This is a clone of issue OCPBUGS-29858. The following is the description of the original issue:
—
The convention is a format like node-role.kubernetes.io/role: "", not node-role.kubernetes.io: role, however ROSA uses the latter format to indicate the infra role. This changes the node watch code to ignore it, as well as other potential variations like node-role.kubernetes.io/.

The current code panics when run against a ROSA cluster:
 {{ E0209 18:10:55.533265 78 runtime.go:79] Observed a panic: runtime.boundsError{x:24, y:23, signed:true, code:0x3} (runtime error: slice bounds out of range [24:23])
goroutine 233 [running]:
k8s.io/apimachinery/pkg/util/runtime.logPanic({0x7a71840?, 0xc0018e2f48})
k8s.io/apimachinery@v0.27.2/pkg/util/runtime/runtime.go:75 +0x99
k8s.io/apimachinery/pkg/util/runtime.HandleCrash({0x0, 0x0, 0x1000251f9fe?})
k8s.io/apimachinery@v0.27.2/pkg/util/runtime/runtime.go:49 +0x75
panic({0x7a71840, 0xc0018e2f48})
runtime/panic.go:884 +0x213
github.com/openshift/origin/pkg/monitortests/node/watchnodes.nodeRoles(0x7ecd7b3?)
github.com/openshift/origin/pkg/monitortests/node/watchnodes/node.go:187 +0x1e5
github.com/openshift/origin/pkg/monitortests/node/watchnodes.startNodeMonitoring.func1(0}}