Matthias Andreas Benkard | 832a54e | 2019-01-29 09:27:38 +0100 | [diff] [blame] | 1 | /* |
| 2 | Copyright 2015 The Kubernetes Authors. |
| 3 | |
| 4 | Licensed under the Apache License, Version 2.0 (the "License"); |
| 5 | you may not use this file except in compliance with the License. |
| 6 | You may obtain a copy of the License at |
| 7 | |
| 8 | http://www.apache.org/licenses/LICENSE-2.0 |
| 9 | |
| 10 | Unless required by applicable law or agreed to in writing, software |
| 11 | distributed under the License is distributed on an "AS IS" BASIS, |
| 12 | WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 13 | See the License for the specific language governing permissions and |
| 14 | limitations under the License. |
| 15 | */ |
| 16 | |
| 17 | package storage |
| 18 | |
| 19 | import ( |
| 20 | "context" |
| 21 | "fmt" |
| 22 | "net/http" |
| 23 | "reflect" |
| 24 | "sync" |
| 25 | "time" |
| 26 | |
| 27 | "github.com/golang/glog" |
| 28 | |
| 29 | "k8s.io/apimachinery/pkg/api/errors" |
| 30 | "k8s.io/apimachinery/pkg/api/meta" |
| 31 | metav1 "k8s.io/apimachinery/pkg/apis/meta/v1" |
| 32 | "k8s.io/apimachinery/pkg/conversion" |
| 33 | "k8s.io/apimachinery/pkg/fields" |
| 34 | "k8s.io/apimachinery/pkg/labels" |
| 35 | "k8s.io/apimachinery/pkg/runtime" |
| 36 | utilruntime "k8s.io/apimachinery/pkg/util/runtime" |
| 37 | "k8s.io/apimachinery/pkg/util/wait" |
| 38 | "k8s.io/apimachinery/pkg/watch" |
| 39 | "k8s.io/apiserver/pkg/features" |
| 40 | utilfeature "k8s.io/apiserver/pkg/util/feature" |
| 41 | utiltrace "k8s.io/apiserver/pkg/util/trace" |
| 42 | "k8s.io/client-go/tools/cache" |
| 43 | ) |
| 44 | |
| 45 | // CacherConfig contains the configuration for a given Cache. |
| 46 | type CacherConfig struct { |
| 47 | // Maximum size of the history cached in memory. |
| 48 | CacheCapacity int |
| 49 | |
| 50 | // An underlying storage.Interface. |
| 51 | Storage Interface |
| 52 | |
| 53 | // An underlying storage.Versioner. |
| 54 | Versioner Versioner |
| 55 | |
| 56 | // The Cache will be caching objects of a given Type and assumes that they |
| 57 | // are all stored under ResourcePrefix directory in the underlying database. |
| 58 | Type interface{} |
| 59 | ResourcePrefix string |
| 60 | |
| 61 | // KeyFunc is used to get a key in the underlying storage for a given object. |
| 62 | KeyFunc func(runtime.Object) (string, error) |
| 63 | |
| 64 | // GetAttrsFunc is used to get object labels, fields, and the uninitialized bool |
| 65 | GetAttrsFunc func(runtime.Object) (label labels.Set, field fields.Set, uninitialized bool, err error) |
| 66 | |
| 67 | // TriggerPublisherFunc is used for optimizing amount of watchers that |
| 68 | // needs to process an incoming event. |
| 69 | TriggerPublisherFunc TriggerPublisherFunc |
| 70 | |
| 71 | // NewList is a function that creates new empty object storing a list of |
| 72 | // objects of type Type. |
| 73 | NewListFunc func() runtime.Object |
| 74 | |
| 75 | Codec runtime.Codec |
| 76 | } |
| 77 | |
| 78 | type watchersMap map[int]*cacheWatcher |
| 79 | |
| 80 | func (wm watchersMap) addWatcher(w *cacheWatcher, number int) { |
| 81 | wm[number] = w |
| 82 | } |
| 83 | |
| 84 | func (wm watchersMap) deleteWatcher(number int) { |
| 85 | delete(wm, number) |
| 86 | } |
| 87 | |
| 88 | func (wm watchersMap) terminateAll() { |
| 89 | for key, watcher := range wm { |
| 90 | delete(wm, key) |
| 91 | watcher.stop() |
| 92 | } |
| 93 | } |
| 94 | |
| 95 | type indexedWatchers struct { |
| 96 | allWatchers watchersMap |
| 97 | valueWatchers map[string]watchersMap |
| 98 | } |
| 99 | |
| 100 | func (i *indexedWatchers) addWatcher(w *cacheWatcher, number int, value string, supported bool) { |
| 101 | if supported { |
| 102 | if _, ok := i.valueWatchers[value]; !ok { |
| 103 | i.valueWatchers[value] = watchersMap{} |
| 104 | } |
| 105 | i.valueWatchers[value].addWatcher(w, number) |
| 106 | } else { |
| 107 | i.allWatchers.addWatcher(w, number) |
| 108 | } |
| 109 | } |
| 110 | |
| 111 | func (i *indexedWatchers) deleteWatcher(number int, value string, supported bool) { |
| 112 | if supported { |
| 113 | i.valueWatchers[value].deleteWatcher(number) |
| 114 | if len(i.valueWatchers[value]) == 0 { |
| 115 | delete(i.valueWatchers, value) |
| 116 | } |
| 117 | } else { |
| 118 | i.allWatchers.deleteWatcher(number) |
| 119 | } |
| 120 | } |
| 121 | |
| 122 | func (i *indexedWatchers) terminateAll(objectType reflect.Type) { |
| 123 | if len(i.allWatchers) > 0 || len(i.valueWatchers) > 0 { |
| 124 | glog.Warningf("Terminating all watchers from cacher %v", objectType) |
| 125 | } |
| 126 | i.allWatchers.terminateAll() |
| 127 | for index, watchers := range i.valueWatchers { |
| 128 | watchers.terminateAll() |
| 129 | delete(i.valueWatchers, index) |
| 130 | } |
| 131 | } |
| 132 | |
| 133 | type filterWithAttrsFunc func(key string, l labels.Set, f fields.Set, uninitialized bool) bool |
| 134 | |
| 135 | // Cacher is responsible for serving WATCH and LIST requests for a given |
| 136 | // resource from its internal cache and updating its cache in the background |
