crewly 1.20.226 → 1.20.228

package/dist/backend/backend/src/services/apps/app-wake.service.d.ts

@@ -27,6 +27,26 @@
  * - A failed delivery keeps the batch (and the cursor before it) and is
  *   retried with backoff; after WAKE_FAILS_BEFORE_ORC_NOTICE failures the
  *   orchestrator is told once.
+ * - Nothing may stall the loop (2026-10-05, 科技晨报: one poll pass never
+ *   finished, the self-rescheduling loop stopped for good, and five owner
+ *   comments sat in Cloud for hours with nothing logged). Each app's poll
+ *   has a deadline (POLL_APP_DEADLINE_MS; a late result is discarded), a
+ *   pass that still has not finished after POLL_PASS_STALL_MS is abandoned
+ *   and a fresh one starts, and a delivery that has not settled after
+ *   WAKE_DELIVER_TIMEOUT_MS counts as failed and is retried (a late success
+ *   still settles the batch).
+ * - A batch being delivered is detached: changes arriving meanwhile open the
+ *   next batch instead of being marked delivered without being in the
+ *   message.
+ * - A watchdog on every scheduled pass re-arms a batch whose timer is gone
+ *   and, once per batch, logs and tells the orchestrator about owner changes
+ *   that have not reached their agent after WAKE_STUCK_NOTICE_MS.
+ * - Owner changes are delivered as owner-authored (`owner: true`, crewly#1105
+ *   queue priority): a busy agent gets them queued ahead of system traffic,
+ *   not after it goes idle.
+ * - Batches live in memory; the persisted cursor stays before the first
+ *   change of every batch not yet delivered, so after a restart the same
+ *   changes are read again and batched again.
  * - An `ask` reaches a named agent only when it is in the publisher's team
  *   and already running; it never starts a stopped agent.
  *
@@ -46,6 +66,10 @@ export interface AppWakeClient {
 export interface WakeDeliveryOptions {
     /** Start the agent when its session is down (only for the app's publisher) */
     activate: boolean;
+    /** The owner made (some of) these changes: queue ahead of system traffic (crewly#1105) */
+    owner?: boolean;
+    /** Identity of this batch (`app:<id>:<first>-<last>`): a second queued copy is a duplicate */
+    ref?: string;
 }
 /** Constructor dependencies. */
 export interface AppWakeServiceDeps {
@@ -78,7 +102,16 @@ export declare class AppWakeService {
     private timer;
     private running;
     private ticking;
+    /** When the running pass started, and its generation (an abandoned pass stops at its next step) */
+    private tickStartedAt;
+    private generation;
+    private passStallNotified;
+    /** Current poll of each app; a poll whose token was replaced or dropped discards its result */
+    private readonly pollTokens;
+    /** Batches collecting changes, by `appId\0recipient` */
     private readonly batches;
+    /** Batches handed to `deliver` and not yet confirmed (in flight, or failed and waiting to retry) */
+    private readonly outgoing;
     private readonly lastWakeAt;
     /** Highest seq read per app (the registry holds what is safe to persist) */
     private readonly fetched;
@@ -103,10 +136,12 @@ export declare class AppWakeService {
     private schedule;
     /**
      * One pass over the apps that are due. Exposed for tests and for an immediate poll.
+     * A pass still running after POLL_PASS_STALL_MS is abandoned and a fresh one starts.
      */
     tick(): Promise<void>;
     private runTick;
     private pollOne;
+    private isCurrentPoll;
     private deliveredSet;
     private pollApp;
     /**
@@ -131,14 +166,39 @@ export declare class AppWakeService {
     private recipientFor;
     private handle;
     private arm;
+    /** Whether a batch holds only visitor submissions (or only a skipped-submissions notice). */
+    private static visitorOnly;
+    /** Fold `from` (newer) into `into` (older): one message covers both. */
+    private static merge;
     /**
-     * Send one batch now (timer callback; tests). On failure the batch stays
-     * pending and is retried with backoff.
+     * Send one collecting batch now (timer callback; tests). On failure the
+     * batch stays pending and is retried with backoff.
      *
      * @param key - Batch key
      * @returns Whether a message was delivered
      */
     flush(key: string): Promise<boolean>;
+    /** Retry an outgoing batch (its retry timer), folding in whatever collected meanwhile. */
+    private retry;
+    /**
+     * Deliver an outgoing batch once, bounded by WAKE_DELIVER_TIMEOUT_MS.
+     *
+     * @param key - Batch key
+     * @param batch - The batch (already in `outgoing`)
+     * @returns Whether it was delivered
+     */
+    private send;
+    private succeeded;
+    /** Tell the orchestrator something, without waiting on it for long. */
+    private notifyOrc;
+    /**
+     * Runs before every scheduled pass: a batch with no timer and no delivery
+     * running is re-armed (it would otherwise wait forever), and owner changes
+     * that have not reached their agent after WAKE_STUCK_NOTICE_MS are logged
+     * and reported to the orchestrator, once per batch. Visitor-only batches
+     * waiting for their agent to run are expected to wait and are left alone.
