はじめに
前回はPodのスケジューリング制御方法としてNode Affinityなどの動作を確認しました。今回は似たような機能でTaints/Tolerationsの動作を確認したいと思います。
Node AffinityなどはPodデプロイ時にスケジューリングするNodeを引き付ける動作でしたが、Taints/Tolerationsは逆になります。Nodeに対してTaints（直訳すると汚れ、汚染）を設定し、PodにTolerations（許容範囲）を設定することで、Taintsを許容できるPodのみをNodeが許可します。
Effect
TaintsとTolerationsが合わないときの挙動を「Effect」で指定します。Effectは以下の3種類があります。
| Effect | 概要 |
|---|---|
| PreferNoSchedule | Taintsを許容できないPodを可能な限りスケジューリングしませんが、他にスケジューリングできるノードがなければスケジューリングされます。 |
| NoSchedule | Taintsを許容できなPodをスケジューリングしません。ただし、すでにデプロイされているPodはそのままです。 |
| Noexecute | Taintsを許容できなPodをスケジューリングしません。さらに、すでにデプロイされているPodは追い出されます。 |
NoSchedule
各Effectの動作を確認します。順番が異なりますが、まずはNoScheduleから確認します。
Taintsの設定
Nodeに対してTaintsを設定します。今回はk8s-worker01のみに設定します。
$ kubectl taint node k8s-worker01 env=prd:NoSchedule
node/k8s-worker01 tainted
$ kubectl describe node k8s-worker01 | grep Taints
Taints: env=prd:NoSchedule
$ kubectl describe node k8s-worker02 | grep Taints
Taints: <none>
Tolerationsの設定とPodのデプロイ
TolerationsはPodのspec.tolerations配下に指定します。
nginx-prd.yaml
apiVersion: apps/v1
kind: Deployment
metadata:
name: nginx-prd
spec:
replicas: 4
selector:
matchLabels:
app: app1
template:
metadata:
labels:
app: app1
spec:
containers:
- name: nginx
image: nginx:latest
tolerations:
- key: "env"
operator: "Equal" #Equal/Existsいずれかを指定。Equal:keyとvalueが等しい/Exists:keyが存在する
value: "prd"
effect: "NoSchedule"
このほかに、valueをstgにした「nginx-stg」とTolerationsを指定していない「nginx-unspecified」を用意しました。
各Deploymentをapplyします。
$ kubectl apply -f nginx-prd.yaml
deployment.apps/nginx-prd created
$ kubectl apply -f nginx-stg.yaml
deployment.apps/nginx-stg created
$ kubectl apply -f nginx-unspecified.yaml
deployment.apps/nginx-unspecified created
$ kubectl get pod -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
nginx-prd-686bbc6cfd-9b688 1/1 Running 0 96s 192.168.79.78 k8s-worker01 <none> <none>
nginx-prd-686bbc6cfd-dttdl 1/1 Running 0 96s 192.168.69.218 k8s-worker02 <none> <none>
nginx-prd-686bbc6cfd-q9p62 1/1 Running 0 96s 192.168.69.227 k8s-worker02 <none> <none>
nginx-prd-686bbc6cfd-rsvcd 1/1 Running 0 96s 192.168.79.125 k8s-worker01 <none> <none>
nginx-stg-74c9c9d964-f5pjj 1/1 Running 0 56s 192.168.69.237 k8s-worker02 <none> <none>
nginx-stg-74c9c9d964-sn7xs 1/1 Running 0 56s 192.168.69.234 k8s-worker02 <none> <none>
nginx-stg-74c9c9d964-svt86 1/1 Running 0 56s 192.168.69.232 k8s-worker02 <none> <none>
nginx-stg-74c9c9d964-tf8zd 1/1 Running 0 56s 192.168.69.236 k8s-worker02 <none> <none>
nginx-unspecified-5589d85476-5rxnj 1/1 Running 0 22s 192.168.69.246 k8s-worker02 <none> <none>
nginx-unspecified-5589d85476-t5psw 1/1 Running 0 22s 192.168.69.248 k8s-worker02 <none> <none>
nginx-unspecified-5589d85476-vn6b8 1/1 Running 0 22s 192.168.69.241 k8s-worker02 <none> <none>
nginx-unspecified-5589d85476-z8m5p 1/1 Running 0 22s 192.168.69.243 k8s-worker02 <none> <none>
nginx-prdは2つのノードに分散されていて、stgとunspecifiedはk8s-worker02に寄せられていますね。
Taintsを許容できないPodをスケジューリングしませんが、許容できるPodをすべてスケジューリングするわけではないですね。
Taintsを設定したNodeしかない場合の動作
今度はTaintsを設定したNodeしかない場合の動作を確認します。
いったん、全てのDeploymentを削除し、k8s-worker02をスケジューリング対象から外します。
$ kubectl delete -f .
