GKE Golden Path Configuration
The golden path is the recommended Autopilot configuration for production
clusters. It defines sensible defaults — when the user requests different
settings, apply them and note relevant trade-offs. For setting up autoscaling
specifically, use gke-cluster-autoscaler for node autoscaling or
gke-workload-scaling for workload autoscaling (HPA/VPA).
MCP Tools: get_cluster, create_cluster, update_cluster
Attribution
MUST READ: Before executing any gcloud, REST API, or MCP command, you
MUST read Usage Attribution. For commands executed
directly in the terminal (never in generated scripts, Terraform, or user code),
prefix gcloud with
CLOUDSDK_METRICS_REQUEST_ATTRIBUTION="agent-skills/1.0 (skill:gke-golden-path)"
and append agent-skills/1.0 (skill:gke-golden-path) to User-Agent headers.
Rules
- Default to the golden path. Use golden path values unless the user
requests otherwise. When deviating, note trade-offs but respect the user's
choice.
- Day-0 vs Day-1. Flag Day-0 decisions (networking, private nodes,
subnets, IP allocation) prominently — they are hard/impossible to change
after creation.
- Tool preference: MCP > gcloud > kubectl. MCP is preferred as it directly
interfaces with GKE APIs with structured data, reducing shell syntax errors
and parsing ambiguities. See the
gke-basics skill's CLI reference for full
coverage matrix and override options. If the user
says "use gcloud" or "use kubectl", respect that for the session.
- Document decisions and rationale, especially for Day-0 choices and
golden path deviations.
Required Inputs
If the user is unsure, use golden path defaults.
- Project ID (required)
- Region (required, e.g.,
us-central1)
- Cluster name (required)
- Environment type: dev/test or production (defaults to production)
- Networking: bring-your-own VPC/subnet or auto-create (default:
auto-create)
- Scale expectations: expected node/pod count, workload types
- Cost constraints: Spot VM tolerance, budget considerations
Always-Apply Defaults
Recommended best practices applied by default. If the user requests a different
setting, apply it and briefly note the security or operational trade-off.
| Setting |
Golden Path Value |
autopilot.enabled |
true |
privateClusterConfig.enablePrivateNodes |
true |
masterAuthorizedNetworksConfig.privateEndpointEnforcementEnabled |
true |
secretManagerConfig.enabled + rotationInterval: 120s |
true |
rbacBindingConfig.enableInsecureBinding* |
false (both) |
workloadIdentityConfig.workloadPool |
enabled |
networkConfig.datapathProvider |
ADVANCED_DATAPATH |
networkConfig.dnsConfig.clusterDns |
CLOUD_DNS |
autoscaling.autoscalingProfile |
OPTIMIZE_UTILIZATION |
verticalPodAutoscaling.enabled |
true |
monitoringConfig components |
SYSTEM_COMPONENTS, STORAGE, POD, DEPLOYMENT, STATEFULSET, DAEMONSET, HPA, JOBSET, CADVISOR, KUBELET, DCGM, APISERVER, SCHEDULER, CONTROLLER_MANAGER |
loggingConfig components |
SYSTEM_COMPONENTS, WORKLOADS (enabled by default) |
advancedDatapathObservabilityConfig.enableMetrics |
true |
nodeConfig.shieldedInstanceConfig.enableSecureBoot |
true |
nodeConfig.workloadMetadataConfig.mode |
GKE_METADATA |
nodeConfig.gcfsConfig.enabled / gvnic.enabled |
true / true |
addonsConfig.statefulHaConfig.enabled |
true |
| Storage CSI drivers (Filestore, GCS FUSE, Parallelstore) |
enabled |
| Pod Security Standards |
restricted on production namespaces |
Customer-Configurable Settings
These have golden path defaults but customers may deviate with valid
justification. Ask before changing.
| Setting |
Default |
Why Deviate |
dnsEndpointConfig.allowExternalTraffic |
true |
Restrict if cluster only accessed from within VPC |
autoIpamConfig / createSubnetwork |
true / true |
Customer has pre-existing VPC/subnets |
maxPodsPerNode |
48 (this golden path's choice) |
Halves per-node IP consumption (/25 instead of /24). Not a GKE default (Standard defaults to 110, Autopilot to 32); raise for high pod-density at the cost of more CIDR space |
subnetwork |
auto-created |
Customer brings existing subnets |
| Release channel + maintenance windows |
REGULAR channel with a recurring maintenance window |
Add targeted maintenance exclusions (keep under ~6 months) only for critical freezes — see the gke-upgrades skill |
nodeConfig.bootDisk.diskType |
pd-balanced |
pd-ssd for I/O-intensive, pd-standard for cost |
Note: Autopilot selects node machine types automatically (e.g.,
ek-standard-8 may appear in describe output); the machine type is not
customer-configurable in Autopilot. Steer workload placement via
ComputeClasses instead.
Guardrails
- Do not request or output secrets (tokens, keys, service account JSON).
- Resolve project/cluster context from the conversation, MCP tools, or
gcloud config get-value project; ask the user only if it cannot be
resolved.
- For Day-0 decisions, always ask clarifying questions before proceeding.
