Repository navigation
Description
Environment
- omarchy 4.0.2-1
- hyprland 0.56.2-1
- quickshell 0.3.1-1
- Single monitor: HDMI-A-1 (Acer ED270R, 1920x1080@60)
- Arch Linux
Summary
omarchy-shell (Quickshell) entered a crash-restart loop that ran continuously for ~2 days (4,477 relaunches, every ~20-45s — see log below). Every keybinding that opens an Omarchy menu (SUPER+SPACE, SUPER+K, etc., including the keybindings-help menu itself) eventually stopped working entirely, because the loop filled $XDG_RUNTIME_DIR (a 1.6G tmpfs) with 4,475 leftover quickshell/by-id/<id>/ log directories (1.5G). Once full, Quickshell couldn't write its own log or serve IPC calls (omarchy-shell <target> <method> → "omarchy-shell is not running").
Root cause (from the journal, see excerpt below)
- The shell ran normally for over an hour (
lock-requested/lock-pending: screen-stabilizingcycling normally, PID 7094). - Then a screen event produced a placeholder "FALLBACK" Wayland output not backed by a real screen (
Layershell screen does not correspond to a real screen), and Hyprland reported it removed (Got removal for monitor "FALLBACK" which was not previously tracked). - This triggered
The Wayland connection experienced a fatal error: Invalid argument, and Quickshell exited via Qt's_exit()(status 255) — no signal, soomarchy-crash-watchnever reported it as a crash. - Hyprland's
LOCKsolitaryBlockedByblocker on the monitor is (peromarchy-hyprland-session-locked's own comment) never cleared by an abnormal client exit, so it stays stuck. - Every relaunch's
strandedLockCheckProc(inplugins/lock/Service.qml) sees that stale "locked" state viaomarchy-hyprland-session-locked, treats it as a stranded lock (lock-stranded: recovering), and tries to recreate a lock surface — which hits the same FALLBACK-screen condition and crashes again. Indefinitely. omarchy-launch-shell's safety net (Giving up on the Omarchy shell after $attempts relaunches in under a minute) never triggers at this crash cadence: its 60s sliding window resets before 5 attempts accumulate whenever crashes are spaced more than 12s apart (they're ~20-45s apart here), so it loops forever instead of giving up.- Each crash leaves its
by-id/<instance>log directory behind uncleaned, slowly filling the runtime tmpfs until Quickshell can't function at all.
Impact
The entire Omarchy menu system (bar, keybinding popups, omarchy-menu-*) silently stops working, with no crash notification, until a user notices and manually kills+restarts the shell and clears the runtime directory.
Suggested fixes (any of these would help)
- Guard the
strandedLock/requestSessionLockpath inplugins/lock/Service.qmlagainst attempting lock-surface creation on a screen not backed by a real Wayland output. - Clear the stale
LOCKsolitaryBlockedByflag once its owning client is confirmed dead, rather than leaving it stuck indefinitely (in Hyprland itself, or by havingomarchy-hyprland-session-lockedaccount for it). - Fix the give-up window in
omarchy-launch-shell: track attempts over a longer/rolling window (e.g. 5 attempts in 5 minutes) instead of one that resets before fast repeat-crashes accumulate. - Have the launcher prune its own
by-idruntime directories on each launch, so a crash loop — whatever the cause — can't fill the tmpfs.
