@songsid/agend 2.1.1-beta.1 → 2.1.1-beta.3

package/dist/fleet-manager.js

@@ -42,6 +42,7 @@ import { handleSettingsRequest } from "./settings-api.js";
 import { setLocale, detectLocale, t } from "./locale.js";
 import { handleAgentRequest } from "./agent-endpoint.js";
 import { ClassicChannelManager, getClassicBackendChoices, isSelectableClassicBackend, readClassicLastActivityAt } from "./classic-channel-manager.js";
+import { validateFleetConfig } from "./config-validator.js";
 import { readLastInboundAt } from "./daemon.js";
 import { clearPausedMarker } from "./pause-marker.js";
 import { getTmuxSession } from "./config.js";
@@ -88,6 +89,8 @@ const MODEL_SELECT_CALLBACK_PREFIX = "model-select:";
 const CLI_ENV_TTL_MS = 24 * 60 * 60 * 1000; // /model reads cached CLI env within 24h
 export class FleetManager {
     dataDir;
+    static signalTarget = null;
+    static sighupHandlerInstalled = false;
     children = new Map();
     lifecycle;
     /** @deprecated Use lifecycle.daemons — kept for backward compat */
@@ -112,6 +115,12 @@ export class FleetManager {
     instanceIpcClients = new Map();
     scheduler = null;
     configPath = "";
+    /** SIGHUPs received before startAll finishes are replayed once startup is safe. */
+    startupComplete = false;
+    /** Coalesces one or more SIGHUPs into at most one follow-up reconciliation. */
+    reloadPending = false;
+    /** A running reconciliation; only one may mutate lifecycle/config state at a time. */
+    reconcileInFlight = null;
     logger = createLogger("info");
     topicCommands;
     // sessionName → instanceName mapping for external sessions
@@ -174,11 +183,46 @@ export class FleetManager {
     viewToken = null;
     constructor(dataDir) {
         this.dataDir = dataDir;
+        FleetManager.signalTarget = this;
+        if (!FleetManager.sighupHandlerInstalled) {
+            process.on("SIGHUP", () => FleetManager.signalTarget?.handleSighup());
+            FleetManager.sighupHandlerInstalled = true;
+        }
         this.lifecycle = new InstanceLifecycle(this);
         this.topicCommands = new TopicCommands(this);
         this.topicArchiver = new TopicArchiver(this);
         this.statuslineWatcher = new StatuslineWatcher(this);
     }
+    handleSighup() {
+        this.logger.info("Received SIGHUP, hot-reloading config...");
+        if (!this.startupComplete) {
+            this.reloadPending = true;
+            this.logger.info("Fleet startup is still in progress — queued config reload");
+            return;
+        }
+        this.scheduleReconcile();
+    }
+    scheduleReconcile() {
+        if (this.reconcileInFlight) {
+            this.reloadPending = true;
+            this.logger.info("Config reconciliation already running — coalesced reload request");
+            return;
+        }
+        this.reloadPending = false;
+        this.reconcileInFlight = this.reconcileInstances()
+            .catch(err => this.logger.error({ err }, "SIGHUP config reload failed"))
+            .finally(() => {
+            this.reconcileInFlight = null;
+            if (this.reloadPending && this.startupComplete) {
+                this.scheduleReconcile();
+            }
+        });
+    }
+    finishStartup() {
+        this.startupComplete = true;
+        if (this.reloadPending)
+            this.scheduleReconcile();
+    }
     // ── ArchiverContext bridge ────────────────────────────────────────────
     lastActivityMs(name) {
         return this.lastActivity.get(name) ?? 0;
@@ -796,6 +840,8 @@ export class FleetManager {
     }
     /** Start all instances from fleet config */
     async startAll(configPath) {
+        FleetManager.signalTarget = this;
+        this.startupComplete = false;
         this.configPath = configPath;
         this.loadEnvFile();
         // Rotate fleet.log if oversized (before any logging)
@@ -1184,14 +1230,6 @@ export class FleetManager {
             this.checkForUpdates();
             this.updateCheckTimer = setInterval(() => this.checkForUpdates(), 24 * 60 * 60 * 1000);
         }, 60 * 60 * 1000);
-        // SIGHUP: hot-reload instance config (add/remove/restart instances)
-        const onSighup = () => {
-            this.logger.info("Received SIGHUP, hot-reloading config...");
-            this.reconcileInstances()
-                .catch(err => this.logger.error({ err }, "SIGHUP config reload failed"));
-            process.once("SIGHUP", onSighup);
-        };
-        process.once("SIGHUP", onSighup);
         const onRestart = () => {
             this.logger.info("Received SIGUSR2, initiating graceful restart...");
             this.restartInstances()
@@ -1213,6 +1251,10 @@ export class FleetManager {
             });
         };
         process.once("SIGUSR1", onFullRestart);
+        // A SIGHUP may arrive after the PID/general is available but before the
+        // rest of startup finishes. Replay one coalesced reload only after all
+        // startup-owned lifecycle work and signal handlers are in place.
+        this.finishStartup();
     }
     /**
      * Delete inbox files older than retentionDays (by mtime). Cleans the shared
@@ -4744,6 +4786,8 @@ When users create specialized instances, suggest these configurations:
         return t("classic.stopped");
     }
     async stopAll() {
+        this.startupComplete = false;
+        this.reloadPending = false;
         this.ipcStoppingInstances.add("__fleet_stopping__");
         sdNotify("STOPPING=1");
         if (this.watchdogTimer) {
@@ -4951,7 +4995,40 @@ When users create specialized instances, suggest these configurations:
         if (!this.configPath)
             return;
         const oldConfig = this.fleetConfig;
-        this.loadConfig(this.configPath);
+        const previousRawConfig = this.rawFleetConfig;
+        const previousRawDocument = this.rawFleetDocument;
+        const previousSavedSnapshot = this.savedFleetConfigSnapshot;
+        try {
+            this.loadConfig(this.configPath);
+        }
+        catch (err) {
+            this.fleetConfig = oldConfig;
+            this.rawFleetConfig = previousRawConfig;
+            this.rawFleetDocument = previousRawDocument;
+            this.savedFleetConfigSnapshot = previousSavedSnapshot;
+            throw err;
+        }
+        const validation = validateFleetConfig(this.rawFleetConfig);
+        const oldCount = Object.keys(oldConfig?.instances ?? {}).length;
+        const newCount = Object.keys(this.fleetConfig?.instances ?? {}).length;
+        const removedRatio = oldCount > 0 && newCount < oldCount
+            ? (oldCount - newCount) / oldCount
+            : 0;
+        const unsafeEmpty = oldCount > 0 && newCount === 0;
+        const unsafeBulkRemoval = removedRatio > 0.5;
+        if (!validation.valid || unsafeEmpty || unsafeBulkRemoval) {
+            this.fleetConfig = oldConfig;
+            this.rawFleetConfig = previousRawConfig;
+            this.rawFleetDocument = previousRawDocument;
+            this.savedFleetConfigSnapshot = previousSavedSnapshot;
+            this.logger.error({
+                oldCount,
+                newCount,
+                removedRatio,
+                validationErrors: validation.errors,
+            }, "Refusing unsafe fleet config reload; running configuration was kept");
+            return;
+        }
         this.routing.rebuild(this.fleetConfig);
         this.reregisterClassicChannels();
         this.scheduler?.reload();