dsh-remote-gpu-monitoring

Remote multi-server GPU status board for DeepSeek Harness Web: read-only nvidia-smi over SSH with ControlMaster-muxed connections, a host-side cached snapshot, per-server selection, and gpu_overview / gpu_refresh agent tools.

包名

dsh-remote-gpu-monitoring

版本

1.0.0

许可证

MIT

Configuration

Everything below is optional. The row ships with hosts: 'auto' (discover every concrete alias in ~/.ssh/config). To watch an explicit list instead, override the row in your profile's cordis.patch.yml:

- id: remote-gpu-monitoring
  name: dsh-remote-gpu-monitoring
  config:
    hosts: ['gpu-a', 'gpu-b', 'train-01']

All row-config keys merge over the defaults:

Key	Default	Meaning
`hosts`	`'auto'`	`'auto'` = every concrete alias in `~/.ssh/config`, or a `string[]` of aliases
`intervalMs`	`5000`	per-host collection interval
`sshTimeoutMs`	`9000`	per-probe timeout
`busyMemPct`	`80`	memory % at/above which a card counts as busy
`busyUtilPct`	`50`	utilization % at/above which a card counts as busy
`noGpuRetryMs`	`300000`	how often a GPU-less host is re-probed
`backoffBaseMs` / `maxBackoffMs`	`5000` / `60000`	failure backoff ramp

dsh-remote-gpu-monitoring

安装

说明文档

Configuration