Networking problems

Common networking errors on VMStack nodes and servers, from DHCP, IPv6 and bridges to NAT, private networks, DNS and the server firewall, with fixes.

These are the problems most often seen while testing VMStack networking on real nodes, grouped by area. Each one gives the symptom, the cause and the fix. Many were fixed in a later agent, so first check that the node runs the current agent: see Agent updates.

Addresses and routed networking#

Symptom Cause Fix
Servers that ask for DHCP on a routed node get no lease; the node's dnsmasq log says "no address range available for DHCP request via vmbr0" dnsmasq 2.90 and 2.91 (Debian 13) do not match the node's /32 on the bridge Update the agent; it writes shared-network lines for dnsmasq. See DHCP for routed servers.
cloud-init status says "degraded done" on AlmaLinux, Rocky or Fedora with a routed /32 cloud-init's schema check does not know the on-link key Harmless: the network works. See What the server gets.
Removing an address from a running server fails with live_update_failed The guest did not take the change live Shut the server down and remove the address again. See Several addresses per server.
A free address is never given out It is marked In use elsewhere: something else on the network answers for it Use Recheck, or Clear mark once you know the address is free. See In use elsewhere marks.
The block page says "A router answers for this block's addresses" A gateway, HSRP or VRRP MAC answers for every address, so the check cannot tell Turn the address check off for the block, or add the block as routed. See Routers that answer for every address.
A server got another address than the one first shown The first address was found in use at create and replaced Billing modules should read the address again once the server is ready. See At create and Add address.

IPv6#

Symptom Cause Fix
An IPv6 block is refused, or servers fail with ipv6_unavailable ("router advertisements handled outside the kernel") NetworkManager or another tool takes the uplink's router advertisements Give the node a static IPv6 address and gateway, or let the kernel take the advertisements. systemd-networkd is handled from agent 0.42.0. See What the agent changes on the node.
The node loses its IPv6 default route, or forwarded IPv6 stops after a network reload accept_ra was overridden (for example in /etc/sysctl.conf or 99-sysctl.conf), or the uplink lost its link-local address (netplan link-local: []) Remove the override. The agent adds the link-local address back every minute and writes a networkd drop-in. See IPv6.
IPv6 servers have no connectivity The node's IPv6 default route is not the provider's gateway Run the command the block page shows, then make it permanent in the node's network configuration. See The gateway check.

The gateway command the block page gives has this form:

ip -6 route replace default via 2001:db8::1 dev eth0 onlink

Replace 2001:db8::1 with your provider's gateway and eth0 with the node's uplink.

Bridged networking and VLANs#

Symptom Cause Fix
Setting up the uplink bridge ends in rolled_back with a reason The 60-second local check failed (addresses or default route not on the bridge, gateway silent), the panel could not confirm in time, the node rebooted while it waited, or the node's network setup is not supported Fix the cause the reason names and try again. See What happens during the switch.
The node is unreachable after a bridge change The rollback did not run From the provider's console, run the command below.
A server stays "Waiting for OVHcloud to make the virtual MAC", then fails with virtual_mac_failed The dns worker is not running, the OVHcloud service name is missing or wrong, the IP is not routed to that server, the MAC quota is reached, or 30 minutes passed Start the worker, correct the service name or fix it at OVHcloud, then create the server again. A release stuck in remove_failed has Try again and Forget. See OVHcloud virtual MACs: create, wait, release.
A bridged block says there is no free address although it has free rows The bridge is not active, the address has no MAC, its MAC is still being removed, or anti-spoofing is off Set up the bridge, enter the MACs, and check Edit block. See Which servers are bridged.
A Windows or ISO install on a bridged or VLAN address finds no DHCP The agent has no bridged DHCP responder yet, or the responder cannot open its socket (shown in red on the node's DHCP line) Update the agent and check that the DHCP line says "· N bridged". Otherwise set the address and gateway by hand. See DHCP on bridged servers.
A VLAN does not come up: "VLAN N is already on this host as …", "… is a bridge", "… is a port of bond0", or the MTU is too high The tag is already used, the VLAN interface is a bridge or a bond member, or the MTU is above the parent's Use another tag, name the physical port or the bond as the VLAN interface, or lower the MTU. See Refusals and problems.
A DHCP server on the VLAN's own router does not reach VMStack servers DHCP requests of servers on a VLAN are always dropped at the bridge Give the addresses through VMStack's IP block for the VLAN. See Protections.
vmstack-agent bridge rollback -config /etc/vmstack/agent.yaml

NAT and port forwards#

Symptom Cause Fix
A changed or deleted forward keeps forwarding, or a port cannot move to another server ("Could not process rule: File exists") Agent 0.25.0 did not empty the NAT maps on reload Update the agent to 0.25.1 or later.
A forward shows Not active because the host uses the port A service on the node listens on that public port on a wildcard or public address Choose another port, or bind the service elsewhere. See Rules a forward must follow.
A forward stays "Waiting: the public port is still in use on the host" Another server's forward on the node still holds the port for a moment Usually nothing: the panel sends it again. If it stays, look at the node's Port forwards tab. See States of a forward.
Open connections survive a forward change conntrack is missing on the node Install the conntrack package.
Forwards disappear after systemctl restart nftables The restart flushes every table Nothing to do: the agent loads its tables again within a minute.
The forward is active but the service does not answer The server firewall drops it: it matches after the port forward, on the server's port Accept the server port, not the node's public port. See When the firewall drops a forward.

