Fix docker health state

Author: thielepaul

container/docker/handler.go

@@ -68,9 +68,8 @@ type dockerContainerHandler struct {
 	creationTime time.Time
 
 	// Metadata associated with the container.
-	envs         map[string]string
-	labels       map[string]string
-	healthStatus string
+	envs   map[string]string
+	labels map[string]string
 
 	// Image name used for this container.
 	image string
@@ -93,6 +92,9 @@ type dockerContainerHandler struct {
 	reference info.ContainerReference
 
 	libcontainerHandler *containerlibcontainer.Handler
+
+	// the docker client is needed to inspect the container and get the health status
+	client docker.APIClient
 }
 
 var _ container.ContainerHandler = &dockerContainerHandler{}
@@ -201,10 +203,7 @@ func newDockerContainerHandler(
 		labels:             ctnr.Config.Labels,
 		includedMetrics:    metrics,
 		zfsParent:          zfsParent,
-	}
-	// Health status may be nil if no health check is configured
-	if ctnr.State.Health != nil {
-		handler.healthStatus = ctnr.State.Health.Status
+		client:             client,
 	}
 	// Timestamp returned by Docker is in time.RFC3339Nano format.
 	handler.creationTime, err = time.Parse(time.RFC3339Nano, ctnr.Created)
@@ -331,7 +330,16 @@ func (h *dockerContainerHandler) GetStats() (*info.ContainerStats, error) {
 	if err != nil {
 		return stats, err
 	}
-	stats.Health.Status = h.healthStatus
+
+	// We assume that if Inspect fails then the container is not known to docker.
+	ctnr, err := h.client.ContainerInspect(context.Background(), h.reference.Id)
+	if err != nil {
+		return nil, fmt.Errorf("failed to inspect container %q: %v", h.reference.Id, err)
+	}
+
+	if ctnr.State.Health != nil {
+		stats.Health.Status = ctnr.State.Health.Status
+	}
 
 	// Get filesystem stats.
 	err = FsStats(stats, h.machineInfoFactory, h.includedMetrics, h.storageDriver,

metrics/prometheus.go

@@ -139,19 +139,7 @@ func NewPrometheusCollector(i infoProvider, f ContainerLabelsFunc, includedMetri
 				name:      "container_health_state",
 				help:      "The result of the container's health check",
 				valueType: prometheus.GaugeValue,
-				getValues: func(s *info.ContainerStats) metricValues {
-					return metricValues{{
-						// inline if to check if s.health.status = healthy
-						value: func(s *info.ContainerStats) float64 {
-							if s.Health.Status == "healthy" {
-								return 1
-							} else {
-								return 0
-							}
-						}(s),
-						timestamp: s.Timestamp,
-					}}
-				},
+				getValues: getContainerHealthState,
 			},
 		},
 		includedMetrics: includedMetrics,
@@ -2109,3 +2097,18 @@ func getMinCoreScalingRatio(s *info.ContainerStats) metricValues {
 	}
 	return values
 }
+
+func getContainerHealthState(s *info.ContainerStats) metricValues {
+	value := float64(0)
+	switch s.Health.Status {
+	case "healthy":
+		value = 1
+	case "": // if container has no health check defined
+		value = -1
+	default: // starting or unhealthy
+	}
+	return metricValues{{
+		value:     value,
+		timestamp: s.Timestamp,
+	}}
+}

metrics/prometheus_test.go

@@ -336,3 +336,23 @@ func TestGetMinCoreScalingRatio(t *testing.T) {
 	assert.Contains(t, values, 0.5)
 	assert.Contains(t, values, 0.3)
 }
+
+func TestGetContainerHealthState(t *testing.T) {
+	testCases := []struct {
+		name           string
+		containerStats *info.ContainerStats
+		expectedValue  float64
+	}{
+		{name: "healthy", expectedValue: 1.0, containerStats: &info.ContainerStats{Health: info.Health{Status: "healthy"}}},
+		{name: "unhealthy", expectedValue: 0.0, containerStats: &info.ContainerStats{Health: info.Health{Status: "unhealthy"}}},
+		{name: "starting", expectedValue: 0.0, containerStats: &info.ContainerStats{Health: info.Health{Status: "unknown"}}},
+		{name: "empty", expectedValue: -1.0, containerStats: &info.ContainerStats{}},
+	}
+	for _, tc := range testCases {
+		t.Run(tc.name, func(t *testing.T) {
+			metricVals := getContainerHealthState(tc.containerStats)
+			assert.Equal(t, 1, len(metricVals))
+			assert.Equal(t, tc.expectedValue, metricVals[0].value)
+		})
+	}
+}