The patch operator helps with defining patches in a declarative way. This operator has two main features:
- ability to patch an object at creation time via a mutating webhook
- ability to enforce patches on one or more objects via a controller
- Patch Operator
Why apply a patch at creation time when you could directly create the correct object? The reason is that sometime the correct value depends on configuration set on the specific cluster in which the object is being deployed. For example, an ingress/route hostname might depend on the specific cluster. Consider the following example based on cert-manager:
apiVersion: cert-manager.io/v1
kind: ClusterIssuer
metadata:
name: letsencrypt-issuer
spec:
acme:
server: 'https://acme-v02.api.letsencrypt.org/directory'
email: {{ .Values.letsencrypt.email }}
privateKeySecretRef:
name: letsencrypt-staging
solvers:
- dns01:
route53:
accessKeyID: << access_key >>
secretAccessKeySecretRef:
name: cert-manager-dns-credentials
key: aws_secret_access_key
region: << region >>
hostedZoneID: << hosted_zone_id >>In this example the fields: << access_key >>, << region >> and << hosted_zone_id >> are dependent on the specific region in which the cluster is being deployed and in many cases they are discoverable from other configurations already present in the cluster. If you want to deploy the above Cluster Issuer object with a gitops approach, then there is no easy way to discover those values. The solution so far is to manually discover those values and create a different gitops configuration for each cluster. But consider if you could look up values at deploy time based on the cluster you are deploying to. Here is how this object might look:
apiVersion: cert-manager.io/v1
kind: ClusterIssuer
metadata:
name: letsencrypt-issuer
namespace: {{ .Release.Namespace }}
annotations:
"redhat-cop.redhat.io/patch": |
spec:
acme:
- dns01:
route53:
accessKeyID: {{ (lookup "v1" "Secret" .metadata.namespace "cert-manager-dns-credentials").data.aws_access_key_id | b64dec }}
secretAccessKeySecretRef:
name: cert-manager-dns-credentials
key: aws_secret_access_key
region: {{ (lookup "config.openshift.io/v1" "Infrastructure" "" "cluster").status.platformStatus.aws.region }}
hostedZoneID: {{ (lookup "config.openshift.io/v1" "DNS" "" "cluster").spec.publicZone.id }}
spec:
acme:
server: 'https://acme-v02.api.letsencrypt.org/directory'
email: {{ .Values.letsencrypt.email }}
privateKeySecretRef:
name: letsencrypt-staging
solvers:
- dns01:
route53:
accessKeyID: << access_key >>
secretAccessKeySecretRef:
name: cert-manager-dns-credentials
key: aws_secret_access_key
region: << region >>
hostedZoneID: << hosted_zone_id >>The annotation specifies a patch that will be applied by a MutatingWebhook, as you can see the three values are now looked up from different configurations in the cluster.
Two annotations influences the behavior of this MutatingWebhook
- "redhat-cop.redhat.io/patch" : this is the patch itself. The patch is evaluated as a template with the object itself as it's only parameter. The template is expressed in golang template notation and supports the same functions as helm template including the lookup function which plays a major role here. The patch must be expressed in yaml for readability. It will be converted to json by the webhook logic.
- "redhat-cop.redhat.io/patch-type" : this is the type of json patch. The possible values are:
application/json-patch+json,application/merge-patch+jsonandapplication/strategic-merge-patch+json. If this annotation is omitted it defaults to strategic merge.
The lookup function, if used by the template, is executed with a client which impersonates the user issuing the object creation/update request. This should prevent security permission leakage.
The creation time webhook is not installed by the operator. This is because there is no way to know which specific object type should be intercepted and intercepting all of the types would be too inefficient. It's up to the administrator then to install the webhook. Here is some guidance.
The configurations that needs to be applied in order to support creation time webhooks depends on how the operator was installed (OLM or Helm chart).
If you installed the operator via OLM, the certificate that the OLM generates to expose the webhook need to be applied dynamically to the MutatingWebhookConfiguration resource. To support this action, we can use the Patch operator to perform the needed configurations.
