# R595 Driver Post-Reboot Readiness

Status: `READY_FOR_MANUAL_CH20_21_RECOVERY`

Recorded at: `2026-08-31 12:41:01 KST`

## Reboot Reflection

- Reboot request count: 1
- Boot time: `2026-08-31 12:34 KST`
- Kernel: `6.8.0-117-generic`
- GRUB one-time `next_entry`: consumed
- Driver: `595.84`
- GPU: NVIDIA RTX A6000
- Module: NVIDIA Open Kernel Module, `Dual MIT/GPL`
- `nvidia-smi`: PASS
- Additional reboot: 0

## DKMS·Package

- Kernel 117 DKMS: installed
- Kernel 138 DKMS: installed
- Kernel 117/138 module version: `595.84`
- Installed R595 package set: 22
- Installed R580 driver stack: 0
- `dpkg --audit`: empty
- `apt-get check`: PASS
- Secure Boot: disabled
- Reboot-required flag: absent
- Rollback applied: 0

## NVIDIA Runtime

- NVIDIA Container Toolkit: `1.18.1`
- Triton container GPU Driver readback: `595.84`
- Triton process: PASS
- Triton local health: PASS
- ai-worker PyTorch CUDA available: true
- ai-worker CUDA runtime: `12.9`
- ai-worker GPU: NVIDIA RTX A6000

## Container Recovery

- `triton-server` and GNOME Remote Desktop returned automatically.
- Initial `api-endpoint` start failed because maintenance-rebooted `itone-redis` was still stopped; this was
  dependency ordering, not a Driver regression.
- The exact pre-reboot baseline support containers were restored without Channel API access.
- Baseline container set running: 13/13
- OOMKilled: 0
- `api-endpoint` Gunicorn process count: 5
- `ai-worker` Celery process count: 18
- Triton process count: 1
- Bounded recent startup errors: 0

## Channel Boundary

- Channel API access: 0
- Frame access: 0
- Channel Start/Stop: 0
- ch20: manual recovery target
- ch21: pre-existing input issue; Driver acceptance blocker로 사용하지 않음

## Next

사용자가 ch20/ch21을 수동 복구한다. ch21의 기존 input 문제는 별도 요청에서 진단한다.