Repository navigation
Description
Summary
After the in-app auto-update relaunched Orca (1.4.187 → 1.4.191), the new main process hung permanently within its first second of life. The window renders but shows only "no workspace found", the app cannot be quit via Cmd+Q or the Dock (the main thread is dead), and Orca's own hang watchdog recorded the hang with selfRecovered: false.
A sample of the hung process shows the cause: a main-thread self-deadlock inside AppKit/FrontBoardServices scene-settings handling for the NSStatusItem (menu-bar item). While the main thread is executing a callout from the FBSWorkspace callout queue, an NSStatusItem display update triggers an NSWindow frame change, whose notification re-enters -[FBSScene _updateClientSettings:], which calls -[BSServiceDispatchQueue performAsyncAndWait:] back onto the queue the thread is already running a callout from → __DISPATCH_WAIT_FOR_QUEUE__ blocks forever.
All 2,292 samples (3s @ 1ms) show this identical stack; no other thread is doing anything (NSEventThread idle in mach_msg, all Chromium/libuv workers parked). This is an OS-framework re-entrancy deadlock, but it is presumably triggered by how/when Orca creates or updates its status item during startup window setup, and it makes the post-update relaunch a brick.
"No workspace found" appears to be a symptom, not the fault: the renderer came up, but the frozen main process never answered its workspace queries.
Environment
- Orca 1.4.191 (updated from 1.4.187 via in-app updater), packaged build
- Electron 43.1.0 (per crashpad annotations)
- macOS 26.5.1 (25F80), Darwin 25.5.0, Apple Silicon (ARM64)
Timeline (from logs/main.trace.ndjson crash breadcrumbs, local time)
| Time | Event |
|---|---|
| 13:34:58 | updater_will_quit_cleanup_started — old 1.4.187 process quits for the update (daemonTeardown: "disconnect") |
| 13:35:24 | New 1.4.191 process launches (per sample header launch time) |
| 13:35:36 | main_process_lifecycle_started breadcrumb written; skill.recovery span completes 67 ms later — these are the last two trace entries the process ever writes |
| 13:35:36 | Daemon accepts the new app's control/stream hellos (daemon itself, still running the 1.4.187 entry script, is unaffected and healthy) |
| 13:36:26 | Watchdog writes main-thread-hang.json |
| +6 min | Still frozen when sampled; identical stack across the whole sample window |
main-thread-hang.json:
{"detectedAt":1787942186051,"parentPid":<pid>,"unresponsiveMs":49715,"selfRecovered":false}Main-thread stack (from sample, 2292/2292 samples identical)
Full main-thread call graph
2292 Thread_16379519 DispatchQueue_1: com.apple.main-thread (serial)
+ 2292 start (in dyld) + 6992 [0x189c97e00]
+ 2292 ElectronMain (in Electron Framework) + 200 [0x10e30dc5c]
+ 2292 ElectronMain (in Electron Framework) + 1832 [0x10e30e2bc]
+ 2292 v8::Context::UseDefaultSecurityToken() (in Electron Framework) + 5272 [0x10de1f030]
+ 2292 node::LibuvStreamWrap::HasDoTryWrite() const (in Electron Framework) + 55700 [0x10e329814]
+ 2292 v8::internal::OptimizingCompileTaskExecutor::EnsureStarted() (in Electron Framework) + 5632 [0x10db93d40]
+ 2292 v8::internal::OptimizingCompileTaskExecutor::EnsureStarted() (in Electron Framework) + 8952 [0x10db94a38]
+ 2292 v8::internal::OptimizingCompileTaskExecutor::EnsureStarted() (in Electron Framework) + 9180 [0x10db94b1c]
+ 2292 v8::internal::OptimizingCompileTaskExecutor::EnsureStarted() (in Electron Framework) + 9728 [0x10db94d40]
+ 2292 v8::SharedMemoryStatistics::SharedMemoryStatistics() (in Electron Framework) + 122952 [0x10db14634]
