Hi Everyone,

I am facing a challenging networking and client management issue with our
on-prem Kafka cluster and would appreciate your insights.

Currently, we have various clients (producers and consumers) connecting to
our on-prem Kafka cluster from external OpenShift (OCP) and Kubernetes
(K8s) environments. The core problem is that our Kafka cluster only sees
the K8s worker node's IP address rather than the originating pod's IP.

Occasionally, application teams deploy misconfigured pods and abandon them.
These rogue pods continuously trigger SASL_SSL handshake errors and
authentication
failures for hours at a time. This constant spamming heavily strains the
Kafka cluster's network.

Because we only see the worker node IP—which simultaneously routes traffic
for multiple other legitimate clients—I cannot simply block the IP or apply
network-level rate limits. Doing so would effectively block the entire
worker node and cause outages for other healthy applications.

What is the recommended approach or best practice to resolve this issue? Is
there a standard way to expose the actual pod IP to the on-prem Kafka
cluster, or are there alternative architectural/configuration patterns to
effectively isolate and throttle these misconfigured clients without
impacting others?

Thanks in advance for your time and suggestions.

Best regards,

M. Cagri Aktas

Reply via email to