Merge branch 'main' into egomes/new-gateway-management

kurtosis-tech · Sep 26, 2024 · 34ba9ba · 34ba9ba
2 parents a7dd50f + 54bdc21
commit 34ba9ba
Show file tree

Hide file tree

Showing 4 changed files with 137 additions and 26 deletions.
diff --git a/.github/workflows/ci-e2e-tests.yml b/.github/workflows/ci-e2e-tests.yml
@@ -88,36 +88,76 @@ jobs:
 
       - name: Deploy Kardinal manager
         run: |
-          KARDINAL_CLI_DEV_MODE=TRUE /tmp/kardinal-cli manager deploy local-minikube
+          KARDINAL_CLI_DEV_MODE=TRUE /tmp/kardinal-cli manager deploy local-kardinal-kontrol
 
+          # Check that the three kardinal manager service pods are running and ready
+          while [ $(kubectl get pods -n default --no-headers -o custom-columns=NAMESPACE:metadata.namespace,POD:metadata.name,PodIP:status.podIP,READY-true:status.containerStatuses[*].ready | grep "true" | wc -l) -ne 3 ]
+          do
+              echo "Waiting for kardinal manager pods to run..."
+              kubectl get pods -n default -o custom-columns=NAMESPACE:metadata.namespace,POD:metadata.name,PodIP:status.podIP,READY-true:status.containerStatuses[*].ready
+              ((c++)) && ((c==12)) && exit 1
+              sleep 10
+          done
+          apps=$(kubectl get pods -n default -o custom-columns=:metadata.labels.app | tr " " "\n" | sort -g | tr "\n" " " | xargs)
+          echo ${apps}
+          if [ "${apps}" != "kardinal-manager trace-router trace-router-redis" ]; then exit 1; fi
+          
+          # Check for errors in the kardinal manager logs
+          if kubectl logs -n default -l dev.kardinal.app-id=kardinal-manager | grep -q "ERRO"
+          then
+            echo "Errors found in the kardinal manager logs"
+            kubectl logs -n default -l dev.kardinal.app-id=kardinal-manager | grep "ERRO"
+          fi
+ 
       - name: Deploy boutique demo manifest
         run: |
           KARDINAL_CLI_DEV_MODE=TRUE /tmp/kardinal-cli deploy -k ci/obd-demo.yaml
       
       - name: Validate that Kardinal manager applied the changes
         run: |
           # Check that the four baseline service pods are running and ready
-          while [ $(kubectl get pods -n baseline -o custom-columns=NAMESPACE:metadata.namespace,POD:metadata.name,PodIP:status.podIP,READY-true:status.containerStatuses[*].ready | grep "true,true" | wc -l) -ne 4 ]
+          while [ $(kubectl get pods -n baseline --no-headers -o custom-columns=NAMESPACE:metadata.namespace,POD:metadata.name,PodIP:status.podIP,READY-true:status.containerStatuses[*].ready | grep "true,true" | wc -l) -ne 4 ]
           do
               echo "Waiting for baseline pods to run..."
               kubectl get pods -n baseline -o custom-columns=NAMESPACE:metadata.namespace,POD:metadata.name,PodIP:status.podIP,READY-true:status.containerStatuses[*].ready
               ((c++)) && ((c==12)) && exit 1
               sleep 10
           done
-          kubectl get pods -n baseline
-
-          # Check the right services are running.
           apps=$(kubectl get pods -n baseline -o custom-columns=:metadata.labels.app | tr " " "\n" | sort -g | tr "\n" " " | xargs)
           echo ${apps}
           if [ "${apps}" != "cartservice frontend postgres productcatalogservice" ]; then exit 1; fi
         
           # Check for errors in the kardinal manager logs
-          if kubectl logs -n default -l dev.kardinal.app-id=kardinal-manager | grep "ERRO"
+          if kubectl logs -n default -l dev.kardinal.app-id=kardinal-manager | grep -q "ERRO"
           then
-            echo "Errors found in the kardinal manager"
+            echo "Errors found in the kardinal manager logs"
             kubectl logs -n default -l dev.kardinal.app-id=kardinal-manager | grep "ERRO"
           fi
 