Websites by domain#

Symptom Cause Fix
A domain stays on "Waiting for the TXT record" The TXT record is missing or at the wrong name, automatic checks stopped after 7 days, or another server verified the name first Add the TXT record exactly as the page shows it and use Check now. See Prove the domain is yours.
"port 80 (or 443) is used by another program on this node" Another web server on the node holds the port Stop or rebind it; the agent tries again every minute. See Before you start.
Every request fails after turning on PROXY protocol The server's web server does not expect the PROXY header Configure it (nginx listen ... proxy_protocol, Apache RemoteIPProxyProtocol On) or turn PROXY protocol off. See PROXY protocol.

Private networks#

Symptom Cause Fix
The node "cannot carry private networks: wg not found: install wireguard-tools" (or IPv6 is disabled, or the uplink MTU is too small) WireGuard tools missing, IPv6 off on the host, or the uplink MTU below the network MTU plus 150 Install wireguard-tools, enable IPv6 on the host, or lower the private network MTU. See Prepare the nodes.
Servers on the same node reach each other, servers on different nodes do not UDP 51820 is blocked between the nodes Open UDP 51820 between the node addresses at the provider's firewall. See Check it on the node.
A server added while running has no address on its private card The card is hot-plugged; the address applies at the next start Restart the server, or run the commands its Network tab shows. See What happens inside the server.

To check the tunnel between nodes:

wg show vmstack-wg

A peer without a recent handshake means its UDP 51820 is not reachable.

Floating IPs and load balancers#

Symptom Cause Fix
Moving a floating IP between servers of one node loses about 20 seconds of traffic Agent 0.59.0 left a stale ARP entry Update to agent 0.59.1 or later; a move then takes a few seconds.
A floating IP shows "On a server" but does not answer The address is not configured inside the server Use Reconfigure network, or add it by hand (below). See Use it inside the server.
A load balancer's targets show down The targets' firewall does not accept connections from the balancer's address Accept the balancer's address on the target ports. See Troubleshooting.
A listener is refused: "Port N is taken on node …" The port is the agent's or SSH, or is used by another service Choose another listener port.
ip addr add 203.0.113.50/32 dev eth0

Replace 203.0.113.50 with the floating IP and eth0 with the server's public interface.

DNS and reverse DNS#

Symptom Cause Fix
DNS records and reverse names stay Waiting The dns queue worker is not running systemctl status vmstack-worker@dns-1; start it.
"DNS record not created yet: The name is already used in the domain's DNS" The name already has a record the panel did not make Remove the other record or rename the server, then Retry. See Records the panel does not own.
OVHcloud refuses a name with 400 "Cannot check if … resolves to …" OVHcloud sets a PTR only once the name resolves to the address Create the A or AAAA record first, then Retry name. Names under the panel's domain wait for the panel's own record automatically.
ReliableSite's Check fails The API key works only from addresses allowed in ReliableSite's control panel Allow the panel's public IP there.
A name shows Failed after 8 attempts The provider keeps refusing it Read the provider's message on the block page, fix the cause, then Retry name.

Firewall#

Symptom Cause Fix
SSH is locked out after turning the firewall on The default inbound action is drop and no rule accepts TCP 22 From the console or as an administrator, add an accept rule for TCP 22 on the server's Firewall tab. A server in rescue takes firewall changes.
"This node's agent does not support the firewall yet; upgrade the agent." The node's agent is older than 0.16.0 Update the agent. Changes are kept and sent afterwards.
Rate limit, countries and counters are missing The node's agent is older than 0.45.0 Update the agent.
A country rule has no effect The node has not downloaded the country ranges yet, or the download failed Wait for the Country data: line to show the month. A node without internet access needs firewall_countries_url pointed at a mirror.
The firewall cannot be turned on for a bridged or VLAN server Their traffic is bridged, not routed through the node Not supported in this version; the node's network rules on the bridge apply instead. See Not supported.
The firewall shows Failed with invalid_firewall The node refused the rules, for example more than 100 from required templates Read the reason, remove rules, and the panel sends the firewall again.
A forwarded port of a NAT server is dropped Rules match after the forward, on the server's own port Accept the server port, not the node's public port.

To see the firewall on the node:

nft list table inet vmstack_fw
journalctl -k | grep vmfw

The second command shows dropped packets when Log dropped packets is on. For the node's own firewall, see Node firewall.