On 09/30/2014 01:28 AM, Marcela Mašláňová wrote:
Cool, we should do it better. From what I see every image contains just the main package downloaded from upstream. If those users are using cpan/pypi/npm, there's not much added value. I wouldn't say they are outdated at all. They are missing just released node.js 0.12 and quite fresh Ruby 2.2 (from interpreted languages).
Do you have vision what should be done? (Like package in dockerfile at least what we provide for RSHCL). Otherwise, we could create dockerfile for every existing framework, but we probably don't want to go this way and support all existing combinations. Too much maintenance. It might be good to use combinations, which we are providing in DevAssistant or pick one supported frameworks for every language, but someone has to pick the One...
I think this combinatorial explosion problem is why offering a lot of prebuilt images is the wrong way to go with this. The desire to do so is a reflection of the fact that the tooling for building and managing Docker images is still relatively immature - I don't know exactly when Koji was written, but I'd suggest that the current state of Docker image management is, at best, where the RPM build process would have been pre-Koji.
Instead of offering a wide array of prebuilt images, I think OpenShift Origin are going in the right direction with v3, taking full advantage of the notion of composable layers in the design of Docker: https://openshift.github.io/documentation/openshift-pep-010-docker-cartridge...
I don't think that's something we want to reinvent independently at the Fedora layer - instead, I'd encourage folks interested in the concept of a "Docker build service" to get involved with OpenShift Origin v3, and ensure that it's at least possible to get the built images *out* for direct deployment with Docker, rather than *only* allow them to be deployed into the OpenShift instance. Ideally, the image build service and image registry components would be independently deployable, without a full OpenShift instance around them. Either way though, the OpenShift Origin project has a relatively high concentration of Docker expertise at this point, and also a pressing need for an image build service as part of the v3 redesign, so it's likely a good direction to explore.
Regards, Nick.