Huawei Cloud Cce Cost Optimization Advisor

Huawei Cloud CCE Cost Optimization Advisor

Overview

Analyze CCE (Cloud Container Engine) cluster cost optimization opportunities. This skill performs read-only analysis and generates configuration suggestions — it does not directly modify HPA, autoscaler, node pools, or workload requests. All configuration changes require explicit user confirmation.

Analysis scope:

• 24-hour and 7-day node CPU/memory utilization trends
• Low-utilization node detection (below cluster average or below 30%)
• Oversized resource request detection (business workloads only)
• HPA and node autoscaler status review and recommendations
• Cost optimization report with execution plan

Architecture: Python SDK v3 → CCE API + AOM PromQL → Inventory + Metrics → Cost Analysis → Report

Security Constraints

Dangerous Operation Confirmation Mechanism

This skill enforces a strict read-only-by-default policy. All write operations require confirm=true.

Operations Requiring Confirmation

Write operations without confirm=true return a preview only. The huawei_configure_cce_hpa tool returns a manifest preview and risk warning when called without confirm=true. Only after explicit user approval can it be called with confirm=true to apply the configuration.

Workflow

Step 1: Preview Operation — Call without confirm=true

python3 scripts/huawei-cloud.py huawei_configure_cce_hpa \
  region=cn-north-4 \
  cluster_id=xxx \
  workload_name=my-deploy \
  namespace=default \
  min_replicas=1 \
  max_replicas=3 \
  target_cpu_utilization=60

Returns: HPA manifest preview, risk warning, confirmation hint

Step 2: Confirm Execution — Call with confirm=true after user approval

python3 scripts/huawei-cloud.py huawei_configure_cce_hpa \
  region=cn-north-4 \
  cluster_id=xxx \
  workload_name=my-deploy \
  namespace=default \
  min_replicas=1 \
  max_replicas=3 \
  target_cpu_utilization=60 \
  confirm=true

Prohibited Actions

• No automatic node pool scale-down — never delete nodes or shrink node pools automatically
• No workload request modification — never change CPU/memory requests directly
• No automatic HPA installation/update — never apply HPA without explicit user confirmation
• No autoscaler enable/disable — never toggle autoscaler without user approval

Allowed Actions

• Read-only queries: nodes, node pools, pods, deployments, metrics, AOM PromQL
• Generate HPA YAML manifests, autoscaler parameter suggestions, and execution plans
• huawei_configure_cce_hpa without confirm=true returns preview only

Credential Security

1. No persistent credential storage — AK/SK exists only during API calls
2. No credential leakage — never includes AK/SK in logs, responses, or errors
3. Environment variable preferred — HW_ACCESS_KEY / HW_SECRET_KEY / HW_REGION_NAME

---

Prerequisites

Python Environment

• Python 3.8+
• Install SDKs: pip install huaweicloudsdkcce huaweicloudsdkcore huaweicloudsdkces
• Optional for HPA operations: pip install kubernetes
• Optional for dashboard charts: pip install matplotlib numpy

Environment Variables (Recommended)

export HW_ACCESS_KEY="your-access-key-id"
export HW_SECRET_KEY="your-secret-access-key"
export HW_REGION_NAME="cn-north-4"

IAM Permission Policies

Ensure the IAM user has the minimum required permissions:

---

Core Commands

Recommended: Combined Analysis

Prefer huawei_analyze_cce_cost_optimization for comprehensive analysis. Only use individual tools below for supplementing details, reviewing specific metrics, or manually generating HPA YAML.

Resource Inventory

Metrics Analysis

Elasticity Policy

HPA configuration workflow:

1. Use huawei_generate_cce_hpa_manifest or huawei_configure_cce_hpa without confirm=true to generate a preview
2. Review the manifest with the user
3. Only after explicit user approval, call huawei_configure_cce_hpa with confirm=true

HPA recommendations must be based on request sizing. If requests are clearly oversized, first recommend calibrating requests, then configure HPA.

Dashboard

---

Parameter Reference

Common Parameters

All tools accept these common parameters for authentication and region:

* cluster_id is not required for huawei_list_cce_clusters (lists all clusters in region).