+ 2292 v8::SharedMemoryStatistics::SharedMemoryStatistics() (in Electron Framework) + 124524 [0x10db14c58]
+ 2292 uv__getaddrinfo_translate_error (in Electron Framework) + 1996 [0x10e7b35bc]
+ 2292 ares_dns_rr_get_ttl (in Electron Framework) + 3039188 [0x111b2cc40]
+ 2292 -[NSApplication run] (in AppKit) + 368 [0x18e53313c]
+ 2292 -[NSApplication(NSEventRouting) nextEventMatchingMask:untilDate:inMode:dequeue:] (in AppKit) + 72 [0x18f0d55bc]
+ 2292 -[NSApplication(NSEventRouting) _nextEventMatchingEventMask:untilDate:inMode:dequeue:] (in AppKit) + 688 [0x18f0d58b0]
+ 2292 _DPSNextEvent (in AppKit) + 576 [0x18e540084]
+ 2292 _DPSBlockUntilNextEventMatchingListInMode (in AppKit) + 228 [0x18ebec35c]
+ 2292 _BlockUntilNextEventMatchingListInMode (in HIToolbox) + 48 [0x19708414c]
+ 2292 ReceiveNextEventCommon (in HIToolbox) + 488 [0x196efa8bc]
+ 2292 RunCurrentEventLoopInMode (in HIToolbox) + 320 [0x196ef7560]
+ 2292 _CFRunLoopRunSpecificWithOptions (in CoreFoundation) + 532 [0x18a1e41c4]
+ 2292 __CFRunLoopRun (in CoreFoundation) + 820 [0x18a111854]
+ 2292 __CFRunLoopDoSources0 (in CoreFoundation) + 332 [0x18a112c94]
+ 2292 __CFRunLoopDoSource0 (in CoreFoundation) + 172 [0x18a112ec4]
+ 2292 __CFRUNLOOP_IS_CALLING_OUT_TO_A_SOURCE0_PERFORM_FUNCTION__ (in CoreFoundation) + 28 [0x18a112f30]
+ 2292 BSServiceMainRunLoopSourceHandler (in BoardServices) + 224 [0x1a4678690]
+ 2292 __BSSERVICEMAINRUNLOOPQUEUE_IS_CALLING_OUT_TO_A_BLOCK__ (in BoardServices) + 48 [0x1a467990c]
+ 2292 _dispatch_block_invoke_direct (in libdispatch.dylib) + 284 [0x189e9f714]
+ 2292 _dispatch_client_callout (in libdispatch.dylib) + 16 [0x189eb44b0]
+ 2292 -[FBSWorkspace _calloutQueue_executeCalloutFromSource:withBlock:] (in FrontBoardServices) + 188 [0x1a4567b78]
+ 2292 __94-[FBSWorkspaceScenesClient _queue_updateScene:withSettings:diff:transitionContext:completion:]_block_invoke_2 (in FrontBoardServices) + 100 [0x1a45b06d8]
+ 2292 __94-[FBSWorkspaceScenesClient _queue_updateScene:withSettings:diff:transitionContext:completion:]_block_invoke_2.cold.1 (in FrontBoardServices) + 388 [0x1a45f0cb0]
+ 2292 -[FBSScene updater:didUpdateSettings:withDiff:transitionContext:completion:] (in FrontBoardServices) + 780 [0x1a458294c]
+ 2292 -[FBSScene _callOutQueue_maybeCoalesceClientSettingsUpdates:] (in FrontBoardServices) + 136 [0x1a45816b4]
+ 2292 __76-[FBSScene updater:didUpdateSettings:withDiff:transitionContext:completion:]_block_invoke.138 (in FrontBoardServices) + 832 [0x1a4582e9c]
+ 2292 -[FBSSceneObserver scene:didUpdateSettings:] (in FrontBoardServices) + 148 [0x1a45f728c]
+ 2292 -[NSSceneStatusItem scene:didUpdateSettings:] (in AppKit) + 52 [0x18efa10bc]
+ 2292 -[NSSceneStatusItem _updateForDiff:] (in AppKit) + 124 [0x18efa0dc0]
+ 2292 -[NSStatusItem _setDisplayIdentifier:] (in AppKit) + 100 [0x18e74d9c0]
+ 2292 -[NSWindow _setFrameCommon:display:fromServer:] (in AppKit) + 2724 [0x18e582bb8]
+ 2292 -[NSNotificationCenter postNotificationName:object:userInfo:] (in Foundation) + 88 [0x18b90b4a4]
+ 2292 _CFXNotificationPost (in CoreFoundation) + 740 [0x18a0d63d0]
+ 2292 _CFXRegistrationPost (in CoreFoundation) + 440 [0x18a193fd0]
+ 2292 ___CFXRegistrationPost_block_invoke (in CoreFoundation) + 92 [0x18a19408c]
+ 2292 __CFNOTIFICATIONCENTER_IS_CALLING_OUT_TO_AN_OBSERVER__ (in CoreFoundation) + 128 [0x18a1079a8]
+ 2292 ??? (in AppKit) load address 0x18e507000 + 0x684878 [0x18eb8b878]
+ 2292 ??? (in AppKit) load address 0x18e507000 + 0x4befa8 [0x18e9c5fa8]
