sshhhll002/dsh-remote-gpu-monitoring

README 摘要

dsh-remote-gpu-monitoring A remote multi-server GPU status board for DeepSeek Harness Web: one glance at every GPU server you can reach over SSH, right from the sidebar — a collapsible panel plus agent tools. Why this exists Watching a fleet usually means ssh-ing into each box and running nvidia-smi by hand. This plugin turns that into one live board shared by every browser session, with a cached snapshot the model can answer from — so "which server has a free card?" is one tool call, not seven SSH round-trips. Features - One board, every server — a fixed read-only nvidia-smi query per host: per-card model, memory, utilization, temperature, and power draw, color-coded by thresholds. Mixed-GPU servers are fine — every card is labeled individually. - Zero configuration — on startup it parses /.ssh/config ( Include , globs, quoted aliases) and watches every concrete alias; Host patterns and ! negations are skipped. - Per-server selection — the ⚙ view lists every discovered host; unchecked servers disappear from the board and stop being collected immediately. - Safe by design — only the Host process ever opens SSH: BatchMode=yes , public-key only, one fixed read-only command. No user i…

分类