skip reachability test in airgapped (#31)

* disable some prereq tests in airgapped * disable some prereq tests in airgapped * disable some prereq tests in airgapped * disable some prereq tests in airgapped * disable some prereq tests in airgapped * PR comments
run-ai · Aug 4, 2024 · 219cdd6 · 219cdd6
1 parent cd0e059
commit 219cdd6
Show file tree

Hide file tree

Showing 9 changed files with 92 additions and 62 deletions.
diff --git a/README.md b/README.md
@@ -47,7 +47,7 @@ chmod +x ./preinstall-diagnostics-darwin-arm64 && \
       --domain ${CONTROL_PLANE_FQDN} \
       --cluster-domain ${CLUSTER_FQDN} \
       --image-pull-secret ${IMAGE_PULL_SECRET_NAME} \
-      --image ${PRIVATE_REGISTRY_IMAGE_URL}    
+      --image ${PRIVATE_REGISTRY_IMAGE_URL} --airgapped    
 ```
 
 ### Example
@@ -88,6 +88,9 @@ Usage of ./preinstall-diagnostics-darwin-arm64:
     	URL the Run:AI service to check connectivity to (default "https://app.run.ai")
   -version
     	Prints the binary version
+  -airgapped
+    	skip reachability checks to external servers 
+
 ```
 
 ## Build

diff --git a/cmd/preinstall-diagnostics/main.go b/cmd/preinstall-diagnostics/main.go
@@ -25,6 +25,7 @@ const (
 	runaiSaasArgName              = "saas-address"
 	outputArgName                 = "output"
 	versionArgName                = "version"
+	airgappedArgName              = "airgapped"
 )
 
 const (
@@ -33,19 +34,18 @@ const (
 
 var (
 	runInternalClusterTests bool
-
-	clean                  bool
-	backendDomainFQDN      string
-	clusterDomainFQDN      string
-	image                  string
-	imagePullSecretName    string
-	dryRun                 bool
-	runaiContainerRegistry string
-	runaiSaas              string
-	output                 string
-	version                bool
-
-	outputFile *os.File
+	clean                   bool
+	backendDomainFQDN       string
+	clusterDomainFQDN       string
+	image                   string
+	imagePullSecretName     string
+	dryRun                  bool
+	runaiContainerRegistry  string
+	runaiSaas               string
+	output                  string
+	version                 bool
+	airgapped               bool
+	outputFile              *os.File
 )
 
 func init() {
@@ -62,7 +62,7 @@ func init() {
 	flag.StringVar(&runaiSaas, runaiSaasArgName, saas.RunAISaasAddress, "URL the Run:AI service to check connectivity to")
 	flag.StringVar(&output, outputArgName, defaultOutputFileName, "File to save the output to")
 	flag.BoolVar(&version, versionArgName, false, "Prints the binary version")
-
+	flag.BoolVar(&airgapped, airgappedArgName, false, "skip tests that require network access")
 	flag.Parse()
 }
 
@@ -84,6 +84,6 @@ func main() {
 		logger := log.NewLogger(outputFile)
 
 		cli.Main(clean, dryRun, backendDomainFQDN, clusterDomainFQDN, image,
-			imagePullSecretName, runaiContainerRegistry, runaiSaas, version, logger)
+			imagePullSecretName, runaiContainerRegistry, runaiSaas, version, airgapped, logger)
 	}
 }
diff --git a/internal/cmd/cli/cli.go b/internal/cmd/cli/cli.go
@@ -21,14 +21,17 @@ import (
 )
 
 func Main(clean, dryRun bool, backendFQDN, clusterFQDN, image, imagePullSecretName, imageRegistry,
-	runaiSaas string, version bool, logger *log.Logger) {
+	runaiSaas string, version bool, airgapped bool, logger *log.Logger) {
+	if airgapped {
+		fmt.Println("airgapped")
+	}
 	if version {
 		fmt.Println(ver.Version)
 		return
 	}
 
 	creationOrder, deletionOrder := resources.TemplateResources(backendFQDN, image,
-		imagePullSecretName, imageRegistry, runaiSaas)
+		imagePullSecretName, imageRegistry, runaiSaas, airgapped)
 
 	if dryRun {
 		err := resources.PrintResources(creationOrder)

diff --git a/internal/cmd/job/internal_cluster_tests.go b/internal/cmd/job/internal_cluster_tests.go
@@ -2,13 +2,14 @@ package job
 
 import (
 	v2 "github.com/run-ai/preinstall-diagnostics/internal"
+	"github.com/run-ai/preinstall-diagnostics/internal/env"
 	internal_cluster_tests2 "github.com/run-ai/preinstall-diagnostics/internal/internal-cluster-tests"
 	"github.com/run-ai/preinstall-diagnostics/internal/log"
+	"strconv"
 )
 
 func runTestsAndAppendResults(logger *log.Logger) []v2.TestResult {
 	var testResults []v2.TestResult
-
 	osInfo, err := internal_cluster_tests2.ShowOSInfo()
 	if err != nil {
 		testResults = append(testResults, v2.TestResult{
@@ -61,51 +62,56 @@ func runTestsAndAppendResults(logger *log.Logger) []v2.TestResult {
 		})
 	}
 