| 137 | // based on the underlying storage contents. |
| 138 | // Cacher implements storage.Interface (although most of the calls are just |
| 139 | // delegated to the underlying storage). |
| 140 | type Cacher struct { |
| 141 | // HighWaterMarks for performance debugging. |
| 142 | // Important: Since HighWaterMark is using sync/atomic, it has to be at the top of the struct due to a bug on 32-bit platforms |
| 143 | // See: https://golang.org/pkg/sync/atomic/ for more information |
| 144 | incomingHWM HighWaterMark |
| 145 | // Incoming events that should be dispatched to watchers. |
| 146 | incoming chan watchCacheEvent |
| 147 | |
| 148 | sync.RWMutex |
| 149 | |
| 150 | // Before accessing the cacher's cache, wait for the ready to be ok. |
| 151 | // This is necessary to prevent users from accessing structures that are |
| 152 | // uninitialized or are being repopulated right now. |
| 153 | // ready needs to be set to false when the cacher is paused or stopped. |
| 154 | // ready needs to be set to true when the cacher is ready to use after |
| 155 | // initialization. |
| 156 | ready *ready |
| 157 | |
| 158 | // Underlying storage.Interface. |
| 159 | storage Interface |
| 160 | |
| 161 | // Expected type of objects in the underlying cache. |
| 162 | objectType reflect.Type |
| 163 | |
| 164 | // "sliding window" of recent changes of objects and the current state. |
| 165 | watchCache *watchCache |
| 166 | reflector *cache.Reflector |
| 167 | |
| 168 | // Versioner is used to handle resource versions. |
| 169 | versioner Versioner |
| 170 | |
| 171 | // triggerFunc is used for optimizing amount of watchers that needs to process |
| 172 | // an incoming event. |
| 173 | triggerFunc TriggerPublisherFunc |
| 174 | // watchers is mapping from the value of trigger function that a |
| 175 | // watcher is interested into the watchers |
| 176 | watcherIdx int |
| 177 | watchers indexedWatchers |
| 178 | |
| 179 | // Defines a time budget that can be spend on waiting for not-ready watchers |
| 180 | // while dispatching event before shutting them down. |
| 181 | dispatchTimeoutBudget *timeBudget |
| 182 | |
| 183 | // Handling graceful termination. |
| 184 | stopLock sync.RWMutex |
| 185 | stopped bool |
| 186 | stopCh chan struct{} |
| 187 | stopWg sync.WaitGroup |
| 188 | } |
| 189 | |
| 190 | // Create a new Cacher responsible for servicing WATCH and LIST requests from |
| 191 | // its internal cache and updating its cache in the background based on the |
| 192 | // given configuration. |
| 193 | func NewCacherFromConfig(config CacherConfig) *Cacher { |
| 194 | watchCache := newWatchCache(config.CacheCapacity, config.KeyFunc, config.GetAttrsFunc) |
| 195 | listerWatcher := newCacherListerWatcher(config.Storage, config.ResourcePrefix, config.NewListFunc) |
| 196 | reflectorName := "storage/cacher.go:" + config.ResourcePrefix |
| 197 | |
| 198 | // Give this error when it is constructed rather than when you get the |
| 199 | // first watch item, because it's much easier to track down that way. |
| 200 | if obj, ok := config.Type.(runtime.Object); ok { |
| 201 | if err := runtime.CheckCodec(config.Codec, obj); err != nil { |
| 202 | panic("storage codec doesn't seem to match given type: " + err.Error()) |
| 203 | } |
| 204 | } |
| 205 | |
| 206 | stopCh := make(chan struct{}) |
| 207 | cacher := &Cacher{ |
| 208 | ready: newReady(), |
| 209 | storage: config.Storage, |
| 210 | objectType: reflect.TypeOf(config.Type), |
| 211 | watchCache: watchCache, |
| 212 | reflector: cache.NewNamedReflector(reflectorName, listerWatcher, config.Type, watchCache, 0), |
| 213 | versioner: config.Versioner, |
| 214 | triggerFunc: config.TriggerPublisherFunc, |
| 215 | watcherIdx: 0, |
| 216 | watchers: indexedWatchers{ |
| 217 | allWatchers: make(map[int]*cacheWatcher), |
| 218 | valueWatchers: make(map[string]watchersMap), |
| 219 | }, |
| 220 | // TODO: Figure out the correct value for the buffer size. |
| 221 | incoming: make(chan watchCacheEvent, 100), |
| 222 | dispatchTimeoutBudget: newTimeBudget(stopCh), |
| 223 | // We need to (potentially) stop both: |
| 224 | // - wait.Until go-routine |
| 225 | // - reflector.ListAndWatch |
| 226 | // and there are no guarantees on the order that they will stop. |
| 227 | // So we will be simply closing the channel, and synchronizing on the WaitGroup. |
| 228 | stopCh: stopCh, |
| 229 | } |
| 230 | watchCache.SetOnEvent(cacher.processEvent) |
| 231 | go cacher.dispatchEvents() |
| 232 | |
| 233 | cacher.stopWg.Add(1) |
| 234 | go func() { |
| 235 | defer cacher.stopWg.Done() |
| 236 | wait.Until( |
| 237 | func() { |
| 238 | if !cacher.isStopped() { |
| 239 | cacher.startCaching(stopCh) |
| 240 | } |
| 241 | }, time.Second, stopCh, |
| 242 | ) |
| 243 | }() |
| 244 | return cacher |
| 245 | } |
| 246 | |
| 247 | func (c *Cacher) startCaching(stopChannel <-chan struct{}) { |
| 248 | // The 'usable' lock is always 'RLock'able when it is safe to use the cache. |
| 249 | // It is safe to use the cache after a successful list until a disconnection. |
