[
https://issues.apache.org/jira/browse/SPARK-35623?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Dipanjan Kailthya updated SPARK-35623:
--------------------------------------
Description:
Dear Spark Developers,
Hello from the Netherlands! Sending this here as I still haven't gotten
accepted to post in the spark dev mailing list.
My team is planning to use spark with Kubernetes support on our shared
(multi-tenant) on premise Kubernetes cluster. However we would like to have
certain scheduling features like fair-share and preemption which as we
understand are not built into the current spark-kubernetes resource manager
yet. We have been working on and are close to a first successful prototype
integration with Volcano ([https://volcano.sh/en/docs/]). Briefly this means a
new resource manager component with lots in common with existing
spark-kubernetes resource manager, but instead of pods it launches Volcano jobs
which delegate the driver and executor pod creation and lifecycle management to
Volcano. We are interested in contributing this to open source, either directly
in spark or as a separate project.
So, two questions:
1. Do the spark maintainers see this as a valuable contribution to the
mainline spark codebase? If so, can we have some guidance on how to publish the
changes?
2. Are any other developers / organizations interested to contribute to this
effort? If so, please get in touch.
Best,
Dipanjan
was:
Dear Spark Developers,
Hello from the Netherlands! Sending this here as I still haven't gotten
accepted to post in the spark dev mailing list.
My team is planning to use spark with Kubernetes support on our shared
(multi-tenant) on premise Kubernetes cluster. However we would like to have
certain scheduling features like fair-share and preemption which as we
understand are not built into the current spark-kubernetes resource manager
yet. We have been working on and are close to a first successful prototype
integration with Volcano ([https://volcano.sh/en/docs/]). Briefly this means a
new resource manager component with lots in common with existing
spark-kubernetes resource manager, but instead of pods it launches Volcano jobs
which delegates the pod creation and lifecycle management to Volcano. We are
interested in contributing this to open source, either directly in spark or as
a separate project.
So, two questions:
1. Do the spark maintainers see this as a valuable contribution to the
mainline spark codebase? If so, can we have some guidance on how to publish the
changes?
2. Are any other developers / organizations interested to contribute to this
effort? If so, please get in touch.
Best,
Dipanjan
> Volcano resource manager for Spark on Kubernetes
> ------------------------------------------------
>
> Key: SPARK-35623
> URL: https://issues.apache.org/jira/browse/SPARK-35623
> Project: Spark
> Issue Type: Brainstorming
> Components: Kubernetes
> Affects Versions: 3.1.1, 3.1.2
> Reporter: Dipanjan Kailthya
> Priority: Minor
> Labels: kubernetes, resourcemanager
>
> Dear Spark Developers,
>
> Hello from the Netherlands! Sending this here as I still haven't gotten
> accepted to post in the spark dev mailing list.
>
> My team is planning to use spark with Kubernetes support on our shared
> (multi-tenant) on premise Kubernetes cluster. However we would like to have
> certain scheduling features like fair-share and preemption which as we
> understand are not built into the current spark-kubernetes resource manager
> yet. We have been working on and are close to a first successful prototype
> integration with Volcano ([https://volcano.sh/en/docs/]). Briefly this means
> a new resource manager component with lots in common with existing
> spark-kubernetes resource manager, but instead of pods it launches Volcano
> jobs which delegate the driver and executor pod creation and lifecycle
> management to Volcano. We are interested in contributing this to open source,
> either directly in spark or as a separate project.
>
> So, two questions:
>
> 1. Do the spark maintainers see this as a valuable contribution to the
> mainline spark codebase? If so, can we have some guidance on how to publish
> the changes?
>
> 2. Are any other developers / organizations interested to contribute to this
> effort? If so, please get in touch.
>
> Best,
> Dipanjan
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
