feat: explain Jobs, CronJobs, PersistentVolumeClaims, and Nodes

[nklmilojevic]

Summary

X only had dedicated analysis for workloads and pods. Every other kind fell back to its Ready condition. A failed Job read as "exposes no Ready condition to assess", and a Pending claim or a broken CronJob got no explanation at all. Three separate discussions asked for this: #677, #678 and #690.

• Job: the Failed or Complete condition and its reason, progress, failures used against backoffLimit while it runs, and failed pods with their exit reasons.

• CronJob: suspension, schedule and time zone, last scheduled and successful runs, runs skipped by concurrencyPolicy: Forbid, its five newest Jobs (each a target for E), and the pods of the latest Job.

• PersistentVolumeClaim: why a claim is not bound:

  • a missing StorageClass, or no default class
  • WaitForFirstConsumer with no pod using the claim
  • an unprovisioned volume
  • a pending bind to a named volume

  Also covered: a resize that has not finished, a ReadWriteOnce claim used on several nodes, and the pods that mount the claim. Listing StorageClasses needs cluster-scope RBAC. Without it, a class is reported as unknown, never as missing.

• Node: Ready state and since when, cordon, taints, pod capacity, and up to ten unhealthy pods on the node. Pressure-condition handling is unchanged.

Evidence gathering for these kinds lives in one gather_evidence that both explain and :bundle use, so diagnostic bundles get the same analysis.

Discussions: #677, #678, #690

Test plan

• just check: fmt, clippy -D warnings, 2002 tests
• Unit tests for each new analysis in src/explain.rs
• X key tests for a Pending claim (class lookup, consumer pods filtered by claim), a CronJob (only its own Jobs), and a Node (cordon, unhealthy pods)
• Manual: X on a failed Job, a Pending PVC, and a cordoned node on a real cluster

[kritikal-github]

Kritika Review

Explains Jobs, CronJobs, claims, and Nodes with shared evidence gathering.

No findings

Confidence 4/5 · medium risk: src/app/explain.rs skips Job pods when a valid selector uses only matchExpressions, leaving failures and restarts out of the explanation; this case is not covered by the reported review. Risk is medium because this affects application diagnostics rather than cluster state.

Summary

The change adds dedicated diagnostics for four resource kinds and shares evidence gathering between Explain and bundles. The latest adjustment reads StorageClasses independently of Pod discovery and marks unavailable Pod evidence as unknown, preserving the distinction between missing and unlisted resources.

What's good

• Shared evidence gathering keeps interactive explanations and bundles consistent.
• Tests exercise missing Pod discovery and the distinction between unreadable and absent evidence.

58 context chunk(s) left out of the prompt to fit its budget.

_{Reviews (4) · Last reviewed commit: "fix: treat pods as unknown when discover..." · kritika with chatgpt/gpt-6-sol}

[greptile-apps]

Confidence Score: 5/5

[Medium risk] Adds diagnostic analysis for four Kubernetes resource types.

The PR appears safe to merge; no new actionable issue was identified.

Summary

The PR adds dedicated explanations for Jobs, CronJobs, PersistentVolumeClaims, and Nodes, and shares evidence gathering between Explain and diagnostic bundles.

• Related pods, Jobs, StorageClasses, and events inform the new findings.
• The latest changes distinguish unavailable pod evidence from an empty pod list.

Diagram

%%{init: {'theme': 'neutral'}}%%
flowchart LR
  A[Selected resource] --> B[gather_evidence]
  B --> C[Related resources and events]
  C --> D[Kind-specific explanation]
  D --> E[Explain view]
  D --> F[Diagnostic bundle]

Loading

_{Reviews (4) · Last reviewed commit: "fix: treat pods as unknown when discover..."}