woodpecker/docs/versioned_docs/version-3.0/30-administration/22-backends/40-kubernetes.md

---
toc_max_heading_level: 2
---

# Kubernetes backend

The Kubernetes backend executes steps inside standalone Pods. A temporary PVC is created for the lifetime of the pipeline to transfer files between steps.

## Images from private registries

In addition to [registries specified in the UI](../../20-usage/41-registries.md), you may provide [registry credentials in Kubernetes Secrets](https://kubernetes.io/docs/tasks/configure-pod-container/pull-image-private-registry/) to pull private container images defined in your pipeline YAML.

Place these Secrets in namespace defined by `WOODPECKER_BACKEND_K8S_NAMESPACE` and provide the Secret names to Agents via `WOODPECKER_BACKEND_K8S_PULL_SECRET_NAMES`.

## Step specific configuration

### Resources

The Kubernetes backend also allows for specifying requests and limits on a per-step basic, most commonly for CPU and memory.
We recommend to add a `resources` definition to all steps to ensure efficient scheduling.

Here is an example definition with an arbitrary `resources` definition below the `backend_options` section:

```yaml
steps:
  - name: 'My kubernetes step'
    image: alpine
    commands:
      - echo "Hello world"
    backend_options:
      kubernetes:
        resources:
          requests:
            memory: 200Mi
            cpu: 100m
          limits:
            memory: 400Mi
            cpu: 1000m
```

You can use [Limit Ranges](https://kubernetes.io/docs/concepts/policy/limit-range/) if you want to set the limits by per-namespace basis.

### Runtime class

`runtimeClassName` specifies the name of the RuntimeClass which will be used to run this Pod. If no `runtimeClassName` is specified, the default RuntimeHandler will be used.
See the [Kubernetes documentation](https://kubernetes.io/docs/concepts/containers/runtime-class/) for more information on specifying runtime classes.

### Service account

`serviceAccountName` specifies the name of the ServiceAccount which the Pod will mount. This service account must be created externally.
See the [Kubernetes documentation](https://kubernetes.io/docs/concepts/security/service-accounts/) for more information on using service accounts.

```yaml
steps:
  - name: 'My kubernetes step'
    image: alpine
    commands:
      - echo "Hello world"
    backend_options:
      kubernetes:
        # Use the service account `default` in the current namespace.
        # This usually the same as wherever woodpecker is deployed.
        serviceAccountName: default
```

To give steps access to the Kubernetes API via service account, take a look at [RBAC Authorization](https://kubernetes.io/docs/reference/access-authn-authz/rbac/)

### Node selector

`nodeSelector` specifies the labels which are used to select the node on which the step will be executed.

Labels defined here will be appended to a list which already contains `"kubernetes.io/arch"`.
By default `"kubernetes.io/arch"` is inferred from the agents' platform. One can override it by setting that label in the `nodeSelector` section of the `backend_options`.
Without a manual overwrite, builds will be randomly assigned to the runners and inherit their respective architectures.

To overwrite this, one needs to set the label in the `nodeSelector` section of the `backend_options`.
A practical example for this is when running a matrix-build and delegating specific elements of the matrix to run on a specific architecture.
In this case, one must define an arbitrary key in the matrix section of the respective matrix element:

```yaml
matrix:
  include:
    - NAME: runner1
      ARCH: arm64
```

And then overwrite the `nodeSelector` in the `backend_options` section of the step(s) using the name of the respective env var:

```yaml
[...]
    backend_options:
      kubernetes:
        nodeSelector:
          kubernetes.io/arch: "${ARCH}"
```

You can use [WOODPECKER_BACKEND_K8S_POD_NODE_SELECTOR](#woodpecker_backend_k8s_pod_node_selector) if you want to set the node selector per Agent
or [PodNodeSelector](https://kubernetes.io/docs/reference/access-authn-authz/admission-controllers/#podnodeselector) admission controller if you want to set the node selector by per-namespace basis.

