| 1 | // SPDX-License-Identifier: GPL-3.0-or-later |
| 2 | |
| 3 | package k8s_state |
| 4 | |
| 5 | import ( |
| 6 | "context" |
| 7 | "os" |
| 8 | "sync" |
| 9 | "time" |
| 10 | |
| 11 | "github.com/netdata/netdata/go/plugins/logger" |
| 12 | |
| 13 | appsv1 "k8s.io/api/apps/v1" |
| 14 | batchv1 "k8s.io/api/batch/v1" |
| 15 | corev1 "k8s.io/api/core/v1" |
| 16 | metav1 "k8s.io/apimachinery/pkg/apis/meta/v1" |
| 17 | "k8s.io/apimachinery/pkg/runtime" |
| 18 | "k8s.io/apimachinery/pkg/watch" |
| 19 | "k8s.io/client-go/kubernetes" |
| 20 | "k8s.io/client-go/tools/cache" |
| 21 | "k8s.io/client-go/util/workqueue" |
| 22 | ) |
| 23 | |
| 24 | type discoverer interface { |
| 25 | run(ctx context.Context, in chan<- resource) |
| 26 | ready() bool |
| 27 | stopped() bool |
| 28 | } |
| 29 | |
| 30 | func newKubeDiscovery(client kubernetes.Interface, l *logger.Logger) *kubeDiscovery { |
| 31 | return &kubeDiscovery{ |
| 32 | client: client, |
| 33 | Logger: l, |
| 34 | readyCh: make(chan struct{}), |
| 35 | stopCh: make(chan struct{}), |
| 36 | } |
| 37 | } |
| 38 | |
| 39 | type kubeDiscovery struct { |
| 40 | *logger.Logger |
| 41 | client kubernetes.Interface |
| 42 | discoverers []discoverer |
| 43 | readyCh chan struct{} |
| 44 | stopCh chan struct{} |
| 45 | } |
| 46 | |
| 47 | func (d *kubeDiscovery) run(ctx context.Context, in chan<- resource) { |
| 48 | d.Info("kube_discoverer is started") |
| 49 | defer func() { close(d.stopCh); d.Info("kube_discoverer is stopped") }() |
| 50 | |
| 51 | d.discoverers = d.setupDiscoverers(ctx) |
| 52 | |
| 53 | var wg sync.WaitGroup |
| 54 | updates := make(chan resource) |
| 55 | |
| 56 | for _, dd := range d.discoverers { |
| 57 | wg.Add(1) |
| 58 | go func(dd discoverer) { defer wg.Done(); dd.run(ctx, updates) }(dd) |
| 59 | } |
| 60 | |
| 61 | wg.Go(func() { d.runDiscover(ctx, updates, in) }) |
| 62 | |
| 63 | close(d.readyCh) |
| 64 | wg.Wait() |
| 65 | <-ctx.Done() |
| 66 | } |
| 67 | |
| 68 | func (d *kubeDiscovery) ready() bool { |
| 69 | if !isChanClosed(d.readyCh) { |
| 70 | return false |
| 71 | } |
| 72 | for _, dd := range d.discoverers { |
| 73 | if !dd.ready() { |
| 74 | return false |
| 75 | } |
| 76 | } |
| 77 | return true |
| 78 | } |
| 79 | |
| 80 | func (d *kubeDiscovery) stopped() bool { |
| 81 | if !isChanClosed(d.stopCh) { |
| 82 | return false |
| 83 | } |
| 84 | for _, dd := range d.discoverers { |
| 85 | if !dd.stopped() { |
| 86 | return false |
| 87 | } |
| 88 | } |
| 89 | return true |
| 90 | } |
| 91 | |
| 92 | func (d *kubeDiscovery) runDiscover(ctx context.Context, updates chan resource, in chan<- resource) { |
| 93 | for { |
| 94 | select { |
| 95 | case <-ctx.Done(): |
| 96 | return |
| 97 | case r := <-updates: |
| 98 | select { |
| 99 | case <-ctx.Done(): |
| 100 | return |
| 101 | case in <- r: |
| 102 | } |
| 103 | } |
| 104 | } |
| 105 | } |
| 106 | |
| 107 | const resyncPeriod = 10 * time.Minute |
| 108 | |
| 109 | var ( |
| 110 | myNodeName = os.Getenv("MY_NODE_NAME") |
| 111 | ) |
| 112 | |
| 113 | func (d *kubeDiscovery) setupDiscoverers(ctx context.Context) []discoverer { |
| 114 | node := d.client.CoreV1().Nodes() |
| 115 | nodeWatcher := cache.ToListWatcherWithWatchListSemantics(&cache.ListWatch{ |
| 116 | ListWithContextFunc: func(_ context.Context, options metav1.ListOptions) (runtime.Object, error) { |
| 117 | return node.List(ctx, options) |
| 118 | }, |
| 119 | WatchFuncWithContext: func(_ context.Context, options metav1.ListOptions) (watch.Interface, error) { |
| 120 | return node.Watch(ctx, options) |
| 121 | }, |
