Please sign your commits following these rules:
https://github.com/moby/moby/blob/master/CONTRIBUTING.md#sign-your-work
The easiest way to do this is to amend the last commit:

$ git clone -b "restart_monitor" git@github.com:kkroo/linuxkit.git somewhere
$ cd somewhere
$ git commit --amend -s --no-edit
$ git push -f

Amending updates the existing PR. You DO NOT need to open a new one.

Hi, we can't merge PRs unless they are signed-off-by as per instructions above.

I think restart policy should probably be configurable.

containerd/containerd#2318 shows it using container.Update to set it after starting, also using WithNoRestart on teardown. However I'm not completely convinced this is intended for this use case. From containerd/containerd#2318 (comment):

The usecase is for system level containers, it will allow remote snapshotters, runtimes, and such to be kept running without any external deps on a system. Its not built for the kube or docker restart policies, just a simple internal function to keep low level system containers running, maybe something like kubelet in a container ;)

It seems like it is for containerd internal functionality not 3rd party use (as much as some linuxkit containers could be considered "system level containers").

Addressing comments regarding race conditions. Regarding the use of this interface, it is a plugin with public interfaces and seems very well suited for improving the availability of linuxkit's system services.

This needs to be configurable for each service. Not all those who wander are lost. Not all services that stop are broken.

This needs to be configurable for each service. Not all those who wander are lost. Not all services that stop are broken.

Long running services are typically ones where any exit code would be anomalous. Yes, we need to continue to support one off execution. This change would obviously change that and restart each service that has exited. One off executables are still supported and probably better served under the onboot section of the runtime definition where there is reliable execution ordering for dependencies. This would provide reliable long running services to Linuxkit without the need to restart the whole system. Besides, I think it would be interesting to have a service that exits as part of its nominal behavior be monitored by this change to implement watchdogs or custom polling functionality.

@medic15 and @justincormack are correct, this needs to be configurable per service.

I still have concerns regarding this use of a plugin which is "The usecase is for system level containers" and "Its not built for the kube or docker restart policies, just a simple internal function to keep low level system containers running". Whether or not it happens to currently meet the use case here if we are using it outside its "support envelope" we have no guarantee that it will continue to be suitable in the future. I'd be happy to be overruled by the other maintainers on that though.

At present a few of the services we run may terminate (eg rngd if no rng is supported) but this could be changed.

My use case involves populating a Redis in-memory database using data from several disparate sources. This can't be done in the onboot phase since my init service and Redis cannot be running at the same time.

@medic15 yes that seems a reasonable use case where it should be configurable.

@medic15 @justincormack I still don't understand the use case and why such a one-off can't be added to the end of the onboot with some bash script that would wait for redis to come up. Hopefully you can help clarify.

In any case, I've added a boolean flag noRestart to the runtime configuration section than can disable the restart. PTAL!

Services are not started until all onboot containers have exited.

closing this as there was no activity for over a year and some changes were requested.