kuber Overview

The Python Kubernetes client exists to provide low-level access to the Kubernetes API. However, low-level access can be clunky to use and require an additional effort to achieve parity with common workflows provided by configuration-driven tooling.

kuber is a higher-level abstraction designed to be compliant with the general usage level of someone comfortable working with Kubernetes configuration files and managing them with tools like kubectl and/or helm.

Configuring Individual Resources

kuber allows Kubernetes resources to be defined entirely in Python code, or defined in configuration files and loaded and modified by code. Examples of the two approaches are shown below:

The Pure Python Approach

Here’s an example of how a Deployment can be created with kuber:

from kuber.latest import apps_v1

# Create a deployment using the most recent stable Kubernetes version
# from the apps/v1 API version.
d = apps_v1.Deployment()

with d.metadata as md:
    md.name = "my-deployment"
    md.namespace = "my-app"
    md.labels.update(app="foo", component="application")

d.spec.selector.match_labels.update(app="foo")
d.spec.template.metadata.labels.update(app="foo")

d.append_container(
    name="app",
    image="my-app:1.0",
    ports=[apps_v1.ContainerPort(container_port=8080, host_port=80)],
    tty=True,
    image_pull_policy="Always",
    resources=apps_v1.ResourceRequirements(
        limits={"cpu": "1.5", "memory": "1Gi"},
        requests={"cpu": "1.5", "memory": "800Mi"},
    )
)

# Render the results to YAML.
print(d.to_yaml())

The printed output of executing this would be:

apiVersion: apps/v1
kind: Deployment
metadata:
  labels:
    app: foo
    component: application
  name: my-deployment
  namespace: my-app
spec:
  template:
    spec:
      containers:
      - image: my-app:1.0
        imagePullPolicy: Always
        name: app
        ports:
        - containerPort: 8080
          hostPort: 80
        resources:
          limits:
            cpu: "1"
            memory: 1Gi
          requests:
            cpu: "1"
            memory: 800Mi
        tty: true

The Hybrid Approach

In many cases it is convenient to use standard Kubernetes configuration as a base template. The common approach in these cases used by projects like Helm is to introduce a templating language into the configuration files that gets rendered prior to using the configuration. However, a templated approach has a number of drawbacks - a primary one being that if the template doesn’t support a necessary piece custom configuration it means forking that template and managing yourself. Instead kuber facilitates flexible modification and augmentation of resource configurations that have been loaded from configuration files.

Following from the example above, let’s say we have a YAML resource configuration file my-deployment.yaml with part of the contents from the example above:

apiVersion: apps/v1
kind: Deployment
metadata:
  labels:
    component: application
  name: my-deployment
  namespace: my-app
spec:
  template:
    spec:
      containers:
      - image: my-app:1.0
        imagePullPolicy: Always
        name: app
        tty: true

We want to load that configuration file and modify the loaded definition to match the results from the Pure Python Approach example in the previous section. That would look like this:

import kuber
from kuber.latest import apps_v1

# Load YAML configuration file into a Deployment object.
d: apps_v1.Deployment = kuber.from_yaml_file("./my-deployment.yaml")

d.metadata.labels.update(app="foo")

with d.get_container("app") as c:
    c.resources.limits.update(cpu="1.5", memory="1Gi")
    c.resources.requests.update(cpu="1.5", memory="800Mi")
    c.ports.append(apps_v1.ContainerPort(container_port=8080, host_port=80))

# Render the results to YAML.
print(d.to_yaml())

The printed configuration matches the configuration printed in the previous example.

Managing Multiple Resources

Often times multiple resources are needed to support a single application within a Kubernetes cluster. This is where explicit configuration can get increasingly complex and has resulted in a number of tools, like Helm, that try to simplify the process. kuber supports high-level constructs as well that make it easier to manage multiple resources but without having to rely on templating.

import kuber
from kuber.latest import apps_v1
from kuber.latest import core_v1

# Load all YAML and/or JSON configuration files in the specified directory
# and return a kuber ResourceBundle object that contains those loaded
# resources.
bundle = kuber.from_directory("../my-application")

# Add environment label to all loaded resources.
for r in bundle.resources:
    r.metadata.labels.update(environment="production")

# Change the number of replicas in the deployment named "my-app" that has
# the label `component=web`.
d: apps_v1.Deployment = bundle.get(
    name="my-app",
    kind="Deployment",
    component="web"
)
d.spec.replicas = 20

# Change the service port to 443 for the service named "my-app" that has the
# label `component=web`.
s: core_v1.Service = bundle.get(
    name="my-app",
    kind="Service",
    component="web"
)
s.spec.ports = [core_v1.ServicePort(port=443, target_port=8080)]

# Render to consolidated YAML configuration file
print(bundle.render_yaml_bundle())

The flexibility of this approach comes in part from the ability to define a working base configuration in standard configuration files, but then load and modify that configuration before deployment.