| Jan Schär | a48bd3c | 2024-07-29 17:22:18 +0200 | [diff] [blame] | 1 | package kubernetes |
| 2 | |
| 3 | // Taken and modified from the Kubernetes plugin of CoreDNS, under Apache 2.0. |
| 4 | |
| 5 | import ( |
| 6 | "context" |
| 7 | "errors" |
| 8 | "sync/atomic" |
| 9 | "time" |
| 10 | |
| 11 | api "k8s.io/api/core/v1" |
| 12 | discovery "k8s.io/api/discovery/v1" |
| 13 | meta "k8s.io/apimachinery/pkg/apis/meta/v1" |
| 14 | "k8s.io/apimachinery/pkg/runtime" |
| 15 | "k8s.io/apimachinery/pkg/watch" |
| 16 | "k8s.io/client-go/kubernetes" |
| 17 | "k8s.io/client-go/tools/cache" |
| 18 | |
| 19 | "source.monogon.dev/osbase/net/dns/kubernetes/object" |
| 20 | ) |
| 21 | |
| 22 | const ( |
| 23 | svcIPIndex = "ServiceIP" |
| 24 | epNameNamespaceIndex = "EndpointNameNamespace" |
| 25 | epIPIndex = "EndpointsIP" |
| 26 | ) |
| 27 | |
| 28 | // epLabelSelector selects EndpointSlices that belong to headless services. |
| 29 | // Endpoint DNS records are only served for headless services, |
| 30 | // and we can save resources by not fetching other EndpointSlices. |
| 31 | const epLabelSelector = api.IsHeadlessService |
| 32 | |
| 33 | type dnsController interface { |
| 34 | GetSvc(string) *object.Service |
| 35 | SvcIndexReverse(string) []*object.Service |
| 36 | EpIndex(string) []*object.Endpoints |
| 37 | EpIndexReverse(string) []*object.Endpoints |
| 38 | NamespaceExists(string) bool |
| 39 | |
| 40 | Start(<-chan struct{}) |
| 41 | HasSynced() bool |
| 42 | |
| 43 | // Modified returns the timestamp of the most recent changes to services. |
| 44 | Modified() int64 |
| 45 | } |
| 46 | |
| 47 | type dnsControl struct { |
| 48 | // modified tracks timestamp of the most recent changes |
| 49 | modified atomic.Int64 |
| 50 | hasSynced atomic.Bool |
| 51 | |
| 52 | svcController cache.Controller |
| 53 | epController cache.Controller |
| 54 | nsController cache.Controller |
| 55 | |
| 56 | svcLister cache.Indexer |
| 57 | epLister cache.Indexer |
| 58 | nsLister cache.Store |
| 59 | } |
| 60 | |
| 61 | // newdnsController creates a controller. |
| 62 | func newdnsController(ctx context.Context, kubeClient kubernetes.Interface) *dnsControl { |
| 63 | dns := dnsControl{} |
| 64 | |
| 65 | dns.svcLister, dns.svcController = object.NewIndexerInformer( |
| 66 | &cache.ListWatch{ |
| 67 | ListFunc: serviceListFunc(ctx, kubeClient, api.NamespaceAll), |
| 68 | WatchFunc: serviceWatchFunc(ctx, kubeClient, api.NamespaceAll), |
| 69 | }, |
| 70 | &api.Service{}, |
| 71 | &dns, |
| 72 | cache.Indexers{svcIPIndex: svcIPIndexFunc}, |
| 73 | object.DefaultProcessor(object.ToService), |
| 74 | ) |
| 75 | |
| 76 | dns.epLister, dns.epController = object.NewIndexerInformer( |
| 77 | &cache.ListWatch{ |
| 78 | ListFunc: endpointSliceListFunc(ctx, kubeClient, api.NamespaceAll), |
| 79 | WatchFunc: endpointSliceWatchFunc(ctx, kubeClient, api.NamespaceAll), |
| 80 | }, |
| 81 | &discovery.EndpointSlice{}, |
| 82 | &dns, |
| 83 | cache.Indexers{epNameNamespaceIndex: epNameNamespaceIndexFunc, epIPIndex: epIPIndexFunc}, |
| 84 | object.DefaultProcessor(object.EndpointSliceToEndpoints), |
| 85 | ) |
| 86 | |
| 87 | dns.nsLister, dns.nsController = object.NewIndexerInformer( |
| 88 | &cache.ListWatch{ |
| 89 | ListFunc: namespaceListFunc(ctx, kubeClient), |
| 90 | WatchFunc: namespaceWatchFunc(ctx, kubeClient), |
| 91 | }, |
| 92 | &api.Namespace{}, |
| 93 | &dns, |
| 94 | cache.Indexers{}, |
| 95 | object.DefaultProcessor(object.ToNamespace), |
| 96 | ) |
| 97 | |
| 98 | return &dns |
| 99 | } |
| 100 | |
| 101 | func svcIPIndexFunc(obj interface{}) ([]string, error) { |
| 102 | svc, ok := obj.(*object.Service) |
| 103 | if !ok { |
| 104 | return nil, errObj |
| 105 | } |
| 106 | idx := make([]string, len(svc.ClusterIPs)) |
