fleet cli json output

cmd/fleetcli/main.go

@@ -3,6 +3,10 @@ package main
 
 import (
 	// Ensure GVKs are registered
+
+	"os"
+	"strings"
+
 	_ "github.com/rancher/fleet/pkg/generated/controllers/fleet.cattle.io"
 	_ "github.com/rancher/wrangler/v3/pkg/generated/controllers/apiextensions.k8s.io"
 	_ "github.com/rancher/wrangler/v3/pkg/generated/controllers/apps"
@@ -22,7 +26,16 @@ func main() {
 	ctx := signals.SetupSignalContext()
 	cmd := cmds.App()
 	if err := cmd.ExecuteContext(ctx); err != nil {
-		logrus.Fatal(err)
+		if strings.ToLower(os.Getenv("FLEET_JSON_OUTPUT")) == "true" {
+			log := logrus.New()
+			log.SetFormatter(&logrus.JSONFormatter{})
+			// use a fleet specific field name so we are sure logs from other libraries
+			// are not considered.
+			log.WithFields(logrus.Fields{
+				"fleetErrorMessage": err.Error(),
+			}).Fatal("Fleet cli failed")
+		} else {
+			logrus.Fatal(err)
+		}
 	}
-
 }

integrationtests/gitjob/controller/controller_test.go

@@ -542,10 +542,15 @@ var _ = Describe("GitJob controller", func() {
 					job.Status.StartTime = &metav1.Time{Time: time.Now().Add(-1 * time.Hour)}
 					job.Status.Conditions = []batchv1.JobCondition{
 						{
-							Type:    "Failed",
+							// using Stalled because the Compute function uses Stalled
+							// for returning the condition message and it's simpler.
+							// For testing it in a different way we would need to setup a more complex
+							// scenario defining the job pods
+							// We are simulating job failures.
+							Type:    "Stalled",
 							Status:  "True",
 							Reason:  "BackoffLimitExceeded",
-							Message: "Job has reached the specified backoff limit",
+							Message: `{"fleetErrorMessage":"fleet error message","level":"fatal","msg":"Fleet cli failed","time":"2025-04-15T14:53:15+02:00"}`,
 						},
 						{
 							Type:   batchv1.JobFailureTarget,
@@ -566,7 +571,7 @@ var _ = Describe("GitJob controller", func() {
 					// check the conditions related to the job
 					// Failed.... Stalled=true and Reconcilling=false
 					g.Expect(checkCondition(&gitRepo, "Reconciling", corev1.ConditionFalse, "")).To(BeTrue())
-					g.Expect(checkCondition(&gitRepo, "Stalled", corev1.ConditionTrue, "Job Failed. failed: 1/1")).To(BeTrue())
+					g.Expect(checkCondition(&gitRepo, "Stalled", corev1.ConditionTrue, "fleet error message")).To(BeTrue())
 
 					// check the rest
 					g.Expect(checkCondition(&gitRepo, "GitPolling", corev1.ConditionTrue, "")).To(BeTrue())

internal/cmd/controller/gitops/reconciler/gitjob_controller.go

@@ -959,6 +959,10 @@ func argsAndEnvs(
 			Name:  "HOME",
 			Value: fleetHomeDir,
 		},
+		{
+			Name:  "FLEET_JSON_OUTPUT",
+			Value: "true",
+		},
 		{
 			Name:  fleetcli.FleetApplyConflictRetriesEnv,
 			Value: strconv.Itoa(fleetApplyRetries),
@@ -1278,7 +1282,13 @@ func (r *GitJobReconciler) newGitCloner(
 		args = append(args, "--ca-bundle-file", "/gitjob/cabundle/"+bundleCAFile)
 	}
 
-	env := proxyEnvVars()
+	env := []corev1.EnvVar{
+		{
+			Name:  "FLEET_JSON_OUTPUT",
+			Value: "true",
+		},
+	}
+	env = append(env, proxyEnvVars()...)
 
 	// If strict host key checks are enabled but no entries are available, another error will be shown by the known
 	// hosts getter, as that means that the Fleet deployment is incomplete.
@@ -1495,7 +1505,7 @@ func setStatusFromGitjob(ctx context.Context, c client.Client, gitRepo *v1alpha1
 	//   - Terminating
 	switch result.Status {
 	case status.FailedStatus:
-		kstatus.SetError(gitRepo, terminationMessage)
+		kstatus.SetError(gitRepo, filterFleetCLIJobOutput(terminationMessage))
 	case status.CurrentStatus:
 		if strings.Contains(result.Message, "Job Completed") {
 			gitRepo.Status.Commit = job.Annotations["commit"]
@@ -1632,3 +1642,57 @@ func separatorInBundleDefinitions(gitrepo v1alpha1.GitRepo, sep rune) bool {
 
 	return false
 }
+
+func filterFleetCLIJobOutput(output string) string {
+	// first split the output in lines
+	lines := strings.Split(output, "\n")
+	s := ""
+	for _, l := range lines {
+		s = s + getFleetCLIErrorsFromLine(l)
+	}
+
+	s = strings.Trim(s, "\n")
+	// in the case that all the messages from fleet apply are from libraries
+	// we just report an unknown error
+	if s == "" {
+		s = "Unknown error"
+	}
+
+	return s
+}
+
+func getFleetCLIErrorsFromLine(l string) string {
+	type LogEntry struct {
+		Level         string `json:"level"`
+		FleetErrorMsg string `json:"fleetErrorMessage"`
+		Time          string `json:"time"`
+		Msg           string `json:"msg"`
+	}
+	s := ""
+	open := strings.IndexByte(l, '{')
+	if open == -1 {
+		// line does not contain a valid json string
+		return ""
+	}
+	close := strings.IndexByte(l, '}')
+	if close != -1 {
+		if close < open {
+			// looks like there is some garbage before a possible json string
+			// ignore everything up to that closing bracked and try again
+			return getFleetCLIErrorsFromLine(l[close+1:])
+		}
+	} else if close == -1 {
+		// line does not contain a valid json string
+		return ""
+	}
+	var entry LogEntry
+	if err := json.Unmarshal([]byte(l[open:close+1]), &entry); err == nil {
+		s = s + entry.FleetErrorMsg + "\n"
+	}
+	// check if there's more to parse
+	if close+1 < len(l) {
+		s = s + getFleetCLIErrorsFromLine(l[close+1:])
+	}
+
+	return s
+}