FQDN check should pass/fail after retrying mutliple times to reach the node

components/automate-cli/pkg/verifyserver/constants/fqdn.go

@@ -1,20 +1,25 @@
 package constants
 
 const (
-	FQDN_TITLE                     = "FQDN is reachable"
-	FQDN_ERROR_MESSAGE             = "FQDN is not reachable"
-	FQDN_RESOLUTION_MESSAGE        = "Ensure FQDN is reachable and mapped to load balancer.Also, ensure your Port 443 is open and load balancer is able to reach to the machine on port 443. Review security group or firewall settings."
-	NODE_TITLE                     = "Nodes are reachable"
-	NODE_SUCCESS_MESSAGE           = "All nodes are reachable"
-	NODE_ERROR_MESSAGE             = "%v is not reachable"
-	NODE_RESOLUTION_MESSAGE        = "Ensure your Port 443 is open. Review security group or firewall settings."
-	CERTIFICATE_TITLE              = "Certificate validity for FQDN"
-	CERTIFICATE_SUCCESS_MESSAGE    = "FQDN has with valid certificates"
-	CERTIFICATE_ERROR_MESSAGE      = "FQDN certificate is not valid."
-	CERTIFICATE_RESOLUTION_MESSAGE = "Generate new valid certificates and provide those."
-	IP_TO_HASH_FAIL_MESSAGE        = "Failed to hash the IP."
-	DEFAULT_HTTPS_PORT             = "443"
-	SERVER_IP_HEADER_KEY           = "x-server-ip"
-	CHAN_RESULT_ERROR_MESSAGE      = "error recieved"
-	MIN_NUMBER_OF_CALLS            = 50
+	FQDN_TITLE                           = "FQDN is reachable"
+	FQDN_ERROR_MESSAGE                   = "FQDN is not reachable"
+	FQDN_RESOLUTION_MESSAGE              = "Ensure FQDN is reachable and mapped to load balancer. Also, ensure your Port 443 is open and load balancer is able to reach to the machine on port 443. Review security group or firewall settings."
+	CERT_CN_MISMATCH_RESOLUTION_MESSAGE  = "Ensure the certificate provided is Valid. In case of self-signed certificate, make sure the DNS provided in \"subjectAltName\" matches with \"CN\" (Common Name)"
+	CERT_CN_MISMATCH_ERROR_PATTERN       = "x509: certificate is not valid for any names, but wanted to match"
+	INVALID_FQDN_CERT_RESOLUTION_MESSAGE = "Ensure the certificate provided is Valid. In case of self-signed certificate, make sure the root-ca and the certificate provided in LB belongs to same CA"
+	INVALID_FQDN_CERT_ERROR_PATTERN      = "x509: certificate signed by unknown authority"
+	GENERIC_FQDN_CERT_RESOLUTION_MESSAGE = "Ensure the certificate provided is Valid. Also check if the FQDN is reachable, and mapped to load balancer. Review security group or firewall settings for the load-balance."
+	NODE_TITLE                           = "Nodes are reachable"
+	NODE_SUCCESS_MESSAGE                 = "All nodes are reachable"
+	NODE_ERROR_MESSAGE                   = "%v is not reachable"
+	NODE_RESOLUTION_MESSAGE              = "Ensure your Port 443 is open. Review security group or firewall settings of the node."
+	CERTIFICATE_TITLE                    = "Certificate validity for FQDN"
+	CERTIFICATE_SUCCESS_MESSAGE          = "FQDN has with valid certificates"
+	CERTIFICATE_ERROR_MESSAGE            = "FQDN certificate is not valid."
+	CERTIFICATE_RESOLUTION_MESSAGE       = "Generate new valid certificates and provide those."
+	IP_TO_HASH_FAIL_MESSAGE              = "Failed to hash the IP."
+	DEFAULT_HTTPS_PORT                   = "443"
+	SERVER_IP_HEADER_KEY                 = "x-server-ip"
+	CHAN_RESULT_ERROR_MESSAGE            = "error recieved"
+	MIN_NUMBER_OF_CALLS                  = 50
 )

components/automate-cli/pkg/verifyserver/server/api/v1/fqdn.go

@@ -37,6 +37,6 @@ func (h *Handler) CheckFqdn(c *fiber.Ctx) error {
 		return fiber.NewError(http.StatusBadRequest, "node_type should be automate or chef-infra-server, Please provide node_type.")
 	}
 
