gke-cost

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

GKE Cost Optimization

GKE成本优化

This reference covers strategies for reducing GKE costs while maintaining the golden path security and reliability posture.

MCP Tools:

get_k8s_resource

describe_k8s_resource

apply_k8s_manifest

patch_k8s_resource

get_cluster

本参考文档涵盖了在保持黄金路径安全与可靠性的前提下降低GKE成本的策略。

MCP工具：

get_k8s_resource

、

describe_k8s_resource

、

apply_k8s_manifest

、

patch_k8s_resource

、

get_cluster

Golden Path Cost Features

黄金路径成本优化特性

The golden path already includes cost-optimizing settings:

Setting	Value	Impact
`autoscalingProfile`	`OPTIMIZE_UTILIZATION`	Aggressive node
: : : scale-down reduces idle :
: : : compute :
`verticalPodAutoscaling`	`enabled`	VPA recommendations
: : : prevent :
: : : over-provisioning :
Autopilot pricing	Pay per pod request	No charge for unused
: : : node capacity :
Node Auto Provisioning	enabled	Right-sized node pools
: : : created automatically :

黄金路径已内置以下成本优化配置：

配置项	取值	影响
`autoscalingProfile`	`OPTIMIZE_UTILIZATION`	主动缩容减少闲置计算资源
`verticalPodAutoscaling`	`enabled`	VPA建议避免过度配置资源
Autopilot定价模式	按Pod请求量付费	闲置节点容量不收费
Node Auto Provisioning	enabled	自动创建规格适配的节点池

Cost Optimization Strategies

成本优化策略

1. Spot VMs via ComputeClasses

1. 通过ComputeClass使用Spot VM

Use Spot VMs for fault-tolerant workloads (60-90% cost reduction).

yaml

apiVersion: cloud.google.com/v1
kind: ComputeClass
metadata:
  name: spot-with-fallback
spec:
  activeMigration:
    optimizeRulePriority: true
  priorities:
  - machineFamily: n4
    spot: true
  - machineFamily: n4
    spot: false

Spot-suitable workloads:

Workload	Spot-Suitable?
Batch / data processing	Yes
Dev / test environments	Yes
Stateless web/API (replicas >= 2)	Yes (with PDBs)
Jobs with checkpointing	Yes
Stateful workloads (databases)	No
Single-replica critical services	No

Handling eviction:

yaml

spec:
  template:
    spec:
      terminationGracePeriodSeconds: 25  # Must be < 30s for Spot
      containers:
      - name: app
        lifecycle:
          preStop:
            exec:
              command: ["/bin/sh", "-c", "sleep 5"]

为容错型工作负载使用Spot VM（可降低60-90%成本）。

yaml

apiVersion: cloud.google.com/v1
kind: ComputeClass
metadata:
  name: spot-with-fallback
spec:
  activeMigration:
    optimizeRulePriority: true
  priorities:
  - machineFamily: n4
    spot: true
  - machineFamily: n4
    spot: false

适合Spot VM的工作负载：

工作负载类型	是否适合Spot VM
批量/数据处理任务	是
开发/测试环境	是
无状态Web/API（副本数≥2）	是（需配置PDB）
带检查点的任务	是
有状态工作负载（数据库）	否
单副本核心服务	否

处理驱逐场景：

yaml

spec:
  template:
    spec:
      terminationGracePeriodSeconds: 25  # Spot VM场景下必须小于30秒
      containers:
      - name: app
        lifecycle:
          preStop:
            exec:
              command: ["/bin/sh", "-c", "sleep 5"]

1.1 Direct Workload Spot Selection (NodeSelector)

1.1 直接为工作负载选择Spot VM（NodeSelector）

For stateless or batch workloads in GKE Autopilot, you do not need to create a custom ComputeClass to use Spot VMs. You can target the built-in Spot capacity directly by adding a

nodeSelector

to your Pod or Deployment spec.

[!WARNING] Preemption Warning: Spot VMs are interruptible and can be preempted at any time with a 30-second notice. Workloads must be fault-tolerant and run with at least 2 replicas for high availability. Always explicitly warn users about this preemption risk when recommending Spot VMs.

yaml

apiVersion: apps/v1
kind: Deployment
metadata:
  name: stateless-spot-app
spec:
  replicas: 2
  template:
    spec:
      nodeSelector:
        cloud.google.com/gke-provisioning: Spot
      terminationGracePeriodSeconds: 25  # Must be < 30s to allow graceful shutdown before preemption
      containers:
      - name: app
        image: <IMAGE>

在GKE Autopilot中，对于无状态或批量工作负载，无需创建自定义ComputeClass即可使用Spot VM。只需在Pod或Deployment配置中添加

nodeSelector

，直接指向内置的Spot资源池。

[!WARNING] 抢占警告：Spot VM为可中断实例，随时可能收到30秒通知后被抢占。工作负载必须具备容错能力，且至少运行2个副本以保证高可用性。推荐Spot VM时务必明确告知用户该抢占风险。

yaml

apiVersion: apps/v1
kind: Deployment
metadata:
  name: stateless-spot-app
spec:
  replicas: 2
  template:
    spec:
      nodeSelector:
        cloud.google.com/gke-provisioning: Spot
      terminationGracePeriodSeconds: 25  # 必须小于30秒，以便在抢占前完成优雅关闭
      containers:
      - name: app
        image: <IMAGE>

2. Pod Rightsizing

2. Pod规格调整

Use VPA recommendations to reduce over-provisioned requests.

