CPU utilization is the proportion of processor capacity occupied by executing work during a measurement interval. In Kafka operations, sustained high utilization can prevent requests from being processed, while brief spikes may only reduce performance. section 15.1
Interpret utilization as a trend rather than a single reading. Partition rebalancing can temporarily raise broker resource consumption without indicating a lasting problem; persistently high CPU, memory, network, or storage usage more strongly suggests growth and a need to scale. section 15.1
Use resource-utilization measurements to guide operational decisions:
- Track CPU utilization alongside memory, network bandwidth, and storage rather than treating CPU as an isolated signal. section 15.1 section 15.5.1
- Account for deployment context: physical hardware requires hardware monitoring, virtual machines add resource contention, and Kubernetes requires resource tracking with monitoring tools. section 15.6 section 15.5.1
- Set meaningful thresholds and smooth metrics so alerts identify sustained problems instead of normal temporary variation. section 15.6
Reading a CPU spike
A broker shows high CPU during partition rebalancing. Treat the spike as potentially temporary, then investigate scaling if CPU remains high after rebalancing or repeatedly prevents requests from being processed. section 15.1
