add process metrics

Signed-off-by: Akhil Mohan <[email protected]>

Author: akhilerm

internal/cri/server/list_metric_descriptors.go

@@ -22,4 +22,5 @@ const (
 	NetworkUsageMetrics = "network"
 	DiskIOMetrics       = "diskIO"
 	DiskUsageMetrics    = "disk"
+	ProcessMetrics      = "process"
 )

internal/cri/server/list_metric_descriptors_linux.go

@@ -265,6 +265,38 @@ func (c *criService) getMetricDescriptors() map[string][]*runtime.MetricDescript
 				LabelKeys: append(baseLabelKeys, "device", "major", "minor", "operation"),
 			},
 		},
+		ProcessMetrics: {
+			{
+				Name:      "container_processes",
+				Help:      "Number of processes running inside the container.",
+				LabelKeys: baseLabelKeys,
+			},
+			/*{
+				Name:      "container_file_descriptors",
+				Help:      "Number of open file descriptors for the container.",
+				LabelKeys: baseLabelKeys,
+			},
+			{
+				Name:      "container_sockets",
+				Help:      "Number of open sockets for the container.",
+				LabelKeys: baseLabelKeys,
+			},*/
+			{
+				Name:      "container_threads_max",
+				Help:      "Maximum number of threads allowed inside the container, infinity if value is zero",
+				LabelKeys: baseLabelKeys,
+			},
+			/*{
+				Name:      "container_threads",
+				Help:      "Number of threads running inside the container",
+				LabelKeys: baseLabelKeys,
+			},
+			{
+				Name:      "container_ulimits_soft",
+				Help:      "Soft ulimit values for the container root process. Unlimited if -1, except priority and nice",
+				LabelKeys: append(baseLabelKeys, "ulimit"),
+			},*/
+		},
 	}
 	return descriptors
 }

internal/cri/server/list_pod_sandbox_metrics_linux.go

@@ -275,6 +275,14 @@ func (c *criService) collectContainerMetrics(ctx context.Context, container cont
 		containerMetrics.Metrics = append(containerMetrics.Metrics, fsMetrics...)
 	}
 
+	// Collect process metrics
+	processMetrics, err := c.extractProcessMetrics(stats, containerLabels, timestamp)
+	if err != nil {
+		log.G(ctx).WithField("containerid", container.ID).WithError(err).Debug("failed to extract process metrics")
+	} else {
+		containerMetrics.Metrics = append(containerMetrics.Metrics, processMetrics...)
+	}
+
 	return containerMetrics, nil
 }
 
@@ -805,6 +813,58 @@ func (c *criService) extractDiskMetrics(stats interface{}, labels []string, time
 	return metrics, nil
 }
 
+// extractProcessMetrics extracts process-related metrics from container stats
+func (c *criService) extractProcessMetrics(stats interface{}, labels []string, timestamp int64) ([]*runtime.Metric, error) {
+	var metrics []*runtime.Metric
+
+	switch s := stats.(type) {
+	case *cg1.Metrics:
+		if s.Pids != nil {
+			metrics = append(metrics, []*runtime.Metric{
+				{
+					Name:        "container_processes",
+					Timestamp:   timestamp,
+					MetricType:  runtime.MetricType_GAUGE,
+					LabelValues: labels,
+					Value:       &runtime.UInt64Value{Value: s.Pids.Current},
+				},
+				{
+					Name:        "container_threads_max",
+					Timestamp:   timestamp,
+					MetricType:  runtime.MetricType_GAUGE,
+					LabelValues: labels,
+					Value:       &runtime.UInt64Value{Value: s.Pids.Limit},
+				},
+			}...)
+		}
+
+	case *cg2.Metrics:
+		if s.Pids != nil {
+			metrics = append(metrics, []*runtime.Metric{
+				{
+					Name:        "container_processes",
+					Timestamp:   timestamp,
+					MetricType:  runtime.MetricType_GAUGE,
+					LabelValues: labels,
+					Value:       &runtime.UInt64Value{Value: s.Pids.Current},
+				},
+				{
+					Name:        "container_threads_max",
+					Timestamp:   timestamp,
+					MetricType:  runtime.MetricType_GAUGE,
+					LabelValues: labels,
+					Value:       &runtime.UInt64Value{Value: s.Pids.Limit},
+				},
+			}...)
+		}
+
+	default:
+		return nil, fmt.Errorf("unexpected metrics type: %T from %s", s, reflect.TypeOf(s).Elem().PkgPath())
+	}
+
+	return metrics, nil
+}
+
 // extractFilesystemMetrics extracts filesystem-related metrics from container stats
 func (c *criService) extractFilesystemMetrics(ctx context.Context, containerID string, labels []string, timestamp int64) ([]*runtime.Metric, error) {
 	var metrics []*runtime.Metric