[
https://issues.apache.org/jira/browse/FLINK-38101?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Santwana Verma updated FLINK-38101:
-----------------------------------
Description:
The Kubernetes API (for example,
{{{}Pod.spec.containers.resources.limits.cpu{}}}) generally supports fractional
cores in _both_ of these ways:
* As a SI-format string (e.g. “500m” is 500 millicores = 0.5 cores)
* As a decimal-format string (e.g. “0.5” is 0.5 cores, i.e., same value)
I.e., Kubernetes resource API calls can use either format, and the Kubernetes
API will _store_ the configuration in the SI-format (“500m”).
But FKO doesn't support the SI-Format (Relevant code is
[here|https://github.com/apache/flink-kubernetes-operator/blob/5b3856b9a781b508641ee9d901625953d8f15bf0/flink-kubernetes-operator-api/src/main/java/org/apache/flink/kubernetes/operator/api/spec/Resource.java#L35]).
The more K8s native way would be to have it act as a pass through to the K8s.
This will allow customers to have a seamless experience.
was:
The Kubernetes API (for example,
{{{}Pod.spec.containers.resources.limits.cpu{}}}) generally supports fractional
cores in _both_ of these ways:
* As a SI-format string (e.g. “500m” is 500 millicores = 0.5 cores)
* As a decimal-format string (e.g. “0.5” is 0.5 cores, i.e., same value)
I.e., Kubernetes resource API calls can use either format, and the Kubernetes
API will _store_ the configuration in the SI-format (“500m”).
But FKO doesn't support the SI-Format (Relevant code is
[here|https://github.com/apache/flink-kubernetes-operator/blob/5b3856b9a781b508641ee9d901625953d8f15bf0/flink-kubernetes-operator-api/src/main/java/org/apache/flink/kubernetes/operator/api/spec/Resource.java#L35]).
The more K8s native way would be to have it act as a pass through to the K8s.
This will allow customers to have a seamless experience.
(My understanding is that the double format helps FKO in some resource
management, but I am not sure for that.)
> Support SI Format for CPU resources
> -----------------------------------
>
> Key: FLINK-38101
> URL: https://issues.apache.org/jira/browse/FLINK-38101
> Project: Flink
> Issue Type: Improvement
> Components: Kubernetes Operator
> Reporter: Santwana Verma
> Priority: Major
>
> The Kubernetes API (for example,
> {{{}Pod.spec.containers.resources.limits.cpu{}}}) generally supports
> fractional cores in _both_ of these ways:
> * As a SI-format string (e.g. “500m” is 500 millicores = 0.5 cores)
> * As a decimal-format string (e.g. “0.5” is 0.5 cores, i.e., same value)
> I.e., Kubernetes resource API calls can use either format, and the Kubernetes
> API will _store_ the configuration in the SI-format (“500m”).
> But FKO doesn't support the SI-Format (Relevant code is
> [here|https://github.com/apache/flink-kubernetes-operator/blob/5b3856b9a781b508641ee9d901625953d8f15bf0/flink-kubernetes-operator-api/src/main/java/org/apache/flink/kubernetes/operator/api/spec/Resource.java#L35]).
>
> The more K8s native way would be to have it act as a pass through to the K8s.
> This will allow customers to have a seamless experience.
>
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