deployment.apps "nginx-prd" deleted
deployment.apps "nginx-stg" deleted
deployment.apps "nginx-unspecified" deleted
$ kubectl cordon k8s-worker02
node/k8s-worker02 cordoned
$ kubectl get node
NAME STATUS ROLES AGE VERSION
k8s-master Ready master 107d v1.17.3
k8s-worker01 Ready <none> 107d v1.17.3
k8s-worker02 Ready,SchedulingDisabled <none> 107d v1.17.3
再度、全てのDeploymentをapplyします。
$ kubectl apply -f .
deployment.apps/nginx-prd created
deployment.apps/nginx-stg created
deployment.apps/nginx-unspecified created
$ kubectl get pod -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
nginx-prd-686bbc6cfd-2p2qv 1/1 Running 0 73s 192.168.79.91 k8s-worker01 <none> <none>
nginx-prd-686bbc6cfd-fvvk6 1/1 Running 0 73s 192.168.79.95 k8s-worker01 <none> <none>
nginx-prd-686bbc6cfd-lmrb5 1/1 Running 0 73s 192.168.79.100 k8s-worker01 <none> <none>
nginx-prd-686bbc6cfd-tt9d5 1/1 Running 0 73s 192.168.79.83 k8s-worker01 <none> <none>
nginx-stg-74c9c9d964-2k8hw 0/1 Pending 0 73s <none> <none> <none> <none>
nginx-stg-74c9c9d964-stn65 0/1 Pending 0 73s <none> <none> <none> <none>
nginx-stg-74c9c9d964-tqlb4 0/1 Pending 0 73s <none> <none> <none> <none>
nginx-stg-74c9c9d964-v48zf 0/1 Pending 0 73s <none> <none> <none> <none>
nginx-unspecified-5589d85476-5hbl9 0/1 Pending 0 73s <none> <none> <none> <none>
nginx-unspecified-5589d85476-76s62 0/1 Pending 0 73s <none> <none> <none> <none>
nginx-unspecified-5589d85476-bljlw 0/1 Pending 0 73s <none> <none> <none> <none>
nginx-unspecified-5589d85476-c8cpn 0/1 Pending 0 73s <none> <none> <none> <none>
stgとunspecifiedはスケジューリングできるNodeがないので、Pendingで止まっています。
PreferNoSchedule
次にPreferNoScheduleの動作を確認します。
Deploymentを削除し、k8s-worker02をスケジューリング対象に戻しておきます。
$ kubectl delete -f .
deployment.apps "nginx-prd" deleted
deployment.apps "nginx-stg" deleted
deployment.apps "nginx-unspecified" deleted
$ kubectl uncordon k8s-worker02
node/k8s-worker02 uncordoned
Taintsの設定
k8s-worker01のTaintsを削除してから、再設定します。削除する際には「<key>-」を指定します。
$ kubectl taint node k8s-worker01 env-
node/k8s-worker01 untainted
$ kubectl taint node k8s-worker01 env=prd:PreferNoSchedule
node/k8s-worker01 tainted
$ kubectl describe node k8s-worker01 | grep Taint
Taints: env=prd:PreferNoSchedule
Podのデプロイ
nginx-prd/nginx-stg のマニフェストのeffectを「PreferNoSchedule」に変更してapplyします。
$ kubectl apply -f .
