kubectl ome status Unreleased: coming in v1.3

kubectl ome status shows in one report why an InferenceService is or isn't ready. It covers the Ready condition, each component's pods, the rollout, autoscaling, placement and traffic, the runtime the service runs with and the accelerator it asks for, and recent Warning events. It only reads. Start with it when a service misbehaves, then run the detail command for the part that looks wrong.

kubectl ome status INFERENCESERVICE [flags]

It reads the InferenceService, its pods, and the Warning events of both. When the service names a runtime and isn't a VirtualDeployment, it also reads the model, the serving runtime and the runtimes it inherits from, and the runtime revisions the service names. Required RBAC lists the permissions it needs.

Flags¶

Plus the standard kubeconfig flags, such as -n and --context.

Set --ome-namespace when OME runs in another namespace: the CLI reads runtime revisions there.

Output fields¶

The table has two columns, FIELD and VALUE, with one row per fact, in the order of the sections below. It cuts long values and ends them with .... -o wide adds the rows marked Wide. -o json and -o yaml print the whole report, a StatusReport with the sections below under content, and keep longer values. Output formats covers what every command shares.

parent status in a value means the CLI read it from the InferenceService's own status.

The CLI masks the reason and message of the Ready condition and of events. A value that contains :// shows as [OMITTED], and text that looks like a credential shows as [REDACTED]. To read the full text, use kubectl describe or kubectl get events.

The report's sections come in this order. When one looks wrong, run the command in the last column, with the service's name and the same -n:

In a healthy report, Ready and each component row start with True / Valid, every pod is Ready, and no Issue row appears. Some rows read Unknown or Unavailable on a healthy service too:

• Rollout shows Unknown on most services, with the rollout state in reported=.
• Traffic shows Unavailable when the service declares no traffic behavior.
• Runtime active and Accelerator show AutoSelectionNotProbed when the service names a model but no runtime.

The report marks how the CLI got each value with an evidence level:

The service¶

Ready is one of:

Ready doesn't check status.observedGeneration, so True / Valid can describe the service as it was before your last change. Compare the two numbers in the Generation row.

Condition inspection is Complete when every condition checks out, Partial when the CLI found a problem, and LimitExceeded when there are too many conditions to inspect. The Condition warning rows are:

Pods and components¶

The CLI counts the pods labelled ome.io/inferenceservice=<name> in the service's namespace that have a component label of engine, decoder or router. An Issue row names each reason it left a pod out (see Warning events and issues).

A component's <ready> / <validity> comes from its EngineReady, DecoderReady or RouterReady condition, which the CLI checks the same way as Ready, so it takes the same values. It's NotRecorded / Unavailable until the controller sets the condition.

For a component on OMENative, these rows count pods, not Instances. An Instance can be a leader pod and its workers. kubectl ome instance list shows each Instance with its phase, pods and revision.

Pod observation and Event observation show one of these states:

truncated=true means the read stopped at a limit or failed partway. The reason is one of:

Rollout¶

rollout status explains each value, and what each rollout issue means. The state is Unknown whenever the rollout evidence is Reported, because the CLI can't tie the controller's status to the current generation. That's the case for a service with rollout groups, and for most others. Read the reported= value instead.

Autoscaling¶

The autoscaling rows show the component autoscaling status that the controller recorded in status.components. To compare it with the live HorizontalPodAutoscaler or KEDA ScaledObject, run kubectl ome autoscale status with --live-scaler.

Autoscaling is one of:

kubectl ome autoscale status explains the values in a Scale row and the Autoscale issue codes. The state in Scale target and Scale conditions is Reported, NotReported, Unavailable or Invalid. PolicyEvidenceInvalid and PolicyConditionConflict come from an AutoscalerPolicy, which is alpha.

Placement¶

A service without placement shows Placement as NotConfigured / NotApplicable / NotRecorded. Placement homes, Reported cluster, Placement endpoint and Placement replicas appear when the state is Reported or Partial.

The placement state is:

The mode is Single, All, Split, Unknown or NotApplicable. kubectl ome placement status explains the phase, candidate and endpoint values.

Traffic¶

The traffic rows come from status.traffic on the InferenceService, which the controller writes for the traffic policy in spec.traffic.

The traffic state is:

On Envoy Gateway, Traffic stays Pending even after the gateway accepts the policy. To see the gateway's verdict, follow Read the traffic status. This is a known bug.

In Traffic policy Ready, the freshness is Current for the current generation and Stale for an older one. It's Unverifiable when the condition records no generation or a newer one, and Unavailable when the CLI couldn't read it.

Runtime and accelerator¶

Runtime active is one of:

Accelerator is one of:

In an Accelerator <component> row, the selection is NotReported, Unavailable, Invalid or NotConfigured, and the class is NotRequested. In Declared class, the intent is Class for a class by name, Policy for a selection policy, NotConfigured or Invalid. kubectl ome accelerator explain explains these values.

Warning events and issues¶

The report shows the Warning events of the InferenceService and its counted pods, newest first, within the limits. For Normal events and events on InferenceReplicas, run kubectl get events -n demo. An event can remain after the problem is fixed: -o wide shows when it was last seen.

