Federated General Assembly » FGA Devops

Planetary has some VPS space available...

please pat/sam, correct me if i'm wrong, but the issue isn't just general server space. what we really need is an API through which we can provision new instances automatically, so that we can do it as a scripted step in a triggered build process. the sort of thing that rackspace & ec2 provide, openstack ideally is supposed to provide, and a few other folks have APIs for as well.

at least, that's what we need for devops phase 2. some general vps space (or baremetal) would be fine for phase 1.

Well, there's always "knife bootstrap", so an explicit API isn't critical. For me, the biggest concern is that we have access to cloud instances that we can spin up and blow away without having to go through 3 layers of support to reprovision -- partly why I'm big into infra in code is because I'm still learning, and I eff up servers like crazy and, rather than troubleshoot server-level things I'm not an expert in, I just blow them away and start fresh. If we don't have cloud servers, I'll probably just do all my dev and testing on my own rackspace servers, which is no big deal, except that I can't be as open :/

yep, that's essentially what i was trying to describe - cloud servers and their associated spin up-and-down controllable via a API. and for those reasons: there needs to be zero human intervention required for setup and teardown of the build process.

4m reminded me about the VPS space planetary has available. at minimum we could move the central build & static test instances over to there, i imagine. we don't really need a ton of resources there, i don't think - couple gig of RAM on a separate VM should give us plenty of headroom. and i see no reason we couldn't just run it all on a single instance, for the 'phase 1' plan.

after that, we'll still have to deal with some kind of service through which we can dynamically spin up and down the VMs.

Sam Boyer wrote:

please pat/sam, correct me if i'm wrong, but the issue isn't just general server space. what we really need is an API through which we can provision new instances automatically, so that we can do it as a scripted step in a triggered build process. the sort of thing that rackspace & ec2 provide, openstack ideally is supposed to provide, and a few other folks have APIs for as well.

at least, that's what we need for devops phase 2. some general vps space (or baremetal) would be fine for phase 1.

Is it not possible to use something like Aegir(http://community.aegirproject.org) (or extend it to those needs)? I think they talked implementation of "alien" platforms support.

no, Aegir is solving a different problem. the need here is managing the creation/destruction of a virtual server resource in the first place, not what we do with it once its created. Aegir is about provisioning Drupal instances once the server resource is already known and available. It's a bit of a mix of configuration management + drupal configuration wrapped into one.

In any case, separate problem, and not one that helps us much since we need to set up additional services (e.g., PuSH hubs) as part of the overall FGA stack. And the context of this discussion is one where we're creating/destroying these server resources as part of the testing process - and is largely independent of our strategy for hosting production instances.

So unfortunately, unlikely that it'd be useful for any of our purposes. But as we get closer to initial launch...well, I told anarcat about this project back in November, and we'll ask him to weigh in directly when it's time to ask that question :)

Sam Boyer wrote:

no, Aegir is solving a different problem. the need here is managing the creation/destruction of a virtual server resource in the first place, not what we do with it once its created. Aegir is about provisioning Drupal instances once the server resource is already known and available. It's a bit of a mix of configuration management + drupal configuration wrapped into one.

In any case, separate problem, and not one that helps us much since we need to set up additional services (e.g., PuSH hubs) as part of the overall FGA stack. And the context of this discussion is one where we're creating/destroying these server resources as part of the testing process - and is largely independent of our strategy for hosting production instances.

So unfortunately, unlikely that it'd be useful for any of our purposes. But as we get closer to initial launch...well, I told anarcat about this project back in November, and we'll ask him to weigh in directly when it's time to ask that question :)

Ok I got it.
Something like the Gandi.net API (http://doc.rpc.gandi.net/) perhaps or should it be also independent from a hosting company?
Could the system creating/destructing those server be plug able, a kind of abstraction layer?

yep, gandi is precisely one such service we could use, and via that API. there's a number of others, too.

ideally it would be pluggable, yes (and i believe our current thinking around using knife to do it means it WILL be largely pluggable), but my judgment call right now for FGA's purposes is that we focus on making it work with just one virtualization service.