Defining cluster service versions (CSVs) - Developing Operators | Operators

A cluster service version (CSV), defined by a ClusterServiceVersion object, is a YAML manifest created from Operator metadata that assists Operator Lifecycle Manager (OLM) in running the Operator in a cluster. It is the metadata that accompanies an Operator container image, used to populate user interfaces with information such as its logo, description, and version. It is also a source of technical information that is required to run the Operator, like the RBAC rules it requires and which custom resources (CRs) it manages or depends on.

The Operator SDK includes the CSV generator to generate a CSV for the current Operator project, customized using information contained in YAML manifests and Operator source files.

A CSV-generating command removes the responsibility of Operator authors having in-depth OLM knowledge in order for their Operator to interact with OLM or publish metadata to the Catalog Registry. Further, because the CSV spec will likely change over time as new Kubernetes and OLM features are implemented, the Operator SDK is equipped to easily extend its update system to handle new CSV features going forward.

How CSV generation works

Operator bundle manifests, which include cluster service versions (CSVs), describe how to display, create, and manage an application with Operator Lifecycle Manager (OLM). The CSV generator in the Operator SDK, called by the generate bundle subcommand, is the first step towards publishing your Operator to a catalog and deploying it with OLM. The subcommand requires certain input manifests to construct a CSV manifest; all inputs are read when the command is invoked, along with a CSV base, to idempotently generate or regenerate a CSV.

Typically, the generate kustomize manifests subcommand would be run first to generate the input Kustomize bases that are consumed by the generate bundle subcommand. However, the Operator SDK provides the make bundle command, which automates several tasks, including running the following subcommands in order:

generate kustomize manifests
generate bundle
bundle validate

Additional resources

See Bundling an Operator for a full procedure that includes generating a bundle and CSV.

Generated files and resources

The make bundle command creates the following files and directories in your Operator project:

A bundle manifests directory named bundle/manifests that contains a ClusterServiceVersion (CSV) object
A bundle metadata directory named bundle/metadata
All custom resource definitions (CRDs) in a config/crd directory
A Dockerfile bundle.Dockerfile

The following resources are typically included in a CSV:

Role: Defines Operator permissions within a namespace.
ClusterRole: Defines cluster-wide Operator permissions.
Deployment: Defines how an Operand of an Operator is run in pods.
CustomResourceDefinition (CRD): Defines custom resources that your Operator reconciles.
Custom resource examples: Examples of resources adhering to the spec of a particular CRD.

Version management

The --version flag for the generate bundle subcommand supplies a semantic version for your bundle when creating one for the first time and when upgrading an existing one.

By setting the VERSION variable in your Makefile, the --version flag is automatically invoked using that value when the generate bundle subcommand is run by the make bundle command. The CSV version is the same as the Operator version, and a new CSV is generated when upgrading Operator versions.

Manually-defined CSV fields

Many CSV fields cannot be populated using generated, generic manifests that are not specific to Operator SDK. These fields are mostly human-written metadata about the Operator and various custom resource definitions (CRDs).

Operator authors must directly modify their cluster service version (CSV) YAML file, adding personalized data to the following required fields. The Operator SDK gives a warning during CSV generation when a lack of data in any of the required fields is detected.

The following tables detail which manually-defined CSV fields are required and which are optional.

Table 1. Required
Field	Description
`metadata.name`	A unique name for this CSV. Operator version should be included in the name to ensure uniqueness, for example `app-operator.v0.1.1`.
`metadata.capabilities`	The capability level according to the Operator maturity model. Options include `Basic Install`, `Seamless Upgrades`, `Full Lifecycle`, `Deep Insights`, and `Auto Pilot`.
`spec.displayName`	A public name to identify the Operator.
`spec.description`	A short description of the functionality of the Operator.
`spec.keywords`	Keywords describing the Operator.
`spec.maintainers`	Human or organizational entities maintaining the Operator, with a `name` and `email`.
`spec.provider`	The provider of the Operator (usually an organization), with a `name`.
`spec.labels`	Key-value pairs to be used by Operator internals.
`spec.version`	Semantic version of the Operator, for example `0.1.1`.
`spec.customresourcedefinitions`	Any CRDs the Operator uses. This field is populated automatically by the Operator SDK if any CRD YAML files are present in `deploy/`. However, several fields not in the CRD manifest spec require user input: `description`: description of the CRD. `resources`: any Kubernetes resources leveraged by the CRD, for example `Pod` and `StatefulSet` objects. `specDescriptors`: UI hints for inputs and outputs of the Operator.

