Cluster/broker throughput shows 0 B/s despite per-topic rates being populated

[Sangyong-Jeon]

Describe the bug

On the cluster dashboard and brokers list, cluster-wide and per-broker "Bytes In/Out" show 0 B/s, even though each topic's throughput is displayed correctly on the topic details page. This happens when the broker is scraped via JMX (default METRICS_TYPE=JMX) and does not surface the topic-less kafka.server:type=BrokerTopicMetrics,name=BytesInPerSec aggregate in the scraped metric set.

Set up

• kafka-ui: latest (also reproduces on current main)
• Broker: Confluent cp-kafka 7.9.7 (Kafka 3.9), JMX enabled
• METRICS_TYPE: JMX (default)

Steps to Reproduce

1. Connect kafka-ui to a cp-kafka broker over JMX with live traffic.
2. Open the cluster dashboard / brokers list.
3. Observe cluster & broker "Bytes In/Out" = 0, while topic details show non-zero throughput.

Root cause

• IoRatesMetricsScanner#updateBrokerIOrates populates brokerBytes*FifteenMinuteRate only from the topic-less BrokerTopicMetrics metric (labels.size() == 1).
• InternalClusterState derives cluster bytesInPerSec/bytesOutPerSec purely as the sum of the per-broker map; when that map is empty → reduce(...).orElse(null) → 0/null in the UI.
• On this broker the topic-less aggregate is absent from the scraped set (only per-topic metrics are present), so the per-broker map stays empty and cluster throughput collapses to 0 — even though topicBytes*PerSec is fully populated (the topic details page shows correct rates).
• Verified: the broker DOES register kafka.server:type=BrokerTopicMetrics,name=BytesInPerSec (read via kafka.tools.JmxTool), but it does not appear in the JMX-scraped set kafka-ui builds. Most likely it is dropped during the JMX→Prometheus conversion because the same metric family carries inconsistent label sets: global {name} vs per-topic {name,topic}.

Expected behavior

Cluster/broker throughput should reflect actual traffic. Since per-topic rates are available and bytes in/out are additive across topics, the all-topics aggregate can be computed as their sum.

Proposed fix

Fall back to summing topicBytes*PerSec when the per-broker aggregate map is empty (mathematically equal to the all-topics broker aggregate).

---

Suggested labels (for triage): scope/backend, type/bug, area/internal

[kapybro]

AI Summary

The issue describes a bug in kafka-ui where cluster-wide and per-broker "Bytes In/Out" metrics show 0 B/s despite per-topic throughput being correctly displayed. This occurs because the topic-less BrokerTopicMetrics aggregate is missing from the JMX-scraped metrics, causing the per-broker map to remain empty and the cluster throughput to incorrectly sum to zero. The proposed fix is to fall back to summing the per-topic topicBytes*PerSec metrics when the per-broker aggregate map is empty.

[github-actions]

Hi Sangyong-Jeon! 👋

Welcome, and thank you for opening your first issue in the repo!

Please wait for triaging by our maintainers.

As development is carried out in our spare time, you can support us by sponsoring our activities or even funding the development of specific issues.
Sponsorship link

If you plan to raise a PR for this issue, please take a look at our contributing guide.