Skip to content

Add a network maintenance guide for host machines - #939

Open
jacobmattern wants to merge 1 commit into
mainfrom
docs/host-network-maintenance
Open

Add a network maintenance guide for host machines#939
jacobmattern wants to merge 1 commit into
mainfrom
docs/host-network-maintenance

Conversation

@jacobmattern

Copy link
Copy Markdown
Contributor

New page host/network-maintenance.mdx, added to the Maintenance group and linked from the Network requirements in Verification Stages.

Why

Networking is the requirement hosts drift away from without anything appearing to break. A link renegotiates down after a cable gets bumped, a reboot moves the machine off the LAN address the router forwards to, a firmware update re-enters a forwarding rule over a shorter range, a machine gains GPUs and outgrows the range sized for it. In each case the machine stays online, the daemon stays connected, and it keeps taking rentals while quietly failing the requirement.

host/hosting-overview.mdx already names untested networking as the most common host mistake, and Vast does not offer support for getting a machine working, so the docs are the support.

What it covers

The routine check for that drift, plus the network changes that need a maintenance window. Three sections, so a host reads the general checks and then the one matching their setup:

  • General checks — line speed against 500/500 measured on the machine, negotiated link speed and duplex, interface error counters read as a rate against the previous window, public IPv4 and the NAT layers above it, LAN address pinning, port range sized against current GPU count, firewall and ephemeral port range, both ends of the range tested from outside, packet loss and DNS resolver redundancy.
  • Single-machine hosts — what drifts on the router: rules lost to firmware work, and the router itself as a bottleneck through NAT table and forwarding-entry limits.
  • Multi-machine hosts on one network — the capacity, switching and allocation problems that only exist once there is more than one machine and get worse silently as the fleet grows: aggregate uplink measured under load, the switch trunk as the real bottleneck, non-overlapping port range allocation, shared DHCP and conntrack limits.

Scope

Maintenance, not troubleshooting. Symptom-driven diagnosis — a machine renters cannot reach at all, CGNAT disqualification, ISP escalation — is deliberately out of scope and tracked separately. host/machine-offline.mdx remains the page for a machine that has dropped off the platform, and this page defers to it rather than restating it.

Nothing from the gated host setup guide is reproduced here; the setup-side half of resizing a range or picking up a changed address links out to it.

Checks

  • mint validate and mint broken-links both pass.
  • All 10 same-page anchors verified against a mint dev render, since mint broken-links does not check them (a deliberately bogus anchor passes it clean).
  • Commands verified against Ubuntu Server 22.04 and 24.04 behaviour, including that ufw ships installed but disabled, that netplan config is root-only on 24.04, and that nc sets SO_REUSEPORT so a listener can bind alongside an instance.

Note for review

host/verification-stages.mdx requires 5 forwarded ports per GPU; the host setup guide states at least 3 per GPU. Both agree on 100 recommended. This page is written against the verification bar and labels it as such, so it does not contradict either — but the two source documents disagree and one of them should be corrected.

Networking is the requirement hosts drift away from without anything
appearing to break: a link renegotiates down, a reboot moves the machine
off the address the router forwards to, a firmware update re-enters a rule
over a shorter range. The machine stays online and keeps taking rentals
while failing the requirement.

The page is the routine check for that drift, plus the network changes
that need a maintenance window. It is split three ways so a host reads the
general checks and then the section matching their setup: single machine
behind one router, or several machines sharing an uplink, where capacity,
switching and port allocation problems only appear once the fleet grows.

Symptom-driven diagnosis stays out of scope. Offline Machine covers a
machine that has dropped off the platform, and reactive network triage is
tracked separately.

Add the page to the Maintenance group and link it from the Network
requirements in Verification Stages.
@mintlify

mintlify Bot commented Sep 3, 2026

Copy link
Copy Markdown
Contributor

Preview deployment for your docs. Learn more about Mintlify Previews.

Project Status Preview Updated
vastai-80aa3a82 🟢 Ready View Preview Sep 3, 2026, 11:12 PM

💡 Tip: Enable Automations to automatically generate PRs for you.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant