OVN-Kubernetes node metrics¶
This page documents metrics produced by ovnkube-controller, ovnkube-node, and their libovsdb clients. Metrics collected from the OVN and OVS daemons are listed separately in the OVN and Open vSwitch catalogs.
OVN-Kubernetes Controller Metrics¶
All metrics in this section are rooted at ovnkube_controller_.
Lifecycle and Resource Processing¶
| Name | Type | Labels | Description |
|---|---|---|---|
build_info | Gauge | version, revision, branch, build_user, build_date, goversion | Build information. The value is always 1. |
ready_duration_seconds | Gauge | None | Time for ovnkube-controller to become ready. |
sync_duration_seconds | Gauge | resource_name | Time to complete initial synchronization and set up handlers for a resource. |
resource_update_total | Counter | name, event | Resource add, update, and delete events handled. |
resource_add_latency_seconds | Histogram | None | Deprecated: This metric will be removed in the next release. Time to process all handlers for an add event. |
resource_update_latency_seconds | Histogram | None | Deprecated: This metric will be removed in the next release. Time to process all handlers for an update event. |
resource_delete_latency_seconds | Histogram | None | Deprecated: This metric will be removed in the next release. Time to process all handlers for a delete event. |
logfile_size_bytes | Gauge | logfile_name | Size of the configured ovnkube-controller log file. No series is produced when file logging is not configured. |
Pod and Service Programming¶
| Name | Type | Labels | Description |
|---|---|---|---|
pod_creation_latency_seconds | Histogram | None | Time from pod scheduling until logical switch port configuration completes. User-defined network pods are not currently recorded. |
pod_first_seen_lsp_created_duration_seconds | Histogram | None | Time from first observing a pod until its logical switch port is created. |
pod_lsp_created_port_binding_duration_seconds | Histogram | None | Time from logical switch port creation until the port binding is observed. |
pod_port_binding_port_binding_chassis_duration_seconds | Histogram | None | Time from observing the port binding until its chassis assignment is observed. |
pod_port_binding_chassis_port_binding_up_duration_seconds | Histogram | None | Time from chassis assignment until the port binding is observed as up. |
requeue_service_total | Counter | None | Service reconciliations requeued after failing to synchronize with OVN. |
sync_service_total | Counter | None | Service synchronizations with OVN load balancers. |
sync_service_latency_seconds | Histogram | None | Time to synchronize a service with OVN load balancers. |
OVN State and Configuration¶
| Name | Type | Labels | Description |
|---|---|---|---|
nb_e2e_timestamp | Gauge | None | Unix timestamp written by this controller to the northbound database. |
sb_e2e_timestamp | Gauge | None | Unix timestamp observed in the southbound database. |
ipsec_enabled | Gauge | None | Whether IPsec is enabled: 1 for enabled and 0 otherwise. |
egress_routing_via_host | Gauge | None | Gateway mode: 0 for shared, 1 for local, and 2 for an invalid mode. |
num_egress_firewalls | Gauge | None | Number of egress firewall policies. |
num_egress_firewall_rules | Gauge | None | Number of egress firewall rules. |
Admin Network Policy¶
The rule collectors are available when the AdminNetworkPolicy controller is running. Database object series are updated when AdminNetworkPolicy is enabled.
| Name | Type | Labels | Description |
|---|---|---|---|
admin_network_policies | Gauge | None | Number of AdminNetworkPolicy resources. |
baseline_admin_network_policies | Gauge | None | Number of BaselineAdminNetworkPolicy resources. |
admin_network_policies_rules | Gauge | direction, action | Number of rules across AdminNetworkPolicy resources. Actions are Allow, Deny, and Pass. |
baseline_admin_network_policies_rules | Gauge | direction, action | Number of rules in the BaselineAdminNetworkPolicy. Actions are Allow and Deny. |
admin_network_policies_db_objects | Gauge | table_name | OVN northbound database objects owned by the AdminNetworkPolicy controller. |
baseline_admin_network_policies_db_objects | Gauge | table_name | OVN northbound database objects owned by the BaselineAdminNetworkPolicy controller. |
Configuration Duration Recorder¶
These metrics are registered only when --metrics-enable-config-duration is enabled. They report an upper bound: unrelated work and the slowest node may increase a measurement.