Combined Analysis Parameters (huawei_analyze_cce_cost_optimization)

HPA Parameters (huawei_generate_cce_hpa_manifest / huawei_configure_cce_hpa)

Metrics Parameters (huawei_get_cce_node_metrics_topN / huawei_get_cce_pod_metrics_topN)

* Only required for single-entity metrics tools.

Dashboard Parameters (huawei_generate_monitor_dashboard)

---

Analysis Workflow

See references/workflow.md for detailed analysis steps, thresholds, and decision logic.

Quick Summary

1. Scope: Confirm region, cluster_id, namespace range, and exclusion rules (default: exclude kube-system)
2. Node utilization: Analyze 24h and 7d windows for CPU/memory usage per node and cluster average
3. Low-utilization detection: Flag nodes below cluster average by 20 percentage points or below 60% of cluster average; cluster average below 30% signals overall over-provisioning
4. Oversized requests: Compare business workload request vs actual p95 usage; mark as high (p95 < 33% of request), optimize (p95 < 50%), or observe (short-window only)
5. Elasticity review: Check node pool autoscaling and HPA status; generate recommendations
6. Output: Summary, utilization tables, oversized request list, HPA/autoscaler recommendations, risks, and verification steps

---

Risk Rules

See references/risk-rules.md for complete safety boundaries.

Key constraints:

• Auto-execution limited to R1 read-only queries only
• No automatic scale-down, request modification, or HPA/autoscaler changes
• Must reference both 24h and 7d windows before recommending scale-down
• Cost optimization suggestions must include rollback strategy and verification metrics
• Data gaps (missing metrics, missing requests, invisible HPA) must be flagged in the report

---

Output Schema

See references/output-schema.md for the complete JSON report structure.

All tools return JSON with:

• status / success: operation result
• data: analysis results, metrics, or configuration preview
• message: human-readable description
• warning: risk warning for write operations (preview mode only)
• files: paths to generated summary JSON and report markdown

---

Supported Regions

---

Best Practices

1. Run the combined analysis first — use huawei_analyze_cce_cost_optimization for a complete picture before drilling into individual tools; avoid piecemeal queries that miss cross-resource dependencies.
2. Always check both time windows — rely on 7-day data for stable optimization decisions; use 24-hour data only for short-term fluctuation observation, never as the sole basis for scale-down recommendations.
3. Exclude kube-system by default — system workloads have fixed sizing requirements; analyzing them produces misleading oversized-request signals and wastes analysis capacity.
4. Calibrate requests before configuring HPA — HPA scales based on request percentages; if requests are oversized, HPA will trigger premature scaling. Fix requests first, then set HPA targets.
5. Use environment variables for credentials — prefer HW_ACCESS_KEY / HW_SECRET_KEY over passing AK/SK as parameters to avoid credential leakage in command history and logs.
6. Review HPA preview before confirming — always call huawei_configure_cce_hpa without confirm=true first; inspect the manifest YAML and risk warning with the user before applying.
7. Include rollback strategy in every recommendation — cost optimization changes can impact availability; every suggestion must specify how to revert and how to verify the change was safe.
8. Flag data gaps explicitly — if metrics are missing, requests are absent, or HPA status is invisible, report these as data gaps; do not infer optimization decisions from incomplete data.
9. Set top_n appropriately — use top_n=50 for large clusters (100+ pods) to capture all significant outliers; reduce to top_n=10 for focused analysis of specific namespaces.
10. Save outputs to a persistent directory — use output_dir to write the summary JSON and report markdown to a known location; this enables later review and comparison across multiple analysis runs.

---

Common Pitfalls

---

Output Format

All tools return JSON with status, success, data, message, warning, and iles fields. See references/output-schema.md for the complete report structure.

Verification

See Verification Method for step-by-step verification.

Cross-Skill References

---

Reference Documents

---

Notes

• Ensure AK/SK has correct IAM permissions (CCE read + AOM read)
• Default analysis excludes kube-system namespace
• HPA recommendations require request sizing to be reasonable first
• Node scale-down suggestions require both 24h and 7d data confirmation
• Cost optimization reports must include rollback strategy
• Data gaps must be explicitly flagged