| 1 |
/* |
| 2 |
Copyright 2021 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 request |
| 18 |
|
| 19 |
import ( |
| 20 |
"errors" |
| 21 |
"sync" |
| 22 |
"time" |
| 23 |
|
| 24 |
"k8s.io/apimachinery/pkg/util/wait" |
| 25 |
"k8s.io/klog/v2" |
| 26 |
"k8s.io/utils/clock" |
| 27 |
) |
| 28 |
|
| 29 |
const ( |
| 30 |
// type deletion (it applies mostly to CRD) is not a very frequent |
| 31 |
// operation so we can afford to prune the cache at a large interval. |
| 32 |
// at the same time, we also want to make sure that the scalability |
| 33 |
// tests hit this code path. |
| 34 |
pruneInterval = 1 * time.Hour |
| 35 |
|
| 36 |
// the storage layer polls for object count at every 1m interval, we will allow |
| 37 |
// up to 2-3 transient failures to get the latest count for a given resource. |
| 38 |
staleTolerationThreshold = 3 * time.Minute |
| 39 |
) |
| 40 |
|
| 41 |
var ( |
| 42 |
// ObjectCountNotFoundErr is returned when the object count for |
| 43 |
// a given resource is not being tracked. |
| 44 |
ObjectCountNotFoundErr = errors.New("object count not found for the given resource") |
| 45 |
|
| 46 |
// ObjectCountStaleErr is returned when the object count for a |
| 47 |
// given resource has gone stale due to transient failures. |
| 48 |
ObjectCountStaleErr = errors.New("object count has gone stale for the given resource") |
| 49 |
) |
| 50 |
|
| 51 |
// StorageObjectCountTracker is an interface that is used to keep track of |
| 52 |
// of the total number of objects for each resource. |
| 53 |
// {group}.{resource} is used as the key name to update and retrieve |
| 54 |
// the total number of objects for a given resource. |
| 55 |
type StorageObjectCountTracker interface { |
| 56 |
// Set is invoked to update the current number of total |
| 57 |
// objects for the given resource |
| 58 |
Set(string, int64) |
| 59 |
|
| 60 |
// Get returns the total number of objects for the given resource. |
| 61 |
// The following errors are returned: |
| 62 |
// - if the count has gone stale for a given resource due to transient |
| 63 |
// failures ObjectCountStaleErr is returned. |
| 64 |
// - if the given resource is not being tracked then |
| 65 |
// ObjectCountNotFoundErr is returned. |
| 66 |
Get(string) (int64, error) |
| 67 |
|
| 68 |
// RunUntil starts all the necessary maintenance. |
| 69 |
RunUntil(stopCh <-chan struct{}) |
| 70 |
} |
| 71 |
|
| 72 |
// NewStorageObjectCountTracker returns an instance of |
| 73 |
// StorageObjectCountTracker interface that can be used to |
| 74 |
// keep track of the total number of objects for each resource. |
| 75 |
func NewStorageObjectCountTracker() StorageObjectCountTracker { |
| 76 |
return &objectCountTracker{ |
| 77 |
clock: &clock.RealClock{}, |
| 78 |
counts: map[string]*timestampedCount{}, |
| 79 |
} |
| 80 |
} |
| 81 |
|
| 82 |
// timestampedCount stores the count of a given resource with a last updated |
| 83 |
// timestamp so we can prune it after it goes stale for certain threshold. |
| 84 |
type timestampedCount struct { |
| 85 |
count int64 |
| 86 |
lastUpdatedAt time.Time |
| 87 |
} |
| 88 |
|
| 89 |
// objectCountTracker implements StorageObjectCountTracker with |
| 90 |
// reader/writer mutual exclusion lock. |
| 91 |
type objectCountTracker struct { |
| 92 |
clock clock.PassiveClock |
| 93 |
|
| 94 |
lock sync.RWMutex |
| 95 |
counts map[string]*timestampedCount |
| 96 |
} |
| 97 |
|
| 98 |
func (t *objectCountTracker) Set(groupResource string, count int64) { |
| 99 |
if count <= -1 { |
| 100 |
// a value of -1 indicates that the 'Count' call failed to contact |
| 101 |
// the storage layer, in most cases this error can be transient. |
| 102 |
// we will continue to work with the count that is in the cache |
| 103 |
// up to a certain threshold defined by staleTolerationThreshold. |
| 104 |
// in case this becomes a non transient error then the count for |
| 105 |
// the given resource will will eventually be removed from |
| 106 |
// the cache by the pruner. |
| 107 |
return |
| 108 |
} |
| 109 |
|
| 110 |
now := t.clock.Now() |
| 111 |
|
| 112 |
// lock for writing |
| 113 |
t.lock.Lock() |
| 114 |
defer t.lock.Unlock() |
| 115 |
|
| 116 |
if item, ok := t.counts[groupResource]; ok { |
| 117 |
item.count = count |
| 118 |
item.lastUpdatedAt = now |
| 119 |
return |
| 120 |
} |
| 121 |
|
| 122 |
t.counts[groupResource] = ×tampedCount{ |
| 123 |
count: count, |
| 124 |
lastUpdatedAt: now, |
| 125 |
} |
| 126 |
} |
| 127 |
|
| 128 |
func (t *objectCountTracker) Get(groupResource string) (int64, error) { |
| 129 |
staleThreshold := t.clock.Now().Add(-staleTolerationThreshold) |
| 130 |
|
| 131 |
t.lock.RLock() |
| 132 |
defer t.lock.RUnlock() |
| 133 |
|
| 134 |
if item, ok := t.counts[groupResource]; ok { |
| 135 |
if item.lastUpdatedAt.Before(staleThreshold) { |
| 136 |
return item.count, ObjectCountStaleErr |
| 137 |
} |
| 138 |
return item.count, nil |
| 139 |
} |
| 140 |
return 0, ObjectCountNotFoundErr |
| 141 |
} |
| 142 |
|
| 143 |
// RunUntil runs all the necessary maintenance. |
| 144 |
func (t *objectCountTracker) RunUntil(stopCh <-chan struct{}) { |
| 145 |
wait.PollUntil( |
| 146 |
pruneInterval, |
| 147 |
func() (bool, error) { |
| 148 |
// always prune at every pruneInterval |
| 149 |
return false, t.prune(pruneInterval) |
| 150 |
}, stopCh) |
| 151 |
klog.InfoS("StorageObjectCountTracker pruner is exiting") |
| 152 |
} |
| 153 |
|
| 154 |
func (t *objectCountTracker) prune(threshold time.Duration) error { |
| 155 |
oldestLastUpdatedAtAllowed := t.clock.Now().Add(-threshold) |
| 156 |
|
| 157 |
// lock for writing |
| 158 |
t.lock.Lock() |
| 159 |
defer t.lock.Unlock() |
| 160 |
|
| 161 |
for groupResource, count := range t.counts { |
| 162 |
if count.lastUpdatedAt.After(oldestLastUpdatedAtAllowed) { |
| 163 |
continue |
| 164 |
} |
| 165 |
delete(t.counts, groupResource) |
| 166 |
} |
| 167 |
|
| 168 |
return nil |
| 169 |
} |