perf(stagehand): take the hot list from the broker, without REST checks

E2E_USER_STORIES.md

@@ -455,10 +455,11 @@ runs `source` and `mosquitto` for this, and the specs drive state with a purpose
 `drivers/place/e2e/state_harness.cr` (a UDP device whose `connected` and other statuses are
 set through `exec`). See "MQTT" in `e2e/README.md`.
 
-**The broker is asserted directly, not only through the page.** Stagehand re-checks every
-MQTT alert over REST before showing it (`DashboardsService._updateDashboardAlertList`), which
-hides wrong data on the broker. Removing that re-check — it is what makes stagehand slow — is
-only safe once the MQTT rows below are all green.
+**The broker is asserted directly, not only through the page.** Stagehand builds the hot list
+from the broker's retained state alone: the latest value of each topic, with an emptied topic
+removed. It used to re-check every alert over REST (a module list and a state read per alert, on
+every message), which made it slow and hid wrong data on the broker; that went once the MQTT rows
+below were green.
 
 | ID | P | Story | Status |
 |----|---|-------|--------|
@@ -479,7 +480,8 @@ only safe once the MQTT rows below are all green.
 | SH-MQTT-19 | P0 | Two quick edits to one system leave source's mapping matching the latest. | **done** — the lifecycle setup makes them (two modules added back to back). pg-orm and placeos-resource handle each change on its own fiber, so source processed them out of order and kept the older module list. Found while building this coverage. |
 | SH-MQTT-20 | P1 | After a broker restart, every status is published again. | **done, not automated** — verified by hand (restart mosquitto: 32 of 32 topics back within 2 s; before the fix, 0 until each value changed). |
 | SH-MQTT-21 | P1 | Retained topics nothing maps to (left by a module deleted while source was down, or by an older source) are removed when source starts. | **done, not automated** — verified by hand with a planted topic. source spec `sweeps retained topics that nothing maps to`. |
-| SH-HOT-02 | P1 | A module added to a system after stagehand has loaded it can still raise an alert. | **done** — `local/stagehand-hotlist.spec.ts`. Stagehand cached each system's module list forever (`_system_modules_cache`); it now refetches, at most every 5 s, before dropping an alert for an unknown device. |
+| SH-HOT-02 | P1 | A module added to a system after stagehand has loaded it can still raise an alert. | **done** — `local/stagehand-hotlist.spec.ts`. Was a module list cached forever by the REST re-check, which is gone. |
+| SH-HOT-03 | P0 | A deleted module leaves the hot list, the module after it is listed at its index, and the page makes no per-alert REST calls. | **done** — `local/stagehand-hotlist.spec.ts`. Red against the REST re-checking stagehand: an emptied topic failed to parse, so the deleted module stayed listed. |
 
 **The lifecycle specs need source v1.22.1 or later** (PlaceOS/source#65). Against
 `placeos-2.2610.1` SH-MQTT-10 … 19 and SH-HOT-02 are red, which is what they are for. Until a

apps/stagehand/e2e/local/stagehand-hotlist.spec.ts

@@ -6,9 +6,9 @@
  *   harness reports connected: false                -> listed
  *   harness reports connected: true                 -> gone again
  *
- * The broker is asserted alongside the page at every step. When stagehand
- * stops re-checking alerts over REST, this spec is what says the page is still
- * right; until then, the broker assertions are what say the data is.
+ * The broker is asserted alongside the page at every step. Stagehand takes the
+ * hot list straight from the broker's retained state, with no REST checks, so
+ * these are what say the page follows the broker.
  *
  * "Not listed" needs a control (see e2e/README.md, Conventions): the page shows
  * a spinner, not an empty table, until its first alert. So the system also
@@ -29,6 +29,7 @@ import { expect, test } from '../../../../e2e/support/stagehand/fixtures';
 import {
     addHarness,
     createSystem,
+    deleteModule,
     deleteSystem,
     setConnected,
     sweepSystems,
@@ -84,10 +85,6 @@ test('a module is listed while it reports connected: false, and only then', asyn
             MQTT_PROPAGATION_MS,
         );
 
-        // NOTE: both modules exist before the system's first alert. Stagehand
-        // caches a system's module list on its first alert and never refreshes
-        // it, so a module added after that cannot alert until a reload
-        // (SH-HOT-02). This spec is about the working path.
         await setConnected(adminApi, control.id, false);
         await broker.waitForValue(
             'Projector',
@@ -194,3 +191,65 @@ test('SH-HOT-02: a module added after the page loaded the system can still be li
         await deleteSystem(adminApi, system.id);
     }
 });
+
+test('SH-HOT-03: a deleted module leaves the list, and the module after it takes its index, without REST checks', async ({
+    page,
+    adminApi,
+    stagehandAdminToken,
+    harnessDriverId,
+    zones,
+}) => {
+    const system = await createSystem(adminApi, zones, 'delete-listed');
+    const broker = await BrokerView.open(
+        stagehandAdminToken,
+        systemFilter(zones.org, system.id),
+    );
+    // the per-alert checks stagehand used to make: a system's modules, and a
+    // module's state
+    const rest_checks: string[] = [];
+    page.on('request', (r) => {
+        const url = r.url();
+        if (
+            /\/api\/engine\/v2\/modules\?.*control_system_id=/.test(url) ||
+            new RegExp(`/api/engine/v2/systems/${system.id}/[^/?]+_\\d+`).test(
+                url,
+            )
+        ) {
+            rest_checks.push(`${r.method()} ${url}`);
+        }
+    });
+    try {
+        await openHotList(page);
+
+        const first = await addHarness(adminApi, system.id, harnessDriverId);
+        const second = await addHarness(adminApi, system.id, harnessDriverId);
+        await waitForConnected(adminApi, first.id, true);
+        await waitForConnected(adminApi, second.id, true);
+        await setConnected(adminApi, first.id, false);
+        await setConnected(adminApi, second.id, false);
+        await expectHotList(page, system.id, ['Display_1', 'Display_2']);
+
+        // the second becomes Display_1; nothing is left at Display_2
+        await deleteModule(adminApi, first.id);
+        await expectHotList(page, system.id, ['Display_1']);
+
+        // and it is the second module listed there, so reconnecting it clears it
+        await setConnected(adminApi, second.id, true);
+        await broker.waitForValue(
+            'Display',
+            1,
+            'connected',
+            true,
+            MQTT_PROPAGATION_MS,
+        );
+        await expectHotList(page, system.id, []);
+
+        expect(rest_checks).toEqual([]);
+    } finally {
+        await test
+            .info()
+            .attach('broker messages', { body: broker.describeLog() });
+        await broker.close();
+        await deleteSystem(adminApi, system.id);
+    }
+});