-	res := h.FqdnService.CheckFqdnReachability(*req, constants.DEFAULT_HTTPS_PORT, time.Minute*1)
+	res := h.FqdnService.CheckFqdnReachability(*req, constants.DEFAULT_HTTPS_PORT, time.Second*30)
 	return c.JSON(response.BuildSuccessResponse(res))
 }

components/automate-cli/pkg/verifyserver/services/batchcheckservice/trigger/triggerapi.go

@@ -121,7 +121,7 @@ func TriggerCheckAPI(endPoint, host, nodeType, method string, output chan<- mode
 	}
 
 	client := http.Client{
-		Timeout: 35 * time.Second,
+		Timeout: 65 * time.Second,
 	}
 	resp, err := client.Do(req)
 	if err != nil {

components/automate-cli/pkg/verifyserver/services/fqdnservice/fqdnservice.go

@@ -149,25 +149,45 @@ func (fq *FqdnService) MakeConcurrentCalls(url string, client *http.Client, setN
 	return errors.New("nodes are not reachable")
 }
 
-func (fq *FqdnService) triggerRequest(client *http.Client, url string) error {
-	res, err := client.Get(url)
-	if err != nil {
-		fq.log.Error(err.Error())
-		return err
-	}
+func (fq *FqdnService) triggerRequest(client *http.Client, url string, duration time.Duration) error {
+	timeout := time.After(duration)
+
+	var fqdnError error
+loop:
+	for {
+		select {
+		case <-timeout:
+			fq.log.Debugf("Stopped making API calls after %v seconds.", duration)
+			if fqdnError == nil {
+				return errors.New(constants.FQDN_ERROR_MESSAGE)
+			}
+			return fqdnError
+		default:
+			res, err := client.Get(url)
+			if err != nil {
+				fq.log.Error("Iteration logs: ", err.Error())
+				fqdnError = err
+				time.Sleep(3 * time.Second)
+				continue
+			}
+			fq.log.Debug("Status Code: ", res.StatusCode)
+			if res.StatusCode != 200 {
+				fq.log.Errorf("%v is not reachable.", url)
+			}
+			if res.StatusCode == 200 {
+				fq.log.Debug("FQDN is reachable")
+				// return nil
+				break loop
+			}
+		}
+		time.Sleep(3 * time.Second)
 
-	fq.log.Debug("Status Code: ", res.StatusCode)
-	if res.StatusCode != 200 {
-		fq.log.Debugf("%v is not reachable.", url)
-		return errors.New("fqdn is not reachable")
 	}
-
-	fq.log.Debug("Fqdn is Reachable.")
 	return nil
 }
 
 // fqdnReachable function will check that are we able to hit the load balancer fqdn or not.
-func (fq *FqdnService) fqdnReachable(fqdn, rootCert, nodeType string, isAfterDeployment bool, port string) models.Checks {
+func (fq *FqdnService) fqdnReachable(fqdn, rootCert, nodeType string, isAfterDeployment bool, port string, duration time.Duration) models.Checks {
 	fq.log.Debug("Checking Fqdn Reachability...")
 	client := fq.createClient(rootCert)
 	var url string
@@ -179,9 +199,17 @@ func (fq *FqdnService) fqdnReachable(fqdn, rootCert, nodeType string, isAfterDep
 	}
 	fq.log.Debug("URL: ", url)
 