deployment.apps/nginx-prd created
deployment.apps/nginx-stg created
deployment.apps/nginx-unspecified created
$ kubectl get pod -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
nginx-prd-5678ddf968-5hck2 1/1 Running 0 38s 192.168.69.193 k8s-worker02 <none> <none>
nginx-prd-5678ddf968-h9hw4 1/1 Running 0 38s 192.168.69.245 k8s-worker02 <none> <none>
nginx-prd-5678ddf968-qg9th 1/1 Running 0 38s 192.168.79.65 k8s-worker01 <none> <none>
nginx-prd-5678ddf968-spz2k 1/1 Running 0 38s 192.168.79.81 k8s-worker01 <none> <none>
nginx-stg-cb887fb9d-84vkr 1/1 Running 0 38s 192.168.69.200 k8s-worker02 <none> <none>
nginx-stg-cb887fb9d-9mlvc 1/1 Running 0 38s 192.168.69.198 k8s-worker02 <none> <none>
nginx-stg-cb887fb9d-cl5m2 1/1 Running 0 38s 192.168.79.89 k8s-worker01 <none> <none>
nginx-stg-cb887fb9d-n7sqv 1/1 Running 0 38s 192.168.69.196 k8s-worker02 <none> <none>
nginx-unspecified-5589d85476-hkw9b 1/1 Running 0 38s 192.168.69.235 k8s-worker02 <none> <none>
nginx-unspecified-5589d85476-p5nm9 1/1 Running 0 38s 192.168.69.201 k8s-worker02 <none> <none>
nginx-unspecified-5589d85476-t86h4 1/1 Running 0 38s 192.168.79.68 k8s-worker01 <none> <none>
nginx-unspecified-5589d85476-zrc7x 1/1 Running 0 38s 192.168.69.199 k8s-worker02 <none> <none>
Taintsを許容できなPodは可能な限りスケジューリングしないはずですが、stgとunspecifiedも1つのPodはk8s-worker01にスケジューリングされています。
クラスタのNodeが2Nodeなので、許容できないNodeにもスケジューリングされたのかなと想像しています。
NoExecute
最後にNoExecuteの動作を確認します。
Taintsの設定
Taintsを削除し、NoExecuteに再設定します。
$ kubectl taint node k8s-worker01 env-
node/k8s-worker01 untainted
$ kubectl taint node k8s-worker01 env=prd:NoExecute
node/k8s-worker01 tainted
Podの移動
Taintsを設定する際に、別ターミナルでPodの状態を監視していました。
Taintsを再設定したら、Podが以下のように移動しました。
$ kubectl get pod -o wide -w
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
nginx-prd-5678ddf968-5hck2 1/1 Running 0 2m34s 192.168.69.193 k8s-worker02 <none> <none>
nginx-prd-5678ddf968-h9hw4 1/1 Running 0 2m34s 192.168.69.245 k8s-worker02 <none> <none>
nginx-prd-5678ddf968-qg9th 1/1 Running 0 2m34s 192.168.79.65 k8s-worker01 <none> <none>
nginx-prd-5678ddf968-spz2k 1/1 Running 0 2m34s 192.168.79.81 k8s-worker01 <none> <none>
nginx-stg-cb887fb9d-84vkr 1/1 Running 0 2m34s 192.168.69.200 k8s-worker02 <none> <none>
nginx-stg-cb887fb9d-9mlvc 1/1 Running 0 2m34s 192.168.69.198 k8s-worker02 <none> <none>
nginx-stg-cb887fb9d-cl5m2 1/1 Running 0 2m34s 192.168.79.89 k8s-worker01 <none> <none>
nginx-stg-cb887fb9d-n7sqv 1/1 Running 0 2m34s 192.168.69.196 k8s-worker02 <none> <none>
nginx-unspecified-5589d85476-hkw9b 1/1 Running 0 2m34s 192.168.69.235 k8s-worker02 <none> <none>
nginx-unspecified-5589d85476-p5nm9 1/1 Running 0 2m34s 192.168.69.201 k8s-worker02 <none> <none>
nginx-unspecified-5589d85476-t86h4 1/1 Running 0 2m34s 192.168.79.68 k8s-worker01 <none> <none>
