[
https://issues.apache.org/jira/browse/SPARK-36058?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17714556#comment-17714556
]
Hu Ziqian commented on SPARK-36058:
-----------------------------------
hi [~holden], i have a question about the statefulsetPodsAllocator.
I understand that with dynamic allocation, the driver will delete executor who
has idle exceed timeout. For example, we have executor 0 to 9, and the executor
5 is idle. the driver will delete executor 5 and adjust target pod number from
10 to 9. But with stateful set, the k8s will try to delete pod with max index,
for example executor 9.
So there is a conflict between deletion from driver and deletion from
controller manager of k8s.
I want to know is there any limitation when use statefulset pod allocator. If
not, how to avoid the conflict above?
> Support replicasets/job API
> ---------------------------
>
> Key: SPARK-36058
> URL: https://issues.apache.org/jira/browse/SPARK-36058
> Project: Spark
> Issue Type: Sub-task
> Components: Kubernetes
> Affects Versions: 3.2.0, 3.3.0
> Reporter: Holden Karau
> Assignee: Holden Karau
> Priority: Major
> Fix For: 3.3.0
>
>
> Volcano & Yunikorn both support scheduling invidual pods, but they also
> support higher level abstractions similar to the vanilla Kube replicasets
> which we can use to improve scheduling performance.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
