Description
A Linux desktop's sole job is to connect users to their applications.
There used to be a time where one application was a process. 'ps' would show one easy to manage entry. Now Discord in a flatpak is 13(!) procesess. Have two electron instances and a system-monitor is useless. We have backgrounds services that linger with indecipherable names. It doesn't scale anymore on the desktop. We need metadata.
This is a solved problem with new kernel features. CGroups are used commonly for your system services. Any sysadmin won't deal with an Apache process, but the whole webserver as one cgroup. By embracing this not only do we have metadata providing a single-truth of the launched context, but also expose all of the kernel features reserved for cgroups. Separate network namespaces, hard CPU quotas, Out-of-memory protection that works, we can expose this either as transient modifiers or as sysadmin overrides.
We've been tackling this on both the GNOME and KDE side. In this talk we come together to talk about what we have each done on our respective sides. Where we've found common ground to common problems, as well as our plans for where we take the linux application ecosystem in the future.