nodepool/doc/source/kubernetes.rst
James E. Blair c78fe769f2 Allow custom k8s pod specs
This change adds the ability to use the k8s (and friends) drivers
to create pods with custom specs.  This will allow nodepool admins
to define labels that create pods with options not otherwise supported
by Nodepool, as well as pods with multiple containers.

This can be used to implement the versatile sidecar pattern, which,
in a system where it is difficult to background a system process (such
as a database server or container runtime) is useful to run jobs with
such requirements.

It is still the case that a single resource is returned to Zuul, so
a single pod will be added to the inventory.  Therefore, the expectation
that it should be possible to shell into the first container in the
pod is documented.

Change-Id: I4a24a953a61239a8a52c9e7a2b68a7ec779f7a3d
2024-01-30 15:59:34 -08:00

18 KiB

zuul

Kubernetes Driver

Selecting the kubernetes driver adds the following options to the providers section of the configuration.

providers.[kubernetes]

A Kubernetes provider's resources are partitioned into groups called pools (see providers.[kubernetes].pools for details), and within a pool, the node types which are to be made available are listed (see providers.[kubernetes].pools.labels for details).

Note

For documentation purposes the option names are prefixed providers.[kubernetes] to disambiguate from other drivers, but [kubernetes] is not required in the configuration (e.g. below providers.[kubernetes].pools refers to the pools key in the providers section when the kubernetes driver is selected).

Example:

providers:
  - name: kubespray
    driver: kubernetes
    context: admin-cluster.local
    pools:
      - name: main
        labels:
          - name: kubernetes-namespace
            type: namespace
          - name: pod-fedora
            type: pod
            image: docker.io/fedora:28

context

Name of the context configured in kube/config.

Before using the driver, Nodepool either needs a kube/config file installed with a cluster admin context, in which case this setting is required, or if Nodepool is running inside Kubernetes, this setting and the kube/config file may be omitted and Nodepool will use a service account loaded from the in-cluster configuration path.

launch-retries

The number of times to retry launching a node before considering the job failed.

max-cores

Maximum number of cores usable from this provider's pools by default. This can be used to limit usage of the kubernetes backend. If not defined nodepool can use all cores up to the limit of the backend.

max-servers

Maximum number of pods spawnable from this provider's pools by default. This can be used to limit the number of pods. If not defined nodepool can create as many servers the kubernetes backend allows.

max-ram

Maximum ram usable from this provider's pools by default. This can be used to limit the amount of ram allocated by nodepool. If not defined nodepool can use as much ram as the kubernetes backend allows.

max-resources

A dictionary of other quota resource limits applicable to this provider's pools by default. Arbitrary limits may be supplied with the providers.[kubernetes].pools.labels.extra-resources attribute.

pools

A pool defines a group of resources from a Kubernetes provider.

name

Namespaces are prefixed with the pool's name.

priority

The priority of this provider pool (a lesser number is a higher priority). Nodepool launchers will yield requests to other provider pools with a higher priority as long as they are not paused. This means that in general, higher priority pools will reach quota first before lower priority pools begin to be used.

This setting may be specified at the provider level in order to apply to all pools within that provider, or it can be overridden here for a specific pool.

node-attributes

A dictionary of key-value pairs that will be stored with the node data in ZooKeeper. The keys and values can be any arbitrary string.

max-cores

Maximum number of cores usable from this pool. This can be used to limit usage of the kubernetes backend. If not defined nodepool can use all cores up to the limit of the backend.

max-servers

Maximum number of pods spawnable from this pool. This can be used to limit the number of pods. If not defined nodepool can create as many servers the kubernetes backend allows.

max-ram

Maximum ram usable from this pool. This can be used to limit the amount of ram allocated by nodepool. If not defined nodepool can use as much ram as the kubernetes backend allows.

max-resources

A dictionary of other quota resource limits applicable to this pool. Arbitrary limits may be supplied with the providers.[kubernetes].pools.labels.extra-resources attribute.

