What it is
A cgroup v2 directory exposes the counters the kernel keeps for a container: cpu.stat counts bandwidth periods and throttled periods, memory.stat splits memory into anonymous, page cache and kernel parts, memory.events counts limit hits and OOM kills, pids.events counts refused forks, and the pressure files report how long tasks were stalled. This tool reads a paste of those files and explains them; with a second sample it turns cumulative counters into rates.
When to use it
Use it when a container is slow although its average CPU is low, when it is OOM-killed although its RSS is small, when forks fail, or when you need to turn a pile of cat output from a node into a diagnosis.
Gotchas & limits
Thresholds such as 5% throttled or 90% memory are rules of thumb. memory.current includes reclaimable page cache; the working set (memory.current minus inactive_file) is closer to what leads to an OOM kill. nr_throttled is counted per scheduler queue. A single sample only gives totals since the cgroup was created.