| 250 | // We start with usable (write) locked. The below OnReplace function will |
| 251 | // unlock it after a successful list. The below defer will then re-lock |
| 252 | // it when this function exits (always due to disconnection), only if |
| 253 | // we actually got a successful list. This cycle will repeat as needed. |
| 254 | successfulList := false |
| 255 | c.watchCache.SetOnReplace(func() { |
| 256 | successfulList = true |
| 257 | c.ready.set(true) |
| 258 | }) |
| 259 | defer func() { |
| 260 | if successfulList { |
| 261 | c.ready.set(false) |
| 262 | } |
| 263 | }() |
| 264 | |
| 265 | c.terminateAllWatchers() |
| 266 | // Note that since onReplace may be not called due to errors, we explicitly |
| 267 | // need to retry it on errors under lock. |
| 268 | // Also note that startCaching is called in a loop, so there's no need |
| 269 | // to have another loop here. |
| 270 | if err := c.reflector.ListAndWatch(stopChannel); err != nil { |
| 271 | glog.Errorf("unexpected ListAndWatch error: %v", err) |
| 272 | } |
| 273 | } |
| 274 | |
| 275 | // Implements storage.Interface. |
| 276 | func (c *Cacher) Versioner() Versioner { |
| 277 | return c.storage.Versioner() |
| 278 | } |
| 279 | |
| 280 | // Implements storage.Interface. |
| 281 | func (c *Cacher) Create(ctx context.Context, key string, obj, out runtime.Object, ttl uint64) error { |
| 282 | return c.storage.Create(ctx, key, obj, out, ttl) |
| 283 | } |
| 284 | |
| 285 | // Implements storage.Interface. |
| 286 | func (c *Cacher) Delete(ctx context.Context, key string, out runtime.Object, preconditions *Preconditions) error { |
| 287 | return c.storage.Delete(ctx, key, out, preconditions) |
| 288 | } |
| 289 | |
| 290 | // Implements storage.Interface. |
| 291 | func (c *Cacher) Watch(ctx context.Context, key string, resourceVersion string, pred SelectionPredicate) (watch.Interface, error) { |
| 292 | watchRV, err := c.versioner.ParseWatchResourceVersion(resourceVersion) |
| 293 | if err != nil { |
| 294 | return nil, err |
| 295 | } |
| 296 | |
| 297 | c.ready.wait() |
| 298 | |
| 299 | // We explicitly use thread unsafe version and do locking ourself to ensure that |
| 300 | // no new events will be processed in the meantime. The watchCache will be unlocked |
| 301 | // on return from this function. |
| 302 | // Note that we cannot do it under Cacher lock, to avoid a deadlock, since the |
| 303 | // underlying watchCache is calling processEvent under its lock. |
| 304 | c.watchCache.RLock() |
| 305 | defer c.watchCache.RUnlock() |
| 306 | initEvents, err := c.watchCache.GetAllEventsSinceThreadUnsafe(watchRV) |
| 307 | if err != nil { |
| 308 | // To match the uncached watch implementation, once we have passed authn/authz/admission, |
| 309 | // and successfully parsed a resource version, other errors must fail with a watch event of type ERROR, |
| 310 | // rather than a directly returned error. |
| 311 | return newErrWatcher(err), nil |
| 312 | } |
| 313 | |
| 314 | triggerValue, triggerSupported := "", false |
| 315 | // TODO: Currently we assume that in a given Cacher object, any <predicate> that is |
| 316 | // passed here is aware of exactly the same trigger (at most one). |
| 317 | // Thus, either 0 or 1 values will be returned. |
| 318 | if matchValues := pred.MatcherIndex(); len(matchValues) > 0 { |
| 319 | triggerValue, triggerSupported = matchValues[0].Value, true |
| 320 | } |
| 321 | |
| 322 | // If there is triggerFunc defined, but triggerSupported is false, |
| 323 | // we can't narrow the amount of events significantly at this point. |
| 324 | // |
| 325 | // That said, currently triggerFunc is defined only for Pods and Nodes, |
| 326 | // and there is only constant number of watchers for which triggerSupported |
| 327 | // is false (excluding those issues explicitly by users). |
| 328 | // Thus, to reduce the risk of those watchers blocking all watchers of a |
| 329 | // given resource in the system, we increase the sizes of buffers for them. |
| 330 | chanSize := 10 |
| 331 | if c.triggerFunc != nil && !triggerSupported { |
| 332 | // TODO: We should tune this value and ideally make it dependent on the |
| 333 | // number of objects of a given type and/or their churn. |
| 334 | chanSize = 1000 |
| 335 | } |
| 336 | |
| 337 | c.Lock() |
| 338 | defer c.Unlock() |
| 339 | forget := forgetWatcher(c, c.watcherIdx, triggerValue, triggerSupported) |
| 340 | watcher := newCacheWatcher(watchRV, chanSize, initEvents, filterWithAttrsFunction(key, pred), forget, c.versioner) |
| 341 | |
| 342 | c.watchers.addWatcher(watcher, c.watcherIdx, triggerValue, triggerSupported) |
| 343 | c.watcherIdx++ |
| 344 | return watcher, nil |
| 345 | } |
| 346 | |
| 347 | // Implements storage.Interface. |
| 348 | func (c *Cacher) WatchList(ctx context.Context, key string, resourceVersion string, pred SelectionPredicate) (watch.Interface, error) { |
| 349 | return c.Watch(ctx, key, resourceVersion, pred) |
| 350 | } |
| 351 | |
| 352 | // Implements storage.Interface. |
| 353 | func (c *Cacher) Get(ctx context.Context, key string, resourceVersion string, objPtr runtime.Object, ignoreNotFound bool) error { |
