specification: API Commons Rate Limits
specificationVersion: '0.1'
schema: https://raw.githubusercontent.com/api-evangelist/interface-research/main/schema/api-commons.yml#/$defs/RateLimits
provider: Kubernetes Services
providerId: kubernetes-services
created: '2026-05-04'
# Provenance stamped 2026-08-11: this artifact was written by the API Evangelist
# bulk sweep dated 2026-05-04, not harvested from the provider. See roadmap#35.
method: generated
modified: '2026-05-05'
reconciled: true
tags:
  - Rate Limiting
  - Kubernetes
  - Networking
  - Open Source
description: The Kubernetes Services API itself does not impose request-per-second limits on data-plane
  traffic to a Service. Control-plane operations (creating, updating, listing Services through the kube-apiserver)
  are subject to the API server's API Priority and Fairness (APF) just like any other Kubernetes API
  resource. Data-plane throughput depends on the chosen Service type and the underlying cloud-provider
  load balancer.
sources:
  - https://kubernetes.io/docs/concepts/cluster-administration/flow-control/
  - https://kubernetes.io/docs/concepts/services-networking/service/
responseCodes:
  throttled: 429
  serviceUnavailable: 503
limits:
  - name: kube-apiserver Service CRUD (control plane)
    scope: priority-level
    metric: concurrent_requests
    limit: 'governed by API Priority and Fairness PriorityLevelConfiguration'
  - name: Service data-plane throughput
    scope: load-balancer
    metric: varies
    limit: 'depends on cloud-provider LB type / instance class — see provider quotas'
policies:
  - name: APF on control plane
    description: Service create/update/delete/list operations against kube-apiserver are subject to API
      Priority and Fairness; high-frequency reconciliation should use watches and informers rather than
      polling.
  - name: kube-proxy / iptables / IPVS scaling
    description: ClusterIP and NodePort throughput depends on the kube-proxy mode (iptables, IPVS, nftables)
      and node kernel networking; not a documented numeric limit.
  - name: External LB quotas
    description: Type=LoadBalancer Services provision a cloud-provider LB; per-LB and per-account quotas
      (NLB connection limits, ALB rule limits, GCLB forwarding-rule limits) apply.
  - name: Endpoint slice fan-out
    description: Very large Services (thousands of endpoints) should rely on EndpointSlice (default in
      modern Kubernetes) rather than the legacy Endpoints object to avoid kube-apiserver load.
maintainers:
  - FN: Kin Lane
    email: kin@apievangelist.com