Node settings

A node's tabs and its Settings tab: capacity, caps, dedicated CPUs, network rules, traffic, processor, disks, new servers and NAT.

Each node's page has the tabs Overview, Servers, Settings, Port forwards, Domains, Storage and Maintenance. This page covers the Settings tab and the panel-wide defaults behind it.

Tab What it holds
Overview usage, hardware and software, the Agent, Certificate and Node firewall panels, IP addresses, images and ISOs on the node, private networks, recent tasks
Servers bars of memory, vCPUs and storage sold and free IPv4, and the node's servers
Settings capacity, caps and every node setting (this page)
Port forwards, Domains NAT forwards and websites by domain on this node (see NAT)
Storage the node's storages and disks (see Storage)
Maintenance no new servers, maintenance mode, emptying the node (see Maintenance)

Panel defaults and node values#

Every setting has a panel-wide default in System › Settings › Node defaults. A node inherits it or sets its own value. For a few settings a panel plan, or an administrator on one server, can override the node.

The most specific value wins: server, then plan, then node, then panel default. A node that inherits follows the default when you change it later.

On the node's Settings tab each field's first choice is Panel default (…), with the default's current value. A field set on the node says Set on this node and offers Use panel default. Under each field the page says when a change takes effect: at once; for new servers only; or at each server's next power off and start. Reset all to panel defaults clears every value of the node.

First apply on an existing node#

A node that has never been sent its settings runs as before, with none of the panel's network rules. Its Settings tab says Not applied yet and shows Apply panel defaults. The confirmation lists the network rules that come on and how many servers lose nested virtualization at their next power off and start. Nothing changes on the node until you apply. After that, the panel keeps the node in step and sends it again whenever it differs.

On a node whose agent is too old for a setting, the field says which agent version it needs; values are saved and sent once the agent is updated.

Capacity#

What may be sold sets the node's overselling ratios:

Field Default Meaning
Memory that may be sold (%) 100 of the memory left after the reserve; above 100 oversells it
Memory kept for the node (GB) 4 never sold: the node's own system, the agent, QEMU's overhead
vCPUs per CPU thread 4 total vCPUs sold may be this many times the node's threads; a server never gets more vCPUs than the node has threads
Server storage that may be sold (%) 100 disks take space as they fill; above 100 oversells it

Click Save capacity. Storages have their own sell percent on the Storage tab. How placement uses these figures is in Locations, node groups and capacity.

Dedicated CPUs#

CPUs for dedicated servers (for example 8-15,24-31) reserves host CPUs for servers of plans with dedicated CPU: each such server gets these CPUs, one per vCPU, and every other server runs on the rest. Empty: the node takes no dedicated servers. Leave at least one CPU, ideally a whole core, for the node itself. The section shows the node's CPU topology and NUMA nodes and which servers hold which CPUs.

Network rules#

These rules run in the node's own nftables tables on the servers' traffic. None of them touches the node's own traffic (SSH, the agent port).

Setting Default Levels What it does
ARP and neighbour discovery flood protection on panel, node drops ARP and IPv6 neighbour discovery frames past 50 a second from each server
Block DHCP servers on servers on panel, node drops DHCP and DHCPv6 offers a server sends
Block the cloud-init (metadata) address on panel, node drops traffic from servers to 169.254.169.254 and fd00:ec2::254
Block outbound SMTP (port 25) off panel, node, plan, server drops servers' connections to port 25; 465 and 587 stay open
Block inbound ping (ICMP echo) off panel, node, plan, server drops ping requests to servers; their own pings out keep working
Block inbound RPC portmapper (port 111) off panel, node drops TCP and UDP to port 111 of servers (off because servers serving NFS need it)

Network rules take effect at once. Spoof protection is always on and has no setting: a server may only send from the addresses the panel gave it. An administrator can allow extra source prefixes per server on the server's Manage tab. The section says how many servers on the node set their own SMTP and ping values.

Traffic#

Traffic counted towards the allowance: In and out (default), In only or Out only. Also settable per plan. A change counts the current month again from its first day, so servers may pass their allowance, or come back under it, at once; the page asks you to confirm. Closed months keep what they counted. See Traffic.

Caps#

Hard limits on top of the ratios, used by placement for new servers and upgrades:

Setting Meaning
Most servers servers in any state but deleted
Most memory sold on top of the memory ratio: the smaller applies
Most vCPUs sold on top of the vCPU ratio
Most disk sold on top of the storage ratio

On the node, empty inherits the panel default and 0 means no cap on this node. A node beyond a cap shows Over its cap.

Processor#

Setting Default Levels Takes effect
Nested virtualization (vmx/svm) Off panel, node, plan, server next power off and start
AMD SMT topology (topoext) as the host's CPU model panel, node, plan, server next power off and start
CPU model host model panel, node, plan, server next power off and start
CPU name shown to customers empty panel, node at once
  • Nested virtualization is off by default because of CVE-2026-53359, a KVM guest-to-host escape through nested virtualization. Turn it on only for plans that sell it, on patched nodes. The Security box lists the servers that still expose it, with Restart now.
  • AMD SMT topology is shown on AMD nodes only.
  • CPU model: host model (the host's CPU as a stable model), host passthrough (exactly the host's CPU, fastest, but the server cannot move to another CPU later), or a named model the node lists, which is the same on every node that has it. Windows activation and licensed software may notice a new CPU.
  • CPU name shown to customers: what the server's Overview calls its processor. Empty: the CPU the node reports, or "Virtual CPU" and the model's name.

Disks#

Setting Default Takes effect
Disk format of new servers qcow2 new servers only
Disk cache QEMU's default next power off and start
Disk IO driver QEMU's default next power off and start
An I/O thread for each VirtIO disk off next power off and start
  • Format: qcow2, or raw for I/O-heavy servers (no qcow2 metadata on the I/O path). Snapshots and backups work with both. Existing disks never change, a reinstall keeps the format, and a server's extra disks take the format of its first disk.
  • Cache and IO driver apply to every disk without its own value (set per disk on the server's Storage tab). The IO driver native needs the cache none or directsync; the panel refuses any other pair. unsafe ignores flushes: data written may be lost when the node fails.

New servers#

Setting Default Meaning
Machine type q35 (PCI Express) or pc (i440FX) for old systems without q35 support; existing servers keep theirs
Give every new server IPv6 off a new server whose plan gives no IPv6 gets one from the node's IPv6 blocks

NAT#

Setting Default Meaning
Public port range 1024-65535 the node's ports customers may forward to their NAT servers
Reserved ports none ports and ranges no forward may take, such as 3306, 5432, 6000-6100 (up to 50 entries)

Both apply to new port forwards; existing ones are kept. See NAT and port forwards.

Click Save settings at the end of the form. Node changes are audited as node.settings_changed; a node that refuses its settings is audited as node.settings_failed and the tab shows what the node reported.