Is this a known issue? If not I can create a JIRA.

 

For example, here’s a screenshot of the process-total metric for one processor node and task under stream-processor-node-metrics. Processor-total goes down over time when it shouldn’t. Same thing for total metrics under stream-metrics, stream-*-state-metrics, stream-task-metrics. Total metrics look okay for producer and consumers that I’ve looked at.

 

Technically most streams total metrics aren’t documented (https://docs.confluent.io/current/streams/monitoring.html), so maybe they’re not officially supported but they are mentioned in KIP 187 https://cwiki.apache.org/confluence/display/KAFKA/KIP-187+-+Add+cumulative+count+metric+for+all+Kafka+rate+metrics.

 

cid:image001.png@01D429B1.2CED5AD0