- For Day-1 features, propose golden path defaults with trade-offs and let the
customer confirm.
- Do not promise zero downtime — see Upgrade Disruption below for what to
advise instead.
- When auditing existing clusters, compare against golden path and report
deviations with severity and remediation.
Upgrade Disruption
Never promise zero downtime for node upgrades, on any configuration. Node
upgrades cordon and drain nodes, which evicts Pods. Draining honors
PodDisruptionBudgets and terminationGracePeriodSeconds for up to one hour,
after which GKE forcefully evicts the remaining Pods so the upgrade can proceed.
A PDB narrows the window; it cannot veto the upgrade. Say so plainly rather than
implying the disruption can be eliminated.
What to recommend, all four — not a subset:
PodDisruptionBudgets with minAvailable set so eviction cannot take the
last healthy replica. A PDB that can never be satisfied stalls the drain for
an hour and then loses anyway.
At least 2 replicas, spread across zones with topology spread
constraints. A single-replica Deployment has downtime by definition.
Readiness probes that reflect real serving health, so traffic drains
before the Pod dies.
Surge upgrade settings on the node pool. Surge is the default strategy;
the default is maxSurge=1, maxUnavailable=0 — one extra node is created
and made ready before an old one is drained.
| Setting |
Controls |
Default |
maxSurge |
Additional nodes added per zone during the upgrade |
1 |
maxUnavailable |
Nodes simultaneously unavailable per zone |
0 |
Nodes upgraded at once is the sum of the two, capped at 20 (Autopilot)
and 100 (Standard). Multi-zone node pools upgrade one zone at a time. Raising
maxUnavailable trades availability for speed; raising maxSurge trades
cost for availability.
Caveat: externalTrafficPolicy: Local does not work with parallel node
drains, so it constrains aggressive surge configurations.
For rollback procedures and maintenance windows, see the gke-upgrades skill.
Golden Path Config
See golden-path-autopilot.yaml for the
full cluster-level policy settings.
1---2name: gke-golden-path-23description: >-4 Provides GKE golden path configuration defaults, production readiness5 checklists, and cluster default patterns. Use when designing GKE clusters,6 verifying GKE production readiness, or checking configurations against7 GKE defaults. Don't use for setting up workload autoscaling specifically (use8 gke-workload-scaling instead).9metadata:10 source: https://skillmd.com/skills/google/gke-golden-path-211 version: "1.1.1"12 category: Containers13---1415# GKE Golden Path Configuration1617The golden path is the recommended Autopilot configuration for production18clusters. It defines sensible defaults — when the user requests different19settings, apply them and note relevant trade-offs. For setting up autoscaling20specifically, use `gke-cluster-autoscaler` for node autoscaling or21`gke-workload-scaling` for workload autoscaling (HPA/VPA).2223> **MCP Tools:** `get_cluster`, `create_cluster`, `update_cluster`2425## Attribution2627**MUST READ**: Before executing any `gcloud`, REST API, or MCP command, you28MUST read [Usage Attribution](references/attribution.md). For commands executed29directly in the terminal (never in generated scripts, Terraform, or user code),30prefix `gcloud` with31`CLOUDSDK_METRICS_REQUEST_ATTRIBUTION="agent-skills/1.0 (skill:gke-golden-path)"`32and append `agent-skills/1.0 (skill:gke-golden-path)` to `User-Agent` headers.3334## Rules35361. **Default to the golden path.** Use golden path values unless the user37 requests otherwise. When deviating, note trade-offs but respect the user's38 choice.392. **Day-0 vs Day-1.** Flag Day-0 decisions (networking, private nodes,40 subnets, IP allocation) prominently — they are hard/impossible to change41 after creation.423. **Tool preference: MCP > gcloud > kubectl.** MCP is preferred as it directly43 interfaces with GKE APIs with structured data, reducing shell syntax errors44 and parsing ambiguities. See the `gke-basics` skill's CLI reference for full45 coverage matrix and override options. If the user46 says "use gcloud" or "use kubectl", respect that for the session.474. **Document decisions and rationale**, especially for Day-0 choices and48 golden path deviations.4950## Required Inputs5152If the user is unsure, use golden path defaults.5354- **Project ID** (required)55- **Region** (required, e.g., `us-central1`)56- **Cluster name** (required)57- **Environment type**: dev/test or production (defaults to production)58- **Networking**: bring-your-own VPC/subnet or auto-create (default:59 auto-create)60- **Scale expectations**: expected node/pod count, workload types61- **Cost constraints**: Spot VM tolerance, budget considerations6263## Always-Apply Defaults6465Recommended best practices applied by default. If the user requests a different66setting, apply it and briefly note the security or operational trade-off.6768Setting | Golden Path Value69------------------------------------------------------------------ | -----------------70`autopilot.enabled` | `true`71`privateClusterConfig.enablePrivateNodes` | `true`72`masterAuthorizedNetworksConfig.privateEndpointEnforcementEnabled` | `true`73`secretManagerConfig.enabled` + `rotationInterval: 