Document telemetry, backtrace only on crash, ignore submit errors (#388)

* Document telemetry in README and improve crash message redaction

* Update README.md

Co-authored-by: Tom Wieczorek <twz123@users.noreply.github.com>

* Only send a cleaned up backtrace upon crash

* Suppress analytics error messages and ignore any failures

* The test panic was not meant to be commited

Co-authored-by: Tom Wieczorek <twz123@users.noreply.github.com>

README.md

@@ -88,13 +88,30 @@ k0sctl completion > /usr/local/share/zsh/site-functions/_k0sctl
 k0sctl completion > ~/.config/fish/completions/k0sctl.fish
 ```
 
+## Anonymous telemetry
+
+K0sctl sends anonymized telemetry data when it is used. This can be disabled via the `--disable-telemetry` flag or by setting the environment variable `DISABLE_TELEMETRY=true`.
+
+The telemetry data includes:
+
+- K0sctl version
+- Operating system + CPU architecture ("linux x86", "darwin arm64", ...)
+- An anonymous machine ID generated by [denisbrodbeck/machineid](https://github.com/denisbrodbeck/machineid) or if that fails, an md5 sum of the hostname
+- Event information:
+  * Phase name ("Connecting to hosts", "Gathering facts", ...) and the duration how long it took to finish
+  * Cluster UUID (`kubectl get -n kube-system namespace kube-system -o template={{.metadata.uid}}`)
+  * Was k0s dynamic config enabled (true/false)
+  * Was a custom or the default k0s configuration used (true/false)
+  * In case of a crash, a backtrace with source filenames and line numbers only
+
+The data is used to estimate the number of users and to identify failure hotspots.
+
 ## Development status
 
 K0sctl is ready for use and in continuous development. It is still at a stage where maintaining backwards compatibility is not a high priority goal.
 
 Missing major features include at least:
 
-* Windows targets are not yet supported
 * The released binaries have not been signed
 * Nodes can't be removed
 * The configuration specification and command-line interface options are still evolving

analytics/analytics.go

@@ -5,7 +5,7 @@ import (
 )
 
 type publisher interface {
-	Publish(string, map[string]interface{}) error
+	Publish(string, map[string]interface{})
 	Close()
 }
 
@@ -21,9 +21,8 @@ func (c *NullClient) Initialize() error {
 }
 
 // Publish would send a tracking event
-func (c *NullClient) Publish(event string, props map[string]interface{}) error {
+func (c *NullClient) Publish(event string, props map[string]interface{}) {
 	log.Tracef("analytics event %s - properties: %+v", event, props)
-	return nil
 }
 
 // Close the analytics connection

analytics/phase.go

@@ -54,5 +54,7 @@ func (p *Phase) After(result error) error {
 		event = "phase-failure"
 	}
 
-	return Client.Publish(event, p.props)
+	Client.Publish(event, p.props)
+
+	return nil
 }

cmd/apply.go

@@ -78,12 +78,10 @@ var applyCommand = &cli.Command{
 			&phase.Disconnect{},
 		)
 
-		if err := analytics.Client.Publish("apply-start", map[string]interface{}{}); err != nil {
-			return err
-		}
+		analytics.Client.Publish("apply-start", map[string]interface{}{})
 
 		if err := manager.Run(); err != nil {
-			_ = analytics.Client.Publish("apply-failure", map[string]interface{}{"clusterID": manager.Config.Spec.K0s.Metadata.ClusterID})
+			analytics.Client.Publish("apply-failure", map[string]interface{}{"clusterID": manager.Config.Spec.K0s.Metadata.ClusterID})
 			if lf, err := LogFile(); err == nil {
 				if ln, ok := lf.(interface{ Name() string }); ok {
 					log.Errorf("apply failed - log file saved to %s", ln.Name())
@@ -92,7 +90,7 @@ var applyCommand = &cli.Command{
 			return err
 		}
 
-		_ = analytics.Client.Publish("apply-success", map[string]interface{}{"duration": time.Since(start), "clusterID": manager.Config.Spec.K0s.Metadata.ClusterID})
+		analytics.Client.Publish("apply-success", map[string]interface{}{"duration": time.Since(start), "clusterID": manager.Config.Spec.K0s.Metadata.ClusterID})
 
 		duration := time.Since(start).Truncate(time.Second)
 		text := fmt.Sprintf("==> Finished in %s", duration)

cmd/backup.go

@@ -39,12 +39,10 @@ var backupCommand = &cli.Command{
 			&phase.Disconnect{},
 		)
 
-		if err := analytics.Client.Publish("backup-start", map[string]interface{}{}); err != nil {
-			return err
-		}
+		analytics.Client.Publish("backup-start", map[string]interface{}{})
 
 		if err := manager.Run(); err != nil {
