Fixing breaking unit test with proposed changes to IsConsideredHealthy and getWorkerPeersResponse for issue #251

Author: mark-dell-usa

pkg/apicheck/check.go

@@ -168,8 +168,10 @@ func (c *ApiConnectivityCheck) Start(ctx context.Context) error {
 }
 
 // isConsideredHealthy keeps track of the number of errors reported, and when a certain amount of error occur within a certain
-// time, ask peers if this node is healthy. Returns if the node is considered to be healthy or not.
+// time, ask peers if this node is healthy. Returns if the node is considered to be healthy or not.  It is usable
+// whether this is a control plane node or a worker node
 func (c *ApiConnectivityCheck) isConsideredHealthy() bool {
+
 	isControlPlaneManagerNil := c.controlPlaneManager == nil
 
 	isWorkerNode := isControlPlaneManagerNil || !c.controlPlaneManager.IsControlPlane()
@@ -178,31 +180,52 @@ func (c *ApiConnectivityCheck) isConsideredHealthy() bool {
 		"isControlPlaneManagerNil", isControlPlaneManagerNil,
 		"isWorkerNode", isWorkerNode)
 
-	workerPeersResponse := c.getWorkerPeersResponse()
+	var peersResponse peers.Response
 
 	if isWorkerNode {
-		c.config.Log.Info("isConsideredHealthy: returning result from getWorkerPeersResponse",
-			"workerPeersResponse.IsHealthy", workerPeersResponse.IsHealthy)
-		return workerPeersResponse.IsHealthy
+		peersResponse = c.getPeersResponse(peers.Worker)
+
 	} else {
-		canOtherControlPlanesBeReached := c.canOtherControlPlanesBeReached()
-		isControlPlaneHealthy := c.controlPlaneManager.IsControlPlaneHealthy(workerPeersResponse, canOtherControlPlanesBeReached)
-		c.config.Log.Info("isConsideredHealthy: returning result from IsControlPlaneHealthy",
-			"c.canOtherControlPlanesBeReached()", canOtherControlPlanesBeReached,
-			"c.controlPlaneManager.IsControlPlaneHealthy", isControlPlaneHealthy)
-		return isControlPlaneHealthy
+		// MES: On a control plane node, when it can reach the other nodes, and the control plane is healthy,
+		//	that's overriding any sort of peer response from them
+
+		//canOtherControlPlanesBeReached := c.canOtherControlPlanesBeReached()
+		peersResponse = c.getPeersResponse(peers.ControlPlane)
+
+		// MES: This does not appear to have any actual relevance.  To me, it appears that all the necessary
+		//	logic is contained in getPeersResponse(peers.ControlPlane)
+		//isControlPlaneHealthy := c.controlPlaneManager.IsControlPlaneHealthy(controlPlanePeersResponse,
+		//	canOtherControlPlanesBeReached)
+
+		//c.config.Log.Info("isConsideredHealthy: returning result from IsControlPlaneHealthy",
+		//	"c.canOtherControlPlanesBeReached()", canOtherControlPlanesBeReached,
+		//	"c.controlPlaneManager.IsControlPlaneHealthy", isControlPlaneHealthy)
+		//return isControlPlaneHealthy
+
+		//c.config.Log.Info("isConsideredHealthy: returning result from getPeersResponse",
+		//	"workerPeersResponse.IsHealthy", workerPeersResponse.IsHealthy)
+		//return workerPeersResponse.IsHealthy
 	}
 
+	c.config.Log.Info("isConsideredHealthy: returning result from getPeersResponse",
+		"workerPeersResponse.IsHealthy", peersResponse.IsHealthy)
+
+	return peersResponse.IsHealthy
+
 }
 
-func (c *ApiConnectivityCheck) getWorkerPeersResponse() peers.Response {
+func (c *ApiConnectivityCheck) getPeersResponse(role peers.Role) peers.Response {
 	c.errorCount++
 	if c.errorCount < c.config.MaxErrorsThreshold {
 		c.config.Log.Info("Ignoring api-server error, error count below threshold", "current count", c.errorCount, "threshold", c.config.MaxErrorsThreshold)
 		return peers.Response{IsHealthy: true, Reason: peers.HealthyBecauseErrorsThresholdNotReached}
 	}
+	c.config.Log.Info("Error count was above threshold, we will continue and attempt to get the addressess" +
+		" for our peers, I consider myself a WORKER at the moment")
 
-	peersToAsk := c.config.Peers.GetPeersAddresses(peers.Worker)
+	// MES: This gets called even if the current node is a control plane node.  Hopefully
+	//	in an actual environment it is returning actual worker peers
+	peersToAsk := c.config.Peers.GetPeersAddresses(role)
 
 	c.config.Log.Info("Error count exceeds threshold, trying to ask other peer nodes if I'm healthy",
 		"minPeersRequired", c.config.MinPeersForRemediation, "actualNumPeersFound", len(peersToAsk))