quadlet kube: correctly mark unit as failed

When no containers could be started we need to make sure the unit status
reflects this. This means we should not send the READ=1 message and not
keep the service container running when we were unable to start any
container.

There is the question what should happen when only a subset was started.
For systemd we can only be either running or failed. And as podman kube
play also just keeps the partial started pods running I opted to let
systemd keep considering this as success.

Fixes #20667
Fixes https://issues.redhat.com/browse/RHEL-80471

Signed-off-by: Paul Holzinger <pholzing@redhat.com>

Author: Luap99

pkg/domain/infra/abi/play.go

@@ -283,6 +283,19 @@ func (ic *ContainerEngine) PlayKube(ctx context.Context, body io.Reader, options
 	var configMaps []v1.ConfigMap
 
 	ranContainers := false
+	// set the ranContainers bool to true if at least one container was successfully started.
+	setRanContainers := func(r *entities.PlayKubeReport) {
+		if !ranContainers {
+			for _, p := range r.Pods {
+				// If the list of container errors is less then the total number of pod containers then we know it didn't start.
+				if len(p.ContainerErrors) < len(p.Containers)+len(p.InitContainers) {
+					ranContainers = true
+					break
+				}
+			}
+		}
+	}
+
 	// FIXME: both, the service container and the proxies, should ideally
 	// be _state_ of an object. The Kube code below is quite Spaghetti-code
 	// which we should refactor at some point to make it easier to extend
@@ -364,7 +377,7 @@ func (ic *ContainerEngine) PlayKube(ctx context.Context, body io.Reader, options
 
 			report.Pods = append(report.Pods, r.Pods...)
 			validKinds++
-			ranContainers = true
+			setRanContainers(r)
 		case "DaemonSet":
 			var daemonSetYAML v1apps.DaemonSet
 
@@ -380,7 +393,7 @@ func (ic *ContainerEngine) PlayKube(ctx context.Context, body io.Reader, options
 
 			report.Pods = append(report.Pods, r.Pods...)
 			validKinds++
-			ranContainers = true
+			setRanContainers(r)
 		case "Deployment":
 			var deploymentYAML v1apps.Deployment
 
@@ -396,7 +409,7 @@ func (ic *ContainerEngine) PlayKube(ctx context.Context, body io.Reader, options
 
 			report.Pods = append(report.Pods, r.Pods...)
 			validKinds++
-			ranContainers = true
+			setRanContainers(r)
 		case "Job":
 			var jobYAML v1.Job
 
@@ -412,7 +425,7 @@ func (ic *ContainerEngine) PlayKube(ctx context.Context, body io.Reader, options
 
 			report.Pods = append(report.Pods, r.Pods...)
 			validKinds++
-			ranContainers = true
+			setRanContainers(r)
 		case "PersistentVolumeClaim":
 			var pvcYAML v1.PersistentVolumeClaim
 
@@ -473,10 +486,14 @@ func (ic *ContainerEngine) PlayKube(ctx context.Context, body io.Reader, options
 		return nil, fmt.Errorf("YAML document does not contain any supported kube kind")
 	}
 
+	if !options.ServiceContainer {
+		return report, nil
+	}
+
 	// If we started containers along with a service container, we are
 	// running inside a systemd unit and need to set the main PID.
 
-	if options.ServiceContainer && ranContainers {
+	if ranContainers {
 		switch len(notifyProxies) {
 		case 0: // Optimization for containers/podman/issues/17345
 			// No container needs sdnotify, so we can mark the
@@ -509,6 +526,12 @@ func (ic *ContainerEngine) PlayKube(ctx context.Context, body io.Reader, options
 		}
 
 		report.ServiceContainerID = serviceContainer.ID()
+	} else if serviceContainer != nil {
+		// No containers started, make sure to stop the service container.
+		// Note because the pods still do exists and are not removed by default we cannot remove it.
+		if err := serviceContainer.StopWithTimeout(0); err != nil {
+			logrus.Errorf("Failed to stop service container: %v", err)
+		}
 	}
 
 	return report, nil

test/system/252-quadlet.bats

@@ -1107,6 +1107,59 @@ EOF
     service_cleanup $QUADLET_SERVICE_NAME inactive
 }
 
+# https://github.com/containers/podman/issues/20667
+@test "quadlet kube - start error" {
+    local port=$(random_free_port)
+    # Create the YAMl file
+    pod_name="p-$(safename)"
+    container_name="c-$(safename)"
+    yaml_source="$PODMAN_TMPDIR/start_err$(safename).yaml"
+    cat >$yaml_source <<EOF
+apiVersion: v1
+kind: Pod
+metadata:
+  name: $pod_name
+spec:
+  containers:
+  - command:
+    - "top"
+    image: $IMAGE
+    name: $container_name
+    ports:
+    - containerPort: 80
+      hostPort: $port
+EOF
+
+    # Bind the port to force a an error when starting the pod
+    timeout --foreground -v --kill=10 10 nc -l 127.0.0.1 $port &
+    nc_pid=$!
+
+    # Create the Quadlet file
+    local quadlet_file=$PODMAN_TMPDIR/start_err_$(safename).kube
+    cat > $quadlet_file <<EOF
+[Kube]
+Yaml=${yaml_source}
+EOF
+
+    run_quadlet "$quadlet_file"
+
+    run -0 systemctl daemon-reload
+
+    echo "$_LOG_PROMPT systemctl start $QUADLET_SERVICE_NAME"
+    run systemctl start $QUADLET_SERVICE_NAME
+    echo $output
+    assert $status -eq 1 "systemctl start should report failure"
+
+    run -0 systemctl show --property=ActiveState $QUADLET_SERVICE_NAME
+    assert "$output" == "ActiveState=failed" "unit must be in failed state"
+
+    echo "$_LOG_PROMPT journalctl -u $QUADLET_SERVICE_NAME"
+    run -0 journalctl -eu $QUADLET_SERVICE_NAME
+    assert "$output" =~ "$port: bind: address already in use" "journal contains the real podman start error"
+
+    kill "$nc_pid"
+}
+
 @test "quadlet - image files" {
     local quadlet_tmpdir=$PODMAN_TMPDIR/quadlets