Table 2. Optional
Field	Description
`spec.replaces`	The name of the CSV being replaced by this CSV.
`spec.links`	URLs (for example, websites and documentation) pertaining to the Operator or application being managed, each with a `name` and `url`.
`spec.selector`	Selectors by which the Operator can pair resources in a cluster.
`spec.icon`	A base64-encoded icon unique to the Operator, set in a `base64data` field with a `mediatype`.
`spec.maturity`	The level of maturity the software has achieved at this version. Options include `planning`, `pre-alpha`, `alpha`, `beta`, `stable`, `mature`, `inactive`, and `deprecated`.

Further details on what data each field above should hold are found in the CSV spec.

Several YAML fields currently requiring user intervention can potentially be parsed from Operator code.

Additional resources

Operator maturity model

Operator metadata annotations

Operator developers can manually define certain annotations in the metadata of a cluster service version (CSV) to enable features or highlight capabilities in user interfaces (UIs), such as OperatorHub.

The following table lists Operator metadata annotations that can be manually defined using metadata.annotations fields.

Field Description

alm-examples

Provide custom resource definition (CRD) templates with a minimum set of configuration. Compatible UIs pre-fill this template for users to further customize.

operatorframework.io/initialization-resource

Specify a single required custom resource by adding operatorframework.io/initialization-resource annotation to the cluster service version (CSV) during Operator installation. The user is then prompted to create the custom resource through a template provided in the CSV. Must include a template that contains a complete YAML definition.

operatorframework.io/suggested-namespace

Set a suggested namespace where the Operator should be deployed.

operators.openshift.io/infrastructure-features

Infrastructure features supported by the Operator. Users can view and filter by these features when discovering Operators through OperatorHub in the web console. Valid, case-sensitive values:

disconnected: Operator supports being mirrored into disconnected catalogs, including all dependencies, and does not require internet access. All related images required for mirroring are listed by the Operator.
cnf: Operator provides a Cloud-native Network Functions (CNF) Kubernetes plugin.
cni: Operator provides a Container Network Interface (CNI) Kubernetes plugin.
csi: Operator provides a Container Storage Interface (CSI) Kubernetes plugin.
fips: Operator accepts the FIPS mode of the underlying platform and works on nodes that are booted into FIPS mode.

The use of FIPS Validated / Modules in Process cryptographic libraries is only supported on OKD deployments on the x86_64 architecture.

proxy-aware: Operator supports running on a cluster behind a proxy. Operator accepts the standard proxy environment variables HTTP_PROXY and HTTPS_PROXY, which Operator Lifecycle Manager (OLM) provides to the Operator automatically when the cluster is configured to use a proxy. Required environment variables are passed down to Operands for managed workloads.

operators.openshift.io/valid-subscription

Free-form array for listing any specific subscriptions that are required to use the Operator. For example, '["3Scale Commercial License", "Red Hat Managed Integration"]'.

operators.operatorframework.io/internal-objects

Hides CRDs in the UI that are not meant for user manipulation.

Example use cases

Operator supports disconnected and proxy-aware

operators.openshift.io/infrastructure-features: '["disconnected", "proxy-aware"]'

Operator requires an OKD license

operators.openshift.io/valid-subscription: '["OpenShift Container Platform"]'

Operator requires a 3scale license

operators.openshift.io/valid-subscription: '["3Scale Commercial License", "Red Hat Managed Integration"]'

Operator supports disconnected and proxy-aware, and requires an OKD license

operators.openshift.io/infrastructure-features: '["disconnected", "proxy-aware"]'
operators.openshift.io/valid-subscription: '["OpenShift Container Platform"]'

Additional resources

Enabling your Operator for restricted network environments

As an Operator author, your Operator must meet additional requirements to run properly in a restricted network, or disconnected, environment.

Operator requirements for supporting disconnected mode

In the cluster service version (CSV) of your Operator:
- List any related images, or other container images that your Operator might require to perform their functions.
- Reference all specified images by a digest (SHA) and not by a tag.
All dependencies of your Operator must also support running in a disconnected mode.
Your Operator must not require any off-cluster resources.

For the CSV requirements, you can make the following changes as the Operator author.

Prerequisites

An Operator project with a CSV.

Procedure

Use SHA references to related images in two places in the CSV for your Operator:

Update spec.relatedImages:

...
spec:
  relatedImages: (1)
    - name: etcd-operator (2)
      image: quay.io/etcd-operator/operator@sha256:d134a9865524c29fcf75bbc4469013bc38d8a15cb5f41acfddb6b9e492f556e4 (3)
    - name: etcd-image
      image: quay.io/etcd-operator/etcd@sha256:13348c15263bd8838ec1d5fc4550ede9860fcbb0f843e48cbccec07810eebb68
...

1	Create a `relatedImages` section and set the list of related images.
2	Specify a unique identifier for the image.
3	Specify each image by a digest (SHA), not by an image tag.

