> ## Documentation Index
> Fetch the complete documentation index at: https://docs.kguardian.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# compute findings

> List compute findings — throttled, contended and noisy-neighbour pods — for a namespace, a node, or the cluster

## Synopsis

List the broker's current compute findings. Read-only: the command calls
`GET /compute/findings` and prints the result; it generates nothing and
applies nothing.

```bash theme={null}
kubectl kguardian compute findings [flags]
```

## Flags

| Flag              | Type   | Description                                     | Default        |
| ----------------- | ------ | ----------------------------------------------- | -------------- |
| `-n, --namespace` | string | Only findings whose victim is in this namespace | all namespaces |
| `--node`          | string | Only findings whose victim runs on this node    | all nodes      |
| `-o, --output`    | string | `table` or `json`                               | `table`        |

Global flags (`--kubeconfig`, `--context`, `--broker-namespace`,
`--broker-service`, `--debug`) are listed in the [CLI overview](/cli/overview).

## Output

The table has one row per finding:

```
SEVERITY  KIND            VICTIM                       CULPRIT                 SHARE  MESSAGE
high      noisy-neighbor  payments/api-7c9d8f6b5-x2k4q  batch/etl-1-x           71%    payments/api is starved for CPU by batch/etl-1-x (71% of its wait); etl-1-x is using 1.9 cores against a 0.5-core request.
high      noisy-neighbor  logging/fluentd-4m2pq         system:kubelet.service  55%    logging/fluentd is starved for CPU by kubelet.service (55% of its wait).
medium    cpu-throttled   payments/worker-5f7c9-b8n3d   -                       -      payments/worker is throttled 34% of the time by its own 500m limit.
```

`VICTIM` and `CULPRIT` are printed as `namespace/pod`; the container is in
the JSON output. A `system` culprit prints as `system:<unit>` (for example
`system:kubelet.service`), a `kernel` culprit as `kernel`. `CULPRIT` and
`SHARE` are `-` for the kinds that never name one (`cpu-throttled`,
`cpu-contended`, `memory-limit-thrash`). `SHARE` is the culprit's share of
the victim's scheduler wait for `noisy-neighbor`, and its share of the
node's memory overage for `memory-pressure`.

`-o json` prints the broker's response unchanged (`findings`, `truncated`,
`victims_evaluated`, `history_disabled`); the schema is in the
[compute reference](/reference/compute#findings-response). In table mode the
CLI prints a notice when `truncated` is set (narrow the scope with `-n` or
`--node`), when `history_disabled` is set (`retentionDays: 0` — no findings
are possible), and how many victims were evaluated.

## Examples

```bash theme={null}
# Everything in one namespace
kubectl kguardian compute findings -n payments

# One node, cluster-wide, as JSON for scripting
kubectl kguardian compute findings --node worker-3 -o json

# Only noisy neighbours
kubectl kguardian compute findings -o json | jq '.findings[] | select(.kind == "noisy-neighbor")'
```

***

See [Compute Contention](/concepts/compute-contention) for what each kind
means and why a throttled pod is never blamed on a neighbour.