nginx-unspecified-5589d85476-zrc7x 1/1 Running 0 2m34s 192.168.69.199 k8s-worker02 <none> <none>
nginx-unspecified-5589d85476-t86h4 1/1 Terminating 0 3m27s 192.168.79.68 k8s-worker01 <none> <none>
nginx-prd-5678ddf968-qg9th 1/1 Terminating 0 3m27s 192.168.79.65 k8s-worker01 <none> <none>
nginx-prd-5678ddf968-spz2k 1/1 Terminating 0 3m27s 192.168.79.81 k8s-worker01 <none> <none>
nginx-stg-cb887fb9d-cl5m2 1/1 Terminating 0 3m27s 192.168.79.89 k8s-worker01 <none> <none>
nginx-stg-cb887fb9d-xs45w 0/1 Pending 0 0s <none> <none> <none> <none>
nginx-prd-5678ddf968-t2znv 0/1 Pending 0 0s <none> <none> <none> <none>
nginx-unspecified-5589d85476-9cmk9 0/1 Pending 0 0s <none> <none> <none> <none>
nginx-stg-cb887fb9d-xs45w 0/1 Pending 0 0s <none> k8s-worker02 <none> <none>
nginx-prd-5678ddf968-t2znv 0/1 Pending 0 0s <none> k8s-worker02 <none> <none>
nginx-unspecified-5589d85476-9cmk9 0/1 Pending 0 0s <none> k8s-worker02 <none> <none>
nginx-prd-5678ddf968-6rt79 0/1 Pending 0 0s <none> <none> <none> <none>
nginx-prd-5678ddf968-6rt79 0/1 Pending 0 0s <none> k8s-worker02 <none> <none>
nginx-stg-cb887fb9d-xs45w 0/1 ContainerCreating 0 0s <none> k8s-worker02 <none> <none>
nginx-prd-5678ddf968-t2znv 0/1 ContainerCreating 0 0s <none> k8s-worker02 <none> <none>
nginx-unspecified-5589d85476-9cmk9 0/1 ContainerCreating 0 0s <none> k8s-worker02 <none> <none>
nginx-prd-5678ddf968-6rt79 0/1 ContainerCreating 0 0s <none> k8s-worker02 <none> <none>
nginx-stg-cb887fb9d-xs45w 0/1 ContainerCreating 0 2s <none> k8s-worker02 <none> <none>
nginx-stg-cb887fb9d-cl5m2 0/1 Terminating 0 3m29s 192.168.79.89 k8s-worker01 <none> <none>
nginx-prd-5678ddf968-6rt79 0/1 ContainerCreating 0 2s <none> k8s-worker02 <none> <none>
nginx-prd-5678ddf968-t2znv 0/1 ContainerCreating 0 2s <none> k8s-worker02 <none> <none>
nginx-unspecified-5589d85476-9cmk9 0/1 ContainerCreating 0 2s <none> k8s-worker02 <none> <none>
nginx-prd-5678ddf968-spz2k 0/1 Terminating 0 3m30s 192.168.79.81 k8s-worker01 <none> <none>
nginx-stg-cb887fb9d-cl5m2 0/1 Terminating 0 3m30s 192.168.79.89 k8s-worker01 <none> <none>
nginx-stg-cb887fb9d-cl5m2 0/1 Terminating 0 3m30s 192.168.79.89 k8s-worker01 <none> <none>
nginx-stg-cb887fb9d-cl5m2 0/1 Terminating 0 3m30s 192.168.79.89 k8s-worker01 <none> <none>
nginx-prd-5678ddf968-qg9th 0/1 Terminating 0 3m30s 192.168.79.65 k8s-worker01 <none> <none>
nginx-unspecified-5589d85476-t86h4 0/1 Terminating 0 3m30s <none> k8s-worker01 <none> <none>
nginx-prd-5678ddf968-spz2k 0/1 Terminating 0 3m32s 192.168.79.81 k8s-worker01 <none> <none>
nginx-prd-5678ddf968-spz2k 0/1 Terminating 0 3m32s 192.168.79.81 k8s-worker01 <none> <none>
nginx-stg-cb887fb9d-xs45w 1/1 Running 0 6s 192.168.69.255 k8s-worker02 <none> <none>
nginx-prd-5678ddf968-6rt79 1/1 Running 0 9s 192.168.69.254 k8s-worker02 <none> <none>
nginx-prd-5678ddf968-qg9th 0/1 Terminating 0 3m38s 192.168.79.65 k8s-worker01 <none> <none>