| 354 | if resourceVersion == "" { |
| 355 | // If resourceVersion is not specified, serve it from underlying |
| 356 | // storage (for backward compatibility). |
| 357 | return c.storage.Get(ctx, key, resourceVersion, objPtr, ignoreNotFound) |
| 358 | } |
| 359 | |
| 360 | // If resourceVersion is specified, serve it from cache. |
| 361 | // It's guaranteed that the returned value is at least that |
| 362 | // fresh as the given resourceVersion. |
| 363 | getRV, err := c.versioner.ParseListResourceVersion(resourceVersion) |
| 364 | if err != nil { |
| 365 | return err |
| 366 | } |
| 367 | |
| 368 | if getRV == 0 && !c.ready.check() { |
| 369 | // If Cacher is not yet initialized and we don't require any specific |
| 370 | // minimal resource version, simply forward the request to storage. |
| 371 | return c.storage.Get(ctx, key, resourceVersion, objPtr, ignoreNotFound) |
| 372 | } |
| 373 | |
| 374 | // Do not create a trace - it's not for free and there are tons |
| 375 | // of Get requests. We can add it if it will be really needed. |
| 376 | c.ready.wait() |
| 377 | |
| 378 | objVal, err := conversion.EnforcePtr(objPtr) |
| 379 | if err != nil { |
| 380 | return err |
| 381 | } |
| 382 | |
| 383 | obj, exists, readResourceVersion, err := c.watchCache.WaitUntilFreshAndGet(getRV, key, nil) |
| 384 | if err != nil { |
| 385 | return err |
| 386 | } |
| 387 | |
| 388 | if exists { |
| 389 | elem, ok := obj.(*storeElement) |
| 390 | if !ok { |
| 391 | return fmt.Errorf("non *storeElement returned from storage: %v", obj) |
| 392 | } |
| 393 | objVal.Set(reflect.ValueOf(elem.Object).Elem()) |
| 394 | } else { |
| 395 | objVal.Set(reflect.Zero(objVal.Type())) |
| 396 | if !ignoreNotFound { |
| 397 | return NewKeyNotFoundError(key, int64(readResourceVersion)) |
| 398 | } |
| 399 | } |
| 400 | return nil |
| 401 | } |
| 402 | |
| 403 | // Implements storage.Interface. |
| 404 | func (c *Cacher) GetToList(ctx context.Context, key string, resourceVersion string, pred SelectionPredicate, listObj runtime.Object) error { |
| 405 | pagingEnabled := utilfeature.DefaultFeatureGate.Enabled(features.APIListChunking) |
| 406 | if resourceVersion == "" || (pagingEnabled && (len(pred.Continue) > 0 || pred.Limit > 0)) { |
| 407 | // If resourceVersion is not specified, serve it from underlying |
| 408 | // storage (for backward compatibility). If a continuation or limit is |
| 409 | // requested, serve it from the underlying storage as well. |
| 410 | return c.storage.GetToList(ctx, key, resourceVersion, pred, listObj) |
| 411 | } |
| 412 | |
| 413 | // If resourceVersion is specified, serve it from cache. |
| 414 | // It's guaranteed that the returned value is at least that |
| 415 | // fresh as the given resourceVersion. |
| 416 | listRV, err := c.versioner.ParseListResourceVersion(resourceVersion) |
| 417 | if err != nil { |
| 418 | return err |
| 419 | } |
| 420 | |
| 421 | if listRV == 0 && !c.ready.check() { |
| 422 | // If Cacher is not yet initialized and we don't require any specific |
| 423 | // minimal resource version, simply forward the request to storage. |
| 424 | return c.storage.GetToList(ctx, key, resourceVersion, pred, listObj) |
| 425 | } |
| 426 | |
| 427 | trace := utiltrace.New(fmt.Sprintf("cacher %v: List", c.objectType.String())) |
| 428 | defer trace.LogIfLong(500 * time.Millisecond) |
| 429 | |
| 430 | c.ready.wait() |
| 431 | trace.Step("Ready") |
| 432 | |
| 433 | // List elements with at least 'listRV' from cache. |
| 434 | listPtr, err := meta.GetItemsPtr(listObj) |
| 435 | if err != nil { |
| 436 | return err |
| 437 | } |
| 438 | listVal, err := conversion.EnforcePtr(listPtr) |
| 439 | if err != nil || listVal.Kind() != reflect.Slice { |
| 440 | return fmt.Errorf("need a pointer to slice, got %v", listVal.Kind()) |
| 441 | } |
| 442 | filter := filterWithAttrsFunction(key, pred) |
| 443 | |
| 444 | obj, exists, readResourceVersion, err := c.watchCache.WaitUntilFreshAndGet(listRV, key, trace) |
| 445 | if err != nil { |
| 446 | return err |
| 447 | } |
| 448 | trace.Step("Got from cache") |
| 449 | |
| 450 | if exists { |
| 451 | elem, ok := obj.(*storeElement) |
| 452 | if !ok { |
| 453 | return fmt.Errorf("non *storeElement returned from storage: %v", obj) |
| 454 | } |
| 455 | if filter(elem.Key, elem.Labels, elem.Fields, elem.Uninitialized) { |
| 456 | listVal.Set(reflect.Append(listVal, reflect.ValueOf(elem.Object).Elem())) |
| 457 | } |
| 458 | } |
| 459 | if c.versioner != nil { |
| 460 | if err := c.versioner.UpdateList(listObj, readResourceVersion, ""); err != nil { |
| 461 | return err |
| 462 | } |
| 463 | } |
| 464 | return nil |
| 465 | } |
| 466 | |
| 467 | // Implements storage.Interface. |
| 468 | func (c *Cacher) List(ctx context.Context, key string, resourceVersion string, pred SelectionPredicate, listObj runtime.Object) error { |
| 469 | pagingEnabled := utilfeature.DefaultFeatureGate.Enabled(features.APIListChunking) |
| 470 | hasContinuation := pagingEnabled && len(pred.Continue) > 0 |
