Hard host freeze on Arch Linux with nvidia-open-dkms 595.58.03 during Plex Docker GPU workloads

NVIDIA Open GPU Kernel Modules Version

nvidia-open-dkms 595.58.03-1

Please confirm this issue does not happen with the proprietary driver (of the same version). This issue tracker is only for bugs specific to the open kernel driver.

• I confirm that this does not happen with the proprietary driver package.

Operating System and Version

Arch Linux

Kernel Release

6.18.21-1-lts

Please confirm you are running a stable release kernel (e.g. not a -rc). We do not accept bug reports for unreleased kernels.

• I am running on a stable kernel release.

Hardware: GPU

NVIDIA RTX 2000 Ada Generation (UUID: GPU-8de1bc21-ab8b-71ba-c6af-9f5895aaabaf)

Describe the bug

The host intermittently hard-freezes while using the NVIDIA Open Kernel Module with a Dockerized Plex workload.

This is not just a container or application failure. When the issue happens, the entire Arch Linux host becomes unresponsive:

• ping stops responding
• the local console no longer reacts
• only a hard reset recovers the machine

The system uses:

• nvidia-open-dkms 595.58.03-1
• linux-lts 6.18.21-1
• NVIDIA RTX 2000 Ada Generation

The GPU is passed into a Plex container through the NVIDIA runtime.

The issue happens after some uptime and is not an immediate boot failure.

To Reproduce

• Boot Arch Linux with kernel 6.18.21-1-lts
• Load the NVIDIA Open Kernel Module version 595.58.03
• Start Docker and run Plex with:
  • --runtime=nvidia
  • --gpus all
  • NVIDIA_VISIBLE_DEVICES=all
  • NVIDIA_DRIVER_CAPABILITIES=all
• Let Plex perform normal media activity such as:
  • media analysis
  • chapter thumbnail generation
  • commercial skipping
  • transcoding-related ffmpeg workloads
• After some time or after some days of uptime, the host may hard-freeze
• When the problem occurs:
  • the host does not respond to ping
  • the local console is also unresponsive
• Recovery is only possible by hard reset

I do not currently have a minimal deterministic reproducer. The problem is intermittent.

Bug Incidence

Sometimes

nvidia-bug-report.log.gz

Attached: nvidia-bug-report.log.gz

I collected this report after observing the issue condition and while the NVIDIA Open Kernel Module environment was active.

nvidia-bug-report.log.gz

More Info

Environment

• OS: Arch Linux
• Kernel: 6.18.21-1-lts
• NVIDIA packages:
  • nvidia-open-dkms 595.58.03-1
  • nvidia-utils 595.58.03-1
  • nvidia-container-toolkit 1.19.0-1
• GPU: NVIDIA RTX 2000 Ada Generation
• PCI ID: 10de:28b0
• NVIDIA module is installed through DKMS

Important observations

• No clear OOM condition was present when diagnostics were collected
• The system had ample free memory
• The previous boot logs did not show an obvious NVIDIA Xid
• However, the machine had clearly crashed hard, because filesystems required recovery after reboot
• Docker reported one NVIDIA-related warning:
  • CDI device injection failed: failed to stat CDI host device "/dev/nvidiactl": no such file or directory
• The Plex container uses the NVIDIA runtime and GPU device access successfully in normal operation

Notes

I cannot yet prove with certainty that NVIDIA is the only cause, because the issue is intermittent and there is no clear Xid in the collected previous-boot logs.

However, NVIDIA is my primary suspicion because the freezes happen on a host using the open NVIDIA kernel module for Plex GPU workloads.

If needed, I can provide additional:

• journalctl logs
• Docker container details
• Plex runtime information

Start with the attached nvidia-bug-report.log.gz and the previous-boot journalctl logs. Reproduce the Dockerized Plex GPU workload using the listed kernel and open-module versions, paying attention to the missing /dev/nvidiactl warning and the absence of a clear Xid. Done means identifying a reproducible driver or integration cause and validating a fix without host freezes.