|
| 1 | +package pkg |
| 2 | + |
| 3 | +import ( |
| 4 | + "context" |
| 5 | + "errors" |
| 6 | + "net/http" |
| 7 | + "strings" |
| 8 | + "sync" |
| 9 | + "time" |
| 10 | + |
| 11 | + corev1 "k8s.io/api/core/v1" |
| 12 | + |
| 13 | + "github.com/prometheus/client_golang/prometheus" |
| 14 | + "github.com/prometheus/client_golang/prometheus/promhttp" |
| 15 | + "k8s.io/klog/v2" |
| 16 | +) |
| 17 | + |
| 18 | +const ( |
| 19 | + metricsPath = "/metrics" |
| 20 | + evictionMetricName = "soft_pod_memory_evicter_evicted_pods_total" |
| 21 | + evictionMetricHelp = "Total number of Pods evicted by soft-pod-memory-evicter." |
| 22 | + metricsShutdownGrace = 5 * time.Second |
| 23 | + metricsReadHeaderTimeout = 4 * time.Second |
| 24 | + |
| 25 | + metricLabelNamespace = "affected_namespace" |
| 26 | + metricLabelAppName = "affected_app_kubernetes_io_name" |
| 27 | + metricLabelAppInstance = "affected_app_kubernetes_io_instance" |
| 28 | +) |
| 29 | + |
| 30 | +type metricsRecorder interface { |
| 31 | + Start(ctx context.Context) |
| 32 | + RecordPodEviction(pod *corev1.Pod) |
| 33 | + RecordObservedNamespace(pod *corev1.Pod) |
| 34 | +} |
| 35 | + |
| 36 | +func newMetricsRecorder(opts Options) metricsRecorder { |
| 37 | + if !opts.EnableMetrics { |
| 38 | + return noopMetricsRecorder{} |
| 39 | + } |
| 40 | + |
| 41 | + bindAddr := strings.TrimSpace(opts.MetricsBindAddress) |
| 42 | + if bindAddr == "" { |
| 43 | + klog.Warning("Metrics exporter enabled but no bind address provided; disabling exporter") |
| 44 | + return noopMetricsRecorder{} |
| 45 | + } |
| 46 | + |
| 47 | + counter := prometheus.NewCounterVec( |
| 48 | + prometheus.CounterOpts{ |
| 49 | + Name: evictionMetricName, |
| 50 | + Help: evictionMetricHelp, |
| 51 | + }, |
| 52 | + []string{ |
| 53 | + metricLabelNamespace, |
| 54 | + metricLabelAppName, |
| 55 | + metricLabelAppInstance, |
| 56 | + }, |
| 57 | + ) |
| 58 | + |
| 59 | + registry := prometheus.NewRegistry() |
| 60 | + if err := registry.Register(counter); err != nil { |
| 61 | + klog.Errorf("unable to register metrics exporter: %v", err) |
| 62 | + return noopMetricsRecorder{} |
| 63 | + } |
| 64 | + |
| 65 | + mux := http.NewServeMux() |
| 66 | + mux.Handle(metricsPath, promhttp.HandlerFor(registry, promhttp.HandlerOpts{})) |
| 67 | + |
| 68 | + return &prometheusMetricsRecorder{ |
| 69 | + server: &http.Server{ |
| 70 | + Addr: bindAddr, |
| 71 | + Handler: mux, |
| 72 | + ReadHeaderTimeout: metricsReadHeaderTimeout, |
| 73 | + }, |
| 74 | + counter: counter, |
| 75 | + } |
| 76 | +} |
| 77 | + |
| 78 | +type noopMetricsRecorder struct{} |
| 79 | + |
| 80 | +func (noopMetricsRecorder) Start(context.Context) {} |
| 81 | + |
| 82 | +func (noopMetricsRecorder) RecordPodEviction(*corev1.Pod) {} |
| 83 | + |
| 84 | +func (noopMetricsRecorder) RecordObservedNamespace(*corev1.Pod) {} |
| 85 | + |
| 86 | +type prometheusMetricsRecorder struct { |
| 87 | + server *http.Server |
| 88 | + counter *prometheus.CounterVec |
| 89 | + once sync.Once |
| 90 | +} |
| 91 | + |
| 92 | +func (r *prometheusMetricsRecorder) Start(ctx context.Context) { |
| 93 | + r.once.Do(func() { |
| 94 | + go func() { |
| 95 | + <-ctx.Done() |
| 96 | + |
| 97 | + shutdownCtx, cancel := context.WithTimeout(context.Background(), metricsShutdownGrace) |
| 98 | + defer cancel() |
| 99 | + |
| 100 | + if err := r.server.Shutdown(shutdownCtx); err != nil { |
| 101 | + klog.Errorf("unable to shutdown metrics exporter: %v", err) |
| 102 | + } |
| 103 | + }() |
| 104 | + |
| 105 | + go func() { |
| 106 | + klog.Infof("Prometheus exporter listening on %s%s", r.server.Addr, metricsPath) |
| 107 | + if err := r.server.ListenAndServe(); err != nil && !errors.Is(err, http.ErrServerClosed) { |
| 108 | + klog.Errorf("Prometheus exporter failed: %v", err) |
| 109 | + } |
| 110 | + }() |
| 111 | + }) |
| 112 | +} |
| 113 | + |
| 114 | +func (r *prometheusMetricsRecorder) RecordPodEviction(pod *corev1.Pod) { |
| 115 | + r.counter.WithLabelValues( |
| 116 | + pod.Namespace, |
| 117 | + pod.Labels[appKubernetesNameLabel], |
| 118 | + pod.Labels[appKubernetesInstanceLabel], |
| 119 | + ).Inc() |
| 120 | +} |
| 121 | + |
| 122 | +func (r *prometheusMetricsRecorder) RecordObservedNamespace(pod *corev1.Pod) { |
| 123 | + r.counter.WithLabelValues( |
| 124 | + pod.Namespace, |
| 125 | + pod.Labels[appKubernetesNameLabel], |
| 126 | + pod.Labels[appKubernetesInstanceLabel], |
| 127 | + ) |
| 128 | +} |
0 commit comments