Server Operations
Server operation note는 Linux server, GPU driver, storage, account, networking, monitoring의 일반적인 troubleshooting pattern을 다룹니다.
공개 버전은 sanitize된 runbook처럼 읽혀야 합니다.
Public Writing Rules
login-node,gpu-node,shared-storage같은 generic name을 사용합니다.- exact path는
/path/to/project같은 placeholder로 바꿉니다. - private topology를 노출하지 않고 symptom, diagnosis, prevention을 설명합니다.
- live security setting, user list, credential, port, hostname을 공개하지 않습니다.
Runbook Shape
| Section | Write |
|---|---|
| Symptom | user 또는 job이 관찰한 것 |
| Scope | account, storage, GPU, network, scheduler 중 어떤 generic layer가 관련되는가 |
| Evidence | private output이 아니라 public-safe command class 또는 log class |
| Immediate action | reversible 또는 low-risk fix를 먼저 |
| Root cause | 어떤 misconfiguration 또는 resource failure class가 원인인가 |
| Prevention | monitoring, backup, access boundary, documentation, smoke check |
| Sanitization | publishing 전에 제거해야 하는 것 |
Notes
- Admin Usage Patterns: daily server/HPC admin commands grouped by read-only checks, state-changing policy changes, and sensitive evidence boundaries
- GPU driver and CUDA debugging
- Storage mounts and permissions
- Account and group management
- Monitoring shared machines
- Operations Command Cookbook: network, disk IO, GPU Xid, auth logs, Slurm inspection, and public-safe command patterns
- Incident response
- Backup and restore
- Access boundary
- Environment modules and containers
- GPU utilization