@decentnetwork/beagle 0.1.96 → 0.1.97

package/dist/desktop/app.js

@@ -1,4 +1,4 @@
-window.__DK_UI_VERSION="0.1.96";
+window.__DK_UI_VERSION="0.1.97";
 "use strict";
 (() => {
   // packages/beagle-ui/dist/beagle-ui.js

package/dist/ipc.d.ts

@@ -44,9 +44,13 @@ export declare function ipcSocketPath(dataDir: string, platform?: NodeJS.Platfor
  *  Windows named pipes, so on Windows we connect and let net report. */
 export declare function ipcSocketSupportsFsExistenceCheck(platform?: NodeJS.Platform): boolean;
 /** How long the liveness probe waits for a connect before calling it dead.
- *  Local IPC: a listening socket accepts in microseconds, so this is a
- *  generous ceiling that still keeps the 1500ms watch tick honest. */
-export declare const DAEMON_PROBE_TIMEOUT_MS = 400;
+ *  A listening local socket accepts in well under a millisecond, so this
+ *  budget never measured the daemon — it measured how long OUR event loop took
+ *  to run the connect callback. At 400 ms a GC pause or a big JSON body in
+ *  beagle read as "daemon gone": 814 false flaps on mac-dev while the daemon's
+ *  pid never changed. One miss is not believed either; see DAEMON_MISS_STREAK
+ *  in lan-handoff.ts. */
+export declare const DAEMON_PROBE_TIMEOUT_MS = 2000;
 /**
  * Is a decentlan daemon actually LISTENING for this identity?
  *

package/dist/ipc.js

@@ -28,9 +28,13 @@ export function ipcSocketSupportsFsExistenceCheck(platform = process.platform) {
     return platform !== "win32";
 }
 /** How long the liveness probe waits for a connect before calling it dead.
- *  Local IPC: a listening socket accepts in microseconds, so this is a
- *  generous ceiling that still keeps the 1500ms watch tick honest. */
-export const DAEMON_PROBE_TIMEOUT_MS = 400;
+ *  A listening local socket accepts in well under a millisecond, so this
+ *  budget never measured the daemon — it measured how long OUR event loop took
+ *  to run the connect callback. At 400 ms a GC pause or a big JSON body in
+ *  beagle read as "daemon gone": 814 false flaps on mac-dev while the daemon's
+ *  pid never changed. One miss is not believed either; see DAEMON_MISS_STREAK
+ *  in lan-handoff.ts. */
+export const DAEMON_PROBE_TIMEOUT_MS = 2000;
 /**
  * Is a decentlan daemon actually LISTENING for this identity?
  *

package/dist/lan-handoff.d.ts

@@ -6,6 +6,13 @@ export type HandoffState = "embedded" | "releasing" | "daemon" | "waiting";
  *  gives up on it and takes the identity back. `agentnet restart` and a
  *  launchd/systemd kickstart are both back well inside this. */
 export declare const DAEMON_GRACE_MS: number;
+/** One missed probe is not a verdict. The probe runs on beagle's own event
+ *  loop, so any stall here (GC, a large /api/desktop body, an npm child) fires
+ *  its timer before the queued `connect` callback runs, and a healthy daemon
+ *  reads as gone — 814 times on mac-dev while daemon pid 75338 never changed.
+ *  Only a streak of misses that also spans real time is believed. */
+export declare const DAEMON_MISS_STREAK = 3;
+export declare const DAEMON_MISS_WINDOW_MS = 5000;
 export interface SwitchableHostOptions {
     dataDir: string;
     /** Build a fresh embedded host. Called on first start and on every fallback,
@@ -37,6 +44,9 @@ export declare class SwitchablePeerHost implements PeerHost {
     get kind(): PeerHost["kind"];
     get hasVirtualLan(): boolean;
     get state(): HandoffState;
+    /** The daemon's data dir (socket and pidfile), for callers that need to
+     *  watch a daemon restart through. */
+    get dataDir(): string;
     /** Seconds left before `releasing` gives up; 0 when not releasing. */
     get releaseSecondsLeft(): number;
     call(req: IpcRequest): Promise<IpcResponse>;

