Initial support for GCP cluster health.

openshift · Feb 15, 2023 · 658b97f · 658b97f
1 parent 9cb3934
commit 658b97f
Show file tree

Hide file tree

Showing 3 changed files with 149 additions and 43 deletions.
diff --git a/cmd/cluster/health.go b/cmd/cluster/health.go
@@ -1,6 +1,7 @@
 package cluster
 
 import (
+	"errors"
 	"fmt"
 	"log"
 	"strconv"
@@ -11,6 +12,7 @@ import (
 	"github.com/openshift/osdctl/pkg/osdCloud"
 	"github.com/openshift/osdctl/pkg/utils"
 	"github.com/spf13/cobra"
+	"google.golang.org/api/iterator"
 	"gopkg.in/yaml.v2"
 	cmdutil "k8s.io/kubectl/pkg/cmd/util"
 )
@@ -96,59 +98,123 @@ func (o *healthOptions) run() error {
 		healthObject.Expected.Worker = int(cluster.Nodes().Compute())
 	}
 
-	awsClient, err := osdCloud.GenerateAWSClientForCluster(o.awsProfile, o.clusterID)
-	if err != nil {
-		return err
-	}
-
-	instances, err := awsClient.DescribeInstances(&ec2.DescribeInstancesInput{})
-	if err != nil {
-		return err
-	}
 	runningMasters := 0
 	runningInfra := 0
 	runningWorkers := 0
 	totalStopped := 0
 	totalCluster := 0
 
-	//Here we count the number of customer's running worker, infra and master instances in the cluster in the given region. To decide if the instance belongs to the cluster we are checking the Name Tag on the instance.
-	for idx := range instances.Reservations {
-		for _, inst := range instances.Reservations[idx].Instances {
-			tags := inst.Tags
-			for _, t := range tags {
-				if *t.Key == "Name" {
-					if strings.HasPrefix(*t.Value, cluster.InfraID()) && strings.Contains(*t.Value, "master") {
-						totalCluster += 1
-						if *inst.State.Name == "running" {
-							runningMasters += 1
-						}
-						if *inst.State.Name == "stopped" {
-							totalStopped += 1
-						}
-
-					} else if strings.HasPrefix(*t.Value, cluster.InfraID()) && strings.Contains(*t.Value, "infra") {
-						totalCluster += 1
-						if *inst.State.Name == "running" {
-							runningInfra += 1
-						}
-						if *inst.State.Name == "stopped" {
-							totalStopped += 1
-						}
-					} else if strings.HasPrefix(*t.Value, cluster.InfraID()) && strings.Contains(*t.Value, "worker") {
-						totalCluster += 1
-						if *inst.State.Name == "running" {
-							runningWorkers += 1
-						}
-						if *inst.State.Name == "stopped" {
-							totalStopped += 1
-						}
-
+	if cluster.CloudProvider().ID() == "gcp" {
+		clusterResources, err := ocmClient.ClustersMgmt().V1().Clusters().Cluster(o.clusterID).Resources().Live().Get().Send()
+		if err != nil {
+			return err
+		}
+		projectClaimRaw, found := clusterResources.Body().Resources()["gcp_project_claim"]
+		if !found {
+			return fmt.Errorf("The gcp_project_claim was not found in the ocm resource")
+		}
+		projectClaim, err := osdCloud.ParseGcpProjectClaim(projectClaimRaw)
+		if err != nil {
+			log.Printf("Unmarshalling GCP projectClaim failed: %v\n", err)
+			return err
+		}
+		projectId := projectClaim.Spec.GcpProjectID
+		zones := cluster.Nodes().AvailabilityZones()
+		if projectId == "" || len(zones) == 0 {
+			return fmt.Errorf("ProjectID or Zones empty - aborting")
+		}
+		gcpClient, err := osdCloud.GenerateGCPComputeInstancesClient()
+		defer gcpClient.Close()
+		if err != nil {
+			return err
+		}
+		ownedLabel := "kubernetes-io-cluster-" + cluster.InfraID()
+		for _, zone := range zones {
+			instances := osdCloud.ListInstances(gcpClient, projectId, zone)
+			for {
+				instance, err := instances.Next()
+				if err == iterator.Done {
+					break
+				}
+				if err != nil {
+					return err
+				}
+				name := instance.GetName()
+				state := instance.GetStatus()
+				labels := instance.GetLabels()
+				belongsToCluster := false
+				for label := range labels {
+					if label == ownedLabel {
+						belongsToCluster = true
+					}
+				}
+				if !belongsToCluster {
+					log.Printf("Skipping a machine not belonging to the cluster: %s\n", name)
+					continue
+				}
+				totalCluster += 1
+				if state != "RUNNING" {
+					totalStopped += 1
+				} else {
+					if strings.HasPrefix(name, cluster.InfraID()) && strings.Contains(name, "master") {
+						runningMasters += 1
+					} else if strings.HasPrefix(name, cluster.InfraID()) && strings.Contains(name, "infra") {
+						runningInfra += 1
+					} else if strings.HasPrefix(name, cluster.InfraID()) && strings.Contains(name, "worker") {
+						runningWorkers += 1
 					}
 				}
 			}
+		}
+	} else if cluster.CloudProvider().ID() == "aws" {
+		awsClient, err := osdCloud.GenerateAWSClientForCluster(o.awsProfile, o.clusterID)
+		if err != nil {
+			return err
+		}
 