| 471 | hasLimit := pagingEnabled && pred.Limit > 0 && resourceVersion != "0" |
| 472 | if resourceVersion == "" || hasContinuation || hasLimit { |
| 473 | // If resourceVersion is not specified, serve it from underlying |
| 474 | // storage (for backward compatibility). If a continuation is |
| 475 | // requested, serve it from the underlying storage as well. |
| 476 | // Limits are only sent to storage when resourceVersion is non-zero |
| 477 | // since the watch cache isn't able to perform continuations, and |
| 478 | // limits are ignored when resource version is zero. |
| 479 | return c.storage.List(ctx, key, resourceVersion, pred, listObj) |
| 480 | } |
| 481 | |
| 482 | // If resourceVersion is specified, serve it from cache. |
| 483 | // It's guaranteed that the returned value is at least that |
| 484 | // fresh as the given resourceVersion. |
| 485 | listRV, err := c.versioner.ParseListResourceVersion(resourceVersion) |
| 486 | if err != nil { |
| 487 | return err |
| 488 | } |
| 489 | |
| 490 | if listRV == 0 && !c.ready.check() { |
| 491 | // If Cacher is not yet initialized and we don't require any specific |
| 492 | // minimal resource version, simply forward the request to storage. |
| 493 | return c.storage.List(ctx, key, resourceVersion, pred, listObj) |
| 494 | } |
| 495 | |
| 496 | trace := utiltrace.New(fmt.Sprintf("cacher %v: List", c.objectType.String())) |
| 497 | defer trace.LogIfLong(500 * time.Millisecond) |
| 498 | |
| 499 | c.ready.wait() |
| 500 | trace.Step("Ready") |
| 501 | |
| 502 | // List elements with at least 'listRV' from cache. |
| 503 | listPtr, err := meta.GetItemsPtr(listObj) |
| 504 | if err != nil { |
| 505 | return err |
| 506 | } |
| 507 | listVal, err := conversion.EnforcePtr(listPtr) |
| 508 | if err != nil || listVal.Kind() != reflect.Slice { |
| 509 | return fmt.Errorf("need a pointer to slice, got %v", listVal.Kind()) |
| 510 | } |
| 511 | filter := filterWithAttrsFunction(key, pred) |
| 512 | |
| 513 | objs, readResourceVersion, err := c.watchCache.WaitUntilFreshAndList(listRV, trace) |
| 514 | if err != nil { |
| 515 | return err |
| 516 | } |
| 517 | trace.Step(fmt.Sprintf("Listed %d items from cache", len(objs))) |
| 518 | if len(objs) > listVal.Cap() && pred.Label.Empty() && pred.Field.Empty() { |
| 519 | // Resize the slice appropriately, since we already know that none |
| 520 | // of the elements will be filtered out. |
| 521 | listVal.Set(reflect.MakeSlice(reflect.SliceOf(c.objectType.Elem()), 0, len(objs))) |
| 522 | trace.Step("Resized result") |
| 523 | } |
| 524 | for _, obj := range objs { |
| 525 | elem, ok := obj.(*storeElement) |
| 526 | if !ok { |
| 527 | return fmt.Errorf("non *storeElement returned from storage: %v", obj) |
| 528 | } |
| 529 | if filter(elem.Key, elem.Labels, elem.Fields, elem.Uninitialized) { |
| 530 | listVal.Set(reflect.Append(listVal, reflect.ValueOf(elem.Object).Elem())) |
| 531 | } |
| 532 | } |
| 533 | trace.Step(fmt.Sprintf("Filtered %d items", listVal.Len())) |
| 534 | if c.versioner != nil { |
| 535 | if err := c.versioner.UpdateList(listObj, readResourceVersion, ""); err != nil { |
| 536 | return err |
| 537 | } |
| 538 | } |
| 539 | return nil |
| 540 | } |
| 541 | |
| 542 | // Implements storage.Interface. |
| 543 | func (c *Cacher) GuaranteedUpdate( |
| 544 | ctx context.Context, key string, ptrToType runtime.Object, ignoreNotFound bool, |
| 545 | preconditions *Preconditions, tryUpdate UpdateFunc, _ ...runtime.Object) error { |
| 546 | // Ignore the suggestion and try to pass down the current version of the object |
| 547 | // read from cache. |
| 548 | if elem, exists, err := c.watchCache.GetByKey(key); err != nil { |
| 549 | glog.Errorf("GetByKey returned error: %v", err) |
| 550 | } else if exists { |
| 551 | currObj := elem.(*storeElement).Object.DeepCopyObject() |
| 552 | return c.storage.GuaranteedUpdate(ctx, key, ptrToType, ignoreNotFound, preconditions, tryUpdate, currObj) |
| 553 | } |
| 554 | // If we couldn't get the object, fallback to no-suggestion. |
| 555 | return c.storage.GuaranteedUpdate(ctx, key, ptrToType, ignoreNotFound, preconditions, tryUpdate) |
| 556 | } |
| 557 | |
| 558 | func (c *Cacher) Count(pathPrefix string) (int64, error) { |
| 559 | return c.storage.Count(pathPrefix) |
| 560 | } |
| 561 | |
| 562 | func (c *Cacher) triggerValues(event *watchCacheEvent) ([]string, bool) { |
| 563 | // TODO: Currently we assume that in a given Cacher object, its <c.triggerFunc> |
| 564 | // is aware of exactly the same trigger (at most one). Thus calling: |
| 565 | // c.triggerFunc(<some object>) |
| 566 | // can return only 0 or 1 values. |
| 567 | // That means, that triggerValues itself may return up to 2 different values. |
| 568 | if c.triggerFunc == nil { |
| 569 | return nil, false |
| 570 | } |
| 571 | result := make([]string, 0, 2) |
| 572 | matchValues := c.triggerFunc(event.Object) |
| 573 | if len(matchValues) > 0 { |
| 574 | result = append(result, matchValues[0].Value) |
| 575 | } |
| 576 | if event.PrevObject == nil { |
| 577 | return result, len(result) > 0 |
| 578 | } |
