[
https://issues.apache.org/jira/browse/YUNIKORN-564?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Wilfred Spiegelenburg resolved YUNIKORN-564.
--------------------------------------------
Fix Version/s: 0.10
Resolution: Fixed
StateAware now considers all applications when deciding what to mark for scale
up. This means that only for the scale up check StateAware falls back to FIFO.
This allows the scale up to look ahead and scale up as much as possible as soon
as possible while still honouring the queue quota.
The optimisation is implemented in the queue logic and does not change the
sorter itself.
> Improve the upscaling performance for StateAware policy
> -------------------------------------------------------
>
> Key: YUNIKORN-564
> URL: https://issues.apache.org/jira/browse/YUNIKORN-564
> Project: Apache YuniKorn
> Issue Type: Improvement
> Components: core - scheduler
> Reporter: Weiwei Yang
> Assignee: Weiwei Yang
> Priority: Major
> Labels: pull-request-available
> Fix For: 0.10
>
>
> When calculating the outstanding requests that potentially need to trigger
> the upscaling, we use the same sortingApp function that used by the core
> scheduling logic. When the sorting policy is StateAware, this policy filters
> out apps to limit num of apps that can be scheduled at each moment. This is
> designed for the scheduling phase, however, for trigging the upscaling, we do
> not need to filter those apps, as long as the requests can fit into the
> queues' headroom, we should allow them to trigger the upscaling. This will
> optimize the upscaling perf when the StateAware policy is used.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
