Redeploy unit in place

[jcderr]

It would be quite nice to be able to replace an existing unit in one go, if a change must be made, preserving its location.

• fleetctl submit some.unit.service
• fleetctl run some.unit.service
• edit some.unit.service
• fleetctl reload some.unit.service

Fleet should replace the unit conf in-place, stop the running unit, and restart it using the new conf.

[bcwaldon]

@jcderr How do we handle changes in [X-Fleet] parameters? The new version of the unit file may not be able to run on the same machine.

[jcderr]

I would say to re-evaluate it. If required, go ahead and reschedule it to another machine, but prefer to stay on the same machine if necessary.

[bcwaldon]

I think we can distill this down to a scheduling optimization. When the engine reconciler runs, it needs to identify when the contents of a given unit have changed and prefer the currently-scheduled host (if one exists).

[nnnnathann]

I would love to see reload, or even just restart features of systemctl make their way up into fleetctl. In my case, I would use it for reloading a docker image from the repository in place similar to the following example unit:

[Unit]
Description=hello

[Service]
EnvironmentFile=/etc/environment

# Pull the image from the repository
ExecStartPre=/usr/bin/docker pull localhost:5000/hello
ExecStartPre=-/usr/bin/docker kill hello
ExecStartPre=-/usr/bin/docker rm hello

ExecStart=/usr/bin/docker run --name="hello" -p 80:80 -e HOST_IP=${COREOS_PUBLIC_IPV4} localhost:5000/hello

ExecStop=/usr/bin/docker stop hello

# ExecReload could run on fleetctl restart,
# in this case it would pull the repository again
# and restart with updated image

ExecReload=/usr/bin/docker pull localhost:5000/hello
ExecReload=-/usr/bin/docker kill hello
ExecReload=-/usr/bin/docker rm hello
ExecReload=/usr/bin/docker run --name="hello" -p 80:80 -e HOST_IP=${COREOS_PUBLIC_IPV4} localhost:5000/hello

[X-Fleet]
X-Conflicts=hello.service

This particular unit is kind of awkward because there are so many repeated commands, but the way this works is if you call systemctl restart hello.service on the host machine, it will re-pull, and re-deploy the docker image with minimal downtime between the kill and run

I'm new to this whole thing, so I might be missing something obvious too.

Thanks for the great work on coreos to everyone involved.