Operator SDK Version upgrade guide

A guide to upgrading the Operator SDK version for an existing operator project from v0.2.x all the way through to 0.17.x.

In most cases the upgrading the SDK version should only entail updating the operator’s SDK dependency version in the Gopkg.toml or go.mod file. For some versions it might also be necessary to update the upstream Kubernetes and controller-runtime dependencies.

The full list of dependencies and their versions required by a particular version of the SDK can be viewed by using the operator-sdk print-deps command. Use this command to update the project Gopkg.toml/go.mod file accordingly as you upgrade to a new SDK version.

For some SDK versions after v0.1.0 there might be minor breaking changes in the controller-runtime APIs, operator-sdk CLI or the expected project layout and file names. These breaking changes will usually be outlined in the CHANGELOG/release-notes for each release version.

For releases that have a large number of breaking changes or involve a significant refactoring of the APIs and project layout there will be a migration guide similar to the v0.1.0 migration guide.

The following sections outline the upgrade steps for each SDK version along with steps necessary for any associated breaking changes.

v0.2.x

  • Update the SDK constraint in Gopkg.toml to version v0.2.1 and run dep ensure to update the vendor directory.

    1. [[constraint]]
    2. name = "github.com/operator-framework/operator-sdk"
    3. version = "=v0.2.1"

v0.3.x

  • Update the SDK constraint in Gopkg.toml to version v0.3.0, the kubernetes dependencies to kubernetes-1.12.3 revisions, and the controller-runtime version to v0.1.8. Then run dep ensure to update the vendor directory.

    1. [[override]]
    2. name = "k8s.io/code-generator"
    3. **revision for tag "kubernetes-1.12.3"**
    4. revision = "3dcf91f64f638563e5106f21f50c31fa361c918d"
    5. [[override]]
    6. name = "k8s.io/api"
    7. **revision for tag "kubernetes-1.12.3"**
    8. revision = "b503174bad5991eb66f18247f52e41c3258f6348"
    9. [[override]]
    10. name = "k8s.io/apiextensions-apiserver"
    11. **revision for tag "kubernetes-1.12.3"**
    12. revision = "0cd23ebeb6882bd1cdc2cb15fc7b2d72e8a86a5b"
    13. [[override]]
    14. name = "k8s.io/apimachinery"
    15. **revision for tag "kubernetes-1.12.3"**
    16. revision = "eddba98df674a16931d2d4ba75edc3a389bf633a"
    17. [[override]]
    18. name = "k8s.io/client-go"
    19. **revision for tag "kubernetes-1.12.3"**
    20. revision = "d082d5923d3cc0bfbb066ee5fbdea3d0ca79acf8"
    21. [[override]]
    22. name = "sigs.k8s.io/controller-runtime"
    23. version = "=v0.1.8"
    24. [[constraint]]
    25. name = "github.com/operator-framework/operator-sdk"
    26. version = "=v0.3.0"

v0.4.x

  • Update the SDK constraint in Gopkg.toml to version v0.4.1 and run dep ensure to update the vendor directory.

    1. [[constraint]]
    2. name = "github.com/operator-framework/operator-sdk"
    3. version = "=v0.4.1"

v0.5.x

  • Update the SDK constraint in Gopkg.toml to version v0.5.0, the kubernetes dependencies to kubernetes-1.13.1 revisions, and the controller-runtime version to v0.1.10.

    1. [[override]]
    2. name = "k8s.io/code-generator"
    3. **revision for tag "kubernetes-1.13.1"**
    4. revision = "c2090bec4d9b1fb25de3812f868accc2bc9ecbae"
    5. [[override]]
    6. name = "k8s.io/api"
    7. **revision for tag "kubernetes-1.13.1"**
    8. revision = "05914d821849570fba9eacfb29466f2d8d3cd229"
    9. [[override]]
    10. name = "k8s.io/apiextensions-apiserver"
    11. **revision for tag "kubernetes-1.13.1"**
    12. revision = "0fe22c71c47604641d9aa352c785b7912c200562"
    13. [[override]]
    14. name = "k8s.io/apimachinery"
    15. **revision for tag "kubernetes-1.13.1"**
    16. revision = "2b1284ed4c93a43499e781493253e2ac5959c4fd"
    17. [[override]]
    18. name = "k8s.io/client-go"
    19. **revision for tag "kubernetes-1.13.1"**
    20. revision = "8d9ed539ba3134352c586810e749e58df4e94e4f"
    21. [[override]]
    22. name = "sigs.k8s.io/controller-runtime"
    23. version = "=v0.1.10"
    24. [[constraint]]
    25. name = "github.com/operator-framework/operator-sdk"
    26. version = "=v0.5.0"
  • Append the following new constraints to your Gopkg.toml.

    1. [[override]]
    2. name = "k8s.io/kube-openapi"
    3. revision = "0cf8f7e6ed1d2e3d47d02e3b6e559369af24d803"
    4. [[override]]
    5. name = "github.com/go-openapi/spec"
    6. branch = "master"
    7. [[override]]
    8. name = "sigs.k8s.io/controller-tools"
    9. version = "=v0.1.8"
  • Update the required dependencies in Gopkg.toml to include sigs.k8s.io/controller-tools/pkg/crd/generator and change k8s.io/code-generator/cmd/openapi-gen to k8s.io/kube-openapi/cmd/openapi-gen. Then run dep ensure to update the vendor directory.

    1. required = [
    2. "k8s.io/code-generator/cmd/defaulter-gen",
    3. "k8s.io/code-generator/cmd/deepcopy-gen",
    4. "k8s.io/code-generator/cmd/conversion-gen",
    5. "k8s.io/code-generator/cmd/client-gen",
    6. "k8s.io/code-generator/cmd/lister-gen",
    7. "k8s.io/code-generator/cmd/informer-gen",
    8. "k8s.io/kube-openapi/cmd/openapi-gen",
    9. "k8s.io/gengo/args",
    10. "sigs.k8s.io/controller-tools/pkg/crd/generator",
    11. ]

v0.6.x

  • Update the SDK constraint in Gopkg.toml to version v0.6.0 and run dep ensure to update the vendor directory.

    1. [[constraint]]
    2. name = "github.com/operator-framework/operator-sdk"
    3. version = "=v0.6.0"
  • The operator-sdk olm-catalog command now expects and generates manifests in the operator-registry manifest format. Generate your CSVs in the new layout by using operator-sdk olm-catalog gen-csv command, or modify the layout of your existing CSV manifests directory similar to the example below.

    1. $ tree deploy/olm-catalog
    2. deploy/olm-catalog/
    3. └── memcached-operator
    4. ├── 0.1.0
    5. └── memcached-operator.v0.1.0.clusterserviceversion.yaml
    6. ├── 0.2.0
    7. └── memcached-operator.v0.2.0.clusterserviceversion.yaml
    8. └── memcached-operator.package.yaml

v0.7.x

  • Update the SDK constraint in Gopkg.toml to version v0.7.1 and run dep ensure to update the vendor directory.

    1. [[constraint]]
    2. name = "github.com/operator-framework/operator-sdk"
    3. version = "=v0.7.1"

v0.8.x

The SDK version v0.8.x supports scaffolding projects to use Go modules by default. It is recommended that you migrate your operator project to use modules for dependency management, however you can choose to keep using dep. The upgrade steps for both are outlined below:

dep

  • Update the SDK constraint in Gopkg.toml to version v0.8.2.

    1. [[constraint]]
    2. name = "github.com/operator-framework/operator-sdk"
    3. version = "=v0.8.2"
  • Pin the controller-tools dependency to the following revision. See the release notes or #1278 for why this is needed.

