Resource Abuse

Stop one VPS slowing down the whole node

01 · Measure

Read from the node, never from inside the VPS

CPU as a share of each VPS's own vCPUs, so a 2-core VPS at 100% and a 32-core VPS at 100% read the same.
Disk throughput in MB/s and disk operations per second, read and write together.
Memory in use, shown beside the rest. Never alerted on: a healthy Linux VPS fills its memory with cache.
Read every 10 seconds from each VPS's control group on KVM (Proxmox, VirtFusion, Virtualizor, SolusVM, plain libvirt) and LXC.
Xen and OpenVZ keep their VPSes outside the node's control groups, so Frabs says they are not measured instead of guessing.

node-ams-14 · 32 vCPUs, 88% busy

185.231.66.1999%4 MB/sCPU Limited

185.231.66.4271%38 MB/s

185.231.67.812%410 MB/sWatching

185.231.66.1056%1 MB/s

02 · Judge

Detector	Low	Medium	High	Critical
Sustained High CPU% of its own vCPUs	80over 1 h	90over 1 h	95over 3 h	98over 6 h
Heavy Disk ThroughputMB/s read and written	75over 30 min	150over 30 min	300over 1 h	500over 3 h
Heavy Disk Operationsdisk operations/s	1.5kover 30 min	3kover 30 min	6kover 1 h	10kover 3 h

A build, an update, a reindex or a nightly backup runs flat out for minutes. Abuse runs for hours, so High needs three hours at 95% CPU.
Averaged over the whole window, so a VPS that started an hour ago cannot reach a three-hour level.

Backups, migrations, snapshots and clones run by Proxmox are the node's own work and never count against the customer.
When many VPSes spike together, Frabs holds every automatic limit: that is a host-wide event, not abuse.

03 · Act

A VPS is limited only when it slows others down: the node's CPUs at least 70% busy, or the VPS's disk busy at least 60% of the time.
On a node with room to spare, heavy use is reported to your team and never limited.
CPU Limit caps the VPS at half of its vCPUs; Disk Limit at 50 MB/s and 1,000 operations a second. Each lasts an hour and is renewed while the heavy use lasts.

Never below 10% of its vCPUs or 5 MB/s, so the VPS always stays usable for its owner.
Its network is never touched, and a resource finding never isolates or suspends a VPS.
Every limit is written to the node's own control group and put back exactly when it ends, or when the Frabs service stops.

04 · Control

Run Resource Abuse only at set hours, such as weekday evenings when your nodes are busiest.
Schedule a change for a set time: turn it on from Monday 09:00, or off for a planned maintenance window.