k8smith 0.4.0

A transparent, lightweight Kubernetes manifest generator

k8smith

A transparent, lightweight Kubernetes manifest generator.

Philosophy

• Transparency over abstraction — Generated YAML is predictable; no hidden defaults
• Composition over inheritance — Build complex resources from simple pieces
• Type safety at the boundary — Validate inputs with Pydantic, output plain dicts
• Cloud-agnostic core — Vanilla K8s resources work anywhere; cloud extensions are optional

Installation

pip install k8smith
# or
uv add k8smith

Quick Start

from k8smith import (
    DeploymentSpec,
    Container,
    PodSpec,
    PodTemplateSpec,
    ResourceRequirements,
    ResourceQuantity,
    build_deployment,
    Manifest,
)

spec = DeploymentSpec(
    name="web",
    namespace="production",
    replicas=3,
    template=PodTemplateSpec(
        spec=PodSpec(
            containers=[
                Container(
                    name="web",
                    image="nginx:1.25",
                    resources=ResourceRequirements(
                        requests=ResourceQuantity(cpu="100m", memory="128Mi"),
                        limits=ResourceQuantity(memory="256Mi"),
                    ),
                )
            ]
        )
    ),
)

deployment = build_deployment(spec)

# Collect resources into a manifest
manifest = Manifest()
manifest.add(deployment)
print(manifest.to_yaml())

GPU Support (no workarounds needed!)

from k8smith import (
    Container,
    ResourceRequirements,
    ResourceQuantity,
)

container = Container(
    name="ml-worker",
    image="ml-training:v1",
    resources=ResourceRequirements(
        requests=ResourceQuantity(memory="4Gi", cpu="1"),
        limits=ResourceQuantity(
            memory="8Gi",
            extended={"nvidia.com/gpu": "1"},  # Just works!
        ),
    ),
)

GKE Extensions

from k8smith.gke import (
    GatewaySpec,
    HTTPRouteSpec,
    build_gateway,
    build_httproute,
)

gateway = build_gateway(GatewaySpec(
    name="main-gateway",
    namespace="gateway",
    gateway_class_name="gke-l7-global-external-managed",
    listeners=[{
        "name": "https",
        "port": 443,
        "protocol": "HTTPS",
    }],
))

Core Resources

Resource	Builder
Deployment	build_deployment(DeploymentSpec)
Service	build_service(ServiceSpec)
Ingress	build_ingress(IngressSpec)
StatefulSet	build_statefulset(StatefulSetSpec)
DaemonSet	build_daemonset(DaemonSetSpec)
CronJob	build_cronjob(CronJobSpec)
ConfigMap	build_configmap(ConfigMapSpec)
Secret	build_secret(SecretSpec)
HPA	build_hpa(HPASpec)
PDB	build_pdb(PDBSpec)
ServiceAccount	build_serviceaccount(ServiceAccountSpec)
Namespace	build_namespace(NamespaceSpec)
Role	build_role(RoleSpec)
ClusterRole	build_clusterrole(ClusterRoleSpec)
RoleBinding	build_rolebinding(RoleBindingSpec)
ClusterRoleBinding	build_clusterrolebinding(ClusterRoleBindingSpec)

GKE Resources

Resource	Builder
Gateway	build_gateway(GatewaySpec)
HTTPRoute	build_httproute(HTTPRouteSpec)
HealthCheckPolicy	build_healthcheckpolicy(HealthCheckPolicySpec)
GCPBackendPolicy	build_gcp_backend_policy(GCPBackendPolicySpec)
PodMonitoring	build_pod_monitoring(PodMonitoringSpec)
ClusterPodMonitoring	build_cluster_pod_monitoring(ClusterPodMonitoringSpec)

Why not cdk8s?

k8smith was created to replace cdk8s after encountering these issues:

cdk8s Problem	k8smith Solution
Size.gibibytes(6) outputs 6144Mi	Strings stay as strings: "6Gi"
Can't add nvidia.com/gpu via resources	Built into ResourceQuantity.extended
Adds hostNetwork: false when not set	Only outputs what you specify
Requires label cleanup post-processing	No generated labels
JSII/Node.js dependency	Pure Python

License

GPL-3.0 - See LICENSE for details.