Capacity Planning · 容量規劃

---
doc_type: "capacity-planning"
variant: "light"
status: "draft"
owner: "<your-name>"
last_updated: "YYYY-MM-DD"
upstream:
  required: ["slo", "historical-metrics"]
  optional: ["business-forecast"]
---

# Capacity Planning: <service-name>

**Status:** Draft v0.X · **Owner:** <SRE name> · **Last updated:** YYYY-MM-DD

> [!IMPORTANT]
> **AI 填寫規則：** 本範本 6 段（編號 1, 2, 4, 5, 7, 9），全部必填——刻意沿用完整版章節編號讓兩版可對照。每結論行內加 `（依據：metric §XXX / forecast §YYY）`；每量化欄位帶 `[H]/[M]/[L]` confidence badge；缺資料寫 `_TODO: 需要 XXX_` 不編造；headroom 必須 ≥ 30%（洪峰緩衝）；至少涵蓋 CPU / RAM / DB / queue 四維度。

---

## 1. Executive Summary

<!-- ai-fill: 3-5 行，主管 30 秒讀完。內容：forecast horizon、最快撐不住的資源、預估擴容月份、cost delta -->

<3-5 行說明>

> **TL;DR:** <一句話：N 月內，X 資源會先撐不住，建議 Y 動作>

---

## 2. Forecast Horizon & Growth Model

<!-- ai-rule: horizon 對應預算週期（3m / 6m / 12m）；growth model 必須附公式與信賴區間 -->

| Field | Value | Confidence |
|---|---|---|
| **Forecast horizon** | <3m / 6m / 12m> | **[H]** |
| **Growth model** | linear / exponential / seasonal / step | **[M]** |
| **MoM growth rate** | <e.g. +12%> | **[M]** |
| **Peak multiplier** | <e.g. 黑五 ×3.5> | **[L]** |
| **Confidence interval** | <e.g. ±15%> | **[M]** |

---

## 4. Headroom per Resource

<!-- ai-rule: 輕量版至少 4 維度（CPU / RAM / DB / queue）；headroom_target ≥ 30%；status breach 必須對應一條 bottleneck -->

| Resource | Current p95 | Forecast p95 | Headroom target | Status | Confidence |
|---|---|---|---|---|---|
| <checkout-api CPU> | 62% | 88% | ≥ 30% | warn | **[H]** |
| <checkout-api RAM> | ... | ... | ≥ 30% | safe | **[H]** |
| <payment-db conn> | ... | ... | ≥ 30% | breach | **[M]** |
| <order-queue depth> | ... | ... | ≥ 30% | safe | **[M]** |

---

## 5. Scaling Triggers

<!-- ai-rule: 每個 breach / warn 資源都要列 trigger；含 auto + manual + lead_time -->

| Resource | Auto scale-up | Auto scale-down | Manual review | Lead time |
|---|---|---|---|---|
| <checkout-api> | p95 CPU > 70% for 5m | p95 CPU < 30% for 30m | forecast headroom < 30% | 4 weeks |

---

## 7. Cost Envelope

<!-- ai-rule: 必填 current / forecast / delta / error band；無預算文件寫 _TODO_ -->

| Field | Value | Confidence |
|---|---|---|
| **Current monthly** | $X | **[H]** |
| **Forecast at horizon** | $Y | **[M]** |
| **Delta** | +Z% | **[M]** |
| **RI / SP saving** | $W | **[L]** |
| **Forecast error band** | ±15% | **[M]** |

---

## 9. Decision Log（key 2-3 條）

<!-- ai-rule: 每條必含 chosen + 至少 1 個 rejected + 拒絕原因 -->

| Date | Decision | Options | Chosen | Rejected why | Confidence |
|---|---|---|---|---|---|
| YYYY-MM-DD | <vertical vs horizontal scale> | vertical / horizontal / sharding | horizontal | vertical (碰到 instance ceiling)、sharding (lead time 不足) | **[H]** |

---

> [!CAUTION]
> **輸出前 AI 自檢：**
> - [ ] 6 段 H2 章節齊全（編號 1, 2, 4, 5, 7, 9）
> - [ ] Headroom 表至少 4 維度（CPU / RAM / DB / queue）
> - [ ] 每個 headroom 行含 `[H/M/L]` badge + status
> - [ ] 每個 breach / warn 資源有對應 scaling trigger
> - [ ] Cost envelope 含 error band
> - [ ] Decision Log ≥ 1 條，每條有 rejected reason
> - [ ] 無 YAML / JSON schema 輸出（capacity plan 是給人讀的 markdown）