-	err := fq.triggerRequest(client, url)
+	err := fq.triggerRequest(client, url, duration)
 	if err != nil {
-		return createCheck(constants.FQDN_TITLE, false, "", constants.FQDN_ERROR_MESSAGE, constants.FQDN_RESOLUTION_MESSAGE)
+		resoultion_message := ""
+		if strings.Contains(err.Error(), constants.CERT_CN_MISMATCH_ERROR_PATTERN) {
+			resoultion_message = constants.CERT_CN_MISMATCH_RESOLUTION_MESSAGE
+		} else if strings.Contains(err.Error(), constants.INVALID_FQDN_CERT_ERROR_PATTERN) {
+			resoultion_message = constants.INVALID_FQDN_CERT_RESOLUTION_MESSAGE
+		} else {
+			resoultion_message = constants.GENERIC_FQDN_CERT_RESOLUTION_MESSAGE
+		}
+		return createCheck(constants.FQDN_TITLE, false, "", err.Error(), resoultion_message)
 	}
 	return createCheck(constants.FQDN_TITLE, true, constants.FQDN_TITLE, "", "")
 }
@@ -266,7 +294,7 @@ func (fq *FqdnService) CheckFqdnReachability(req models.FqdnRequest, port string
 	}
 
 	if certificateValidityCheck {
-		check = fq.fqdnReachable(req.Fqdn, req.RootCert, req.NodeType, req.IsAfterDeployment, port)
+		check = fq.fqdnReachable(req.Fqdn, req.RootCert, req.NodeType, req.IsAfterDeployment, port, duration)
 	} else {
 		check = createCheck(constants.FQDN_TITLE, false, "", constants.FQDN_ERROR_MESSAGE, constants.FQDN_RESOLUTION_MESSAGE)
 	}

components/automate-cli/pkg/verifyserver/services/fqdnservice/fqdnservice_test.go

@@ -325,8 +325,8 @@ func TestCheckFqdnReachability(t *testing.T) {
 						Title:         constants.FQDN_TITLE,
 						Passed:        false,
 						SuccessMsg:    "",
-						ErrorMsg:      constants.FQDN_ERROR_MESSAGE,
-						ResolutionMsg: constants.FQDN_RESOLUTION_MESSAGE,
+						ErrorMsg:      "Get \"https://localhost2:5345\": dial tcp: lookup localhost2: no such host",
+						ResolutionMsg: constants.GENERIC_FQDN_CERT_RESOLUTION_MESSAGE,
 					},
 					{
 						Title:         constants.NODE_TITLE,
@@ -545,8 +545,8 @@ func TestCheckFqdnReachability(t *testing.T) {
 						Title:         constants.FQDN_TITLE,
 						Passed:        false,
 						SuccessMsg:    "",
-						ErrorMsg:      constants.FQDN_ERROR_MESSAGE,
-						ResolutionMsg: constants.FQDN_RESOLUTION_MESSAGE,
+						ErrorMsg:      "Get \"https://localhost2:5345/_status\": dial tcp: lookup localhost2: no such host",
+						ResolutionMsg: constants.GENERIC_FQDN_CERT_RESOLUTION_MESSAGE,
 					},
 					{
 						Title:         constants.NODE_TITLE,
@@ -609,8 +609,8 @@ func TestCheckFqdnReachability(t *testing.T) {
 						Title:         constants.FQDN_TITLE,
 						Passed:        false,
 						SuccessMsg:    "",
-						ErrorMsg:      constants.FQDN_ERROR_MESSAGE,
-						ResolutionMsg: constants.FQDN_RESOLUTION_MESSAGE,
+						ErrorMsg:      "Get \"https://localhost2:5345\": dial tcp: lookup localhost2: no such host",
+						ResolutionMsg: constants.GENERIC_FQDN_CERT_RESOLUTION_MESSAGE,
 					},
 					{
 						Title:         constants.NODE_TITLE,
@@ -713,7 +713,7 @@ func TestCheckFqdnReachability(t *testing.T) {
 						Passed:        false,
 						SuccessMsg:    "",
 						ErrorMsg:      constants.FQDN_ERROR_MESSAGE,
-						ResolutionMsg: constants.FQDN_RESOLUTION_MESSAGE,
+						ResolutionMsg: constants.GENERIC_FQDN_CERT_RESOLUTION_MESSAGE,
 					},
 					{
 						Title:         constants.NODE_TITLE,