    1. [[override]]
    2. name = "sigs.k8s.io/controller-tools"
    3. revision = "9d55346c2bde73fb3326ac22eac2e5210a730207"
  • Run dep ensure to update the vendor directory.

modules

To get familiar with Go modules read the modules wiki. In particular the section on migrating to modules.

  • Ensure that you have Go 1.12+ and Mercurial 3.9+ installed.
  • Activate Go modules support for your project in $GOPATH/src by setting the env GO111MODULE=on. See activating modules for more details.
  • Initialize a new go.mod file by running go mod init.
  • Append the following to the end of your go.mod file to pin the operator-sdk and other upstream dependencies to the required versions.
  1. // Pinned to kubernetes-1.13.1
  2. replace (
  3. k8s.io/api => k8s.io/api v0.0.0-20181213150558-05914d821849
  4. k8s.io/apiextensions-apiserver => k8s.io/apiextensions-apiserver v0.0.0-20181213153335-0fe22c71c476
  5. k8s.io/apimachinery => k8s.io/apimachinery v0.0.0-20181127025237-2b1284ed4c93
  6. k8s.io/client-go => k8s.io/client-go v0.0.0-20181213151034-8d9ed539ba31
  7. )
  8. replace (
  9. github.com/coreos/prometheus-operator => github.com/coreos/prometheus-operator v0.29.0
  10. k8s.io/code-generator => k8s.io/code-generator v0.0.0-20181117043124-c2090bec4d9b
  11. k8s.io/kube-openapi => k8s.io/kube-openapi v0.0.0-20180711000925-0cf8f7e6ed1d
  12. sigs.k8s.io/controller-runtime => sigs.k8s.io/controller-runtime v0.1.10
  13. sigs.k8s.io/controller-tools => sigs.k8s.io/controller-tools v0.1.11-0.20190411181648-9d55346c2bde
  14. )
  15. replace github.com/operator-framework/operator-sdk => github.com/operator-framework/operator-sdk v0.8.2
  • Run go mod tidy to clean up the go.mod file.
    • In case of any go module loading errors, consult the default v0.8.2 go.mod dependencies scaffolded by the operator-sdk to resolve any differences. You can also view this file by scaffolding a new project with operator-sdk v0.8.2.
  • Ensure that you can build the project with operator-sdk build
  • Finally remove Gopkg.lock, Gopkg.toml and the vendor directory.

Breaking changes

Upon updating the project to v0.8.2 the following breaking changes apply:

  • On running the command operator-sdk generate openapi, the CRD manifests at deploy/crds/<group>_<version>_<kind>.crd for all API types will now be regenerated based on their source files pkg/apis/..._types.go. So if you have made any manual edits to the default generated CRD manifest, e.g manually written the validation block or specified the naming (spec.names), then that information be overwritten when the CRD is regenerated.

    The correct way to specify CRD fields like naming, validation, subresources etc is by using // +kubebuilder marker comments. Consult the legacy kubebuilder documentation to see what CRD fields can be generated via // +kubebuilder marker comments.

    Note: The version of controller-tools tied to this release does not support settting the spec.scope field of the CRD. Use the marker comment +genclient:nonNamespaced to set spec.scope=Cluster if necessary. See the example below:

    1. // MemcachedSpec defines the desired state of Memcached
    2. type MemcachedSpec struct {
    3. // +kubebuilder:validation:Maximum=5
    4. // +kubebuilder:validation:Minimum=1
    5. Size int32 `json:"size"`
    6. }
    7. // MemcachedStatus defines the observed state of Memcached
    8. type MemcachedStatus struct {
    9. // +kubebuilder:validation:MaxItems=5
    10. // +kubebuilder:validation:MinItems=1
    11. Nodes []string `json:"nodes"`
    12. }
    13. // +k8s:deepcopy-gen:interfaces=k8s.io/apimachinery/pkg/runtime.Object
    14. // Memcached is the Schema for the memcacheds API
    15. // +kubebuilder:subresource:status
    16. // +kubebuilder:resource:shortName="mc"
    17. // +genclient:nonNamespaced
    18. type Memcached struct {
    19. metav1.TypeMeta `json:",inline"`
    20. metav1.ObjectMeta `json:"metadata,omitempty"`
    21. Spec MemcachedSpec `json:"spec,omitempty"`
    22. Status MemcachedStatus `json:"status,omitempty"`
    23. }

v0.9.x

  • The function ExposeMetricsPort() has been replaced with CreateMetricsService() #1560.

    Replace the following line in cmd/manager/main.go

    1. _, err = metrics.ExposeMetricsPort(ctx, metricsPort)

    with

    1. servicePorts := []v1.ServicePort{
    2. {Port: metricsPort, Name: metrics.OperatorPortName, Protocol: v1.ProtocolTCP, TargetPort: intstr.IntOrString{Type: intstr.Int, IntVal: metricsPort}},
    3. {Port: operatorMetricsPort, Name: metrics.CRPortName, Protocol: v1.ProtocolTCP, TargetPort: intstr.IntOrString{Type: intstr.Int, IntVal: operatorMetricsPort}},
    4. }
    5. _, err = metrics.CreateMetricsService(ctx, servicePorts)

dep

  • Update the SDK constraint in Gopkg.toml to version v0.9.0, the kubernetes dependencies to kubernetes-1.13.4 revisions, and the controller-runtime version to v0.1.12.

    1. [[override]]
    2. name = "k8s.io/api"
    3. **revision for tag "kubernetes-1.13.4"**
    4. revision = "5cb15d34447165a97c76ed5a60e4e99c8a01ecfe"
    5. [[override]]
    6. name = "k8s.io/apiextensions-apiserver"
    7. **revision for tag "kubernetes-1.13.4"**
    8. revision = "d002e88f6236312f0289d9d1deab106751718ff0"
    9. [[override]]
    10. name = "k8s.io/apimachinery"
    11. **revision for tag "kubernetes-1.13.4"**
    12. revision = "86fb29eff6288413d76bd8506874fddd9fccdff0"
    13. [[override]]
    14. name = "k8s.io/client-go"
    15. **revision for tag "kubernetes-1.13.4"**
    16. revision = "b40b2a5939e43f7ffe0028ad67586b7ce50bb675"
    17. [[override]]
    18. name = "github.com/coreos/prometheus-operator"
    19. version = "=v0.29.0"
    20. [[override]]
    21. name = "sigs.k8s.io/controller-runtime"
    22. version = "=v0.1.12"
    23. [[constraint]]
    24. name = "github.com/operator-framework/operator-sdk"
    25. version = "=v0.9.0"
  • Append the contraint for k8s.io/kube-state-metrics.

    1. [[override]]
    2. name = "k8s.io/kube-state-metrics"
    3. version = "v1.6.0"
  • Run dep ensure to update the vendor directory.

modules

  • Update the replace directives in your go.mod file for the SDK, kubernetes, controller-runtime and kube-state metrics dependencies to the following versions.