| 122 | }, d.client) |
| 123 | |
| 124 | pod := d.client.CoreV1().Pods(corev1.NamespaceAll) |
| 125 | podWatcher := cache.ToListWatcherWithWatchListSemantics(&cache.ListWatch{ |
| 126 | ListWithContextFunc: func(_ context.Context, options metav1.ListOptions) (runtime.Object, error) { |
| 127 | if myNodeName != "" { |
| 128 | options.FieldSelector = "spec.nodeName=" + myNodeName |
| 129 | } |
| 130 | return pod.List(ctx, options) |
| 131 | }, |
| 132 | WatchFuncWithContext: func(_ context.Context, options metav1.ListOptions) (watch.Interface, error) { |
| 133 | if myNodeName != "" { |
| 134 | options.FieldSelector = "spec.nodeName=" + myNodeName |
| 135 | } |
| 136 | return pod.Watch(ctx, options) |
| 137 | }, |
| 138 | }, d.client) |
| 139 | |
| 140 | deploy := d.client.AppsV1().Deployments(corev1.NamespaceAll) |
| 141 | deployWatcher := cache.ToListWatcherWithWatchListSemantics(&cache.ListWatch{ |
| 142 | ListWithContextFunc: func(_ context.Context, options metav1.ListOptions) (runtime.Object, error) { |
| 143 | return deploy.List(ctx, options) |
| 144 | }, |
| 145 | WatchFuncWithContext: func(_ context.Context, options metav1.ListOptions) (watch.Interface, error) { |
| 146 | return deploy.Watch(ctx, options) |
| 147 | }, |
| 148 | }, d.client) |
| 149 | |
| 150 | cj := d.client.BatchV1().CronJobs(corev1.NamespaceAll) |
| 151 | cjWatcher := cache.ToListWatcherWithWatchListSemantics(&cache.ListWatch{ |
| 152 | ListWithContextFunc: func(_ context.Context, options metav1.ListOptions) (runtime.Object, error) { |
| 153 | return cj.List(ctx, options) |
| 154 | }, |
| 155 | WatchFuncWithContext: func(_ context.Context, options metav1.ListOptions) (watch.Interface, error) { |
| 156 | return cj.Watch(ctx, options) |
| 157 | }, |
| 158 | }, d.client) |
| 159 | |
| 160 | jobs := d.client.BatchV1().Jobs(corev1.NamespaceAll) |
| 161 | jobsWatcher := cache.ToListWatcherWithWatchListSemantics(&cache.ListWatch{ |
| 162 | ListWithContextFunc: func(_ context.Context, options metav1.ListOptions) (runtime.Object, error) { |
| 163 | return jobs.List(ctx, options) |
| 164 | }, |
| 165 | WatchFuncWithContext: func(_ context.Context, options metav1.ListOptions) (watch.Interface, error) { |
| 166 | return jobs.Watch(ctx, options) |
| 167 | }, |
| 168 | }, d.client) |
| 169 | |
| 170 | return []discoverer{ |
| 171 | newNodeDiscoverer(cache.NewSharedInformer(nodeWatcher, &corev1.Node{}, resyncPeriod), d.Logger), |
| 172 | newPodDiscoverer(cache.NewSharedInformer(podWatcher, &corev1.Pod{}, resyncPeriod), d.Logger), |
| 173 | newDeploymentDiscoverer(cache.NewSharedInformer(deployWatcher, &appsv1.Deployment{}, resyncPeriod), d.Logger), |
| 174 | newCronJobDiscoverer(cache.NewSharedInformer(cjWatcher, &batchv1.CronJob{}, resyncPeriod), d.Logger), |
| 175 | newJobDiscoverer(cache.NewSharedInformer(jobsWatcher, &batchv1.Job{}, resyncPeriod), d.Logger), |
| 176 | } |
| 177 | } |
| 178 | |
| 179 | func enqueue(queue *workqueue.Typed[string], obj any) { |
| 180 | key, err := cache.DeletionHandlingMetaNamespaceKeyFunc(obj) |
| 181 | if err != nil { |
| 182 | return |
| 183 | } |
| 184 | queue.Add(key) |
| 185 | } |
| 186 | |
| 187 | func send(ctx context.Context, in chan<- resource, r resource) { |
| 188 | if r == nil { |
| 189 | return |
| 190 | } |
| 191 | select { |
| 192 | case <-ctx.Done(): |
| 193 | case in <- r: |
| 194 | } |
| 195 | } |
| 196 | |
| 197 | func isChanClosed(ch chan struct{}) bool { |
| 198 | select { |
| 199 | case <-ch: |
| 200 | return true |
| 201 | default: |
| 202 | return false |
| 203 | } |
| 204 | } |