Cluster Update Advisor

Purpose

Assess cluster update readiness and produce a structured risk report with actionable prerequisites, blockers, and recommendations.

The proposal request includes pre-collected cluster readiness data (JSON) gathered by the Cluster Version Operator. Analyze this data, classify findings, and produce a decision with evidence. Do not re-collect cluster data — it is already in the request.

Inputs

The proposal request contains:

• Current and target version metadata
• Channel and update path information
• Cluster readiness JSON — cluster health checks with context relevant to preparing for the update

The readiness JSON is embedded in the request between ```json markers under the "Cluster Readiness Data" heading. Parse it to begin analysis.

Readiness JSON structure:

{
  "current_version": "4.21.5",
  "target_version": "4.21.8",
  "checks": {
    "cluster_conditions":    { "_status": "ok", "summary": {...}, ... },
    "operator_health":       { "_status": "ok", "summary": {...}, ... },
    "api_deprecations":      { "_status": "ok", "summary": {...}, ... },
    "node_capacity":         { "_status": "ok", "summary": {...}, ... },
    "pdb_drain":             { "_status": "ok", "summary": {...}, ... },
    "etcd_health":           { "_status": "ok", "summary": {...}, ... },
    "network":               { "_status": "ok", "summary": {...}, ... },
    "crd_compat":            { "_status": "ok", "summary": {...}, ... },
    "olm_operator_lifecycle": { "_status": "ok", "summary": {...}, ... }
  }
}

Each check contains _status (ok or error) and check-specific data with a summary section for quick parsing.

Evaluation

Parse readiness data

Extract the JSON from the proposal request. Count checks with _status ok vs error for completeness.

Verify data completeness

Any check with _status error represents a gap in visibility. Note incomplete areas — they reduce confidence.

Evaluate findings in detail

If the system prompt includes organization-specific policy (thresholds, scheduling preferences, risk tolerance), apply those constraints. Otherwise use sensible defaults. Walk through each check's summary and detail data:

• Compare numeric thresholds (node headroom, etcd backup age)
• Evaluate conditional update risks against cluster state
• Identify compounding risks (e.g., paused MCP + cert expiry)
• Estimate update duration (~10 min/node)

Classify findings

Assign each finding a severity per the classification table.

For other checks, treat an issue as a blocker if would cause data loss, a performance regression, or a failed update. Treat the issue as a warning if would cause temporary disruption or slow updates.

Investigate with other skills

If additional information or context is needed to classify a finding, these skills may be useful:

• openshift-docs — Read official OpenShift update docs for version-specific procedures and breaking changes.
• prometheus — Query cluster metrics for trend analysis (etcd latency, CPU headroom, firing alerts).
• jira — Search Red Hat Jira for bugs and known issues affecting the target version.
• product-lifecycle — Query Red Hat Product Life Cycle API to check support status and OCP compatibility for installed operators. Use the operator's package name from OLM readiness data to look up entries via the package field (exact match). Flag operators whose product version is End of life or whose openshift_compatibility does not include the target OCP version.

Classify overall recommendation

Aggregate finding classification, and and make a decision on the overall assessment:

• escalate — insufficient data for confident assessment.
• block — findings must be resolved before update.
• warn — findings exist but manageable with prerequisites.
• recommend — all checks pass within acceptable thresholds.

Produce a structured risk report

The output schema is enforced by the OlsAgent CR's outputSchema field — the operator handles structured output compliance via the LLM API.

Failure Modes — What NOT to Do

1. Never recommend updating without analyzing the readiness data. The JSON in the request is the source of truth.
2. Never dismiss conditional update risks. If the update path is conditional, evaluate each risk against the cluster.
3. Never skip the API deprecation check. Workloads using removed APIs will break after the update.
4. Never assume etcd is healthy. Always check member health in the readiness data.
5. Never fabricate Jira issue keys, KB article IDs, or CVE numbers. Use the redhat-support skill to get real data.
6. Never recommend skipping an update version unless the readiness data shows that path exists.
7. Never recommend force-updating. If the standard path is blocked, report it.