-	reachable, err := internal_cluster_tests2.RunAIAuthProviderReachable()
+	airgapped, err := strconv.ParseBool(env.EnvOrDefault(env.AirgappedEnvVar, "false"))
 	if err != nil {
-		testResults = append(testResults, v2.TestResult{
-			Name:    "RunAI Auth Provider Reachable",
-			Result:  false,
-			Message: err.Error(),
-		})
-	} else {
-		testResults = append(testResults, v2.TestResult{
-			Name:    "RunAI Auth Provider Reachable",
-			Result:  reachable,
-			Message: "",
-		})
+		airgapped = false
 	}
+	if !airgapped {
+		reachable, err := internal_cluster_tests2.RunAIAuthProviderReachable()
+		if err != nil {
+			testResults = append(testResults, v2.TestResult{
+				Name:    "RunAI Auth Provider Reachable",
+				Result:  false,
+				Message: err.Error(),
+			})
+		} else {
+			testResults = append(testResults, v2.TestResult{
+				Name:    "RunAI Auth Provider Reachable",
+				Result:  reachable,
+				Message: "",
+			})
+		}
 
-	reachable, err = internal_cluster_tests2.RunAIPrometheusReachable()
-	if err != nil {
-		testResults = append(testResults, v2.TestResult{
-			Name:    "RunAI Prometheus Reachable",
-			Result:  false,
-			Message: err.Error(),
-		})
-	} else {
-		testResults = append(testResults, v2.TestResult{
-			Name:    "RunAI Prometheus Reachable",
-			Result:  reachable,
-			Message: "",
-		})
-	}
+		reachable, err = internal_cluster_tests2.RunAIPrometheusReachable()
+		if err != nil {
+			testResults = append(testResults, v2.TestResult{
+				Name:    "RunAI Prometheus Reachable",
+				Result:  false,
+				Message: err.Error(),
+			})
+		} else {
+			testResults = append(testResults, v2.TestResult{
+				Name:    "RunAI Prometheus Reachable",
+				Result:  reachable,
+				Message: "",
+			})
+		}
 
-	reachable, err = internal_cluster_tests2.RunAIBackendReachable()
-	if err != nil {
-		testResults = append(testResults, v2.TestResult{
-			Name:    "RunAI Backend Reachable",
-			Result:  false,
-			Message: err.Error(),
-		})
-	} else {
-		testResults = append(testResults, v2.TestResult{
-			Name:    "RunAI Backend Reachable",
-			Result:  reachable,
-			Message: "",
-		})
+		reachable, err = internal_cluster_tests2.RunAIBackendReachable()
+		if err != nil {
+			testResults = append(testResults, v2.TestResult{
+				Name:    "RunAI Backend Reachable",
+				Result:  false,
+				Message: err.Error(),
+			})
+		} else {
+			testResults = append(testResults, v2.TestResult{
+				Name:    "RunAI Backend Reachable",
+				Result:  reachable,
+				Message: "",
+			})
+		}
 	}
-
 	err = internal_cluster_tests2.CheckNodeConnectivity(logger)
 	if err != nil {
 		testResults = append(testResults, v2.TestResult{

diff --git a/internal/cmd/job/job.go b/internal/cmd/job/job.go
@@ -3,7 +3,6 @@ package job
 import (
 	"context"
 	"encoding/json"
-
 	v2 "github.com/run-ai/preinstall-diagnostics/internal"
 	"github.com/run-ai/preinstall-diagnostics/internal/env"
 	"github.com/run-ai/preinstall-diagnostics/internal/k8sclient"

diff --git a/internal/env/env.go b/internal/env/env.go
@@ -17,6 +17,7 @@ const (
 	RegistryEnvVar = "REGISTRY"
 
 	RunAISaasEnvVar = "RUNAI_SAAS"
+	AirgappedEnvVar = "AIRGAPPED"
 )
 
 func EnvOrError(envVar string) (string, error) {

diff --git a/internal/internal-cluster-tests/nodes.go b/internal/internal-cluster-tests/nodes.go
@@ -112,6 +112,9 @@ func WaitForJobPodsToBeRunning(logger *log.Logger) error {
 		logger.LogF("waiting for pods to be available...")
 		availablePods := 0
 		for _, job := range jobs.Items {
+			if job.Status.Ready == nil {
+				continue
+			}
 			availablePods += int(*job.Status.Ready)
 		}
 

diff --git a/internal/resources/functions.go b/internal/resources/functions.go
@@ -93,7 +93,7 @@ func CreateResources(objs []client.Object, kubeDynamicClient dynamic.Interface)
 }
 
 func TemplateResources(backendFQDN, image, imagePullSecretName,
-	imageRegistry, runaiSaas string) (creationOrder, deletionOrder []client.Object) {
+	imageRegistry, runaiSaas string, airgapped bool) (creationOrder, deletionOrder []client.Object) {
 	creationOrder = []client.Object{}
 
 	k8s, err := k8sclient.ClientSet()
@@ -121,6 +121,21 @@ func TemplateResources(backendFQDN, image, imagePullSecretName,
 				},
 			}
 		}
+		if airgapped {
+			job.Spec.Template.Spec.Containers[0].Env =
+				append(job.Spec.Template.Spec.Containers[0].Env,
+					v1.EnvVar{
+						Name:  env.AirgappedEnvVar,
+						Value: "true",
+					})
+		} else {
+			job.Spec.Template.Spec.Containers[0].Env =
+				append(job.Spec.Template.Spec.Containers[0].Env,
+					v1.EnvVar{
+						Name:  env.AirgappedEnvVar,
+						Value: "false",
+					})
+		}
 
 		if imageRegistry != "" {
 			job.Spec.Template.Spec.Containers[0].Env =

diff --git a/makefile b/makefile
@@ -2,7 +2,7 @@ OUT_DIR=_out
 BIN=preinstall-diagnostics
 
 REGISTRY=gcr.io/run-ai-lab
-VERSION=v2.16.19
+VERSION=v2.18.14
 
 IMAGE_DOCKER_FILE=cmd/preinstall-diagnostics/Dockerfile
 IMAGE=$(REGISTRY)/preinstall-diagnostics:$(VERSION)