fix: idle quiet watch for disconnected-only workers

Keep disconnected observations for episode and reminder bookkeeping without admitting recurring sweeps. Existing topology and status changes wake discovery on reconnect. Cover disconnected fleets, reconnect grace and budgets, mixed workers, and teardown through the real watch.

Co-authored-by: Taras Kornichuk <dread9ko@gmail.com>

src/main/runtime/orchestration/workers/reports/worker-quiet-admission.test.ts

@@ -1,6 +1,8 @@
 import { expect, it, vi } from 'vitest'
 import { runtimeFixture } from './worker-quiet-runtime.test-fixture'
 import { startWorkerQuietWatch } from './worker-quiet-watch'
+import { TerminalIdentitySource } from '../../../terminal/identity/terminal-identity-source'
+import { PANE } from '../../../../agent-hooks/server/testing/server.test-fixtures'
 
 const fixture = runtimeFixture()
 
@@ -64,6 +66,63 @@ it.each([
   }
 )
 
+it.each([100, 1000])(
+  'leaves %i disconnected working workers idle until reconnect wakes discovery',
+  (population) => {
+    const { db, chat, changes } = fixture
+    const { run } = fixture.workers(population)
+    chat.handles.clear()
+    const terminals = new TerminalIdentitySource(changes.notify)
+    for (let n = 0; n < population; n++) {
+      const paneKey = n === 0 ? PANE : `offline-pane-${n}`
+      const pty = fixture.terminal(paneKey, `term_worker_${n}`)
+      terminals.createPty({ ...pty, ptyId: paneKey, connected: false })
+    }
+    fixture.store.ingestRemote(
+      {
+        paneKey: PANE,
+        source: 'codex',
+        hookEventName: 'UserPromptSubmit',
+        payload: { state: 'working', prompt: 'work', agentType: 'codex' }
+      },
+      'ssh-owner'
+    )
+    const inventory = vi.fn(() => terminals.ptys.keys())
+    const local = vi.spyOn(db, 'getActiveDispatchForTerminal')
+    fixture.start({
+      readPaneKeys: inventory,
+      readPty: (key) => terminals.ptys.get(key) ?? null
+    })
+    // One startup pass retains each episode; none can produce a notice while disconnected.
+    expect(inventory).toHaveBeenCalledTimes(1)
+    expect(local).toHaveBeenCalledTimes(population)
+    vi.advanceTimersByTime(3_600_000)
+    expect(inventory).toHaveBeenCalledTimes(1)
+    expect(local).toHaveBeenCalledTimes(population)
+    expect(db.getUnreadRunMailbox(run.id, 100, ['status'])).toHaveLength(0)
+    const paneKey = PANE
+    const pty = terminals.ptys.get(paneKey)!
+    terminals.setPtyConnected(pty, true)
+    vi.advanceTimersByTime(0)
+    expect(inventory).toHaveBeenCalledTimes(2)
+    expect(local).toHaveBeenCalledTimes(population * 2)
+    vi.advanceTimersByTime(840_000)
+    expect(db.getUnreadRunMailbox(run.id, 100, ['status'])).toHaveLength(0)
+    vi.advanceTimersByTime(60_000)
+    const notices = db.getUnreadRunMailbox(run.id, 100, ['status'])
+    expect(notices).toHaveLength(1)
+    expect(JSON.parse(notices[0]!.payload!).quietNotice.observedAt).toBe(4_500_000)
+    terminals.setPtyConnected(pty, false)
+    fixture.stop()
+    inventory.mockClear()
+    local.mockClear()
+    terminals.setPtyConnected(pty, true)
+    vi.advanceTimersByTime(3_600_000)
+    expect(inventory).not.toHaveBeenCalled()
+    expect(local).not.toHaveBeenCalled()
+  }
+)
+
 it('admits a newly working chat within one interval without event-driven postponement', async () => {
   const { db, chat, changes } = fixture
   const { run } = fixture.workers(1)

src/main/runtime/orchestration/workers/reports/worker-quiet-attachments.test.ts

@@ -102,46 +102,55 @@ it('recovers acknowledged remote notices only for the current live process after
   expect(rows[1]!.message_id).not.toBe(first.message_id)
 })
 
