.2.0 → 0.3.0

package/dist/index.d.ts

@@ -334,6 +334,8 @@ interface BrokerOptions {
     link?: LinkFactory | undefined;
     /** how often presence that decays without a request (a hook-only endpoint) is checked for changes (default 60 s) */
     presenceEveryMs?: number | undefined;
+    /** how often the broker checks it still owns its socket and lock, stopping once another broker or a wiped state directory took them (default 2 s) */
+    ownershipEveryMs?: number | undefined;
     /** where a judge model's key_env is read (default the broker's environment) */
     env?: Record<string, string | undefined> | undefined;
     /** the judge model client for a policy's `judge.model` (default the provider's HTTP API) */

package/dist/index.js

@@ -706,7 +706,8 @@ function startBridge(options) {
       byMachine.set(view.machine, [...byMachine.get(view.machine) ?? [], view]);
     }
     if (full) {
-      for (const view of core.remoteViews()) if (view.machine !== account && !byMachine.has(view.machine)) byMachine.set(view.machine, []);
+      core.dropMachine(machine);
+      for (const view of core.remoteViews()) if (view.machine !== account && !byMachine.has(view.machine)) core.dropMachine(view.machine);
     }
     for (const [name, list] of byMachine) {
       try {
@@ -807,7 +808,8 @@ function startBridge(options) {
         if (typeof params.machine !== "string" || params.machine === machine) return;
         if (typeof params.own === "boolean") core.setOther(params.machine, !params.own);
         try {
-          core.mirror(params.machine, params.endpoints, { full: params.full === true });
+          if (params.released === true) core.dropMachine(params.machine);
+          else core.mirror(params.machine, params.endpoints, { full: params.full === true });
         } catch (error) {
           log(`relay bridge: the endpoints of ${params.machine} were not applied: ${error instanceof Error ? error.message : String(error)}`);
         }
@@ -1224,6 +1226,7 @@ function openDatabase(path) {
 // src/server.ts
 var SWEEP_MS = 10 * 6e4;
 var PRESENCE_MS = 6e4;
+var OWNERSHIP_MS = 2e3;
 var OBSERVER_METHODS = /* @__PURE__ */ new Set(["list", "snapshot", "watch", "log", "policy", "policy_check", "cloud_status", "bridge_relink"]);
 var SCOPE_METHODS = /* @__PURE__ */ new Set([
   "rules",
@@ -1304,6 +1307,7 @@ async function startBroker(options) {
   let judge;
   let bridge = null;
   let listener;
+  let socketIno;
   try {
     privateDir(dirname2(socket));
     db = openDatabase(join4(dir, "relay.db"));
@@ -1320,6 +1324,7 @@ async function startBroker(options) {
       alive: running
     });
     core.migrate();
+    core.relabelMachine();
     core.resetConnections();
     core.sweep();
     judge = new JudgeGate({ sql, core, now: options.now ?? Date.now, env: options.env ?? process.env, client: options.judgeClient, log, wake });
@@ -1356,6 +1361,7 @@ async function startBroker(options) {
     });
     listener.on("error", (error) => log(`relay broker: the socket listener failed: ${error.message}`));
     chmodSync2(socket, 384);
+    socketIno = lstatSync2(socket).ino;
   } catch (error) {
     listener?.close();
     db?.close();
@@ -1369,16 +1375,39 @@ async function startBroker(options) {
     judge.sweep();
   }, SWEEP_MS);
   const known = /* @__PURE__ */ new Map();
-  for (const view of localViews()) known.set(view.id, view.presence);
+  const wakes = /* @__PURE__ */ new Map();
+  for (const view of localViews()) {
+    known.set(view.id, view.presence);
+    wakes.set(view.id, view.wake);
+  }
   const presenceTimer = setInterval(() => {
     core.endExited();
     const live = /* @__PURE__ */ new Set();
     for (const view of localViews()) {
       live.add(view.id);
       if ((known.get(view.id) ?? "offline") !== view.presence) presenceChanged(view);
+      else if (wakes.has(view.id) && wakes.get(view.id) !== view.wake) bridge?.changed(view.id);
+      wakes.set(view.id, view.wake);
     }
     for (const id of known.keys()) if (!live.has(id)) known.delete(id);
+    for (const id of wakes.keys()) if (!live.has(id)) wakes.delete(id);
