gke-scaling

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

GKE Workload Scaling

GKE工作负载扩缩容

This reference covers scaling workloads on GKE. The golden path enables VPA, OPTIMIZE_UTILIZATION autoscaling profile, and Node Auto Provisioning by default.

MCP Tools:

get_k8s_resource

describe_k8s_resource

apply_k8s_manifest

patch_k8s_resource

get_cluster

update_cluster

update_node_pool

本参考文档介绍GKE上的工作负载扩缩容配置。黄金路径默认启用VPA、OPTIMIZE_UTILIZATION扩缩容配置文件以及节点自动配置（NAP）。

MCP工具：

get_k8s_resource

、

describe_k8s_resource

、

apply_k8s_manifest

、

patch_k8s_resource

、

get_cluster

、

update_cluster

、

update_node_pool

Golden Path Scaling Defaults

黄金路径扩缩容默认配置

Setting	Golden Path Value	Notes
`autoscaling.autoscalingProfile`	`OPTIMIZE_UTILIZATION`	Aggressive scale-down for cost savings
`verticalPodAutoscaling.enabled`	`true`	VPA recommendations available
`autoscaling.enableNodeAutoprovisioning`	`true`	NAP creates node pools on demand
GPU resource limits (T4, A100)	`1000000000` each	NAP can provision GPU nodes

配置项	黄金路径默认值	说明
`autoscaling.autoscalingProfile`	`OPTIMIZE_UTILIZATION`	主动缩容以节省成本
`verticalPodAutoscaling.enabled`	`true`	可获取VPA推荐配置
`autoscaling.enableNodeAutoprovisioning`	`true`	NAP会按需创建节点池
GPU资源限制（T4、A100）	`1000000000` 每个	NAP可配置GPU节点

Scaling Mechanisms

扩缩容机制

1. Manual Scaling

1. 手动扩缩容

kubectl-only — no MCP equivalent for
kubectl scale
. Use kubectl directly.

bash

kubectl scale deployment <DEPLOYMENT> --replicas=<N> -n <NAMESPACE>

仅支持kubectl —— 没有与
kubectl scale
对应的MCP工具。请直接使用kubectl。

bash

kubectl scale deployment <DEPLOYMENT> --replicas=<N> -n <NAMESPACE>

2. Horizontal Pod Autoscaling (HPA)

2. 水平Pod自动扩缩容（HPA）

Scales the number of pods based on metrics.

Quick setup (kubectl-only — no MCP equivalent for
kubectl autoscale
):

bash

kubectl autoscale deployment <DEPLOYMENT> --cpu-percent=50 --min=1 --max=10

Manifest approach (recommended — use MCP
apply_k8s_manifest
):

See assets/hpa-example.yaml for a template.

yaml

apiVersion: autoscaling/v2
kind: HorizontalPodAutoscaler
metadata:
  name: <DEPLOYMENT>-hpa
spec:
  scaleTargetRef:
    apiVersion: apps/v1
    kind: Deployment
    name: <DEPLOYMENT>
  minReplicas: 1
  maxReplicas: 10
  metrics:
  - type: Resource
    resource:
      name: cpu
      target:
        type: Utilization
        averageUtilization: 50

基于指标调整Pod数量。

快速配置（仅支持kubectl —— 没有与
kubectl autoscale
对应的MCP工具）：

bash

kubectl autoscale deployment <DEPLOYMENT> --cpu-percent=50 --min=1 --max=10

清单配置法（推荐 —— 使用MCP工具
apply_k8s_manifest
）：

模板可参考assets/hpa-example.yaml。

yaml

apiVersion: autoscaling/v2
kind: HorizontalPodAutoscaler
metadata:
  name: <DEPLOYMENT>-hpa
spec:
  scaleTargetRef:
    apiVersion: apps/v1
    kind: Deployment
    name: <DEPLOYMENT>
  minReplicas: 1
  maxReplicas: 10
  metrics:
  - type: Resource
    resource:
      name: cpu
      target:
        type: Utilization
        averageUtilization: 50

3. Vertical Pod Autoscaling (VPA)

3. 垂直Pod自动扩缩容（VPA）

Adjusts CPU and memory requests to match actual usage. Enabled by default on golden path.

Update modes:

```
Off
```
— recommendations only (safest, start here)
```
Initial
```
— sets resources only at pod creation
```
Auto
```
— restarts pods to apply new resource values
```
InPlaceOrRecreate
```
— updates resources without restart when possible (GKE 1.34+)

Create VPA in recommendation mode:

yaml

apiVersion: autoscaling.k8s.io/v1
kind: VerticalPodAutoscaler
metadata:
  name: <DEPLOYMENT>-vpa
spec:
  targetRef:
    apiVersion: apps/v1
    kind: Deployment
    name: <DEPLOYMENT>
  updatePolicy:
    updateMode: "Off"

Read recommendations (prefer MCP
describe_k8s_resource
):

undefined

根据实际使用情况调整CPU和内存请求。黄金路径默认启用该功能。

更新模式：

```
Off
```
—— 仅提供推荐配置（最安全，建议从该模式开始）
```
Initial
```
—— 仅在Pod创建时设置资源
```
Auto
```
—— 重启Pod以应用新的资源配置
```
InPlaceOrRecreate
```
—— 尽可能在不重启的情况下更新资源（GKE 1.34+版本支持）

创建推荐模式的VPA：

yaml

apiVersion: autoscaling.k8s.io/v1
kind: VerticalPodAutoscaler
metadata:
  name: <DEPLOYMENT>-vpa
spec:
  targetRef:
    apiVersion: apps/v1
    kind: Deployment
    name: <DEPLOYMENT>
  updatePolicy:
    updateMode: "Off"

