Kafka and KafkaNodePool resources in kafka.strimzi.io have curated tables and drawers. The Kafka drawer shows the reported Kafka and metadata versions, with the declared version alongside when they differ (Strimzi keeps the old reported version while an upgrade has not succeeded), the metadata mode Strimzi reports, the cluster ID, the node pools Strimzi lists in status.kafkaNodePools (each links to its pool), and a listener table that joins declared listeners (spec.kafka.listeners) with the addresses Strimzi reports. The node pool drawer shows the owning Kafka (from the strimzi.io/cluster label), roles, assigned and desired replicas, node IDs, declared storage volumes and resource requests/limits. Listener certificates and authentication secrets are never read, and free-form condition messages stay in YAML because they can echo configuration.
Strimzi marks a failed Kafka reconciliation with NotReady=True, never Ready=False, so Radar reads that condition. A Kafka’s status badge reads Ready only when the operator has processed the current generation and reports Ready=True; it reads Not ready for NotReady=True, Creating while a new cluster’s first reconciliation is still running, and Reconciliation paused, Status stale or Status unknown (never healthy) for a paused, outdated or unreported status. A Warning condition is shown as a warning, not a failure: the Kafka table’s Warnings column lists their reasons, as Strimzi’s own printer columns do, and the drawer of a Ready Kafka shows them in a banner, because some (such as KafkaStorage on a single node with ephemeral storage) mean data loss on restart. Ready means Strimzi reconciled the resource; it does not establish that clients can reach the brokers.
Strimzi never writes Ready or NotReady on a KafkaNodePool. It writes the pool’s status (node IDs, roles, replicas and any Warning conditions) partway through reconciling the owning Kafka, before brokers roll. A pool therefore reads Accepted when that status matches its current generation, Not applied when its current status reports different replicas or roles than its spec (Strimzi refuses a scale-down or role removal while those brokers hold partitions, and records the reason as a ScaleDownPreventionCheck warning on the Kafka only), Warnings when it carries a Warning condition (for example KafkaStorage, meaning a storage change was ignored), Status stale when it is outdated, and Not reconciled when Strimzi has never written it (such as a pool whose strimzi.io/cluster label names no Kafka). None of these say whether the brokers are healthy; that is the owning Kafka’s status. Strimzi honours strimzi.io/pause-reconciliation only on the Kafka, so Radar ignores it on a pool. A pool’s assigned replica count is the number of node IDs Strimzi allocated, not the number of running broker Pods.
A NotReady=True Kafka whose status matches its current generation produces one warning in Issues (including the API and MCP). Paused, stale, missing, creating or terminating status produces none, the condition message is left out of the issue, and KafkaNodePools produce none. Radar Cloud’s default integration-read role does not grant kafkas or kafkanodepools because their specs can carry credentials. Both kinds are watched at startup when installed, so no extra Helm setting is needed beyond the existing rbac.crdGroups.strimzi.
Checked against a running Strimzi 1.3.1 operator, which serves only kafka.strimzi.io/v1. Earlier releases serving v1beta2 are read with the same field paths but were not run.
Not covered: relationships beyond the exact references above (for example Pods managed through StrimziPodSet), KafkaTopic, KafkaUser, and any connection to Kafka itself.