Adding cloud-init to images

README.md

@@ -70,22 +70,28 @@ This will create a new Windows runner pool for the repo with ID `26ae13a1-13e9-4
 
 **NOTE**: If you want to use a custom image that you created, specify the image name in the following format: `projects/my_project/global/images/my-custom-image`
 
-Here is an example for a Linux pool that uses the image specified by its image name:
+Always find a recent image to use. For example, to see available Windows server 2022 images, run something like `gcloud compute images list --filter windows-2022` or just search [here](https://console.cloud.google.com/compute/images).
+
+Linux pools support **ONLY** images with **CLOUD-INIT** already installed. Before using a linux pool, you must be sure that the image has cloud-init installed. Here is an example for a Linux pool that uses a custom image with cloud-init specified by its image name:
 
 ```bash
 garm-cli pool create \
     --os-type linux \
     --os-arch amd64 \
     --enabled=true \
     --flavor e2-medium \
-    --image  projects/debian-cloud/global/images/debian-11-bullseye-v20240110 \
+    --image  projects/garm-testing-424210/global/images/debian-cloud-init \
     --min-idle-runners 0 \
     --repo eb3f78b6-d667-4717-97c4-7aa1f3852138 \
     --tags gcp,linux \
     --provider-name gcp
 ```
 
-Always find a recent image to use. For example, to see available Windows server 2022 images, run something like `gcloud compute images list --filter windows-2022` or just search [here](https://console.cloud.google.com/compute/images).
+**NOTE:** In order to [create a custom image](https://cloud.google.com/compute/docs/images/create-custom#create_image) with cloud-init, you have to *create an instance* with your desired Linux OS. Then connect to that instance and install `cloud-init`. After the install is finished, you can stop that instance and from the `Disk` of that instance, create a custom image. As example, if you use `projects/debian-cloud/global/images/debian-12-bookworm-v20240617`, you can install cloud-init on the instance like this:
+
+```
+sudo apt update && sudo apt install -y cloud-init
+```
 
 ## Tweaking the provider
 

internal/client/gcp.go

@@ -35,7 +35,7 @@ import (
 )
 
 const (
-	linuxStartupScript   string = "startup-script"
+	linuxUserData        string = "user-data"
 	windowsStartupScript string = "sysprep-specialize-script-ps1"
 	accessConfigType     string = "ONE_TO_ONE_NAT"
 )
@@ -293,7 +293,7 @@ func selectStartupScript(osType params.OSType) string {
 	case params.Windows:
 		return windowsStartupScript
 	case params.Linux:
-		return linuxStartupScript
+		return linuxUserData
 	default:
 		return ""
 	}

internal/client/gcp_test.go

@@ -33,7 +33,7 @@ import (
 	"google.golang.org/protobuf/proto"
 )
 
-func TestCreateInstance(t *testing.T) {
+func TestCreateInstanceLinux(t *testing.T) {
 	ctx := context.Background()
 	mockClient := new(MockGcpClient)
 	WaitOp = func(op *compute.Operation, ctx context.Context, opts ...gax.CallOption) error {
@@ -53,6 +53,9 @@ func TestCreateInstance(t *testing.T) {
 
 	mockOperation := &compute.Operation{}
 	mockClient.On("Insert", mock.Anything, mock.Anything, mock.Anything).Return(mockOperation, nil)
+	spec.DefaultCloudConfigFunc = func(bootstrapParams params.BootstrapInstance, tools params.RunnerApplicationDownload, runnerName string) (string, error) {
+		return "MockUserData", nil
+	}
 
 	spec := &spec.RunnerSpec{
 		Zone: "europe-west1-d",
@@ -81,10 +84,100 @@ func TestCreateInstance(t *testing.T) {
 
