On Wed, Jul 17, 2019 at 4:31 PM Jeremy Cline <jeremy@jcline.org> wrote:
>
> On Wed, Jul 17, 2019 at 08:33:02AM -0400, Neal Gompa wrote:
> > On Wed, Jul 17, 2019 at 6:46 AM Pierre-Yves Chibon <pingou@pingoured.fr> wrote:
> > >
> > > Good Morning,
> > >
> > > We posted this [1] blog today and want to open a mailing thread to garner
> > > feedback, field questions and get some thoughts from the Community on
> > > the approach that we in Community Platform Engineering (CPE) are taking.
> > >
> > > [1] https://communityblog.fedoraproject.org/application-service-categories-and-community-handoff/
> > >
> >
> > Two things that concern me at this time:
>
> <snip>
>
> >
> > > Ipsilon — Ipsilon is our identity provider. It supports multiple
> > > authentication protocol (OpenID 2.0, OpenID Connect, SAML 2.0, …)
> > > and multiple backends (FAS, LDAP/FreeIPA, htpasswd, system accounts…).
> > > While it was originally shipped as a tech preview in RHEL it no longer is and
> > > the team working on this application has also been refocused on other projects.
> > > We would like to move all our applications to use OpenID Connect or SAML 2.0
> > > (instead of OpenID 2.0 with (custom) extensions) and replace FAS with an
> > > IPA-based solution, which in turn allows us to replace ipsilon by a more
> > > maintained solution, likely Red Hat Single Sign On. The dependencies
> > > are making this a long term effort. We will need to announce to the community
> > > that this means we will shut down the public OpenID 2.0 endpoints,
> > > which means that any community services that use this protocol need
> > > to be moved to OpenID Connect as well.
> >
> > There are two issues to unpack here:
> >
> > 1. We use a weird custom backend and custom protocol extensions.
> >
> > This should definitely be replaced if it makes sense. It’s more urgent
> > now that RHEL 6 is going EOL next year, and FAS 2 is still a Python
> > 2.6 application. FAS 3 *would* have fixed it, but interest by the FAS
> > developers died a while ago…
> >
> > Naturally, the replacement is equally in a poor state, but may have
> > some legs someday: https://github.com/fedora-infra/noggin
> >
> > 2. Ipsilon development was only considered important as part of being
> > tech preview in RHEL and now it’s not.
> >
> > There are some major problems here. First of all, Ipsilon development
> > has been gated by a single person. That person also seems to have
> > trouble making time to review pull requests. There has been interest
> > from the broader community about using and contributing to Ipsilon,
> > since unlike Keycloak, it is written in an accessible language
> > (Python).
> >
> > Getting Ipsilon to Python 3 would be enough for me to get started on
> > bootstrapping some of the other interested parties onto Ipsilon, and
> > hopefully give us a more sustainable community long-term.
>
> I guess my question to all this is... Why? What's the goal? If Keycloak
> does everything Ipsilon does and more, what's the point of keeping a
> dead project alive instead of contributing to the active, lively one?
>
> If there really, truly is interest from the broader community, why not
> do a friendly fork, get all the work you want in, and see what the
> original maintainer thinks?
>
Keycloak is not generally Fedora contributor friendly. Aside from it
being written in Java (which is problematic with the Java stack in
Fedora slowly imploding...), Keycloak is a lot less flexible and a lot
more tied to aspects of RHEL/Fedora that make it annoying to use in
other environments.
At least with Ipsilon, the Python codebase makes it easy for a broad
base of contributors to hack on the code. It's also much easier to set
up than Keycloak and easier to plug into more environments.
The biggest issue with Ipsilon as a project is the lack of awareness
of its existence. That has allowed the fact that the maintainer hasn't
recently been able to focus on it in a while to be unnoticed. Now that
is changing, and this is a problem, as I outline later...
> For Fedora, though, if FreeIPA can replace FAS, or GitLab can replace
> Pagure, or a generic notification service exists somewhere to replace
> FMN, or whatever, why spend time on such things we could be spending
> developing the few unique tools we need to continue building the Fedora
> distribution? Stopping along the way to build an identity and access
> management platform isn't going to make the distribution better.
>
FreeIPA is a perfectly fine solution, since it's broadly available and
easy to deploy. The non-Java parts (everything but Dogtag) is quite
easy to understand and hack on. Reproducing the tooling and
environment is easy as well. I have no problems with it other than
FAS-specific functions still need implementing on top of FreeIPA
(which in theory, Noggin will do). It's more complicated than I'd like
in some ways, but at least setup is replicable, making it easy to do
development and contribute.
As for GitLab vs Pagure, my reasoning is more or less than same as
mine with Keycloak vs Ipsilon. And there *are* users other than us for
both Pagure and Ipsilon.
The difference between Ipsilon and Pagure, in my view, is the primary
maintainer. Pierre-Yves Chibon has been very accessible and a pleasure
to work with as part of contributing to Pagure. I have virtually no
experience interacting with Patrick Uiterwijk because he's barely
around.
We have a number of projects that we're basically said only the
"security guy" can work on (i.e. Patrick). That's more of a problem
than anything else the CPE team does. The bus factor
The bus factor is a driver for us reshaping our ownership of certain apps. 100+ services each owned by an individual who in turn owns several is not sustainable. We want coverage on the apps that are critical to survive events like PTO, someone leaving the company, someone leaving the project or indeed the bus factor.
of *one* security
person is ridiculous, and the fact we're stretching him across both
CentOS and Fedora now means he doesn't pay enough attention to either
project. If we're going to have this (frankly insane) requirement that
we need to block all this on a security person, we need more than one.
Full stop.
I don't have the history here but I would encourage you to open a thread and let's discuss it. How can we improve it? What tooling or automation could we bring in? How can we get more community involvement? Let's bring solutions forward and see what can be done to improve this.