# Health & upgrades

> Check that every platform component is healthy, and upgrade an install by re-running it — the same idempotent apply that stood it up.

Source: https://openap.org/docs/health-upgrades

A running OAP is a handful of platform components — the operator, authorization, channels, the web server, an
operator-managed SpiceDB, the messaging bus, and the memory stores. Two commands keep them honest:
`oap check` to see health, and re-running the install to upgrade.

## Checking health

`oap check` verifies each component and reports `✓ / ⚠ / ✗` per one, so "is the platform healthy?" has a
concrete answer rather than a guess. `oap check --watch` keeps re-checking every couple of seconds — useful
while an install or upgrade is converging. It's the same readiness the [install](https://openap.org/docs/install-overview) itself
waits on, available on demand.

[Video: `oap check` — each component reports for itself: ✓ healthy, ⚠ degraded, ✗ down.](https://openap.org/media/console-check.mp4)

## Upgrading

An upgrade is a **re-install**: `oap install` applies the current platform manifests over the running one. The
apply is idempotent — an unchanged component is left alone, and only what actually differs is updated — so
re-running it is safe. The resolved [cluster kind](https://openap.org/docs/install-cluster) is stamped onto the components at install,
which is why upgrading a pre-cluster-kind install means re-running `oap install` so the kind is set explicitly.

```bash
oap install --builder-starters user:$(oap identity canonical-id you@example.com)
```

`oap install` needs either `--builder-starters` (who may start Agent Builder) or `--without-builder`; see
[Agent Builder](https://openap.org/docs/agent-builder-install). The starter list replaces what was there, so pass the full list on every
upgrade.

> **Tip: Fail loud, not silent**
>
> The platform is deliberately fail-closed about its own configuration — a missing artifact store, an
> unrecognized cluster kind, or an unset required backend crash-loops the component rather than starting in a
> quietly-degraded state. So a healthy `oap check` means the thing is actually configured, not just
> running.
