ECS Linux Troubleshooting

This skill applies only to Linux GuestOS on Alibaba Cloud ECS and diagnoses abnormal issues on the Alibaba Cloud ECS Linux instance specified by the user (the target instance). The agent may run on the target instance or another machine, and uses the aliyun CLI to remotely diagnose and collect data from the target ECS instance.

When to Use This Skill

• The troubleshooting target specified by the user is an Alibaba Cloud ECS Linux instance.
• The user's description involves abnormal issues such as mismatches between startup/running status and OS state, remote login failures, network connectivity failures, disk expansion/mount issues, performance anomalies, crashes or hangs, clock drift, or configurations not taking effect.
• Not applicable: non-Alibaba Cloud ECS, non-Linux GuestOS, other clouds or physical machines; purely control-plane, billing, or API-side issues with no GuestOS involvement.

Principles and Requirements

1. Troubleshooting target = the Alibaba Cloud ECS Linux instance specified by the user: all checks and conclusions must target that one machine. Commands and paths in references/ are written for the in-instance environment.
2. Use the aliyun CLI for remote diagnosis and data collection: except for the steps in references/utils/guestos-pe-prep.md, all other steps may only use the subcommands listed in references/aliyun-cli-cheatsheet.md. Do not call subcommands that are not listed there.
3. Clarify the problem before investigating: strictly follow the troubleshooting workflow. First narrow the user's description to a phenomenon domain, then follow the corresponding troubleshooting document. Do not skip phases or reorder them, and do not draw a conclusion or stop the workflow early when the evidence does not uniquely point to a single root cause.
4. Prefer the commands described in the references/<slug>.md troubleshooting document. Do not blindly guess commands on your own.
5. External reference material: when a troubleshooting document contains a URL, such as an Alibaba Cloud help page, you must fetch the content of that link and use it as the basis for the investigation. Do not judge based only on the link title or on prior knowledge.
6. Multiple-instance scenarios: troubleshoot each ECS instance separately. Do not use results from instance A to draw conclusions about instance B.
7. Least-privilege permissions: before running this skill, ensure the caller has only the required RAM actions for the selected workflow. Use references/ram-policies.md as the permission source of truth.