    1. // Pinned to kubernetes-1.13.4
    2. replace (
    3. k8s.io/api => k8s.io/api v0.0.0-20190222213804-5cb15d344471
    4. k8s.io/apiextensions-apiserver => k8s.io/apiextensions-apiserver v0.0.0-20190228180357-d002e88f6236
    5. k8s.io/apimachinery => k8s.io/apimachinery v0.0.0-20190221213512-86fb29eff628
    6. k8s.io/client-go => k8s.io/client-go v0.0.0-20190228174230-b40b2a5939e4
    7. )
    8. replace (
    9. github.com/coreos/prometheus-operator => github.com/coreos/prometheus-operator v0.29.0
    10. sigs.k8s.io/controller-runtime => sigs.k8s.io/controller-runtime v0.1.12
    11. sigs.k8s.io/controller-tools => sigs.k8s.io/controller-tools v0.1.11-0.20190411181648-9d55346c2bde
    12. k8s.io/kube-state-metrics => k8s.io/kube-state-metrics v1.6.0
    13. )
    14. replace github.com/operator-framework/operator-sdk => github.com/operator-framework/operator-sdk v0.9.0

v0.10.x

  • The scorecard configuration format for the operator-sdk scorecard command has changed. See doc/test-framework/scorecard for more info.
  • The CSV config field role-path is now role-paths and takes a list of strings. Replace:

    1. role-path: path/to/role.yaml

    with:

    1. role-paths:
    2. - path/to/role.yaml

modules

  • Ensure the the following replace directives are present in your go.mod file:

    1. replace (
    2. github.com/coreos/prometheus-operator => github.com/coreos/prometheus-operator v0.29.0
    3. // Pinned to v2.9.2 (kubernetes-1.13.1) so https://proxy.golang.org can
    4. // resolve it correctly.
    5. github.com/prometheus/prometheus => github.com/prometheus/prometheus v0.0.0-20190424153033-d3245f150225
    6. k8s.io/kube-state-metrics => k8s.io/kube-state-metrics v1.6.0
    7. sigs.k8s.io/controller-runtime => sigs.k8s.io/controller-runtime v0.1.12
    8. sigs.k8s.io/controller-tools => sigs.k8s.io/controller-tools v0.1.11-0.20190411181648-9d55346c2bde
    9. )
    10. replace github.com/operator-framework/operator-sdk => github.com/operator-framework/operator-sdk v0.10.0

v0.11.x

NOTE: this version uses Kubernetes v1.14.x and controller-runtime v0.2.x, both of which have breaking API changes. See the changelog for more details.

dep

  • Remove the required = [ ... ] section and comment from the top of your Gopkg.toml file.
  • Update the following overrides in Gopkg.toml:

    1. [[override]]
    2. name = "k8s.io/api"
    3. **revision for tag "kubernetes-1.14.1"**
    4. revision = "6e4e0e4f393bf5e8bbff570acd13217aa5a770cd"
    5. [[override]]
    6. name = "k8s.io/apiextensions-apiserver"
    7. **revision for tag "kubernetes-1.14.1"**
    8. revision = "727a075fdec8319bf095330e344b3ccc668abc73"
    9. [[override]]
    10. name = "k8s.io/apimachinery"
    11. **revision for tag "kubernetes-1.14.1"**
    12. revision = "6a84e37a896db9780c75367af8d2ed2bb944022e"
    13. [[override]]
    14. name = "k8s.io/client-go"
    15. **revision for tag "kubernetes-1.14.1"**
    16. revision = "1a26190bd76a9017e289958b9fba936430aa3704"
    17. [[override]]
    18. name = "github.com/coreos/prometheus-operator"
    19. version = "=v0.31.1"
    20. [[override]]
    21. name = "sigs.k8s.io/controller-runtime"
    22. version = "=v0.2.2"
    23. [[constraint]]
    24. name = "github.com/operator-framework/operator-sdk"
    25. version = "=v0.11.0"
  • Append an override for gopkg.in/fsnotify.v1, which is required when resolving controller-runtime dependencies:

    1. [[override]]
    2. name = "gopkg.in/fsnotify.v1"
    3. source = "https://github.com/fsnotify/fsnotify.git"
  • Remove the k8s.io/kube-state-metrics override.

  • Run dep ensure to update the vendor directory.

modules

  • Ensure the the following replace directives are present in your go.mod file:

    1. // Pinned to kubernetes-1.14.1
    2. replace (
    3. k8s.io/api => k8s.io/api kubernetes-1.14.1
    4. k8s.io/apiextensions-apiserver => k8s.io/apiextensions-apiserver kubernetes-1.14.1
    5. k8s.io/apimachinery => k8s.io/apimachinery kubernetes-1.14.1
    6. k8s.io/client-go => k8s.io/client-go kubernetes-1.14.1
    7. k8s.io/cloud-provider => k8s.io/cloud-provider kubernetes-1.14.1
    8. )
    9. replace (
    10. // Indirect operator-sdk dependencies use git.apache.org, which is frequently
    11. // down. The github mirror should be used instead.
    12. // Locking to a specific version (from 'go mod graph'):
    13. git.apache.org/thrift.git => github.com/apache/thrift v0.0.0-20180902110319-2566ecd5d999
    14. github.com/coreos/prometheus-operator => github.com/coreos/prometheus-operator v0.31.1
    15. // Pinned to v2.10.0 (kubernetes-1.14.1) so https://proxy.golang.org can
    16. // resolve it correctly.
    17. github.com/prometheus/prometheus => github.com/prometheus/prometheus d20e84d0fb64aff2f62a977adc8cfb656da4e286
    18. )
    19. replace github.com/operator-framework/operator-sdk => github.com/operator-framework/operator-sdk v0.11.0

Import updates

  • Replace import sigs.k8s.io/controller-runtime/pkg/runtime/scheme with sigs.k8s.io/controller-runtime/pkg/scheme in:
    • ./pkg/apis/<group>/<version>/register.go
  • Replace import sigs.k8s.io/controller-runtime/pkg/runtime/log with sigs.k8s.io/controller-runtime/pkg/log in:
    • cmd/manager/main.go
    • ./pkg/controller/<kind>/<kind>_controller.go
  • Replace import sigs.k8s.io/controller-runtime/pkg/runtime/signals with sigs.k8s.io/controller-runtime/pkg/manager/signals in:
    • cmd/manager/main.go
  • Remove import sigs.k8s.io/controller-tools/pkg/crd/generator from:
    • tools.go

controller-runtime API updates

All method signatures for sigs.k8s.io/controller-runtime/pkg/client.Client and sigs.k8s.io/controller-runtime/pkg/client.StatusWriter (except for Client.Get()) have been updated. Each now uses a variadic option interface parameter typed for each method.

OpenAPI updates

  • Run the command operator-sdk generate openapi and ensure that no errors such as API rule violation are raised. For further information see the API rules documentation.

NOTE: You may need to add or remove markers (code annotations) to fix issues found when running generate openapi. Usage of markers in API code is discussed in the kubebuilder CRD generation documentation and in marker documentation. A full list of OpenAPIv3 validation markers can be found here.

TIPS:

  • If the +kubebuilder:validation:Pattern has commas, then surround the expressions in backticks.
  • If you are using +kubebuilder:validation:Enum then either surround the expression list in curly braces and quote each expression, or separate each expression using semicolons.

Operator SDK updates

  • pkg/test.FrameworkClient List() and Delete() method invocations should be updated to match those of Client.List() and Client.Delete(), described above.
  • The annotation to assign a scope to your CRD has changed. For the following changes, note that <resource> is the plural lower-case CRD Kind found at spec.names.plural.
    • For Namespaced-scoped operators, add a +kubebuilder:resource:path=<resource>,scope=Namespaced comment above your kind type in pkg/apis/<group>/<version>/<kind>_types.go.
    • For Cluster-scoped operators, replace the +genclient:nonNamespaced comment above your kind type in pkg/apis/<group>/<version>/<kind>_types.go with +kubebuilder:resource:path=<resource>,scope=Cluster.
  • CRD file names now have the form <full group>_<resource>_crd.yaml, and CR file names now have the form <full group>_<version>_<kind>_cr.yaml. <full group> is the full group name of your CRD found at spec.group, and <resource> is the plural lower-case CRD Kind found at spec.names.plural. To migrate:
    • Run operator-sdk generate openapi. CRD manifest files with new names containing versioned validation and subresource blocks will be generated.
    • Delete the old CRD manifest files.
    • Rename CR manifest file names from <group>_<version>_<kind>_cr.yaml to <full group>_<version>_<kind>_cr.yaml.