### Tolerations

When you use `nodeSelector` and the node pool is configured with Taints, you need to specify the Tolerations. Tolerations allow the scheduler to schedule Pods with matching taints.
See the [Kubernetes documentation](https://kubernetes.io/docs/concepts/scheduling-eviction/taint-and-toleration/) for more information on using tolerations.

Example pipeline configuration:

```yaml
steps:
  - name: build
    image: golang
    commands:
      - go get
      - go build
      - go test
    backend_options:
      kubernetes:
        serviceAccountName: 'my-service-account'
        resources:
          requests:
            memory: 128Mi
            cpu: 1000m
          limits:
            memory: 256Mi
        nodeSelector:
          beta.kubernetes.io/instance-type: Standard_D2_v3
        tolerations:
          - key: 'key1'
            operator: 'Equal'
            value: 'value1'
            effect: 'NoSchedule'
            tolerationSeconds: 3600
```

### Volumes

To mount volumes a PersistentVolume (PV) and PersistentVolumeClaim (PVC) are needed on the cluster which can be referenced in steps via the `volumes` option.

Persistent volumes must be created manually. Use the Kubernetes [Persistent Volumes](https://kubernetes.io/docs/concepts/storage/persistent-volumes/) documentation as a reference.

_If your PVC is not highly available or NFS-based, you may also need to integrate affinity settings to ensure that your steps are executed on the correct node._

NOTE: If you plan to use this volume in more than one workflow concurrently, make sure you have configured the PVC in `RWX` mode. Keep in mind that this feature must be supported by the used CSI driver:

```yaml
accessModes:
  - ReadWriteMany
```

Assuming a PVC named `woodpecker-cache` exists, it can be referenced as follows in a plugin step:

```yaml
steps:
  - name: "Restore Cache"
    image: meltwater/drone-cache
    volumes:
      - woodpecker-cache:/woodpecker/src/cache
    settings:
      mount:
        - "woodpecker-cache"
    [...]
```

Or as follows when using a normal image:

```yaml
steps:
  - name: "Edit cache"
    image: alpine:latest
    volumes:
      - woodpecker-cache:/woodpecker/src/cache
    commands:
      - echo "Hello World" > /woodpecker/src/cache/output.txt
    [...]
```

### Security context

Use the following configuration to set the [Security Context](https://kubernetes.io/docs/tasks/configure-pod-container/security-context/) for the Pod/container running a given pipeline step:

```yaml
steps:
  - name: test
    image: alpine
    commands:
      - echo Hello world
    backend_options:
      kubernetes:
        securityContext:
          runAsUser: 999
          runAsGroup: 999
          privileged: true
    [...]
```

Note that the `backend_options.kubernetes.securityContext` object allows you to set both Pod and container level security context options in one object.
By default, the properties will be set at the Pod level. Properties that are only supported on the container level will be set there instead. So, the
configuration shown above will result in something like the following Pod spec:

<!-- cspell:disable -->

```yaml
kind: Pod
spec:
  securityContext:
    runAsUser: 999
    runAsGroup: 999
  containers:
    - name: wp-01hcd83q7be5ymh89k5accn3k6-0-step-0
      image: alpine
      securityContext:
        privileged: true
  [...]
```

<!-- cspell:enable -->

You can also restrict a syscalls of containers with [seccomp](https://kubernetes.io/docs/tutorials/security/seccomp/) profile.

```yaml
backend_options:
  kubernetes:
    securityContext:
      seccompProfile:
        type: Localhost
        localhostProfile: profiles/audit.json
```

or restrict a container's access to resources by specifying [AppArmor](https://kubernetes.io/docs/tutorials/security/apparmor/) profile

```yaml
backend_options:
  kubernetes:
    securityContext:
      apparmorProfile:
        type: Localhost
        localhostProfile: k8s-apparmor-example-deny-write
```

:::note
The feature requires Kubernetes v1.30 or above.
:::