default-label-cpu

Only used by the providers.[kubernetes].pools.labels.type.pod label type; specifies specifies a default value for providers.[kubernetes].pools.labels.cpu for all labels of this pool that do not set their own value.

default-label-memory

Only used by the providers.[kubernetes].pools.labels.type.pod label type; specifies a default value in MiB for providers.[kubernetes].pools.labels.memory for all labels of this pool that do not set their own value.

default-label-storage

Only used by the providers.[kubernetes].pools.labels.type.pod label type; specifies a default value in MB for providers.[kubernetes].pools.labels.storage for all labels of this pool that do not set their own value.

default-label-extra-resources

Only used by the providers.[kubernetes].pools.labels.type.pod label type; specifies default values for providers.[kubernetes].pools.labels.extra-resources for all labels of this pool that do not set their own value.

default-label-cpu-limit

Only used by the providers.[kubernetes].pools.labels.type.pod label type; specifies specifies a default value for providers.[kubernetes].pools.labels.cpu-limit for all labels of this pool that do not set their own value.

default-label-memory-limit

Only used by the providers.[kubernetes].pools.labels.type.pod label type; specifies a default value in MiB for providers.[kubernetes].pools.labels.memory-limit for all labels of this pool that do not set their own value.

default-label-storage-limit

Only used by the providers.[kubernetes].pools.labels.type.pod label type; specifies a default value in MB for providers.[kubernetes].pools.labels.storage-limit for all labels of this pool that do not set their own value.

labels

Each entry in a pool`s labels section indicates that the corresponding label is available for use in this pool.

Each entry is a dictionary with the following keys

name

Identifier for this label; references an entry in the labels section.

type

The Kubernetes provider supports two types of labels:

namespace

Namespace labels provide an empty namespace configured with a service account that can create pods, services, configmaps, etc.

pod

Pod labels provide a dedicated namespace with a single pod created using the providers.[kubernetes].pools.labels.image parameter and it is configured with a service account that can exec and get the logs of the pod.

image

Only used by the providers.[kubernetes].pools.labels.type.pod label type; specifies the image name used by the pod.

image-pull

The ImagePullPolicy, can be IfNotPresent, Always or Never.

labels

A dictionary of additional values to be added to the namespace or pod metadata. The value of this field is added to the metadata.labels field in Kubernetes. Note that this field contains arbitrary key/value pairs and is unrelated to the concept of labels in Nodepool.

dynamic-labels

Similar to providers.[kubernetes].pools.labels.labels, but is interpreted as a format string with the following values available:

  • request: Information about the request which prompted the creation of this node (note that the node may ultimately be used for a different request and in that case this information will not be updated).
    • id: The request ID.
    • labels: The list of labels in the request.
    • requestor: The name of the requestor.
    • requestor_data: Key/value information from the requestor.
    • relative_priority: The relative priority of the request.
    • event_id: The external event ID of the request.
    • created_time: The creation time of the request.
    • tenant_name: The name of the tenant associated with the request.

For example:

labels:
  - name: pod-fedora
    dynamic-labels:
      request_info: "{request.id}"

annotations

A dictionary of additional values to be added to the pod metadata. The value of this field is added to the metadata.annotations field in Kubernetes. This field contains arbitrary key/value pairs that can be accessed by tools and libraries. E.g custom schedulers can make use of this metadata.

python-path

The path of the default python interpreter. Used by Zuul to set ansible_python_interpreter. The special value auto will direct Zuul to use inbuilt Ansible logic to select the interpreter on Ansible >=2.8, and default to /usr/bin/python2 for earlier versions.

shell-type

The shell type of the node's default shell executable. Used by Zuul to set ansible_shell_type. This setting should only be used

