gke-cost

Name: gke-cost
Author: google

1.3kinstalls1.1kforks

How to install gke-cost?

google/skills/gke-cost

$npx -y skills add google/skills --skill gke-cost

Installs into the current project.

Summary

Optimizes GKE costs, rightsizes workloads, and configures Spot VMs and CUDs. Use when optimizing GKE costs, rightsizing GKE workloads, or configuring GKE Spot VMs. Don't use for general compute class provisioning or GPU Selection (use gke-compute-classes instead).

Files · 1

View on GitHub

SKILL.md

1	# GKE Cost Optimization
2
3	This reference covers strategies for reducing GKE costs while maintaining the
4	golden path security and reliability posture.
5
6	> MCP Tools: `get_k8s_resource`, `describe_k8s_resource`,
7	> `apply_k8s_manifest`, `patch_k8s_resource`, `get_cluster`
8
9	## Golden Path Cost Features
10
11	The golden path already includes cost-optimizing settings:
12
13	\| Setting \| Value \| Impact \|
14	\| ------------------------ \| ---------------------- \| ----------------------- \|
15	\| `autoscalingProfile` \| `OPTIMIZE_UTILIZATION` \| Aggressive node \|
16	: : : scale-down reduces idle :
17	: : : compute :
18	\| `verticalPodAutoscaling` \| `enabled` \| VPA recommendations \|
19	: : : prevent :
20	: : : over-provisioning :
21	\| Autopilot pricing \| Pay per pod request \| No charge for unused \|
22	: : : node capacity :
23	\| Node Auto Provisioning \| enabled \| Right-sized node pools \|
24	: : : created automatically :
25
26	## Cost Optimization Strategies
27
28	### 1. Spot VMs via ComputeClasses
29
30	Use Spot VMs for fault-tolerant workloads (60-90% cost reduction).
31
32	```yaml
33	apiVersion: cloud.google.com/v1
34	kind: ComputeClass
35	metadata:
36	name: spot-with-fallback
37	spec:
38	activeMigration:
39	optimizeRulePriority: true
40	priorities:
41	- machineFamily: n4
42	spot: true
43	- machineFamily: n4
44	spot: false
45	```
46
47	Spot-suitable workloads:
48
49	Workload \| Spot-Suitable?
50	--------------------------------- \| ---------------
51	Batch / data processing \| Yes
52	Dev / test environments \| Yes
53	Stateless web/API (replicas >= 2) \| Yes (with PDBs)
54	Jobs with checkpointing \| Yes
55	Stateful workloads (databases) \| No
56	Single-replica critical services \| No
57
58	Handling eviction:
59
60	```yaml
61	spec:
62	template:
63	spec:
64	terminationGracePeriodSeconds: 25 # Must be < 30s for Spot
65	containers:
66	- name: app
67	lifecycle:
68	preStop:
69	exec:
70	command: ["/bin/sh", "-c", "sleep 5"]
71	```
72
73	### 1.1 Direct Workload Spot Selection (NodeSelector)
74
75	For stateless or batch workloads in GKE Autopilot, you do not need to create a
76	custom ComputeClass to use Spot VMs. You can target the built-in Spot capacity
77	directly by adding a `nodeSelector` to your Pod or Deployment spec.
78
79	> [!WARNING] Preemption Warning: Spot VMs are interruptible and can be
80	> preempted at any time with a 30-second notice. Workloads must be
81	> fault-tolerant and run with at least 2 replicas for high availability. Always
82	> explicitly warn users about this preemption risk when recommending Spot VMs.
83
84	```yaml
85	apiVersion: apps/v1
86	kind: Deployment
87	metadata:
88	name: stateless-spot-app
89	spec:
90	replicas: 2
91	template:
92	spec:
93	nodeSelector:
94	cloud.google.com/gke-provisioning: Spot
95	terminationGracePeriodSeconds: 25 # Must be < 30s to allow graceful shutdown before preemption
96	containers:
97	- name: app
98	image: <IMAGE>
99	```
100
101	### 2. Pod Rightsizing
102
103	Use VPA recommendations to reduce over-provisioned requests.
104
105	```bash
106	# 1. Deploy VPA in recommendation mode
107	kubectl apply -f - <<EOF
108	apiVersion: autoscaling.k8s.io/v1
109	kind: VerticalPodAutoscaler
110	metadata:
111	name: <DEPLOYMENT>-vpa
112	spec:
113	targetRef:
114	apiVersion: apps/v1
115	kind: Deployment
116	name: <DEPLOYMENT>
117	updatePolicy:
118	updateMode: "Off"
119	EOF
120
121	# 2. Wait 24+ hours for data collection
122
123	# 3. Read recommendations
124	kubectl get vpa <DEPLOYMENT>-vpa -o jsonpath='{.status.recommendation}'
125	```
126
127	Optimization rules:
128
129	Condition \| Action \| Savings
130	----------------------------- \| ---------------------------------- \| -------
131	CPU request >5x P95 actual \| Reduce to `P95 * 1.2` \| High
132	Memory request >3x P95 actual \| Reduce to `P95 * 1.2` \| High
133	CPU request >2x P95 actual \| Reduce to `P95 * 1.2` \| Medium
134	No resource requests set \| Add requests (enables bin-packing) \| Medium
135
136	### 3. Machine Type Selection
137
138	\| Family \| Use Case \| Relative Cost \|
139	\| ------------- \| -------------------------------------------- \| ------------- \|
140	\| e2 \| General purpose, burstable \| Lowest \|
141	\| t2a / t2d \| Scale-out (Arm/AMD), pri