v0.12.x

Go version

  • Ensure that you are using a go version 1.13+

dep

Using dep is no longer supported. Follow Go’s official blog post about migrating to modules to learn how to migrate your project.

modules

  • Ensure the the following require modules and replace directives with the specific versions are present in your go.mod file:
  1. require (
  2. github.com/go-openapi/spec v0.19.0
  3. github.com/operator-framework/operator-sdk v0.12.1-0.20191112211508-82fc57de5e5b
  4. github.com/spf13/pflag v1.0.3
  5. k8s.io/api v0.0.0
  6. k8s.io/apimachinery v0.0.0
  7. k8s.io/client-go v11.0.0+incompatible
  8. k8s.io/kube-openapi v0.0.0-20190918143330-0270cf2f1c1d
  9. sigs.k8s.io/controller-runtime v0.3.0
  10. )
  11. // Pinned to kubernetes-1.15.4
  12. replace (
  13. k8s.io/api => k8s.io/api v0.0.0-20190918195907-bd6ac527cfd2
  14. k8s.io/apiextensions-apiserver => k8s.io/apiextensions-apiserver v0.0.0-20190918201827-3de75813f604
  15. k8s.io/apimachinery => k8s.io/apimachinery v0.0.0-20190817020851-f2f3a405f61d
  16. k8s.io/apiserver => k8s.io/apiserver v0.0.0-20190918200908-1e17798da8c1
  17. k8s.io/cli-runtime => k8s.io/cli-runtime v0.0.0-20190918202139-0b14c719ca62
  18. k8s.io/client-go => k8s.io/client-go v0.0.0-20190918200256-06eb1244587a
  19. k8s.io/cloud-provider => k8s.io/cloud-provider v0.0.0-20190918203125-ae665f80358a
  20. k8s.io/cluster-bootstrap => k8s.io/cluster-bootstrap v0.0.0-20190918202959-c340507a5d48
  21. k8s.io/code-generator => k8s.io/code-generator v0.0.0-20190612205613-18da4a14b22b
  22. k8s.io/component-base => k8s.io/component-base v0.0.0-20190918200425-ed2f0867c778
  23. k8s.io/cri-api => k8s.io/cri-api v0.0.0-20190817025403-3ae76f584e79
  24. k8s.io/csi-translation-lib => k8s.io/csi-translation-lib v0.0.0-20190918203248-97c07dcbb623
  25. k8s.io/kube-aggregator => k8s.io/kube-aggregator v0.0.0-20190918201136-c3a845f1fbb2
  26. k8s.io/kube-controller-manager => k8s.io/kube-controller-manager v0.0.0-20190918202837-c54ce30c680e
  27. k8s.io/kube-proxy => k8s.io/kube-proxy v0.0.0-20190918202429-08c8357f8e2d
  28. k8s.io/kube-scheduler => k8s.io/kube-scheduler v0.0.0-20190918202713-c34a54b3ec8e
  29. k8s.io/kubelet => k8s.io/kubelet v0.0.0-20190918202550-958285cf3eef
  30. k8s.io/legacy-cloud-providers => k8s.io/legacy-cloud-providers v0.0.0-20190918203421-225f0541b3ea
  31. k8s.io/metrics => k8s.io/metrics v0.0.0-20190918202012-3c1ca76f5bda
  32. k8s.io/sample-apiserver => k8s.io/sample-apiserver v0.0.0-20190918201353-5cc279503896
  33. )

NOTE: Check here an example of this upgrade to see the changes from the version 0.11.0 to 0.12.0.

  • Run go mod tidy to update the project modules
  • Run the command operator-sdk generate k8s to ensure that your resources will be updated
  • Run the command operator-sdk generate openapi and ensure that no errors such as API rule violation are raised. For further information see the API rules documentation.

(Optional) Update your operator to print its version

In v0.12.0, the SDK team updated the scaffold for cmd/manager/main.go to include the operator’s version in the output produced by the printVersion() function. See #1953

To add this feature to your operator, add the following lines in <project>/cmd/manager/main.go:

  1. import (
  2. ...
  3. "<your_module_path>/version"
  4. ...
  5. )
  6. func printVersion() {
  7. log.Info(fmt.Sprintf("Operator Version: %s", version.Version))
  8. ...
  9. }

v0.13.x

modules

  • Ensure the the following require modules and replace directives with the specific versions are present in your go.mod file:
  1. require (
  2. github.com/operator-framework/operator-sdk v0.13.0
  3. sigs.k8s.io/controller-runtime v0.4.0
  4. )
  5. // Pinned to kubernetes-1.16.2
  6. replace (
  7. k8s.io/api => k8s.io/api v0.0.0-20191016110408-35e52d86657a
  8. k8s.io/apiextensions-apiserver => k8s.io/apiextensions-apiserver v0.0.0-20191016113550-5357c4baaf65
  9. k8s.io/apimachinery => k8s.io/apimachinery v0.0.0-20191004115801-a2eda9f80ab8
  10. k8s.io/apiserver => k8s.io/apiserver v0.0.0-20191016112112-5190913f932d
  11. k8s.io/cli-runtime => k8s.io/cli-runtime v0.0.0-20191016114015-74ad18325ed5
  12. k8s.io/client-go => k8s.io/client-go v0.0.0-20191016111102-bec269661e48
  13. k8s.io/cloud-provider => k8s.io/cloud-provider v0.0.0-20191016115326-20453efc2458
  14. k8s.io/cluster-bootstrap => k8s.io/cluster-bootstrap v0.0.0-20191016115129-c07a134afb42
  15. k8s.io/code-generator => k8s.io/code-generator v0.0.0-20191004115455-8e001e5d1894
  16. k8s.io/component-base => k8s.io/component-base v0.0.0-20191016111319-039242c015a9
  17. k8s.io/cri-api => k8s.io/cri-api v0.0.0-20190828162817-608eb1dad4ac
  18. k8s.io/csi-translation-lib => k8s.io/csi-translation-lib v0.0.0-20191016115521-756ffa5af0bd
  19. k8s.io/kube-aggregator => k8s.io/kube-aggregator v0.0.0-20191016112429-9587704a8ad4
  20. k8s.io/kube-controller-manager => k8s.io/kube-controller-manager v0.0.0-20191016114939-2b2b218dc1df
  21. k8s.io/kube-proxy => k8s.io/kube-proxy v0.0.0-20191016114407-2e83b6f20229
  22. k8s.io/kube-scheduler => k8s.io/kube-scheduler v0.0.0-20191016114748-65049c67a58b
  23. k8s.io/kubectl => k8s.io/kubectl v0.0.0-20191016120415-2ed914427d51
  24. k8s.io/kubelet => k8s.io/kubelet v0.0.0-20191016114556-7841ed97f1b2
  25. k8s.io/legacy-cloud-providers => k8s.io/legacy-cloud-providers v0.0.0-20191016115753-cf0698c3a16b
  26. k8s.io/metrics => k8s.io/metrics v0.0.0-20191016113814-3b1a734dba6e
  27. k8s.io/sample-apiserver => k8s.io/sample-apiserver v0.0.0-20191016112829-06bb3c9d77c9
  28. )
  • Run go mod tidy to update the project modules
  • Run the command operator-sdk generate k8s to ensure that your resources will be updated
  • Run the command operator-sdk generate crds to regenerate CRDs

(Optional) Update the roles.yaml file

Replace * per verbs in order to solve the issue 671 and make clear the permissions used.