First, create the following resources which will create a ServiceAccount and RBAC policies so that the operator can apply the needed configurations:
---
apiVersion: v1
kind: ServiceAccount
metadata:
creationTimestamp: null
name: mutatingwebhook-patcher
namespace: patch-operator
---
apiVersion: rbac.authorization.k8s.io/v1
kind: ClusterRole
metadata:
labels:
name: mutatingwebhookconfiguration-patcher
rules:
- apiGroups:
- ""
resources:
- secrets
verbs:
- get
- list
- watch
- apiGroups:
- "admissionregistration.k8s.io"
resources:
- mutatingwebhookconfigurations
verbs:
- get
- list
- watch
- patch
- update
---
apiVersion: rbac.authorization.k8s.io/v1
kind: ClusterRoleBinding
metadata:
name: mutatingwebhookconfiguration-patcher
roleRef:
apiGroup: rbac.authorization.k8s.io
kind: ClusterRole
name: mutatingwebhookconfiguration-patcher
subjects:
- kind: ServiceAccount
name: mutatingwebhook-patcher
namespace: patch-operatorNext, apply the following Patch resource which will look up the Secret the OLM created containing the CA used by the webhook
apiVersion: redhatcop.redhat.io/v1alpha1
kind: Patch
metadata:
name: patch-operator-mutatingwebhookconfiguration
namespace: patch-operator
spec:
serviceAccountRef:
name: mutatingwebhook-patcher
patches:
patch-operator-mutatingwebhookconfigurations:
targetObjectRef:
apiVersion: admissionregistration.k8s.io/v1
kind: MutatingWebhookConfiguration
name: patch-operator-inject
patchTemplate: '[{"op": "replace", "path": "/webhooks/0/clientConfig/caBundle", "value":"{{ (index (index . 1).data "olmCAKey") }}"}]'
patchType: application/json-patch+json
sourceObjectRefs:
- apiVersion: v1
kind: Secret
name: patch-operator-controller-manager-service-cert
namespace: patch-operatorHere is the webhook:
---
apiVersion: admissionregistration.k8s.io/v1
kind: MutatingWebhookConfiguration
metadata:
name: patch-operator-inject
webhooks:
- admissionReviewVersions:
- v1
clientConfig:
service:
name: patch-operator-controller-manager-service
namespace: patch-operator
path: /inject
caBundle: Cg==
failurePolicy: Fail
name: patch-operator-inject.redhatcop.redhat.io
rules:
- << add your intercepted objects here >>
sideEffects: NoneIf you installed the operator via the Helm chart and are using cert-manager, use the following webhook template:
apiVersion: admissionregistration.k8s.io/v1
kind: MutatingWebhookConfiguration
metadata:
name: patch-operator-inject
annotations:
cert-manager.io/inject-ca-from: '{{ .Release.Namespace }}/webhook-server-cert'
webhooks:
- admissionReviewVersions:
- v1
clientConfig:
service:
name: patch-operator-webhook-service
namespace: patch-operator
path: /inject
failurePolicy: Fail
name: patch-operator-inject.redhatcop.redhat.io
rules:
- << add your intercepted objects here >>
sideEffects: NoneNo additional steps are needed since as cert-manager manages setting the caBundle field on the MutatingWebhookConfiguration
For the rules that apply to webhooks, you should need to enable the webhook only for CREATE operations. So for example to enable the webhook on configmaps:
rules:
- apiGroups:
- ""
apiVersions:
- v1
operations:
- CREATE
resources:
- configmapsThere are situations when we need to patch pre-existing objects. Again this is a use case that is hard to model with gitops operators which will work only on object that they own. Especially with sophisticated Kubernetes distributions, it is not uncommon that a Kubernetes instance, at installation time, is configured with some default settings. Changing those configurations means patching those objects. For example, let's take the case of OpenShift Oauth configuration. This object is present by default and it is expected to be patched with any newly enabled authentication mechanism. This is how it looks like after installation:
apiVersion: config.openshift.io/v1
kind: OAuth
metadata:
name: cluster
ownerReferences:
- apiVersion: config.openshift.io/v1
kind: ClusterVersion
name: version
uid: 9a9d450b-3076-4e30-ac05-a889d6341fc3
resourceVersion: '20405124'
spec: []If we need to patch it we can use the patch controller and the Patch object as follows:
apiVersion: redhatcop.redhat.io/v1alpha1
kind: Patch
metadata:
name: gitlab-ocp-oauth-provider
namespace: openshift-config
spec:
serviceAccountRef:
name: default
patches:
gitlab-ocp-oauth-provider:
targetObjectRef:
apiVersion: config.openshift.io/v1
kind: OAuth
name: cluster
patchTemplate: |
spec:
identityProviders:
- name: my-github
mappingMethod: claim
type: GitHub
github:
clientID: "{{ (index . 1).data.client_id | b64dec }}"
clientSecret:
name: ocp-github-app-credentials
organizations:
- my-org
teams: []
patchType: application/merge-patch+json
sourceObjectRefs:
- apiVersion: v1
kind: Secret
name: ocp-github-app-credentials
namespace: openshift-configThis will cause the OAuth object to be patched and the patch to be enforced. That means that if anything changes on the secret that we use a parameter (which may be rotated) or in Oauth object itself, the patch will be reapplied. In this case we are adding a gitlab authentication provider.
