What does a managed Kubernetes SLA actually cover?
A managed Kubernetes SLA covers the availability of the cluster’s Kubernetes API server, measured per month, and nothing else by default. Your nodes fall under the provider’s compute SLA, and your workloads are not covered at all. As of November 2026, Amazon EKS, Google GKE and Azure AKS commit to between 99.5% and 99.95% for standard control planes, and the only remedy for a breach is a service credit on that cluster’s fee.
What the SLA measures
Each provider defines the covered service as the control plane endpoint:
- The Amazon EKS SLA covers “the EKS Cluster control plane run by AWS”. A cluster counts as unavailable when all connection requests to its Kubernetes endpoint fail during a 5-minute interval. EC2 instances attached to the cluster fall under the Amazon Compute SLA instead.
- The GKE SLA covers the Kubernetes API of the cluster, and only while the cluster runs a version offered in the Stable, Regular or Extended release channel. Downtime counts only in periods of five or more consecutive minutes and excludes scheduled maintenance and failures of nodes or the Pods on them.
- The AKS pricing tiers page states its SLA as availability of the Kubernetes API server.
A cluster can therefore meet its SLA while your application is down. If nodes fail, a zone loses capacity or a bad deploy crashes your Pods, the API server is still reachable and no SLA applies.
Commitments by provider and tier
| Service and tier | Monthly uptime commitment | Notes |
|---|---|---|
| EKS, standard control plane | 99.95% | Applies to every EKS cluster |
| EKS, Provisioned Control Plane | 99.99%, measured in 1-minute intervals | Paid scaling tier, priced per hour on top of the cluster fee |
| GKE zonal cluster | 99.5% | Single control plane in one zone |
| GKE regional cluster | 99.95% | Control plane replicas in several zones |
| GKE Autopilot cluster | 99.95% control plane, 99.9% for Pods spread across several zones | The only tier in this table that covers Pods |
| AKS Free tier | None | Best effort, no financially backed SLA |
| AKS Standard and Premium | 99.95% with availability zones, 99.9% without | AKS Automatic adds a Pod readiness SLA |
GKE values apply to most regions; Stockholm has lower commitments for regional and Autopilot clusters. AKS Automatic’s Pod readiness SLA promises that 99.9% of qualifying Pod readiness operations complete within five minutes.
What the percentages mean in minutes
Uptime percentages are easier to judge as allowed downtime. For a 30-day month (43,200 minutes):
| Commitment | Allowed downtime per month |
|---|---|
| 99.5% | 216 minutes (3 hours 36 minutes) |
| 99.9% | 43.2 minutes |
| 99.95% | 21.6 minutes |
| 99.99% | 4.3 minutes |
Measurement rules shrink the effective protection further. On GKE, outages shorter than five consecutive minutes do not count toward downtime. On EKS, an interval counts only if all connection requests fail, so a partially degraded API server can stay inside the SLA.
The remedy is a service credit
SLAs do not pay damages. If the provider misses the commitment, you can claim a credit against future bills for the affected service.
| Monthly uptime | EKS credit | GKE regional or Autopilot credit |
|---|---|---|
| Below the commitment, at least 99.0% | 10% | 10% |
| Below 99.0%, at least 95.0% | 25% | 25% |
| Below 95.0% | 100% | 50% |
Credits apply to the charge for the covered service, which for a standard control plane is the cluster fee, about $73 a month on EKS or GKE. A 10% credit is about $7. GKE caps all credits in a month at 50% of the bill for the covered service. Credits are not automatic: AWS requires a support case by the end of the second billing cycle after the incident, and Google requires notice within 30 days along with logs showing the errors. AKS credits are defined in Microsoft’s Service Level Agreements for Online Services.
The practical value of an SLA is less the credit than the architecture behind it. On GKE and AKS, the 99.95% commitment is tied to a control plane replicated across zones, while single-zone control planes get 99.5% or 99.9%.
What the SLA does not cover
- Nodes. Virtual machines fall under the compute SLA, which has its own terms and usually requires instances spread across zones.
- Your workloads. Pod health, application errors and data loss are your responsibility, except Autopilot’s Pod commitment and AKS Automatic’s Pod readiness SLA.
- Load balancers, VPNs and storage. These have separate SLAs. GKE’s SLA names load balancing and VPN tunnels as covered elsewhere.
- Support response times. How fast a provider answers a ticket is set by a support plan, which is a separate contract with its own price.
- Versions outside support. GKE excludes clusters on versions not offered in its covered release channels.
Who handles what is set by the provider’s shared responsibility model, not the SLA. When comparing services, ask for both documents. The checklist for choosing a managed Kubernetes service lists the questions, and making Kubernetes highly available across clusters covers the parts of availability the SLA leaves to you. For background on what the API server does during an outage, see the Kubernetes control plane.
Cloudfleet SLAs
Cloudfleet Basic clusters have a best-effort control plane, Pro clusters have a 99.95% monthly control plane uptime SLA with multi-zone replication, and Enterprise clusters have a 99.99% SLA with custom terms. Support response SLAs are part of the tier: 8 hours on Pro and 1 hour with a named technical account manager on Enterprise. See cluster types for details and the shared responsibility model for an example of how the operational split is documented.