   }, options.presenceEveryMs ?? PRESENCE_MS);
+  const ownership = setInterval(() => {
+    if (closing || ownsState()) return;
+    log(`relay broker: ${socket} or its lock was removed or taken over; stopping`);
+    void stop();
+  }, options.ownershipEveryMs ?? OWNERSHIP_MS);
+  ownership.unref?.();
+  function ownsSocket() {
+    return lstatSync2(socket, { throwIfNoEntry: false })?.ino === socketIno;
+  }
+  function ownsState() {
+    try {
+      return ownsSocket() && readFileSync6(lock.path, "utf8") === lock.content;
+    } catch {
+      return false;
+    }
+  }
   const remoteHooks = {
     wake,
     fleet: notifyFleet,
@@ -1467,6 +1496,26 @@ async function startBroker(options) {
   function localViews() {
     return core.list({ all: true }).filter((view) => core.machineOf(view.id) === null);
   }
+  function setWaker(s, endpoint) {
+    const was = s.data.waker;
+    if (was === endpoint) return;
+    s.data.waker = endpoint;
+    if (was) {
+      const wake2 = core.whoami(was).wake;
+      core.waker(was, false, { grace: s.data.transient === true });
+      if (!presenceMoved(was) && core.whoami(was).wake !== wake2) bridge?.changed(was);
+    }
+    if (endpoint) {
+      core.waker(endpoint, true);
+      presenceMoved(endpoint);
+    }
+  }
+  function presenceMoved(id) {
+    const view = core.whoami(id);
+    if ((known.get(id) ?? "offline") === view.presence) return false;
+    presenceChanged(view);
+    return true;
+  }
   function presenceChanged(view) {
     known.set(view.id, view.presence);
     notifyFleet({ type: "presence", endpoint: view.address, presence: view.presence, harness: view.harness, labels: view.labels });
@@ -1618,6 +1667,7 @@ async function startBroker(options) {
     s.data.closed = true;
     clearTimeout(s.data.lineTimer);
     connections.delete(s);
+    if (!closing) setWaker(s, void 0);
     if (s.data.endpoint && !closing && !s.data.transient && !s.data.lineage) {
       const before = core.whoami(s.data.endpoint).presence;
       const settled = core.disconnect(s.data.endpoint, { settle: !s.data.brief });
@@ -1911,7 +1961,8 @@ async function startBroker(options) {
       case "wake": {
         if (s.data.lineage && !s.data.own) throw new RelayError3("unauthorized", "only the endpoint's own session says how it wakes");
         const changed = core.setWake(me, params.via, hookOf(s));
-        if (changed) bridge?.changed(me);
+        setWaker(s, params.via !== "none" && core.whoami(me).wake === params.via ? me : void 0);
+        if (!presenceMoved(me) && changed) bridge?.changed(me);
         return { changed };
       }
       case "rename": {
@@ -1948,11 +1999,13 @@ async function startBroker(options) {
         return view;
       }
       case "event": {
+        const wake2 = core.whoami(me).wake;
         const row = core.event(me, params, hookOf(s));
         notifyFleet({ type: "activity", ...row });
         bridge?.activity(me, row);
         const after = core.whoami(me);
         if ((known.get(me) ?? "offline") !== after.presence) presenceChanged(after);
+        else if (after.wake !== wake2) bridge?.changed(me);
         return row;
       }
       case "inject": {
@@ -2214,6 +2267,7 @@ async function startBroker(options) {
       closing = true;
       clearInterval(sweeper);
       clearInterval(presenceTimer);
+      clearInterval(ownership);
       await bridge?.stop();
       if (idleTimer) clearTimeout(idleTimer);
       for (const set of [...waiters.values()]) for (const fn of [...set]) fn();
@@ -2221,11 +2275,13 @@ async function startBroker(options) {
       while (pending > 0 && Date.now() < deadline) await sleep2(5);
       await Promise.race([judge.settled(), sleep2(Math.max(0, deadline - Date.now()))]);
       for (const s of connections) s.end();
-      listener.close();
-      try {
-        unlinkSync2(socket);
-      } catch {
-      }
+      if (ownsSocket()) {
+        listener.close();
+        try {
+          unlinkSync2(socket);
+        } catch {
+        }
+      } else listener.unref();
       if (socket !== join4(dir, "relay.sock")) {
         try {
           rmdirSync(dirname2(socket));