package/dist/lan-handoff.js

@@ -23,6 +23,13 @@ const RELEASE_TIMEOUT_MS = 10 * 60 * 1000;
  *  gives up on it and takes the identity back. `agentnet restart` and a
  *  launchd/systemd kickstart are both back well inside this. */
 export const DAEMON_GRACE_MS = 60 * 1000;
+/** One missed probe is not a verdict. The probe runs on beagle's own event
+ *  loop, so any stall here (GC, a large /api/desktop body, an npm child) fires
+ *  its timer before the queued `connect` callback runs, and a healthy daemon
+ *  reads as gone — 814 times on mac-dev while daemon pid 75338 never changed.
+ *  Only a streak of misses that also spans real time is believed. */
+export const DAEMON_MISS_STREAK = 3;
+export const DAEMON_MISS_WINDOW_MS = 5000;
 /**
  * A PeerHost whose backend can change under it.
  *
@@ -59,6 +66,11 @@ export class SwitchablePeerHost {
     get state() {
         return this.#state;
     }
+    /** The daemon's data dir (socket and pidfile), for callers that need to
+     *  watch a daemon restart through. */
+    get dataDir() {
+        return this.#opts.dataDir;
+    }
     /** Seconds left before `releasing` gives up; 0 when not releasing. */
     get releaseSecondsLeft() {
         if (this.#state !== "releasing")
@@ -130,14 +142,33 @@ export class SwitchablePeerHost {
         }, WATCH_INTERVAL_MS);
         this.#timer.unref?.();
     }
+    /** Consecutive missed probes; 0 the moment the daemon answers. */
+    #missStreak = 0;
+    /** A probe can take its full 2 s timeout against a 1.5 s watch interval.
+     *  Overlapping ticks would each act on the other's half-made decision. */
+    #ticking = false;
     async #tick() {
-        if (this.#busy)
+        if (this.#busy || this.#ticking)
             return;
+        this.#ticking = true;
+        try {
+            await this.#step();
+        }
+        finally {
+            this.#ticking = false;
+        }
+    }
+    async #step() {
         const daemonUp = await this.#probe(this.#opts.dataDir);
-        if (daemonUp)
+        if (daemonUp) {
             this.#daemonMissingSince = 0;
-        else if (this.#daemonMissingSince === 0)
-            this.#daemonMissingSince = Date.now();
+            this.#missStreak = 0;
+        }
+        else {
+            if (this.#daemonMissingSince === 0)
+                this.#daemonMissingSince = Date.now();
+            this.#missStreak++;
+        }
         // A tick can await; make sure nothing started while we were probing.
         if (this.#busy)
             return;
@@ -152,7 +183,14 @@ export class SwitchablePeerHost {
         // and rm the pidfile.
         if (this.#opts.force === "daemon") {
             if (this.#state === "daemon" && !daemonUp) {
-                this.#log.info("the daemon went away — waiting for it to come back (forced daemon backend)");
+                // Keep serving until the absence is confirmed. Every call made during
+                // a false `waiting` failed with "the daemon is restarting", which the
+                // UI showed as an empty friend list and a hung /api/desktop.
+                const missedMs = Date.now() - this.#daemonMissingSince;
+                if (this.#missStreak < DAEMON_MISS_STREAK || missedMs < DAEMON_MISS_WINDOW_MS)
+                    return;
+                this.#log.info(`the daemon went away — waiting for it to come back (forced daemon backend; ` +
+                    `${this.#missStreak} probes missed over ${(missedMs / 1000).toFixed(1)} s)`);
                 const old = this.#inner;
                 this.#inner = null;
                 this.#setState("waiting");