+     */
+    watchdog(): void;
     /** @returns Keys of batches waiting to be sent (tests) */
     pendingKeys(): string[];
 }

package/dist/backend/backend/src/services/apps/app-wake.service.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"app-wake.service.d.ts","sourceRoot":"","sources":["../../../../../../backend/src/services/apps/app-wake.service.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAiCG;AAGH,OAAO,EAAiB,KAAK,eAAe,EAAE,MAAM,2BAA2B,CAAC;AAEhF,OAAO,KAAK,EAAE,mBAAmB,EAAsC,MAAM,4BAA4B,CAAC;AAK1G,2CAA2C;AAC3C,MAAM,WAAW,aAAa;IAC5B,WAAW,IAAI,OAAO,CAAC;IACvB,OAAO,CAAC,CAAC,EAAE,MAAM,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,CAAC,EAAE;QAAE,KAAK,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,GAAG,MAAM,GAAG,SAAS,CAAC,CAAC;QAAC,SAAS,CAAC,EAAE,MAAM,CAAA;KAAE,GAAG,OAAO,CAAC,CAAC,CAAC,CAAC;CAC1I;AAED,+BAA+B;AAC/B,MAAM,WAAW,mBAAmB;IAClC,8EAA8E;IAC9E,QAAQ,EAAE,OAAO,CAAC;CACnB;AAED,gCAAgC;AAChC,MAAM,WAAW,kBAAkB;IACjC,MAAM,EAAE,aAAa,CAAC;IACtB,QAAQ,EAAE,mBAAmB,CAAC;IAC9B,uFAAuF;IACvF,OAAO,EAAE,CAAC,OAAO,EAAE,MAAM,GAAG,IAAI,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,mBAAmB,KAAK,OAAO,CAAC,OAAO,CAAC,CAAC;IAC/F,wFAAwF;IACxF,YAAY,CAAC,EAAE,CAAC,IAAI,EAAE,MAAM,EAAE,SAAS,EAAE,MAAM,KAAK,OAAO,CAAC,MAAM,GAAG,IAAI,CAAC,CAAC;IAC3E,gDAAgD;IAChD,SAAS,CAAC,EAAE,CAAC,OAAO,EAAE,MAAM,KAAK,OAAO,CAAC;IACzC,8DAA8D;IAC9D,UAAU,EAAE,MAAM,CAAC;IACnB,GAAG,CAAC,EAAE,MAAM,MAAM,CAAC;IACnB,MAAM,CAAC,EAAE,eAAe,CAAC;CAC1B;AA4BD,UAAU,UAAU;IAClB,QAAQ,EAAE,MAAM,CAAC;IACjB,MAAM,EAAE,MAAM,CAAC;CAChB;AAED,uEAAuE;AACvE,eAAO,MAAM,aAAa,mBAAuB,CAAC;AAElD;;GAEG;AACH,qBAAa,cAAc;IAiBb,OAAO,CAAC,QAAQ,CAAC,IAAI;IAhBjC,OAAO,CAAC,QAAQ,CAAC,GAAG,CAAe;IACnC,OAAO,CAAC,QAAQ,CAAC,MAAM,CAAkB;IACzC,OAAO,CAAC,KAAK,CAA8C;IAC3D,OAAO,CAAC,OAAO,CAAS;IACxB,OAAO,CAAC,OAAO,CAA8B;IAC7C,OAAO,CAAC,QAAQ,CAAC,OAAO,CAA4B;IACpD,OAAO,CAAC,QAAQ,CAAC,UAAU,CAA6B;IACxD,4EAA4E;IAC5E,OAAO,CAAC,QAAQ,CAAC,OAAO,CAA6B;IACrD,4DAA4D;IAC5D,OAAO,CAAC,QAAQ,CAAC,SAAS,CAAkC;IAC5D,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAiC;IACzD,6EAA6E;IAC7E,OAAO,CAAC,QAAQ,CAAC,YAAY,CAAuC;IACpE,OAAO,CAAC,QAAQ,CAAC,YAAY,CAAqB;gBAErB,IAAI,EAAE,kBAAkB;IAKrD,qDAAqD;IACrD,KAAK,IAAI,IAAI;IAMb,gHAAgH;IAChH,IAAI,IAAI,IAAI;IAQZ;;;;;OAKG;IACH,UAAU,CAAC,KAAK,EAAE,MAAM,GAAG,UAAU,GAAG,IAAI;IAK5C,OAAO,CAAC,QAAQ;IAShB;;OAEG;IACG,IAAI,IAAI,OAAO,CAAC,IAAI,CAAC;YAQb,OAAO;YAcP,OAAO;IAkBrB,OAAO,CAAC,YAAY;YASN,OAAO;IAyBrB;;;;OAIG;YACW,eAAe;IAe7B,OAAO,CAAC,KAAK;IAIb;;;OAGG;IACH,OAAO,CAAC,YAAY;YAgBN,mBAAmB;IAQjC;;;OAGG;IACH,OAAO,CAAC,sBAAsB;IAa9B,OAAO,CAAC,QAAQ;YAsBF,YAAY;YAWZ,MAAM;IAiEpB,OAAO,CAAC,GAAG;IAMX;;;;;;OAMG;IACG,KAAK,CAAC,GAAG,EAAE,MAAM,GAAG,OAAO,CAAC,OAAO,CAAC;IAwF1C,0DAA0D;IAC1D,WAAW,IAAI,MAAM,EAAE;CAGxB"}
+{"version":3,"file":"app-wake.service.d.ts","sourceRoot":"","sources":["../../../../../../backend/src/services/apps/app-wake.service.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAqDG;AAGH,OAAO,EAAiB,KAAK,eAAe,EAAE,MAAM,2BAA2B,CAAC;AAEhF,OAAO,KAAK,EAAE,mBAAmB,EAAsC,MAAM,4BAA4B,CAAC;AAK1G,2CAA2C;AAC3C,MAAM,WAAW,aAAa;IAC5B,WAAW,IAAI,OAAO,CAAC;IACvB,OAAO,CAAC,CAAC,EAAE,MAAM,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,CAAC,EAAE;QAAE,KAAK,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,GAAG,MAAM,GAAG,SAAS,CAAC,CAAC;QAAC,SAAS,CAAC,EAAE,MAAM,CAAA;KAAE,GAAG,OAAO,CAAC,CAAC,CAAC,CAAC;CAC1I;AAED,+BAA+B;AAC/B,MAAM,WAAW,mBAAmB;IAClC,8EAA8E;IAC9E,QAAQ,EAAE,OAAO,CAAC;IAClB,0FAA0F;IAC1F,KAAK,CAAC,EAAE,OAAO,CAAC;IAChB,8FAA8F;IAC9F,GAAG,CAAC,EAAE,MAAM,CAAC;CACd;AAED,gCAAgC;AAChC,MAAM,WAAW,kBAAkB;IACjC,MAAM,EAAE,aAAa,CAAC;IACtB,QAAQ,EAAE,mBAAmB,CAAC;IAC9B,uFAAuF;IACvF,OAAO,EAAE,CAAC,OAAO,EAAE,MAAM,GAAG,IAAI,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,mBAAmB,KAAK,OAAO,CAAC,OAAO,CAAC,CAAC;IAC/F,wFAAwF;IACxF,YAAY,CAAC,EAAE,CAAC,IAAI,EAAE,MAAM,EAAE,SAAS,EAAE,MAAM,KAAK,OAAO,CAAC,MAAM,GAAG,IAAI,CAAC,CAAC;IAC3E,gDAAgD;IAChD,SAAS,CAAC,EAAE,CAAC,OAAO,EAAE,MAAM,KAAK,OAAO,CAAC;IACzC,8DAA8D;IAC9D,UAAU,EAAE,MAAM,CAAC;IACnB,GAAG,CAAC,EAAE,MAAM,MAAM,CAAC;IACnB,MAAM,CAAC,EAAE,eAAe,CAAC;CAC1B;AAiED,UAAU,UAAU;IAClB,QAAQ,EAAE,MAAM,CAAC;IACjB,MAAM,EAAE,MAAM,CAAC;CAChB;AAED,uEAAuE;AACvE,eAAO,MAAM,aAAa,mBAAuB,CAAC;AAElD;;GAEG;AACH,qBAAa,cAAc;IA0Bb,OAAO,CAAC,QAAQ,CAAC,IAAI;IAzBjC,OAAO,CAAC,QAAQ,CAAC,GAAG,CAAe;IACnC,OAAO,CAAC,QAAQ,CAAC,MAAM,CAAkB;IACzC,OAAO,CAAC,KAAK,CAA8C;IAC3D,OAAO,CAAC,OAAO,CAAS;IACxB,OAAO,CAAC,OAAO,CAA8B;IAC7C,mGAAmG;IACnG,OAAO,CAAC,aAAa,CAAK;IAC1B,OAAO,CAAC,UAAU,CAAK;IACvB,OAAO,CAAC,iBAAiB,CAAS;IAClC,+FAA+F;IAC/F,OAAO,CAAC,QAAQ,CAAC,UAAU,CAA6B;IACxD,wDAAwD;IACxD,OAAO,CAAC,QAAQ,CAAC,OAAO,CAA4B;IACpD,oGAAoG;IACpG,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAA4B;IACrD,OAAO,CAAC,QAAQ,CAAC,UAAU,CAA6B;IACxD,4EAA4E;IAC5E,OAAO,CAAC,QAAQ,CAAC,OAAO,CAA6B;IACrD,4DAA4D;IAC5D,OAAO,CAAC,QAAQ,CAAC,SAAS,CAAkC;IAC5D,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAiC;IACzD,6EAA6E;IAC7E,OAAO,CAAC,QAAQ,CAAC,YAAY,CAAuC;IACpE,OAAO,CAAC,QAAQ,CAAC,YAAY,CAAqB;gBAErB,IAAI,EAAE,kBAAkB;IAKrD,qDAAqD;IACrD,KAAK,IAAI,IAAI;IAMb,gHAAgH;IAChH,IAAI,IAAI,IAAI;IAWZ;;;;;OAKG;IACH,UAAU,CAAC,KAAK,EAAE,MAAM,GAAG,UAAU,GAAG,IAAI;IAK5C,OAAO,CAAC,QAAQ;IAchB;;;OAGG;IACG,IAAI,IAAI,OAAO,CAAC,IAAI,CAAC;YAyBb,OAAO;YAmBP,OAAO;IAyBrB,OAAO,CAAC,aAAa;IAIrB,OAAO,CAAC,YAAY;YASN,OAAO;IA4BrB;;;;OAIG;YACW,eAAe;IAe7B,OAAO,CAAC,KAAK;IAIb;;;OAGG;IACH,OAAO,CAAC,YAAY;YAgBN,mBAAmB;IAQjC;;;OAGG;IACH,OAAO,CAAC,sBAAsB;IAa9B,OAAO,CAAC,QAAQ;YA0BF,YAAY;YAWZ,MAAM;IAiEpB,OAAO,CAAC,GAAG;IAUX,6FAA6F;IAC7F,OAAO,CAAC,MAAM,CAAC,WAAW;IAI1B,wEAAwE;IACxE,OAAO,CAAC,MAAM,CAAC,KAAK;IAiBpB;;;;;;OAMG;IACG,KAAK,CAAC,GAAG,EAAE,MAAM,GAAG,OAAO,CAAC,OAAO,CAAC;IAiC1C,0FAA0F;YAC5E,KAAK;IAcnB;;;;;;OAMG;YACW,IAAI;YAoFJ,SAAS;IAiCvB,uEAAuE;IACvE,OAAO,CAAC,SAAS;IAUjB;;;;;;OAMG;IACH,QAAQ,IAAI,IAAI;IAgChB,0DAA0D;IAC1D,WAAW,IAAI,MAAM,EAAE;CAGxB"}

package/dist/backend/backend/src/services/apps/app-wake.service.js

@@ -27,6 +27,26 @@
  * - A failed delivery keeps the batch (and the cursor before it) and is
  *   retried with backoff; after WAKE_FAILS_BEFORE_ORC_NOTICE failures the
  *   orchestrator is told once.
+ * - Nothing may stall the loop (2026-10-05, 科技晨报: one poll pass never
+ *   finished, the self-rescheduling loop stopped for good, and five owner
+ *   comments sat in Cloud for hours with nothing logged). Each app's poll
+ *   has a deadline (POLL_APP_DEADLINE_MS; a late result is discarded), a
+ *   pass that still has not finished after POLL_PASS_STALL_MS is abandoned
+ *   and a fresh one starts, and a delivery that has not settled after
+ *   WAKE_DELIVER_TIMEOUT_MS counts as failed and is retried (a late success
+ *   still settles the batch).
+ * - A batch being delivered is detached: changes arriving meanwhile open the
