[Bug]: A hung desktop main process leaves no record and blocks macOS logout until force-quit #21301

Before submitting

• I searched existing issues and did not find a duplicate. Searched "quit hangs", "blocks logout", "not responding", "crashReporter". Related: [Bug]: Linux/Wayland desktop main process aborts with V8 heap OOM (SIGTRAP), no preview/CDP activity #21189 and [Bug]: macOS main process aborts with V8 OOM in JSON.stringify while relaying preview CDP events (SIGTRAP) #20395 (main-process V8 OOM aborts); [Bug]: Concurrent agent test runs exhaust memory and crash macOS; agent memory limits are Linux-only #20329 and When agents keep memory pressure high, stop one agent's command instead of stalling every agent #21288 (no agent memory limits on macOS, which caused the pressure here); [Bug]: Windows desktop typing lags even in empty chats; browser interface is smooth #12689 (the renderer's synchronous topology IPC, which is why the whole UI froze along with main).
• I included enough detail to reproduce or investigate the problem.

Area

apps/desktop

Steps to reproduce

Not deterministic. Seen on a 16 GB Mac under critical memory pressure (28 GB swap) while ~12 agent threads ran test suites on the app's local server:

1. The desktop main process stops making progress.
2. Log out of macOS. Logout waits on T3 Code until you force-quit it.
3. Afterwards, nothing in the app's logs records the hang.

Untested check for the missing record: kill -STOP main for 60 s, then kill -CONT; desktop.trace.ndjson should show only a gap.

Expected behavior

• A stuck main process leaves a record: a stall entry, a stack or a crash dump.
• Quitting or logging out finishes within a bounded time.

Actual behavior

• There is no crash report, hang report or dump.
• The embedded server process logged event-loop stalls of 54 s, 54 s and 420 s, the last ending at 14:16:43.
• After the workload was stopped (~14:19), main stayed silent for 36 more minutes.

Root cause of the hang: unknown (no stack was captured). Unverified candidates:

• V8 heap thrashing through swap (compare [Bug]: Linux/Wayland desktop main process aborts with V8 heap OOM (SIGTRAP), no preview/CDP activity #21189 and [Bug]: macOS main process aborts with V8 OOM in JSON.stringify while relaying preview CDP events (SIGTRAP) #20395).
• A blocking native call on the main thread, e.g. the telemetry sampler's synchronous getSystemIdleTime/getSystemIdleState/isOnBatteryPower/getAppMetrics (apps/desktop/src/telemetry/DesktopTelemetryPublisher.ts:207-216).
• A starved libuv pool, which would stop all async log writes even if JS kept running.

What the code does. Line numbers are at the tag and match main @ 3adac94 unless noted.

Quit has no overall deadline:

• before-quit is cancelled (app/DesktopLifecycle.ts:116), then requestDesktopShutdownAndWait (:84-94) waits on flushMainWindowBounds, a file-write join (window/DesktopWindow.ts:474-481; :489-496 on main), and on shutdown.awaitComplete.
• Completion comes only after stopAllPoolInstances (app/DesktopApp.ts:341-351). Its 5 s timeout covers only closeRun: stop waits for the instance mutex untimed (backend/DesktopBackendManager.ts:1017), and the layer finalizer calls stop() with no timeout (:1121).
• If the JS thread itself is blocked, none of this runs. The records can't say which case this was.

There are no hang or crash diagnostics:

• Only render-process-gone is handled (window/DesktopWindow.ts:755; :777 on main).
• There is no unresponsive listener and no crashReporter.
• apps/desktop has no event-loop monitor (the server's apps/server/src/observability/EventLoopMonitor.ts produced the stall records above).

Suggested diagnostics:

• Watchdog: a worker_thread that main heartbeats every second. After 10 s without a heartbeat it records the time, last span, event-loop delay and last heap/RSS, then runs /usr/bin/sample <mainPid> 5 -file … once. The resulting native and JS frames would tell the candidates apart.
• Abort coverage: unresponsive logging plus a local-only crashReporter.
• Quit deadline: about 10 s, then app.exit(), so logout cannot be blocked.

Impact

Major degradation or frequent failure

Version or commit

T3 Code (Nightly) 0.1.1-nightly.20261018.3315. The same code is on main @ 3adac94.

Environment

macOS 27.0.1 (26A434), MacBook Pro M5, 16 GB. The desktop app was connected to its local server and to two Linux servers over T3 Connect.

Logs or stack traces

2026-10-18 13:59:15.490 WindowServer: pid 61441 failed to act on a ping it dequeued before timing out.
2026-10-18 14:53:35.677 loginwindow: -[GUIApplicationLogout loadProccessesToQuitArray] | Adding: T3 Code (Nightly) to gui quit list
2026-10-18 14:55:03.015 loginwindow: T3 Code (Nightly) [61441] force quit (caller responsible for termination)
2026-10-18 14:55:09.070 runningboardd: [app<application.com.t3tools.t3code...>:61441] termination reported by launchd (2, 9, 9)

Workaround

Force quit. Before force-quitting, run sample <mainPid> 10 -file ~/Desktop/t3-main.txt to capture the stuck stack.