Documentation ¶
Index ¶
- Constants
- func RequestDelegated(ctx context.Context)
- func WatchInitialized(ctx context.Context)
- func WithInitializationSignal(ctx context.Context, signal InitializationSignal) context.Context
- type DroppedRequestsTracker
- type ForgetWatchFunc
- type InitializationSignal
- type Interface
- type MaxSeatsTracker
- type RequestDigest
- type StartFunction
- type TestableConfig
- type WatchTracker
Constants ¶
const ConfigConsumerAsFieldManager = "api-priority-and-fairness-config-consumer-v1"
ConfigConsumerAsFieldManager is how the config consuminng controller appears in an ObjectMeta ManagedFieldsEntry.Manager
const MinTarget = 0.001
Variables ¶
This section is empty.
Functions ¶
func RequestDelegated ¶ added in v0.22.4
RequestDelegated informs the priority and fairness dispatcher that a given request has been delegated to an aggregated API server. No-op when priority and fairness is disabled.
func WatchInitialized ¶ added in v0.22.0
WatchInitialized sends a signal to priority and fairness dispatcher that a given watch request has already been initialized.
func WithInitializationSignal ¶ added in v0.22.0
func WithInitializationSignal(ctx context.Context, signal InitializationSignal) context.Context
WithInitializationSignal creates a copy of parent context with priority and fairness initialization signal value.
Types ¶
type DroppedRequestsTracker ¶ added in v0.28.0
type DroppedRequestsTracker interface { // RecordDroppedRequest records a request that was just // dropped from processing. RecordDroppedRequest(plName string) // GetRetryAfter returns the current suggested value of // RetryAfter value. GetRetryAfter(plName string) int64 }
DroppedRequestsTracker is an interface that allows tracking a history od dropped requests in the system for the purpose of adjusting RetryAfter header to avoid system overload.
func NewDroppedRequestsTracker ¶ added in v0.28.0
func NewDroppedRequestsTracker() DroppedRequestsTracker
NewDroppedRequestsTracker is creating a new instance of DroppedRequestsTracker.
type ForgetWatchFunc ¶ added in v0.22.0
type ForgetWatchFunc func()
ForgetWatchFunc is a function that should be called to forget the previously registered watch from the watch tracker.
type InitializationSignal ¶ added in v0.22.0
type InitializationSignal interface { // Signal notifies the dispatcher about finished initialization. Signal() // Wait waits for the initialization signal. Wait() }
InitializationSignal is an interface that allows sending and handling initialization signals.
func NewInitializationSignal ¶ added in v0.22.0
func NewInitializationSignal() InitializationSignal
type Interface ¶
type Interface interface { // Handle takes care of queuing and dispatching a request // characterized by the given digest. The given `noteFn` will be // invoked with the results of request classification. // The given `workEstimator` is called, if at all, after noteFn. // `workEstimator` will be invoked only when the request // is classified as non 'exempt'. // 'workEstimator', when invoked, must return the // work parameters for the request. // If the request is queued then `queueNoteFn` will be called twice, // first with `true` and then with `false`; otherwise // `queueNoteFn` will not be called at all. If Handle decides // that the request should be executed then `execute()` will be // invoked once to execute the request; otherwise `execute()` will // not be invoked. // Handle() should never return while execute() is running, even if // ctx is cancelled or times out. Handle(ctx context.Context, requestDigest RequestDigest, noteFn func(fs *flowcontrol.FlowSchema, pl *flowcontrol.PriorityLevelConfiguration, flowDistinguisher string), workEstimator func() fcrequest.WorkEstimate, queueNoteFn fq.QueueNoteFn, execFn func(), ) // Run monitors config objects from the main apiservers and causes // any needed changes to local behavior. This method ceases // activity and returns after the given channel is closed. Run(stopCh <-chan struct{}) error // Install installs debugging endpoints to the web-server. Install(c *mux.PathRecorderMux) // WatchTracker provides the WatchTracker interface. WatchTracker // MaxSeatsTracker is invoked from the work estimator to track max seats // that can be occupied by a request for a priority level. MaxSeatsTracker }
Interface defines how the API Priority and Fairness filter interacts with the underlying system.
