[
https://issues.apache.org/jira/browse/MESOS-7398?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Benjamin Mahler updated MESOS-7398:
-----------------------------------
Component/s: multitenancy
> HierarchicalAllocatorProcess::allocatable make strong assumptions about both
> resource providers and users
> ---------------------------------------------------------------------------------------------------------
>
> Key: MESOS-7398
> URL: https://issues.apache.org/jira/browse/MESOS-7398
> Project: Mesos
> Issue Type: Improvement
> Components: allocation
> Reporter: Benjamin Bannier
> Labels: multitenancy, tech-debt
>
> The function {{HierarchicalAllocatorProcess::allocatable}} is used in the
> allocator to decide whether a set of resources will be considered when
> calculating offers. It currently hardcodes minimal requirements for a number
> of common resource kinds.
> While it seems to have in the past be used to enforce the offer side of
> minimal task resources by not offering resources which we didn't want to be
> used for tasks, it now seems to mainly help to minimize performance overhead
> from too many small offers (instead too small resource amounts are kept out
> of the offer pool until they became accumulated into larger resources).
> While {{allocatable}} has already in the past prevented allocating sets of
> only certain resource kinds (e.g., a {{Resources}} holding only GPU is not
> {{allocatable}}; the same holds for custom resource kinds), the current
> approach breaks down with the introduction of resource providers with
> MESOS-7235 which might provide a single kind of resource each and which in
> the case of external resource providers might never "reside" on the same
> agent as e.g., CPU.
> It seems that we need to separate the different concerns of {{allocatable}}
> into dedicated functions, and adjust it to remain useful in a world of
> (external) resource providers.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)