Troubleshooting And Log Files

Troubleshooting Network Service Problems

If the Networking Service appears Unhealthy in the Private Cloud Director Service Health page, this likely means that the networking service components running on one or more hosts are experiencing an issue. Reference the Networking Service Log Files to identify the specific error.

Log Files

The Networking Service component log files are located at:

/var/log/pf9/pf9-neutron-ovn-metadata-agent.log

/var/log/pf9/ovn/ovn-controller.log

Agent naming

pcdctl network agent list reports these components as the OVN Neutron agent and the OVN Controller Gateway agent. The on-host package, service, and log file names have not changed and still use the earlier pf9-neutron-ovn-metadata-agent naming shown above. This naming difference is expected and does not indicate a failed or incomplete deployment.

## VM Cloud-Init Fails to Retrieve Metadata on a Physical Network

Symptom

A virtual machine on a physical network boots without its SSH keys injected or its user data applied. The VM's console log shows repeated Missing mandatory path warnings for meta_data.json, ending in a DataSourceNone fallback message from cloud-init.

Cause

A known cause is a flat physical network whose host interface also carries the hypervisor's own management traffic. When the host's management address falls inside the subnet's allocation pool, return traffic for the metadata connection can be misdirected back onto the physical uplink instead of the local metadata proxy, so the connection never completes before cloud-init exhausts its retries. This has been observed on flat networks; tagged (VLAN) physical networks and virtual networks are not known to be affected, since they do not share a physical bridge with host management traffic in the same way.

Resolution

1. Check whether the host's management address is already allocated to a port:

   pcdctl port list --fixed-ip ip-address=<host-management-ip>

   If a port already owns that address, resolve the conflict before changing the allocation pool.
2. Narrow the subnet's allocation pool to exclude the host's management address and any other infrastructure addresses on the same interface:

   pcdctl subnet set <subnet-id> --no-allocation-pool
   pcdctl subnet set <subnet-id> \
     --allocation-pool start=<pool-start-ip>,end=<pool-end-ip>

3. For a more thorough fix, keep hypervisor management interfaces on a dedicated management network, separate from any physical network used for tenant workloads. This removes the shared-bridge condition rather than just working around it.

See Physical Network for background on allocation pools.

Related Pages

• Physical Network: allocation pool configuration for physical networks