Make machineConfigLabels-related misconfiguration more visible

NTO Tuned CRs support machine config labels (mcLabels) based matching.
This involves finding a MachineConfigPool with machineConfigSelector
matching mcLabels and setting a TuneD profile on all nodes that are
assigned the found MachineConfigPool.  We do not support configurations,
where more than 1 MachineConfigPool matches the mcLabels.  While this
configuration is not supported, users often unwittingly configure
their clusters in this way.  In this case, NTO issues an error message
in the operator logs.

This is often not sufficient, so we make this misconfiguration more
visible by making ClusterOperator/node-tuning object Degraded.

Author: jmencak

pkg/operator/controller.go

@@ -87,6 +87,10 @@ type Controller struct {
 	// tracked as having kernel command-line conflict due to belonging
 	// to the same MCP.
 	bootcmdlineConflict map[string]bool
+
+	// mcLabelsAcrossMCP is the internal operator's cache of Profiles
+	// tracked as using machineConfigLabels that match across multiple MCPs.
+	mcLabelsAcrossMCP map[string]bool
 }
 
 type wqKey struct {
@@ -118,6 +122,7 @@ func NewController() (*Controller, error) {
 	}
 
 	controller.bootcmdlineConflict = map[string]bool{}
+	controller.mcLabelsAcrossMCP = map[string]bool{}
 
 	// Initial event to bootstrap CR if it doesn't exist.
 	controller.workqueue.AddRateLimited(wqKey{kind: wqKindTuned, name: tunedv1.TunedDefaultResourceName})
@@ -592,6 +597,7 @@ func (c *Controller) syncProfile(tuned *tunedv1.Tuned, nodeName string) error {
 
 	profileMf.Name = nodeName
 	delete(c.bootcmdlineConflict, nodeName)
+	delete(c.mcLabelsAcrossMCP, nodeName)
 	nodeLabels, err := c.pc.nodeLabelsGet(nodeName)
 	if err != nil {
 		// Remove Profiles for Nodes which no longer exist.
@@ -775,6 +781,7 @@ func (c *Controller) syncMachineConfig(labels map[string]string, profile *tunedv
 		// Log an error and do not requeue, this is a configuration issue.
 		klog.Errorf("profile %v uses machineConfigLabels that match across multiple MCPs (%v); this is not supported",
 			profile.Name, printMachineConfigPoolsNames(pools))
+		c.mcLabelsAcrossMCP[profile.Name] = true
 		return nil
 	}
 

pkg/operator/status.go

@@ -163,6 +163,20 @@ func (c *Controller) numProfilesWithBootcmdlineConflict(profileList []*tunedv1.P
 	return numConflict
 }
 
+// numMCLabelsAcrossMCP returns the total number
+// of Profiles in the internal operator's cache (mcLabelsAcrossMCP)
+// tracked as using machineConfigLabels that match across multiple MCPs.
+func (c *Controller) numMCLabelsAcrossMCP(profileList []*tunedv1.Profile) int {
+	n := 0
+	for _, profile := range profileList {
+		if c.mcLabelsAcrossMCP[profile.Name] {
+			n++
+		}
+	}
+
+	return n
+}
+
 // computeStatus computes the operator's current status.
 func (c *Controller) computeStatus(tuned *tunedv1.Tuned, conditions []configv1.ClusterOperatorStatusCondition) ([]configv1.ClusterOperatorStatusCondition, string, error) {
 	const (
@@ -318,6 +332,14 @@ func (c *Controller) computeStatus(tuned *tunedv1.Tuned, conditions []configv1.C
 			degradedCondition.Message = fmt.Sprintf("%v/%v Profiles with bootcmdline conflict", numConflict, len(profileList))
 		}
 
+		numMCLabelsAcrossMCP := c.numMCLabelsAcrossMCP(profileList)
+		if numMCLabelsAcrossMCP > 0 {
+			klog.Infof("%v/%v Profiles use machineConfigLabels that match across multiple MCPs", numMCLabelsAcrossMCP, len(profileList))
+			degradedCondition.Status = configv1.ConditionTrue
+			degradedCondition.Reason = "MCLabelsAcrossMCPs"
+			degradedCondition.Message = fmt.Sprintf("%v/%v Profiles use machineConfigLabels that match across multiple MCPs", numMCLabelsAcrossMCP, len(profileList))
+		}
+
 		// If the operator is not available for an extensive period of time, set the Degraded operator status.
 		conditions = clusteroperator.SetStatusCondition(conditions, &availableCondition)
 		now := metav1.Now().Unix()