+ *   next batch instead of being marked delivered without being in the
+ *   message.
+ * - A watchdog on every scheduled pass re-arms a batch whose timer is gone
+ *   and, once per batch, logs and tells the orchestrator about owner changes
+ *   that have not reached their agent after WAKE_STUCK_NOTICE_MS.
+ * - Owner changes are delivered as owner-authored (`owner: true`, crewly#1105
+ *   queue priority): a busy agent gets them queued ahead of system traffic,
+ *   not after it goes idle.
+ * - Batches live in memory; the persisted cursor stays before the first
+ *   change of every batch not yet delivered, so after a restart the same
+ *   changes are read again and batched again.
  * - An `ask` reaches a named agent only when it is in the publisher's team
  *   and already running; it never starts a stopped agent.
  *
@@ -37,6 +57,32 @@ import { LoggerService } from '../core/logger.service.js';
 import { AppsCloudError } from './apps-cloud.client.js';
 import { buildAppWakeMessage, safeAppName } from './app-wake-message.js';
 const C = CREWLY_APPS_CONSTANTS;
+/** Settles like `promise`, or like `onTimeout()` (its value, or what it throws) after `ms` — whichever comes first. */
+function withDeadline(promise, ms, onTimeout) {
+    let timer = null;
+    const deadline = new Promise((resolve, reject) => {
+        timer = setTimeout(() => {
+            try {
+                resolve(onTimeout());
+            }
+            catch (err) {
+                reject(err);
+            }
+        }, ms);
+        timer.unref?.();
+    });
+    return Promise.race([promise, deadline]).finally(() => {
+        if (timer)
+            clearTimeout(timer);
+    });
+}
+/** Thrown when one app's poll misses its deadline. */
+class PollDeadlineError extends Error {
+    constructor(ms) {
+        super(`poll did not finish within ${Math.round(ms / 1000)} s`);
+        this.name = 'PollDeadlineError';
+    }
+}
 /** Recipient key for the orchestrator (never a valid session name). */
 export const ORC_RECIPIENT = '\u0000orchestrator';
 /**
@@ -49,7 +95,16 @@ export class AppWakeService {
     timer = null;
     running = false;
     ticking = null;
+    /** When the running pass started, and its generation (an abandoned pass stops at its next step) */
+    tickStartedAt = 0;
+    generation = 0;
+    passStallNotified = false;
+    /** Current poll of each app; a poll whose token was replaced or dropped discards its result */
+    pollTokens = new Map();
+    /** Batches collecting changes, by `appId\0recipient` */
     batches = new Map();
+    /** Batches handed to `deliver` and not yet confirmed (in flight, or failed and waiting to retry) */
+    outgoing = new Map();
     lastWakeAt = new Map();
     /** Highest seq read per app (the registry holds what is safe to persist) */
     fetched = new Map();
@@ -77,10 +132,13 @@ export class AppWakeService {
         if (this.timer)
             clearTimeout(this.timer);
         this.timer = null;
-        for (const b of this.batches.values())
+        this.generation++;
+        for (const b of [...this.batches.values(), ...this.outgoing.values()])
             if (b.timer)
                 clearTimeout(b.timer);
         this.batches.clear();
+        this.outgoing.clear();
+        this.pollTokens.clear();
     }
     /**
      * Per-app backoff state (tests).
