# Availability and the SLA

How {{product}} measures a cluster's API availability every minute, what the standard tier's 99.9 % commitment covers, and how credits are applied.

## How a minute is measured

Deutsche Cloud measures the API of every cluster that is `active`, `updating` or `restoring` once a
minute, two ways:

- **The machines' reports:** whether a control-plane machine reported a serving API.
- **From outside:** `GET /readyz` on the endpoint from Deutsche Cloud's control plane and its probe
  locations, with TLS pinned to the cluster's own CA. The outside measurement fails when a strict
  majority of the locations that answered failed, with at least two answering.

A minute is **unavailable** when either says so. Minutes Deutsche Cloud did not measure are not
counted, and neither is the time a cluster is being created or deleted, the time after its
Kubernetes minor's end of support, or the first 30 minutes of a restore you started (see
[Restore a cluster from a snapshot](restore-a-cluster.md)).

**Deutsche Cloud's blind spot:** a minute with no machine report at all, while at least two outside
locations asked and every one got an answer, is listed with the source `ews-blind-spot` and
counted as available (`blindSpotMinutes`).

With the public endpoint off there are no outside probes, and availability is measured from the
machines' reports alone (up to 90 seconds later at the start of an outage). While the public
endpoint is on, the probes' addresses (`platformAddresses`) are let through whatever the allowed
networks say, and they only request `/readyz`.

## Where to see it

Both need `eks:DescribeCluster` on the cluster.

- The cluster's **Monitoring** tab: this month's availability, minutes measured and unavailable and
  which measurement saw them, each probe location's last result, last month's result.
- `dcloud eks clusters availability my-cluster --month 2026-11`
  (`GET /eks/api/clusters/{cluster}/availability`).

## The SLA

The `standard` tier commits to 99.9 % API availability per calendar month (UTC). The `dev` tier has
no SLA, and neither does a cluster on a minor past its end of support. The SLA covers the API
endpoint; your workloads run on your nodes and keep running while the API is down.

| Monthly availability | Credit, as a share of that month's control-plane fee |
| -------------------- | ---------------------------------------------------- |
| below 99.9 %         | 10 %                                                 |
| below 99.0 %         | 25 %                                                 |
| below 95.0 %         | 100 %                                                |

Credits are automatic. On the 1st of each month Deutsche Cloud closes the previous month for every
cluster: the availability, the credit tier and the credit, which is never more than the fee. The
fee is the cluster's charged `standard` control-plane hours of the month while its minor was
supported. A month whose fee was €0 (a preview or free hours) earns no credit, but its result still
shows the tier it reached. The result is on the **Monitoring** tab and in `closed` of the availability answer.

## Related

- [Monitoring a cluster](monitoring.md)
- `eks.clusters.availability` · `dcloud eks clusters availability`
