| 1 |
/* |
| 2 |
Copyright 2019 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 flowcontrol |
| 18 |
|
| 19 |
import ( |
| 20 |
"context" |
| 21 |
"strconv" |
| 22 |
"time" |
| 23 |
|
| 24 |
endpointsrequest "k8s.io/apiserver/pkg/endpoints/request" |
| 25 |
"k8s.io/apiserver/pkg/server/httplog" |
| 26 |
"k8s.io/apiserver/pkg/server/mux" |
| 27 |
fq "k8s.io/apiserver/pkg/util/flowcontrol/fairqueuing" |
| 28 |
"k8s.io/apiserver/pkg/util/flowcontrol/fairqueuing/eventclock" |
| 29 |
fqs "k8s.io/apiserver/pkg/util/flowcontrol/fairqueuing/queueset" |
| 30 |
"k8s.io/apiserver/pkg/util/flowcontrol/metrics" |
| 31 |
fcrequest "k8s.io/apiserver/pkg/util/flowcontrol/request" |
| 32 |
kubeinformers "k8s.io/client-go/informers" |
| 33 |
"k8s.io/klog/v2" |
| 34 |
"k8s.io/utils/clock" |
| 35 |
|
| 36 |
flowcontrol "k8s.io/api/flowcontrol/v1" |
| 37 |
flowcontrolclient "k8s.io/client-go/kubernetes/typed/flowcontrol/v1" |
| 38 |
) |
| 39 |
|
| 40 |
// ConfigConsumerAsFieldManager is how the config consuminng |
| 41 |
// controller appears in an ObjectMeta ManagedFieldsEntry.Manager |
| 42 |
const ConfigConsumerAsFieldManager = "api-priority-and-fairness-config-consumer-v1" |
| 43 |
|
| 44 |
// Interface defines how the API Priority and Fairness filter interacts with the underlying system. |
| 45 |
type Interface interface { |
| 46 |
// Handle takes care of queuing and dispatching a request |
| 47 |
// characterized by the given digest. The given `noteFn` will be |
| 48 |
// invoked with the results of request classification. |
| 49 |
// The given `workEstimator` is called, if at all, after noteFn. |
| 50 |
// `workEstimator` will be invoked only when the request |
| 51 |
// is classified as non 'exempt'. |
| 52 |
// 'workEstimator', when invoked, must return the |
| 53 |
// work parameters for the request. |
| 54 |
// If the request is queued then `queueNoteFn` will be called twice, |
| 55 |
// first with `true` and then with `false`; otherwise |
| 56 |
// `queueNoteFn` will not be called at all. If Handle decides |
| 57 |
// that the request should be executed then `execute()` will be |
| 58 |
// invoked once to execute the request; otherwise `execute()` will |
| 59 |
// not be invoked. |
| 60 |
// Handle() should never return while execute() is running, even if |
| 61 |
// ctx is cancelled or times out. |
| 62 |
Handle(ctx context.Context, |
| 63 |
requestDigest RequestDigest, |
| 64 |
noteFn func(fs *flowcontrol.FlowSchema, pl *flowcontrol.PriorityLevelConfiguration, flowDistinguisher string), |
| 65 |
workEstimator func() fcrequest.WorkEstimate, |
| 66 |
queueNoteFn fq.QueueNoteFn, |
| 67 |
execFn func(), |
| 68 |
) |
| 69 |
|
| 70 |
// Run monitors config objects from the main apiservers and causes |
| 71 |
// any needed changes to local behavior. This method ceases |
| 72 |
// activity and returns after the given channel is closed. |
| 73 |
Run(stopCh <-chan struct{}) error |
| 74 |
|
| 75 |
// Install installs debugging endpoints to the web-server. |
| 76 |
Install(c *mux.PathRecorderMux) |
| 77 |
|
| 78 |