-it('preserves a disconnected worker episode, excluding offline time and retaining its notice budget', () => {
-  const { run, workers } = fixture.workers(2)
-  fixture.chat.handles.delete('term_worker_0')
-  const pty = fixture.terminal(PANE, 'term_worker_0')
-  fixture.store.ingestRemote(
-    {
-      paneKey: PANE,
-      source: 'codex',
-      hookEventName: 'UserPromptSubmit',
-      payload: { state: 'working', prompt: 'work', agentType: 'codex' }
-    },
-    'ssh-owner'
-  )
-  expect(fixture.store.getStatusSnapshotForPane(PANE)[0]).toMatchObject({
-    state: 'working',
-    turnStartedAt: 0
-  })
-  const noticeTimes = () =>
-    fixture.db
-      .getUnreadRunMailbox(run.id, 100, ['status'])
-      .map((message) => JSON.parse(message.payload!).quietNotice)
-      .filter((notice) => notice.dispatchId === workers[0]!.id)
-      .map((notice) => notice.observedAt)
-  fixture.start()
-  vi.advanceTimersByTime(960_000)
-  expect(noticeTimes()).toEqual([900_000])
-  fixture.panes.set(PANE, { ...pty, connected: false })
-  fixture.changes.notify(new Set(['status']))
-  vi.advanceTimersByTime(1_200_000)
-  fixture.panes.set(PANE, { ...pty, connected: true })
-  fixture.changes.notify(new Set(['status']))
-  vi.advanceTimersByTime(840_000)
-  expect(noticeTimes()).toEqual([900_000])
-  vi.advanceTimersByTime(960_000)
-  expect(noticeTimes()).toEqual([900_000, 3_900_000])
-  fixture.panes.set(PANE, { ...pty, connected: false })
-  fixture.changes.notify(new Set(['status']))
-  vi.advanceTimersByTime(2_040_000)
-  fixture.panes.set(PANE, { ...pty, connected: true })
-  fixture.changes.notify(new Set(['status']))
-  vi.advanceTimersByTime(7_200_000)
-  expect(noticeTimes()).toEqual([900_000, 3_900_000])
-})
+it.each([
+  [0, 3_840_000],
+  [1, 3_900_000]
+])(
+  'preserves a disconnected episode and budget with %i other working chats',
+  (chats, reminderAt) => {
+    const { run, workers } = fixture.workers(1 + chats)
+    fixture.chat.handles.delete('term_worker_0')
+    const pty = fixture.terminal(PANE, 'term_worker_0')
+    fixture.store.ingestRemote(
+      {
+        paneKey: PANE,
+        source: 'codex',
+        hookEventName: 'UserPromptSubmit',
+        payload: { state: 'working', prompt: 'work', agentType: 'codex' }
+      },
+      'ssh-owner'
+    )
+    expect(fixture.store.getStatusSnapshotForPane(PANE)[0]).toMatchObject({
+      state: 'working',
+      turnStartedAt: 0
+    })
+    const noticeTimes = () =>
+      fixture.db
+        .getUnreadRunMailbox(run.id, 100, ['status'])
+        .map((message) => JSON.parse(message.payload!).quietNotice)
+        .filter((notice) => notice.dispatchId === workers[0]!.id)
+        .map((notice) => notice.observedAt)
+    const inventory = vi.fn(() => fixture.panes.keys())
+    fixture.start({ readPaneKeys: inventory })
+    vi.advanceTimersByTime(960_000)
+    expect(noticeTimes()).toEqual([900_000])
+    fixture.panes.set(PANE, { ...pty, connected: false })
+    inventory.mockClear()
+    fixture.changes.notify(new Set(['status']))
+    vi.advanceTimersByTime(1_200_000)
+    expect(inventory).toHaveBeenCalledTimes(chats ? 20 : 1)
+    fixture.panes.set(PANE, { ...pty, connected: true })
+    fixture.changes.notify(new Set(['status']))
+    vi.advanceTimersByTime(840_000)
+    expect(noticeTimes()).toEqual([900_000])
+    vi.advanceTimersByTime(960_000)
+    expect(noticeTimes()).toEqual([900_000, reminderAt])
+    fixture.panes.set(PANE, { ...pty, connected: false })
+    fixture.changes.notify(new Set(['status']))
+    vi.advanceTimersByTime(2_040_000)
+    fixture.panes.set(PANE, { ...pty, connected: true })
+    fixture.changes.notify(new Set(['status']))
+    vi.advanceTimersByTime(7_200_000)
+    expect(noticeTimes()).toEqual([900_000, reminderAt])
+  }
+)

src/main/runtime/orchestration/workers/reports/worker-quiet-runtime.ts

@@ -63,7 +63,7 @@ export function startWorkerQuietSweep(
       Date.now(),
       [...candidates.keys()]
     )
-    return workers.length > 0
+    return workers.some(({ observation }) => observation.connected !== false)
   }
   return startWorkerQuietWatch({
     sweep: run,