### Annotations and labels

You can specify arbitrary [annotations](https://kubernetes.io/docs/concepts/overview/working-with-objects/annotations/) and [labels](https://kubernetes.io/docs/concepts/overview/working-with-objects/labels/) to be set on the Pod definition for a given workflow step using the following configuration:

```yaml
backend_options:
  kubernetes:
    annotations:
      workflow-group: alpha
      io.kubernetes.cri-o.Devices: /dev/fuse
    labels:
      environment: ci
      app.kubernetes.io/name: builder
```

In order to enable this configuration you need to set the appropriate environment variables to `true` on the woodpecker agent:
[WOODPECKER_BACKEND_K8S_POD_ANNOTATIONS_ALLOW_FROM_STEP](#woodpecker_backend_k8s_pod_annotations_allow_from_step) and/or [WOODPECKER_BACKEND_K8S_POD_LABELS_ALLOW_FROM_STEP](#woodpecker_backend_k8s_pod_labels_allow_from_step).

## Tips and tricks

### CRI-O

CRI-O users currently need to configure the workspace for all workflows in order for them to run correctly. Add the following at the beginning of your configuration:

```yaml
workspace:
  base: '/woodpecker'
  path: '/'
```

See [this issue](https://github.com/woodpecker-ci/woodpecker/issues/2510) for more details.

### `KUBERNETES_SERVICE_HOST` environment variable

Like the below env vars used for configuration, this can be set in the environment for configuration of the agent.
It configures the address of the Kubernetes API server to connect to.

If running the agent within Kubernetes, this will already be set and you don't have to add it manually.

## Configuration

These env vars can be set in the `env:` sections of the agent.

### `WOODPECKER_BACKEND_K8S_NAMESPACE`

> Default: `woodpecker`

The namespace to create worker Pods in.

### `WOODPECKER_BACKEND_K8S_VOLUME_SIZE`

> Default: `10G`

The volume size of the pipeline volume.

### `WOODPECKER_BACKEND_K8S_STORAGE_CLASS`

> Default: empty

The storage class to use for the pipeline volume.

### `WOODPECKER_BACKEND_K8S_STORAGE_RWX`

> Default: `true`

Determines if `RWX` should be used for the pipeline volume's [access mode](https://kubernetes.io/docs/concepts/storage/persistent-volumes/#access-modes). If false, `RWO` is used instead.

### `WOODPECKER_BACKEND_K8S_POD_LABELS`

> Default: empty

Additional labels to apply to worker Pods. Must be a YAML object, e.g. `{"example.com/test-label":"test-value"}`.

### `WOODPECKER_BACKEND_K8S_POD_LABELS_ALLOW_FROM_STEP`

> Default: `false`

Determines if additional Pod labels can be defined from a step's backend options.

### `WOODPECKER_BACKEND_K8S_POD_ANNOTATIONS`

> Default: empty

Additional annotations to apply to worker Pods. Must be a YAML object, e.g. `{"example.com/test-annotation":"test-value"}`.

### `WOODPECKER_BACKEND_K8S_POD_ANNOTATIONS_ALLOW_FROM_STEP`

> Default: `false`

Determines if Pod annotations can be defined from a step's backend options.

### `WOODPECKER_BACKEND_K8S_POD_NODE_SELECTOR`

> Default: empty

Additional node selector to apply to worker pods. Must be a YAML object, e.g. `{"topology.kubernetes.io/region":"eu-central-1"}`.

### `WOODPECKER_BACKEND_K8S_SECCTX_NONROOT` <!-- cspell:ignore SECCTX NONROOT -->

> Default: `false`

Determines if containers must be required to run as non-root users.

### `WOODPECKER_BACKEND_K8S_PULL_SECRET_NAMES`

> Default: empty

Secret names to pull images from private repositories. See, how to [Pull an Image from a Private Registry](https://kubernetes.io/docs/tasks/configure-pod-container/pull-image-private-registry/).