Reading the OVN logs with Neutron. ovn-controller, ovn-northd, the OVN databases and the Neutron OVN driver¶
With ML2/OVN there are no OVS, L3 or DHCP agents. Neutron writes the wanted state into the OVN Northbound database,
ovn-northd translates it into the Southbound database, and ovn-controller on each chassis (compute or gateway
node) programs Open vSwitch from it. A port that never comes up has stopped at one of those steps, and each step
has its own log. Messages are quoted from OVN 25.09.5, the Open vSwitch library (lib/reconnect.c, 3.6.0) that OVN
daemons use for their database connections, and Neutron 2026.1.
ML2/OVS deployments: reading the Neutron logs with ML2/OVS.
Where the logs are¶
Kolla-Ansible writes all OVN logs in /var/log/kolla/openvswitch/ (fluentd's input/10-openvswitch.conf.j2
lists them):
| Daemon | Container | Log | Where |
|---|---|---|---|
| ovn-controller | ovn_controller |
ovn-controller.log |
every chassis (computes, network nodes) |
| ovn-northd | ovn_northd |
ovn-northd.log |
controllers |
| Northbound DB | ovn_nb_db |
ovn-nb-db.log |
controllers |
| Southbound DB | ovn_sb_db |
ovn-sb-db.log |
controllers |
| Southbound relays | ovn_sb_db_relay_<n> |
ovn-sb-relay-<n>.log |
controllers |
Neutron's side is in /var/log/kolla/neutron/: neutron-server.log, the OVN maintenance worker
(neutron_ovn_maintenance_worker) and, on the chassis, neutron-ovn-metadata-agent.log.
Kolla-Ansible 2026.1 enables Southbound relays with OVN by default (enable_ovn_sb_db_relay), one relay per 50
chassis (ovn_sb_db_relay_compute_per_relay), listening on ports 1664<n>. Southbound clients configured with
ovn_sb_connection, ovn-controller among them, connect to a relay, not to the Southbound cluster directly
(group_vars/all/ovn.yml). neutron-server and ovn-northd are configured with ovn_sb_connection_no_relay
(ml2_conf.ini.j2, ovn-northd.json.j2): they connect to the Southbound cluster on port 6642, so their connection
lines name a different target.
Format. OVN and OVS daemons write <date>T<time>.<ms>Z|<sequence>|<module>|<LEVEL>|<message>, in UTC
(fluentd's parser for these files). oslo.log lines in Neutron's logs are in the host's local time: line them up
with care.
# Kolla-Ansible, on a compute
tail -50 /var/log/kolla/openvswitch/ovn-controller.log
grep -E '\|(WARN|ERR|EMER)\|' /var/log/kolla/openvswitch/ovn-controller.log | tail -20
Database connections (every OVN daemon)¶
OVN daemons use Open vSwitch's reconnect logic for each database connection, and log its transitions
(lib/reconnect.c). <target> is the connection string, for example tcp:<address>:6642 (Southbound),
tcp:<address>:1664<n> (a relay), or unix:/run/openvswitch/db.sock (the local OVS database).
| Line (message text) | Level | What it means |
|---|---|---|
<target>: connecting... |
INFO | Opening the connection. |
<target>: connected |
INFO | Connected. |
<target>: connection dropped (<error>) |
WARN | The connection broke: the remote database restarted, the network failed, or a TLS problem. |
<target>: waiting <n> seconds before reconnect |
INFO | Backing off. The interval grows up to the configured maximum. |
<target>: continuing to reconnect in the background but suppressing further logging |
INFO | The backoff reached its maximum; reconnection attempts continue but are no longer logged. Silence after this line does not mean the connection came back: look for connected. |
<target>: no response to inactivity probe after <n> seconds, disconnecting |
ERR | The remote side did not answer the keep-alive probe: it is overloaded, stuck, or the network drops packets. A classic sign of an overloaded Southbound database or relay. |
grep -E "connection dropped|inactivity probe|connected$" /var/log/kolla/openvswitch/ovn-controller.log | tail -20
ovn-controller: port binding on the chassis¶
When an instance's tap device appears on br-int with the Neutron port id in its external_ids:iface-id, the local
ovn-controller claims the matching logical port. Neutron names the logical switch port after the Neutron port id
(lport_name=port['id'] in ovn_client.py), so these lines can be grepped by port id (controller/binding.c):
| Line (message text) | Level | What it means |
|---|---|---|
Claiming lport <port-id> for this chassis. |
INFO | This chassis takes the port. |
<port-id>: Claiming <mac> <ip> ... |
INFO | The addresses claimed with it. |
Claiming lport <port-id> for this additional chassis. |
INFO | During a live migration the destination usually claims the port as an additional chassis first. |
Changing chassis for lport <port-id> from <old> to <new>. |
INFO | The port moves to this chassis (live migration, or a port bound on two chassis at once). Repeated back and forth between two chassis: both claim it. |
Setting lport <port-id> ovn-installed in OVS |
INFO | The flows for the port are installed locally. |
Setting lport <port-id> up in Southbound |
INFO | ovn-controller marks the port up; Neutron will see it. |
Releasing lport <port-id> from this chassis (sb_readonly=<n>) |
INFO | The port left this chassis (instance deleted or moved). |
Setting lport <port-id> down in Southbound |
INFO | The port is marked down. |
# On the compute that hosts the instance
grep "<port-id>" /var/log/kolla/openvswitch/ovn-controller.log
No Claiming lport line for a new instance's port means ovn-controller never saw the interface with that
iface-id: check sudo docker exec openvswitch_vswitchd ovs-vsctl --columns=name,external_ids list Interface on the compute (Kolla-Ansible), and libvirt.