-			_ = analytics.Client.Publish("backup-failure", map[string]interface{}{"clusterID": manager.Config.Spec.K0s.Metadata.ClusterID})
+			analytics.Client.Publish("backup-failure", map[string]interface{}{"clusterID": manager.Config.Spec.K0s.Metadata.ClusterID})
 			if lf, err := LogFile(); err == nil {
 				if ln, ok := lf.(interface{ Name() string }); ok {
 					log.Errorf("backup failed - log file saved to %s", ln.Name())
@@ -53,7 +51,7 @@ var backupCommand = &cli.Command{
 			return err
 		}
 
-		_ = analytics.Client.Publish("backup-success", map[string]interface{}{"duration": time.Since(start), "clusterID": manager.Config.Spec.K0s.Metadata.ClusterID})
+		analytics.Client.Publish("backup-success", map[string]interface{}{"duration": time.Since(start), "clusterID": manager.Config.Spec.K0s.Metadata.ClusterID})
 
 		duration := time.Since(start).Truncate(time.Second)
 		text := fmt.Sprintf("==> Finished in %s", duration)

cmd/config_edit.go

@@ -46,9 +46,7 @@ var configEditCommand = &cli.Command{
 			return fmt.Errorf("output is not a terminal")
 		}
 
-		if err := analytics.Client.Publish("config-edit-start", map[string]interface{}{}); err != nil {
-			return err
-		}
+		analytics.Client.Publish("config-edit-start", map[string]interface{}{})
 
 		editor, err := shellEditor()
 		if err != nil {

cmd/config_status.go

@@ -29,9 +29,7 @@ var configStatusCommand = &cli.Command{
 	Before: actions(initLogging, startCheckUpgrade, initConfig, initAnalytics),
 	After:  actions(reportCheckUpgrade, closeAnalytics),
 	Action: func(ctx *cli.Context) error {
-		if err := analytics.Client.Publish("config-status-start", map[string]interface{}{}); err != nil {
-			return err
-		}
+		analytics.Client.Publish("config-status-start", map[string]interface{}{})
 
 		c := ctx.Context.Value(ctxConfigKey{}).(*v1beta1.Cluster)
 		h := c.Spec.K0sLeader()

cmd/reset.go

@@ -63,16 +63,14 @@ var resetCommand = &cli.Command{
 			&phase.Disconnect{},
 		)
 
-		if err := analytics.Client.Publish("reset-start", map[string]interface{}{}); err != nil {
-			return err
-		}
+		analytics.Client.Publish("reset-start", map[string]interface{}{})
 
 		if err := manager.Run(); err != nil {
-			_ = analytics.Client.Publish("reset-failure", map[string]interface{}{"clusterID": manager.Config.Spec.K0s.Metadata.ClusterID})
+			analytics.Client.Publish("reset-failure", map[string]interface{}{"clusterID": manager.Config.Spec.K0s.Metadata.ClusterID})
 			return err
 		}
 
-		_ = analytics.Client.Publish("reset-success", map[string]interface{}{"duration": time.Since(start), "clusterID": manager.Config.Spec.K0s.Metadata.ClusterID})
+		analytics.Client.Publish("reset-success", map[string]interface{}{"duration": time.Since(start), "clusterID": manager.Config.Spec.K0s.Metadata.ClusterID})
 
 		duration := time.Since(start).Truncate(time.Second)
 		text := fmt.Sprintf("==> Finished in %s", duration)

integration/segment/segment.go

@@ -51,14 +51,17 @@ func NewClient() (*Client, error) {
 }
 
 // Publish enqueues the sending of a tracking event
-func (c Client) Publish(event string, props map[string]interface{}) error {
+func (c Client) Publish(event string, props map[string]interface{}) {
 	log.Tracef("segment event %s - properties: %+v", event, props)
-	return c.client.Enqueue(segment.Track{
+	err := c.client.Enqueue(segment.Track{
 		Context:     ctx,
 		AnonymousId: c.machineID,
 		Event:       event,
 		Properties:  props,
 	})
+	if err != nil {
+		log.Debugf("failed to submit telemetry: %s", err)
+	}
 }
 
 // Close the analytics connection

main.go

@@ -1,8 +1,9 @@
 package main
 
 import (
-	"fmt"
 	"os"
+	"runtime"
+	"strings"
 
 	"github.com/k0sproject/k0sctl/analytics"
 	"github.com/k0sproject/k0sctl/cmd"
@@ -11,8 +12,25 @@ import (
 
 func handlepanic() {
 	if err := recover(); err != nil {
-		_ = analytics.Client.Publish("panic", map[string]interface{}{"error": fmt.Sprint(err)})
-		log.Fatalf("PANIC: %s", err)
+		buf := make([]byte, 1<<16)
+		ss := runtime.Stack(buf, true)
+		msg := string(buf[:ss])
+		var bt []string
+		for _, row := range strings.Split(msg, "\n") {
+			if !strings.HasPrefix(row, "\t") {
+				continue
+			}
+			if strings.Contains(row, "main.") {
+				continue
+			}
+			if strings.Contains(row, "panic") {
+				continue
+			}
+			bt = append(bt, strings.TrimSpace(row))
+		}
+
+		analytics.Client.Publish("panic", map[string]interface{}{"backtrace": strings.Join(bt, "\n")})
+		log.Fatalf("PANIC: %v\n", err)
 	}
 }