Update the env section in the deployment when declaring environment variables that inject the image that the Operator should use:

spec:
  install:
    spec:
      deployments:
      - name: etcd-operator-v3.1.1
        spec:
          replicas: 1
          selector:
            matchLabels:
              name: etcd-operator
          strategy:
            type: Recreate
          template:
            metadata:
              labels:
                name: etcd-operator
            spec:
              containers:
              - args:
                - /opt/etcd/bin/etcd_operator_run.sh
                env:
                - name: WATCH_NAMESPACE
                  valueFrom:
                    fieldRef:
                      fieldPath: metadata.annotations['olm.targetNamespaces']
                - name: ETCD_OPERATOR_DEFAULT_ETCD_IMAGE (1)
                  value: quay.io/etcd-operator/etcd@sha256:13348c15263bd8838ec1d5fc4550ede9860fcbb0f843e48cbccec07810eebb68 (2)
                - name: ETCD_LOG_LEVEL
                  value: INFO
                image: quay.io/etcd-operator/operator@sha256:d134a9865524c29fcf75bbc4469013bc38d8a15cb5f41acfddb6b9e492f556e4 (3)
                imagePullPolicy: IfNotPresent
                livenessProbe:
                  httpGet:
                    path: /healthy
                    port: 8080
                  initialDelaySeconds: 10
                  periodSeconds: 30
                name: etcd-operator
                readinessProbe:
                  httpGet:
                    path: /ready
                    port: 8080
                  initialDelaySeconds: 10
                  periodSeconds: 30
                resources: {}
              serviceAccountName: etcd-operator
    strategy: deployment

1	Inject the images referenced by the Operator by using environment variables.
2	Specify each image by a digest (SHA), not by an image tag.
3	Also reference the Operator container image by a digest (SHA), not by an image tag.

When configuring probes, the timeoutSeconds value must be lower than the periodSeconds value. The timeoutSeconds default value is 1. The periodSeconds default value is 10.

Add the disconnected annotation, which indicates that the Operator works in a disconnected environment:
```
metadata:
  annotations:
    operators.openshift.io/infrastructure-features: '["disconnected"]'
```
Operators can be filtered in OperatorHub by this infrastructure feature.

Enabling your Operator for multiple architectures and operating systems

Operator Lifecycle Manager (OLM) assumes that all Operators run on Linux hosts. However, as an Operator author, you can specify whether your Operator supports managing workloads on other architectures, if worker nodes are available in the OKD cluster.

If your Operator supports variants other than AMD64 and Linux, you can add labels to the cluster service version (CSV) that provides the Operator to list the supported variants. Labels indicating supported architectures and operating systems are defined by the following:

labels:
    operatorframework.io/arch.<arch>: supported (1)
    operatorframework.io/os.<os>: supported (2)

1	Set `<arch>` to a supported string.
2	Set `<os>` to a supported string.

Only the labels on the channel head of the default channel are considered for filtering package manifests by label. This means, for example, that providing an additional architecture for an Operator in the non-default channel is possible, but that architecture is not available for filtering in the PackageManifest API.

If a CSV does not include an os label, it is treated as if it has the following Linux support label by default:

labels:
    operatorframework.io/os.linux: supported

If a CSV does not include an arch label, it is treated as if it has the following AMD64 support label by default:

labels:
    operatorframework.io/arch.amd64: supported

If an Operator supports multiple node architectures or operating systems, you can add multiple labels, as well.

Prerequisites

An Operator project with a CSV.
To support listing multiple architectures and operating systems, your Operator image referenced in the CSV must be a manifest list image.
For the Operator to work properly in restricted network, or disconnected, environments, the image referenced must also be specified using a digest (SHA) and not by a tag.

Procedure

Add a label in the metadata.labels of your CSV for each supported architecture and operating system that your Operator supports:
```
labels:
  operatorframework.io/arch.s390x: supported
  operatorframework.io/os.zos: supported
  operatorframework.io/os.linux: supported (1)
  operatorframework.io/arch.amd64: supported (1)
```
1 After you add a new architecture or operating system, you must also now include the default os.linux and arch.amd64 variants explicitly.

Additional resources

See the Image Manifest V 2, Schema 2 specification for more information on manifest lists.

Architecture and operating system support for Operators

The following strings are supported in Operator Lifecycle Manager (OLM) on OKD when labeling or filtering Operators that support multiple architectures and operating systems:

Table 4. Architectures supported on OKD
Architecture	String
AMD64	`amd64`
64-bit PowerPC little-endian	`ppc64le`
IBM Z	`s390x`

Table 5. Operating systems supported on OKD
Operating system	String
Linux	`linux`
z/OS	`zos`

Different versions of OKD and other Kubernetes-based distributions might support a different set of architectures and operating systems.