func New ¶
func New( informerFactory kubeinformers.SharedInformerFactory, flowcontrolClient flowcontrolclient.FlowcontrolV1beta3Interface, serverConcurrencyLimit int, ) Interface
New creates a new instance to implement API priority and fairness
func NewTestable ¶
func NewTestable(config TestableConfig) Interface
NewTestable is extra flexible to facilitate testing
type MaxSeatsTracker ¶ added in v0.25.13
type MaxSeatsTracker interface { // GetMaxSeats returns the maximum seats a request should occupy for a given priority level. GetMaxSeats(priorityLevelName string) uint64 // SetMaxSeats configures max seats for a priority level. SetMaxSeats(priorityLevelName string, maxSeats uint64) // ForgetPriorityLevel removes max seats tracking for a priority level. ForgetPriorityLevel(priorityLevelName string) }
MaxSeatsTracker is used to track max seats allocatable per priority level from the work estimator
func NewMaxSeatsTracker ¶ added in v0.25.13
func NewMaxSeatsTracker() MaxSeatsTracker
type RequestDigest ¶
type RequestDigest struct { RequestInfo *request.RequestInfo User user.Info }
RequestDigest holds necessary info from request for flow-control
func (RequestDigest) GoString ¶
func (rd RequestDigest) GoString() string
GoString produces a golang source expression of the value.
type StartFunction ¶
type StartFunction func(ctx context.Context, hashValue uint64) (execute bool, afterExecution func())
StartFunction begins the process of handling a request. If the request gets queued then this function uses the given hashValue as the source of entropy as it shuffle-shards the request into a queue. The descr1 and descr2 values play no role in the logic but appear in log messages. This method does not return until the queuing, if any, for this request is done. If `execute` is false then `afterExecution` is irrelevant and the request should be rejected. Otherwise the request should be executed and `afterExecution` must be called exactly once.
type TestableConfig ¶ added in v0.21.0
type TestableConfig struct { // Name of the controller Name string // Clock to use in timing deliberate delays Clock clock.PassiveClock // AsFieldManager is the string to use in the metadata for // server-side apply. Normally this is // `ConfigConsumerAsFieldManager`. This is exposed as a parameter // so that a test of competing controllers can supply different // values. AsFieldManager string // FoundToDangling maps the boolean indicating whether a // FlowSchema's referenced PLC exists to the boolean indicating // that FlowSchema's status should indicate a dangling reference. // This is a parameter so that we can write tests of what happens // when servers disagree on that bit of Status. FoundToDangling func(bool) bool // InformerFactory to use in building the controller InformerFactory kubeinformers.SharedInformerFactory // FlowcontrolClient to use for manipulating config objects FlowcontrolClient flowcontrolclient.FlowcontrolV1beta3Interface // ServerConcurrencyLimit for the controller to enforce ServerConcurrencyLimit int // GaugeVec for metrics about requests, broken down by phase and priority_level ReqsGaugeVec metrics.RatioedGaugeVec // RatioedGaugePairVec for metrics about seats occupied by all phases of execution ExecSeatsGaugeVec metrics.RatioedGaugeVec // QueueSetFactory for the queuing implementation QueueSetFactory fq.QueueSetFactory }
TestableConfig carries the parameters to an implementation that is testable
type WatchTracker ¶ added in v0.22.0
type WatchTracker interface { // RegisterWatch reqisters a watch based on the provided http.Request // in the tracker. It returns the function that should be called // to forget the watcher once it is finished. RegisterWatch(r *http.Request) ForgetWatchFunc // GetInterestedWatchCount returns the number of watches that are // potentially interested in a request with a given RequestInfo // for the purpose of estimating cost of that request. GetInterestedWatchCount(requestInfo *request.RequestInfo) int }
WatchTracker is an interface that allows tracking the number of watches in the system for the purpose of estimating the cost of incoming mutating requests.
func NewWatchTracker ¶ added in v0.22.0
func NewWatchTracker() WatchTracker