feat: reporting back individual node upgrades

cmd/helmvm/join.go

@@ -75,6 +75,13 @@ var joinCommand = &cli.Command{
 			metrics.ReportJoinFailed(c.Context, hvmtoken.ClusterID, err)
 			return err
 		}
+		fpath := defaults.PathToConfig(".cluster-id")
+		cid := hvmtoken.ClusterID.String()
+		if err := os.WriteFile(fpath, []byte(cid), 0644); err != nil {
+			err := fmt.Errorf("unable to write cluster id to disk: %w", err)
+			metrics.ReportJoinFailed(c.Context, hvmtoken.ClusterID, err)
+			return err
+		}
 		metrics.ReportJoinSucceeded(c.Context, hvmtoken.ClusterID)
 		return nil
 	},

cmd/helmvm/upgrade.go

@@ -13,6 +13,7 @@ import (
 	"github.com/replicatedhq/helmvm/pkg/addons"
 	"github.com/replicatedhq/helmvm/pkg/defaults"
 	"github.com/replicatedhq/helmvm/pkg/goods"
+	"github.com/replicatedhq/helmvm/pkg/metrics"
 	"github.com/replicatedhq/helmvm/pkg/preflights"
 	"github.com/replicatedhq/helmvm/pkg/prompts"
 )
@@ -103,34 +104,49 @@ var upgradeCommand = &cli.Command{
 		},
 	},
 	Action: func(c *cli.Context) error {
+		metrics.ReportNodeUpgradeStarted(c.Context)
 		if err := canRunUpgrade(c); err != nil {
+			metrics.ReportNodeUpgradeFailed(c.Context, err)
 			return err
 		}
 		logrus.Infof("Materializing binaries")
 		if err := goods.Materialize(); err != nil {
-			return fmt.Errorf("unable to materialize binaries: %w", err)
+			err := fmt.Errorf("unable to materialize binaries: %w", err)
+			metrics.ReportNodeUpgradeFailed(c.Context, err)
+			return err
 		}
 		if err := runHostPreflightsLocally(c); err != nil {
-			return fmt.Errorf("unable to run host preflights locally: %w", err)
+			err := fmt.Errorf("unable to run host preflights locally: %w", err)
+			metrics.ReportNodeUpgradeFailed(c.Context, err)
+			return err
 		}
 		logrus.Infof("Stopping %s", defaults.BinaryName())
 		if err := stopHelmVM(); err != nil {
-			return fmt.Errorf("unable to stop: %w", err)
+			err := fmt.Errorf("unable to stop: %w", err)
+			metrics.ReportNodeUpgradeFailed(c.Context, err)
+			return err
 		}
 		logrus.Infof("Installing binary")
 		if err := installK0sBinary(); err != nil {
-			return fmt.Errorf("unable to install k0s binary: %w", err)
+			err := fmt.Errorf("unable to install k0s binary: %w", err)
+			metrics.ReportNodeUpgradeFailed(c.Context, err)
+			return err
 		}
 		logrus.Infof("Starting service")
 		if err := startK0sService(); err != nil {
-			return fmt.Errorf("unable to start service: %w", err)
+			err := fmt.Errorf("unable to start k0s service: %w", err)
+			metrics.ReportNodeUpgradeFailed(c.Context, err)
+			return err
 		}
 		kcfg := defaults.PathToConfig("kubeconfig")
 		if _, err := os.Stat(kcfg); err != nil {
 			if os.IsNotExist(err) {
+				metrics.ReportNodeUpgradeSucceeded(c.Context)
 				return nil
 			}
-			return fmt.Errorf("unable to stat kubeconfig: %w", err)
+			err := fmt.Errorf("unable to read kubeconfig: %w", err)
+			metrics.ReportNodeUpgradeFailed(c.Context, err)
+			return err
 		}
 		os.Setenv("KUBECONFIG", kcfg)
 		logrus.Infof("Upgrading addons")
@@ -142,8 +158,11 @@ var upgradeCommand = &cli.Command{
 			opts = append(opts, addons.WithoutAddon(addon))
 		}
 		if err := addons.NewApplier(opts...).Apply(c.Context); err != nil {
-			return fmt.Errorf("unable to apply addons: %w", err)
+			err := fmt.Errorf("unable to apply addons: %w", err)
+			metrics.ReportNodeUpgradeFailed(c.Context, err)
+			return err
 		}
+		metrics.ReportNodeUpgradeSucceeded(c.Context)
 		logrus.Infof("Upgrade complete")
 		return nil
 	},

