Fatal error on terminal job

pkg/engine/health/health.go

@@ -16,6 +16,30 @@ import (
 	kudov1beta1 "github.com/kudobuilder/kudo/pkg/apis/kudo/v1beta1"
 )
 
+// IsTerminallyFailed returns whether an object is in a terminal failed state and has no chance to reach healthy
+func IsTerminallyFailed(obj runtime.Object) (bool, string) {
+	if obj == nil {
+		return false, ""
+	}
+
+	switch obj := obj.(type) {
+	case *batchv1.Job:
+		return isJobTerminallyFailed(obj)
+	default:
+		return false, ""
+	}
+}
+
+func isJobTerminallyFailed(job *batchv1.Job) (bool, string) {
+	for _, c := range job.Status.Conditions {
+		if c.Type == batchv1.JobFailed && c.Status == corev1.ConditionTrue {
+			log.Printf("HealthUtil: Job \"%v\" has failed: %s", job.Name, c.Message)
+			return true, c.Message
+		}
+	}
+	return false, ""
+}
+
 // IsHealthy returns whether an object is healthy. Must be implemented for each type.
 func IsHealthy(obj runtime.Object) error {
 	if obj == nil {

pkg/engine/task/task.go

@@ -53,6 +53,7 @@ var (
 	dummyTaskError          = "DummyTaskError"
 	resourceUnmarshalError  = "ResourceUnmarshalError"
 	resourceValidationError = "ResourceValidationError"
+	failedTerminalState     = "FailedTerminalStateError"
 )
 
 // Build factory method takes an v1beta1.Task and returns a corresponding Tasker object

pkg/engine/task/task_apply.go

@@ -58,7 +58,9 @@ func (at ApplyTask) Run(ctx Context) (bool, error) {
 	// 4. - Check health for all resources -
 	err = isHealthy(applied)
 	if err != nil {
-		// so far we do not distinguish between unhealthy resources and other errors that might occur during a health check
+		if fatal := isTerminallyFailed(applied); fatal != nil {
+			return false, fatalExecutionError(fatal, failedTerminalState, ctx.Meta)
+		}
 		// an error during a health check is not treated task execution error
 		log.Printf("TaskExecution: %v", err)
 		return false, nil
@@ -237,6 +239,16 @@ func isHealthy(ro []runtime.Object) error {
 	return nil
 }
 
+func isTerminallyFailed(ro []runtime.Object) error {
+	for _, r := range ro {
+		if failed, msg := health.IsTerminallyFailed(r); failed {
+			key, _ := client.ObjectKeyFromObject(r)
+			return fmt.Errorf("object %s/%s has failed: %s", key.Namespace, key.Name, msg)
+		}
+	}
+	return nil
+}
+
 // copy from k8s.io/kubectl@v0.16.6/pkg/util/apply.go, with adjustments
 // GetOriginalConfiguration retrieves the original configuration of the object
 // from the annotation.

pkg/engine/task/task_pipe.go

@@ -97,6 +97,9 @@ func (pt PipeTask) Run(ctx Context) (bool, error) {
 	// once the pod is Ready, it means that its initContainer finished successfully and we can copy
 	// out the generated files. An error during a health check is not treated as task execution error
 	if err != nil {
+		if fatal := isTerminallyFailed(podObj); fatal != nil {
+			return false, fatalExecutionError(fatal, failedTerminalState, ctx.Meta)
+		}
 		return false, nil
 	}
 

test/e2e/terminal-failed-job/00-assert.yaml

@@ -0,0 +1,11 @@
+apiVersion: kudo.dev/v1beta1
+kind: Instance
+metadata:
+  labels:
+    kudo.dev/operator: failjob-operator
+spec:
+  operatorVersion:
+    name: failjob-operator-0.1.0
+status:
+  aggregatedStatus:
+    status: FATAL_ERROR

test/e2e/terminal-failed-job/00-install.yaml

@@ -0,0 +1,4 @@
+apiVersion: kudo.dev/v1beta1
+kind: TestStep
+kubectl:
+- kudo install --instance failjob-operator-instance ./failjob-operator

test/e2e/terminal-failed-job/01-assert.yaml

@@ -0,0 +1,11 @@
+apiVersion: kudo.dev/v1beta1
+kind: Instance
+metadata:
+  labels:
+    kudo.dev/operator: job-timeout-operator
+spec:
+  operatorVersion:
+    name: job-timeout-operator-0.1.0
+status:
+  aggregatedStatus:
+    status: FATAL_ERROR

test/e2e/terminal-failed-job/01-install-timeout.yaml

@@ -0,0 +1,5 @@
+apiVersion: kudo.dev/v1beta1
+kind: TestStep
+kubectl:
+- kudo uninstall --instance failjob-operator-instance
+- kudo install --instance job-timout-operator-instance ./job-timeout-operator

test/e2e/terminal-failed-job/failjob-operator/operator.yaml

@@ -0,0 +1,25 @@
+apiVersion: kudo.dev/v1beta1
+name: "failjob-operator"
+operatorVersion: "0.1.0"
+appVersion: "1.7.9"
+kubernetesVersion: 1.13.0
+maintainers:
+  - name: Your name
+    email: <your@email.com>
+url: https://kudo.dev
+tasks:
+  - name: app
+    kind: Apply
+    spec:
+      resources:
+        - failjob.yaml
+plans:
+  deploy:
+    strategy: serial
+    phases:
+      - name: main
+        strategy: parallel
+        steps:
+          - name: everything
+            tasks:
+              - app

test/e2e/terminal-failed-job/failjob-operator/params.yaml

@@ -0,0 +1,2 @@
+apiVersion: kudo.dev/v1beta1
+parameters:

test/e2e/terminal-failed-job/failjob-operator/templates/failjob.yaml

@@ -0,0 +1,14 @@
+apiVersion: batch/v1
+kind: Job
+metadata:
+  name: failing-job
+  namespace: {{ .Namespace }}
+spec:
+  template:
+    spec:
+      containers:
+        - name: main
+          image: alpine
+          command: ["false"]
+      restartPolicy: Never
+  backoffLimit: 0

test/e2e/terminal-failed-job/job-timeout-operator/operator.yaml

@@ -0,0 +1,25 @@
+apiVersion: kudo.dev/v1beta1
+name: "job-timeout-operator"
+operatorVersion: "0.1.0"
+appVersion: "1.7.9"
+kubernetesVersion: 1.13.0
+maintainers:
+  - name: Your name
+    email: <your@email.com>
+url: https://kudo.dev
+tasks:
+  - name: app
+    kind: Apply
+    spec:
+      resources:
+        - timeoutjob.yaml
+plans:
+  deploy:
+    strategy: serial
+    phases:
+      - name: main
+        strategy: parallel
+        steps:
+          - name: everything
+            tasks:
+              - app

test/e2e/terminal-failed-job/job-timeout-operator/params.yaml

@@ -0,0 +1,2 @@
+apiVersion: kudo.dev/v1beta1
+parameters:

test/e2e/terminal-failed-job/job-timeout-operator/templates/timeoutjob.yaml

@@ -0,0 +1,15 @@
+apiVersion: batch/v1
+kind: Job
+metadata:
+  name: timeout-job
+  namespace: {{ .Namespace }}
+spec:
+  template:
+    spec:
+      containers:
+        - name: main
+          image: alpine
+          command: ["sleep", "1000"]
+      restartPolicy: Never
+  backoffLimit: 3
+  activeDeadlineSeconds: 60