+      - name: Start gateway to the baseline flow and validate access
+        run: |
+          KARDINAL_CLI_DEV_MODE=TRUE /tmp/kardinal-cli gateway baseline > kardinal.out &
+          cli_pid=$!
+          while ! grep "Proxy server for host" kardinal.out
+          do
+              echo "Waiting for gateway to start..."
+              cat kardinal.out
+              ((c++)) && ((c==12)) && exit 1
+              sleep 10
+          done
+          flow_url=$(grep "http" kardinal.out | awk '{print $NF}')
+          status_code=$(curl -s -o /dev/null -w "%{http_code}" ${flow_url})
+          echo "${flow_url} returned status code ${status_code}"
+          if [ "${status_code}" != "200" ]; then exit 1; fi
+          kill ${cli_pid}
+
+          # Check for errors in the trace router logs
+          if kubectl logs -n default -l app=trace-router | grep -q "ERRO"
+          then
+            echo "Errors found in the trace router logs"
+            kubectl logs -n default -l app=trace-router | grep "ERRO"
+          fi
+
       - name: Create, validate and delete flow
         run: |
           KARDINAL_CLI_DEV_MODE=TRUE /tmp/kardinal-cli flow create frontend kurtosistech/frontend:demo-frontend > kardinal.out
@@ -126,21 +166,54 @@ jobs:
           KARDINAL_CLI_DEV_MODE=TRUE /tmp/kardinal-cli flow ls | grep ${flow_id}
           
           # Check that the dev service pod is running and ready
-          while ! kubectl get pods -n baseline -o custom-columns=NAMESPACE:metadata.namespace,POD:metadata.name,PodIP:status.podIP,READY-true:status.containerStatuses[*].ready | grep true | grep "frontend-${flow_id}"
+          while ! kubectl get pods -n baseline --no-headers -o custom-columns=NAMESPACE:metadata.namespace,POD:metadata.name,PodIP:status.podIP,READY-true:status.containerStatuses[*].ready | grep true | grep "frontend-${flow_id}"
           do
               echo "Waiting for dev frontend pod to run..."
               kubectl get pods -n baseline -o custom-columns=NAMESPACE:metadata.namespace,POD:metadata.name,PodIP:status.podIP,READY-true:status.containerStatuses[*].ready
               ((c++)) && ((c==12)) && exit 1
               sleep 10
           done
           kubectl get pods -n baseline
+         
+          # Start gateway to the flow and validate access
+          KARDINAL_CLI_DEV_MODE=TRUE /tmp/kardinal-cli gateway ${flow_id} > kardinal.out &
+          cli_pid=$!
+          while ! grep "Proxy server for host" kardinal.out
+          do
+              echo "Waiting for gateway to start..."
+              cat kardinal.out
+              ((c++)) && ((c==12)) && exit 1
+              sleep 10
+          done
+          flow_url=$(grep "http" kardinal.out | awk '{print $NF}')
+          status_code=$(curl -s -o /dev/null -w "%{http_code}" ${flow_url})
+          echo "${flow_url} returned status code ${status_code}"
+          if [ "${status_code}" != "200" ]; then exit 1; fi
+          kill ${cli_pid}
           
+          # Check for errors in the trace router logs
+          if kubectl logs -n default -l app=trace-router | grep -q "ERRO"
+          then
+            echo "Errors found in the trace router logs"
+            kubectl logs -n default -l app=trace-router | grep "ERRO"
+          fi
+ 
           KARDINAL_CLI_DEV_MODE=TRUE /tmp/kardinal-cli flow delete ${flow_id}
-        
+
+          # Check that the dev service pod is not running anymore
+          while kubectl get pods -n baseline | grep "frontend-${flow_id}"
+          do
+              echo "Waiting for dev frontend pod to terminate..."
+              kubectl get pods -n baseline
+              ((c++)) && ((c==12)) && exit 1
+              sleep 10
+          done
+          kubectl get pods -n baseline
+
           # Check for errors in the kardinal manager logs
-          if kubectl logs -n default -l dev.kardinal.app-id=kardinal-manager | grep "ERRO"
+          if kubectl logs -n default -l dev.kardinal.app-id=kardinal-manager | grep -q "ERRO"
           then
-            echo "Errors found in the kardinal manager"
+            echo "Errors found in the kardinal manager logs"
             kubectl logs -n default -l dev.kardinal.app-id=kardinal-manager | grep "ERRO"
           fi
 
