@songsid/agend 2.1.2-beta.47 → 2.1.2-beta.48

package/dist/fleet-manager.js

@@ -51,6 +51,7 @@ import { readLastInboundAt } from "./daemon.js";
 import { clearPausedMarker } from "./pause-marker.js";
 import { releaseProcessFleetLock } from "./fleet-lock.js";
 import { GENERAL_PAUSE_ERROR, isGeneralInstance } from "./general-instance.js";
+import { loadOrCreateWebToken, WEB_TOKEN_INVALID_MESSAGE } from "./web-auth.js";
 import { getTmuxSession } from "./config.js";
 export function resolveReplyThreadId(argsThreadId, instanceConfig) {
     if (typeof argsThreadId === "string" && argsThreadId.length > 0) {
@@ -285,6 +286,7 @@ export class FleetManager {
     sseClients = new Set();
     webToken = null;
     viewToken = null;
+    healthServerListening = false;
     constructor(dataDir) {
         this.dataDir = dataDir;
         FleetManager.signalTarget = this;
@@ -1060,6 +1062,17 @@ export class FleetManager {
      * TODO: per-instance startup timeout (existing issue, not introduced here)
      */
     async startInstancesWithConcurrency(entries, topicMode) {
+        // Persisted pauses are intentionally preserved across fleet restarts. Filter
+        // them before grouping/staggering: startInstance() retains its own guard as
+        // a final backstop, but putting a no-op entry in this queue still consumes a
+        // full stagger slot for every distinct working directory.
+        const runnableEntries = entries.filter(([name]) => !this.lifecycle.isPaused(name));
+        const pausedCount = entries.length - runnableEntries.length;
+        if (pausedCount > 0) {
+            this.logger.info({ pausedCount }, "Paused instances excluded from startup queue");
+        }
+        if (runnableEntries.length === 0)
+            return;
         const raw = this.fleetConfig?.defaults?.startup;
         const explicitConcurrency = raw?.concurrency;
         const staggerMs = Math.max(0, Math.min(30_000, raw?.stagger_delay_ms ?? 500));
@@ -1074,10 +1087,10 @@ export class FleetManager {
         else {
             const freeMemMB = Math.round(freemem() / (1024 * 1024));
             concurrency = Math.max(2, Math.min(10, Math.floor(freeMemMB / ESTIMATED_MB_PER_INSTANCE)));
-            this.logger.info({ concurrency, freeMemMB: freeMemMB, totalInstances: entries.length }, "Adaptive startup concurrency");
+            this.logger.info({ concurrency, freeMemMB: freeMemMB, totalInstances: runnableEntries.length }, "Adaptive startup concurrency");
         }
         const byWorkDir = new Map();
-        for (const [name, config] of entries) {
+        for (const [name, config] of runnableEntries) {
             const dir = config.working_directory;
             if (!byWorkDir.has(dir))
                 byWorkDir.set(dir, []);
@@ -1225,6 +1238,21 @@ export class FleetManager {
             process.env[key] = value;
         }
     }
+    /** Initialize auth before any adapter can answer /dashboard. */
+    initializeWebAuthTokens() {
+        this.webToken = loadOrCreateWebToken(this.dataDir);
+        this.viewToken = randomBytes(24).toString("hex");
+        const viewTokenPath = join(this.dataDir, "view.token");
+        writeFileSync(viewTokenPath, this.viewToken, { encoding: "utf8", mode: 0o600 });
+        try {
+            chmodSync(viewTokenPath, 0o600);
+        }
+        catch { /* best effort */ }
+        this.healthServerListening = false;
+    }
+    getDashboardAccess() {
+        return { ready: this.healthServerListening, token: this.webToken };
+    }
     /** Start all instances from fleet config */
     async startAll(configPath) {
         FleetManager.signalTarget = this;
@@ -1238,6 +1266,7 @@ export class FleetManager {
         rotateLogIfNeeded(join(this.dataDir, "fleet.log"));
         const fleet = this.loadConfig(configPath);
         setLocale(detectLocale(fleet)); // user-facing text language (fleet.yaml defaults.locale / timezone)
+        this.initializeWebAuthTokens();
         const topicMode = fleet.channel?.mode === "topic" || !!fleet.channels?.some(ch => ch.mode === "topic");
         // Set tmux socket isolation for custom AGEND_HOME
         const { getTmuxSocketName: getSocket } = await import("./paths.js");
@@ -6331,6 +6360,7 @@ When users create specialized instances, suggest these configurations:
         this.controlClient?.stop();
         this.controlClient = null;
         if (this.healthServer) {
+            this.healthServerListening = false;
             this.healthServer.close();
             this.healthServer = null;
         }
@@ -6799,26 +6829,11 @@ When users create specialized instances, suggest these configurations:
     // ── Health HTTP endpoint ─────────────────────────────────────────────
     startHealthServer(port) {
         this.startedAt = Date.now();
-        // Generate web token before server starts so auth is enforced from the first request.