Example

  1. verbs:
  2. - create
  3. - delete
  4. - get
  5. - list
  6. - patch
  7. - update
  8. - watch

Notable changes

  • Deprecated: Deprecated the operator-sdk generate openapi command. CRD generation is still supported with operator-sdk generate crds. It is now recommended to use openapi-gen directly for OpenAPI code generation. The generate openapi subcommand will be removed in a future release.
  • Breaking change: An existing CSV’s spec.customresourcedefinitions.owned is now always overwritten except for each name, version, and kind on invoking olm-catalog gen-csv when Go API code annotations are present.
  • Potentially Breaking change: Be aware that there are potentially other breaking changes due to the controller-runtime and Kubernetes version be upgraded from v0.4.0 to `v1.16.2, respectively. There may be breaking changes to Go client code due to both of those changes.

For further detailed information see CHANGELOG

v0.14.x

modules

  • Ensure the the following require modules and replace directives with the specific versions are present in your go.mod file:
  1. require (
  2. github.com/operator-framework/operator-sdk v0.14.1
  3. sigs.k8s.io/controller-runtime v0.4.0
  4. )
  5. // Pinned to kubernetes-1.16.2
  6. replace (
  7. k8s.io/api => k8s.io/api v0.0.0-20191016110408-35e52d86657a
  8. k8s.io/apiextensions-apiserver => k8s.io/apiextensions-apiserver v0.0.0-20191016113550-5357c4baaf65
  9. k8s.io/apimachinery => k8s.io/apimachinery v0.0.0-20191004115801-a2eda9f80ab8
  10. k8s.io/apiserver => k8s.io/apiserver v0.0.0-20191016112112-5190913f932d
  11. k8s.io/cli-runtime => k8s.io/cli-runtime v0.0.0-20191016114015-74ad18325ed5
  12. k8s.io/client-go => k8s.io/client-go v0.0.0-20191016111102-bec269661e48
  13. k8s.io/cloud-provider => k8s.io/cloud-provider v0.0.0-20191016115326-20453efc2458
  14. k8s.io/cluster-bootstrap => k8s.io/cluster-bootstrap v0.0.0-20191016115129-c07a134afb42
  15. k8s.io/code-generator => k8s.io/code-generator v0.0.0-20191004115455-8e001e5d1894
  16. k8s.io/component-base => k8s.io/component-base v0.0.0-20191016111319-039242c015a9
  17. k8s.io/cri-api => k8s.io/cri-api v0.0.0-20190828162817-608eb1dad4ac
  18. k8s.io/csi-translation-lib => k8s.io/csi-translation-lib v0.0.0-20191016115521-756ffa5af0bd
  19. k8s.io/kube-aggregator => k8s.io/kube-aggregator v0.0.0-20191016112429-9587704a8ad4
  20. k8s.io/kube-controller-manager => k8s.io/kube-controller-manager v0.0.0-20191016114939-2b2b218dc1df
  21. k8s.io/kube-proxy => k8s.io/kube-proxy v0.0.0-20191016114407-2e83b6f20229
  22. k8s.io/kube-scheduler => k8s.io/kube-scheduler v0.0.0-20191016114748-65049c67a58b
  23. k8s.io/kubectl => k8s.io/kubectl v0.0.0-20191016120415-2ed914427d51
  24. k8s.io/kubelet => k8s.io/kubelet v0.0.0-20191016114556-7841ed97f1b2
  25. k8s.io/legacy-cloud-providers => k8s.io/legacy-cloud-providers v0.0.0-20191016115753-cf0698c3a16b
  26. k8s.io/metrics => k8s.io/metrics v0.0.0-20191016113814-3b1a734dba6e
  27. k8s.io/sample-apiserver => k8s.io/sample-apiserver v0.0.0-20191016112829-06bb3c9d77c9
  28. )
  29. replace github.com/docker/docker => github.com/moby/moby v0.7.3-0.20190826074503-38ab9da00309 // Required by Helm
  • Run go mod tidy to update the project modules
  • Run the command operator-sdk generate k8s to ensure that your resources will be updated
  • Run the command operator-sdk generate crds to regenerate CRDs

(Optional) Skip metrics logs when the operator is running locally

There are changes to the default implementation of the metrics export. These changes require cmd/manager/main.go to be updated as follows.

Update imports:

  1. import (
  2. ...
  3. "errors"
  4. ...
  5. )

Replace:

  1. func main() {
  2. ...
  3. if err = serveCRMetrics(cfg); err != nil {
  4. log.Info("Could not generate and serve custom resource metrics", "error", err.Error())
  5. }
  6. // Add to the below struct any other metrics ports you want to expose.
  7. servicePorts := []v1.ServicePort{
  8. {Port: metricsPort, Name: metrics.OperatorPortName, Protocol: v1.ProtocolTCP, TargetPort: intstr.IntOrString{Type: intstr.Int, IntVal: metricsPort}},
  9. {Port: operatorMetricsPort, Name: metrics.CRPortName, Protocol: v1.ProtocolTCP, TargetPort: intstr.IntOrString{Type: intstr.Int, IntVal: operatorMetricsPort}},
  10. }
  11. // Create Service object to expose the metrics port(s).
  12. service, err := metrics.CreateMetricsService(ctx, cfg, servicePorts)
  13. if err != nil {
  14. log.Info("Could not create metrics Service", "error", err.Error())
  15. }
  16. // CreateServiceMonitors will automatically create the prometheus-operator ServiceMonitor resources
  17. // necessary to configure Prometheus to scrape metrics from this operator.
  18. services := []*v1.Service{service}
  19. _, err = metrics.CreateServiceMonitors(cfg, namespace, services)
  20. if err != nil {
  21. log.Info("Could not create ServiceMonitor object", "error", err.Error())
  22. // If this operator is deployed to a cluster without the prometheus-operator running, it will return
  23. // ErrServiceMonitorNotPresent, which can be used to safely skip ServiceMonitor creation.
  24. if err == metrics.ErrServiceMonitorNotPresent {
  25. log.Info("Install prometheus-operator in your cluster to create ServiceMonitor objects", "error", err.Error())
  26. }
  27. }

With:

  1. func main() {
  2. ...
  3. // Add the Metrics Service
  4. addMetrics(ctx, cfg, namespace)

And then, add implementation for addMetrics:

  1. // addMetrics will create the Services and Service Monitors to allow the operator export the metrics by using
  2. // the Prometheus operator
  3. func addMetrics(ctx context.Context, cfg *rest.Config, namespace string) {
  4. if err := serveCRMetrics(cfg); err != nil {
  5. if errors.Is(err, k8sutil.ErrRunLocal) {
  6. log.Info("Skipping CR metrics server creation; not running in a cluster.")
  7. return
  8. }
  9. log.Info("Could not generate and serve custom resource metrics", "error", err.Error())
  10. }
  11. // Add to the below struct any other metrics ports you want to expose.
  12. servicePorts := []v1.ServicePort{
  13. {Port: metricsPort, Name: metrics.OperatorPortName, Protocol: v1.ProtocolTCP, TargetPort: intstr.IntOrString{Type: intstr.Int, IntVal: metricsPort}},
  14. {Port: operatorMetricsPort, Name: metrics.CRPortName, Protocol: v1.ProtocolTCP, TargetPort: intstr.IntOrString{Type: intstr.Int, IntVal: operatorMetricsPort}},
  15. }
  16. // Create Service object to expose the metrics port(s).
  17. service, err := metrics.CreateMetricsService(ctx, cfg, servicePorts)
  18. if err != nil {
  19. log.Info("Could not create metrics Service", "error", err.Error())
  20. }
  21. // CreateServiceMonitors will automatically create the prometheus-operator ServiceMonitor resources
  22. // necessary to configure Prometheus to scrape metrics from this operator.
  23. services := []*v1.Service{service}
  24. _, err = metrics.CreateServiceMonitors(cfg, namespace, services)
  25. if err != nil {
  26. log.Info("Could not create ServiceMonitor object", "error", err.Error())
  27. // If this operator is deployed to a cluster without the prometheus-operator running, it will return
  28. // ErrServiceMonitorNotPresent, which can be used to safely skip ServiceMonitor creation.
  29. if err == metrics.ErrServiceMonitorNotPresent {
  30. log.Info("Install prometheus-operator in your cluster to create ServiceMonitor objects", "error", err.Error())
  31. }
  32. }
  33. }
  34. // serveCRMetrics gets the Operator/CustomResource GVKs and generates metrics based on those types.
  35. ...