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@scp-dev/broker",
-  "version": "0.2.0",
+  "version": "0.3.0",
   "description": "Swarm Communication Protocol local broker: a unix socket JSON-RPC daemon over the relay core, its client and the bridge to an account hub.",
   "license": "Apache-2.0",
   "homepage": "https://scp.dev",
@@ -37,8 +37,8 @@
     "typecheck": "tsc -p ."
   },
   "dependencies": {
-    "@scp-dev/core": "0.1.5",
-    "@scp-dev/hub": "0.5.0"
+    "@scp-dev/core": "0.2.0",
+    "@scp-dev/hub": "0.6.0"
   },
   "devDependencies": {
     "tsup": "^8.5.1"

package/src/bridge.ts

@@ -300,14 +300,20 @@ export function startBridge(options: BridgeOptions): Bridge {
     for (const fn of waitingForLink.splice(0)) fn()
   }
 
-  /** the other machines' endpoints, grouped by machine; with `full`, a known machine missing from the list goes offline */
+  /**
+   * the other machines' endpoints, grouped by machine; with `full`, a known machine the hub no longer lists was released and its rows go,
+   * as do rows mirrored under this machine's own name from before it was renamed back to it
+   */
   function applyRemote(views: RemoteView[], full: boolean) {
     const byMachine = new Map<string, RemoteView[]>()
     for (const view of views) {
       if (view.machine === machine) continue
       byMachine.set(view.machine, [...(byMachine.get(view.machine) ?? []), view])
     }
-    if (full) for (const view of core.remoteViews()) if (view.machine !== account && !byMachine.has(view.machine)) byMachine.set(view.machine, [])
+    if (full) {
+      core.dropMachine(machine)
+      for (const view of core.remoteViews()) if (view.machine !== account && !byMachine.has(view.machine)) core.dropMachine(view.machine)
+    }
     for (const [name, list] of byMachine) {
       try {
         core.mirror(name, list, { full })
@@ -423,7 +429,8 @@ export function startBridge(options: BridgeOptions): Bridge {
         if (typeof params.machine !== "string" || params.machine === machine) return
         if (typeof params.own === "boolean") core.setOther(params.machine, !params.own)
         try {
-          core.mirror(params.machine, params.endpoints, { full: params.full === true })
+          if (params.released === true) core.dropMachine(params.machine)
+          else core.mirror(params.machine, params.endpoints, { full: params.full === true })
         } catch (error) {
           log(`relay bridge: the endpoints of ${params.machine} were not applied: ${error instanceof Error ? error.message : String(error)}`)
         }

package/src/server.ts

@@ -64,6 +64,8 @@ export interface BrokerOptions {
   link?: LinkFactory | undefined
   /** how often presence that decays without a request (a hook-only endpoint) is checked for changes (default 60 s) */
   presenceEveryMs?: number | undefined
+  /** how often the broker checks it still owns its socket and lock, stopping once another broker or a wiped state directory took them (default 2 s) */
+  ownershipEveryMs?: number | undefined
   /** where a judge model's key_env is read (default the broker's environment) */
   env?: Record<string, string | undefined> | undefined
   /** the judge model client for a policy's `judge.model` (default the provider's HTTP API) */
@@ -114,6 +116,8 @@ interface Connection {
   process?: string
   /** a one-shot CLI command's connection: its comings and goings are not fleet presence events */
   brief?: boolean
+  /** the endpoint whose wake this connection set: it counts as its waker until it closes or sets none */
+  waker?: string | undefined
   cwd?: string | undefined
   watching: boolean
   closed: boolean
@@ -126,6 +130,7 @@ interface Connection {
 
