JMX metrics endpoint is slow #7670
Replies: 1 comment
-
Hello Abhinav! Your issue is really similar to the issue we have with Strimzi Metrics. From what I've seen so far, this issue usually appears when the Kafka cluster has quite a lot of topics (10..20k), and the main problem is that metrics are prepared and exposed for each and every topic, enlarging the output drastically, up to ~33.96 MiB in our case (please see logs below). To me it also seems like we spend quite some time calculating/generating the output rather than downloading it (as it's localhost and not a remote server of any kind). The utilization patterns of the VM don't seem to change much during the handling of this request.
Most of the time these requests are failing after approximately 1 minute, like below:
I've found some information on how to limit the metrics output here, but it seems to be relevant only to Strimzi Kafka Exporter, while the main issue to me seems to be with the metrics generated and exposed from Kafka process running on the Broker Pods. Is there any way to exclude some topics from these metrics on Pods, so that preparation and exposition will be done only for the topics we are interested in? |
Beta Was this translation helpful? Give feedback.
Uh oh!
There was an error while loading. Please reload this page.
-
There was some metrics mismatch on grafana compared to what we actual saw on the cli (like number of brokers up), to investigate further, we exec into the pod and ran command curl
localhost:9404/metrics
, it gave some data but the result came after like 5-6 mins, what should we scale, to get metrics faster?Beta Was this translation helpful? Give feedback.
All reactions