A patch has the following fields:
targetObjectRef this refers to the object(s) receiving the patch. Mutliple object can be selected based on the following rules:
| Namespaced Type | Namespace | Name | Selection type |
|---|---|---|---|
| yes | null | null | multiple selection across namespaces |
| yes | null | not null | multiple selection across namespaces where the name corresponds to the passed name |
| yes | not null | null | multiple selection within a namespace |
| yes | not null | not nul | single selection |
| no | N/A | null | multiple selection |
| no | N/A | not null | single selection |
Selection can be further narrowed down by filtering by labels and/or annotations using the labelSelector and annotationSelector fields. The patch will be applied to all of the selected instances.
sourceObjectRefs these are the objects that will be watched and become part of the parameters of the patch template. Name and Namespace of sourceRefObjects are interpreted as golang templates with the current target instance and the only parameter. This allows to select different source object for each target object.
So, for example, with this patch:
apiVersion: redhatcop.redhat.io/v1alpha1
kind: Patch
metadata:
name: multiple-namespaced-targets-patch
spec:
serviceAccountRef:
name: default
patches:
multiple-namespaced-targets-patch:
targetObjectRef:
apiVersion: v1
kind: ServiceAccount
name: deployer
patchTemplate: |
metadata:
annotations:
{{ (index . 1).metadata.uid }}: {{ (index . 1) }}
patchType: application/strategic-merge-patch+json
sourceObjectRefs:
- apiVersion: v1
kind: ServiceAccount
name: default
namespace: "{{ .metadata.namespace }}"
fieldPath: $.metadata.uidThe deployer service accounts from all namespaces are selected as target of this patch, each patch template will receive a different parameter and that is the default service account of the same namespace as the namespace of the deployer service account being processed.
sourceObjectRefs also have the fieldPath field which can contain a jsonpath expression. If a value is passed the jsonpath expression will be calculate for the current source object and the result will be passed as parameter of the template.
patchTemplate This is the the template that will be evaluated. The result must be a valid patch compatible with the requested type and expressed in yaml for readability. The parameters passed to the template are the target object and then the all of the source object. So if you want to refer to the target object in the template you can use this expression (index . 0). Higher indexes refer to the sourceObjectRef array. The template is expressed in golang template notation and supports the same functions as helm template.
patchType is the type of the json patch. The possible values are: application/json-patch+json, application/merge-patch+json and application/strategic-merge-patch+json. If this annotation is omitted it defaults to strategic merge.
The patch enforcement enacted by the patch controller is executed with a client which uses the service account referenced by the serviceAccountRef field. So before a patch object can actually work an administrator must have granted the needed permissions to a service account in the same namespace. The serviceAccountRef will default to the default service account if not specified.
This operator uses the TokenRequest API to get a token to instantiate an internal controller to watch for the target(s) and source(s) of a patch. The token request API returns time bound token. At the moment, this token is refreshed when the patch is changed or when the operator is restarted. It is a responsibility of the administrator to make sure that the token is refreshed before its expiration or the patch will stop being enforced. By default, tokens have a 1 year expiration period. This default can be changed via the SERVICE_ACCOUNT_TOKEN_EXPIRATION_DURATION environment variable. Environment variables can be set following these instructions The expected format is of time.Duration
The patch controller will create a controller-manager and per Patch object and a reconciler for each of the PatchSpec defined in the array on patches in the Patch object.
These reconcilers share the same cached client. In order to be able to watch changes on target and source objects of a PatchSpec, all of the target and source object type instances will be cached by the client. This is a normal behavior of a controller-manager client, but it implies that if you create patches on object types that have many instances in etcd (Secrets, ServiceAccounts, Namespaces for example), the patch operator instance will require a significant amount of memory. A way to contain this issue is to try to aggregate together PatchSpec that deal with the same object types. This will cause those object type instances to cached only once.
This is a cluster-level operator that you can deploy in any namespace, patch-operator is recommended.
It is recommended to deploy this operator via OperatorHub, but you can also deploy it using Helm.
| Arch | Support |
|---|---|
| amd64 | ✅ |
| arm64 | ✅ |
| ppc64le | ✅ |
| s390x | ✅ |
Note: This operator supports being installed disconnected environments
If you want to utilize the Operator Lifecycle Manager (OLM) to install this operator, you can do so in two ways: from the UI or the CLI.