Setting a suggested namespace

Some Operators must be deployed in a specific namespace, or with ancillary resources in specific namespaces, to work properly. If resolved from a subscription, Operator Lifecycle Manager (OLM) defaults the namespaced resources of an Operator to the namespace of its subscription.

As an Operator author, you can instead express a desired target namespace as part of your cluster service version (CSV) to maintain control over the final namespaces of the resources installed for their Operators. When adding the Operator to a cluster using OperatorHub, this enables the web console to autopopulate the suggested namespace for the cluster administrator during the installation process.

Procedure

In your CSV, set the operatorframework.io/suggested-namespace annotation to your suggested namespace:
```
metadata:
  annotations:
    operatorframework.io/suggested-namespace: <namespace> (1)
```
1 Set your suggested namespace.

Enabling Operator conditions

Operator Lifecycle Manager (OLM) provides Operators with a channel to communicate complex states that influence OLM behavior while managing the Operator. By default, OLM creates an OperatorCondition custom resource definition (CRD) when it installs an Operator. Based on the conditions set in the OperatorCondition custom resource (CR), the behavior of OLM changes accordingly.

To support Operator conditions, an Operator must be able to read the OperatorCondition CR created by OLM and have the ability to complete the following tasks:

Get the specific condition.
Set the status of a specific condition.

This can be accomplished by using the operator-lib library. An Operator author can provide a controller-runtime client in their Operator for the library to access the OperatorCondition CR owned by the Operator in the cluster.

The library provides a generic Conditions interface, which has the following methods to Get and Set a conditionType in the OperatorCondition CR:

Get: To get the specific condition, the library uses the client.Get function from controller-runtime, which requires an ObjectKey of type types.NamespacedName present in conditionAccessor.
Set: To update the status of the specific condition, the library uses the client.Update function from controller-runtime. An error occurs if the conditionType is not present in the CRD.

The Operator is allowed to modify only the status subresource of the CR. Operators can either delete or update the status.conditions array to include the condition. For more details on the format and description of the fields present in the conditions, see the upstream Condition GoDocs.

Operator SDK v1.8.0 supports operator-lib v0.3.0.

Prerequisites

An Operator project generated using the Operator SDK.

Procedure

To enable Operator conditions in your Operator project:

In the go.mod file of your Operator project, add operator-framework/operator-lib as a required library:

module github.com/example-inc/memcached-operator

go 1.15

require (
  k8s.io/apimachinery v0.19.2
  k8s.io/client-go v0.19.2
  sigs.k8s.io/controller-runtime v0.7.0
  operator-framework/operator-lib v0.3.0
)

Write your own constructor in your Operator logic that will result in the following outcomes:
- Accepts a controller-runtime client.
- Accepts a conditionType.
- Returns a Condition interface to update or add conditions.
Because OLM currently supports the Upgradeable condition, you can create an interface that has methods to access the Upgradeable condition. For example:
```
import (
  ...
  apiv1 "github.com/operator-framework/api/pkg/operators/v1"
)

func NewUpgradeable(cl client.Client) (Condition, error) {
  return NewCondition(cl, "apiv1.OperatorUpgradeable")
}

cond, err := NewUpgradeable(cl);
```
In this example, the NewUpgradeable constructor is further used to create a variable cond of type Condition. The cond variable would in turn have Get and Set methods, which can be used for handling the OLM Upgradeable condition.

Additional resources

Operator conditions

Defining webhooks

Webhooks allow Operator authors to intercept, modify, and accept or reject resources before they are saved to the object store and handled by the Operator controller. Operator Lifecycle Manager (OLM) can manage the lifecycle of these webhooks when they are shipped alongside your Operator.

The cluster service version (CSV) resource of an Operator can include a webhookdefinitions section to define the following types of webhooks:

Admission webhooks (validating and mutating)
Conversion webhooks

Procedure

Add a webhookdefinitions section to the spec section of the CSV of your Operator and include any webhook definitions using a type of ValidatingAdmissionWebhook, MutatingAdmissionWebhook, or ConversionWebhook. The following example contains all three types of webhooks:

CSV containing webhooks

  apiVersion: operators.coreos.com/v1alpha1
  kind: ClusterServiceVersion
  metadata:
    name: webhook-operator.v0.0.1
  spec:
    customresourcedefinitions:
      owned:
      - kind: WebhookTest
        name: webhooktests.webhook.operators.coreos.io (1)
        version: v1
    install:
      spec:
        deployments:
        - name: webhook-operator-webhook
          ...
          ...
          ...
      strategy: deployment
    installModes:
    - supported: false
      type: OwnNamespace
    - supported: false
      type: SingleNamespace
    - supported: false
      type: MultiNamespace
    - supported: true
      type: AllNamespaces
    webhookdefinitions:
    - type: ValidatingAdmissionWebhook (2)
      admissionReviewVersions:
      - v1beta1
      - v1
      containerPort: 443
      targetPort: 4343
      deploymentName: webhook-operator-webhook
      failurePolicy: Fail
      generateName: vwebhooktest.kb.io
      rules:
      - apiGroups:
        - webhook.operators.coreos.io
        apiVersions:
        - v1
        operations:
        - CREATE
        - UPDATE
        resources:
        - webhooktests
      sideEffects: None
      webhookPath: /validate-webhook-operators-coreos-io-v1-webhooktest
    - type: MutatingAdmissionWebhook (3)
      admissionReviewVersions:
      - v1beta1
      - v1
      containerPort: 443
      targetPort: 4343
      deploymentName: webhook-operator-webhook
      failurePolicy: Fail
      generateName: mwebhooktest.kb.io
      rules:
      - apiGroups:
        - webhook.operators.coreos.io
        apiVersions:
        - v1
        operations:
        - CREATE
        - UPDATE
        resources:
        - webhooktests
      sideEffects: None
      webhookPath: /mutate-webhook-operators-coreos-io-v1-webhooktest
    - type: ConversionWebhook (4)
      admissionReviewVersions:
      - v1beta1
      - v1
      containerPort: 443
      targetPort: 4343
      deploymentName: webhook-operator-webhook
      generateName: cwebhooktest.kb.io
      sideEffects: None
      webhookPath: /convert
      conversionCRDs:
      - webhooktests.webhook.operators.coreos.io (5)
...

1	The CRDs targeted by the conversion webhook must exist here.
2	A validating admission webhook.
3	A mutating admission webhook.
4	A conversion webhook.
5	The `spec.PreserveUnknownFields` property of each CRD must be set to `false` or `nil`.

Additional resources

Types of webhook admission plugins
Kubernetes documentation:

Webhook considerations for OLM

When deploying an Operator with webhooks using Operator Lifecycle Manager (OLM), you must define the following:

The type field must be set to either ValidatingAdmissionWebhook, MutatingAdmissionWebhook, or ConversionWebhook, or the CSV will be placed in a failed phase.
The CSV must contain a deployment whose name is equivalent to the value supplied in the deploymentName field of the webhookdefinition.

When the webhook is created, OLM ensures that the webhook only acts upon namespaces that match the Operator group that the Operator is deployed in.

Certificate authority constraints

OLM is configured to provide each deployment with a single certificate authority (CA). The logic that generates and mounts the CA into the deployment was originally used by the API service lifecycle logic. As a result:

The TLS certificate file is mounted to the deployment at /apiserver.local.config/certificates/apiserver.crt.
The TLS key file is mounted to the deployment at /apiserver.local.config/certificates/apiserver.key.

Admission webhook rules constraints

To prevent an Operator from configuring the cluster into an unrecoverable state, OLM places the CSV in the failed phase if the rules defined in an admission webhook intercept any of the following requests:

Requests that target all groups
Requests that target the operators.coreos.com group
Requests that target the ValidatingWebhookConfigurations or MutatingWebhookConfigurations resources

Conversion webhook constraints

OLM places the CSV in the failed phase if a conversion webhook definition does not adhere to the following constraints:

CSVs featuring a conversion webhook can only support the AllNamespaces install mode.
The CRD targeted by the conversion webhook must have its spec.preserveUnknownFields field set to false or nil.
The conversion webhook defined in the CSV must target an owned CRD.
There can only be one conversion webhook on the entire cluster for a given CRD.

Understanding your custom resource definitions (CRDs)

There are two types of custom resource definitions (CRDs) that your Operator can use: ones that are owned by it and ones that it depends on, which are required.

Owned CRDs

The custom resource definitions (CRDs) owned by your Operator are the most important part of your CSV. This establishes the link between your Operator and the required RBAC rules, dependency management, and other Kubernetes concepts.

It is common for your Operator to use multiple CRDs to link together concepts, such as top-level database configuration in one object and a representation of replica sets in another. Each one should be listed out in the CSV file.