| 579 | prevMatchValues := c.triggerFunc(event.PrevObject) |
| 580 | if len(prevMatchValues) > 0 { |
| 581 | if len(result) == 0 || result[0] != prevMatchValues[0].Value { |
| 582 | result = append(result, prevMatchValues[0].Value) |
| 583 | } |
| 584 | } |
| 585 | return result, len(result) > 0 |
| 586 | } |
| 587 | |
| 588 | func (c *Cacher) processEvent(event *watchCacheEvent) { |
| 589 | if curLen := int64(len(c.incoming)); c.incomingHWM.Update(curLen) { |
| 590 | // Monitor if this gets backed up, and how much. |
| 591 | glog.V(1).Infof("cacher (%v): %v objects queued in incoming channel.", c.objectType.String(), curLen) |
| 592 | } |
| 593 | c.incoming <- *event |
| 594 | } |
| 595 | |
| 596 | func (c *Cacher) dispatchEvents() { |
| 597 | for { |
| 598 | select { |
| 599 | case event, ok := <-c.incoming: |
| 600 | if !ok { |
| 601 | return |
| 602 | } |
| 603 | c.dispatchEvent(&event) |
| 604 | case <-c.stopCh: |
| 605 | return |
| 606 | } |
| 607 | } |
| 608 | } |
| 609 | |
| 610 | func (c *Cacher) dispatchEvent(event *watchCacheEvent) { |
| 611 | triggerValues, supported := c.triggerValues(event) |
| 612 | |
| 613 | c.Lock() |
| 614 | defer c.Unlock() |
| 615 | // Iterate over "allWatchers" no matter what the trigger function is. |
| 616 | for _, watcher := range c.watchers.allWatchers { |
| 617 | watcher.add(event, c.dispatchTimeoutBudget) |
| 618 | } |
| 619 | if supported { |
| 620 | // Iterate over watchers interested in the given values of the trigger. |
| 621 | for _, triggerValue := range triggerValues { |
| 622 | for _, watcher := range c.watchers.valueWatchers[triggerValue] { |
| 623 | watcher.add(event, c.dispatchTimeoutBudget) |
| 624 | } |
| 625 | } |
| 626 | } else { |
| 627 | // supported equal to false generally means that trigger function |
| 628 | // is not defined (or not aware of any indexes). In this case, |
| 629 | // watchers filters should generally also don't generate any |
| 630 | // trigger values, but can cause problems in case of some |
| 631 | // misconfiguration. Thus we paranoidly leave this branch. |
| 632 | |
| 633 | // Iterate over watchers interested in exact values for all values. |
| 634 | for _, watchers := range c.watchers.valueWatchers { |
| 635 | for _, watcher := range watchers { |
| 636 | watcher.add(event, c.dispatchTimeoutBudget) |
| 637 | } |
| 638 | } |
| 639 | } |
| 640 | } |
| 641 | |
| 642 | func (c *Cacher) terminateAllWatchers() { |
| 643 | c.Lock() |
| 644 | defer c.Unlock() |
| 645 | c.watchers.terminateAll(c.objectType) |
| 646 | } |
| 647 | |
| 648 | func (c *Cacher) isStopped() bool { |
| 649 | c.stopLock.RLock() |
| 650 | defer c.stopLock.RUnlock() |
| 651 | return c.stopped |
| 652 | } |
| 653 | |
| 654 | func (c *Cacher) Stop() { |
| 655 | // avoid stopping twice (note: cachers are shared with subresources) |
| 656 | if c.isStopped() { |
| 657 | return |
| 658 | } |
| 659 | c.stopLock.Lock() |
| 660 | if c.stopped { |
| 661 | c.stopLock.Unlock() |
| 662 | return |
| 663 | } |
| 664 | c.stopped = true |
| 665 | c.stopLock.Unlock() |
| 666 | close(c.stopCh) |
| 667 | c.stopWg.Wait() |
| 668 | } |
| 669 | |
| 670 | func forgetWatcher(c *Cacher, index int, triggerValue string, triggerSupported bool) func(bool) { |
| 671 | return func(lock bool) { |
| 672 | if lock { |
| 673 | c.Lock() |
| 674 | defer c.Unlock() |
| 675 | } else { |
| 676 | // false is currently passed only if we are forcing watcher to close due |
| 677 | // to its unresponsiveness and blocking other watchers. |
| 678 | // TODO: Get this information in cleaner way. |
| 679 | glog.V(1).Infof("Forcing watcher close due to unresponsiveness: %v", c.objectType.String()) |
| 680 | } |
| 681 | // It's possible that the watcher is already not in the structure (e.g. in case of |
| 682 | // simultaneous Stop() and terminateAllWatchers(), but it doesn't break anything. |
| 683 | c.watchers.deleteWatcher(index, triggerValue, triggerSupported) |
| 684 | } |
| 685 | } |
| 686 | |
| 687 | func filterWithAttrsFunction(key string, p SelectionPredicate) filterWithAttrsFunc { |
| 688 | filterFunc := func(objKey string, label labels.Set, field fields.Set, uninitialized bool) bool { |
| 689 | if !hasPathPrefix(objKey, key) { |
| 690 | return false |
| 691 | } |
| 692 | return p.MatchesObjectAttributes(label, field, uninitialized) |
| 693 | } |
| 694 | return filterFunc |
| 695 | } |
| 696 | |
| 697 | // Returns resource version to which the underlying cache is synced. |
| 698 | func (c *Cacher) LastSyncResourceVersion() (uint64, error) { |
| 699 | c.ready.wait() |
| 700 | |
| 701 | resourceVersion := c.reflector.LastSyncResourceVersion() |
| 702 | return c.versioner.ParseListResourceVersion(resourceVersion) |
| 703 | } |
| 704 | |
| 705 | // cacherListerWatcher opaques storage.Interface to expose cache.ListerWatcher. |
| 706 | type cacherListerWatcher struct { |
| 707 | storage Interface |
| 708 | resourcePrefix string |
| 709 | newListFunc func() runtime.Object |
| 710 | } |
| 711 | |
| 712 | func newCacherListerWatcher(storage Interface, resourcePrefix string, newListFunc func() runtime.Object) cache.ListerWatcher { |