People who liked this also viewed

6 picks

Type

gke-cost

1.3kinstalls1.1kforks

View on GitHub

How to install gke-cost?

google/skills/gke-cost

$npx -y skills add google/skills --skill gke-cost

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# GKE Cost Optimization
2
3	This reference covers strategies for reducing GKE costs while maintaining the
4	golden path security and reliability posture.
5
6	> MCP Tools: `get_k8s_resource`, `describe_k8s_resource`,
7	> `apply_k8s_manifest`, `patch_k8s_resource`, `get_cluster`
8
9	## Golden Path Cost Features
10
11	The golden path already includes cost-optimizing settings:
12
13	\| Setting \| Value \| Impact \|
14	\| ------------------------ \| ---------------------- \| ----------------------- \|
15	\| `autoscalingProfile` \| `OPTIMIZE_UTILIZATION` \| Aggressive node \|
16	: : : scale-down reduces idle :
17	: : : compute :
18	\| `verticalPodAutoscaling` \| `enabled` \| VPA recommendations \|
19	: : : prevent :
20	: : : over-provisioning :
21	\| Autopilot pricing \| Pay per pod request \| No charge for unused \|
22	: : : node capacity :
23	\| Node Auto Provisioning \| enabled \| Right-sized node pools \|
24	: : : created automatically :
25
26	## Cost Optimization Strategies
27
28	### 1. Spot VMs via ComputeClasses
29
30	Use Spot VMs for fault-tolerant workloads (60-90% cost reduction).
31
32	```yaml
33	apiVersion: cloud.google.com/v1
34	kind: ComputeClass
35	metadata:
36	name: spot-with-fallback
37	spec:
38	activeMigration:
39	optimizeRulePriority: true
40	priorities:
41	- machineFamily: n4
42	spot: true
43	- machineFamily: n4
44	spot: false
45	```
46
47	Spot-suitable workloads:
48
49	Workload \| Spot-Suitable?
50	--------------------------------- \| ---------------
51	Batch / data processing \| Yes
52	Dev / test environments \| Yes
53	Stateless web/API (replicas >= 2) \| Yes (with PDBs)
54	Jobs with checkpointing \| Yes
55	Stateful workloads (databases) \| No
56	Single-replica critical services \| No
57
58	Handling eviction:
59
60	```yaml
61	spec:
62	template:
63	spec:
64	terminationGracePeriodSeconds: 25 # Must be < 30s for Spot
65	containers:
66	- name: app
67	lifecycle:
68	preStop:
69	exec:
70	command: ["/bin/sh", "-c", "sleep 5"]
71	```
72
73	### 1.1 Direct Workload Spot Selection (NodeSelector)
74
75	For stateless or batch workloads in GKE Autopilot, you do not need to create a
76	custom ComputeClass to use Spot VMs. You can target the built-in Spot capacity
77	directly by adding a `nodeSelector` to your Pod or Deployment spec.
78
79	> [!WARNING] Preemption Warning: Spot VMs are interruptible and can be
80	> preempted at any time with a 30-second notice. Workloads must be
81	> fault-tolerant and run with at least 2 replicas for high availability. Always
82	> explicitly warn users about this preemption risk when recommending Spot VMs.
83
84	```yaml
85	apiVersion: apps/v1
86	kind: Deployment
87	metadata:
88	name: stateless-spot-app
89	spec:
90	replicas: 2
91	template:
92	spec:
93	nodeSelector:
94	cloud.google.com/gke-provisioning: Spot
95	terminationGracePeriodSeconds: 25 # Must be < 30s to allow graceful shutdown before preemption
96	containers:
97	- name: app
98	image: <IMAGE>
99	```
100
101	### 2. Pod Rightsizing
102
103	Use VPA recommendations to reduce over-provisioned requests.
104
105	```bash
106	# 1. Deploy VPA in recommendation mode
107	kubectl apply -f - <<EOF
108	apiVersion: autoscaling.k8s.io/v1
109	kind: VerticalPodAutoscaler
110	metadata:
111	name: <DEPLOYMENT>-vpa
112	spec:
113	targetRef:
114	apiVersion: apps/v1
115	kind: Deployment
116	name: <DEPLOYMENT>
117	updatePolicy:
118	updateMode: "Off"
119	EOF
120
121	# 2. Wait 24+ hours for data collection
122
123	# 3. Read recommendations
124	kubectl get vpa <DEPLOYMENT>-vpa -o jsonpath='{.status.recommendation}'
125	```
126
127	Optimization rules:
128
129	Condition \| Action \| Savings
130	----------------------------- \| ---------------------------------- \| -------
131	CPU request >5x P95 actual \| Reduce to `P95 * 1.2` \| High
132	Memory request >3x P95 actual \| Reduce to `P95 * 1.2` \| High
133	CPU request >2x P95 actual \| Reduce to `P95 * 1.2` \| Medium
134	No resource requests set \| Add requests (enables bin-packing) \| Medium
135
136	### 3. Machine Type Selection
137
138	\| Family \| Use Case \| Relative Cost \|
139	\| ------------- \| -------------------------------------------- \| ------------- \|
140	\| e2 \| General purpose, burstable \| Lowest \|
141	\| t2a / t2d \| Scale-out (Arm/AMD), pri

People who liked this also viewed

6 picks

Type