| Name | Type | Labels | Description |
|---|---|---|---|
network_programming_duration_seconds | Histogram | kind | End-to-end time to apply an add, update, or delete for a Kubernetes resource kind. The value includes OVN time when it is measured. |
network_programming_ovn_duration_seconds | Histogram | None | Time for OVN to apply measured configuration to all relevant nodes. |
Scale Metrics¶
These metrics are registered only when --metrics-enable-scale is enabled.
| Name | Type | Labels | Description |
|---|---|---|---|
egress_ips_assign_latency_seconds | Histogram | None | Time to assign an egress IP in the OVN northbound database. |
egress_ips_unassign_latency_seconds | Histogram | None | Time to unassign an egress IP from the OVN northbound database. |
network_policy_event_latency_seconds | Histogram | event | Time to handle a complete network policy create or delete event. |
network_policy_local_pod_event_latency_seconds | Histogram | event | Time to handle a local pod add or delete for network policy. |
network_policy_peer_namespace_event_latency_seconds | Histogram | event | Time to handle a peer namespace add or delete for network policy. |
pod_selector_address_set_pod_event_latency_seconds | Histogram | event | Time to handle a peer pod add or delete for a pod-selector address set. |
pod_selector_address_set_namespace_event_latency_seconds | Histogram | event | Time to handle a peer namespace add or delete for a pod-selector address set. |
pod_event_latency_seconds | Histogram | event | Time to handle a pod add, update, or delete. |
udn_nbdb_programmed_duration_seconds | Histogram | topology | Time to program the northbound database while initializing a UDN, grouped by topology. |
workqueue_depth | Gauge | name | Current workqueue depth. |
workqueue_adds_total | Counter | name | Items added to a workqueue. |
workqueue_queue_duration_seconds | Histogram | name | Time an item remains queued before processing. Buckets are powers of two from 1ms to ~524s. |
workqueue_work_duration_seconds | Histogram | name | Time spent processing an item. Buckets are powers of two from 1ms to ~524s. |
workqueue_unfinished_work_seconds | Gauge | name | Total in-progress work not yet observed by the work-duration metric. |
workqueue_longest_running_processor_seconds | Gauge | name | Runtime of the longest-running workqueue processor. |
workqueue_retries_total | Counter | name | Workqueue retries. |
NetworkQoS¶
NetworkQoS metrics are registered by the NetworkQoS controller package. Duration values use milliseconds, as indicated by the metric names.
| Name | Type | Labels | Description |
|---|---|---|---|
num_network_qoses | Gauge | network | Number of NetworkQoS resources per network. |
nqos_ovn_operation_duration_ms | Histogram | operation | Time spent on an OVN operation during NetworkQoS reconciliation. |
nqos_creation_duration_ms | Histogram | network | Time spent reconciling a NetworkQoS event. |
nqos_deletion_duration_ms | Histogram | network | Time spent reconciling a pod event for NetworkQoS. The historical metric name is retained for compatibility. |
nqos_ns_reconcile_duration_ms | Histogram | network | Time spent applying a namespace change to related NetworkQoS pods. |
nqos_status_patch_duration_ms | Histogram | network | Time spent patching NetworkQoS status. |
Route Import¶
These metrics are registered only when both --enable-route-advertisements and --metrics-enable-scale are enabled. The network label is the network name, and its series are removed when the network is no longer imported.
| Name | Type | Labels | Description |
|---|---|---|---|
route_import_sync_duration_seconds | Histogram | network, result | Time in seconds spent importing BGP routes for a network, covering the netlink route dump, the OVN static route scan, the diff, and the resulting transaction. The result label is success or error. |
route_import_routes | Gauge | network, source | Routes seen by the most recent import sync for a network. Source bgp is what the kernel VRF table holds, source ovn is what the gateway router holds. |
route_import_operations_total | Counter | network, op | Logical router static routes added or deleted by route import. The op label is add or delete. |
OVN-Kubernetes Node Metrics¶
All metrics in this section are rooted at ovnkube_node_.