@@ -157,7 +230,7 @@ jobs:
           KARDINAL_CLI_DEV_MODE=TRUE /tmp/kardinal-cli flow ls | grep ${flow_id}
           
           # Check that the dev service pod is running and ready
-          while ! kubectl get pods -n baseline -o custom-columns=NAMESPACE:metadata.namespace,POD:metadata.name,PodIP:status.podIP,READY-true:status.containerStatuses[*].ready | grep true | grep "frontend-${flow_id}"
+          while ! kubectl get pods -n baseline --no-headers -o custom-columns=NAMESPACE:metadata.namespace,POD:metadata.name,PodIP:status.podIP,READY-true:status.containerStatuses[*].ready | grep true | grep "frontend-${flow_id}"
           do
               echo "Waiting for dev frontend pod to run..."
               kubectl get pods -n baseline -o custom-columns=NAMESPACE:metadata.namespace,POD:metadata.name,PodIP:status.podIP,READY-true:status.containerStatuses[*].ready
@@ -168,10 +241,20 @@ jobs:
 
           KARDINAL_CLI_DEV_MODE=TRUE /tmp/kardinal-cli flow delete ${flow_id}
 
+          # Check that the dev service pod is not running anymore
+          while kubectl get pods -n baseline | grep "frontend-${flow_id}"
+          do
+              echo "Waiting for dev frontend pod to terminate..."
+              kubectl get pods -n baseline
+              ((c++)) && ((c==12)) && exit 1
+              sleep 10
+          done
+          kubectl get pods -n baseline
+
           # Check for errors in the kardinal manager logs
-          if kubectl logs -n default -l dev.kardinal.app-id=kardinal-manager | grep "ERRO"
+          if kubectl logs -n default -l dev.kardinal.app-id=kardinal-manager | grep -q "ERRO"
           then
-            echo "Errors found in the kardinal manager"
+            echo "Errors found in the kardinal manager logs"
             kubectl logs -n default -l dev.kardinal.app-id=kardinal-manager | grep "ERRO"
           fi
       
@@ -196,8 +279,23 @@ jobs:
           kubectl get pods -n baseline
           
           # Check for errors in the kardinal manager logs
-          if kubectl logs -n default -l dev.kardinal.app-id=kardinal-manager | grep "ERRO"
+          if kubectl logs -n default -l dev.kardinal.app-id=kardinal-manager | grep -q "ERRO"
           then
-            echo "Errors found in the kardinal manager"
+            echo "Errors found in the kardinal manager logs"
             kubectl logs -n default -l dev.kardinal.app-id=kardinal-manager | grep "ERRO"
           fi
+      
+      - name: Remove kardinal manager
+        run: |
+          KARDINAL_CLI_DEV_MODE=TRUE /tmp/kardinal-cli manager remove > kardinal.out
+
+          # Check that the services have been terminated
+          while [ "$(kubectl get pods -n default 2>&1 >/dev/null)" != "No resources found in default namespace." ]
+          do
+              echo "Waiting for the services to terminate..."
+              kubectl get pods -n default
+              ((c++)) && ((c==12)) && exit 1
+              sleep 10
+          done
+          kubectl get pods -n default
+ 
diff --git a/kardinal-cli/cmd/root.go b/kardinal-cli/cmd/root.go
@@ -480,9 +480,10 @@ var topologyManifestCmd = &cobra.Command{
 }
 