// WatchTracker provides the WatchTracker interface. |
| 79 |
WatchTracker |
| 80 |
|
| 81 |
// MaxSeatsTracker is invoked from the work estimator to track max seats |
| 82 |
// that can be occupied by a request for a priority level. |
| 83 |
MaxSeatsTracker |
| 84 |
} |
| 85 |
|
| 86 |
// This request filter implements https://github.com/kubernetes/enhancements/blob/master/keps/sig-api-machinery/1040-priority-and-fairness/README.md |
| 87 |
|
| 88 |
// New creates a new instance to implement API priority and fairness |
| 89 |
func New( |
| 90 |
informerFactory kubeinformers.SharedInformerFactory, |
| 91 |
flowcontrolClient flowcontrolclient.FlowcontrolV1Interface, |
| 92 |
serverConcurrencyLimit int, |
| 93 |
) Interface { |
| 94 |
clk := eventclock.Real{} |
| 95 |
return NewTestable(TestableConfig{ |
| 96 |
Name: "Controller", |
| 97 |
Clock: clk, |
| 98 |
AsFieldManager: ConfigConsumerAsFieldManager, |
| 99 |
FoundToDangling: func(found bool) bool { return !found }, |
| 100 |
InformerFactory: informerFactory, |
| 101 |
FlowcontrolClient: flowcontrolClient, |
| 102 |
ServerConcurrencyLimit: serverConcurrencyLimit, |
| 103 |
ReqsGaugeVec: metrics.PriorityLevelConcurrencyGaugeVec, |
| 104 |
ExecSeatsGaugeVec: metrics.PriorityLevelExecutionSeatsGaugeVec, |
| 105 |
QueueSetFactory: fqs.NewQueueSetFactory(clk), |
| 106 |
}) |
| 107 |
} |
| 108 |
|
| 109 |
// TestableConfig carries the parameters to an implementation that is testable |
| 110 |
type TestableConfig struct { |
| 111 |
// Name of the controller |
| 112 |
Name string |
| 113 |
|
| 114 |
// Clock to use in timing deliberate delays |
| 115 |
Clock clock.PassiveClock |
| 116 |
|
| 117 |
// AsFieldManager is the string to use in the metadata for |
| 118 |
// server-side apply. Normally this is |
| 119 |
// `ConfigConsumerAsFieldManager`. This is exposed as a parameter |
| 120 |
// so that a test of competing controllers can supply different |
| 121 |
// values. |
| 122 |
AsFieldManager string |
| 123 |
|
| 124 |
// FoundToDangling maps the boolean indicating whether a |
| 125 |
// FlowSchema's referenced PLC exists to the boolean indicating |
| 126 |
// that FlowSchema's status should indicate a dangling reference. |
| 127 |
// This is a parameter so that we can write tests of what happens |
| 128 |
// when servers disagree on that bit of Status. |
| 129 |
FoundToDangling func(bool) bool |
| 130 |
|
| 131 |
// InformerFactory to use in building the controller |
| 132 |
InformerFactory kubeinformers.SharedInformerFactory |
| 133 |
|
| 134 |
// FlowcontrolClient to use for manipulating config objects |
| 135 |
FlowcontrolClient flowcontrolclient.FlowcontrolV1Interface |
| 136 |
|
| 137 |
// ServerConcurrencyLimit for the controller to enforce |
| 138 |
ServerConcurrencyLimit int |
| 139 |
|
| 140 |
// GaugeVec for metrics about requests, broken down by phase and priority_level |
| 141 |
ReqsGaugeVec metrics.RatioedGaugeVec |
| 142 |
|
| 143 |
// RatioedGaugePairVec for metrics about seats occupied by all phases of execution |
| 144 |
ExecSeatsGaugeVec metrics.RatioedGaugeVec |