| Name | Type | Labels | Description |
|---|---|---|---|
build_info | Gauge | version, revision, branch, build_user, build_date, goversion | Build information. The value is always 1. |
ready_duration_seconds | Gauge | None | Time for ovnkube-node to become ready. |
cni_request_duration_seconds | Histogram | command, err | Duration of CNI server requests, grouped by command and error status. |
nodeport_enabled | Gauge | None | Whether NodePort is enabled on the node: 1 for enabled and 0 otherwise. |
logfile_size_bytes | Gauge | logfile_name | Size of the configured ovnkube log file. No series is produced when file logging is not configured. |
EVPN¶
These metrics are registered only when both --enable-evpn and --metrics-enable-scale are enabled. The vtep label is the VTEP resource name, and its series are removed when the VTEP is deleted.
| Name | Type | Labels | Description |
|---|---|---|---|
evpn_pod_program_duration_seconds | Histogram | result | Time in seconds spent programming the FDB and neighbor entries for a single pod on an EVPN network. The result label is success or error. |
evpn_pod_entries | Gauge | None | Pods on this node with EVPN FDB entries programmed. EVPN route count scales with this, not with node count. |
evpn_neighbor_entries | Gauge | None | Permanent neighbor entries programmed on this node for pods on EVPN networks. |
evpn_vtep_reconcile_duration_seconds | Histogram | vtep, result | Time in seconds spent reconciling the netlink devices for a VTEP. The result label is success or error. |
evpn_networks | Gauge | vtep | EVPN networks this node resolves to a VTEP, as of the last reconcile. |
evpn_vlans_used | Gauge | vtep | VLAN IDs this node maps on a VTEP's single VXLAN device, one per MAC-VRF and one per IP-VRF. The hard limit is 4094. Reported as resolved rather than as programmed, so a count that exceeds what the kernel accepted is still visible; pair with evpn_vtep_reconcile_duration_seconds{result="error"} to tell the two apart. |
libovsdb Client Metrics¶
These metrics describe the OVN northbound and southbound clients used by ovnkube-controller. The primary_model constant label distinguishes the client model. Its values are OVN_Northbound and OVN_Southbound.
All metrics in this section are rooted at ovnkube_master_libovsdb_.
| Name | Type | Labels | Description |
|---|---|---|---|
update_messages_total | Counter | primary_model, database | Monitor update messages processed, grouped by database. |
table_updates_total | Counter | primary_model, database, table | Monitor updates processed, grouped by database and table. |
disconnects_total | Counter | primary_model | libovsdb client disconnects. |
monitors | Gauge | primary_model | Running libovsdb monitors. |
OVSDB Transaction Metrics¶
These metrics are observed by the shared OVSDB transaction path used by ovnkube-controller and ovnkube-node for writes to the OVN northbound, OVN southbound, and Open vSwitch databases. A transaction with no operations, or one whose context is already cancelled, is not submitted and not counted. They are registered only when --metrics-enable-scale is enabled. The db label is the database name, such as OVN_Northbound, OVN_Southbound, or Open_vSwitch.
All metrics in this section are rooted at ovnkube_ovsdb_.
| Name | Type | Labels | Description |
|---|---|---|---|
txn_duration_seconds | Histogram | db, result | Time in seconds taken by an OVSDB transaction, including reconnect retries and result checking. The result label is success or error. |
txn_ops | Histogram | db | Number of operations submitted in a single OVSDB transaction. |
Shared OVN-Kubernetes Metrics¶
All metrics in this section are rooted at ovnkube_.
| Name | Type | Labels | Description |
|---|---|---|---|
resource_retry_failures_total | Counter | None | Resources that reached the maximum reconciliation retry limit and are no longer processed. This family is shared by OVN-Kubernetes components. |