[
https://issues.apache.org/jira/browse/MESOS-3308?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Yan Xu updated MESOS-3308:
--------------------------
Description:
A few motivations:
1) Given the design in MESOS-3004 it became apparent that we need to support
multiple images in a container and these images can be of different image
types. (There are no sufficient reasons or major obstacles that force us not to
allow it and it obviously gives the users more flexibility).
2) Also, even though we currently allow only one backend for each provisioner,
when we update a running slave there can be multiple backends left in each
container that we need to launch tasks with, or at least recover. We should
evaluate in the future whether to support multiple backends and choose among
them dynamically based on image characteristics.
3) Since the rootfs' lifecycle tie with the running containers and should be
cleaned up after containers die, it fits into the pattern of {{word_dir}} and
we can manage them inside the work dir without needing to ask the operator to
specify more flags.
> Define the container rootfs directories within the slave work_dir.
> ------------------------------------------------------------------
>
> Key: MESOS-3308
> URL: https://issues.apache.org/jira/browse/MESOS-3308
> Project: Mesos
> Issue Type: Task
> Reporter: Yan Xu
>
> A few motivations:
> 1) Given the design in MESOS-3004 it became apparent that we need to support
> multiple images in a container and these images can be of different image
> types. (There are no sufficient reasons or major obstacles that force us not
> to allow it and it obviously gives the users more flexibility).
> 2) Also, even though we currently allow only one backend for each
> provisioner, when we update a running slave there can be multiple backends
> left in each container that we need to launch tasks with, or at least
> recover. We should evaluate in the future whether to support multiple
> backends and choose among them dynamically based on image characteristics.
> 3) Since the rootfs' lifecycle tie with the running containers and should be
> cleaned up after containers die, it fits into the pattern of {{word_dir}} and
> we can manage them inside the work dir without needing to ask the operator to
> specify more flags.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