package/dist/server.d.ts

@@ -1,4 +1,4 @@
-import type { IpcRequest, IpcResponse } from "./ipc.js";
+import { type IpcRequest, type IpcResponse } from "./ipc.js";
 /** How this copy of beagle got onto disk — decides what "update" can do.
  *  npx cache paths contain _npx AND node_modules, so test _npx first. */
 export declare function installMode(): "global" | "npx" | "dev" | "unknown";
@@ -32,6 +32,8 @@ export interface BeagleServerOptions {
         readonly hasVirtualLan: boolean;
         readonly state: string;
         readonly releaseSecondsLeft: number;
+        /** The daemon's socket/pidfile dir, so "Update lan" can wait for a new pid. */
+        readonly dataDir?: string;
         armForDaemon: () => Promise<{
             state: string;
             command: string;

package/dist/server.js

@@ -19,6 +19,7 @@ import { fileURLToPath } from "node:url";
 import { dirname, join } from "node:path";
 import { INSTALL_COMMAND as LAN_INSTALL_COMMAND } from "./lan-handoff.js";
 import yaml from "js-yaml";
+import { probeDaemon } from "./ipc.js";
 import { DEFAULT_EXITS } from "./exits.js";
 import * as peerAddr from "@decentnetwork/peer";
 import { bridgeRoster } from "./bridge-roster.js";
@@ -513,10 +514,45 @@ async function npmQuery(args) {
         }
     });
 }
-/** Ask the running agentnet daemon to re-exec itself so it picks up the
- *  just-installed lan version. Tries the PATH binary, then the global
- *  install's CLI entry directly (pm2 PATH often lacks the nvm shims). */
-async function agentnetRestart() {
+/** Ask the agentnet daemon to restart onto the lan that was just installed,
+ *  and report success only when a NEW daemon answers. `agentnet restart`
+ *  exiting 0 means the request was accepted, nothing more: under launchd a
+ *  lan ≥ 0.1.302 daemon exits and launchd starts the new code, while an older
+ *  one re-execs outside launchd — which left mac-dev's KeepAlive job failing
+ *  every ~10 s. Either way the proof is a different pid that accepts a
+ *  connection. Without a data dir there is nothing to watch, so the old
+ *  answer stands. */
+async function agentnetRestart(dataDir) {
+    const pidBefore = dataDir ? readDaemonPid(dataDir) : "";
+    const r = await requestAgentnetRestart();
+    if (!r.ok || !dataDir)
+        return r;
+    return waitForNewDaemon(dataDir, pidBefore);
+}
+const readDaemonPid = (dataDir) => {
+    try {
+        return readFileSync(join(dataDir, "daemon.pid"), "utf-8").trim();
+    }
+    catch {
+        return "";
+    }
+};
+async function waitForNewDaemon(dataDir, pidBefore) {
+    const deadline = Date.now() + 60_000;
+    while (Date.now() < deadline) {
+        const pid = readDaemonPid(dataDir);
+        if (pid && pid !== pidBefore && (await probeDaemon(dataDir)))
+            return { ok: true };
+        await new Promise((r) => setTimeout(r, 1000));
+    }
+    return {
+        ok: false,
+        error: `agentnet accepted the restart, but no new daemon answered within 60 s (pid still ${readDaemonPid(dataDir) || "none"})`,
+    };
+}
+/** Run `agentnet restart`: the PATH binary, then the global install's CLI entry
+ *  directly (pm2's PATH often lacks the nvm shims). */
+async function requestAgentnetRestart() {
     const { spawn } = await import("node:child_process");
     const cliJs = join(dirname(process.execPath), "..", "lib", "node_modules", "@decentnetwork", "lan", "dist", "cli", "index.js");
     const attempts = [
@@ -1826,7 +1862,7 @@ export function startBeagleServer(opts) {
                         sendJson(res, 502, inst);
                         return;
                     }
-                    const rr = await agentnetRestart();
+                    const rr = await agentnetRestart(opts.lanHost?.dataDir);
                     updateCache = null;
                     sendJson(res, 200, { ok: true, restarted: rr.ok, ...(rr.ok ? {} : { note: rr.error }) });
                     return;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@decentnetwork/beagle",
-  "version": "0.1.96",
+  "version": "0.1.97",
   "description": "Beagle — P2P chat, file transfer and calls for regular users, on the Decent Network. No admin privilege required.",
   "type": "module",
   "main": "./dist/index.js",