 const SWEEP_MS = 10 * 60_000
 const PRESENCE_MS = 60_000
+const OWNERSHIP_MS = 2_000
 const OBSERVER_METHODS = new Set(["list", "snapshot", "watch", "log", "policy", "policy_check", "cloud_status", "bridge_relink"])
 const SCOPE_METHODS = new Set([
   "rules",
@@ -219,6 +224,7 @@ export async function startBroker(options: BrokerOptions): Promise<Broker> {
   let judge!: JudgeGate
   let bridge: Bridge | null = null
   let listener!: Server
+  let socketIno: number | bigint | undefined
   try {
     privateDir(dirname(socket))
     db = openDatabase(join(dir, "relay.db"))
@@ -235,6 +241,7 @@ export async function startBroker(options: BrokerOptions): Promise<Broker> {
       alive: running,
     })
     core.migrate()
+    core.relabelMachine()
     core.resetConnections()
     core.sweep()
     judge = new JudgeGate({ sql, core, now: options.now ?? Date.now, env: options.env ?? process.env, client: options.judgeClient, log, wake })
@@ -271,6 +278,7 @@ export async function startBroker(options: BrokerOptions): Promise<Broker> {
     })
     listener.on("error", (error) => log(`relay broker: the socket listener failed: ${error.message}`))
     chmodSync(socket, 0o600)
+    socketIno = lstatSync(socket).ino
   } catch (error) {
     listener?.close()
     db?.close()
@@ -284,16 +292,43 @@ export async function startBroker(options: BrokerOptions): Promise<Broker> {
     judge.sweep()
   }, SWEEP_MS)
   const known = new Map<string, Presence>()
-  for (const view of localViews()) known.set(view.id, view.presence)
+  const wakes = new Map<string, string | null | undefined>()
+  for (const view of localViews()) {
+    known.set(view.id, view.presence)
+    wakes.set(view.id, view.wake)
+  }
   const presenceTimer = setInterval(() => {
     core.endExited()
     const live = new Set<string>()
     for (const view of localViews()) {
       live.add(view.id)
       if ((known.get(view.id) ?? "offline") !== view.presence) presenceChanged(view)
+      else if (wakes.has(view.id) && wakes.get(view.id) !== view.wake) bridge?.changed(view.id)
+      wakes.set(view.id, view.wake)
     }
     for (const id of known.keys()) if (!live.has(id)) known.delete(id)
+    for (const id of wakes.keys()) if (!live.has(id)) wakes.delete(id)
   }, options.presenceEveryMs ?? PRESENCE_MS)
+  const ownership = setInterval(() => {
+    if (closing || ownsState()) return
+    log(`relay broker: ${socket} or its lock was removed or taken over; stopping`)
+    void stop()
+  }, options.ownershipEveryMs ?? OWNERSHIP_MS)
+  ownership.unref?.()
+
+  /** whether the socket path is still this broker's listener */
+  function ownsSocket(): boolean {
+    return lstatSync(socket, { throwIfNoEntry: false })?.ino === socketIno
+  }
+
+  /** whether the socket path and the lock are still this broker's */
+  function ownsState(): boolean {
+    try {
+      return ownsSocket() && readFileSync(lock.path, "utf8") === lock.content
+    } catch {
+      return false
+    }
+  }
   const remoteHooks: BridgeHooks = {
     wake,
     fleet: notifyFleet,
@@ -399,6 +434,30 @@ export async function startBroker(options: BrokerOptions): Promise<Broker> {
     return core.list({ all: true }).filter((view) => core.machineOf(view.id) === null)
   }
 