120s` | `true`74`rbacBindingConfig.enableInsecureBinding*` | `false` (both)75`workloadIdentityConfig.workloadPool` | enabled76`networkConfig.datapathProvider` | `ADVANCED_DATAPATH`77`networkConfig.dnsConfig.clusterDns` | `CLOUD_DNS`78`autoscaling.autoscalingProfile` | `OPTIMIZE_UTILIZATION`79`verticalPodAutoscaling.enabled` | `true`80`monitoringConfig` components | SYSTEM_COMPONENTS, STORAGE, POD, DEPLOYMENT, STATEFULSET, DAEMONSET, HPA, JOBSET, CADVISOR, KUBELET, DCGM, APISERVER, SCHEDULER, CONTROLLER_MANAGER81`loggingConfig` components | SYSTEM_COMPONENTS, WORKLOADS (enabled by default)82`advancedDatapathObservabilityConfig.enableMetrics` | `true`83`nodeConfig.shieldedInstanceConfig.enableSecureBoot` | `true`84`nodeConfig.workloadMetadataConfig.mode` | `GKE_METADATA`85`nodeConfig.gcfsConfig.enabled` / `gvnic.enabled` | `true` / `true`86`addonsConfig.statefulHaConfig.enabled` | `true`87Storage CSI drivers (Filestore, GCS FUSE, Parallelstore) | enabled88Pod Security Standards | `restricted` on production namespaces8990## Customer-Configurable Settings9192These have golden path defaults but customers may deviate with valid93justification. **Ask before changing.**9495Setting | Default | Why Deviate96---------------------------------------- | ----------------------------------- | -----------97`dnsEndpointConfig.allowExternalTraffic` | `true` | Restrict if cluster only accessed from within VPC98`autoIpamConfig` / `createSubnetwork` | `true` / `true` | Customer has pre-existing VPC/subnets99`maxPodsPerNode` | `48` (this golden path's choice) | Halves per-node IP consumption (/25 instead of /24). Not a GKE default (Standard defaults to `110`, Autopilot to `32`); raise for high pod-density at the cost of more CIDR space100`subnetwork` | auto-created | Customer brings existing subnets101Release channel + maintenance windows | `REGULAR` channel with a recurring maintenance window | Add targeted maintenance exclusions (keep under ~6 months) only for critical freezes — see the `gke-upgrades` skill102`nodeConfig.bootDisk.diskType` | `pd-balanced` | `pd-ssd` for I/O-intensive, `pd-standard` for cost103104> **Note**: Autopilot selects node machine types automatically (e.g.,105> `ek-standard-8` may appear in describe output); the machine type is not106> customer-configurable in Autopilot. Steer workload placement via107> ComputeClasses instead.108109## Guardrails110111- Do not request or output secrets (tokens, keys, service account JSON).112- Resolve project/cluster context from the conversation, MCP tools, or113 `gcloud config get-value project`; ask the user only if it cannot be114 resolved.115- For Day-0 decisions, always ask clarifying questions before proceeding.116- For Day-1 features, propose golden path defaults with trade-offs and let the117 customer confirm.118- Do not promise zero downtime — see Upgrade Disruption below for what to119 advise instead.120- When auditing existing clusters, compare against golden path and report121 deviations with severity and remediation.122123## Upgrade Disruption124125**Never promise zero downtime for node upgrades, on any configuration.** Node126upgrades cordon and drain nodes, which evicts Pods. Draining honors127PodDisruptionBudgets and `terminationGracePeriodSeconds` for **up to one hour**,128after which GKE forcefully evicts the remaining Pods so the upgrade can proceed.129A PDB narrows the window; it cannot veto the upgrade. Say so plainly rather than130implying the disruption can be eliminated.131132What to recommend, all four — not a subset:133134- **PodDisruptionBudgets** with `minAvailable` set so eviction cannot take the135 last healthy replica. A PDB that can never be satisfied stalls the drain for136 an hour and then loses anyway.137- **At least 2 replicas**, spread across zones with topology spread138 constraints. A single-replica Deployment has downtime by definition.139- **Readiness probes** that reflect real serving health, so traffic drains140 before the Pod dies.141- **Surge upgrade settings** on the node pool. Surge is the default strategy;142 the default is `maxSurge=1`, `maxUnavailable=0` — one extra node is created143 and made ready before an old one is drained.144145 Setting | Controls | Default146 ---------------- | --------------------------------------------------- | -------147 `maxSurge` | Additional nodes added per zone during the upgrade | `1`148 `maxUnavailable` | Nodes simultaneously unavailable per zone | `0`149150 Nodes upgraded at once is the **sum** of the two, capped at 20 (Autopilot)151 and 100 (Standard). Multi-zone node pools upgrade one zone at a time. Raising152 `maxUnavailable` trades availability for speed; raising `maxSurge` trades153 cost for availability.154155> **Caveat**: `externalTrafficPolicy: Local` does not work with parallel node156> drains, so it constrains aggressive surge configurations.157158For rollback procedures and maintenance windows, see the `gke-upgrades` skill.159160## Golden Path Config161162See [golden-path-autopilot.yaml](./assets/golden-path-autopilot.yaml) for the163full cluster-level policy settings.