What makes Fedora desirable in clouds?

[Major Hayden]

Hey everyone,

While mowing the yard last weekend, something came to mind about Fedora on clouds. We
spend a lot of time thinking about getting Fedora onto more clouds and making it more
interoperable with cloud APIs. Increasing Fedora's usage by making it available in
more clouds seems like a worthwhile goal.

However, I wonder why people choose to use Fedora versus an alternative when they deploy
in public clouds. These questions came to mind:

* Is there something lacking in the Fedora experience?
* Is Fedora more difficult to use or does it have limitations that frustrate users?
* Are we missing docs and blog posts that help users deploy their favorite applications on
Fedora?

I'll admit that I bounce between regular Fedora and Fedora CoreOS in my own
deployments. A lot of that depends on what I plan to run there. If I'm running
containerized applications, CoreOS gets me up and running quickly. If I need to do some
development or run something a bit more complex, I usually reach for Fedora Cloud.

I would love to hear thoughts from others on this topic about how we can improve the
end-user experience for Fedora in public clouds of all sizes.

--
Major Hayden

[David Duncan]

>  While mowing the yard last weekend, something came to mind about
> Fedora on clouds. We
>  spend a lot of time thinking about getting Fedora onto more clouds and making it more
>  interoperable with cloud APIs. Increasing Fedora's usage by making it available in
>  more clouds seems like a worthwhile goal.
>   
This is a great path for your thoughts and I am so grateful that you _do_ think in this
direction when you are really relaxed into a routine. 

>  However, I wonder why people choose to use Fedora versus an
> alternative when they deploy
>  in public clouds. These questions came to mind:
>  
>  * Is there something lacking in the Fedora experience?
>  * Is Fedora more difficult to use or does it have limitations that frustrate users?
>  * Are we missing docs and blog posts that help users deploy their favorite applications
> on
>  Fedora?
>  
>  I'll admit that I bounce between regular Fedora and Fedora CoreOS in my own
>  deployments. A lot of that depends on what I plan to run there. If I'm running
>  containerized applications, CoreOS gets me up and running quickly. If I need to do some
>  development or run something a bit more complex, I usually reach for Fedora Cloud.
>   
I think that's great. They are both incredibly beneficial, but one doesn't answer
all of my problems. Like you, I feel incredibly comfortable using both FCOS and Fedora
Cloud Editions. 

>  I would love to hear thoughts from others on this topic about how we
> can improve the
>  end-user experience for Fedora in public clouds of all sizes. 
There are a lot of variations in workloads, but one of the things that I run into for user
workloads is an ability to predict how they are going to use the tools they have been
using previously in the context of their old process and still deliver big engineering
goals.  

One of the things that makes me excited is building supported models for specific tasks.
It's how I built out support for configurations including Nice-DCV on AWS which
ultimately became the proof of concept that cemented process for Amazon EC2 Accelerated
Instances on RHEL Workstation. CoreOS wasn't going to make it through the commercial
partner application security team review for the organization that I needed for a hand off
to the engineering team. It was the shortest path to complete the concept work and helps
me to build out the next generation support efforts.

Another thing that I enjoy working with the cloud images for is the consistency with the
Workstation experience from the perspective of the filesystem. Bringing btrfs onto the
cloud image could not happen without strong community control, but it delivers a lot of
flexibility in the configurations and a significantly different experience to the image in
terms of experimental configurations. It allows me to build, store, and restore in ways
that I could not do with a single experimental instance before. That is super helpful
building out new solutions too. 

Also, adding packages into the cloud-sig, so that we are working on an ecosystem for the
support model excites me. I don't care if they are beneficial for only CoreOS or
everywhere. I like the idea that we are building towares a shared core of utilities that
supports management of cloud resources and related configurations. Being able to support
the ecosystem is most important for me as a member of the team.

I think that everyone should be focused on application modernization, but not every
problem is going to call for a new solution. When I reach for the tool to get what I want,
I stick to the old adage I learned from my mentor Wayne Walker a long time ago, which is
this: "The right tool for the job is the tool that gets the job done while I still
have a job!" Sometimes, that's means I pull from what I know will work and that
can be different in different situations or it can be, like I mentioned above, to satisfy
external requirements that are not in my control while I still handle the task at hand.

[Miroslav Suchý]

Dne 15. 05. 23 v 16:28 Major Hayden napsal(a):
>  * Is there something lacking in the Fedora experience? 
I was recently forced to migrate my personal server from on-premise to cloud and I tried
several small EU VPS providers. 
The Fedora was not available there. Sometimes CentOS 7, rarely 8. Only Debian or Ubuntu.

>  * Are we missing docs and blog posts that help users deploy their
> favorite applications on Fedora? 
More documentation and examples please.

This

https://cloudinit.readthedocs.io/en/latest/howto/index.html

is better than I remember, but I know cloudinit can do more.

