Object API

Responses from the Client API are usually objects from kr8s.objects which represent Kubernetes resources.

import kr8s

pods = kr8s.get("pods", namespace=kr8s.ALL)
pod = pods[0]
print(type(pod))
# <class 'kr8s.objects.Pod'>

import kr8s

pods = await kr8s.asyncio.get("pods", namespace=kr8s.ALL)
pod = pods[0]
print(type(pod))
# <class 'kr8s.asyncio.objects.Pod'>

In the above example the kr8s.get() function returns a list of Pod objects.

Attributes

These objects contain the raw response at .raw.

print(pod.raw)
# {'metadata': ..., 'spec': ..., 'status': ...}

There are also a selection of other properties including .name, .namespace, .metadata, .labels, .annotations and more to make it convenient to access sections of this raw data.

Note

Attributes work the same in in sync and async APIs.

print(pod.name)
# 'foo'

print(pod.namespace)
# 'default'

print(pod.metadata)
# {...}

print(pod.labels)
# {...}

print(pod.annotations)
# {...}

# See the API reference for a complete list

Methods

Objects also have helper methods like .patch(), .exists(), .refresh() and .delete() for interacting with Kubernetes resources.

# Patch the Pod
pod.patch({"metadata": {"labels": {"foo": "bar"}}})

# Check the Pod exists
pod.exists()
# True

# Update the object with the latest state from the API
pod.refresh()

# Delete the Pod
pod.delete()

# Patch the Pod
await pod.patch({"metadata": {"labels": {"foo": "bar"}}})

# Check the Pod exists
await pod.exists()
# True

# Update the object with the latest state from the API
await pod.refresh()

# Delete the Pod
await pod.delete()

Some objects also have additional methods that are unique to them. For example Pod has .logs(), .ready() and .exec().

# Get Pod logs
logs = [line for line in pod.logs()]

# Check if Pod containers are ready
pod.ready()
# True

# Exec a command in a Pod
pod.exec(["uptime"])
# CompletedExec(args=['uptime'], stdout=..., stderr=..., returncode=0)

# Get Pod logs
logs = [line async for line in pod.logs()]

# Check if Pod containers are ready
await pod.ready()
# True

# Exec a command in a Pod
await pod.exec(["uptime"])
# CompletedExec(args=['uptime'], stdout=..., stderr=..., returncode=0)

Client references

All objects returned by kr8s will have a reference to the API client that created it at Object.api.

You can also create objects yourself from a spec or get existing ones by name with .create().

Methods on objects that require communicating with Kubernetes will create an API client or retrieve one from the cache automatically.

# Create a new Pod
from kr8s.objects import Pod

pod = Pod({
        "apiVersion": "v1",
        "kind": "Pod",
        "metadata": {
            "name": "my-pod",
        },
        "spec": {
            "containers": [{"name": "pause", "image": "gcr.io/google_containers/pause",}]
        },
    })

pod.create()

# Create a new Pod
from kr8s.asyncio.objects import Pod

pod = await Pod({
        "apiVersion": "v1",
        "kind": "Pod",
        "metadata": {
            "name": "my-pod",
        },
        "spec": {
            "containers": [{"name": "pause", "image": "gcr.io/google_containers/pause",}]
        },
    })

await pod.create()

Note

With async objects we need to await them if we don’t pass them an existing client via Pod(..., api=...) as creating the API client may invoke some IO such as reading config from disk or requesting new tokens from a third-party auth provider.

Get a Pod reference by name with .get().

from kr8s.object import Pod

pod = Pod.get("my-pod")

from kr8s.asyncio.object import Pod

pod = await Pod.get("my-pod")

When creating new objects they will not have a client reference because they are created directly. In this case the object will call the kr8s.api factory function which will either create a new client if none exists or will grab the first client from the cache if one was created somewhere else in your code.

import kr8s
from kr8s.objects import Pod

api = kr8s.api(kubeconfig="/foo/bar")

pod = Pod({...})
# pod.api is api due to client caching

import kr8s
from kr8s.asyncio.objects import Pod

api = await kr8s.asyncio.api(kubeconfig="/foo/bar")

pod = await Pod({...})
# pod.api is api due to client caching

You can also explicitly pass an API client to the object when you create it.

import kr8s
from kr8s.objects import Pod

api = kr8s.api(kubeconfig="/foo/bar")

pod = Pod({...}, api=api)

import kr8s
from kr8s.asyncio.objects import Pod

api = await kr8s.api(kubeconfig="/foo/bar")

pod = Pod({...}, api=api)

Note

Note that you can optionally skip calling await when creating this Pod because we are passing an existing API client.

Creating new objects

We have provided common Kubernetes objects like Pod, Service, Deployment, etc in the kr8s.objects submodule but not all objects are represented. This helps to keep the library lightweight and also reduces maintenance overhead in the future as resources are added/removed. There is also no way we can represent everything as custom resources extend the Kubernetes API almost infinitely.

Instead we have focused on making the API extensible so that if there isn’t a built-in object for the resource you want to work with it is quick to add in your own code.

Extending the objects API

To create your own objects you can subclass kr8s.objects.APIObject, however we recommend you use the new_class class factory to ensure all of the required attributes are set. These will be used when constructing API calls by the API client.

from kr8s.objects import new_class

CustomObject = new_class(
        kind="CustomObject",
        version="example.org",
        namespaced=True,
        asyncio=False,
    )

The kr8s.objects.APIObject base class contains helper methods such as .create(), .delete(), .patch(), .exists(), etc.

There are also optional helpers that can be enabled for resources that support them. For example you can enable .scale() for resources which support updating the number of replicas.

from kr8s.objects import new_class

CustomScalableObject = new_class(
        kind="CustomObject",
        version="example.org",
        namespaced=True,
        scalable=True,
        scalable_spec="replicas",  # The spec key to patch when scaling
        asyncio=False,
    )

Using custom objects with other kr8s functions

When using the kr8s API some methods such as kr8s.get("pods") will want to return kr8s objects, in this case a Pod. The API client handles this by looking up all of the subclasses of APIObject and matching the kind against the kind returned by the API. If the API returns a kind of object that there is no kr8s object to deserialize into it will raise an exception.

When you create your own custom objects that subclass APIObject the client is then able to use those objects in its response.

import kr8s
from kr8s.objects import new_class

CustomObject = new_class(
        kind="CustomObject",
        version="example.org",
        namespaced=True,
        asyncio=False,
    )

cos = kr8s.get("customobjects")  # Will return a list of CustomObject instances

Note

If multiple subclasses of APIObject are created with the same API version and kind the first one registered will be used.

Interoperability with other libraries

If you are also using other Kubernetes client libraries including kubernetes, kubernetes-asyncio, pykube-ng or lightkube you can easily convert resource objects from those libraries to kr8s objects.

import pykube

api = pykube.HTTPClient(pykube.KubeConfig.from_file())
pykube_pod = pykube.Pod.objects(api).filter(namespace="gondor-system").get(name="my-pod")

Objects from other libraries can be cast directly to kr8s objects.

from kr8s.objects import Pod

kr8s_pod = Pod(pykube_pod)

from kr8s.asyncio.objects import Pod

kr8s_pod = await Pod(pykube_pod)

For some libraries including pykube-ng and lightkube we also have utility methods that support casting back again.

import pykube

api = pykube.HTTPClient(pykube.KubeConfig.from_file())
pykube_pod = kr8s_pod.to_pykube(api)  # Pykube requires you to provide every object with an instance of HTTPClient so we pass it here