APIM Self-hosted gateway multi instance rate limit sync not working

Question

APIM Self-hosted gateway multi instance rate limit sync not working

Madhu Hiriyadka 0

Hi

I have deployed an API Managment - with self hosted gateway. I am running this as Kubernetes service, with multiple pods running this APIM instance. I have set rate limit policy for one of the POST api. It does not seems to work as expected.

For e.g.:

Locked Question. You can vote on whether it's helpful, but you can't add comments or replies or follow the question.

Christos Panagiotidis 3,301 Reputation points

2026-07-29T12:34:35.58+00:00

You have multiple Kubernetes pods running the same APIM self-hosted gateway, but each pod appears to enforce its own rate limit. Cross-instance counters require the gateways to discover one another and exchange UDP traffic. Create a headless Kubernetes Service that resolves every gateway pod, set the gateway’s neighborhood.host to that service name, and allow bidirectional UDP 4291 for heartbeat discovery and UDP 4290 for rate-limit synchronization across pods. Confirm NetworkPolicy, CNI, and any host firewall permit those flows, and ensure all pods use the same self-hosted gateway resource and a v2 image. Check logs for neighborhood discovery and restart the deployment after changing settings. Synchronization is local to those self-hosted instances; it does not merge counters with the managed gateway or a different gateway resource. Counters are exchanged asynchronously, so brief bursts can slightly exceed the nominal limit. Use a centralized backend if an exact transactional global quota is mandatory.

Christos Panagiotidis 3,301 Reputation points

2026-07-29T12:34:35.58+00:00

You have multiple Kubernetes pods running the same APIM self-hosted gateway, but each pod appears to enforce its own rate limit. Cross-instance counters require the gateways to discover one another and exchange UDP traffic. Create a headless Kubernetes Service that resolves every gateway pod, set the gateway’s neighborhood.host to that service name, and allow bidirectional UDP 4291 for heartbeat discovery and UDP 4290 for rate-limit synchronization across pods. Confirm NetworkPolicy, CNI, and any host firewall permit those flows, and ensure all pods use the same self-hosted gateway resource and a v2 image. Check logs for neighborhood discovery and restart the deployment after changing settings. Synchronization is local to those self-hosted instances; it does not merge counters with the managed gateway or a different gateway resource. Counters are exchanged asynchronously, so brief bursts can slightly exceed the nominal limit. Use a centralized backend if an exact transactional global quota is mandatory.