An issue about a pod or an event means the CLI left it out of the report. The issues are:

Detail rows¶

The report ends with Full safe values, which says Use -o json or -o yaml, and six rows that name the commands in the section table, with NAME for the service's name. -o wide adds three rows: Collected at, when the CLI built the report, as an RFC 3339 time; Source generation, the generation of the InferenceService it read; and Source evidence, always Observed.

Limits¶

The command reads a bounded amount, so it answers quickly even for a large service. These limits show in the output:

When the CLI can read the InferenceService, any other read that fails or stops at a limit still leaves a report, with Partial or Unavailable in the rows it affects.

Examples¶

Show the status of qwen2-5-7b in demo, just after you created it:

kubectl ome status qwen2-5-7b -n demo

FIELD                VALUE
Name                 qwen2-5-7b
Namespace            demo
Ready                NotRecorded / Unavailable
Ready reason
Declared runtime
Model                qwen2-5-7b
Generation           1 observed=0; advisory Unverifiable
Pod observation      Reported count=0 truncated=false
Event observation    Reported count=0 truncated=false
engine               NotRecorded / Unavailable; Ready pods=0/0 restarts=0
Rollout              Unknown reported=Unknown
Rollout evidence     Reported / Unverifiable
Autoscaling          Unavailable / Unavailable parent status
Placement            NotConfigured / NotApplicable / NotRecorded
Traffic              Unavailable / Unavailable parent status
Runtime active       Unavailable / Unavailable AutoSelectionNotProbed
Accelerator          Unavailable / Unavailable AutoSelectionNotProbed
Full safe values     Use -o json or -o yaml
Rollout detail       kubectl ome rollout status NAME
Autoscale detail     kubectl ome autoscale status NAME
Placement detail     kubectl ome placement status NAME
Traffic detail       kubectl ome traffic status NAME
Runtime detail       kubectl ome runtime effective NAME
Accelerator detail   kubectl ome accelerator explain NAME

The controller hasn't written any status yet, so Ready is NotRecorded / Unavailable, and Autoscaling and Traffic are Unavailable. The engine row comes from spec.engine. The service names a model but no runtime, so Runtime active and Accelerator show AutoSelectionNotProbed.

Print the same report with every row:

kubectl ome status qwen2-5-7b -n demo -o wide

FIELD                  VALUE
Name                   qwen2-5-7b
Namespace              demo
Ready                  NotRecorded / Unavailable
Ready reason
Ready message
Condition inspection   Complete 0/0
Declared runtime
Model                  qwen2-5-7b
Generation             1 observed=0; advisory Unverifiable
Pod observation        Reported count=0 truncated=false
Event observation      Reported count=0 truncated=false
Pod targets skipped    0
Event targets skip     0
engine                 NotRecorded / Unavailable; Ready pods=0/0 restarts=0
engine Pod total       0
engine Pod Ready       0
engine restarts        0
engine phases          R=0 P=0 F=0 S=0 U=0 deleting=0
engine evidence        Observed
Rollout                Unknown reported=Unknown
Rollout evidence       Reported / Unverifiable
Coordination Ready     NotApplicable
Rollout issue          ComponentStatusMissing
Rollout issue          EpochUnverifiable
Rollout warning        PartialData
Autoscaling            Unavailable / Unavailable parent status
Placement              NotConfigured / NotApplicable / NotRecorded
Placement evidence     Unavailable / NotApplicable
Placement mode         NotApplicable
Traffic                Unavailable / Unavailable parent status
Runtime active         Unavailable / Unavailable AutoSelectionNotProbed
Accelerator            Unavailable / Unavailable AutoSelectionNotProbed
Full safe values       Use -o json or -o yaml
Rollout detail         kubectl ome rollout status NAME
Autoscale detail       kubectl ome autoscale status NAME
Placement detail       kubectl ome placement status NAME
Traffic detail         kubectl ome traffic status NAME
Runtime detail         kubectl ome runtime effective NAME
Accelerator detail     kubectl ome accelerator explain NAME
Collected at           2026-09-15T12:00:00Z
Source generation      1
Source evidence        Observed

The service has no conditions yet, so Condition inspection is Complete 0/0. The rollout issues say why Rollout is Unknown: status.components has no entry for the engine yet, and the CLI can't tie the rollout status to the current generation.

Print only the Ready section of the JSON report, with jq:

kubectl ome status qwen2-5-7b -n demo -o json | jq '.content.ready'

{
  "status": "NotRecorded",
  "validity": "Unavailable",
  "inspection": {
    "state": "Complete",
    "total": 0,
    "inspected": 0,
    "warnings": []
  }
}

The JSON leaves out reason and message while they're empty.

Ask for a service that doesn't exist:

kubectl ome status qwen2-5-7 -n demo

error: status InferenceService was not found

The command exits with code 1.

Exit codes¶

Errors print to stderr as error: <message>.

A run over 30 seconds, or Ctrl-C, prints only error: status read was cancelled or timed out. To wait for Ready in a script, use kubectl ome wait: kubectl ome wait qwen2-5-7b -n demo --for=condition=Ready.

Related guides¶

• Troubleshoot an InferenceService