| 107 | copy(idx, svc.ClusterIPs) |
| 108 | return idx, nil |
| 109 | } |
| 110 | |
| 111 | func epNameNamespaceIndexFunc(obj interface{}) ([]string, error) { |
| 112 | s, ok := obj.(*object.Endpoints) |
| 113 | if !ok { |
| 114 | return nil, errObj |
| 115 | } |
| 116 | return []string{s.Index}, nil |
| 117 | } |
| 118 | |
| 119 | func epIPIndexFunc(obj interface{}) ([]string, error) { |
| 120 | ep, ok := obj.(*object.Endpoints) |
| 121 | if !ok { |
| 122 | return nil, errObj |
| 123 | } |
| 124 | idx := make([]string, len(ep.Addresses)) |
| 125 | for i, addr := range ep.Addresses { |
| 126 | idx[i] = addr.IP |
| 127 | } |
| 128 | return idx, nil |
| 129 | } |
| 130 | |
| 131 | func serviceListFunc(ctx context.Context, c kubernetes.Interface, ns string) func(meta.ListOptions) (runtime.Object, error) { |
| 132 | return func(opts meta.ListOptions) (runtime.Object, error) { |
| 133 | return c.CoreV1().Services(ns).List(ctx, opts) |
| 134 | } |
| 135 | } |
| 136 | |
| 137 | func endpointSliceListFunc(ctx context.Context, c kubernetes.Interface, ns string) func(meta.ListOptions) (runtime.Object, error) { |
| 138 | return func(opts meta.ListOptions) (runtime.Object, error) { |
| 139 | opts.LabelSelector = epLabelSelector |
| 140 | return c.DiscoveryV1().EndpointSlices(ns).List(ctx, opts) |
| 141 | } |
| 142 | } |
| 143 | |
| 144 | func namespaceListFunc(ctx context.Context, c kubernetes.Interface) func(meta.ListOptions) (runtime.Object, error) { |
| 145 | return func(opts meta.ListOptions) (runtime.Object, error) { |
| 146 | return c.CoreV1().Namespaces().List(ctx, opts) |
| 147 | } |
| 148 | } |
| 149 | |
| 150 | func serviceWatchFunc(ctx context.Context, c kubernetes.Interface, ns string) func(options meta.ListOptions) (watch.Interface, error) { |
| 151 | return func(options meta.ListOptions) (watch.Interface, error) { |
| 152 | return c.CoreV1().Services(ns).Watch(ctx, options) |
| 153 | } |
| 154 | } |
| 155 | |
| 156 | func endpointSliceWatchFunc(ctx context.Context, c kubernetes.Interface, ns string) func(options meta.ListOptions) (watch.Interface, error) { |
| 157 | return func(options meta.ListOptions) (watch.Interface, error) { |
| 158 | options.LabelSelector = epLabelSelector |
| 159 | return c.DiscoveryV1().EndpointSlices(ns).Watch(ctx, options) |
| 160 | } |
| 161 | } |
| 162 | |
| 163 | func namespaceWatchFunc(ctx context.Context, c kubernetes.Interface) func(options meta.ListOptions) (watch.Interface, error) { |
| 164 | return func(options meta.ListOptions) (watch.Interface, error) { |
| 165 | return c.CoreV1().Namespaces().Watch(ctx, options) |
| 166 | } |
| 167 | } |
| 168 | |
| 169 | // Start starts the controller. |
| 170 | func (dns *dnsControl) Start(stopCh <-chan struct{}) { |
| 171 | go dns.svcController.Run(stopCh) |
| 172 | go dns.epController.Run(stopCh) |
| 173 | go dns.nsController.Run(stopCh) |
| 174 | } |
| 175 | |
| 176 | // HasSynced returns true if the initial data has been |
| 177 | // completely loaded into memory. |
| 178 | func (dns *dnsControl) HasSynced() bool { |
| 179 | if dns.hasSynced.Load() { |
| 180 | return true |
| 181 | } |
| 182 | a := dns.svcController.HasSynced() |
| 183 | b := dns.epController.HasSynced() |
| 184 | c := dns.nsController.HasSynced() |
| 185 | if a && b && c { |
| 186 | dns.hasSynced.Store(true) |
| 187 | return true |
| 188 | } |
| 189 | return false |
| 190 | } |
| 191 | |
| 192 | func (dns *dnsControl) GetSvc(key string) *object.Service { |
| 193 | o, exists, err := dns.svcLister.GetByKey(key) |
| 194 | if err != nil || !exists { |
| 195 | return nil |
| 196 | } |
| 197 | s, ok := o.(*object.Service) |
| 198 | if ok { |
| 199 | return s |
| 200 | } |
| 201 | return nil |
| 202 | } |
| 203 | |