| 145 |
|
| 146 |
// QueueSetFactory for the queuing implementation |
| 147 |
QueueSetFactory fq.QueueSetFactory |
| 148 |
} |
| 149 |
|
| 150 |
// NewTestable is extra flexible to facilitate testing |
| 151 |
func NewTestable(config TestableConfig) Interface { |
| 152 |
return newTestableController(config) |
| 153 |
} |
| 154 |
|
| 155 |
func (cfgCtlr *configController) Handle(ctx context.Context, requestDigest RequestDigest, |
| 156 |
noteFn func(fs *flowcontrol.FlowSchema, pl *flowcontrol.PriorityLevelConfiguration, flowDistinguisher string), |
| 157 |
workEstimator func() fcrequest.WorkEstimate, |
| 158 |
queueNoteFn fq.QueueNoteFn, |
| 159 |
execFn func()) { |
| 160 |
fs, pl, isExempt, req, startWaitingTime := cfgCtlr.startRequest(ctx, requestDigest, noteFn, workEstimator, queueNoteFn) |
| 161 |
queued := startWaitingTime != time.Time{} |
| 162 |
if req == nil { |
| 163 |
if queued { |
| 164 |
observeQueueWaitTime(ctx, pl.Name, fs.Name, strconv.FormatBool(req != nil), cfgCtlr.clock.Since(startWaitingTime)) |
| 165 |
} |
| 166 |
klog.V(7).Infof("Handle(%#+v) => fsName=%q, distMethod=%#+v, plName=%q, isExempt=%v, reject", requestDigest, fs.Name, fs.Spec.DistinguisherMethod, pl.Name, isExempt) |
| 167 |
return |
| 168 |
} |
| 169 |
klog.V(7).Infof("Handle(%#+v) => fsName=%q, distMethod=%#+v, plName=%q, isExempt=%v, queued=%v", requestDigest, fs.Name, fs.Spec.DistinguisherMethod, pl.Name, isExempt, queued) |
| 170 |
var executed bool |
| 171 |
idle, panicking := true, true |
| 172 |
defer func() { |
| 173 |
klog.V(7).Infof("Handle(%#+v) => fsName=%q, distMethod=%#+v, plName=%q, isExempt=%v, queued=%v, Finish() => panicking=%v idle=%v", |
| 174 |
requestDigest, fs.Name, fs.Spec.DistinguisherMethod, pl.Name, isExempt, queued, panicking, idle) |
| 175 |
if idle { |
| 176 |
cfgCtlr.maybeReap(pl.Name) |
| 177 |
} |
| 178 |
}() |
| 179 |
idle = req.Finish(func() { |
| 180 |
if queued { |
| 181 |
observeQueueWaitTime(ctx, pl.Name, fs.Name, strconv.FormatBool(req != nil), cfgCtlr.clock.Since(startWaitingTime)) |
| 182 |
} |
| 183 |
metrics.AddDispatch(ctx, pl.Name, fs.Name) |
| 184 |
fqs.OnRequestDispatched(req) |
| 185 |
executed = true |
| 186 |
startExecutionTime := cfgCtlr.clock.Now() |
| 187 |
defer func() { |
| 188 |
executionTime := cfgCtlr.clock.Since(startExecutionTime) |
| 189 |
httplog.AddKeyValue(ctx, "apf_execution_time", executionTime) |
| 190 |
metrics.ObserveExecutionDuration(ctx, pl.Name, fs.Name, executionTime) |
| 191 |
}() |
| 192 |
execFn() |
| 193 |
}) |
| 194 |
if queued && !executed { |
| 195 |
observeQueueWaitTime(ctx, pl.Name, fs.Name, strconv.FormatBool(req != nil), cfgCtlr.clock.Since(startWaitingTime)) |
| 196 |
} |
| 197 |
panicking = false |
| 198 |
} |
| 199 |
|
| 200 |
func observeQueueWaitTime(ctx context.Context, priorityLevelName, flowSchemaName, execute string, waitTime time.Duration) { |
| 201 |
metrics.ObserveWaitingDuration(ctx, priorityLevelName, flowSchemaName, execute, waitTime) |
| 202 |
endpointsrequest.TrackAPFQueueWaitLatency(ctx, waitTime) |
| 203 |
} |