Clusters, topics and consumer groups

A tour of the screens you use every day to watch and operate a Kafka cluster.

Cluster dashboard

The overview of a cluster shows the controller, brokers (missing and unreachable ones included), topics, partitions, offline and under-replicated partitions, the Kafka version and the cluster mode, with alerts when something is wrong, for example when the Kafka authorizer is disabled.

Cluster overview with broker health cards
Cluster overview: brokers, partitions and alerts at a glance.

Topics

Topics list, advanced view
Topics list, advanced view.
Partitions of a topic with leaders and replicas
Partitions, leaders and replicas of a topic.

Consumer groups

Consumer group lag by topic
Lag of a consumer group by topic.

AI assistant (beta)

When turned on (KCM_ASSISTANT_ENABLED, off by default), a consumer group gets a plain English analysis of its lag, members and offsets over a chosen period of up to 7 days, with the history of past analyses. The model runs on your own Ollama server: no data leaves your network. The assistant explains, it never changes anything. See the configuration reference.

ACLs, Kafka users and quotas

The Security page of a cluster manages ACLs (create, delete, filter), SCRAM users, AWS MSK IAM users and client quotas. All of it needs ADMIN on the cluster. On Confluent Cloud clusters, see Confluent Cloud IAM.

Metrics

Starter With metrics turned on for a cluster, KCM computes the traffic of each topic every 15 minutes and shows messages per minute over the last hour, 24 hours, 7 days, 30 days or a custom range. Exclude topics from metrics with regular expressions that match the whole topic name (internal topics are always excluded).

Diagnostics and operations

Starter For cluster ADMINs:

What-if simulation

Simulate the failure of brokers or of a whole rack, with or without a preferred leader election, and compare the simulated state with the real one: which partitions would go offline or under-replicated. Simulations never touch the cluster; results are kept 30 minutes. Not available on Confluent Cloud clusters.