- If you would like to launch this operator from the UI, you'll need to navigate to the OperatorHub tab in the console. Before starting, make sure you've created the namespace that you want to install this operator to with the following:
oc new-project patch-operator- Once there, you can search for this operator by name:
patch operator. This will then return an item for our operator and you can select it to get started. Once you've arrived here, you'll be presented with an option to install, which will begin the process. - After clicking the install button, you can then select the namespace that you would like to install this to as well as the installation strategy you would like to proceed with (
AutomaticorManual). - Once you've made your selection, you can select
Subscribeand the installation will begin. After a few moments you can go ahead and check your namespace and you should see the operator running.
If you'd like to launch this operator from the command line, you can use the manifests contained in this repository by running the following commands from the folder where you cloned the repository:
oc new-project patch-operator
oc apply -f config/operatorhub -n patch-operatorThis will create the appropriate OperatorGroup and Subscription and will trigger OLM to launch the operator in the specified namespace.
Here are the instructions to install the latest release with Helm.
oc new-project patch-operator
helm repo add patch-operator https://redhat-cop.github.io/patch-operator
helm repo update
helm install patch-operator patch-operator/patch-operatorThis can later be updated with the following commands:
helm repo update
helm upgrade patch-operator patch-operator/patch-operatorPrometheus compatible metrics are exposed by the Operator and can be integrated into OpenShift's default cluster monitoring. To enable OpenShift cluster monitoring, label the namespace the operator is deployed in with the label openshift.io/cluster-monitoring="true".
oc label namespace <namespace> openshift.io/cluster-monitoring="true"export operatorNamespace=patch-operator-local # or patch-operator
oc label namespace ${operatorNamespace} openshift.io/cluster-monitoring="true"
oc rsh -n openshift-monitoring -c prometheus prometheus-k8s-0 /bin/bash
export operatorNamespace=patch-operator-local # or patch-operator
curl -v -s -k -H "Authorization: Bearer $(cat /var/run/secrets/kubernetes.io/serviceaccount/token)" https://patch-operator-controller-manager-metrics.${operatorNamespace}.svc.cluster.local:8443/metrics
exitexport repo=raffaelespazzoli
docker login quay.io/$repo
oc new-project patch-operator
oc project patch-operator
envsubst < config/local-development/tilt/env-replace-image.yaml > config/local-development/tilt/replace-image.yaml
tilt upsee here
Define an image and tag. For example...
export imageRepository="quay.io/redhat-cop/patch-operator"
export imageTag="$(git -c 'versionsort.suffix=-' ls-remote --exit-code --refs --sort='version:refname' --tags https://github.com/redhat-cop/patch-operator.git '*.*.*' | tail --lines=1 | cut --delimiter='/' --fields=3)"Deploy chart...
make helmchart IMG=${imageRepository} VERSION=${imageTag}
helm upgrade -i patch-operator-local charts/patch-operator -n patch-operator-local --create-namespaceDelete...
helm delete patch-operator-local -n patch-operator-local
kubectl delete -f charts/patch-operator/crds/crds.yamlexport repo=raffaelespazzoli #replace with yours
docker login quay.io/$repo
make docker-build IMG=quay.io/$repo/patch-operator:latest
make docker-push IMG=quay.io/$repo/patch-operator:latestmake manifests
make bundle IMG=quay.io/$repo/patch-operator:latest
operator-sdk bundle validate ./bundle --select-optional name=operatorhub
make bundle-build BUNDLE_IMG=quay.io/$repo/patch-operator-bundle:latest
docker push quay.io/$repo/patch-operator-bundle:latest
operator-sdk bundle validate quay.io/$repo/patch-operator-bundle:latest --select-optional name=operatorhub
oc new-project patch-operator
oc label namespace patch-operator openshift.io/cluster-monitoring="true"
operator-sdk cleanup patch-operator -n patch-operator
operator-sdk run bundle --install-mode AllNamespaces -n patch-operator quay.io/$repo/patch-operator-bundle:latestgit tag -a "<tagname>" -m "<commit message>"
git push upstream <tagname>If you need to remove a release:
git tag -d <tagname>
git push upstream --delete <tagname>If you need to "move" a release to the current main
git tag -f <tagname>
git push upstream -f <tagname>operator-sdk cleanup patch-operator -n patch-operator
oc delete operatorgroup operator-sdk-og
oc delete catalogsource patch-operator-catalog
![@renovate[bot]](https://avatars.githubusercontent.com/in/2740?s=64&v=4){kind=small}