dokku is also meant to build a custom image on deploy : rather than using heroku's buildpacks, you can put a Dockerfile at the root of your project and it will be used instead.
So basically, you could put a Dockerfile file container just FROM and MAINTAINER, referring the image you want to use in the FROM, and dokku will download and execute it on `git push` (provided it can access to the image repository).
This is good at first, but I deployed on a very small machine, that couldn't handle the live application + the Docker image being built, thus why I wanted to build the image in my CI pipelines.
+1 for this -- requiring the docker image to be built/managed on the machine it's being deployed on is the simpler architectural choice (easier to debug, etc), but it doesn't necessarily make sense for production.
I wonder if there's a ticket about this on dokku already
[EDIT] - Couldn't find anything... Some tickets about how the containers are built and changing the base image but not much about.
I wonder if you could jury rig something like kraken[0] and make sure wherever your building images is a peer or something... Of course the simpler solution might be to add a CI step that just pushes the image (via the working `docker save` method) to the deployment machine(s)? Maybe if you have a staging environment, let CI push there, then if that machine is peered (via something like kraken) with production, production will get the image (though it may never run the image).
So basically, you could put a Dockerfile file container just FROM and MAINTAINER, referring the image you want to use in the FROM, and dokku will download and execute it on `git push` (provided it can access to the image repository).