Hey all,

I've heard from a number of different sources that trying to figure
out the LXD side of our plan has been a major decelerator. None of us
have built custom LXD images before, and figuring out the networking
configuration for that has also been probing unknown waters.

With that in mind, I'd like to propose a pivot in our plan:

- Let's focus on writing systemd service files for the services that
we want to run. These service files should load their settings from
config files.
- We can then build some mage-based automation to generate the config
files for a particular host (imagine `mage configure
aspen.arbor.chat`, which emits a relay config file, a webserver config
file, and whatever else we need).
- We can install all of our service definitions and binaries on each
server, then use mage to configure the host as needed to perform its
particular function.

I think that this is achievable in a much shorter timeframe, and still
gives us some nice idempotence if we ensure that all of the changes to
the host take the form of service files or config files that we
provide via automation.

Once we have a system like this set up, we can look back at the LXD
side of things. This is an incremental step that will help us bring
some services online and gain familiarity with configuring those
services without also going through the LXD learning curve.

What do you all think? This is just a proposal, not a decision. I'm
just throwing this plan out as a starting point for discussion.

Remember, when responding, use "reply all" to ensure everyone gets a
copy of your message. Also ensure you write in plain text mode so that
srht doesn't bounce your email.

Cheers,
Chris

I agree. I think using a systemd service would be desirable, especially as a first pass. LXD looks promising and I’d certainly like to revisit it, but it is a barrier to knowing what the desired state will look like, and that makes things more difficult starting off. 

Cheers,

Andrew Thorp

It's more important to get _something_ over nothing. We can revisit
the topic at a later date or when we find we actually need some of the
features that LXD provides. Thanks for trying to work through the LXD
stuff!

J

On Tue, Feb 2, 2021 at 10:33 PM Andrew Thorp <andrew.thorp.dev@gmail.com> wrote:
>
> I agree. I think using a systemd service would be desirable, especially as a first pass. LXD looks promising and I’d certainly like to revisit it, but it is a barrier to knowing what the desired state will look like, and that makes things more difficult starting off.
>
> Cheers,
>
> Andrew Thorp
>


-- 
Thank you,
Josh Whetton

Y'all do whatever you think is best, just tell me what I need to do to
support it. I finally have some motivation and time tonight so
arbor-infra ought to get a push or two.

Danny

Josh Whetton wrote:
> It's more important to get _something_ over nothing. We can revisit
> the topic at a later date or when we find we actually need some of the
> features that LXD provides. Thanks for trying to work through the LXD
> stuff!
>
> J
>
> On Tue, Feb 2, 2021 at 10:33 PM Andrew Thorp <andrew.thorp.dev@gmail.com> wrote:
>> I agree. I think using a systemd service would be desirable, especially as a first pass. LXD looks promising and I’d certainly like to revisit it, but it is a barrier to knowing what the desired state will look like, and that makes things more difficult starting off.
>>
>> Cheers,
>>
>> Andrew Thorp
>>
>

On Fri, Feb 5, 2021 at 8:05 PM Daniel Wilkins <tekk@linuxmail.org> wrote:
>
> Y'all do whatever you think is best, just tell me what I need to do to
> support it. I finally have some motivation and time tonight so
> arbor-infra ought to get a push or two.

At this stage, there isn't a lot you'd need to do. If you felt
inspired, you could build out some mage targets for installing systemd
units. That's about it.