From Fedora Project Wiki
Warning.png
This is a draft document!
This document is a work-in-progress draft and has not been agreed upon yet.

Discussion

Draft Use Cases

  1. Users of Fedora server will be able to install - at their option - a select set of software with graphical interfaces, and they will be able to successfully use these graphical interfaces via trusted X-forwarding (ssh -Y). [1]
  2. Provide a platform for acting as a node in an OpenStack rack.
  3. Provide a platform and simple setup for certain infrastructure services, e.g.
    • FreeIPA Domain Controller
    • BIND DNS
    • DHCP
    • Database server (both free and commercial).
      • (Should Fedora care about proprietary ones? We shouldn't be hostile, but neither cater for them too much, most of the contributors wouldn't have the means to do it anyway, and proprietary vendors can better install on RHEL and interact with RH. --simo)
    • Samba Ad Domain Controller
      • (that is mostly blocked due to upstream work, and I am not sure we can do much here, beyond encouraging upstream. --simo)
  4. Provide simple setup of a file-server (on par with Windows).
  5. Platform for deploying web applications with high-value frameworks.
    • Frameworks:
      • Ruby on Rails
      • Django
      • Turbogears
      • Node.js
    • (Are we going to do this via Software Collections? The main issue I see with these is that various apps requires at times, conflicting versions of the same base framework/language. --simo )
  6. Make Fedora the best platform to deploy JBoss applications.
  7. Come up with standardized mechanisms for centralized monitoring
  8. Come up with standardized mechanisms for centralized configuration and management
  9. Simple enrollment into FreeIPA and Active Directory domains
  10. Provide the best platform for secure application deployment
  11. Isolation of OS from applications
  12. Isolation of applications from each other
  13. Isolation of application users from each other
  14. Management of application resource consumption
  15. Simplify management and deployment.[1]
  16. Deliver the world's best leading edge DevOps platform.
  17. Container host management, including at scale (10k containers per host with container hibernation and migration capability)
  18. Working with upstream and packagers to have first-class systemd support, including native units and outsourcing privilege dropping, logging, and socket listening to systemd. Maximize use of isolation for capabilities, privileges, and namespaces. Continually raise the bar for uniformity of configuration and management tools.
  19. No packaged config shipping to /etc (system and services should use defaults with empty /etc)

Questions / Discussion Points

Items under "Provide the best platform for secure application deployment

are cgroups / containers meant here?

Difference between server and devops?

Difference between Server and Cloud products vs compute nodes?

  • "Traditional servers [like pets] have names, personalities, and are lovingly cared for. When they get sick, you diagnose the problem and carefully nurse them back to health, sometimes at great expense. Like pets. On the other hand, cattle sre numbered, and thought of as basically identical, and if they get sick, you put them down and get another one." --mattdm
  • "This was my thinking, I thought the cloud WG is about building lean images that do not have much in the way of wizards, installers, and hand-holding software, but are focused toward rapid deployment, either as lean hosts or as computing/elastic guests, all controlled via some configuration engine like puppet etc... basically single task machines. I see the server WG more about building heavier, long term, multipurpose servers (be it on bare metal or as guest)." --simo
  • "I think it would make sense for the Base Design and Environments/Stacks WGs to define how to install, build and package the runtimes and the applications that use them (including the conflicting/multiple versions issue), without focusing on a specific use (e.g. treating web applications, GUI applications and CLI applications equally), and for the Server WG to handle deploying the web applications within the web server and managing deployed web applications." --mitr
  • "I would say single purpose servers/vm/containers all fall under the server WG as well but I think we should be looking at this from application stand point as in which of those services/daemon fall under the server WG and that means we could make up to about 500 550 applications or "products" that can be deployed on bare metal in vm's or containers we would be delivering." --johannbg
  • "I think it would make most sense for Cloud and Server to "share applications", i.e. the same application package can be deployed either within a single-purpose Cloud image (automatically managed for horizontal scaling), or as a single instance within a Server (one of many applications running on this particular Server). Given that, I think the Server WG should indeed choose a very limited set of "applications" / "services" to include within the Server product and to make management of this limited set of services really good." --mitr

Balance between OS packages and more rapidly moving frameworks in collections...

Visibility of packages

  • "Would it be radical to suggest that "packages" should be invisible to an admin that doesn't want to see them? "Enable the DNS server", configure what it is serving, *product's magic here*, the DNS server runs." --mitr

Overlap between other working groups

  • "Basically where I stand any application that runs daemon/service as in it's an application (or set of applications) that runs in the background waiting to be used, or carrying out essential tasks on an pyshical/vm or in container or in other words basically it's an systemd/upstart/sysv unit/service or an container that can be started and enable with the systemd and service commands and is not part of the desktop/graphical target (such as gdm.service which thus makes it part of the workstation group ), as well is not part of the base/coreOS ( like device mapper etc ) it belongs within the server WG." --johannbg
  • "I tentatively agree, although I guess there may be desktop-oriented daemons we may not care about. Say a desktop-oriented backup daemon, that is sort of single user or anyway ill-suited for a multi-user server. Also should we care much about Graphical UIs ? Or should we freeze early and maintain whatever version was considered stable in the Desktop WG at the start of the cycle ? And who is going to maintain it if we do so and happen to have a longer term cycle than the desktop ?" --simo
  • "I agree that there will be situations where an administrator will want to install a GUI on a server (even if it's just because they have one machine in a rack that they use to fix things up when things go sideways)." --sgallagh

Headless only or support GUI?

