菜单
Maintain the UVACompute hub (DO droplet) including disk cleanup, service deployment, and KubeVirt management. Use when hub has disk pressure, services need deployment, or KubeVirt is unhealthy.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Debug UVACompute infrastructure including the DO droplet (orchestrator/hub), workstation node, KubeVirt VMs, and GPU mode. Use when troubleshooting VM creation failures, GPU issues, node connectivity, or checking system status.
Set up GPU for VFIO passthrough to KubeVirt VMs. Use when GPU VMs fail with "no GPU nodes available", "failed to create GPU host-devices", or "group is not viable" errors.
Use when building Go applications requiring concurrent programming, microservices architecture, or high-performance systems. Invoke for goroutines, channels, Go generics, gRPC integration.
Remove AI-generated code slop from the current branch by checking the diff against main
| name | hub-maintenance |
| description | Maintain the UVACompute hub (DO droplet) including disk cleanup, service deployment, and KubeVirt management. Use when hub has disk pressure, services need deployment, or KubeVirt is unhealthy. |
| allowed-tools | Bash, Read |
The hub (uvacompute-ssh) runs on a DigitalOcean droplet with limited disk (8.6G).
ssh root@***REDACTED_IP*** "df -h /; kubectl get nodes; kubectl get pods -n kubevirt"
node.kubernetes.io/disk-pressure:NoSchedulessh root@***REDACTED_IP*** "
df -h /
kubectl describe node uvacompute-ssh | grep -A2 Taints
"
ssh root@***REDACTED_IP*** "
# Prune unused container images
crictl rmi --prune
# Vacuum old journal logs
journalctl --vacuum-size=100M
# Truncate large log files
truncate -s 0 /var/log/auth.log /var/log/kern.log /var/log/syslog
# Remove old rotated logs
rm -f /var/log/*.1 /var/log/*.gz
# Check result
df -h /
"
ssh root@***REDACTED_IP*** "kubectl taint nodes uvacompute-ssh node.kubernetes.io/disk-pressure-"
cd apps/vm-orchestration-service
ssh root@***REDACTED_IP*** "systemctl stop vm-orchestration"
make build-linux && make deploy-hub
The binary was likely copied but systemd wasn't reloaded:
ssh root@***REDACTED_IP*** "systemctl daemon-reload && systemctl restart vm-orchestration && systemctl status vm-orchestration"
ssh root@***REDACTED_IP*** "
systemctl status vm-orchestration
journalctl -u vm-orchestration -n 20 --no-pager
"
Usually disk pressure. Fix disk first, then:
ssh root@***REDACTED_IP*** "kubectl delete pods -n kubevirt --all"
Restart the virt-operator:
ssh root@***REDACTED_IP*** "
kubectl delete pod -n kubevirt -l kubevirt.io=virt-operator
sleep 15
kubectl get pods -n kubevirt
"
ssh root@***REDACTED_IP*** "kubectl delete pods -n kubevirt --field-selector=status.phase=Failed"
ssh root@***REDACTED_IP*** "kubectl get kubevirt -n kubevirt -o yaml | head -50"
# List VMs
ssh root@***REDACTED_IP*** "kubectl get vm -n uvacompute"
# Delete all VMs
ssh root@***REDACTED_IP*** "kubectl delete vm -n uvacompute --all"
# Delete specific VM
ssh root@***REDACTED_IP*** "kubectl delete vm VM_ID -n uvacompute"
| Method | Command |
|---|---|
| SSH (public IP) | ssh root@***REDACTED_IP*** |
| SSH (Tailscale) | Check tailscale status for current IP |
| File | Path |
|---|---|
| Service binary | /usr/local/bin/vm-orchestration |
| Systemd unit | /etc/systemd/system/vm-orchestration.service |
| Service logs | journalctl -u vm-orchestration |