  • For a windows pod with the experimental connection-type ssh, in which case cmd or powershell should be set and reflect the node's DefaultShell configuration.
  • If the default shell is not Bourne compatible (sh), but instead e.g. csh or fish, and the user is aware that there is a long-standing issue with ansible_shell_type in combination with become

cpu

Only used by the providers.[kubernetes].pools.labels.type.pod label type; specifies the number of cpu to request for the pod. If no limit is specified, this will also be used as the limit.

memory

Only used by the providers.[kubernetes].pools.labels.type.pod label type; specifies the amount of memory in MiB to request for the pod. If no limit is specified, this will also be used as the limit.

storage

Only used by the providers.[kubernetes].pools.labels.type.pod label type; specifies the amount of ephemeral-storage in MB to request for the pod. If no limit is specified, this will also be used as the limit.

extra-resources

Only used by the providers.[kubernetes].pools.labels.type.pod label type; specifies any extra resources that Nodepool should consider in its quota calculation other than the resources described above (cpu, memory, storage).

cpu-limit

Only used by the providers.[kubernetes].pools.labels.type.pod label type; specifies the cpu limit for the pod.

memory-limit

Only used by the providers.[kubernetes].pools.labels.type.pod label type; specifies the memory limit in MiB for the pod.

storage-limit

Only used by the providers.[kubernetes].pools.labels.type.pod label type; specifies the ephemeral-storage limit in MB for the pod.

gpu

Only used by the providers.[kubernetes].pools.labels.type.pod label type; specifies the amount of gpu allocated to the pod. This will be used to set both requests and limits to the same value, based on how kubernetes assigns gpu resources: https://kubernetes.io/docs/tasks/manage-gpus/scheduling-gpus/.

gpu-resource

Only used by the providers.[kubernetes].pools.labels.type.pod label type; specifies the custom schedulable resource associated with the installed gpu that is available in the cluster.

env

Only used by the providers.[kubernetes].pools.labels.type.pod label type; A list of environment variables to pass to the Pod.

name

The name of the environment variable passed to the Pod.

value

The value of the environment variable passed to the Pod.

node-selector

Only used by the providers.[kubernetes].pools.labels.type.pod label type; A map of key-value pairs to ensure the Kubernetes scheduler places the Pod on a node with specific node labels.

scheduler-name

Only used by the providers.[kubernetes].pools.labels.type.pod label type. Sets the schedulerName field on the container. Normally left unset for the Kubernetes default.

privileged

Only used by the providers.[kubernetes].pools.labels.type.pod label type. Sets the securityContext.privileged flag on the container. Normally left unset for the Kubernetes default.

volumes

Only used by the providers.[kubernetes].pools.labels.type.pod label type. Sets the volumes field on the pod. If supplied, this should be a list of Kubernetes Pod Volume definitions.

volume-mounts

Only used by the providers.[kubernetes].pools.labels.type.pod label type. Sets the volumeMounts flag on the container. If supplied, this should be a list of Kubernetes Container VolumeMount definitions.

spec

This attribute is exclusive with all other label attributes except providers.[kubernetes].pools.labels.name, providers.[kubernetes].pools.labels.type, providers.[kubernetes].pools.labels.annotations, providers.[kubernetes].pools.labels.labels and providers.[kubernetes].pools.labels.dynamic-labels. If a spec is provided, then Nodepool will supply the contents of this value verbatim to Kubernetes as the spec attribute of the Kubernetes Pod definition. No other Nodepool attributes are used, including any default values set at the provider level (such as default-label-cpu and similar).

This attribute allows for the creation of arbitrary complex pod definitions but the user is responsible for ensuring that they are suitable. The first container in the pod is expected to be a long-running container that hosts a shell environment for running commands. The following minimal definition matches what Nodepool itself normally creates and is recommended as a starting point:

labels:
  - name: custom-pod
    type: pod
    spec:
      containers:
        - name: custom-pod
          image: ubuntu:jammy
          imagePullPolicy: IfNotPresent
          command: ["/bin/sh", "-c"]
          args: ["while true; do sleep 30; done;"]