-        this.webToken = randomBytes(24).toString("hex");
-        const tokenPath = join(this.dataDir, "web.token");
-        writeFileSync(tokenPath, this.webToken, { mode: 0o600 });
-        // Defensive: if file existed previously with looser perms, tighten it.
-        try {
-            chmodSync(tokenPath, 0o600);
-        }
-        catch {
-            // best-effort
-        }
-        // Separate read-only token for the /view page: grants terminal-view + profile
-        // read, but never write (POSTs still require the full web token).
-        this.viewToken = randomBytes(24).toString("hex");
-        const viewTokenPath = join(this.dataDir, "view.token");
-        writeFileSync(viewTokenPath, this.viewToken, { mode: 0o600 });
-        try {
-            chmodSync(viewTokenPath, 0o600);
-        }
-        catch { /* best-effort */ }
+        this.healthServerListening = false;
+        this.healthPortRetried = false;
+        // Defensive for direct/unit callers; normal startup initializes these before adapters.
+        if (!this.webToken || !this.viewToken)
+            this.initializeWebAuthTokens();
         this.healthServer = createServer((req, res) => {
             res.setHeader("Content-Type", "application/json");
             // Public health probe — no auth required.
@@ -6845,7 +6860,7 @@ When users create specialized instances, suggest these configurations:
                     ?? (typeof headerToken === "string" ? headerToken : null);
                 if (!this.webToken || providedToken !== this.webToken) {
                     res.writeHead(401);
-                    res.end(JSON.stringify({ error: "Unauthorized" }));
+                    res.end(JSON.stringify({ error: WEB_TOKEN_INVALID_MESSAGE }));
                     return;
                 }
             }
@@ -7060,10 +7075,20 @@ When users create specialized instances, suggest these configurations:
             res.writeHead(404);
             res.end(JSON.stringify({ error: "not found" }));
         });
+        const markListening = (afterTakeover = false) => {
+            this.healthServerListening = true;
+            this.logger.info({ port }, afterTakeover
+                ? "Health endpoint listening (after takeover)"
+                : "Health endpoint listening");
+            this.logger.info({ url: `http://localhost:${port}/ui?token=${this.webToken}` }, "Web UI available");
+            this.logger.info({ url: `http://localhost:${port}/view?token=${this.viewToken}` }, "Web View available");
+        };
         this.healthServer.on("error", (err) => {
+            this.healthServerListening = false;
             if (err.code === "EADDRINUSE") {
                 if (this.healthPortRetried) {
-                    this.logger.debug({ port }, "Health port still in use after takeover — skipping health endpoint");
+                    this.logger.error({ err, port }, "Health port still in use after takeover — dashboard disabled");
+                    this.notifyFleetError(`⚠️ Dashboard unavailable — health port ${port} is already in use. Stop the conflicting process or configure a different health_port.`);
                     return;
                 }
                 this.healthPortRetried = true;
@@ -7084,19 +7109,14 @@ When users create specialized instances, suggest these configurations:
                 setTimeout(() => {
                     if (!this.healthServer)
                         return;
-                    this.healthServer.listen(port, "127.0.0.1", () => {
-                        this.logger.info({ port }, "Health endpoint listening (after takeover)");
-                    });
+                    this.healthServer.listen(port, "127.0.0.1", () => markListening(true));
                 }, 1500);
                 return;
             }
             this.logger.error({ err, port }, "Health server error");
+            this.notifyFleetError(`⚠️ Dashboard unavailable — health server failed: ${err.message}`);
         });
-        this.healthServer.listen(port, "127.0.0.1", () => {
-            this.logger.info({ port }, "Health endpoint listening");
-        });
-        this.logger.info({ url: `http://localhost:${port}/ui?token=${this.webToken}` }, "Web UI available");
-        this.logger.info({ url: `http://localhost:${port}/view?token=${this.viewToken}` }, "Web View available");
+        this.healthServer.listen(port, "127.0.0.1", () => markListening());
     }
     getUiStatus() {
         const fleetNames = Object.keys(this.fleetConfig?.instances ?? {});