+ 2292 ??? (in AppKit) load address 0x18e507000 + 0x4bed5c [0x18e9c5d5c]
+ 2292 -[FBSScene _updateClientSettings:] (in FrontBoardServices) + 672 [0x1a457f6a8]
+ 2292 -[FBSScene _sendUpdate:] (in FrontBoardServices) + 1120 [0x1a457fbac]
+ 2292 -[FBSWorkspaceScenesClient scene:didUpdateClientSettings:withDiff:transitionContext:] (in FrontBoardServices) + 256 [0x1a4568d18]
+ 2292 -[BSServiceDispatchQueue performAsyncAndWait:] (in BoardServices) + 84 [0x1a4677c30]
+ 2292 _dispatch_async_and_wait_f (in libdispatch.dylib) + 100 [0x189ea105c]
+ 2292 _dispatch_async_and_wait_f_slow (in libdispatch.dylib) + 44 [0x189eaac24]
+ 2292 __DISPATCH_WAIT_FOR_QUEUE__ (in libdispatch.dylib) + 432 [0x189eaa6ac]
+ 2292 _dispatch_event_loop_wait_for_ownership (in libdispatch.dylib) + 460 [0x189ebe474]
+ 2292 _dispatch_kq_poll (in libdispatch.dylib) + 220 [0x189ebda64]
+ 2292 kevent_id (in libsystem_kernel.dylib) + 8 [0x18a013a74]
Key frames, inner to outer:
kevent_id
_dispatch_kq_poll
_dispatch_event_loop_wait_for_ownership
__DISPATCH_WAIT_FOR_QUEUE__ ← waits forever
_dispatch_async_and_wait_f_slow
-[BSServiceDispatchQueue performAsyncAndWait:]
-[FBSWorkspaceScenesClient scene:didUpdateClientSettings:withDiff:transitionContext:]
-[FBSScene _sendUpdate:] / _updateClientSettings: ← re-entry
... (NSNotificationCenter post from window frame change)
-[NSWindow _setFrameCommon:display:fromServer:]
-[NSStatusItem _setDisplayIdentifier:]
-[NSSceneStatusItem _updateForDiff:] / scene:didUpdateSettings:
-[FBSScene updater:didUpdateSettings:withDiff:transitionContext:completion:]
-[FBSWorkspace _calloutQueue_executeCalloutFromSource:withBlock:] ← already on this queue
BSServiceMainRunLoopSourceHandler (main run loop source0)
-[NSApplication run] → ElectronMain
Additional observations
- No crash report: Crashpad
completed/andpending/are empty — the process hung, it never crashed, so nothing was auto-reported. The hang watchdog file + asamplewere the only diagnostics. - The app process is parented to launchd and holds the
SingletonLock; only SIGKILL/Force Quit clears it. - The Orca daemon (
Orca Helperrunningdaemon-entry.js, started by the previous version, uptime 4+ days) and all terminal sessions under it survived both the update quit and the hang — they are separate processes and reattach fine. - Because the deadlock fires inside status-item scene-settings handling on macOS 26.5.1, a possible mitigation on the app side is deferring
NSStatusItemcreation/updates out of the initial window-frame/scene-update window, or avoiding synchronous scene-settings work re-entrancy at startup.
Steps to reproduce (as experienced)
- Run Orca 1.4.187 on macOS 26.5.1 with an active workspace and terminal sessions.
- Accept the in-app auto-update to 1.4.191; the app quits and relaunches itself.
- Relaunched app shows "no workspace found" and is permanently frozen (not repro-verified a second time — the hung instance was kept alive for diagnostics).
Happy to provide the full sample file, trace tails, or run further diagnostics on the hung instance if requested.
Activity
srs-adamr commentedon Aug 28, 2026
Additional observation that may explain the trigger: after force-killing the hung instance and relaunching 1.4.191, macOS immediately showed the keychain authorization prompt — "Orca wants to use your confidential information stored in 'orca Safe Storage' in your keychain" (Electron safeStorage). It prompted twice; once the login keychain password was entered, the app started up completely normally (workspace loaded, hang watchdog marker was cleaned up).