Table 6. Owned CRD fields
Field	Description	Required/optional
`Name`	The full name of your CRD.	Required
`Version`	The version of that object API.	Required
`Kind`	The machine readable name of your CRD.	Required
`DisplayName`	A human readable version of your CRD name, for example `MongoDB Standalone`.	Required
`Description`	A short description of how this CRD is used by the Operator or a description of the functionality provided by the CRD.	Required
`Group`	The API group that this CRD belongs to, for example `database.example.com`.	Optional
`Resources`	Your CRDs own one or more types of Kubernetes objects. These are listed in the `resources` section to inform your users of the objects they might need to troubleshoot or how to connect to the application, such as the service or ingress rule that exposes a database. It is recommended to only list out the objects that are important to a human, not an exhaustive list of everything you orchestrate. For example, do not list config maps that store internal state that are not meant to be modified by a user.	Optional
`SpecDescriptors`, `StatusDescriptors`, and `ActionDescriptors`	These descriptors are a way to hint UIs with certain inputs or outputs of your Operator that are most important to an end user. If your CRD contains the name of a secret or config map that the user must provide, you can specify that here. These items are linked and highlighted in compatible UIs. There are three types of descriptors: `SpecDescriptors`: A reference to fields in the `spec` block of an object. `StatusDescriptors`: A reference to fields in the `status` block of an object. `ActionDescriptors`: A reference to actions that can be performed on an object. All descriptors accept the following fields: `DisplayName`: A human readable name for the `Spec`, `Status`, or `Action`. `Description`: A short description of the `Spec`, `Status`, or `Action` and how it is used by the Operator. `Path`: A dot-delimited path of the field on the object that this descriptor describes. `X-Descriptors`: Used to determine which "capabilities" this descriptor has and which UI component to use. See the openshift/console project for a canonical list of React UI X-Descriptors for OKD. Also see the openshift/console project for more information on Descriptors in general.	Optional

The following example depicts a MongoDB Standalone CRD that requires some user input in the form of a secret and config map, and orchestrates services, stateful sets, pods and config maps:

Example owned CRD

      - displayName: MongoDB Standalone
        group: mongodb.com
        kind: MongoDbStandalone
        name: mongodbstandalones.mongodb.com
        resources:
          - kind: Service
            name: ''
            version: v1
          - kind: StatefulSet
            name: ''
            version: v1beta2
          - kind: Pod
            name: ''
            version: v1
          - kind: configmap
            name: ''
            version: v1
        specDescriptors:
          - description: Credentials for Ops Manager or Cloud Manager.
            displayName: Credentials
            path: credentials
            x-descriptors:
              - 'urn:alm:descriptor:com.tectonic.ui:selector:core:v1:Secret'
          - description: Project this deployment belongs to.
            displayName: Project
            path: project
            x-descriptors:
              - 'urn:alm:descriptor:com.tectonic.ui:selector:core:v1:configmap'
          - description: MongoDB version to be installed.
            displayName: Version
            path: version
            x-descriptors:
              - 'urn:alm:descriptor:com.tectonic.ui:label'
        statusDescriptors:
          - description: The status of each of the pods for the MongoDB cluster.
            displayName: Pod Status
            path: pods
            x-descriptors:
              - 'urn:alm:descriptor:com.tectonic.ui:podStatuses'
        version: v1
        description: >-
          MongoDB Deployment consisting of only one host. No replication of
          data.

Required CRDs

Relying on other required CRDs is completely optional and only exists to reduce the scope of individual Operators and provide a way to compose multiple Operators together to solve an end-to-end use case.

An example of this is an Operator that might set up an application and install an etcd cluster (from an etcd Operator) to use for distributed locking and a Postgres database (from a Postgres Operator) for data storage.

Operator Lifecycle Manager (OLM) checks against the available CRDs and Operators in the cluster to fulfill these requirements. If suitable versions are found, the Operators are started within the desired namespace and a service account created for each Operator to create, watch, and modify the Kubernetes resources required.

Table 7. Required CRD fields
Field	Description	Required/optional
`Name`	The full name of the CRD you require.	Required
`Version`	The version of that object API.	Required
`Kind`	The Kubernetes object kind.	Required
`DisplayName`	A human readable version of the CRD.	Required
`Description`	A summary of how the component fits in your larger architecture.	Required

Example required CRD

    required:
    - name: etcdclusters.etcd.database.coreos.com
      version: v1beta2
      kind: EtcdCluster
      displayName: etcd Cluster
      description: Represents a cluster of etcd nodes.

CRD upgrades

OLM upgrades a custom resource definition (CRD) immediately if it is owned by a singular cluster service version (CSV). If a CRD is owned by multiple CSVs, then the CRD is upgraded when it has satisfied all of the following backward compatible conditions:

All existing serving versions in the current CRD are present in the new CRD.
All existing instances, or custom resources, that are associated with the serving versions of the CRD are valid when validated against the validation schema of the new CRD.

Adding a new CRD version

Procedure

To add a new version of a CRD to your Operator:

Add a new entry in the CRD resource under the versions section of your CSV.

For example, if the current CRD has a version v1alpha1 and you want to add a new version v1beta1 and mark it as the new storage version, add a new entry for v1beta1:
```
versions:
  - name: v1alpha1
    served: true
    storage: false
  - name: v1beta1 (1)
    served: true
    storage: true
```
1 New entry.

