epic: GC improvements

[mjudeikis]

Demo Objective

We want to demonstrate a larger kcp instance where each workspace uses less memory.

Goal

Reduce the per-workspace memory footprint in the current kcp implementation so that memory usage scales more gracefully with the number of workspaces.

This epic is scoped to garbage collection (GC) behavior and explicitly does not change cross-workspace ownership semantics.

Background

Today, the GC implementation creates multiple goroutines per workspace, so memory usage grows linearly with the number of workspaces.
GC also does not support cross-workspace ownership because ownership is currently restricted to a single workspace.

Objectives

• Make GC memory usage scale better with the number of workspaces (e.g., by reducing per-workspace goroutines and related overhead).
• Preserve existing ownership semantics for now; cross-workspace ownership will be handled in a separate effort.

Tasks

• Reuse CEL machinery: kcp-dev/kubernetes@4b3495f ✅
• Rework GC and quota patches to reduce per-workspace overhead:
  kcp-dev/kubernetes@c4290e3

Demo Steps

No response

Action Items

• Scope of the current demo is necessary to fit in the prototype boundaries
• Contribute to the final demo script and recording

Stories

• Reuse cel machinery kcp-dev/kubernetes@4b3495f ✅
• Rework GC and quota patches kcp-dev/kubernetes@c4290e3
• Add an initial implementation of a kcp native garbage collector #3768

[mjudeikis]

@ntnn

[ntnn]

GC test improvements: #4015