NOTE: For more information check the PR which is responsible for the above changes #2190.

Deprecations

The github.com/operator-framework/operator-sdk/pkg/restmapper package was deprecated in favor of the DynamicRESTMapper implementation in controller-runtime. Users should migrate to controller-runtime’s implementation, which is a drop-in replacement.

Replace:

  1. github.com/operator-framework/operator-sdk/pkg/restmapper.DynamicRESTMapper

With:

  1. sigs.k8s.io/controller-runtime/pkg/client/apiutil.DynamicRESTMapper

Breaking Changes

Add operator_sdk.util Ansible collection

The Ansible module k8s_status was extracted and is now provided by the operator_sdk.util Ansible collection. See developer_guide for new usage.

To use the collection in a role, declare it at the root level in meta/main.yaml:

  1. collections:
  2. - operator_sdk.util

To use it in a playbook, declare it in the play:

  1. - hosts: all
  2. collections:
  3. - operator_sdk.util
  4. tasks:
  5. - k8s_status:
  6. api_version: app.example.com/v1
  7. kind: Foo
  8. name: "{{ meta.name }}"
  9. namespace: "{{ meta.namespace }}"
  10. status:
  11. foo: bar

You can also use the fully-qualified name without declaring the collection:

  1. - operator_sdk.util.k8s_status:
  2. api_version: app.example.com/v1
  3. kind: Foo
  4. name: "{{ meta.name }}"
  5. namespace: "{{ meta.namespace }}"
  6. status:
  7. foo: bar

Notable Changes

These notable changes contain just the most important user-facing changes. See the CHANGELOG for details of the release.

Ansible version update

The Ansible version in the init projects was upgraded from 2.6 to 2.9 for collections support. Update the meta/main.yaml file.

Replace:

  1. ...
  2. min_ansible_version: 2.6
  3. ...

With:

  1. ...
  2. min_ansible_version: 2.9
  3. ...

Helm Upgrade to V3

The Helm operator packages and base image were upgraded from Helm v2 to Helm v3. Note that cluster state for pre-existing CRs using Helm v2-based operators will be automatically migrated to Helm v3’s new release storage format, and existing releases may be upgraded due to changes in Helm v3’s label injection.

If you are using any external helm v2 tooling with the your helm operator-managed releases, you will need to upgrade to the equivalent helm v3 tooling.

v0.15.x

modules

  • Ensure the the following require modules and replace directives with the specific versions are present in your go.mod file:
  1. require (
  2. github.com/operator-framework/operator-sdk v0.15.2
  3. sigs.k8s.io/controller-runtime v0.4.0
  4. )
  5. // Pinned to kubernetes-1.16.2
  6. replace (
  7. k8s.io/api => k8s.io/api v0.0.0-20191016110408-35e52d86657a
  8. k8s.io/apiextensions-apiserver => k8s.io/apiextensions-apiserver v0.0.0-20191016113550-5357c4baaf65
  9. k8s.io/apimachinery => k8s.io/apimachinery v0.0.0-20191004115801-a2eda9f80ab8
  10. k8s.io/apiserver => k8s.io/apiserver v0.0.0-20191016112112-5190913f932d
  11. k8s.io/cli-runtime => k8s.io/cli-runtime v0.0.0-20191016114015-74ad18325ed5
  12. k8s.io/client-go => k8s.io/client-go v0.0.0-20191016111102-bec269661e48
  13. k8s.io/cloud-provider => k8s.io/cloud-provider v0.0.0-20191016115326-20453efc2458
  14. k8s.io/cluster-bootstrap => k8s.io/cluster-bootstrap v0.0.0-20191016115129-c07a134afb42
  15. k8s.io/code-generator => k8s.io/code-generator v0.0.0-20191004115455-8e001e5d1894
  16. k8s.io/component-base => k8s.io/component-base v0.0.0-20191016111319-039242c015a9
  17. k8s.io/cri-api => k8s.io/cri-api v0.0.0-20190828162817-608eb1dad4ac
  18. k8s.io/csi-translation-lib => k8s.io/csi-translation-lib v0.0.0-20191016115521-756ffa5af0bd
  19. k8s.io/kube-aggregator => k8s.io/kube-aggregator v0.0.0-20191016112429-9587704a8ad4
  20. k8s.io/kube-controller-manager => k8s.io/kube-controller-manager v0.0.0-20191016114939-2b2b218dc1df
  21. k8s.io/kube-proxy => k8s.io/kube-proxy v0.0.0-20191016114407-2e83b6f20229
  22. k8s.io/kube-scheduler => k8s.io/kube-scheduler v0.0.0-20191016114748-65049c67a58b
  23. k8s.io/kubectl => k8s.io/kubectl v0.0.0-20191016120415-2ed914427d51
  24. k8s.io/kubelet => k8s.io/kubelet v0.0.0-20191016114556-7841ed97f1b2
  25. k8s.io/legacy-cloud-providers => k8s.io/legacy-cloud-providers v0.0.0-20191016115753-cf0698c3a16b
  26. k8s.io/metrics => k8s.io/metrics v0.0.0-20191016113814-3b1a734dba6e
  27. k8s.io/sample-apiserver => k8s.io/sample-apiserver v0.0.0-20191016112829-06bb3c9d77c9
  28. )
  29. replace github.com/docker/docker => github.com/moby/moby v0.7.3-0.20190826074503-38ab9da00309 // Required by Helm
  30. replace github.com/openshift/api => github.com/openshift/api v0.0.0-20190924102528-32369d4db2ad // Required until https://github.com/operator-framework/operator-lifecycle-manager/pull/1241 is resolved
  • Run go mod tidy to update the project modules
  • Run the command operator-sdk generate k8s to ensure that your resources will be updated
  • Run the command operator-sdk generate crds to regenerate CRDs

Breaking Changes on Commands

This release contains breaking changes in some commands.

  • The operator-sdk olm-catalog gen-csv was replaced by operator-sdk generate csv
  • The operator-sdk up local is now operator-sdk run --local. However, all functionality of this command is retained.
  • And then, the operator-sdk alpha olm [sub-commands] [flags] was moved from alpha to its own sub-command. However, all functionality of this command is retained. To check run; operator-sdk olm --help.

Breaking Changes for Helm and Ansible

The operator-sdk run ansible/helm are now hidden commands in exec-entrypoint ansible/helm. However, all functionality of each sub-command is still the same. If you are using this feature then you will need to replace the run for exec-entrypoint as the following examples.