Log excerpt (first crash + one full repeat cycle, journal -t omarchy-shell, noise-filtered)
Aug 31 11:54:29 host omarchy-shell[1332]: INFO: Launching config: "/usr/share/omarchy/shell/shell.qml"
Aug 31 11:54:29 host omarchy-shell[1332]: INFO: Shell ID: "c0240dfe8c8e8a421e1f8db23a03fc60" Path ID "c0240dfe8c8e8a421e1f8db23a03fc60"
Aug 31 11:56:05 host omarchy-shell[4228]: INFO: Launching config: "/usr/share/omarchy/shell/shell.qml"
Aug 31 11:56:05 host omarchy-shell[4228]: INFO: Shell ID: "c0240dfe8c8e8a421e1f8db23a03fc60" Path ID "c0240dfe8c8e8a421e1f8db23a03fc60"
Aug 31 11:57:57 host omarchy-shell[7094]: INFO: Launching config: "/usr/share/omarchy/shell/shell.qml"
Aug 31 11:57:57 host omarchy-shell[7094]: INFO: Shell ID: "c0240dfe8c8e8a421e1f8db23a03fc60" Path ID "c0240dfe8c8e8a421e1f8db23a03fc60"
Aug 31 12:40:43 host omarchy-shell[7094]: DEBUG qml: omarchy lock ... lock-requested
Aug 31 12:40:43 host omarchy-shell[7094]: DEBUG qml: omarchy lock ... lock-pending: screen-stabilizing
Aug 31 12:41:54 host omarchy-shell[7094]: DEBUG qml: omarchy lock ... lock-requested
Aug 31 12:41:54 host omarchy-shell[7094]: DEBUG qml: omarchy lock ... lock-pending: screen-stabilizing
Aug 31 12:43:35 host omarchy-shell[7094]: DEBUG qml: omarchy lock ... lock-requested
Aug 31 12:43:35 host omarchy-shell[7094]: DEBUG qml: omarchy lock ... lock-pending: screen-stabilizing
Aug 31 13:07:25 host omarchy-shell[7094]: DEBUG qml: omarchy lock ... lock-requested
Aug 31 13:07:25 host omarchy-shell[7094]: DEBUG qml: omarchy lock ... lock-pending: screen-stabilizing
Aug 31 13:07:46 host omarchy-shell[7094]: WARN: Layershell screen does not correspond to a real screen. Letting the compositor pick.
Aug 31 13:07:46 host omarchy-shell[7094]: WARN: Layershell screen does not correspond to a real screen. Letting the compositor pick.
Aug 31 13:07:46 host omarchy-shell[7094]: WARN: Layershell screen does not correspond to a real screen. Letting the compositor pick.
Aug 31 13:07:46 host omarchy-shell[7094]: WARN: Layershell screen does not correspond to a real screen. Letting the compositor pick.
Aug 31 13:07:47 host omarchy-shell[7094]: WARN quickshell.hyprland.ipc: Got removal for monitor "FALLBACK" which was not previously tracked.
Aug 31 13:07:47 host omarchy-shell[7094]: WARN: The Wayland connection experienced a fatal error: Invalid argument
Aug 31 13:07:47 host omarchy-shell[293466]: Omarchy shell exited with status 255; relaunching.
Aug 31 13:07:48 host omarchy-shell[293708]: INFO: Launching config: "/usr/share/omarchy/shell/shell.qml"
Aug 31 13:07:48 host omarchy-shell[293708]: INFO: Shell ID: "c0240dfe8c8e8a421e1f8db23a03fc60" Path ID "c0240dfe8c8e8a421e1f8db23a03fc60"
Aug 31 13:07:48 host omarchy-shell[293708]: DEBUG qml: omarchy lock ... lock-stranded: recovering
Aug 31 13:07:48 host omarchy-shell[293708]: DEBUG qml: omarchy lock ... lock-requested
Aug 31 13:07:48 host omarchy-shell[293708]: DEBUG qml: omarchy lock ... lock-pending: screen-stabilizing
Aug 31 13:08:09 host omarchy-shell[293708]: WARN: Layershell screen does not correspond to a real screen. Letting the compositor pick.
Aug 31 13:08:09 host omarchy-shell[293708]: WARN: Layershell screen does not correspond to a real screen. Letting the compositor pick.
Aug 31 13:08:09 host omarchy-shell[293708]: WARN: Layershell screen does not correspond to a real screen. Letting the compositor pick.
Aug 31 13:08:09 host omarchy-shell[293708]: WARN: Layershell screen does not correspond to a real screen. Letting the compositor pick.
Aug 31 13:08:10 host omarchy-shell[293708]: WARN quickshell.hyprland.ipc: Got removal for monitor "FALLBACK" which was not previously tracked.
Aug 31 13:08:10 host omarchy-shell[293708]: WARN: The Wayland connection experienced a fatal error: Invalid argument
Aug 31 13:08:10 host omarchy-shell[298070]: Omarchy shell exited with status 255; relaunching.
Aug 31 13:08:11 host omarchy-shell[298318]: INFO: Launching config: "/usr/share/omarchy/shell/shell.qml"
This exact lock-stranded → FALLBACK screen → fatal Wayland error → exit 255 → relaunch cycle then repeated roughly every 20-45s continuously from Aug 31 13:07 through Sep 2 10:13 (4,477 total relaunches) until the shell was manually killed and the runtime directory cleared. Hostname/PIDs anonymized above; happy to provide the full raw journal privately if useful.
Not reliably reproducible on demand — it followed a screen/output event during normal use and then self-sustained.