 	expectedInstance := &computepb.Instance{
 		Name: proto.String("garm-instance"),
+		Metadata: &computepb.Metadata{
+			Items: []*computepb.Items{
+				{
+					Key:   proto.String("runner_name"),
+					Value: proto.String("garm-instance"),
+				},
+				{
+					Key:   proto.String(linuxUserData),
+					Value: proto.String("MockUserData"),
+				},
+			},
+		},
+	}
+	result, err := gcpCli.CreateInstance(ctx, spec)
+	assert.NoError(t, err)
+	assert.Equal(t, expectedInstance.Name, result.Name)
+	for key, value := range expectedInstance.Metadata.Items {
+		assert.Equal(t, *expectedInstance.Metadata.Items[key].Key, *value.Key)
+		assert.Equal(t, *expectedInstance.Metadata.Items[key].Value, *value.Value)
+	}
+	mockClient.AssertExpectations(t)
+}
+
+func TestCreateInstanceWindows(t *testing.T) {
+	ctx := context.Background()
+	mockClient := new(MockGcpClient)
+	WaitOp = func(op *compute.Operation, ctx context.Context, opts ...gax.CallOption) error {
+		return nil
+	}
+	gcpCli := &GcpCli{
+		cfg: &config.Config{
+			Zone:             "europe-west1-d",
+			ProjectId:        "my-project",
+			NetworkID:        "my-network",
+			SubnetworkID:     "my-subnetwork",
+			CredentialsFile:  "path/to/credentials.json",
+			ExternalIPAccess: true,
+		},
+		client: mockClient,
+	}
+
+	mockOperation := &compute.Operation{}
+	mockClient.On("Insert", mock.Anything, mock.Anything, mock.Anything).Return(mockOperation, nil)
+	spec.DefaultRunnerInstallScriptFunc = func(bootstrapParams params.BootstrapInstance, tools params.RunnerApplicationDownload, runnerName string) ([]byte, error) {
+		return []byte("MockUserData"), nil
+	}
+
+	spec := &spec.RunnerSpec{
+		Zone: "europe-west1-d",
+		Tools: params.RunnerApplicationDownload{
+			OS:           proto.String("windows"),
+			Architecture: proto.String("amd64"),
+			DownloadURL:  proto.String("MockURL"),
+			Filename:     proto.String("garm-runner"),
+		},
+		NetworkID:      "my-network",
+		SubnetworkID:   "my-subnetwork",
+		ControllerID:   "my-controller",
+		NicType:        "VIRTIO_NET",
+		DiskSize:       50,
+		CustomLabels:   map[string]string{"key1": "value1"},
+		NetworkTags:    []string{"tag1", "tag2"},
+		SourceSnapshot: "projects/garm-testing/global/snapshots/garm-snapshot",
+		BootstrapParams: params.BootstrapInstance{
+			Name:   "garm-instance",
+			Flavor: "n1-standard-1",
+			Image:  "projects/garm-testing/global/images/garm-image",
+			OSType: params.Windows,
+			OSArch: "amd64",
+		},
+	}
+
+	expectedInstance := &computepb.Instance{
+		Name: proto.String("garm-instance"),
+		Metadata: &computepb.Metadata{
+			Items: []*computepb.Items{
+				{
+					Key:   proto.String("runner_name"),
+					Value: proto.String("garm-instance"),
+				},
+				{
+					Key:   proto.String(windowsStartupScript),
+					Value: proto.String("MockUserData"),
+				},
+			},
+		},
 	}
 	result, err := gcpCli.CreateInstance(ctx, spec)
 	assert.NoError(t, err)
 	assert.Equal(t, expectedInstance.Name, result.Name)
+	for key, value := range expectedInstance.Metadata.Items {
+		assert.Equal(t, *expectedInstance.Metadata.Items[key].Key, *value.Key)
+		assert.Equal(t, *expectedInstance.Metadata.Items[key].Value, *value.Value)
+	}
 	mockClient.AssertExpectations(t)
 }
 

internal/spec/spec.go

@@ -16,15 +16,12 @@
 package spec
 
 import (
-	"encoding/base64"
 	"encoding/json"
 	"fmt"
 	"maps"
 	"regexp"
-	"strings"
 
 	"github.com/cloudbase/garm-provider-common/cloudconfig"
-	"github.com/cloudbase/garm-provider-common/defaults"
 	"github.com/cloudbase/garm-provider-common/params"
 	"github.com/cloudbase/garm-provider-common/util"
 	"github.com/cloudbase/garm-provider-gcp/config"
@@ -104,6 +101,8 @@ const (
 type ToolFetchFunc func(osType params.OSType, osArch params.OSArch, tools []params.RunnerApplicationDownload) (params.RunnerApplicationDownload, error)
 
 var DefaultToolFetch ToolFetchFunc = util.GetTools
+var DefaultCloudConfigFunc = cloudconfig.GetCloudConfig
+var DefaultRunnerInstallScriptFunc = cloudconfig.GetRunnerInstallScript
 
 func jsonSchemaValidation(schema json.RawMessage) error {
 	schemaLoader := gojsonschema.NewStringLoader(jsonSchema)
@@ -286,34 +285,18 @@ func (r RunnerSpec) ComposeUserData() (string, error) {
 
 	switch r.BootstrapParams.OSType {
 	case params.Linux:
-		udata, err := cloudconfig.GetRunnerInstallScript(r.BootstrapParams, r.Tools, r.BootstrapParams.Name)
+		// Get the cloud-init config
+		udata, err := DefaultCloudConfigFunc(bootstrapParams, r.Tools, bootstrapParams.Name)
 		if err != nil {
 			return "", fmt.Errorf("failed to generate userdata: %w", err)
 		}
+		return udata, nil
 
-		asBase64 := base64.StdEncoding.EncodeToString(udata)
-		scriptCommands := []string{
-			"sudo useradd -m " + defaults.DefaultUser + " || true",
-			// Create the runner home directory if it doesn't exist
-			"sudo mkdir -p /home/" + defaults.DefaultUser,
-			// Add user to sudoers
-			"sudo usermod -aG sudo " + defaults.DefaultUser,
-			// Check curl and tar are installed
-			"sudo apt-get update && sudo apt-get install -y curl tar",
-			// Install the runner
-			"echo " + asBase64 + " | base64 -d > /install_runner.sh",
-			"chmod +x /install_runner.sh",
-			"echo 'runner  ALL=(ALL) NOPASSWD:ALL' > /etc/sudoers.d/garm",
-			"su -l -c /install_runner.sh " + defaults.DefaultUser,
-		}
-		script := strings.Join(scriptCommands, "\n")
-		return script, nil
 	case params.Windows:
-		udata, err := cloudconfig.GetRunnerInstallScript(bootstrapParams, r.Tools, bootstrapParams.Name)
+		udata, err := DefaultRunnerInstallScriptFunc(bootstrapParams, r.Tools, bootstrapParams.Name)
 		if err != nil {
 			return "", fmt.Errorf("failed to generate userdata: %w", err)
 		}
-
 		return string(udata), nil
 	}
 	return "", fmt.Errorf("unsupported OS type for cloud config: %s", r.BootstrapParams.OSType)