bash

undefined

使用VPA建议减少过度配置的资源请求。

bash

undefined

1. Deploy VPA in recommendation mode

1. 以推荐模式部署VPA

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

2. Wait 24+ hours for data collection

2. 等待24小时以上收集数据

3. Read recommendations

3. 查看推荐配置

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


**Optimization rules:**

Condition                     | Action                             | Savings
----------------------------- | ---------------------------------- | -------
CPU request >5x P95 actual    | Reduce to `P95 * 1.2`              | High
Memory request >3x P95 actual | Reduce to `P95 * 1.2`              | High
CPU request >2x P95 actual    | Reduce to `P95 * 1.2`              | Medium
No resource requests set      | Add requests (enables bin-packing) | Medium

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


**优化规则：**

条件                     | 操作                             | 成本节省幅度
----------------------------- | ---------------------------------- | -------
CPU请求量 > 实际P95值的5倍    | 调整为`P95 * 1.2`              | 高
内存请求量 > 实际P95值的3倍 | 调整为`P95 * 1.2`              | 高
CPU请求量 > 实际P95值的2倍    | 调整为`P95 * 1.2`              | 中
未设置资源请求                | 添加资源请求（启用装箱优化） | 中

3. Machine Type Selection

3. 机器类型选择

Family	Use Case	Relative Cost
e2	General purpose, burstable	Lowest
t2a / t2d	Scale-out (Arm/AMD), price-performance	Low
: : optimized : :
n4a	Axion Arm-based, general-purpose	Low
: : price-performance : :
n4 / n4d	General purpose (Intel/AMD), flexible shapes	Low-Medium
c4a	Compute-optimized (Arm), high efficiency	Medium-High
c3 / c4	Compute-optimized (Intel)	Medium-High
c3d / c4d	Compute-optimized (AMD), high-performance	Medium-High
: : throughput : :
ek-standard	Autopilot enhanced (golden path)	Medium
m3 / x4	Memory-optimized, SAP HANA, large databases	High
g2 (L4 GPU)	AI inference	High
a3 (H100 GPU)	AI training	Highest
a4 / a4x	Ultra-scale AI (Blackwell GPUs)	Highest

In Autopilot, machine type is managed. Use ComputeClasses to influence selection.

系列	使用场景	相对成本
e2	通用型、可突发性能	最低
t2a / t2d	横向扩展型（Arm/AMD）、性价比优化	低
n4a	Axion Arm架构、通用型、性价比优化	低
n4 / n4d	通用型（Intel/AMD）、灵活规格	中低
c4a	计算优化型（Arm）、高能效	中高
c3 / c4	计算优化型（Intel）	中高
c3d / c4d	计算优化型（AMD）、高性能吞吐量	中高
ek-standard	Autopilot增强型（黄金路径）	中
m3 / x4	内存优化型、SAP HANA、大型数据库	高
g2 (L4 GPU)	AI推理场景	高
a3 (H100 GPU)	AI训练场景	最高
a4 / a4x	超大规模AI场景（Blackwell GPU）	最高

在Autopilot模式下，机器类型由系统管理。可通过ComputeClass影响类型选择。

4. Committed Use Discounts (CUDs)

4. 承诺使用折扣（CUD）

For steady-state workloads, purchase 1-year or 3-year CUDs:

1-year: ~20-30% discount
3-year: ~50-55% discount
Applied automatically to matching usage in the region
Purchase via Google Cloud Console > Billing > Committed use discounts

对于稳定运行的工作负载，购买1年或3年的CUD：

1年期：约20-30%折扣
3年期：约50-55%折扣
自动应用于区域内匹配的资源使用
通过Google Cloud控制台 > 计费 > 承诺使用折扣进行购买

5. Cluster Management

5. 集群管理

Stop/start dev clusters: Idle dev clusters cost money even with no workloads (control plane fee).
Right-size node pools (Standard): Use Cluster Autoscaler with appropriate min/max.
Multi-tenant clusters: Share a single cluster across teams instead of per-team clusters (see the
```
gke-multitenancy
```
skill).

启停开发集群：闲置的开发集群即使没有工作负载也会产生控制平面费用。
调整节点池规格（Standard模式）：配合Cluster Autoscaler设置合适的最小/最大节点数。
多租户集群：多个团队共享单个集群，而非每个团队单独部署集群（详见
```
gke-multitenancy
```
技能）。

Cost Monitoring

成本监控

bash

undefined

bash

undefined

View cluster cost breakdown (requires Cost Management API)

查看集群成本明细（需启用Cost Management API）

gcloud billing budgets list --billing-account=<BILLING_ACCOUNT> --quiet

View node utilization

查看节点利用率

kubectl top nodes

View pod resource usage vs requests

查看Pod资源使用量 vs 请求量

kubectl top pods --all-namespaces --containers

undefined

kubectl top pods --all-namespaces --containers

undefined

Dev/Test Cost Savings

开发/测试环境成本节省

For non-production environments, these golden path deviations are acceptable:

| Setting | Production (Golden | Dev/Test | : : Path) : : | ----------------------- | ------------------ | ----------------------------- | | Cluster mode | Autopilot | Autopilot (cheaper with fewer | : : : pods) : | Release channel | Regular | Rapid (get fixes faster) | | Private nodes | Required | Optional (simpler access) | | Monitoring components | Full suite | SYSTEM_COMPONENTS only | | Secret Manager rotation | 120s | Disabled | | Maintenance windows | Configured | Not needed |

对于非生产环境，以下黄金路径偏差是可接受的：

配置项	生产环境（黄金路径）	开发/测试环境
集群模式	Autopilot	Autopilot（Pod数量更少时更经济）
发布通道	Regular	Rapid（更快获取修复更新）
私有节点	必填	可选（简化访问）
监控组件	完整套件	仅启用SYSTEM_COMPONENTS
Secret Manager轮换	120s	禁用
维护窗口	已配置	无需配置