ovn-northd¶
Several ovn-northd instances run, one active, the others on standby, by holding a lock in the Southbound database
(northd/ovn-northd.c):
| Line (message text) | Level | What it means |
|---|---|---|
ovn-northd lock acquired. This ovn-northd instance is now active. |
INFO | This instance now translates NB into SB. |
ovn-northd lock lost. This ovn-northd instance is now on standby. |
INFO | Another instance took over. Frequent flips mean the instances lose their SB connection. |
OVN SB IDL reconnected, force recompute. |
INFO | After a reconnection, a full recomputation. On a large deployment it takes time and CPU. |
grep -h "lock acquired\|lock lost" /var/log/kolla/openvswitch/ovn-northd.log | tail
Exactly one controller should have logged lock acquired last.
The Neutron OVN driver¶
| Line (message text) | Level | What it means | Next |
|---|---|---|---|
Refusing to bind port <port-id> due to no OVN chassis for host: <host> |
WARNING | Neutron found no ovn-controller registered for the host Nova asked for: its ovn-controller is down or registered under another hostname. The port ends binding_failed. |
openstack network agent list --host <host>; ovn-controller on that host |
Refusing to bind port <port-id> to dead agent: <agent> |
WARNING | The chassis exists but its agent (ovn-controller) is considered dead. | same |
Refusing to bind port <port-id> on host <host> due to the OVN chassis bridge mapping physical networks [...] not supporting physical network: <physnet> |
INFO | A flat or VLAN network whose physical network is not in that chassis's ovn-bridge-mappings. |
the chassis's bridge mappings |
OVN reports status up for port: <port-id> |
INFO | Neutron saw the Southbound port go up (after Setting lport … up in Southbound). Neutron then notifies Nova (network-vif-plugged). |
— |
OVN reports status down for port: <port-id> |
INFO | The port went down. | — |
Maintenance task: … (inconsistencies found, fixing resource …) |
DEBUG / ERROR on failure | The maintenance worker repairs differences between Neutron's database and OVN's Northbound. Maintenance task: Failed to fix resource … with a traceback means a resource stays out of sync. |
the traceback |
Sources: mech_driver.py, ovsdb/maintenance.py (Neutron 2026.1).
The chain for a new port, by port id: Claiming lport <port-id> (ovn-controller, compute) →
Setting lport <port-id> up in Southbound → OVN reports status up for port: <port-id> (Neutron) →
Creating event network-vif-plugged:<port-id> … (nova-api). The step after the last line you find is where it
stopped; if Nova timed out, see waiting for vif_plugged.
Reading an OVN problem in order¶
- Database connections of the affected ovn-controller:
connection dropped,inactivity probe. - The port on its chassis:
Claiming lportandup in Southboundfor the port id. - Neutron:
Refusing to bind port …(binding) orOVN reports status up(the event to Nova). - The control plane: one active
ovn-northd; the NB/SB database and relay logs for restarts or elections.
Validity and limits¶
- Messages are quoted from OVN 25.09.5, Open vSwitch 3.6.0
lib/reconnect.cand Neutron 2026.1. The OVN and OVS versions in a Kolla 2026.1 image come from its distribution repositories and may differ slightly. No line was captured on a lab. A lab request asks for real samples. - Which Neutron process logs
OVN reports status up(API workers, the maintenance worker) was not observed. - The NB and SB database servers' own logs (RAFT elections, compaction) are not detailed here.
Related¶
- Reading the Neutron logs with ML2/OVS
- Waiting for vif_plugged
- Where the OpenStack logs are, and how to read one log line
- Logs index
Provenance: written by agent:ozarland-author, from OVN, Open vSwitch, Neutron and Kolla-Ansible source code. Reviewed in its pull request (see the page history). Checked by a human operator: not yet.
Worked for you, or not? Tell us in one line: your release, your deployment method, and what happened.