Activity
AndrijaSkontra commentedon Sep 2, 2026
Confirmation on a second machine (Intel iGPU desktop)
Same crash-loop → leftover
quickshell/by-iddirs →$XDG_RUNTIME_DIRfills → menus/keybindings die.User-facing report after leaving the PC overnight (~14–18h; last systemd suspend was a short S3, then the loop ran unattended until morning):
System details
Timeline (journal
-b -1)There are no outputs - creating placeholder screen/Got removal for monitor "FALLBACK"/ fatal Wayland error /exited with status 255; relaunching/lock-stranded: recovering. Same cycle as this issue and omarchy-shell crashes and relaunches on every wake from idle lock (FALLBACK monitor removal, fatal Wayland error) #7380 / Shell lock client dies on resume from suspend, showing Hyprland 'screensaver died' failsafe #9156.omarchy-launch-shell's give-up window never trips at this cadence.libgtk-4Wayland dispatch (--gtk-single-instance=true --initial-window=false). Related: Ghostty terminal keybind launches single-instance mode that can crash all windows #5447 / Ghostty bug closes all sessions #6567. Super+Return iso.bind("SUPER + RETURN", "Terminal", { omarchy = "terminal" })→xdg-terminal-exec→ that same single-instance process, so the keybind becomes a silent no-op.Failed to copy log from memfd ... "No space left on device"(ENOSPC). 2,838 ENOSPC lines by morning.qs ipc ... call -- shell toggle omarchy.menu. Each IPC client SIGSEGVs while sorting leftover instance lockfiles (21 coredumps in five minutes). That is why the launcher appears completely dead from both keyboard and mouse.This also happened on the previous long uptime: 2026-08-31 07:43, same
qs ipcstack (toggle omarchy.menu '{"menu":"root"}'), then reboot.Extra crash signature not in the original report
qs ipcdoes not just fail with "omarchy-shell is not running". With thousands of staleInstanceLockInfoentries it dies in Quickshell's instance picker:Same stack on the Aug 31 morning occurrence. So the leftover
by-iddirectories do not only fill the tmpfs — they also poisonqs ipcinstance selection with invalidQDateTimes.Ghostty (why Super+Return was dead)
Same GTK/Wayland stack on earlier dumps (Aug 26, Aug 28, Aug 31). Once that daemon is gone, Super+Return has nothing to attach a new window to.
Happy to attach coredumpctl info privately if useful. Reboot cleared
$XDG_RUNTIME_DIRand restored the launcher.omarchybot commentedon Sep 2, 2026
Triaged: your crash is #7380, not the SIGABRT four sibling reports share — plus two defects that are new here
Four other reports filed in the same window — #9654, #9709, #9727, #9736, all duplicates of #6888 — carry
FATAL: Tried to show lockscreen surfaces without active lockand die on SIGABRT. Yours shares thelock-stranded: recovering/lock-requested/lock-pending: screen-stabilizingprefix but ends differently:The Wayland connection experienced a fatal error: Invalid argumentandexited with status 255, with no signal and no FATAL lockscreen line anywhere in your log. So this is not that abort, and your diagnosis is the right one.1. The crash loop itself — already open as #7380
The
lock-stranded → FALLBACK screen → fatal Wayland error → exit 255 → relaunchcycle you describe is the signature of #7380 ("omarchy-shell crashes and relaunches on every wake from idle lock (FALLBACK monitor removal, fatal Wayland error)"), and #9156 is the same family. Your steps 2-5 are consistent with what quickshell does:WlSessionLock::updateSurfaces()(src/wayland/session_lock.cpp:55-70) drops screens whose Wayland output is a placeholder, and the connection dies with EINVAL around that.omarchy-launch-shellwas taught to supervise exactly those signal-less deaths in 1e7bb66, but supervising them is what turns one crash into a loop when the underlying condition persists.2. The give-up window never trips — confirmed, and it is Omarchy's
You are right about
bin/omarchy-launch-shell, and the arithmetic checks out. The window is reset before the counter can reach its limit:With the
sleep 1backoff, a crash roughly every 11 s or less accumulates six attempts inside one 60 s window and the launcher gives up as intended. Any slower — your loop ran at 20-45 s — and the window resets first, soattemptsnever gets past 5 and it relaunches forever. Simulated across the spacings in your log: give-up fires at 10 s, and never at 12 s, 13 s, 20 s, 30 s or 45 s. The exact threshold shifts a little with how long each run lasts, but the 20-45 s case is unambiguous.3. The leftover
by-iddirectories — quickshell's leak, Omarchy's exposureNothing in Omarchy creates or removes those.
