On 1/7/19 12:41 PM, Matthew Miller wrote:
On Mon, Jan 07, 2019 at 09:26:25AM -0800, Kevin Fenzi wrote:
> I am not sure if this really follows from the goal above.
> Sure, low level infrastructure takes some time, but I don't think it's
> all that much, and also some of it's unavoidable.
As I understand it, there was just a multi-month, multi-person project to
update OpenStack. That's the kind of thing I'm talking about.
Yes. I'd be 1100% fine with never dealing with openstack again, but I
think it's largely the exception/outlyer here.
> So, what do you mean running non free tools "in Fedora's
> here? As noted above we always said upstreams for applications or the
> like can do as the people doing the work decide, but does this mean we
> should (for example) consider running oracle db if we feel it would be
> better for our database loads? Can you clarify here? We don't currently
> run any non free software actually in infrastructure.
Hmmm; I'm not super sure about that exact wording in re-reading it. I think
"infrastructure" there is to be read very broadly. We meant things like:
hosting VMs in a cloud service where the cloud service stack might not be
open source all the way to bare metal (GCE, Azure, AWS). We'd want to make
sure we aren't locked in (for example, by using OpenShift on top of that).
This certainly wasn't meant to be an instruction to go look for proprietary
replacements for infrastructure. And "we feel it would be better" is
certainly not the same as "not viable or not available".
ok. That makes me feel better. :)
You really can't be 100% open source and provide network services (look
at all those non free routers handling your packets, etc). I just think
we need to make sure and use as much open source as we can and avoid
building on something we can't get out of.
Although we aren't in this situation now, one thing I can imagine is back
before Amazon released open-licensed versions of their tool to upload
images. I think it probably would have been okay for Fedora to use that
proprietary tool in order to make Fedora Cloud available to people.
>> #action contyk, FESCo to work with Infra to examine current
>> applications and determine: 1. which applications can be moved out of
>> the datacenter immediately or in the short term,
> Moved out of the datacenter to... where? Cloud instances?
> Hosted solutions? Based on what critera?
Any of those things. I think in general, if there's an open source hosted
version of something (like Taiga) that should absolutely be the default.
Otherwise, yes, cloud instances — or better, running containers in a
I'm not sure how much control that gives up to do things that way, but I
guess we can investigate it.
> This is going to cause a lot of short term work for... well,
> Especially in the case of things we don't spend much time on now, since
> everyone will have to adjust their workflow and we will have to spend
> time migrating things.
I recognize that this is non-trivial, and it's possible that it's easier to
just let some things sit where they are. But a lot of those normally
low-maintenance things are tech-debt time bombs. Or, they're really
Sure, but I am objecting to the "move it all now!"
How about "move it all when it makes sense?"
>> 2. Which applications
>> have industry-standard open source or proprietary alternatives that we
>> could move to.
> I'm not coming up with too long a list here. Most of our apps are highly
> specialised for distro making, but:
> * fedocal (which we pretty much spend no time on)
Yeah, this'd be a good candidate. It was a great, useful tool for its time
but is rather dated. It could use a *lot* of development work, and
realistically, we all know it's not going to get it.
> * pagure (which we want to keep? or no?)
I think we're pretty committed to pagure for src.fpo. But for general git
hosting... I think it's a great project — but I also see that Debian, GNOME,
Linux Kernel, are evaluating it and picking GitLab. Meanwhile GitLab (and
GitHub) are growing new useful features rapidly and there's no way we can
invest development to keep up. I don't think anything like "kill pagure!"
but I also don't think that Fedora *as a project* can afford to invest
> * nagios (pagerduty?)
I had a great conversation with Jef Spatela who is now at Sensu at
AnsibleFest. Maybe Sensu. :)
well, my tenative plan was to just look at configuring the Prometheus
that comes with openshift to handle those things in openshift and then
move as much as makes sense there... but we could look at it.
> * magazine /communityblog ( wordpress.org? but there's very
> spent on these too)
All of the "very little times" do add up.
I suppose, but the migration of them is much more time.
> * mailing lists (discourse?)
Or a more traditional mailing list provider — maybe ponee.io?
Never heard of it, but ok.
I think this is a place where moving will also be pretty disruptive,
especially if people need to change filters, but perhaps we could avoid
> * openstack (The big consumer here is copr... if it could move to
> cloud or somewhere and we didn't need a openstack I would be very happy).
Yes. I think this is something we can make happen. Possibly we can work
something out with Mass Open Cloud.
Well, the sticking point is the storage. It needs a lot.
> I think things like koji are particularly ill suited to move
> since we need to run Fedora on the builders, they have to run virt, and
> we have lots of arches that aren't commonly available in cloud.
Yeah, multi-arch is going to be a sticking point for a lot of things.
Yep. Also things that are specialized to our needs and not generic
enough to have anyone who deals with them.