actually in isolated mode we only care about RequestHandlerAvgIdlePercent. The other two metrics are useless.
On Mon, Aug 25, 2025 at 4:25 PM Tony Tang <zt...@confluent.io> wrote: > Hey Justine, > > Thanks for the review. > > > > *In other words, we have a per node metric that combines all node > pools(RequestHandlerAvgIdlePercent) and then a metric per the specified > pools(broker vs controller)? *yes > > in combined mode those three metrics are different (1 for all pools, 2 for > separate pools) > in isolated mode they're identical or we can make the broker pool metric > invalid in controller node/controller pool metric invalid in broker node. I > prefer the second option. > > Best, > Zhiyan > > > On Fri, Aug 22, 2025 at 12:02 PM Justine Olshan > <jols...@confluent.io.invalid> wrote: > >> Hey Tony, >> >> Thanks for the KIP. >> I might have missed it but can you clarify how these metrics work in >> combined vs isolated mode? >> Will both modes have the metrics, but for combined mode, we will have all >> three metrics on the nodes? >> >> In other words, we have a per node metric that combines all node pools >> (RequestHandlerAvgIdlePercent) and then a metric per the specified pools >> (broker vs controller)? >> >> Thanks, >> Justine >> >> On Wed, Aug 20, 2025 at 8:52 AM Kevin Wu <kevin.wu2...@gmail.com> wrote: >> >> > Hi Tony, >> > >> > Thanks for the KIP! I agree that we should keep and fix the existing >> > metric. Adding a broker and controller-specific metric makes it more >> > explicit for operators who monitor this metric. One thing to keep in >> mind >> > when fixing the existing `RequestHandlerAvgIdlePercent` metric is that >> we >> > should not couple its denominator value with the mode of kafka being run >> > (isolated or combined). We could very well introduce another thread >> pool in >> > the future. >> > >> > Best, >> > Kevin Wu >> > >> > On Tue, Aug 19, 2025 at 5:16 PM Tony Tang <zt...@confluent.io.invalid> >> > wrote: >> > >> > > Hello everyone, I'd like to discuss a KIP regarding adding new metrics >> > for >> > > request handler pool and fixing the anomaly of a JMX metric. KIP link >> > > < >> > > >> > >> https://cwiki.apache.org/confluence/display/KAFKA/KIP-1207%3A+Fix+anomaly+of+JMX+metrics+RequestHandlerAvgIdlePercent+in+kraft+combined+mode >> > > > >> > > >> > > Thanks, >> > > Tony >> > > >> > >> >
