-
Notifications
You must be signed in to change notification settings - Fork 134
/
resource_cleanup.go
190 lines (165 loc) · 6.48 KB
/
resource_cleanup.go
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
package service
import (
"sort"
"time"
log "github.com/sirupsen/logrus"
v1 "k8s.io/api/core/v1"
"k8s.io/client-go/tools/cache"
"github.com/armadaproject/armada/internal/executor/configuration"
clusterContext "github.com/armadaproject/armada/internal/executor/context"
"github.com/armadaproject/armada/internal/executor/util"
)
type ResourceCleanupService struct {
clusterContext clusterContext.ClusterContext
kubernetesConfiguration configuration.KubernetesConfiguration
}
func NewResourceCleanupService(
clusterContext clusterContext.ClusterContext,
kubernetesConfiguration configuration.KubernetesConfiguration,
) (*ResourceCleanupService, error) {
service := &ResourceCleanupService{
clusterContext: clusterContext,
kubernetesConfiguration: kubernetesConfiguration,
}
/*
The purpose of this is to remove any associated ingress and services resources as soon as the pod enters a terminal state
There is a limited number of nodeports + clusterIPs in a cluster and we keep pods around after they complete (especially failed pods can live a long time)
If we wait for the pod to be cleaned up to delete the associated ingresses + services, we risk exhausting the number of nodeports/clusterIPs available
- Especially in the case someone submits lots of jobs that require these resources but fail instantly
We do set ownerreference on the services to point to the pod.
So in the case the cleanup below fails, the ownerreference will ensure it is cleaned up when the pod is
*/
_, err := clusterContext.AddPodEventHandler(cache.ResourceEventHandlerFuncs{
UpdateFunc: func(oldObj, newObj interface{}) {
pod, ok := newObj.(*v1.Pod)
if !ok {
log.Errorf("Failed to process pod event due to it being an unexpected type. Failed to process %+v", newObj)
return
}
if util.IsManagedPod(pod) && util.IsInTerminalState(pod) && util.HasIngress(pod) {
go service.removeAnyAssociatedIngress(pod)
}
},
})
if err != nil {
return nil, err
}
return service, nil
}
func (i *ResourceCleanupService) removeAnyAssociatedIngress(pod *v1.Pod) {
log.Infof("Removing any ingresses associated with pod %s (%s)", pod.Name, pod.Namespace)
services, err := i.clusterContext.GetServices(pod)
if err != nil {
log.Errorf("Failed to get associated services for pod %s (%s) because %s", pod.Name, pod.Namespace, err)
} else {
for _, service := range services {
err = i.clusterContext.DeleteService(service)
if err != nil {
log.Errorf("Failed to remove associated ingress for pod %s (%s) because %s", pod.Namespace, pod.Namespace, err)
continue
}
}
}
ingresses, err := i.clusterContext.GetIngresses(pod)
if err != nil {
log.Errorf("Failed to get associated ingresses for pod %s (%s) because %s", pod.Name, pod.Namespace, err)
} else {
for _, ingress := range ingresses {
err = i.clusterContext.DeleteIngress(ingress)
if err != nil {
log.Errorf("Failed to remove associated ingress for pod %s (%s) because %s", pod.Namespace, pod.Namespace, err)
continue
}
}
}
}
// CleanupResources
/*
* This function finds and delete old resources. It does this in two ways:
* - By deleting all expired terminated pods
* - Deleting non-expired terminated pods when then MaxTerminatedPods limit is exceeded
*/
func (r *ResourceCleanupService) CleanupResources() {
pods, err := r.clusterContext.GetActiveBatchPods()
if err != nil {
log.WithError(err).Errorf("Skipping pod cleanup due to an error while loading active batch pods")
return
}
allTerminatedPods := util.FilterPods(pods, util.IsPodFinishedAndReported)
// Expired pods are ones who are older than configured retention period
expiredTerminatedPods := util.FilterPods(allTerminatedPods, r.canPodBeRemoved)
nonExpiredTerminatedPods := util.RemovePodsFromList(allTerminatedPods, expiredTerminatedPods)
r.clusterContext.DeletePods(expiredTerminatedPods)
if len(nonExpiredTerminatedPods) > r.kubernetesConfiguration.MaxTerminatedPods {
numberOfPodsToDelete := len(nonExpiredTerminatedPods) - r.kubernetesConfiguration.MaxTerminatedPods
// We get the oldest pods from queues that have the terminated pods
// This means each queue has a "share" of terminated pods
// so one bad queue doesn't cause everyone to lose their terminated pod logs early
podsToDelete := getOldestPodsWithQueueFairShare(nonExpiredTerminatedPods, numberOfPodsToDelete)
r.clusterContext.DeletePods(podsToDelete)
}
}
func getOldestPodsWithQueueFairShare(pods []*v1.Pod, numberOfPodsLimit int) []*v1.Pod {
if len(pods) <= numberOfPodsLimit {
return pods
}
podsByQueue := groupPodsByQueueAndSortByPodAge(pods)
podsToReturn := make([]*v1.Pod, 0, numberOfPodsLimit)
for i := 0; i < numberOfPodsLimit; i++ {
queueWithMostPods := getQueueWithTheMostPods(podsByQueue)
if queueWithMostPods == "" {
log.Warn("Getting oldest pods for queues ended early due to lack of pods")
break
}
lastElementIndex := len(podsByQueue[queueWithMostPods]) - 1
oldestPod := podsByQueue[queueWithMostPods][lastElementIndex]
podsByQueue[queueWithMostPods] = podsByQueue[queueWithMostPods][:lastElementIndex]
podsToReturn = append(podsToReturn, oldestPod)
}
return podsToReturn
}
func getQueueWithTheMostPods(podsByQueue map[string][]*v1.Pod) string {
queueWithMostPods := ""
for queue := range podsByQueue {
if len(podsByQueue[queue]) > len(podsByQueue[queueWithMostPods]) {
queueWithMostPods = queue
}
}
return queueWithMostPods
}
func groupPodsByQueueAndSortByPodAge(pods []*v1.Pod) map[string][]*v1.Pod {
podsByQueue := util.GroupByQueue(pods)
for queue := range podsByQueue {
sort.Slice(podsByQueue[queue], func(i, j int) bool {
pod1 := podsByQueue[queue][i]
pod2 := podsByQueue[queue][j]
pod1LastStateChange, errPod1 := util.LastStatusChange(pod1)
pod2LastStateChange, errPod2 := util.LastStatusChange(pod2)
if errPod1 != nil {
return false
}
if errPod2 != nil {
return true
}
// Sort most recent time to oldest
return pod1LastStateChange.After(pod2LastStateChange)
})
}
return podsByQueue
}
func (r *ResourceCleanupService) canPodBeRemoved(pod *v1.Pod) bool {
if !util.IsPodFinishedAndReported(pod) {
return false
}
lastContainerStart := util.FindLastContainerStartTime(pod)
if lastContainerStart.Add(r.kubernetesConfiguration.MinimumPodAge).After(time.Now()) {
return false
}
if pod.Status.Phase == v1.PodFailed {
lastChange, err := util.LastStatusChange(pod)
if err == nil && lastChange.Add(r.kubernetesConfiguration.FailedPodExpiry).After(time.Now()) {
return false
}
}
return true
}