$XDG_RUNTIME_DIR/quickshell/by-id/<instance-id>/is created by quickshell itself inQsPaths::instanceRunDir()(src/core/paths.cpp:116-143), one per process launch. Quickshell knows which ones are dead —QsPaths::collectInstances()(src/core/paths.cpp:415) classifies them intodeadInstancesvia anfcntllock probe — but nothing anywhere in its source removes them; the onlyremoveRecursivelycall in the tree is for theqsVFS directory. So the directories accumulate for the life of the login session, and 4,477 relaunches is 4,475 of them.That makes the leak quickshell's defect. It is bounded by the login session — the runtime directory goes away with it — but as you found, it can fill long before then. Whether Omarchy should prune anyway is a fair question, since Omarchy is the thing running the relaunch loop that multiplies a slow leak into a full tmpfs — but pruning another process's runtime state has to probe the same instance lock quickshell does, or it will delete a live instance's IPC socket. That is the maintainer's call, not a mechanical fix, so it is written up rather than patched.
Your
qs ipcSIGSEGV insortInstances/compareThreeWay(QDateTime), and the confirmation from a second machine in the comment above, are consistent with the same pile of stale lock files being read back throughQsPaths::checkLock()'sQDataStream— also quickshell's, and worth reporting upstream alongside the leak.What happens next
Nothing was reproduced: the loop needs a real compositor and a real output-removal event, so it would have to be exercised on a disposable VM rather than here, and it is explicitly not reliably reproducible on demand. The crash itself is tracked in #7380. The launcher window and the runtime-directory growth are new here and go to the maintainer with the analysis above. The four SIGABRT reports are duplicates of #6888, with PR #7169 open for the Omarchy half of it — a different bug from yours, so this is not a duplicate of them.
Reviewed twice: by Claude Opus 5, and independently by Codex at xhigh reasoning.
JaviBasulto commentedon Sep 12, 2026
Same failure mode, with two things this thread does not yet have: a count from the compositor's side that matches the shell's exit count 1:1, and evidence that the churn is not confined to the shell.
Environment
4.0.3-1, hyprland0.56.2-2, quickshell0.3.1-1, Linux7.2.3-arch1-3screensaver=150,lock=300)1. The loop, counted on both sides
One boot, one lock cycle, 31 minutes:
Every
Omarchy shell exited with status 255has a matchingdrm: Disabling output HDMI-A-1in the compositor log. The chain is visible in the journal:idle-monitor: idle17:18:45 ->lock-system: lock-timeout17:21:09 -> first shell exit 17:21:25; the loop stopped on its own at 17:52:20, and the shell launched then (pid 2439140) was still alive when I wrote this, with 0 exits since. So it is bounded by the interval where the output was down, not self-sustaining.Other clients saw the same thing in lockstep, which rules out the shell as the odd client out:
kdeconnectd,kdeconnect-indicatorandxdg-desktop-portal-hyprlandall loggedThere are no outputs - creating placeholder screen/ re-discoveringFALLBACKon each pass.2. Not HPD, and the connector never reads disconnected
No
amdgpuconnector/hotplug lines in the kernel log for the whole window, and/sys/class/drm/card1-HDMI-A-1/statusanswersconnectedbefore and after. Aquamarine's own rescan right after disabling the output records:The output is being taken down while the connector reports connected — consistent with the DPMS-off framing above rather than link loss or a flaky cable. What precedes the disable in the same log is the full re-probe +
Modesetting HDMI-A-1 with 1920x1080@60.00Hz+Cannot commit when a page-flip is awaiting, i.e. a re-apply cycle, ~19s apart.3. Runtime tmpfs litter, measured
After that single 31-minute loop,
/run/user/1000/quickshell/by-id/held 100 leftover instance directories, 23 MB, on a 4.7 G/run/user/1000. Same mechanism you describe, just bounded here because the loop stopped after 99 relaunches instead of 4,477 — the accumulation rate is ~0.7 MB per relaunch, which means your 2-day loop and my 31-minute loop are the same leak at different durations.4. Collateral: an unrelated GTK client dumped core mid-loop
The churn is not confined to quickshell. 1.5s after one output-removal pass (17:32:16), a
nautilus 50.3.1instance — unrelated to Omarchy, just a GTK4 app open in the session — died with SIGSEGV/SEGV_MAPERR(17:32:17.529):The fault address (
0x7f6efc00e140) is unmapped at crash time; glibc's tcache fast path read the next pointer of a cached chunk in a region that no longer exists, inside a GTK callback dispatched from a Wayland event — i.e. output teardown running in a client. That is a data point that the blast radius extends past the shell: any GTK client in the session gets one roll of these dice per cycle.Workaround in place
Per the framing in #7478, the only config-level escape is to stop the blank/lock chain from firing (
idle.lockhigh enough thatarmBlankTimer()never runs), giving up automatic locking. I have not verified that independently.Posted by deepseek-v4-flash via Hermes Agent (Nous Research).
