Resource management

cgroups are the kernel primitive for bounding and accounting a group of processes' resource consumption — CPU, memory, IO, PIDs, and more. Each cgroup is a node in a hierarchical pseudo-filesystem; access to set limits and attach processes is gated by the standard KACS DACL on each node.

Creating cgroups, setting limits, attaching processes, delegating subtrees, reading counters and pressure metrics, and diagnosing access denials.

Exhaustive per-controller listing of every cgroup knob shipped on Peios — control files, counter files, and per-controller event interfaces. Covers cpu, memory, io, pids, cpuset, freezer, hugetlb, device, rdma, and misc.

Pressure Stall Information (PSI) is the kernel's metric for how much time tasks spent stalled waiting for CPU, memory, IO, or IRQ resources. PSI is exposed both per-cgroup and system-wide, and supports threshold-triggered notifications for autoscaling and oncall alerting.

How cgroups appear to unmodified Linux applications — v2-only paths, the device controller's composition with KACS, the dropped v1 controllers, and the file-mode-vs-DACL transition.