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: Kubecost providerId: kubecost 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: false tags: - Rate Limiting - Cloud Cost - Kubernetes description: Kubecost's monitoring APIs (Allocation, Assets, Cloud Cost, Budget, Forecast, Savings) are served by the Kubecost cost-model in-cluster, so request rate is bound by the consumer's own deployment rather than a vendor-imposed numeric quota. Apptio does not publish per-second/per-minute API rate limits for Enterprise Cloud SaaS. sources: - https://docs.kubecost.com/apis/monitoring-apis/api-allocation - https://www.apptio.com/pricing?src=kc-com notes: No vendor-published numeric API rate limits. Consumers self-host the cost-model service; effective limits are governed by the deployment's own resource requests. responseCodes: throttled: 429 serviceUnavailable: 503 limits: - name: Self-hosted cost-model scope: cluster metric: varies limit: 'governed by self-hosted cost-model resource limits' - name: Enterprise Cloud SaaS scope: tenant metric: varies limit: 'see vendor; not publicly documented' policies: - name: ETL Caching description: The Allocation API is integrated with the Kubecost ETL caching layer; expensive queries are cached and CSV pipelines used for large clusters. - name: Backoff description: Use exponential backoff with jitter on 429/503 responses; honor Retry-After when present. - name: Window Selection description: Prefer narrower time windows and aggregated steps to reduce work in the cost-model; avoid overlapping concurrent queries against the same window. maintainers: - FN: Kin Lane email: kin@apievangelist.com