Skip to content
This repository has been archived by the owner on Jun 6, 2024. It is now read-only.

User specified GPUType job will not trigger yarn preemption. #445

Closed
qinchen123 opened this issue Apr 9, 2018 · 4 comments
Closed

User specified GPUType job will not trigger yarn preemption. #445

qinchen123 opened this issue Apr 9, 2018 · 4 comments
Assignees
Labels
hadoop-ai Issue or feature relative to hadoop-ai new feature

Comments

@qinchen123
Copy link
Contributor

GPU type is PAI supported feature, but not supported in hadoop yarn.
If user submit job with GPU Type specified, this job will not preempt others job even if his VC quota is occupied by others.

@fanyangCS fanyangCS changed the title User specified GPUType job will not trigge yarn preemption. User specified GPUType job will not trigger yarn preemption. Apr 17, 2018
@hao1939 hao1939 added new feature hadoop-ai Issue or feature relative to hadoop-ai labels Jul 9, 2018
@qinchen123
Copy link
Contributor Author

Update hadoop-ai to hadoop-2.9.0 will resovle it.

@fanyangCS fanyangCS assigned qinchen123 and unassigned fanyangCS Aug 21, 2018
@fanyangCS fanyangCS assigned mzmssg and unassigned qinchen123 Nov 21, 2018
@fanyangCS
Copy link
Contributor

we will remove GPU type support when submitting a job. We will bind GPU type to a VC, and assuming one VC only has one SKUs.

@GeoffreyChen777
Copy link
Contributor

we will remove GPU type support when submitting a job. We will bind GPU type to a VC, and assuming one VC only has one SKUs.

@fanyangCS Is this feature in plan now?

@fanyangCS
Copy link
Contributor

@GeoffreyChen777 , yes. the plan is to introduce a "tag" like feature, when submitting a job, user can specify the tag so that the scheduler knows what to match when finding a qualified node.

Initial design passed, pending release date

Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Labels
hadoop-ai Issue or feature relative to hadoop-ai new feature
Projects
None yet
Development

No branches or pull requests

5 participants