+		instances, err := awsClient.DescribeInstances(&ec2.DescribeInstancesInput{})
+		if err != nil {
+			return err
 		}
 
+		//Here we count the number of customer's running worker, infra and master instances in the cluster in the given region. To decide if the instance belongs to the cluster we are checking the Name Tag on the instance.
+		for idx := range instances.Reservations {
+			for _, inst := range instances.Reservations[idx].Instances {
+				tags := inst.Tags
+				for _, t := range tags {
+					if *t.Key == "Name" {
+						if strings.HasPrefix(*t.Value, cluster.InfraID()) && strings.Contains(*t.Value, "master") {
+							totalCluster += 1
+							if *inst.State.Name == "running" {
+								runningMasters += 1
+							}
+							if *inst.State.Name == "stopped" {
+								totalStopped += 1
+							}
+
+						} else if strings.HasPrefix(*t.Value, cluster.InfraID()) && strings.Contains(*t.Value, "infra") {
+							totalCluster += 1
+							if *inst.State.Name == "running" {
+								runningInfra += 1
+							}
+							if *inst.State.Name == "stopped" {
+								totalStopped += 1
+							}
+						} else if strings.HasPrefix(*t.Value, cluster.InfraID()) && strings.Contains(*t.Value, "worker") {
+							totalCluster += 1
+							if *inst.State.Name == "running" {
+								runningWorkers += 1
+							}
+							if *inst.State.Name == "stopped" {
+								totalStopped += 1
+							}
+						}
+					}
+				}
+			}
+		}
+	} else {
+		return errors.New(fmt.Sprintf("Unknown cloud provider found: %s", cluster.CloudProvider().ID()))
 	}
 
 	healthObject.Actual.Stopped = totalStopped

diff --git a/go.mod b/go.mod
@@ -30,6 +30,8 @@ require (
 	github.com/spf13/cobra v1.6.1
 	github.com/spf13/pflag v1.0.5
 	github.com/spf13/viper v1.13.0
+	google.golang.org/api v0.84.0
+	google.golang.org/genproto v0.0.0-20220616135557-88e70c0c3a90
 	gopkg.in/yaml.v2 v2.4.0
 	k8s.io/api v0.24.3
 	k8s.io/apimachinery v0.24.3
@@ -146,9 +148,7 @@ require (
 	golang.org/x/term v0.3.0 // indirect
 	golang.org/x/text v0.5.0 // indirect
 	golang.org/x/time v0.0.0-20220210224613-90d013bbcef8 // indirect
-	google.golang.org/api v0.84.0 // indirect
 	google.golang.org/appengine v1.6.7 // indirect
-	google.golang.org/genproto v0.0.0-20220616135557-88e70c0c3a90 // indirect
 	google.golang.org/grpc v1.47.0 // indirect
 	google.golang.org/protobuf v1.28.1 // indirect
 	gopkg.in/check.v1 v1.0.0-20201130134442-10cb98267c6c // indirect

diff --git a/pkg/osdCloud/gcp.go b/pkg/osdCloud/gcp.go
@@ -0,0 +1,40 @@
+package osdCloud
+
+import (
+	"context"
+	"encoding/json"
+
+	compute "cloud.google.com/go/compute/apiv1"
+	computepb "google.golang.org/genproto/googleapis/cloud/compute/v1"
+)
+
+type GcpProjectClaimSpec struct {
+	GcpProjectID string `json:"gcpProjectID"`
+}
+type GcpProjectClaim struct {
+	Spec GcpProjectClaimSpec `json:"spec"`
+}
+
+func ParseGcpProjectClaim(raw string) (*GcpProjectClaim, error) {
+	var projectClaim GcpProjectClaim
+	err := json.Unmarshal([]byte(raw), &projectClaim)
+	if err != nil {
+		return nil, err
+	}
+	return &projectClaim, nil
+}
+
+func GenerateGCPComputeInstancesClient() (*compute.InstancesClient, error) {
+	ctx := context.Background()
+	client, err := compute.NewInstancesRESTClient(ctx)
+	return client, err
+}
+
+func ListInstances(client *compute.InstancesClient, projectID, zone string) *compute.InstanceIterator {
+	ctx := context.Background()
+	request := &computepb.ListInstancesRequest{
+		Project: projectID,
+		Zone:    zone,
+	}
+	return client.List(ctx, request)
+}