@bparees - this struck me as something you would find of interest and want
to weigh in on.

I also think it's related to this as one potential means to support
debugging of the container state:

#14561

On Friday, September 25, 2015, David Oppenheimer notifications@github.com
wrote:

We received an interesting suggestion today which has multiple layers to
it but basically the use case was that they would like us to keep Pod state
around for debugging in the case where a Pod is "moved" to another machine.
This would presumably be an option in the Pod, rather than default
behavior, since in the normal case people probably only care about the logs
and not the full state.

[Please read "move" here in the proper cattle-centric way, meaning "kill
old Pod and create a fungible replica on a different machine"]

Today IIUC the only scenario where we "move" a Pod to a new machine is if
the machine fails. So the feature request is that when the machine comes
back up, we want it to still have logs, containers, host directories,
Pod-scoped directories, etc. still intact. I guess this is somewhat related
to the Borg critical data concept, although for a very different purpose.

In the future I imagine we will want some other scenarios to trigger the
Pod to "move" -- for example there is a restart loop (kubelet keeps
restarting the container locally due to repeated OOM or something else that
boils down to "the Pod is just not going to work here, but might work in
another machine"). In this case you'd want to also keep the same state from
the old Pod around despite creating a replacement replica elsewhere.

The last use case that was brought up was debugging deadlock. There would
be an option to start a new replica on another node while keeping the
deadlocked Pod/container running so that you can attach a debugger to the
process that is deadlocked. Of course you'd also like to keep the same
state we've been talking about for the other cases (logs, containers, host
directories, Pod-scoped directories).

@mikedanese https://github.com/mikedanese does this sound like a fairly
accurate description of the request? Anything I left out?

—
Reply to this email directly or view it on GitHub
#14602.

thanks @derekwaynecarr, definitely the ability to get the pod logs seems extremely useful.

in relation to #14561, the ability to do more than what is described here and actually commit the containers of the failed/moved pod and push it to a registry so i can pull the image (basically a snapshot at that point) and do some investigation of the ephemeral filesystem/attempt to start the process, etc would also be extremely useful imho.

Sort of related: #3949

I have a slightly different use-case, still sort of related to what @davidopp mentioned. When implementing #17940 and still to come #17244 I was struck we don't have any option to gracefully terminate a pod. Aforementioned issues/PRs are dealing with a job which should be terminated upon certain conditions (timeout, remote termination). Such pod should then be in a failed state denoting that it was terminated prematurely. @davidopp does this also fits your use-case, or should I rather create a separate issue regarding the topic?

Related: #2789

Note that if you don't care about the resources consumed, it's easy to just keep a pod around by changing its labels to orphan it from its controller.

Note that if you don't care about the resources consumed, it's easy to just keep a pod around by changing its labels to orphan it from its controller.

I guess not. If I'm setting a deadline, I'm rather interested in limiting the resources consumption. At least that's how I see this.

Issues go stale after 30d of inactivity.
Mark the issue as fresh with /remove-lifecycle stale.
Freeze the issue for 90d with /lifecycle frozen.
Stale issues rot after an additional 30d of inactivity and eventually close.

If this issue is safe to close now please do so with /close.

Send feedback to sig-testing, kubernetes/test-infra and/or @fejta.
/lifecycle stale

Stale issues rot after 30d of inactivity.
Mark the issue as fresh with /remove-lifecycle rotten.
Rotten issues close after an additional 30d of inactivity.

If this issue is safe to close now please do so with /close.

Send feedback to sig-testing, kubernetes/test-infra and/or @fejta.
/lifecycle rotten
/remove-lifecycle stale

Rotten issues close after 30d of inactivity.
Reopen the issue with /reopen.
Mark the issue as fresh with /remove-lifecycle rotten.

Send feedback to sig-testing, kubernetes/test-infra and/or fejta.
/close