>  I would love to hear thoughts from others on this topic about how we
> can improve the end-user experience for Fedora in public clouds of all sizes. 
I spin-up VM, do the job and terminate it. While the cloudinit speed improved a lot, I
wonder why it is so serialized. 
Can it be turned to bunch of systemd services and run in paralell? Oh, wait. It is already
bunch of systemd services. 
Nice. But still, cloudinit takes about 10+ seconds. I think there is still place for
improvements. E.g. 
DataSourceEc2Local runs for 6 seconds on my machine.

Miroslav

[Benson Muite]

Very helpful for testing and development.  Can setup a clean, sometimes
relatively powerful environment with fast downloads from caches then
take it down once done.

On 5/16/23 08:48, Miroslav Suchý wrote:
>  Dne 15. 05. 23 v 16:28 Major Hayden napsal(a):
> > * Is there something lacking in the Fedora experience?
>  
>  I was recently forced to migrate my personal server from on-premise to
>  cloud and I tried several small EU VPS providers. The Fedora was not
>  available there. Sometimes CentOS 7, rarely 8. Only Debian or Ubuntu. 
Making this
easier would be great. CentOS stream is not popular for
smaller providers. Rocky and have Alma gained traction there.
>  
> > * Are we missing docs and blog posts that help users deploy their favorite
> applications on Fedora?
>  
>  More documentation and examples please.
>   
This would be helpful.  Documentation would also help for deploying on
smaller VPS providers.  Increasing number of availability zones in AWS
would be good as well, for example South Africa is missing.  Alibaba and
Oracle are other cloud providers for which it would be nice to have
images.  Happy to help with these.
>  This

[Smith, Stewart]

I can add a bit on why people end up picking Amazon Linux, and where these reasons
conflict with Fedora….

>  On May 15, 2023, at 10:29 AM, Major Hayden <major(a)mhtx.net&gt;
> wrote:
>  
>  While mowing the yard last weekend, something came to mind about Fedora on clouds. We
> spend a lot of time thinking about getting Fedora onto more clouds and making it more
> interoperable with cloud APIs. Increasing Fedora's usage by making it available in
> more clouds seems like a worthwhile goal.
>  
One thing we ask ourselves when looking at possible changes to Amazon Linux (including the
“base it on Fedora” one) is “How would this benefit our customers?”

It’s a good thing to think about - *why* do we want to increase Fedora usage if not for
solving problems for those who end up using Fedora?

>  However, I wonder why people choose to use Fedora versus an
> alternative when they deploy in public clouds. These questions came to mind:
>  
>  * Is there something lacking in the Fedora experience?
>  
We do hear from a lot of Amazon Linux customers that the Long Term Support is important to
them. This is, of course, something lacking in the Fedora experience. Perhaps it always
should? It can be in direct conflict with the idea of being first, and is certainly a
hefty investment.

For the Fedora experience on AWS, there are a few more AWS agents and SDKs packaged in
Amazon Linux than Fedora. This is on us (AWS) to fix, and ensure that software we vend is
easy to package for linux distributions and can thus easily get into the hands of all EC2
linux users. We over in the Amazon Linux space are quite motivated to make that happen as
it means there’s lots of time for the teams working on the upstream software in these
packages to ensure they adapt to the changes in the Free Software ecosystem that will make
their way into a future AL version. So us, as users of Fedora to make a downstream Linux
distribution, get incredible value from Fedora being First.

Other things we hear a lot of is a desire for a choice on language runtimes and major leaf
packages. While everyone WISHES they could get rid of older python versions, older php,
older MariaDB, PostgreSQL etc, having a choice and a grace period is seen as a feature.
For some packages in Fedora, they get this already.

Launch times is another one. When customers in a cloud need to scale up, they want to
launch new capacity FAST. There’s a few tricks we have in AL that should make their way
upstream :)

There’s also the security footprint - which comes down to the minimal number of packages
that are on a base image (AMI or container). Fedora packages do tend to enable a lot of
options and functionality, and this makes the base package set fairly large. For AL2023 we
were able to tweak (or add) a whole lot of bconds to help bring this down dramatically,
and we were looking heavily at “what could we do to alter the default images so that
customers would have fewer patching runs required”.

>  * Is Fedora more difficult to use or does it have limitations that
> frustrate users?
>  
We have found that ease of use in a cloud isn’t exactly what you’d expect when thinking of
ease of use for a more traditional server or desktop OS.

One thing we have ensured we started doing was releasing a new AMI for any security
updates. A lot of customers just want to patch with instance replacement, and use the
“there’s a new AMI available” as that trigger. It also means that container builds get
kicked off, as the base layer changes. Running arbitrary “dnf update” on launch or in
container recipes tends to lead to sadness.

Another one is information on what is fixed where, and having accurate and up to date
security patching information. This is, of course, a lot of work.

>  * Are we missing docs and blog posts that help users deploy their
> favorite applications on Fedora?
>  
One thing I’ve been thinking of as lacking is good guidance for ISVs packaging software.
They may need to bundle dependencies and other such things, may even not be open source,
but we still want them to be good citizens in the ecosystem and help there not be needless
complications.