[
https://issues.apache.org/jira/browse/AURORA-1418?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14705844#comment-14705844
]
Maxim Khutornenko commented on AURORA-1418:
-------------------------------------------
The QoS controller is not going to kill a BE tasks (at least not in the
foreseeable future) but rather throttle them down on the host.
Regardless of the CPU behavior, RAM, DISK and PORTS are offered as
non-revocable resources. For example, a host offer could have:
- CPU: 8.0
- CPU: 12.0, Revocable
- RAM: 10GB
- DISK: 250GB
- PORTS: 32400-32500
In other words, only the CPU type will have a Revocable flag set. All other
non-compressible resources are going to be offered in a "shared" way, meaning
both revocable and non-revocable tasks have to compete for them.
The example above addresses a problem where a BE task takes an inordinate
amount of non-compressible resources and starves out non-BE tasks. Our
preemptor policy will not allow that to happen.
> Ignore revocable resources in preemptor
> ---------------------------------------
>
> Key: AURORA-1418
> URL: https://issues.apache.org/jira/browse/AURORA-1418
> Project: Aurora
> Issue Type: Task
> Components: Scheduler
> Reporter: Maxim Khutornenko
> Assignee: Maxim Khutornenko
>
> From design doc:
> {quote}
> Preemptor needs to ignore best effort task resources when searching for
> preemption slots.
> {quote}
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
