Azure Kubernetes Service · Rate Limits

Azure Kubernetes Service Rate Limits

AKS uses a token-bucket throttle on its resource provider APIs (control plane). Buckets have a fixed burst size and a sustained refill rate; throttled requests return HTTP 429 with Retry-After. Cluster-creation quotas are enforced per subscription per region, with subscription-type-specific defaults (EA 100, PAYG 10, Free Trial 3) and self-service raise paths in the Quotas blade.

16 Limits Throttle: 429
ContainersKubernetesRate Limiting

Limits

LIST ManagedClusters (subscription) subscription
requests_burst
500
Burst 500; refill 1 request/second.
LIST ManagedClusters (resource group) resourceGroup
requests_burst
60
Burst 60; refill 1 request/second.
PUT AgentPool agentPool
requests_burst
20
Burst 20; refill 1 request/minute.
PUT ManagedCluster managedCluster
requests_burst
20
Burst 20; refill 1 request/minute.
GET ManagedCluster managedCluster
requests_burst
60
Burst 60; refill 1 request/second.
GET Operation Status subscription
requests_burst
200
Burst 200; refill 2 requests/second.
All Other AKS APIs subscription
requests_burst
60
Burst 60; refill 1 request/second.
AKS clusters per subscription per region (EA) subscription/region
clusters
100
Default for new EA subscriptions; max self-service 1,000.
AKS clusters per subscription per region (PAYG / CSP / MSDN) subscription/region
clusters
10
Default; max self-service 100.
AKS clusters per subscription per region (Free Trial) subscription/region
clusters
3
Hard cap; cannot be raised.
Maximum clusters per subscription globally subscription
clusters
5000
Maximum nodes per cluster (VMSS + Standard LB) cluster
nodes
5000
Maximum nodes per node pool nodePool
nodes
1000
Maximum node pools per cluster cluster
nodePools
100
Maximum pods per node (Azure CNI) node
pods
250
Maximum load-balanced services per cluster cluster
services
300

Policies

Token-bucket throttling
AKS RP APIs use token-bucket; burst then refill. 429 with Retry-After indicates the bucket is empty.
Backoff
Honor Retry-After (delay-seconds). Use exponential backoff with jitter; avoid bursty list-loops in retry logic.
Quota raise
Use the Azure portal Quotas page or REST API to request cluster quota increases. Free Trial subscriptions cannot raise quotas.
ManagedClusters and AgentPools tracked separately
Even within the same AKS cluster, ManagedClusters and AgentPools have independent buckets.

Sources