@@ -97,29 +155,57 @@ export class AppWakeService {
             return;
         this.timer = setTimeout(() => {
             this.timer = null;
-            void this.tick().finally(() => this.schedule());
+            this.watchdog();
+            // The loop never waits on a pass for longer than the stall limit: the
+            // next scheduled pass then abandons it and starts afresh.
+            void withDeadline(this.tick(), C.POLL_PASS_STALL_MS, () => undefined)
+                .catch((err) => this.logger.warn('App change poll pass failed', { error: err instanceof Error ? err.message : String(err) }))
+                .finally(() => this.schedule());
         }, C.POLL_INTERVAL_MS);
         this.timer.unref?.();
     }
     /**
      * One pass over the apps that are due. Exposed for tests and for an immediate poll.
+     * A pass still running after POLL_PASS_STALL_MS is abandoned and a fresh one starts.
      */
     async tick() {
-        if (this.ticking)
-            return this.ticking;
-        this.ticking = this.runTick().finally(() => {
+        if (this.ticking) {
+            const runningFor = this.now() - this.tickStartedAt;
+            if (runningFor < C.POLL_PASS_STALL_MS)
+                return this.ticking;
+            this.logger.warn('App change poll pass did not finish; abandoning it and starting a fresh one', { runningForMs: runningFor });
+            if (!this.passStallNotified) {
+                this.passStallNotified = true;
+                this.notifyOrc(`[APP CHANGES] Crewly's poll of app changes stalled for ${Math.round(runningFor / 60_000)} min and was restarted. ` +
+                    'Owner changes made in that time are being read again now. If this repeats, the backend may need a restart.');
+            }
+            this.generation++;
             this.ticking = null;
+        }
+        const gen = this.generation;
+        this.tickStartedAt = this.now();
+        const pass = this.runTick(gen).finally(() => {
+            if (this.ticking === pass)
+                this.ticking = null;
         });
-        return this.ticking;
+        this.ticking = pass;
+        await pass;
+        if (gen === this.generation)
+            this.passStallNotified = false;
     }
-    async runTick() {
+    async runTick(gen) {
         if (!this.deps.client.isAvailable())
             return;
         const now = this.now();
-        const due = (await this.deps.registry.list()).filter((e) => !e.deleted && (this.backoff.get(e.appId)?.nextAt ?? 0) <= now);
+        const all = await withDeadline(this.deps.registry.list(), C.POLL_APP_DEADLINE_MS, () => null);
+        if (!all) {
+            this.logger.warn('Reading the apps registry did not finish; skipping this poll pass');
+            return;
+        }
+        const due = all.filter((e) => !e.deleted && (this.backoff.get(e.appId)?.nextAt ?? 0) <= now);
         let next = 0;
         const worker = async () => {
-            while (next < due.length) {
+            while (gen === this.generation && next < due.length) {
                 const app = due[next++];
                 await this.pollOne(app);
             }
@@ -127,8 +213,12 @@ export class AppWakeService {
         await Promise.all(Array.from({ length: Math.min(C.POLL_CONCURRENCY, due.length) }, worker));
     }
     async pollOne(app) {
+        const token = Symbol(app.appId);
+        this.pollTokens.set(app.appId, token);
         try {
-            await this.pollApp(app);
+            await withDeadline(this.pollApp(app, token), C.POLL_APP_DEADLINE_MS, () => {
+                throw new PollDeadlineError(C.POLL_APP_DEADLINE_MS);
+            });
             this.backoff.delete(app.appId);
         }
         catch (err) {
@@ -143,6 +233,14 @@ export class AppWakeService {
             this.backoff.set(app.appId, { failures, nextAt: this.now() + wait });
             this.logger.warn('Polling app changes failed', { appId: app.appId, failures, retryInMs: wait, error: err instanceof Error ? err.message : String(err) });
         }
+        finally {
+            // A poll that missed its deadline may still finish later: its token is gone, so it changes nothing.
+            if (this.pollTokens.get(app.appId) === token)
+                this.pollTokens.delete(app.appId);
+        }
+    }
+    isCurrentPoll(appId, token) {
+        return this.pollTokens.get(appId) === token;
     }
     deliveredSet(app) {
         let set = this.delivered.get(app.appId);
@@ -152,11 +250,13 @@ export class AppWakeService {
         }
         return set;
     }
-    async pollApp(app) {
+    async pollApp(app, token) {
         const opts = { timeoutMs: C.POLL_REQUEST_TIMEOUT_MS };
         if (app.cursor === null || app.cursor === undefined) {
             // Never polled: start from now, never replay history.
             const head = await this.deps.client.request('GET', `/apps/${app.appId}/changes`, opts);
+            if (!this.isCurrentPoll(app.appId, token))
+                return;
             this.fetched.set(app.appId, head.seq);
             await this.deps.registry.setProgress(app.appId, head.seq, []);
             return;
@@ -166,6 +266,9 @@ export class AppWakeService {
         let since = Math.max(app.cursor, this.fetched.get(app.appId) ?? app.cursor);
         for (let page = 0; page < C.POLL_MAX_PAGES; page++) {
             const res = await this.deps.client.request('GET', `/apps/${app.appId}/changes`, { ...opts, query: { since, wait: 0 } });
+            // Missed its deadline (or a newer poll of this app started): discard.