| 204 | func (dns *dnsControl) SvcIndexReverse(ip string) (svcs []*object.Service) { |
| 205 | os, err := dns.svcLister.ByIndex(svcIPIndex, ip) |
| 206 | if err != nil { |
| 207 | return nil |
| 208 | } |
| 209 | svcs = make([]*object.Service, 0, len(os)) |
| 210 | for _, o := range os { |
| 211 | s, ok := o.(*object.Service) |
| 212 | if ok { |
| 213 | svcs = append(svcs, s) |
| 214 | } |
| 215 | } |
| 216 | return svcs |
| 217 | } |
| 218 | |
| 219 | func (dns *dnsControl) EpIndex(idx string) (ep []*object.Endpoints) { |
| 220 | os, err := dns.epLister.ByIndex(epNameNamespaceIndex, idx) |
| 221 | if err != nil { |
| 222 | return nil |
| 223 | } |
| 224 | ep = make([]*object.Endpoints, 0, len(os)) |
| 225 | for _, o := range os { |
| 226 | e, ok := o.(*object.Endpoints) |
| 227 | if ok { |
| 228 | ep = append(ep, e) |
| 229 | } |
| 230 | } |
| 231 | return ep |
| 232 | } |
| 233 | |
| 234 | func (dns *dnsControl) EpIndexReverse(ip string) (ep []*object.Endpoints) { |
| 235 | os, err := dns.epLister.ByIndex(epIPIndex, ip) |
| 236 | if err != nil { |
| 237 | return nil |
| 238 | } |
| 239 | ep = make([]*object.Endpoints, 0, len(os)) |
| 240 | for _, o := range os { |
| 241 | e, ok := o.(*object.Endpoints) |
| 242 | if ok { |
| 243 | ep = append(ep, e) |
| 244 | } |
| 245 | } |
| 246 | return ep |
| 247 | } |
| 248 | |
| 249 | // NamespaceExists returns true if a namespace with this name exists. |
| 250 | func (dns *dnsControl) NamespaceExists(name string) bool { |
| 251 | _, exists, _ := dns.nsLister.GetByKey(name) |
| 252 | return exists |
| 253 | } |
| 254 | |
| 255 | func (dns *dnsControl) OnAdd(obj interface{}, isInInitialList bool) { |
| 256 | dns.updateModified() |
| 257 | switch obj := obj.(type) { |
| 258 | case *object.Endpoints: |
| 259 | // Don't record latency during initial sync, because measuring latency only |
| 260 | // makes sense for changes that happen while the service is running. |
| 261 | if !isInInitialList { |
| 262 | recordDNSProgrammingLatency(obj.LastChangeTriggerTime) |
| 263 | } |
| 264 | } |
| 265 | } |
| 266 | |
| 267 | func (dns *dnsControl) OnDelete(obj interface{}) { |
| 268 | dns.updateModified() |
| 269 | // Note: We cannot record programming latency on deletes, because the trigger |
| 270 | // time annotation is not updated when the object is deleted. |
| 271 | } |
| 272 | |
| 273 | func (dns *dnsControl) OnUpdate(oldObj, newObj interface{}) { |
| 274 | // If both objects have the same resource version, they are identical. |
| 275 | if oldObj.(meta.Object).GetResourceVersion() == newObj.(meta.Object).GetResourceVersion() { |
| 276 | return |
| 277 | } |
| 278 | |
| 279 | switch newObj := newObj.(type) { |
| 280 | case *object.Service: |
| 281 | if object.ServiceModified(oldObj.(*object.Service), newObj) { |
| 282 | dns.updateModified() |
| 283 | } |
| 284 | case *object.Endpoints: |
| 285 | oldObj := oldObj.(*object.Endpoints) |
| 286 | if object.EndpointsModified(oldObj, newObj) { |
| 287 | dns.updateModified() |
| 288 | // If the trigger time has not changed, the process that last updated the |
| 289 | // object did not update the trigger time, so we can't know the latency. |
| 290 | if oldObj.LastChangeTriggerTime != newObj.LastChangeTriggerTime { |
| 291 | recordDNSProgrammingLatency(newObj.LastChangeTriggerTime) |
| 292 | } |
| 293 | } |
| 294 | } |
| 295 | } |
| 296 | |
| 297 | func (dns *dnsControl) Modified() int64 { |
| 298 | return dns.modified.Load() |
| 299 | } |
| 300 | |
| 301 | // updateModified set dns.modified to the current time. |
| 302 | func (dns *dnsControl) updateModified() { |
| 303 | unix := time.Now().Unix() |
| 304 | dns.modified.Store(unix) |
| 305 | } |
| 306 | |
| 307 | var errObj = errors.New("obj was not of the correct type") |