[
https://issues.apache.org/jira/browse/MESOS-1777?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14163134#comment-14163134
]
Zia Bhatti commented on MESOS-1777:
-----------------------------------
I have read the attached documents and have tried to build a picture of the
issues and possible solutions being discussed.
My 2 cents:
1. Clearly the desired storage has to be allocated to a task in some fashion
for the duration of the life of the task. No different than other resources
at some level.
2. Conceptually, if the discussion remains in the scope of ephemeral tasks than
a solution will remain complex. The nature of the beast is that some tasks are
not truly ephemeral. In this sense the node can restart or be in maintenance
mode. The task is sticky/persistent and needs to stay on the node along with
the data. The task will need to be restarted on the same node after a restart.
Either after a long timeout or by operator intervention (for
long-term/catastrophic node failures), a task may be made eligible for
reassignment. At that point, it is the task's responsibility to rebuild it's
state.
A large database/cache node should not be transitioned due to short-term node
failures.
If I may dream a bit more then a planned/managed migration feature can appear
in some subsequent version that would require a clean stop of the task,
scheduling the task on another node, data move/copy to the new node and then a
task start.
> Design persistent resources
> ---------------------------
>
> Key: MESOS-1777
> URL: https://issues.apache.org/jira/browse/MESOS-1777
> Project: Mesos
> Issue Type: Task
> Reporter: Jie Yu
> Assignee: Jie Yu
>
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
