[
https://issues.apache.org/jira/browse/MESOS-3421?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14944233#comment-14944233
]
Anindya Sinha commented on MESOS-3421:
--------------------------------------
I am looking into this JIRA and had an observation, and was hoping if someone
can throw some light on the use case of sharing of other resource types:
In addition to shareable persistent volume disk resource, I understand the use
case of non-persistent volume disk resources being shared. Persistent volumes
are gc-ed only after a DESTROY is done, whereas non-persistent volumes can be
gc-ed only after all containers using that persistent volume terminates; and in
both cases, multiple task containers on the same agent can have access to the
same volume simultaneously.
However, what other resource types can be shared? Is there any use case of
sharing cpu, mem, ports and disk (without a volume associated with it) being
shareable? IIUC, shareable resources are relevant not for resource types, but
when a discrete "thing" is associated with the resource type. eg. I understand
sharing a "volume" (persistent or non-persistent) being a valid scenario for
disk resource, but not sharing just a disk resource (since that is more like a
pool from which tasks are given chunks). Maybe I am just misunderstanding the
use cases here.
Regarding *custom* resource types, are there examples which illustrate sharing
of resource types?
If the above is accurate, we should focus on "shareable" attribute in
"DiskInfo" as opposed to it being a part of the higher level "Resource". If and
when there is another resource type identified that can be shared, we should
have it as a part of a subsection within Resource protobuf (similar to
DiskInfo).
> Support sharing persistent volumes across task instances
> --------------------------------------------------------
>
> Key: MESOS-3421
> URL: https://issues.apache.org/jira/browse/MESOS-3421
> Project: Mesos
> Issue Type: Improvement
> Components: general
> Affects Versions: 0.23.0
> Reporter: Anindya Sinha
> Assignee: Anindya Sinha
>
> A service that needs persistent volume needs to have access to the same
> persistent volume (RW) from multiple task(s) instances on the same agent
> node. Currently, a persistent volume once offered to the framework(s) can be
> scheduled to a task and until that tasks terminates, that persistent volume
> cannot be used by another task.
> Explore providing the capability of sharing persistent volumes across task
> instances scheduled on a single agent node.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