Replace:

  1. oprator-sdk run ansible --watches-file=/opt/ansible/watches.yaml

With:

  1. oprator-sdk exec-entrypoint ansible --watches-file=/opt/ansible/watches.yaml

Replace:

  1. oprator-sdk run helm --watches-file=$HOME/watches.yaml

With:

  1. oprator-sdk run exec-entrypoint helm --watches-file=$HOME/watches.yaml

See the CHANGELOG for details of the release.

v0.16.x

modules

  • Ensure that the following require modules and replace directives with the specific versions are present in your go.mod file:
  1. require (
  2. github.com/operator-framework/operator-sdk v0.16.0
  3. sigs.k8s.io/controller-runtime v0.4.0
  4. )
  5. // Pinned to kubernetes-1.16.2
  6. replace (
  7. k8s.io/api => k8s.io/api v0.0.0-20191016110408-35e52d86657a
  8. k8s.io/apiextensions-apiserver => k8s.io/apiextensions-apiserver v0.0.0-20191016113550-5357c4baaf65
  9. k8s.io/apimachinery => k8s.io/apimachinery v0.0.0-20191004115801-a2eda9f80ab8
  10. k8s.io/apiserver => k8s.io/apiserver v0.0.0-20191016112112-5190913f932d
  11. k8s.io/cli-runtime => k8s.io/cli-runtime v0.0.0-20191016114015-74ad18325ed5
  12. k8s.io/client-go => k8s.io/client-go v0.0.0-20191016111102-bec269661e48
  13. k8s.io/cloud-provider => k8s.io/cloud-provider v0.0.0-20191016115326-20453efc2458
  14. k8s.io/cluster-bootstrap => k8s.io/cluster-bootstrap v0.0.0-20191016115129-c07a134afb42
  15. k8s.io/code-generator => k8s.io/code-generator v0.0.0-20191004115455-8e001e5d1894
  16. k8s.io/component-base => k8s.io/component-base v0.0.0-20191016111319-039242c015a9
  17. k8s.io/cri-api => k8s.io/cri-api v0.0.0-20190828162817-608eb1dad4ac
  18. k8s.io/csi-translation-lib => k8s.io/csi-translation-lib v0.0.0-20191016115521-756ffa5af0bd
  19. k8s.io/kube-aggregator => k8s.io/kube-aggregator v0.0.0-20191016112429-9587704a8ad4
  20. k8s.io/kube-controller-manager => k8s.io/kube-controller-manager v0.0.0-20191016114939-2b2b218dc1df
  21. k8s.io/kube-proxy => k8s.io/kube-proxy v0.0.0-20191016114407-2e83b6f20229
  22. k8s.io/kube-scheduler => k8s.io/kube-scheduler v0.0.0-20191016114748-65049c67a58b
  23. k8s.io/kubectl => k8s.io/kubectl v0.0.0-20191016120415-2ed914427d51
  24. k8s.io/kubelet => k8s.io/kubelet v0.0.0-20191016114556-7841ed97f1b2
  25. k8s.io/legacy-cloud-providers => k8s.io/legacy-cloud-providers v0.0.0-20191016115753-cf0698c3a16b
  26. k8s.io/metrics => k8s.io/metrics v0.0.0-20191016113814-3b1a734dba6e
  27. k8s.io/sample-apiserver => k8s.io/sample-apiserver v0.0.0-20191016112829-06bb3c9d77c9
  28. )
  29. replace github.com/docker/docker => github.com/moby/moby v0.7.3-0.20190826074503-38ab9da00309 // Required by Helm
  30. replace github.com/openshift/api => github.com/openshift/api v0.0.0-20190924102528-32369d4db2ad // Required until https://github.com/operator-framework/operator-lifecycle-manager/pull/1241 is resolved
  • Run go mod tidy to update the project modules
  • Run the command operator-sdk generate k8s to ensure that your resources will be updated
  • Run the command operator-sdk generate crds to regenerate CRDs

Bug Fixes and Improvements for Metrics

There are changes to the default implementation of the metrics export. These changes require cmd/manager/main.go to be updated as follows.

Replace:

  1. func main() {
  2. ...
  3. // Add the Metrics Service
  4. addMetrics(ctx, cfg, namespace)

With:

  1. func main() {
  2. ...
  3. // Add the Metrics Service
  4. addMetrics(ctx, cfg)

And then, update the default implementation of addMetrics and serveCRMetrics with:

  1. // addMetrics will create the Services and Service Monitors to allow the operator export the metrics by using
  2. // the Prometheus operator
  3. func addMetrics(ctx context.Context, cfg *rest.Config) {
  4. // Get the namespace the operator is currently deployed in.
  5. operatorNs, err := k8sutil.GetOperatorNamespace()
  6. if err != nil {
  7. if errors.Is(err, k8sutil.ErrRunLocal) {
  8. log.Info("Skipping CR metrics server creation; not running in a cluster.")
  9. return
  10. }
  11. }
  12. if err := serveCRMetrics(cfg, operatorNs); err != nil {
  13. log.Info("Could not generate and serve custom resource metrics", "error", err.Error())
  14. }
  15. // Add to the below struct any other metrics ports you want to expose.
  16. servicePorts := []v1.ServicePort{
  17. {Port: metricsPort, Name: metrics.OperatorPortName, Protocol: v1.ProtocolTCP, TargetPort: intstr.IntOrString{Type: intstr.Int, IntVal: metricsPort}},
  18. {Port: operatorMetricsPort, Name: metrics.CRPortName, Protocol: v1.ProtocolTCP, TargetPort: intstr.IntOrString{Type: intstr.Int, IntVal: operatorMetricsPort}},
  19. }
  20. // Create Service object to expose the metrics port(s).
  21. service, err := metrics.CreateMetricsService(ctx, cfg, servicePorts)
  22. if err != nil {
  23. log.Info("Could not create metrics Service", "error", err.Error())
  24. }
  25. // CreateServiceMonitors will automatically create the prometheus-operator ServiceMonitor resources
  26. // necessary to configure Prometheus to scrape metrics from this operator.
  27. services := []*v1.Service{service}
  28. // The ServiceMonitor is created in the same namespace where the operator is deployed
  29. _, err = metrics.CreateServiceMonitors(cfg, operatorNs, services)
  30. if err != nil {
  31. log.Info("Could not create ServiceMonitor object", "error", err.Error())
  32. // If this operator is deployed to a cluster without the prometheus-operator running, it will return
  33. // ErrServiceMonitorNotPresent, which can be used to safely skip ServiceMonitor creation.
  34. if err == metrics.ErrServiceMonitorNotPresent {
  35. log.Info("Install prometheus-operator in your cluster to create ServiceMonitor objects", "error", err.Error())
  36. }
  37. }
  38. }
  39. // serveCRMetrics gets the Operator/CustomResource GVKs and generates metrics based on those types.
  40. // It serves those metrics on "http://metricsHost:operatorMetricsPort".
  41. func serveCRMetrics(cfg *rest.Config, operatorNs string) error {
  42. // The function below returns a list of filtered operator/CR specific GVKs. For more control, override the GVK list below
  43. // with your own custom logic. Note that if you are adding third party API schemas, probably you will need to
  44. // customize this implementation to avoid permissions issues.
  45. filteredGVK, err := k8sutil.GetGVKsFromAddToScheme(apis.AddToScheme)
  46. if err != nil {
  47. return err
  48. }
  49. // The metrics will be generated from the namespaces which are returned here.
  50. // NOTE that passing nil or an empty list of namespaces in GenerateAndServeCRMetrics will result in an error.
  51. ns, err := kubemetrics.GetNamespacesForMetrics(operatorNs)
  52. if err != nil {
  53. return err
  54. }
  55. // Generate and serve custom resource specific metrics.
  56. err = kubemetrics.GenerateAndServeCRMetrics(cfg, ns, filteredGVK, metricsHost, operatorMetricsPort)
  57. if err != nil {
  58. return err
  59. }
  60. return nil
  61. }

NOTE: For more information check the PRs which are responsible for the above changes #2606,#2603 and #2601.

(Optional) Support for watching multiple namespaces

There are changes to add support for watching multiple namespaces. These changes require cmd/manager/main.go to be updated as follows.