| 713 | return &cacherListerWatcher{ |
| 714 | storage: storage, |
| 715 | resourcePrefix: resourcePrefix, |
| 716 | newListFunc: newListFunc, |
| 717 | } |
| 718 | } |
| 719 | |
| 720 | // Implements cache.ListerWatcher interface. |
| 721 | func (lw *cacherListerWatcher) List(options metav1.ListOptions) (runtime.Object, error) { |
| 722 | list := lw.newListFunc() |
| 723 | if err := lw.storage.List(context.TODO(), lw.resourcePrefix, "", Everything, list); err != nil { |
| 724 | return nil, err |
| 725 | } |
| 726 | return list, nil |
| 727 | } |
| 728 | |
| 729 | // Implements cache.ListerWatcher interface. |
| 730 | func (lw *cacherListerWatcher) Watch(options metav1.ListOptions) (watch.Interface, error) { |
| 731 | return lw.storage.WatchList(context.TODO(), lw.resourcePrefix, options.ResourceVersion, Everything) |
| 732 | } |
| 733 | |
| 734 | // errWatcher implements watch.Interface to return a single error |
| 735 | type errWatcher struct { |
| 736 | result chan watch.Event |
| 737 | } |
| 738 | |
| 739 | func newErrWatcher(err error) *errWatcher { |
| 740 | // Create an error event |
| 741 | errEvent := watch.Event{Type: watch.Error} |
| 742 | switch err := err.(type) { |
| 743 | case runtime.Object: |
| 744 | errEvent.Object = err |
| 745 | case *errors.StatusError: |
| 746 | errEvent.Object = &err.ErrStatus |
| 747 | default: |
| 748 | errEvent.Object = &metav1.Status{ |
| 749 | Status: metav1.StatusFailure, |
| 750 | Message: err.Error(), |
| 751 | Reason: metav1.StatusReasonInternalError, |
| 752 | Code: http.StatusInternalServerError, |
| 753 | } |
| 754 | } |
| 755 | |
| 756 | // Create a watcher with room for a single event, populate it, and close the channel |
| 757 | watcher := &errWatcher{result: make(chan watch.Event, 1)} |
| 758 | watcher.result <- errEvent |
| 759 | close(watcher.result) |
| 760 | |
| 761 | return watcher |
| 762 | } |
| 763 | |
| 764 | // Implements watch.Interface. |
| 765 | func (c *errWatcher) ResultChan() <-chan watch.Event { |
| 766 | return c.result |
| 767 | } |
| 768 | |
| 769 | // Implements watch.Interface. |
| 770 | func (c *errWatcher) Stop() { |
| 771 | // no-op |
| 772 | } |
| 773 | |
| 774 | // cachWatcher implements watch.Interface |
| 775 | type cacheWatcher struct { |
| 776 | sync.Mutex |
| 777 | input chan *watchCacheEvent |
| 778 | result chan watch.Event |
| 779 | done chan struct{} |
| 780 | filter filterWithAttrsFunc |
| 781 | stopped bool |
| 782 | forget func(bool) |
| 783 | versioner Versioner |
| 784 | } |
| 785 | |
| 786 | func newCacheWatcher(resourceVersion uint64, chanSize int, initEvents []*watchCacheEvent, filter filterWithAttrsFunc, forget func(bool), versioner Versioner) *cacheWatcher { |
| 787 | watcher := &cacheWatcher{ |
| 788 | input: make(chan *watchCacheEvent, chanSize), |
| 789 | result: make(chan watch.Event, chanSize), |
| 790 | done: make(chan struct{}), |
| 791 | filter: filter, |
| 792 | stopped: false, |
| 793 | forget: forget, |
| 794 | versioner: versioner, |
| 795 | } |
| 796 | go watcher.process(initEvents, resourceVersion) |
| 797 | return watcher |
| 798 | } |
| 799 | |
| 800 | // Implements watch.Interface. |
| 801 | func (c *cacheWatcher) ResultChan() <-chan watch.Event { |
| 802 | return c.result |
| 803 | } |
| 804 | |
| 805 | // Implements watch.Interface. |
| 806 | func (c *cacheWatcher) Stop() { |
| 807 | c.forget(true) |
| 808 | c.stop() |
| 809 | } |
| 810 | |
| 811 | func (c *cacheWatcher) stop() { |
| 812 | c.Lock() |
| 813 | defer c.Unlock() |
| 814 | if !c.stopped { |
| 815 | c.stopped = true |
| 816 | close(c.done) |
| 817 | close(c.input) |
| 818 | } |
| 819 | } |
| 820 | |
| 821 | var timerPool sync.Pool |
| 822 | |
| 823 | func (c *cacheWatcher) add(event *watchCacheEvent, budget *timeBudget) { |
| 824 | // Try to send the event immediately, without blocking. |
| 825 | select { |
| 826 | case c.input <- event: |
| 827 | return |
| 828 | default: |
| 829 | } |
| 830 | |
| 831 | // OK, block sending, but only for up to <timeout>. |
| 832 | // cacheWatcher.add is called very often, so arrange |
| 833 | // to reuse timers instead of constantly allocating. |
| 834 | startTime := time.Now() |
| 835 | timeout := budget.takeAvailable() |
| 836 | |
| 837 | t, ok := timerPool.Get().(*time.Timer) |
| 838 | if ok { |
| 839 | t.Reset(timeout) |
| 840 | } else { |
| 841 | t = time.NewTimer(timeout) |
| 842 | } |
| 843 | defer timerPool.Put(t) |
| 844 | |
| 845 | select { |
| 846 | case c.input <- event: |
| 847 | stopped := t.Stop() |
| 848 | if !stopped { |
| 849 | // Consume triggered (but not yet received) timer event |
| 850 | // so that future reuse does not get a spurious timeout. |
| 851 | <-t.C |
| 852 | } |
| 853 | case <-t.C: |
| 854 | // This means that we couldn't send event to that watcher. |
| 855 | // Since we don't want to block on it infinitely, |
| 856 | // we simply terminate it. |
| 857 | c.forget(false) |
| 858 | c.stop() |
| 859 | } |
| 860 | |
| 861 | budget.returnUnused(timeout - time.Since(startTime)) |
| 862 | } |
| 863 | |
| 864 | // NOTE: sendWatchCacheEvent is assumed to not modify <event> !!! |
