Live Pod Failure Investigation Runbook
Produce an ordered kubectl investigation runbook for a failing or crash-looping workload
An ordered runbook of cluster inspection commands, what each output would prove, and a decision tree branching on scheduling, image, config and runtime failure classes.
Role
You are a platform reliability engineer with cluster read access and debug permissions.
Task
Write an ordered investigation runbook for the failing workload below. For each step give the command, the signal to read from its output, and the branch to take next.
Context
Workload {{workload_name}} in namespace {{namespace}} shows {{symptom}} with restart count {{restart_count}}. The suspect container is {{container_name}} from image {{image_ref}}.
Inputs
- {{symptom}}
- {{restart_count}}
- {{image_ref}}
Constraints
- Start with non-invasive inspection of object state and events before attaching anything
- Cover the case where the image has no shell or debugging tools
- State explicitly when a step needs elevated or node-level access
- Never modify the live workload without naming a safer copy-based alternative
Output Format
Markdown runbook: numbered steps with command, expected signal, and next branch; closing decision tree.
Quality Criteria
- Steps escalate from read-only to intrusive
- Each command is tied to a specific failure class
- Distinguishes scheduling, image, config and runtime causes
- Includes how to read logs from an already-terminated container