Update imports:

  1. import (
  2. ...
  3. "strings"
  4. ...
  5. "sigs.k8s.io/controller-runtime/pkg/cache"
  6. ...
  7. )

Replace:

  1. func main() {
  2. ...
  3. // Create a new Cmd to provide shared dependencies and start components
  4. mgr, err := manager.New(cfg, manager.Options{
  5. Namespace: namespace,
  6. MetricsBindAddress: fmt.Sprintf("%s:%d", metricsHost, metricsPort),
  7. })
  8. if err != nil {
  9. log.Error(err, "")
  10. os.Exit(1)
  11. }

With:

  1. func main() {
  2. ...
  3. // Set default manager options
  4. options := manager.Options{
  5. Namespace: namespace,
  6. MetricsBindAddress: fmt.Sprintf("%s:%d", metricsHost, metricsPort),
  7. }
  8. // Add support for MultiNamespace set in WATCH_NAMESPACE (e.g ns1,ns2)
  9. // Note that this is not intended to be used for excluding namespaces, this is better done via a Predicate
  10. // Also note that you may face performance issues when using this with a high number of namespaces.
  11. // More Info: https://godoc.org/github.com/kubernetes-sigs/controller-runtime/pkg/cache#MultiNamespacedCacheBuilder
  12. if strings.Contains(namespace, ",") {
  13. options.Namespace = ""
  14. options.NewCache = cache.MultiNamespacedCacheBuilder(strings.Split(namespace, ","))
  15. }
  16. // Create a new manager to provide shared dependencies and start components
  17. mgr, err := manager.New(cfg, options)
  18. if err != nil {
  19. log.Error(err, "")
  20. os.Exit(1)
  21. }

NOTE: For more information check the PR which is responsible for the above changes #2522.

Breaking changes

TestCtx in pkg/test has been deprecated

The type name TestCtx in pkg/test has been deprecated and renamed to Context. Users of the e2e framework should do the following:

  • Replace TestCtx with Context
  • Replace NewTestCtx with NewContext

Scorecard only supports YAML config files

The scorecard feature now only supports YAML config files. Config files with other extensions are no longer supported and should be changed to the YAML format. For further information see scorecard config file

Breaking Changes for Ansible

Remove Ansible container sidecar

The Ansible logs are now output in the operator container, so there is no longer a need for the Ansible container sidecar. To reflect this change, update the deploy/operator.yaml file as follows.

Remove:

  1. - name: ansible
  2. command:
  3. - /usr/local/bin/ao-logs
  4. - /tmp/ansible-operator/runner
  5. - stdout
  6. # Replace this with the built image name**
  7. image: "REPLACE_IMAGE"
  8. imagePullPolicy: "Always"
  9. volumeMounts:
  10. - mountPath: /tmp/ansible-operator/runner
  11. name: runner
  12. readOnly: true

Replace:

  1. - name: operator

With:

  1. - name: {{your operator name which is the value of metadata.name in this file}}

By default the full Ansible logs will not be output, however, you can setup it via the ANSIBLE_DEBUG_LOGS environment variable in the deploy/operator.yaml file. See:

  1. ...
  2. - name: ANSIBLE_DEBUG_LOGS
  3. value: "True"
  4. ...

Migration to Ansible collections

The core Ansible Kubernetes modules have been moved to the community.kubernetes Ansible collection. Future development of the modules will occur there, with only critical bugfixes going into the modules in core. Additionally, the operator_sdk.util collection is no longer installed by default in the base image. Instead, users should add a requirements.yml to their project root, with the following content:

  1. collections:
  2. - community.kubernetes
  3. - operator_sdk.util

Users should then add the following stages to their build/Dockerfile:

  1. COPY requirements.yml ${HOME}/requirements.yml
  2. RUN ansible-galaxy collection install -r ${HOME}/requirements.yml \
  3. && chmod -R ug+rwx ${HOME}/.ansible

v0.17.x

modules

  • Ensure that the following require modules and replace directives with the specific versions are present in your go.mod file:
  1. require (
  2. github.com/operator-framework/operator-sdk v0.17.1
  3. sigs.k8s.io/controller-runtime v0.5.2
  4. )
  5. replace (
  6. k8s.io/client-go => k8s.io/client-go v0.17.4 // Required by prometheus-operator
  7. github.com/Azure/go-autorest => github.com/Azure/go-autorest v13.3.2+incompatible // Required by OLM
  8. )
  • Run go mod tidy to update the project modules
  • Run the command operator-sdk generate k8s to ensure that your resources will be updated
  • Run the command operator-sdk generate crds to regenerate CRDs

Breaking Changes

OpenAPI generation

  • The deprecated operator-sdk generate openapi command has been removed. This command generated CRDs and zz_generated.openapi files for operator APIs.

To generate CRDs, use operator-sdk generate crds.

To generate Go OpenAPI code, use openapi-gen directly. For example:

  1. &#35 Build the latest openapi-gen from source
  2. which ./bin/openapi-gen > /dev/null || go build -o ./bin/openapi-gen k8s.io/kube-openapi/cmd/openapi-gen
  3. &#35 Run openapi-gen for each of your API group/version packages
  4. ./bin/openapi-gen --logtostderr=true \
  5. -i ./pkg/apis/<group>/<version> \
  6. -o "" \
  7. -O zz_generated.openapi \
  8. -p ./pkg/apis/<group>/<version> \
  9. -h ./hack/boilerplate.go.txt \
  10. -r "-"

Molecule Upgrade for Ansible based-operators

The Molecule version for Ansible based-operators was upgraded from 2.22 to 3.0.2. The following changes are required in the default scaffold files.

  • Remove the scenario.name from molecule.yaml and then, ensure that any condition with will look for the folder name which determines the scenario name from now on
  • Replace the lint with newer syntax from documentation. See:

Replace:

  1. lint:
  2. name: yamllint
  3. options:
  4. config-data:
  5. line-length:
  6. max: 120

With:

  1. lint: |
  2. set -e
  3. yamllint -d "{extends: relaxed, rules: {line-length: {max: 120}}}" .

Replace:

  1. lint:
  2. name: ansible-lint

With:

  1. lint: |
  2. set -e
  3. ansible-lint
  • Rename molecule/$SCENARIO/playbook.yml to molecule/$SCENARIO/converge.yml to avoid a deprecation message.
  • Update the .travis.yml file to install the supported lints as follows.

Replace:

  1. install:
  2. - pip3 install docker molecule openshift jmespath

With:

  1. install:
  2. - pip3 install docker molecule ansible-lint yamllint flake8 openshift jmespath

NOTE To know more about how to upgrade your project to use the V3 Molecule version see here.

Deprecations

Test Framework

  • The methods ctx.GetOperatorNamespace() and ctx.GetWatchNamespace() were added to pkg/test in order to replace ctx.GetNamespace() which is deprecated. In this way, replace the use of ctx.GetNamespace() in your project with ctx.GetOperatorNamespace().

  • The --namespace flag from operator-sdk run --local, operator-sdk test --local, and operator-sdk cleanup was deprecated and is replaced by --watch-namespace and --operator-namespace.

    The --operator-namespace flag can be used to set the namespace where the operator will be deployed. It will set the environment variable OPERATOR_NAMESPACE. If this value is not set, then it will be the namespace defined as in your current kubeconfig context.

    The --watch-namespace flag can be used to set the namespace(s) which the operator will watch for changes. It will set the environment variable WATCH_NAMESPACE. Use an explicit empty string to watch all namespaces or a comma-separated list of namespaces (e.g. “ns1,ns2”) to watch multiple namespace when the operator is cluster-scoped. If using a list, then it should contain the namespace where the operator is deployed since the default metrics implementation will manage resources in the Operator’s namespace. By default, --watch-namespace will be set to the operator namespace.

  • If you’ve run operator-sdk bundle create --generate-only, move your bundle Dockerfile at <project-root>/deploy/olm-catalog/<operator-name>/Dockerfile to <project-root>/bundle.Dockerfile and update the first COPY from COPY /*.yaml manifests/ to COPY deploy/olm-catalog/<operator-name>/manifests manifests/.

Last modified July 17, 2020: [docs] Reorganize hugo doc site (#3382) (1248c1ca)