5e13c68a5b
This also forces the RP from Go 1.14 to Go 1.16. Aside from requiring OCP 4.9 / Kubernetes 1.22 modules, the other go.mod changes are all manual workarounds from failed "make vendor" runs. |
||
---|---|---|
.. | ||
apis | ||
clientset/versioned | ||
controllers | ||
deploy | ||
README.md | ||
const.go | ||
generate.go |
README.md
Azure Red Hat OpenShift Operator
Responsibilities
Decentralizing service monitoring
This has the advantage of moving a proportion of monitoring effort to the edge, giving headroom (and the corresponding potential disadvantage of increased management complexity). Doing this helps avoid bloat and complexity risks in central monitoring as well as enabling additional and more complex monitoring use cases. Note that not all monitoring can be decentralised.
In all cases below the status.Conditions will be set.
- periodically check for outbound internet connectivity from both the master and worker nodes.
- periodically validate the cluster Service Principal permissions.
- [TODO] Enumerate daemonset statuses, pod statuses, etc. We currently log diagnostic information associated with these checks in service logs; moving the checks to the edge will make these cluster logs, which is preferable.
Automatic service remediation
There will be use cases where we may want to remediate end user decisions automatically. Carrying out remediation locally is advantageous because it is likely to be simpler, more reliable, and with a shorter time to remediate.
Remediations in place:
- periodically reset NSGs in the master and worker subnets to the defaults (controlled by the reconcileNSGs feature flag)
End user warnings
Decentralizing ARO customization management
A cluster agent provides a centralized location to handle this use case. Many post-install configurations should probably move here.
- monitor and repair mdsd as needed
- set the alertmanager webhook
Controllers and Deployment
The full list of operator controllers with descriptions can be found in the README at the root of the repository.
The static pod resources can be found at pkg/operator/deploy/staticresources
. The
deploy operation kicks off two deployments in the openshift-azure-operator
namespace, one for
master and one for worker. The aro-operator-master
deployment runs all controllers,
while the aro-operator-worker
deployment runs only the internet checker in the worker subnet.
Developer documentation
How to Run a pre built operator image
Add the following to your "env" before running the rp
export ARO_IMAGE=arointsvc.azurecr.io/aro:latest
How to Run the operator locally (out of cluster)
Make sure KUBECONFIG is set:
make admin.kubeconfig
export KUBECONFIG=$(pwd)/admin.kubeconfig
oc scale -n openshift-azure-operator deployment/aro-operator-master --replicas=0
make generate
go run ./cmd/aro operator master
How to run a custom operator image
Add the following to your "env" before running the rp
export ARO_IMAGE=quay.io/asalkeld/aos-init:latest #(change to yours)
make publish-image-aro
#Then run an update
curl -X PATCH -k "https://localhost:8443/subscriptions/$AZURE_SUBSCRIPTION_ID/resourceGroups/$RESOURCEGROUP/providers/Microsoft.RedHatOpenShift/openShiftClusters/$CLUSTER?api-version=admin" --header "Content-Type: application/json" -d "{}"
#check on the deployment
oc -n openshift-azure-operator get all
oc -n openshift-azure-operator get clusters.aro.openshift.io/cluster -o yaml
oc -n openshift-azure-operator logs deployment.apps/aro-operator-master
oc -n openshift-config get secrets/pull-secret -o template='{{index .data ".dockerconfigjson"}}' | base64 -d
How to run operator e2e tests
go test ./test/e2e -v -ginkgo.v -ginkgo.focus="ARO Operator" -tags e2e