pkg/metrics/events.go

@@ -100,3 +100,30 @@ type JoinFailed struct {
 func (e JoinFailed) Title() string {
 	return "JoinFailed"
 }
+
+// NodeUpgradeStarted event is send back home when a node upgrade
+// starts.
+type NodeUpgradeStarted JoinStarted
+
+// Title returns the name of the event.
+func (e NodeUpgradeStarted) Title() string {
+	return "NodeUpgradeStarted"
+}
+
+// NodeUpgradeSucceeded event is send back home when a node upgrade
+// succeeds.
+type NodeUpgradeSucceeded NodeUpgradeStarted
+
+// Title returns the name of the event.
+func (e NodeUpgradeSucceeded) Title() string {
+	return "NodeUpgradeSucceeded"
+}
+
+// NodeUpgradeFailed event is send back home when a node upgrade
+// fails.
+type NodeUpgradeFailed JoinFailed
+
+// Title returns the name of the event.
+func (e NodeUpgradeFailed) Title() string {
+	return "NodeUpgradeFailed"
+}

pkg/metrics/reporter.go

@@ -178,3 +178,33 @@ func ReportApplyFinished(c *cli.Context, err error) {
 	}
 	ReportInstallationSuceeded(ctx)
 }
+
+// ReportNodeUpgradeStarted reports that a node upgrade has started.
+func ReportNodeUpgradeStarted(ctx context.Context) {
+	hostname, err := os.Hostname()
+	if err != nil {
+		logrus.Warnf("unable to get hostname: %s", err)
+		hostname = "unknown"
+	}
+	Send(ctx, NodeUpgradeStarted{ClusterID(), hostname})
+}
+
+// ReportNodeUpgradeSucceeded reports that a node upgrade has finished successfully.
+func ReportNodeUpgradeSucceeded(ctx context.Context) {
+	hostname, err := os.Hostname()
+	if err != nil {
+		logrus.Warnf("unable to get hostname: %s", err)
+		hostname = "unknown"
+	}
+	Send(ctx, NodeUpgradeSucceeded{ClusterID(), hostname})
+}
+
+// ReportNodeUpgradeFailed reports that node upgrade has failed.
+func ReportNodeUpgradeFailed(ctx context.Context, exterr error) {
+	hostname, err := os.Hostname()
+	if err != nil {
+		logrus.Warnf("unable to get hostname: %s", err)
+		hostname = "unknown"
+	}
+	Send(ctx, NodeUpgradeFailed{ClusterID(), hostname, exterr.Error()})
+}

pkg/metrics/sender_test.go

@@ -89,6 +89,28 @@ func TestSend(t *testing.T) {
 				Reason:    "bar",
 			},
 		},
+		{
+			name: "NodeUpgradeStarted",
+			event: NodeUpgradeStarted{
+				ClusterID: uuid.New(),
+				NodeName:  "foo",
+			},
+		},
+		{
+			name: "NodeUpgradeSucceeded",
+			event: NodeUpgradeSucceeded{
+				ClusterID: uuid.New(),
+				NodeName:  "foo",
+			},
+		},
+		{
+			name: "NodeUpgradeFailed",
+			event: NodeUpgradeFailed{
+				ClusterID: uuid.New(),
+				NodeName:  "foo",
+				Reason:    "bar",
+			},
+		},
 	} {
 		t.Run(tt.name, func(t *testing.T) {
 			payload := map[string]Event{"event": tt.event}