[
https://issues.apache.org/jira/browse/FLINK-31871?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17764296#comment-17764296
]
Praneeth Ramesh commented on FLINK-31871:
-----------------------------------------
[~gyfora]
Can you please help review
https://github.com/apache/flink-kubernetes-operator/pull/670
> Interpret Flink MemoryUnits according to the actual user input
> --------------------------------------------------------------
>
> Key: FLINK-31871
> URL: https://issues.apache.org/jira/browse/FLINK-31871
> Project: Flink
> Issue Type: Improvement
> Components: API / Core, Kubernetes Operator
> Reporter: Alexander Fedulov
> Assignee: Praneeth Ramesh
> Priority: Major
> Labels: pull-request-available
>
> Currently all MemorySize.MemoryUnits are interpreted in "bibyte" notation,
> regardless of the units that users specify:
> [https://github.com/apache/flink/blob/release-1.17/flink-core/src/main/java/org/apache/flink/configuration/MemorySize.java#L352-L356]
> (i.e. G = Gi)
> Flink Kubernetes Operator utilizes these units for specifying resources in
> user-facing CR API (taskManager.resource.memory, jobManager.resource.memory).
> In other places this CR requires native K8S units specification (i.e.
> spec.containers[*].ephemeral-storage).
> There are two issues with this:
> * users cannot rely on the same units notation (taskManager.resource.memory
> = 16Gi fails)
> * taskManager.resource.memory = 16G is not interpreted as other units in the
> spec (16G is implicitly converted into 16Gi)
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