Ensure the referencing version of the CRD in the owned section of your CSV is updated if the CSV intends to use the new version:

customresourcedefinitions:
  owned:
  - name: cluster.example.com
    version: v1beta1 (1)
    kind: cluster
    displayName: Cluster

1	Update the `version`.

Push the updated CRD and CSV to your bundle.

Deprecating or removing a CRD version

Operator Lifecycle Manager (OLM) does not allow a serving version of a custom resource definition (CRD) to be removed right away. Instead, a deprecated version of the CRD must be first disabled by setting the served field in the CRD to false. Then, the non-serving version can be removed on the subsequent CRD upgrade.

Procedure

To deprecate and remove a specific version of a CRD:

Mark the deprecated version as non-serving to indicate this version is no longer in use and may be removed in a subsequent upgrade. For example:
```
versions:
  - name: v1alpha1
    served: false (1)
    storage: true
```
1 Set to false.

Switch the storage version to a serving version if the version to be deprecated is currently the storage version. For example:

versions:
  - name: v1alpha1
    served: false
    storage: false (1)
  - name: v1beta1
    served: true
    storage: true (1)

1	Update the `storage` fields accordingly.

To remove a specific version that is or was the storage version from a CRD, that version must be removed from the storedVersion in the status of the CRD. OLM will attempt to do this for you if it detects a stored version no longer exists in the new CRD.

Upgrade the CRD with the above changes.
In subsequent upgrade cycles, the non-serving version can be removed completely from the CRD. For example:
```
versions:
  - name: v1beta1
    served: true
    storage: true
```
Ensure the referencing CRD version in the owned section of your CSV is updated accordingly if that version is removed from the CRD.

CRD templates

Users of your Operator must be made aware of which options are required versus optional. You can provide templates for each of your custom resource definitions (CRDs) with a minimum set of configuration as an annotation named alm-examples. Compatible UIs will pre-fill this template for users to further customize.

The annotation consists of a list of the kind, for example, the CRD name and the corresponding metadata and spec of the Kubernetes object.

The following full example provides templates for EtcdCluster, EtcdBackup and EtcdRestore:

metadata:
  annotations:
    alm-examples: >-
      [{"apiVersion":"etcd.database.coreos.com/v1beta2","kind":"EtcdCluster","metadata":{"name":"example","namespace":"default"},"spec":{"size":3,"version":"3.2.13"}},{"apiVersion":"etcd.database.coreos.com/v1beta2","kind":"EtcdRestore","metadata":{"name":"example-etcd-cluster"},"spec":{"etcdCluster":{"name":"example-etcd-cluster"},"backupStorageType":"S3","s3":{"path":"<full-s3-path>","awsSecret":"<aws-secret>"}}},{"apiVersion":"etcd.database.coreos.com/v1beta2","kind":"EtcdBackup","metadata":{"name":"example-etcd-cluster-backup"},"spec":{"etcdEndpoints":["<etcd-cluster-endpoints>"],"storageType":"S3","s3":{"path":"<full-s3-path>","awsSecret":"<aws-secret>"}}}]

Hiding internal objects

It is common practice for Operators to use custom resource definitions (CRDs) internally to accomplish a task. These objects are not meant for users to manipulate and can be confusing to users of the Operator. For example, a database Operator might have a Replication CRD that is created whenever a user creates a Database object with replication: true.

As an Operator author, you can hide any CRDs in the user interface that are not meant for user manipulation by adding the operators.operatorframework.io/internal-objects annotation to the cluster service version (CSV) of your Operator.

Procedure

Before marking one of your CRDs as internal, ensure that any debugging information or configuration that might be required to manage the application is reflected on the status or spec block of your CR, if applicable to your Operator.

Add the operators.operatorframework.io/internal-objects annotation to the CSV of your Operator to specify any internal objects to hide in the user interface:

Internal object annotation

apiVersion: operators.coreos.com/v1alpha1
kind: ClusterServiceVersion
metadata:
  name: my-operator-v1.2.3
  annotations:
    operators.operatorframework.io/internal-objects: '["my.internal.crd1.io","my.internal.crd2.io"]' (1)
...

1	Set any internal CRDs as an array of strings.

Initializing required custom resources

An Operator might require the user to instantiate a custom resource before the Operator can be fully functional. However, it can be challenging for a user to determine what is required or how to define the resource.

As an Operator developer, you can specify a single required custom resource by adding operatorframework.io/initialization-resource to the cluster service version (CSV) during Operator installation. You are then prompted prompted to create the custom resource through a template that is provided in the CSV. The annotation must include a template that contains a complete YAML definition that is required to initialize the resource during installation.

If this annotation is defined, after installing the Operator from the OKD web console, the user is prompted to create the resource using the template provided in the CSV.