+            if (!this.isCurrentPoll(app.appId, token))
+                return;
             const changes = Array.isArray(res.changes) ? res.changes : [];
             for (const change of changes) {
                 if (typeof change.seq === 'number' && change.seq > since && !done.has(change.seq))
@@ -188,7 +291,7 @@ export class AppWakeService {
         if (fetched === undefined)
             return;
         let safe = fetched;
-        for (const b of this.batches.values()) {
+        for (const b of [...this.batches.values(), ...this.outgoing.values()]) {
             if (b.appId === appId)
                 safe = Math.min(safe, b.firstSeq - 1);
         }
@@ -243,7 +346,7 @@ export class AppWakeService {
             return;
         const session = app.agentSession ?? null;
         const key = `${app.appId}\u0000${session ?? ORC_RECIPIENT}`;
-        if (this.batches.has(key))
+        if (this.batches.has(key) || this.outgoing.has(key))
             return;
         const batch = this.newBatch(app.appId, session, false, Number.POSITIVE_INFINITY);
         batch.skippedNotice = true;
@@ -270,6 +373,10 @@ export class AppWakeService {
             failures: 0,
             orcNotified: false,
             skippedNotice: false,
+            openedAt: this.now(),
+            inFlight: false,
+            reportSkipped: null,
+            stuckNotified: false,
         };
     }
     async recipientFor(app, change) {
@@ -366,12 +473,39 @@ export class AppWakeService {
     arm(key, batch, delay) {
         if (batch.timer)
             clearTimeout(batch.timer);
-        batch.timer = setTimeout(() => void this.flush(key), delay);
+        batch.timer = setTimeout(() => {
+            batch.timer = null;
+            if (this.outgoing.get(key) === batch)
+                void this.retry(key);
+            else
+                void this.flush(key);
+        }, Math.max(0, delay));
         batch.timer.unref?.();
     }
+    /** Whether a batch holds only visitor submissions (or only a skipped-submissions notice). */
+    static visitorOnly(batch) {
+        return batch.dataTotal === 0 && batch.eventsTotal === 0 && batch.commentsTotal === 0;
+    }
+    /** Fold `from` (newer) into `into` (older): one message covers both. */
+    static merge(into, from) {
+        const keep = (list, max) => (list.length > max ? list.slice(list.length - max) : list);
+        into.activate = into.activate || from.activate;
+        into.dataChanges = keep([...into.dataChanges, ...from.dataChanges], C.MAX_BATCH_DATA_CHANGES);
+        into.events = keep([...into.events, ...from.events], C.MAX_EVENTS_PER_WAKE);
+        into.visitorChanges = keep([...into.visitorChanges, ...from.visitorChanges], C.MAX_BATCH_DATA_CHANGES);
+        into.comments = keep([...into.comments, ...from.comments], C.COMMENTS.MAX_PER_WAKE);
+        into.dataTotal += from.dataTotal;
+        into.eventsTotal += from.eventsTotal;
+        into.visitorTotal += from.visitorTotal;
+        into.commentsTotal += from.commentsTotal;
+        into.seqs = keep([...into.seqs, ...from.seqs], C.MAX_DELIVERED_SEQS);
+        into.firstSeq = Math.min(into.firstSeq, from.firstSeq);
+        into.skippedNotice = into.skippedNotice && from.skippedNotice;
+        into.openedAt = Math.min(into.openedAt, from.openedAt);
+    }
     /**
-     * Send one batch now (timer callback; tests). On failure the batch stays
-     * pending and is retried with backoff.
+     * Send one collecting batch now (timer callback; tests). On failure the
+     * batch stays pending and is retried with backoff.
      *
      * @param key - Batch key
      * @returns Whether a message was delivered
@@ -383,23 +517,71 @@ export class AppWakeService {
         if (batch.timer)
             clearTimeout(batch.timer);
         batch.timer = null;
-        const app = await this.deps.registry.get(batch.appId);
-        const recipient = batch.session ?? ORC_RECIPIENT;
+        // The previous batch for this agent is still out: fold this one into it
+        // when it is waiting to retry (its retry carries both), else check again
+        // shortly.
+        const out = this.outgoing.get(key);
+        if (out) {
+            if (out.inFlight) {
+                this.arm(key, batch, C.WAKE_INFLIGHT_RECHECK_MS);
+                return false;
+            }
+            this.batches.delete(key);
+            AppWakeService.merge(out, batch);
+            if (!out.timer)
+                this.arm(key, out, 0);
+            return false;
+        }
         // Visitor submissions alone never start the agent: wait until it runs.