| 865 | func (c *cacheWatcher) sendWatchCacheEvent(event *watchCacheEvent) { |
| 866 | curObjPasses := event.Type != watch.Deleted && c.filter(event.Key, event.ObjLabels, event.ObjFields, event.ObjUninitialized) |
| 867 | oldObjPasses := false |
| 868 | if event.PrevObject != nil { |
| 869 | oldObjPasses = c.filter(event.Key, event.PrevObjLabels, event.PrevObjFields, event.PrevObjUninitialized) |
| 870 | } |
| 871 | if !curObjPasses && !oldObjPasses { |
| 872 | // Watcher is not interested in that object. |
| 873 | return |
| 874 | } |
| 875 | |
| 876 | var watchEvent watch.Event |
| 877 | switch { |
| 878 | case curObjPasses && !oldObjPasses: |
| 879 | watchEvent = watch.Event{Type: watch.Added, Object: event.Object.DeepCopyObject()} |
| 880 | case curObjPasses && oldObjPasses: |
| 881 | watchEvent = watch.Event{Type: watch.Modified, Object: event.Object.DeepCopyObject()} |
| 882 | case !curObjPasses && oldObjPasses: |
| 883 | // return a delete event with the previous object content, but with the event's resource version |
| 884 | oldObj := event.PrevObject.DeepCopyObject() |
| 885 | if err := c.versioner.UpdateObject(oldObj, event.ResourceVersion); err != nil { |
| 886 | utilruntime.HandleError(fmt.Errorf("failure to version api object (%d) %#v: %v", event.ResourceVersion, oldObj, err)) |
| 887 | } |
| 888 | watchEvent = watch.Event{Type: watch.Deleted, Object: oldObj} |
| 889 | } |
| 890 | |
| 891 | // We need to ensure that if we put event X to the c.result, all |
| 892 | // previous events were already put into it before, no matter whether |
| 893 | // c.done is close or not. |
| 894 | // Thus we cannot simply select from c.done and c.result and this |
| 895 | // would give us non-determinism. |
| 896 | // At the same time, we don't want to block infinitely on putting |
| 897 | // to c.result, when c.done is already closed. |
| 898 | |
| 899 | // This ensures that with c.done already close, we at most once go |
| 900 | // into the next select after this. With that, no matter which |
| 901 | // statement we choose there, we will deliver only consecutive |
| 902 | // events. |
| 903 | select { |
| 904 | case <-c.done: |
| 905 | return |
| 906 | default: |
| 907 | } |
| 908 | |
| 909 | select { |
| 910 | case c.result <- watchEvent: |
| 911 | case <-c.done: |
| 912 | } |
| 913 | } |
| 914 | |
| 915 | func (c *cacheWatcher) process(initEvents []*watchCacheEvent, resourceVersion uint64) { |
| 916 | defer utilruntime.HandleCrash() |
| 917 | |
| 918 | // Check how long we are processing initEvents. |
| 919 | // As long as these are not processed, we are not processing |
| 920 | // any incoming events, so if it takes long, we may actually |
| 921 | // block all watchers for some time. |
| 922 | // TODO: From the logs it seems that there happens processing |
| 923 | // times even up to 1s which is very long. However, this doesn't |
| 924 | // depend that much on the number of initEvents. E.g. from the |
| 925 | // 2000-node Kubemark run we have logs like this, e.g.: |
| 926 | // ... processing 13862 initEvents took 66.808689ms |
| 927 | // ... processing 14040 initEvents took 993.532539ms |
| 928 | // We should understand what is blocking us in those cases (e.g. |
| 929 | // is it lack of CPU, network, or sth else) and potentially |
| 930 | // consider increase size of result buffer in those cases. |
| 931 | const initProcessThreshold = 500 * time.Millisecond |
| 932 | startTime := time.Now() |
| 933 | for _, event := range initEvents { |
| 934 | c.sendWatchCacheEvent(event) |
| 935 | } |
| 936 | processingTime := time.Since(startTime) |
| 937 | if processingTime > initProcessThreshold { |
| 938 | objType := "<null>" |
| 939 | if len(initEvents) > 0 { |
| 940 | objType = reflect.TypeOf(initEvents[0].Object).String() |
| 941 | } |
| 942 | glog.V(2).Infof("processing %d initEvents of %s took %v", len(initEvents), objType, processingTime) |
| 943 | } |
| 944 | |
| 945 | defer close(c.result) |
| 946 | defer c.Stop() |
| 947 | for { |
| 948 | event, ok := <-c.input |
| 949 | if !ok { |
| 950 | return |
| 951 | } |
| 952 | // only send events newer than resourceVersion |
| 953 | if event.ResourceVersion > resourceVersion { |
| 954 | c.sendWatchCacheEvent(event) |
| 955 | } |
| 956 | } |
| 957 | } |
| 958 | |
| 959 | type ready struct { |
| 960 | ok bool |
| 961 | c *sync.Cond |
| 962 | } |
| 963 | |
| 964 | func newReady() *ready { |
| 965 | return &ready{c: sync.NewCond(&sync.Mutex{})} |
| 966 | } |
| 967 | |
| 968 | func (r *ready) wait() { |
| 969 | r.c.L.Lock() |
| 970 | for !r.ok { |
| 971 | r.c.Wait() |
| 972 | } |
| 973 | r.c.L.Unlock() |
| 974 | } |
| 975 | |
| 976 | // TODO: Make check() function more sophisticated, in particular |
| 977 | // allow it to behave as "waitWithTimeout". |
| 978 | func (r *ready) check() bool { |
| 979 | r.c.L.Lock() |
| 980 | defer r.c.L.Unlock() |
| 981 | return r.ok |
| 982 | } |
| 983 | |
| 984 | func (r *ready) set(ok bool) { |
| 985 | r.c.L.Lock() |
| 986 | defer r.c.L.Unlock() |
| 987 | r.ok = ok |
| 988 | r.c.Broadcast() |
| 989 | } |