So the post-update relaunch of the updated binary triggers keychain ACL re-authorization at startup. During the original frozen launch this prompt was never visibly presented — plausibly the SecurityAgent dialog presentation is what drove the scene-settings update in which the main thread deadlocked (the NSStatusItem/FBSScene stack in the report), leaving the auth pending and the app permanently stuck at "no workspace found".
Possible repro path: update → binary change → safeStorage keychain prompt due at first launch → main-thread deadlock in scene-settings handling while the modal is being presented. May be worth deferring safeStorage/keychain access and status-item setup off the initial launch path.
AmethystLiang commentedon Sep 27, 2026
Another occurrence on 1.4.215-adhoc.20260926234546 (macOS 26.5.1 / 25F80, Apple Silicon), also on the first launch after an updater relaunch (from 1.4.207-hourly). Full write-up, sample and timeline are in #23279 (closed as a duplicate of this issue).
This one was entered through a different path that the #10473 deferral doesn't cover:
UC::DriverCore::continueProcessing→CA::Transaction::commit) →NSStatusBarButtonCell drawWithFrame:→NSSystemStatusBar drawBackgroundInRect:→NSSceneStatusItem _setSelectedContentFrame:→NSStatusItemScene updateSettings:FBSScene _updateClientSettings:→_sendUpdate:→BSServiceDispatchQueue performAsyncAndWait:→_dispatch_event_loop_wait_for_ownership, with 2513/2513 samples, still identical about 6 min later, at 0% CPUThe update comes from AppKit's own draw of the menu-bar button, not from Orca calling
tray.setImage/setToolTip, so deferring Orca's tray mutations can't prevent it. The process's last log line came 1s after launch, just as ControlCenter created the status-itemReplicant[1]scene during the post-update "reconnect".renderer_bootstrap_startedwas never written. Force-quit and relaunch recovered it.Both occurrences so far happened right after an updater relaunch. Possible directions: delay
new Tray()on that launch (or until after the first window has shown and the ControlCenter reconnect has settled), and let the hang watchdog force-relaunch whenselfRecovered: false.mad-aran commentedon Oct 17, 2026
Similar issue after auto-update relaunch, with a remote Orca server environment.
Versions: macOS desktop <Mac版のバージョン> (after auto-update), remote runtime 1.4.224 on Rocky Linux 9 (headless
orca serve, reached over Tailscale at ws://:6768)What happened
updater_will_quit_cleanup_startedwithdaemonTeardown: "disconnect"and relaunched for the update.ready.Expected
The existing environment and its groups are restored automatically after an update relaunch.
fix(pilot): make the macOS menu-bar item 10s after the window is reve…
FMsongX2 commentedon Nov 7, 2026
Another occurrence, and this one is not tied to an updater relaunch: it hit 5h28m into a normal session on 1.4.251 (Electron 43.8.6), macOS 26.7.2 (25G332), Apple Silicon. Same terminal wait, entered through the window-frame path from electron/electron#52437 rather than the status-item paths above.
Timeline (KST, from the unified log and
main.trace.ndjson)NSPersistentUIManager flushAllChanges, then Orca's main thread sendsNSSceneFenceActionupdates for the menu-bar item scenes (com.apple.controlcenter:…-NSStatusItemView,…-Replicant[1]-Aux[1]-NSStatusItemView)sidebar_worktree_activateSending action(s) in update: NSSceneFenceActionfor the status-item sceneBG-Active[40]→XX-None[0]sample×2, 16 s apart: 813/813 and 810/810 samples on the identical main-thread stack, 0 % CPU. CLI calls time out; the renderer keeps running. Force-killed and relaunched; terminals reattached from the daemon.Main-thread stack (identical in both samples: 813/813 and 810/810; verbatim frames)
Everything above
__CFRunLoopDoSource0is-[NSApplication run]fromElectronMain. There are no Orca or Electron frames inside the deadlock. None of the other 58 threads holds a FrontBoard or BoardServices frame, so nothing else in the process can release the queue.So there are now three doors into one self-wait: the status-item
_setDisplayIdentifier:path (original report), the status-bar draw path (#23279), and the server-delivered window-frame path (this one). Deferring Orca's own AppKit mutations can't close the last two. #39312 lets the existing hang watchdog relaunch Orca only when two livesamplecaptures prove this exact self-wait; every other stall stays observation-only.