  • "We expect headless operation to be the norm, and if graphical interaction is needed, it will usually be done remotely via another system, possibly Fedora Workstation. However, we should also keep in mind the high likelihood that people are going to want to administer these systems from Windows, Macintosh or possibly tablet devices. With this in mind, I'd like to suggest that we focus our efforts around web-based applications and scriptable shell commands. (So things like Katello/Foreman, OpenLMI, FreeIPA WebUI and similar). These should all be consumable from any graphical client." --sgallagh
  • "Not only that, there are some applications that although they are 'server apps' they require a GUI even if just to install them. stupid example but IIRC I remember things like game servers install programs that would run only with the Graphical UI. I also remember proprietary packages that would have a server component that would need X to install. Yes stupid things, but you can't discount them. I am also thinking that installing openoffice in order to do server-side rendering (doc conversions and so on) is going to drag in at least X libraries." --simo

Applications vs. Platform

  • "Our transitioning process needs to be able to cover 500+ applications ( or in other words be as generic as possible ) or so, so it obviously cannot depend on the existence of web fronted otherwise we would be excluding 99% of those server applications." --johannbg
  • "... but the Server shouldn't ship 500 "services" as an integrated part of the product (are there even that many services to provide?). Regarding the "competing products", I'd go as far as to say that the Server should give the users a "good LDAP server" without exposing which upstream project is internally providing the functionality - even possibly switching the upstream projects on an upgrade if one of them started to fall behind." --mitr
  • "I'd like for us to be focusing on a *platform* and a set of standard, visible APIs and working with the Base Design and Environments/Stacks groups to have service packages treated similarly to "apps" in other operating systems. We ourselves don't necessarily need to do all of the porting to accommodate this (though we will probably want to select a group of high-value servers that we use as examples, such as Apache HTTPD and BIND)." --sgallagh
  • "I generally agree - though I'd focus more on getting the "high-value servers" working well than on calling ourselves a "platform" - it's far too easy to make a platform that doesn't "work" without noticing when there are no major users of the platform." --mitr

Versioning

  • "Also, I'd like for us to try to manage this separation so that we can allow our consumers to pick and choose which server they actually want, rather than necessarily the freshest upstream bits. Those fresh copies *must* be available (and probably the default if not otherwise specified), but it would be REALLY nice to be able to hang onto MyServer 2.4 after 3.0 comes out if your other applications aren't ready for it." -sgallagh
  • "For the "high-value" servers (which provide external functionality, not an application API), I strongly disagree. We should be managing the transitions (no functionality dropped, all configuration migrated) so that the user will never want to use the older version. For APIs / runtimes, yes, we have no choice but to provide the older versions when the upstreams make ABI-incompatible changes." --mitr

Configuration Tools

  • "The latest generation, redhat-config-*, were, IIRC, written over a comparatively short period of time (ISTR they all happened within a year!), and covered basically all of the major server functions at the time - networking, httpd, bind, mail, ...) This has been done in the past, and this could be done again, if we really tried." --mitr
  • "Yes; a good UI is a part of the deployment/reliability story. The "desktop application" for managing a server is really out of scope for the Workstation WG as proposed. The management interface and the underlying API would be much better implemented by the same group." --mitr
  • "I wasn't either and I equally don't know, but I can tell you why they weren't useful to me at the time:
    1. no handling of multiple machines (or even remote connections to a single machine)
    2. little cohesive ux design (this did get iteratively better)
    3. some of them didn't work very well (*cough* samba)
    4. generally, you had to commit to using them and never touching the config files by hand
    5. I was operating in an environment with Red Hat Linux, Debian, Other-Linux-Flavor-Of-The-Day, BSDI, NetBSD, netbh Solaris, SunOS, IRIX, Tru64, and, um, VMS. All the various single-vendor GUIs just brought more pain.
    • I would certainly add "no API" to the list of complaints _now_, but I'm not ashamed to admit that that wasn't on list of sysadmin concerns a decade ago." --mattdm
  • "They were abandoned, more or less:
    1. because the hard core admins didn't use them anyway; they would either edit the configs by hand (old days) or just push out their configs with puppet/chef/ansible/cfengine/salt (these days)
    2. becuase the less hard core admins had enough other issues that this wasn't going to win them over
    3. because we were unable to create a larger upstream community that allowed us to drive development forward
    4. because chasing all the options that could be configured for something like this is actually somewhat significant work
    5. because there wasn't any encapsulation for common automation - it was just separate 'click here; do this' sort of tools" --notting
  • "I think we should, over time, move towards a "G"UI (whether local or web is an implementation detail in this) for one-time use, and an actual API used by an actual, current-era, programming language, for automated use. The CLI will obviously stay, both because many users are comfortable with it, and because we can't replace it during this decade." --mitr