Procedure

Add the operatorframework.io/initialization-resource annotation to the CSV of your Operator to specify a required custom resource. For example, the following annotation requires the creation of a StorageCluster resource and provides a full YAML definition:

Initialization resource annotation

apiVersion: operators.coreos.com/v1alpha1
kind: ClusterServiceVersion
metadata:
  name: my-operator-v1.2.3
  annotations:
    operatorframework.io/initialization-resource: |-
        {
            "apiVersion": "ocs.openshift.io/v1",
            "kind": "StorageCluster",
            "metadata": {
                "name": "example-storagecluster"
            },
            "spec": {
                "manageNodes": false,
                "monPVCTemplate": {
                    "spec": {
                        "accessModes": [
                            "ReadWriteOnce"
                        ],
                        "resources": {
                            "requests": {
                                "storage": "10Gi"
                            }
                        },
                        "storageClassName": "gp2"
                    }
                },
                "storageDeviceSets": [
                    {
                        "count": 3,
                        "dataPVCTemplate": {
                            "spec": {
                                "accessModes": [
                                    "ReadWriteOnce"
                                ],
                                "resources": {
                                    "requests": {
                                        "storage": "1Ti"
                                    }
                                },
                                "storageClassName": "gp2",
                                "volumeMode": "Block"
                            }
                        },
                        "name": "example-deviceset",
                        "placement": {},
                        "portable": true,
                        "resources": {}
                    }
                ]
            }
        }
...

Understanding your API services

As with CRDs, there are two types of API services that your Operator may use: owned and required.

Owned API services

When a CSV owns an API service, it is responsible for describing the deployment of the extension api-server that backs it and the group/version/kind (GVK) it provides.

An API service is uniquely identified by the group/version it provides and can be listed multiple times to denote the different kinds it is expected to provide.

Table 8. Owned API service fields
Field	Description	Required/optional
`Group`	Group that the API service provides, for example `database.example.com`.	Required
`Version`	Version of the API service, for example `v1alpha1`.	Required
`Kind`	A kind that the API service is expected to provide.	Required
`Name`	The plural name for the API service provided.	Required
`DeploymentName`	Name of the deployment defined by your CSV that corresponds to your API service (required for owned API services). During the CSV pending phase, the OLM Operator searches the `InstallStrategy` of your CSV for a `Deployment` spec with a matching name, and if not found, does not transition the CSV to the "Install Ready" phase.	Required
`DisplayName`	A human readable version of your API service name, for example `MongoDB Standalone`.	Required
`Description`	A short description of how this API service is used by the Operator or a description of the functionality provided by the API service.	Required
`Resources`	Your API services own one or more types of Kubernetes objects. These are listed in the resources section to inform your users of the objects they might need to troubleshoot or how to connect to the application, such as the service or ingress rule that exposes a database. It is recommended to only list out the objects that are important to a human, not an exhaustive list of everything you orchestrate. For example, do not list config maps that store internal state that are not meant to be modified by a user.	Optional
`SpecDescriptors`, `StatusDescriptors`, and `ActionDescriptors`	Essentially the same as for owned CRDs.	Optional

API service resource creation

Operator Lifecycle Manager (OLM) is responsible for creating or replacing the service and API service resources for each unique owned API service:

Service pod selectors are copied from the CSV deployment matching the DeploymentName field of the API service description.
A new CA key/certificate pair is generated for each installation and the base64-encoded CA bundle is embedded in the respective API service resource.

API service serving certificates

OLM handles generating a serving key/certificate pair whenever an owned API service is being installed. The serving certificate has a common name (CN) containing the hostname of the generated Service resource and is signed by the private key of the CA bundle embedded in the corresponding API service resource.

The certificate is stored as a type kubernetes.io/tls secret in the deployment namespace, and a volume named apiservice-cert is automatically appended to the volumes section of the deployment in the CSV matching the DeploymentName field of the API service description.

If one does not already exist, a volume mount with a matching name is also appended to all containers of that deployment. This allows users to define a volume mount with the expected name to accommodate any custom path requirements. The path of the generated volume mount defaults to /apiserver.local.config/certificates and any existing volume mounts with the same path are replaced.

Required API services

OLM ensures all required CSVs have an API service that is available and all expected GVKs are discoverable before attempting installation. This allows a CSV to rely on specific kinds provided by API services it does not own.

Table 9. Required API service fields
Field	Description	Required/optional
`Group`	Group that the API service provides, for example `database.example.com`.	Required
`Version`	Version of the API service, for example `v1alpha1`.	Required
`Kind`	A kind that the API service is expected to provide.	Required
`DisplayName`	A human readable version of your API service name, for example `MongoDB Standalone`.	Required
`Description`	A short description of how this API service is used by the Operator or a description of the functionality provided by the API service.	Required