 var deployManagerCmd = &cobra.Command{
-	Use:       fmt.Sprintf("deploy [kontrol location] accepted values: %s and %s ", kontrol.KontrolLocationLocalMinikube, kontrol.KontrolLocationKloudKontrol),
+	Use:       fmt.Sprintf("deploy [kardinal-kontrol service location] accepted values: %s and %s ", kontrol.KontrolLocationLocal, kontrol.KontrolLocationKloud),
 	Short:     "Deploy Kardinal manager into the cluster",
-	ValidArgs: []string{kontrol.KontrolLocationLocalMinikube, kontrol.KontrolLocationKloudKontrol},
+	Long:      "The Kardinal Manager retrieves the latest configuration from the Kardinal Kontrol service and applies changes to the user K8S topology.  The Kardinal Kontrol service can be the one running in the Kardinal Cloud or can be the one deployed locally.",
+	ValidArgs: []string{kontrol.KontrolLocationLocal, kontrol.KontrolLocationKloud},
 	Args:      cobra.MatchAll(cobra.ExactArgs(1), cobra.OnlyValidArgs),
 	Run: func(cmd *cobra.Command, args []string) {
 		kontrolLocation := args[0]
@@ -1122,10 +1123,10 @@ func getKontrolBaseURLForManager() (string, error) {
 	)
 
 	switch kontrolLocation {
-	case kontrol.KontrolLocationLocalMinikube:
+	case kontrol.KontrolLocationLocal:
 		scheme = httpSchme
 		host = localMinikubeKontrolAPIHost
-	case kontrol.KontrolLocationKloudKontrol:
+	case kontrol.KontrolLocationKloud:
 		scheme = httpsScheme
 		host = kloudKontrolAPIHost
 	default:

diff --git a/kardinal-cli/deployment/deployment.go b/kardinal-cli/deployment/deployment.go
@@ -79,6 +79,7 @@ spec:
     metadata:
       labels:
         {{.KardinalAppIDLabelKey}}: {{.KardinalManagerAppIDLabelValue}}
+        app: kardinal-manager
     spec:
       serviceAccountName: kardinal-manager
       containers:
@@ -217,9 +218,9 @@ func DeployKardinalManagerInCluster(ctx context.Context, clusterResourcesURL str
 	var imagePullPolicy string
 
 	switch kontrolLocation {
-	case kontrol.KontrolLocationLocalMinikube:
+	case kontrol.KontrolLocationLocal:
 		imagePullPolicy = "IfNotPresent"
-	case kontrol.KontrolLocationKloudKontrol:
+	case kontrol.KontrolLocationKloud:
 		imagePullPolicy = "Always"
 	default:
 		stacktrace.NewError("invalid Kontrol location: %s", kontrolLocation)

diff --git a/kardinal-cli/kontrol/location.go b/kardinal-cli/kontrol/location.go
@@ -1,15 +1,19 @@
 package kontrol
 
 import (
+	"fmt"
+	"os"
+
 	"github.com/kurtosis-tech/stacktrace"
 	"github.com/sirupsen/logrus"
 	"kardinal.cli/host_machine_directories"
-	"os"
 )
 
 const (
-	KontrolLocationLocalMinikube               = "local-minikube"
-	KontrolLocationKloudKontrol                = "kloud-kontrol"
+	KontrolLocationLocal                       = "local-kardinal-kontrol"
+	KontrolLocationKloud                       = "kloud-kardinal-kontrol"
+	OldKontrolLocationLocal                    = "local-minikube"
+	OldKontrolLocationKloud                    = "kloud-kontrol"
 	kontrolLocationFilePermissions os.FileMode = 0644
 )
 
@@ -33,9 +37,10 @@ func GetKontrolLocation() (string, error) {
 		return "", stacktrace.Propagate(err, "An error occurred getting the Kontrol location filepath")
 	}
 
+	fmt.Println(kontrolLocationFilepath)
 	_, err = os.Stat(kontrolLocationFilepath)
 	if err != nil {
-		return "", stacktrace.Propagate(err, "An error occurred getting the Kontrol location  file info")
+		return "", stacktrace.Propagate(err, "An error occurred getting the Kontrol location file info")
 	}
 
 	kontrolLocationFileBytes, err := os.ReadFile(kontrolLocationFilepath)
@@ -44,6 +49,12 @@ func GetKontrolLocation() (string, error) {
 	}
 
 	kontrolLocationFileStr := string(kontrolLocationFileBytes)
+	switch kontrolLocationFileStr {
+	case OldKontrolLocationLocal:
+		kontrolLocationFileStr = KontrolLocationLocal
+	case OldKontrolLocationKloud:
+		kontrolLocationFileStr = KontrolLocationKloud
+	}
 
 	logrus.Infof("Using Kontrol location %s", kontrolLocationFileStr)
 	return kontrolLocationFileStr, nil