# Metrics

> CPU, memory and replica charts for every deployment, plus request rate by status class, p95 latency and network in and out for public web apps. In the dashboard and from okctl.

The **Overview** tab of a deployment charts what it is using and, for a public web app, what traffic it is serving. The charts come from Ownkube's own telemetry; you do not add an agent or a library.

## Resource metrics

Every web app, worker, job, database and cache shows:

| Metric | What it measures |
| --- | --- |
| **CPU** | vCPU in use, against the limit you set. |
| **Memory** | The working set the container holds, against its memory limit. |
| **Replicas** | Running against desired, for example `3 / 3`. |
| **Restarts** | Container restarts in the window. A climbing count means the app is crashing. |

These are the numbers [autoscaling](/docs/deployments/autoscaling) scales on and [alerts](/docs/observability/alerts) fire on. A metered app bills on them too: see [Understanding your bill](/docs/billing/usage).

## Request metrics

A web app with a [public hostname](/docs/networking/public-hostnames) also gets the **Observability** panel, sourced from the region edge:

| Metric | What it measures |
| --- | --- |
| **Requests by status** | Request rate per minute, split into 2xx, 3xx, 4xx and 5xx. |
| **p95 latency** | The response time 95% of requests beat, as seen by the edge. |
| **Network in** | Bytes received from clients. |
| **Network out** | Bytes sent to clients. Billed per GB; see [Regions](/docs/deployments/regions). |

A worker or an app with its public hostname off has no edge traffic, so the panel does not show. Traffic over the [private network](/docs/networking/private) is not counted here.

## The time picker

One picker on the Overview tab sets the window for every chart: the last 15 minutes, hour, 6 hours or 24 hours, or a custom from and to with a step. The same window follows you to the [Logs](/docs/observability/logs) tab, so an incident you spot on a chart is one click from the lines that explain it.

## From the terminal

```bash
okctl deploy telemetry <id>        # CPU, memory and replicas
okctl deploy observability <id>    # requests, latency and errors
```

Both print a table; add `-o json` for scripting. A coding agent connected through the [MCP server](/docs/mcp) can read the same numbers.

## Reading the charts

- **CPU flat at the limit** with latency rising: the app needs a bigger limit or more replicas. Turn on [autoscaling](/docs/deployments/autoscaling) on CPU.
- **Memory climbing between restarts**: a leak. The restart count on the same chart confirms it. Set a [memory alert](/docs/observability/alerts) while you fix it.
- **5xx rising with CPU flat**: the problem is downstream, usually the database. Check its CPU and the [Logs](/docs/observability/logs) for connection errors.
- **4xx rising**: clients, not the app. Look at the paths in the logs.

## Retention

Metrics are kept for 7 days on Pay as you go, 30 days on Personal and 90 days on Team.

- [Alerts](/docs/observability/alerts)
- [Autoscaling](/docs/deployments/autoscaling)
- [Logs](/docs/observability/logs)

---

**Don't see a feature you need?** Email [support@ownkube.io](mailto:support@ownkube.io?subject=Feature%20request). Ownkube is shaped by the teams using it and we ship what our users ask for.

## Search

This site has a search endpoint: https://context7.com/api/v2/ask?siteKey=ask_c64dfe8ffaa8d3ad19576838&query=<URL-encoded question>. It returns documentation that answers the question, with a source link for each part. No API key is needed. If nothing matches, it says so.