Repository navigation
macOS: remote mouse input silently dies after uptime — keyboard survives, fixed by process restart #5874
Description
Is there an existing issue for this?
- I have searched the existing issues
Is your issue described in the documentation?
- I have read the documentation
Is your issue present in the latest beta/pre-release?
This issue is present in the latest release
Describe the Bug
On a macOS host, remote mouse input silently stops being injected while a Moonlight session is active. When it happens:
- Video stream continues normally.
- Remote keyboard input still works.
- Remote mouse is completely dead — no movement, no clicks.
- The physically attached mouse on the Mac keeps working.
- Nothing abnormal appears in
sunshine.logat the default log level.
Restarting the Sunshine process immediately restores remote mouse input:
launchctl kickstart -k "gui/$(id -u)/dev.lizardbyte.Sunshine"The symptom has recurred 3 times over 2 days (Oct 7, Oct 8 twice — once only ~5 h after a restart), always with the same signature: keyboard alive, mouse dead, restart fixes it. Possibly related to #4652 (mouse movement dead on macOS while click/keyboard worked), though in my case there is no cursor jitter — the injected mouse input is simply absent.
Expected Behavior
Mouse input injection should keep working for the lifetime of the Sunshine process.
Additional Context
Accessibility/Screen Recording permissions are granted and had been working (keyboard injection still functioning while mouse was dead suggests this is not a TCC revocation). Happy to reproduce with min_log_level = debug if maintainers tell me which log lines to look for.
Host Operating System
macOS
Operating System Version
27.0.1 (build 26A434)
Architecture
arm64/aarch64
Sunshine commit or version
2026.914.233613 (commit 63d35f7)
Package
macOS - Homebrew
GPU Type
Apple Silicon
GPU Model
Apple M5 Max (Mac17,14)
GPU Driver/Mesa Version
NA
Capture Method
None (default)
Config
address_family = both
encoder = software
locale = ko
native_pen_touch = disabled
sw_preset = ultrafast
sw_tune = zerolatencyApps
(desktop streaming)Relevant log output
Nothing relevant at default log level — the failure is silent.
Client: Moonlight Qt on Arch Linux.
Activity
june8572-design commentedon Oct 8, 2026
Attaching logs from the run that covered the most recent occurrence.
The dead-mouse window was the streaming session 15:38:21 → 15:38:31 below — client connected, remote mouse completely unresponsive (remote keyboard still worked, physical mouse on the Mac unaffected), client disconnected, then Sunshine was restarted via
launchctl kickstart -kat 15:41 which immediately restored input.Note the failure is completely silent at the default log level — nothing but encoder/audio boilerplate between
CLIENT CONNECTEDandCLIENT DISCONNECTED. No input errors, no warnings.sunshine.log excerpt — process run 10:40:05 → 15:41:04, dead-mouse session at 15:38
An earlier dead session the same day (13:00:20 → 13:00:45 connect/disconnect, same signature) is also in this log with zero input-related output.
min_log_level = debugis now enabled and Sunshine restarted, so the next recurrence will capture packet-level input logs — I'll attach them here when it happens.june8572-design commentedon Oct 8, 2026
It recurred, and this time
min_log_level = debugwas armed — so we can see exactly where it breaks.Timeline: Sunshine restarted with debug logging at
2026-10-08 15:47:49. Mouse input died sometime before2026-10-09 01:27— roughly ~9.7 h of process uptime. Process restarted at ~01:33, mouse restored again.Key finding — mouse packets arrive fine, injection is what dies:
During the dead-mouse window, Sunshine received and parsed valid mouse packets with real coordinates and button presses — yet the macOS cursor never moved. So the failure is downstream of packet parsing, inside the macOS event-injection path (CGEvent posting silently becomes a no-op while the process keeps running). Remote keyboard input was still working during the same window.
Dead session excerpt (client connected 01:31:49, mouse dead on host):
(hundreds of move packets follow — coordinates keep changing as expected — and the session ends cleanly at
CLIENT DISCONNECTED01:31:55 with zero errors anywhere in between)Nothing on the host changed between working and dead states — same Sunshine process, no TCC changes, no lock-screen transitions that I noticed. Full 456 KB debug log available if useful; tell me which parts to attach.
ReenigneArcher commentedon Oct 8, 2026
Thanks for reporting this! I have put up a PR at LizardByte/libvirtualhid#183 to try to resolve this. Are you able to test a self compiled build of Sunshine with this version of the library applied?