查看推荐配置（优先使用MCP工具
describe_k8s_resource
）：

undefined

MCP (preferred)

MCP（优先选择）

describe_k8s_resource(parent="...", resourceType="verticalpodautoscaler", name="<DEPLOYMENT>-vpa", namespace="<NAMESPACE>")

kubectl fallback

kubectl备选方案

kubectl get vpa <DEPLOYMENT>-vpa -o jsonpath='{.status.recommendation}'


See [assets/vpa-example.yaml](./assets/vpa-example.yaml) for a full template.

kubectl get vpa <DEPLOYMENT>-vpa -o jsonpath='{.status.recommendation}'


完整模板可参考[assets/vpa-example.yaml](./assets/vpa-example.yaml)。

4. Cluster Autoscaler / Node Auto Provisioning (NAP)

4. 集群自动扩缩容 / 节点自动配置（NAP）

On Autopilot (golden path), node scaling is fully managed. NAP automatically creates and sizes node pools based on workload demands.

For Standard clusters:

bash

undefined

在Autopilot模式（黄金路径）下，节点扩缩容完全由平台管理。NAP会根据工作负载需求自动创建并调整节点池规模。

对于Standard集群：

bash

undefined

Enable cluster autoscaler on a node pool

在节点池上启用集群自动扩缩容

gcloud container clusters update <CLUSTER_NAME> --region <REGION>
--enable-autoscaling --node-pool <POOL_NAME>
--min-nodes <MIN> --max-nodes <MAX>
--quiet

Enable NAP

启用NAP

gcloud container clusters update <CLUSTER_NAME> --region <REGION>
--enable-autoprovisioning
--min-cpu <MIN_CPU> --max-cpu <MAX_CPU>
--min-memory <MIN_MEM> --max-memory <MAX_MEM>
--quiet


**Autoscaling profiles:**

| Profile                | Behavior                             | Golden Path? |
| ---------------------- | ------------------------------------ | ------------ |
| `BALANCED`             | Default GKE; conservative scale-down | No           |
| `OPTIMIZE_UTILIZATION` | Aggressive scale-down; lower idle    | **Yes**      |
:                        : resources                            :              :

gcloud container clusters update <CLUSTER_NAME> --region <REGION>
--enable-autoprovisioning
--min-cpu <MIN_CPU> --max-cpu <MAX_CPU>
--min-memory <MIN_MEM> --max-memory <MAX_MEM>
--quiet


**扩缩容配置文件：**

| 配置文件                | 行为描述                             | 是否为黄金路径默认值 |
| ---------------------- | ------------------------------------ | ------------ |
| `BALANCED`             | GKE默认配置；保守缩容策略 | 否           |
| `OPTIMIZE_UTILIZATION` | 主动缩容；减少闲置资源    | **是**      |

Best Practices

最佳实践

Define resource requests: HPA and VPA rely on accurate requests. Always set them.
Avoid metric conflicts: Do not use HPA and VPA on the same metric. Typical pattern: HPA on CPU, VPA on memory.
Pod Disruption Budgets: Define PDBs for all production workloads to ensure availability during scaling events.
HPA stabilization: HPA has a default 5-minute stabilization window. Tune
```
behavior
```
for faster response if needed.
VPA "Auto" caution: Auto mode restarts pods. Ensure your app handles SIGTERM gracefully. VPA requires at least 2 replicas for evictions by default.
Use ComputeClasses: For workload-specific node targeting (Spot fallback, GPU, specific machine families), use ComputeClasses instead of node selectors.

定义资源请求：HPA和VPA依赖准确的资源请求，请务必配置。
避免指标冲突：不要在同一指标上同时使用HPA和VPA。典型搭配：HPA基于CPU，VPA基于内存。
Pod中断预算（PDB）：为所有生产环境工作负载定义PDB，确保扩缩容事件期间的可用性。
HPA稳定窗口：HPA默认有5分钟稳定窗口，如需更快响应可调整
```
behavior
```
配置。
VPA "Auto"模式注意事项：Auto模式会重启Pod，请确保应用能优雅处理SIGTERM信号。默认情况下，VPA需要至少2个副本才能执行驱逐操作。
使用ComputeClasses：如需针对特定工作负载选择节点（如Spot实例 fallback、GPU、特定机型系列），请使用ComputeClasses而非节点选择器。

Rightsizing Workflow

资源优化流程

Deploy VPA in
```
Off
```
mode for 24+ hours
Read recommendations:
```
kubectl describe vpa <NAME>
```
Compare
```
target
```
values against current
```
requests
```
Apply with 20% buffer:
```
new_request = target * 1.2
```
Use patch format to update Deployment

Condition	Recommendation	Risk
CPU request >5x P95 actual	Reduce to `P95 * 1.2`	Medium
Memory request >3x P95 actual	Reduce to `P95 * 1.2`	Medium
CPU request >2x P95 actual	Rightsizing with 20% buffer	Low
No resource limits set	Add limits to prevent noisy-neighbor	Low

以
```
Off
```
模式部署VPA并运行24小时以上
查看推荐配置：
```
kubectl describe vpa <NAME>
```
对比
```
target
```
值与当前
```
requests
```
配置
预留20%缓冲后应用配置：
```
新请求值 = target * 1.2
```
使用patch格式更新Deployment

场景	建议措施	风险等级
CPU请求 > 实际P95值的5倍	调整为 `P95 * 1.2`	中等
内存请求 > 实际P95值的3倍	调整为 `P95 * 1.2`	中等
CPU请求 > 实际P95值的2倍	预留20%缓冲进行资源优化	低
未设置资源限制	添加限制以避免“噪声邻居”问题	低