nginx-prd-5678ddf968-qg9th 0/1 Terminating 0 3m38s 192.168.79.65 k8s-worker01 <none> <none>
nginx-unspecified-5589d85476-t86h4 0/1 Terminating 0 3m38s <none> k8s-worker01 <none> <none>
nginx-unspecified-5589d85476-t86h4 0/1 Terminating 0 3m38s <none> k8s-worker01 <none> <none>
nginx-prd-5678ddf968-t2znv 1/1 Running 0 14s 192.168.69.252 k8s-worker02 <none> <none>
nginx-unspecified-5589d85476-9cmk9 1/1 Running 0 19s 192.168.69.205 k8s-worker02 <none> <none>
最終的には、以下のように全てのPodがk8s-worker02に移動しました。
これはPodのeffectがPreferNoScheduleのままですので、これも含めて許容できないと判断されてk8s-worker01から追い出されたのだと思います。
$ kubectl get pod -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
nginx-prd-5678ddf968-5hck2 1/1 Running 0 4m30s 192.168.69.193 k8s-worker02 <none> <none>
nginx-prd-5678ddf968-6rt79 1/1 Running 0 63s 192.168.69.254 k8s-worker02 <none> <none>
nginx-prd-5678ddf968-h9hw4 1/1 Running 0 4m30s 192.168.69.245 k8s-worker02 <none> <none>
nginx-prd-5678ddf968-t2znv 1/1 Running 0 63s 192.168.69.252 k8s-worker02 <none> <none>
nginx-stg-cb887fb9d-84vkr 1/1 Running 0 4m30s 192.168.69.200 k8s-worker02 <none> <none>
nginx-stg-cb887fb9d-9mlvc 1/1 Running 0 4m30s 192.168.69.198 k8s-worker02 <none> <none>
nginx-stg-cb887fb9d-n7sqv 1/1 Running 0 4m30s 192.168.69.196 k8s-worker02 <none> <none>
nginx-stg-cb887fb9d-xs45w 1/1 Running 0 63s 192.168.69.255 k8s-worker02 <none> <none>
nginx-unspecified-5589d85476-9cmk9 1/1 Running 0 63s 192.168.69.205 k8s-worker02 <none> <none>
nginx-unspecified-5589d85476-hkw9b 1/1 Running 0 4m30s 192.168.69.235 k8s-worker02 <none> <none>
nginx-unspecified-5589d85476-p5nm9 1/1 Running 0 4m30s 192.168.69.201 k8s-worker02 <none> <none>
nginx-unspecified-5589d85476-zrc7x 1/1 Running 0 4m30s 192.168.69.199 k8s-worker02 <none> <none>
nginx-prd/nginx-stg のマニフェストのeffectを「NoExecute」に変更してapplyします。
$ kubectl apply -f .
deployment.apps/nginx-prd configured
deployment.apps/nginx-stg configured
deployment.apps/nginx-unspecified unchanged
$ kubectl get pod -o wide
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
nginx-prd-8c4f499f6-6cd8d 1/1 Running 0 33s 192.168.79.115 k8s-worker01 <none> <none>
nginx-prd-8c4f499f6-787jr 1/1 Running 0 42s 192.168.79.104 k8s-worker01 <none> <none>
nginx-prd-8c4f499f6-j9tf2 1/1 Running 0 42s 192.168.79.99 k8s-worker01 <none> <none>
nginx-prd-8c4f499f6-w7pn5 1/1 Running 0 36s 192.168.79.114 k8s-worker01 <none> <none>
nginx-stg-7c7f996574-cf4gk 1/1 Running 0 33s 192.168.69.206 k8s-worker02 <none> <none>
nginx-stg-7c7f996574-j5zwd 1/1 Running 0 36s 192.168.69.209 k8s-worker02 <none> <none>
nginx-stg-7c7f996574-m6h6z 1/1 Running 0 42s 192.168.69.244 k8s-worker02 <none> <none>
nginx-stg-7c7f996574-zp8b5 1/1 Running 0 42s 192.168.69.207 k8s-worker02 <none> <none>
