fix: improve handling of metadata service connectivity (#466)

* fix(controller): properly handle error on init

* refactor: simplify error generation

* fix: check for hcloud server

We had some issues where the container could not access the metadata
service and debugging this was hard. By validating this on startup and
logging a warning we can help guide users to fix the issue themselves.

app/app.go

@@ -55,7 +55,7 @@ func CreateListener() (net.Listener, error) {
 	endpoint = endpoint[7:] // strip unix://
 
 	if err := os.Remove(endpoint); err != nil && !os.IsNotExist(err) {
-		return nil, errors.New(fmt.Sprintf("failed to remove socket file at %s: %s", endpoint, err))
+		return nil, fmt.Errorf("failed to remove socket file at %s: %s", endpoint, err)
 	}
 
 	return net.Listen("unix", endpoint)
@@ -145,6 +145,9 @@ func CreateHcloudClient(metricsRegistry *prometheus.Registry, logger log.Logger)
 // GetServer retrieves the hcloud server the application is running on.
 func GetServer(logger log.Logger, hcloudClient *hcloud.Client, metadataClient *metadata.Client) (*hcloud.Server, error) {
 	hcloudServerID, err := getServerID(logger, hcloudClient, metadataClient)
+	if err != nil {
+		return nil, err
+	}
 	level.Debug(logger).Log("msg", "fetching server")
 	server, _, err := hcloudClient.Server.GetByID(context.Background(), hcloudServerID)
 	if err != nil {
@@ -167,7 +170,7 @@ func getServerID(logger log.Logger, hcloudClient *hcloud.Client, metadataClient
 	if s := os.Getenv("HCLOUD_SERVER_ID"); s != "" {
 		id, err := strconv.ParseInt(s, 10, 64)
 		if err != nil {
-			return 0, errors.New(fmt.Sprintf("invalid server id in HCLOUD_SERVER_ID env var: %s", err))
+			return 0, fmt.Errorf("invalid server id in HCLOUD_SERVER_ID env var: %s", err)
 		}
 		level.Debug(logger).Log(
 			"msg", "using server id from HCLOUD_SERVER_ID env var",
@@ -179,7 +182,7 @@ func getServerID(logger log.Logger, hcloudClient *hcloud.Client, metadataClient
 	if s := os.Getenv("KUBE_NODE_NAME"); s != "" {
 		server, _, err := hcloudClient.Server.GetByName(context.Background(), s)
 		if err != nil {
-			return 0, errors.New(fmt.Sprintf("error while getting server through node name: %s", err))
+			return 0, fmt.Errorf("error while getting server through node name: %s", err)
 		}
 		if server != nil {
 			level.Debug(logger).Log(
@@ -199,7 +202,7 @@ func getServerID(logger log.Logger, hcloudClient *hcloud.Client, metadataClient
 	)
 	id, err := metadataClient.InstanceID()
 	if err != nil {
-		return 0, errors.New(fmt.Sprintf("failed to get instance id from metadata service: %s", err))
+		return 0, fmt.Errorf("failed to get instance id from metadata service: %s", err)
 	}
 	return id, nil
 }

cmd/aio/main.go

@@ -34,6 +34,10 @@ func main() {
 
 	metadataClient := metadata.NewClient(metadata.WithInstrumentation(m.Registry()))
 
+	if !metadataClient.IsHcloudServer() {
+		level.Warn(logger).Log("msg", "unable to connect to metadata service, are you sure this is running on a Hetzner Cloud server?")
+	}
+
 	// node
 	serverID, err := metadataClient.InstanceID()
 	if err != nil {

cmd/controller/main.go

@@ -36,6 +36,12 @@ func main() {
 	} else {
 		metadataClient := metadata.NewClient(metadata.WithInstrumentation(m.Registry()))
 
+		if !metadataClient.IsHcloudServer() {
+			level.Warn(logger).Log("msg", "Unable to connect to metadata service. "+
+				"In the current configuration the controller is required to run on a Hetzner Cloud server. "+
+				"You can set HCLOUD_VOLUME_DEFAULT_LOCATION if you want to run it somewhere else.")
+		}
+
 		server, err := app.GetServer(logger, hcloudClient, metadataClient)
 		if err != nil {
 			level.Error(logger).Log(

cmd/node/main.go

@@ -24,6 +24,10 @@ func main() {
 
 	metadataClient := metadata.NewClient(metadata.WithInstrumentation(m.Registry()))
 
+	if !metadataClient.IsHcloudServer() {
+		level.Warn(logger).Log("msg", "unable to connect to metadata service, are you sure this is running on a Hetzner Cloud server?")
+	}
+
 	serverID, err := metadataClient.InstanceID()
 	if err != nil {
 		level.Error(logger).Log("msg", "failed to fetch server ID from metadata service", "err", err)

volumes/mount.go

@@ -1,7 +1,6 @@
 package volumes
 
 import (
-	"errors"
 	"fmt"
 	"os"
 	"path/filepath"
@@ -163,7 +162,7 @@ func (s *LinuxMountService) Publish(targetPath string, devicePath string, opts M
 func (s *LinuxMountService) Unpublish(targetPath string) error {
 	devicePath, _, err := mount.GetDeviceNameFromMount(mount.New(""), targetPath)
 	if err != nil {
-		return errors.New(fmt.Sprintf("failed to determine mount path for %s: %s", targetPath, err))
+		return fmt.Errorf("failed to determine mount path for %s: %s", targetPath, err)
 	}
 
 	level.Info(s.logger).Log(

volumes/resize.go

@@ -1,7 +1,6 @@
 package volumes
 
 import (
-	"errors"
 	"fmt"
 
 	"github.com/go-kit/kit/log"
@@ -36,7 +35,7 @@ func NewLinuxResizeService(logger log.Logger) *LinuxResizeService {
 func (l *LinuxResizeService) Resize(volumePath string) error {
 	devicePath, _, err := mount.GetDeviceNameFromMount(mount.New(""), volumePath)
 	if err != nil {
-		return errors.New(fmt.Sprintf("failed to determine mount path for %s: %s", volumePath, err))
+		return fmt.Errorf("failed to determine mount path for %s: %s", volumePath, err)
 	}
 
 	level.Info(l.logger).Log(