xgfurb commentedon Sep 19, 2026
Additional evidence from Omarchy 4.0.4-1:
Environment:
A lock request occurred while an Omarchy update terminal was open. The idle inhibitor was active, so this was not the normal idle timeout. The lock implementation then armed its five-second display-blank timer.
The journal subsequently recorded:
Got removal for monitor "FALLBACK" which was not previously trackedThe Wayland connection experienced a fatal error: Invalid argumentlock-stranded: recoveringThis cycle continued for more than two hours. It accumulated 62 Quickshell
by-idruntime directories using approximately 36 MiB.During the first output-loss transition, the Ghostty updater terminal also crashed with SIGSEGV. Its symbolized coredump was:
The Wayland queue mutex address was unmapped, consistent with GTK/Ghostty dispatching through a stale or freed Wayland event queue during the output transition.
This was not an OOM event: approximately 17 GiB remained available, swap was unused, and the journal contained no OOM kill. The update also did not replace Ghostty, GTK, Wayland, or Hyprland.
This suggests the #9745 failure persists on Omarchy 4.0.4 and can potentially crash another GTK Wayland client as collateral damage.
Filed by GPT-5 via Codex.
dpan commentedon Sep 19, 2026
The process death itself (placeholder screen → untracked
FALLBACKremoval →Wayland connection experienced a fatal error: Invalid argument→ exit 255) is tracked upstream as quickshell-mirror/quickshell#1155.The Omarchy-specific part of this ticket is what happens after that: stranded-lock recovery retriggering the same crash,
omarchy-launch-shellnever giving up at a 20–45s cadence, and leftoverquickshell/by-iddirs filling$XDG_RUNTIME_DIR.omarchybot commentedon Oct 9, 2026
Re-checked against
quattroat 988f44e: where each part of this now standsThree things have moved since the first triage comment above.
The crash itself. #7380, which that comment named as its tracker, was closed on 2026-09-21 as a duplicate of #6628, so #6628 is where the shell's death on output loss is tracked now. The death itself is in quickshell, reported upstream as quickshell-mirror/quickshell#1155, as noted above. The comment from 2026-09-12 also changes how the loop reads: every one of 99
exited with status 255lines matched adrm: Disabling output HDMI-A-1in Hyprland's log, and the loop stopped by itself once the output stayed up. So each relaunch is a fresh output-loss event that kills a fresh shell. The stranded-lock recovery does not cause the next crash. That recovery (shell/plugins/lock/Service.qml:105-126) is what puts a lock surface back over a session whose locking client died. Without it the session is left behind Hyprland's failsafe, which is the #6628 symptom.The give-up window in
bin/omarchy-launch-shellis unchanged onquattroand still does what the comment above describes: the 60-second window resets instead of sliding, so the launcher only gives up at about one crash every 11-12 seconds or faster. Another report on #7380 saw it fail to trip at 12.3 seconds. No open pull request changes it. It is not patched here because the obvious fix has a cost. A launcher that gives up during a locked session leaves the ext-session-lock with no client, which is the permanently locked screen of #6628. As things stand, the loop ends by itself once the output comes back. Whether a slower loop should stop the shell is the maintainer's decision.The runtime directory filling up now has two open pull requests. #9847 removes dead instances of the Omarchy shell from
quickshell/by-idbefore each launch, usingquickshell listfor liveness, and turns off the uncapped detailed log. It currently has merge conflicts withquattro. #13651 also prunes on launch, and it competes with #9847 for the same fix. Either one stops the tmpfs from filling and the menus from dying. Neither stops the crash loop, so neither is linked to close this issue. Nothing onquattroremoves those directories today.Nothing was reproduced. The loop needs a real output-loss event on real hardware, and the launcher arithmetic and lock code were read in source. This re-check was done by Claude Opus 5.5 alone. The configured second opinion, Codex at medium effort, could not run because its daily review budget was used up, so nobody else has checked these conclusions. This now waits on the maintainer to pick between #9847 and #13651, and to decide whether the launcher should give up on a slower crash loop.