Three separate faults, each producing a black Moonlight stream:
- sunshine's wlr backend takes wlr-screencopy's dmabuf path, which on this
virtio-gpu returns empty buffers. Reproduced locally over 127.0.0.1, so it
is not the network or the client. Switch to kms capture.
- hypridle locked the session and cut dpms, blanking the captured output with
no console to unlock from. Disable it on this host.
- qemu picked the nvidia render node by readdir order, so virgl fell back to
llvmpipe. Pin renderD128 via the VM's args.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Unauthenticated flake input fetches from github.com were hitting 429.
Token lives in /etc/nix/github-token.conf (root:root 600, out-of-band —
never in git or /nix/store) and nix.conf !includes it.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
pve hardcodes egl-headless with no rendernode, and the nvidia egl vendor
outranks mesa, so virgl lands on llvmpipe.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
vga: virtio-gl, serial0 and ovmf are load-bearing and live only on the
hypervisor, so a rebuilt vm would regress silently.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
It leaks a /dev/dri/renderD fd per encoder probe, hits the 1024 default,
and mesa can then no longer dlopen the GBM driver — capture degrades to a
black picture with no error surfaced.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
wlgrab is the name the backend logs under, not a valid config value.
Sunshine rejects it, then starts anyway with no encoder — moonlight sees
a 503.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
zram wedged shutdown faulting stored pages back in, and the VM already
has a real swap partition. aoe-with-web is the same binary plus `serve`.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Needed before switching to vga: virtio-gl, which leaves no QEMU console
surface for noVNC or screendump.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Also overrides the home-manager nh.flake, which sets NH_FLAKE and was
still pointing at the ~/projects default, so bare `nh home switch` never
resolved on this host.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
New NixOS VM on proxmox, replacing the Ubuntu box. Reuses the existing
Hyprland config; remote access is Sunshine/Moonlight rather than xrdp,
since xrdp cannot drive a wayland compositor.
Two fixes here are not yolo-specific and affect any fresh install:
git at system level (nix needs it for the type = "git" hyprland input,
but git only came from home-manager, so neither rebuild could run), and
dropping the yogurt input, whose private repo made home-manager
un-evaluatable without GitHub auth.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
SDDM auto-logs naps62 in and Hyprland locks immediately via hyprlock
(exec-once), so a cold boot lands on the themeable lock screen instead of
the SDDM greeter. Safe here (no FDE makes SDDM-vs-autologin marginal) and it
lets Moonlight reach the box after a cold boot, not just from suspend, since
Sunshine only runs once a session exists. konishi-only.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Arm magic-packet WoL on the Intel igc NIC (re-applied every boot) so
Moonlight can wake the box from suspend to stream. Still needs "Wake on
LAN" enabled in BIOS.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Steam's CEF UI intermittently crashed its GPU child (exit_code=8704) on
NVIDIA+Wayland, falling back to CPU rendering and making Big Picture a
slideshow at 4K/HiDPI. Bake -cef-disable-gpu-sandbox into every steam
invocation via extraArgs so it covers all entry points (desktop, CLI,
Sunshine's Big Picture tile) — launch flags only apply on the invocation
that first starts Steam.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Default nixpkgs sunshine is built -DSUNSHINE_ENABLE_CUDA=FALSE, so every
NVENC probe died with "Couldn't scale frame" and it fell back to CPU
x264. Fix:
- package = sunshine.override { cudaSupport = true; } — gives it the
CUDA scaling/colour-conversion path. Verified: probe now finds
h264_nvenc / hevc_nvenc / av1_nvenc.
- LD_LIBRARY_PATH=/run/opengl-driver/lib on the service so libcuda.so.1
(NVIDIA driver) is discoverable.
- capSysAdmin = false — Hyprland/wlroots capture via wlr-screencopy needs
no CAP_SYS_ADMIN, and the setcap wrapper's secure-exec mode would strip
LD_LIBRARY_PATH and break libcuda loading.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Miguel Palhas