nginx-unspecified-5589d85476-9cmk9 1/1 Running 0 9m58s 192.168.69.205 k8s-worker02 <none> <none>
nginx-unspecified-5589d85476-hkw9b 1/1 Running 0 13m 192.168.69.235 k8s-worker02 <none> <none>
nginx-unspecified-5589d85476-p5nm9 1/1 Running 0 13m 192.168.69.201 k8s-worker02 <none> <none>
nginx-unspecified-5589d85476-zrc7x 1/1 Running 0 13m 192.168.69.199 k8s-worker02 <none> <none>
Taintsを許容できるnginx-prdのみがk8s-worker01にスケジューリングされていますね。
tolerationSecondsフィールド
NoExecutedeでは、tolerationSecondsフィールドを指定してすることで、許容できるPodを一定時間スケジューリングすることができます。
以下のマニフェストで動作を確認します。
nginx-prd.yaml
apiVersion: apps/v1
kind: Deployment
metadata:
name: nginx-prd
spec:
replicas: 2
selector:
matchLabels:
app: app1
template:
metadata:
labels:
app: app1
spec:
containers:
- name: nginx
image: nginx:latest
tolerations:
- key: "env"
operator: "Equal"
value: "prd"
effect: "NoExecute"
tolerationSeconds: 30
マニフェストをapplyして、Podの状態を別ターミナルで監視します。
なお、あらかじめ全てのPodは削除しています。
$ kubectl apply -f nginx-prd.yaml
deployment.apps/nginx-prd created
$ kubectl get pod -o wide -w
NAME READY STATUS RESTARTS AGE IP NODE NOMINATED NODE READINESS GATES
nginx-prd-cc5f68778-hhmx7 0/1 Pending 0 0s <none> <none> <none> <none>
nginx-prd-cc5f68778-hhmx7 0/1 Pending 0 0s <none> k8s-worker01 <none> <none>
nginx-prd-cc5f68778-r8l9p 0/1 Pending 0 0s <none> <none> <none> <none>
nginx-prd-cc5f68778-r8l9p 0/1 Pending 0 0s <none> k8s-worker02 <none> <none>
nginx-prd-cc5f68778-hhmx7 0/1 ContainerCreating 0 0s <none> k8s-worker01 <none> <none>
nginx-prd-cc5f68778-r8l9p 0/1 ContainerCreating 0 0s <none> k8s-worker02 <none> <none>
nginx-prd-cc5f68778-r8l9p 0/1 ContainerCreating 0 1s <none> k8s-worker02 <none> <none>
nginx-prd-cc5f68778-hhmx7 0/1 ContainerCreating 0 1s <none> k8s-worker01 <none> <none>
nginx-prd-cc5f68778-r8l9p 1/1 Running 0 6s 192.168.69.195 k8s-worker02 <none> <none>
nginx-prd-cc5f68778-hhmx7 1/1 Running 0 7s 192.168.79.110 k8s-worker01 <none> <none>
nginx-prd-cc5f68778-hhmx7 1/1 Terminating 0 30s 192.168.79.110 k8s-worker01 <none> <none>
nginx-prd-cc5f68778-l8zkk 0/1 Pending 0 0s <none> <none> <none> <none>
nginx-prd-cc5f68778-l8zkk 0/1 Pending 0 0s <none> k8s-worker01 <none> <none>
nginx-prd-cc5f68778-l8zkk 0/1 ContainerCreating 0 0s <none> k8s-worker01 <none> <none>
nginx-prd-cc5f68778-l8zkk 0/1 ContainerCreating 0 1s <none> k8s-worker01 <none> <none>
nginx-prd-cc5f68778-hhmx7 0/1 Terminating 0 32s <none> k8s-worker01 <none> <none>
nginx-prd-cc5f68778-hhmx7 0/1 Terminating 0 33s <none> k8s-worker01 <none> <none>
nginx-prd-cc5f68778-hhmx7 0/1 Terminating 0 33s <none> k8s-worker01 <none> <none>
nginx-prd-cc5f68778-l8zkk 1/1 Running 0 6s 192.168.79.126 k8s-worker01 <none> <none>
k8s-worker01にスケジューリングされたPodは指定した30秒だけ起動して停止していることがわかります。
Deploymentだと起動停止を繰り返します。指定時間だけ起動して停止させたままにするには、Podでデプロイする必要がありますね。
使いどころが難しそうな印象です。
まとめ
前回のNode Affinityなどと比べると少しわかりづらいですね。使い分けはまだ思いつかないです。