-        const visitorOnly = batch.dataTotal === 0 && batch.eventsTotal === 0 && batch.commentsTotal === 0;
-        if (visitorOnly && batch.session !== null && this.deps.isRunning && !this.deps.isRunning(batch.session)) {
+        if (AppWakeService.visitorOnly(batch) && batch.session !== null && this.deps.isRunning && !this.deps.isRunning(batch.session)) {
             if (this.batches.get(key) === batch)
                 this.arm(key, batch, C.VISITOR_WAKE.PENDING_RECHECK_MS);
             return false;
         }
+        // Detach: changes arriving while this one is delivered open the next batch.
+        this.batches.delete(key);
+        this.outgoing.set(key, batch);
+        return this.send(key, batch);
+    }
+    /** Retry an outgoing batch (its retry timer), folding in whatever collected meanwhile. */
+    async retry(key) {
+        const batch = this.outgoing.get(key);
+        if (!batch || batch.inFlight)
+            return false;
+        if (batch.timer)
+            clearTimeout(batch.timer);
+        batch.timer = null;
+        const newer = this.batches.get(key);
+        if (newer) {
+            if (newer.timer)
+                clearTimeout(newer.timer);
+            this.batches.delete(key);
+            AppWakeService.merge(batch, newer);
+        }
+        return this.send(key, batch);
+    }
+    /**
+     * Deliver an outgoing batch once, bounded by WAKE_DELIVER_TIMEOUT_MS.
+     *
+     * @param key - Batch key
+     * @param batch - The batch (already in `outgoing`)
+     * @returns Whether it was delivered
+     */
+    async send(key, batch) {
+        batch.inFlight = true;
+        const app = await this.deps.registry.get(batch.appId).catch(() => null);
         // Skipped visitor submissions are reported in the next message to the
         // agent visitors wake (the publisher, else the orchestrator).
         const isVisitorRecipient = (app?.agentSession ?? null) === batch.session;
         const vstate = app && isVisitorRecipient ? this.visitorState(app) : null;
-        const skipped = vstate?.skipped ?? 0;
-        if (visitorOnly && batch.visitorTotal === 0 && skipped === 0) {
+        if (batch.reportSkipped === null)
+            batch.reportSkipped = vstate?.skipped ?? 0;
+        const skipped = batch.reportSkipped;
+        if (AppWakeService.visitorOnly(batch) && batch.visitorTotal === 0 && skipped === 0) {
             // A notice-only batch with nothing left to report.
-            this.batches.delete(key);
+            batch.inFlight = false;
+            if (this.outgoing.get(key) === batch)
+                this.outgoing.delete(key);
             return false;
         }
         const text = buildAppWakeMessage({
@@ -417,32 +599,62 @@ export class AppWakeService {
             visitorSkipped: skipped,
             skillsPath: this.deps.skillsPath,
         });
-        let ok = false;
-        try {
-            ok = await this.deps.deliver(batch.session, text, { activate: batch.activate });
-        }
-        catch (err) {
+        const owner = !AppWakeService.visitorOnly(batch);
+        const seqs = batch.seqs.filter((n) => Number.isFinite(n));
+        const ref = seqs.length > 0 ? `app:${batch.appId}:${Math.min(...seqs)}-${Math.max(...seqs)}:${seqs.length}` : undefined;
+        const sentSeqs = new Set(batch.seqs);
+        const attempt = Promise.resolve()
+            .then(() => this.deps.deliver(batch.session, text, { activate: batch.activate, ...(owner ? { owner: true } : {}), ...(ref ? { ref } : {}) }))
+            .catch((err) => {
             this.logger.warn('App change wake failed', { appId: batch.appId, session: batch.session, error: err instanceof Error ? err.message : String(err) });
-        }
-        if (!ok) {
-            batch.failures++;
-            const retry = Math.min(C.WAKE_RETRY_BASE_MS * 2 ** (batch.failures - 1), C.WAKE_RETRY_MAX_MS);
-            this.logger.warn('App change wake was not delivered; will retry', { appId: batch.appId, session: batch.session ?? 'orchestrator', failures: batch.failures, retryInMs: retry });
-            if (batch.failures >= C.WAKE_FAILS_BEFORE_ORC_NOTICE && !batch.orcNotified && batch.session !== null) {
-                batch.orcNotified = true;
-                const notice = `[APP CHANGES] Changes in app "${safeAppName(app?.name ?? batch.appId)}" (${batch.appId}) could not be delivered to ` +
-                    `${batch.session} after ${batch.failures} tries. Crewly keeps retrying. Check whether that agent is stuck or signed out.`;
-                await this.deps.deliver(null, notice, { activate: false }).catch(() => false);
-            }
-            if (this.batches.get(key) === batch)
-                this.arm(key, batch, retry);
             return false;
+        });
+        const outcome = await withDeadline(attempt.then((ok) => (ok ? 'ok' : 'failed')), C.WAKE_DELIVER_TIMEOUT_MS, () => 'timeout');
+        batch.inFlight = false;
+        if (this.outgoing.get(key) !== batch)
+            return false; // stopped meanwhile
+        if (outcome === 'ok')
+            return this.succeeded(key, batch, app, vstate, skipped);
+        if (outcome === 'timeout') {
+            this.logger.warn('App change wake did not finish in time; will retry', {
+                appId: batch.appId,
+                session: batch.session ?? 'orchestrator',
+                timeoutMs: C.WAKE_DELIVER_TIMEOUT_MS,
+            });
+            // A late success still settles the batch, unless a retry is already out
+            // or the batch has grown since (then the retry carries everything).