+  /** counts `s` as the waker of `endpoint`, or of none, and reports the presence that moves */
+  function setWaker(s: Conn<Connection>, endpoint: string | undefined) {
+    const was = s.data.waker
+    if (was === endpoint) return
+    s.data.waker = endpoint
+    if (was) {
+      const wake = core.whoami(was).wake
+      core.waker(was, false, { grace: s.data.transient === true })
+      if (!presenceMoved(was) && core.whoami(was).wake !== wake) bridge?.changed(was)
+    }
+    if (endpoint) {
+      core.waker(endpoint, true)
+      presenceMoved(endpoint)
+    }
+  }
+
+  /** reports an endpoint's presence when it is not the one last reported; whether it was not */
+  function presenceMoved(id: string): boolean {
+    const view = core.whoami(id)
+    if ((known.get(id) ?? "offline") === view.presence) return false
+    presenceChanged(view)
+    return true
+  }
+
   /** an endpoint's presence changed outside a connection's own hello or close: watchers and the other machines hear it */
   function presenceChanged(view: EndpointView) {
     known.set(view.id, view.presence)
@@ -583,6 +642,7 @@ export async function startBroker(options: BrokerOptions): Promise<Broker> {
     s.data.closed = true
     clearTimeout(s.data.lineTimer)
     connections.delete(s)
+    if (!closing) setWaker(s, undefined)
     if (s.data.endpoint && !closing && !s.data.transient && !s.data.lineage) {
       const before = core.whoami(s.data.endpoint).presence
       const settled = core.disconnect(s.data.endpoint, { settle: !s.data.brief })
@@ -891,7 +951,8 @@ export async function startBroker(options: BrokerOptions): Promise<Broker> {
       case "wake": {
         if (s.data.lineage && !s.data.own) throw new RelayError("unauthorized", "only the endpoint's own session says how it wakes")
         const changed = core.setWake(me, params.via, hookOf(s))
-        if (changed) bridge?.changed(me)
+        setWaker(s, params.via !== "none" && core.whoami(me).wake === params.via ? me : undefined)
+        if (!presenceMoved(me) && changed) bridge?.changed(me)
         return { changed }
       }
       case "rename": {
@@ -928,11 +989,13 @@ export async function startBroker(options: BrokerOptions): Promise<Broker> {
         return view
       }
       case "event": {
+        const wake = core.whoami(me).wake
         const row = core.event(me, params, hookOf(s))
         notifyFleet({ type: "activity", ...row })
         bridge?.activity(me, row)
         const after = core.whoami(me)
         if ((known.get(me) ?? "offline") !== after.presence) presenceChanged(after)
+        else if (after.wake !== wake) bridge?.changed(me)
         return row
       }
       case "inject": {
@@ -1210,6 +1273,7 @@ export async function startBroker(options: BrokerOptions): Promise<Broker> {
       closing = true
       clearInterval(sweeper)
       clearInterval(presenceTimer)
+      clearInterval(ownership)
       await bridge?.stop()
       if (idleTimer) clearTimeout(idleTimer)
       for (const set of [...waiters.values()]) for (const fn of [...set]) fn()
@@ -1217,10 +1281,13 @@ export async function startBroker(options: BrokerOptions): Promise<Broker> {
       while (pending > 0 && Date.now() < deadline) await sleep(5)
       await Promise.race([judge.settled(), sleep(Math.max(0, deadline - Date.now()))])
       for (const s of connections) s.end()
-      listener.close()
-      try {
-        unlinkSync(socket)
-      } catch {}
+      // closing a listener unlinks its path, which another broker may own by now
+      if (ownsSocket()) {
+        listener.close()
+        try {
+          unlinkSync(socket)
+        } catch {}
+      } else listener.unref()
       if (socket !== join(dir, "relay.sock")) {
         try {
           rmdirSync(dirname(socket))