Skip to content

Commit a752c27

Browse files
committed
wip
1 parent a8d8cca commit a752c27

File tree

1 file changed

+2
-3
lines changed
  • vertical-pod-autoscaler/enhancements/8515-support-custom-request-to-limit-ratio

1 file changed

+2
-3
lines changed

vertical-pod-autoscaler/enhancements/8515-support-custom-request-to-limit-ratio/README.md

Lines changed: 2 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -141,11 +141,10 @@ The behavior after implementing this feature is as follows:
141141

142142
* The admission controller will **accept** new VPA objects that include a configured `RequestToLimitRatio`.
143143
* For containers targeted by a VPA object using `RequestToLimitRatio`, the admission controller and/or the updater will enforce the configured ratio. Here are some examples of how this may happen:
144-
* **From default to a specific ratio**: This occurs when we have running Pods targeted by a VPA object that does not define `RequestToLimitRatio`. In this case, VPA uses the default ratio derived from the Pod `resources` stanzas. Once we specify a custom ratio using the `RequestToLimitRatio` field, the new ratio is not applied immediately, as the updater still relies on its current behavior to decide when to evict Pods or perform in-place updates. With the `InPlaceOrRecreate` mode, the updater sends patches to the `resize` subresource [as shown here](https://github.com/kubernetes/autoscaler/blob/7b95cb06cb0843c1cd9432a3db893c001e1bc33c/vertical-pod-autoscaler/pkg/updater/restriction/pods_inplace_restriction.go#L137). This can result in two outcomes:
144+
* **From default to a specific ratio**: This occurs when we have a running Pod targeted by a VPA object that does not define `RequestToLimitRatio`. In this case, VPA uses the default ratio derived from the Pod's `resources` stanza (assuming the Pod has a single container). Once we specify a custom ratio using the `RequestToLimitRatio` field, the new ratio is not applied immediately, as the updater still relies on its current behavior to decide when to evict the Pod or perform in-place update. With the `InPlaceOrRecreate` mode, the updater sends patches to the `resize` subresource [as shown here](https://github.com/kubernetes/autoscaler/blob/7b95cb06cb0843c1cd9432a3db893c001e1bc33c/vertical-pod-autoscaler/pkg/updater/restriction/pods_inplace_restriction.go#L137). This can result in two outcomes:
145145
1. If the in-place update fails for any reason - for example, because the change would alter the Pod's QoS class - the updater evicts the Pod.
146146
2. If the in-place update succeeds, the updater is finished for that Pod.
147-
* **From one ratio to another**:
148-
In this case, the default ratio defined in the Pod `resources` stanzas is ignored, and the ratio specified in the `RequestToLimitRatio` field is enforced. The same logic from the first example applies (see points 1 and 2 above).
147+
* **From one ratio to another**: In this case, the default ratio defined in the Pod's `resources` stanza is ignored, and the ratio specified in the `RequestToLimitRatio` field is enforced.
149148

150149

151150
#### When Disabled

0 commit comments

Comments
 (0)