Atchub/addprediagtotelemetry

src/connectedk8s/HISTORY.rst

@@ -2,6 +2,10 @@
 
 Release History
 ===============
+1.10.13
++++++
+* Added telemetry for pre-onboarding diagnostic results, including diagnostics execution failures.
+
 1.10.12
 +++++
 * Removed deprecated '--app-id' and '--app-secret' RBAC parameters from the extension.

src/connectedk8s/azext_connectedk8s/_constants.py

@@ -124,6 +124,11 @@
 PublicKey_Export_Fault_Type = "publickey-export-error"
 PrivateKey_Export_Fault_Type = "privatekey-export-error"
 Install_HelmRelease_Fault_Type = "helm-release-install-error"
+Install_Prediagnostics_Fault_Type = "prediagnostics-failure"
+Install_Prediagnostics_Job_Execution_Error_Fault_Type = (
+    "prediagnostics-job-execution-error"
+)
+Post_Diagnostic_Precheck_Fault_Type = "post-diagnostic-precheck-failure"
 Delete_HelmRelease_Fault_Type = "helm-release-delete-error"
 Check_PodStatus_Fault_Type = "check-pod-status-error"
 Kubernetes_Connectivity_FaultType = "kubernetes-cluster-connection-error"
@@ -418,7 +423,7 @@
 
 # Connect Precheck Diagnoser constants
 Cluster_Diagnostic_Checks_Job_Registry_Path = (
-    "azurearck8s/helmchart/stable/clusterdiagnosticchecks:1.31.2"
+    "azurearck8s/helmchart/stable/clusterdiagnosticchecks:1.33.0"
 )
 Cluster_Diagnostic_Checks_Helm_Install_Failed_Fault_Type = (
     "Error while installing cluster diagnostic checks helm release"
@@ -475,6 +480,10 @@
     "Outbound network connectivity check failed for Cluster Connect"
 )
 DNS_Check_Result_String = "DNS Result:"
+Entra_Connectivity_Check_Result_String = (
+    "Entra Authentication Endpoint Connectivity Check Result"
+)
+CRD_Ownership_Check_Failed_String = "Check Failed: CRD"
 AZ_CLI_ADAL_TO_MSAL_MIGRATE_VERSION = "2.30.0"
 CLIENT_PROXY_VERSION = "1.3.033581"
 CLIENT_PROXY_FOLDER = ".clientproxy"

src/connectedk8s/azext_connectedk8s/_utils.py

@@ -383,6 +383,9 @@ def check_cluster_DNS(
         if (
             "NXDOMAIN" in formatted_dns_log
             or "connection timed out" in formatted_dns_log
+            or "no servers could be reached" in formatted_dns_log
+            or "communications error" in formatted_dns_log
+            or "timed out" in formatted_dns_log
         ):
             logger.warning(
                 "Error: We found an issue with the DNS resolution on your cluster. For details about debugging DNS "

src/connectedk8s/azext_connectedk8s/custom.py

@@ -367,13 +367,20 @@ def create_connectedk8s(
                 filepath_with_timestamp, storage_space_available, 1
             )
 
+            if precheckutils.diagnoser_output:
+                print("\n--- Pre-onboarding Diagnostic Check Results ---")
+                for line in precheckutils.diagnoser_output:
+                    print(line.rstrip())
+                print("--- End of Diagnostic Check Results ---\n")
+
             if storage_space_available is False:
                 logger.warning(
                     "There is no storage space available on your device and hence not saving cluster "
                     "diagnostic check logs on your device"
                 )
 
     except Exception as e:
+        precheckutils.send_prediagnostic_job_execution_error_telemetry(reason=str(e))
         ex_msg = f"An exception occured while trying to execute pre-onboarding diagnostic checks : {e}"
         summ_msg = f"An exception occured while trying to execute pre-onboarding diagnostic checks : {e}"
         telemetry.set_exception(
@@ -401,6 +408,23 @@ def create_connectedk8s(
         and not azure_local_disconnected
         and not lowbandwidth
     ):
+        precheck_failure_summary = precheckutils.get_precheck_failure_summary()
+        precheck_failure_summary_msg = (
+            f" Details: {precheck_failure_summary}" if precheck_failure_summary else ""
+        )
+        if precheckutils.prediagnostic_job_execution_status == "Completed" or (
+            precheckutils.prediagnostic_job_execution_status == "NotCompleted"
+            and precheckutils.prediagnostic_dns_check != "NotApplicable"
+        ):
+            precheckutils.send_prediagnostic_check_failure_telemetry(
+                precheckutils.prediagnostic_dns_check,
+                precheckutils.prediagnostic_outbound_check,
+            )
+        elif precheckutils.prediagnostic_job_execution_status not in (
+            "Completed",
+            "NotCompleted",
+        ):
+            precheckutils.send_prediagnostic_job_execution_error_telemetry()
         if storage_space_available:
             logger.warning(
                 "The pre-check result logs logs have been saved at this path: "
@@ -418,6 +442,7 @@ def create_connectedk8s(
                 "meet the prerequisites - "
                 + consts.Doc_Onboarding_PreRequisites_Url
                 + " and try onboarding again."
+                + precheck_failure_summary_msg
             )
             raise ValidationError(err_msg)
 
@@ -430,6 +455,7 @@ def create_connectedk8s(
         err_msg = (
             "One or more pre-onboarding diagnostic checks failed and hence not proceeding with "
             "cluster onboarding. Please resolve them and try onboarding again."
+            + precheck_failure_summary_msg
         )
         raise ValidationError(err_msg)
 
@@ -450,6 +476,10 @@ def create_connectedk8s(
             fault_type=consts.Linux_Node_Not_Exists,
             summary="Couldn't find any node on the kubernetes cluster with the OS 'linux'",
         )
+        precheckutils.send_post_diagnostic_precheck_failure_telemetry(
+            check_name="LinuxNodeExists",
+            reason="Couldn't find any node on the kubernetes cluster with the OS 'linux'",
+        )
         logger.warning(
             "Please ensure that this Kubernetes cluster has any nodes with OS 'linux', for scheduling the "
             "Arc-Agents onto and connecting to Azure. Learn more at %s",
@@ -468,6 +498,10 @@ def create_connectedk8s(
             fault_type=consts.Cannot_Create_ClusterRoleBindings_Fault_Type,
             summary=summ_msg,
         )
+        precheckutils.send_post_diagnostic_precheck_failure_telemetry(
+            check_name="ClusterRoleBindings",
+            reason=ex_msg,
+        )
         err_msg = (
             "Your credentials doesn't have permission to create clusterrolebindings on this "
             "kubernetes cluster. Please check your permissions."