[Question] Why does the dvdi need to call "unmount" when isolator "cleanup"

[gyliu513]

Here come a question is Why does the dvdi need to call "unmount" when isolator "cleanup"? I did not see much impact if we do not do unmount when isolator cleanup or recover, the only issue is that there will be some garbage mount info in/proc/mounts, any comments?

[clintkitson]

Leaving this one for @cantbewong

[gyliu513]

@cantbewong any comments on this? Thanks.

[cantbewong]

For external mounts, you want to unmount so that the task can be rescheduled to another agent node.

Some storage platforms do suppport pre-emptive mount, but some do not. Even when a pre-emptive mount is supported, it takes more time if the external volume is mounted elsewhere

[cantbewong]

Please close this issue if I answered your question

[gyliu513]

@cantbewong can you please show more detail for pre-emptive mount? Does there an example for this?

[gyliu513]

Also can you please show more detail for For external mounts, you want to unmount so that the task can be rescheduled to another agent node. ? Which framework support task reschedule? Does the reschedule means that if the task failed, find another agent host to re-run the task again? Thanks @cantbewong

[cantbewong]

For detail on pre-emptive mount please see the section titled Pre-emptive Volume Mount right on the readme.md file for this repo.

Marathon supports rescheduling to another agent node and this is what we test with. There is no reason that this can't be done by other Frameworks as well. One use case is a scenario where an agent node crashes or is permanently retired. Another is a batch workloads that can flexibly be run on any agent node offering sufficient resources.

[gyliu513]

@cantbewong thanks.