RE: Bug#696154: cloud.debian.org: Please install 'less' by default on official Debian AMIs.
> De : Anders Jackson [mailto:firstname.lastname@example.org]
> Objet : Re: Bug#696154: cloud.debian.org: Please install 'less' by default on
> official Debian AMIs.
> 2012/12/17 Clint Byrum <email@example.com>:
> > Excerpts from Christopher Gervais's message of 2012-12-17 13:01:54 -0800:
> >> I have to agree with Thomas here. Anything beyond the bare minimum is
> >> a slippery slope. I'd suggest that any debates about including
> >> optional packages should happen on firstname.lastname@example.org, rather
> than here.
> > How is it a slippery slope if it is driven by data?
> Proper automatic installation goes from minimal to full through installation of
> > Also, consider that all packages downloaded from a mirror are going to
> > be metered in some way by most clouds (EC2 for sure). But images are
> > not metered. Given the rapid create/tear-down nature of cloud
> > instances and workloads, I'd say this purity may end up *costing users
Actually, only outbound bandwidth is metered and priced (Amazon/Azure), so the main difference in terms of cost is disk footprint and not bandwidth. The other relevant factor is installation time: so far, my tests (Azure with minimal footprint) were around 2 minutes for the provisioning phase, having download+installation might extend this significantly.
The other issue to consider is how those images will be used. If we start from a barebones image it means every user will have to reconfigure it to its needs, then make his own image from there. It basically means that the "reference" image will only be useable by skilled users (even if the skill boils down to a single command). On the other hand, most cloud providers have a distribution channel for custom images, so we'll see a lot of those emerging. I don't see the value in trying to keep any control there, except maybe to identify the most useful images for first-time users.
The goal of what we're trying to do, in my mind, is to have a reference for cloud providers, and a place where we can quickly and automatically update the base images as needed (be it a debian-initiated change or a cloud provider change). I would entirely and deliberately skip any supplemental package or script, and only keep the absolute minimum to have a machine booted, accessible, and able to interact with the cloud provider's infrastructure. How to transform this into a useable image is entirely another issue.