+            void attempt.then((ok) => {
+                if (!ok || this.outgoing.get(key) !== batch || batch.inFlight)
+                    return;
+                if (!batch.seqs.every((n) => sentSeqs.has(n)))
+                    return;
+                this.logger.info('A timed-out app change wake arrived after all', { appId: batch.appId, session: batch.session ?? 'orchestrator' });
+                if (batch.timer)
+                    clearTimeout(batch.timer);
+                batch.timer = null;
+                void this.succeeded(key, batch, app, vstate, skipped);
+            });
         }
-        this.batches.delete(key);
+        batch.failures++;
+        const retry = Math.min(C.WAKE_RETRY_BASE_MS * 2 ** (batch.failures - 1), C.WAKE_RETRY_MAX_MS);
+        this.logger.warn('App change wake was not delivered; will retry', { appId: batch.appId, session: batch.session ?? 'orchestrator', failures: batch.failures, retryInMs: retry });
+        if (batch.failures >= C.WAKE_FAILS_BEFORE_ORC_NOTICE && !batch.orcNotified && batch.session !== null) {
+            batch.orcNotified = true;
+            this.notifyOrc(`[APP CHANGES] Changes in app "${safeAppName(app?.name ?? batch.appId)}" (${batch.appId}) could not be delivered to ` +
+                `${batch.session} after ${batch.failures} tries. Crewly keeps retrying. Check whether that agent is stuck or signed out.`);
+        }
+        this.arm(key, batch, retry);
+        return false;
+    }
+    async succeeded(key, batch, app, vstate, skipped) {
+        if (this.outgoing.get(key) !== batch)
+            return false;
+        this.outgoing.delete(key);
+        const recipient = batch.session ?? ORC_RECIPIENT;
         const at = this.now();
         this.lastWakeAt.set(key, at);
         await this.deps.registry.setLastWake(batch.appId, recipient, at).catch(() => undefined);
-        if (vstate) {
+        if (vstate && app) {
             if (batch.visitorTotal > 0 || batch.skippedNotice)
                 vstate.count++;
             vstate.skipped = Math.max(0, vstate.skipped - skipped);
@@ -461,13 +673,62 @@ export class AppWakeService {
             comments: batch.commentsTotal,
             visitorSubmissions: batch.visitorTotal,
             ...(skipped > 0 ? { visitorSkippedReported: skipped } : {}),
+            ...(batch.failures > 0 ? { afterFailures: batch.failures } : {}),
         });
+        // Changes that arrived during the delivery wait for the cooldown it started.
+        const next = this.batches.get(key);
+        if (next)
+            this.arm(key, next, C.COOLDOWN_MS);
         await this.persistProgress(batch.appId).catch(() => undefined);
         return true;
     }
+    /** Tell the orchestrator something, without waiting on it for long. */
+    notifyOrc(text) {
+        void withDeadline(Promise.resolve()
+            .then(() => this.deps.deliver(null, text, { activate: false }))
+            .catch(() => false), C.WAKE_DELIVER_TIMEOUT_MS, () => false);
+    }
+    /**
+     * Runs before every scheduled pass: a batch with no timer and no delivery
+     * running is re-armed (it would otherwise wait forever), and owner changes
+     * that have not reached their agent after WAKE_STUCK_NOTICE_MS are logged
+     * and reported to the orchestrator, once per batch. Visitor-only batches
+     * waiting for their agent to run are expected to wait and are left alone.
+     */
+    watchdog() {
+        const now = this.now();
+        for (const [key, batch] of [...this.batches.entries(), ...this.outgoing.entries()]) {
+            if (!batch.timer && !batch.inFlight) {
+                this.logger.warn('App change batch had no timer; re-armed', { appId: batch.appId, session: batch.session ?? 'orchestrator' });
+                this.arm(key, batch, 0);
+            }
+            if (AppWakeService.visitorOnly(batch) || batch.stuckNotified)
+                continue;
+            const ageMs = now - batch.openedAt;
+            if (ageMs < C.WAKE_STUCK_NOTICE_MS)
+                continue;
+            batch.stuckNotified = true;
+            const minutes = Math.round(ageMs / 60_000);
+            this.logger.warn('Owner changes in an app have not reached their agent', {
+                appId: batch.appId,
+                session: batch.session ?? 'orchestrator',
+                minutes,
+                inFlight: batch.inFlight,
+                failures: batch.failures,
+                comments: batch.commentsTotal,
+                dataChanges: batch.dataTotal,
+                events: batch.eventsTotal,
+            });
+            if (batch.session !== null) {
+                this.notifyOrc(`[APP CHANGES] The owner's changes in an app (${batch.appId}: ${batch.commentsTotal} comment(s), ${batch.dataTotal} data change(s), ` +
+                    `${batch.eventsTotal} message(s)) have waited ${minutes} min to reach ${batch.session}. Crewly keeps trying. ` +
+                    'Check whether that agent is stuck.');
+            }
+        }
+    }
     /** @returns Keys of batches waiting to be sent (tests) */
     pendingKeys() {
-        return [...this.batches.keys()];
+        return [...new Set([...this.batches.keys(), ...this.outgoing.keys()])];
     }
 }
 //# sourceMappingURL=app-wake.service.js.map