Configuration - Kubernetes mode

The configuration file of a Ponos agent in Kubernetes mode is an extension of the generic Ponos agent. Only the kubernetes section is added.

`kubernetes.extra_environment`

An optional mapping of extra environment variables set on every new Kubernetes job created by the Ponos agent.

Defaults to an empty mapping.

`kubernetes.extra_volumes`

An optional list of persistent volume claims and their mount paths that will be set on every new Kubernetes job created by the Ponos agent.

Defaults to an empty list.

Each persistent volume claim used in this setting must be previously created and set in the same namespace as kubernetes.namespace described below.

Extra volume definition

An extra volume must have the following fields:

Key Type Required Description

Key	Type	Required	Description
`name`	String	Yes	Name used by the volume and mount points created by the Ponos agent. It’s only descriptive.
`pvc_name`	String	Yes	Name of the pre-existing persistent volume claim whose content will be mounted onto the new jobs.
`mount_path`	String	Yes	Path on the job’s filesystem where the persistent volume claim will be mounted.
`read_only`	Boolean	No	Control if the volume is read-only or read-write (defaults to `False`, so it’s read-write).

name

String

Yes

Name used by the volume and mount points created by the Ponos agent. It’s only descriptive.

pvc_name

String

Yes

Name of the pre-existing persistent volume claim whose content will be mounted onto the new jobs.

mount_path

String

Yes

Path on the job’s filesystem where the persistent volume claim will be mounted.

read_only

Boolean

Control if the volume is read-only or read-write (defaults to False, so it’s read-write).

`kubernetes.gpu_name`

An optional name of a Kubernetes GPU resource limit that will be requested on a Job created by this Ponos agent.

Defaults to nvidia.com/gpu which will allow any GPU available.

Only Nvidia GPUs are currently supported at this moment.

`kubernetes.in_cluster`

Boolean value that configures the connection mode towards Kubernetes cluster. By default, in_cluster is set to true.

When in_cluster is set to true, the Ponos agent will use internal access to reach the Kubernetes API.

Otherwise your local kube configuration is used to reach the cluster (this is mostly used for Ponos agent developers).

`kubernetes.namespace`

The Kubernetes namespace’s name where all jobs managed by the Ponos agent will be running.

It defaults to the default namespace which is always provisioned on Kubernetes clusters.

`kubernetes.pvc_name`

A persistent volume claim name must be configured before running the agent. It will be used to share and cache data between task executions.

The agent will not start if the configured persistent volume claim is not available on the cluster.

You can find more information about the Ponos agent setup on that page.

Example

Below is an example of a YAML configuration file to start a Ponos agent in Kubernetes mode.

# Generic Ponos agent
...

# Ponos agent in Kubernetes mode
kubernetes:
  in_cluster: true
  pvc_name: my-pvc-for-ponos
  namespace: my-ns

  extra_environment:
    CUSTOM_VAR: test1234

  extra_volumes:
    - name: my-app-data
      pvc_name: my-pvc
      mount_path: /app/data
      read_only: false

  gpu_name: nvidia.com/mig-4g.71gb