CharlieIsAwesome
CharlieIsAwesome

Reputation: 83

Flink metric `flink_taskmanager_Status_JVM_CPU_Load{}` is not accurate in Grafana Dashboard

I have a simple Flink job, and the Kafka consumption rate is like 5,000/s.

But from what I observed, the CPU usage is too low. So I think if it's possible that, in this version, the Flink metrics is not accurate. Or maybe it's because I configured wrong PromQL?

Query for the Grafana chart:

flink_jobmanager_Status_JVM_CPU_Load{exported_job='${jobmanager_prome_job}'}

enter image description here

Update task manager chart.

enter image description here

Upvotes: 3

Views: 1052

Answers (1)

Arvid Heise
Arvid Heise

Reputation: 3634

Judging from your title you actually want to check the load on the taskmanager (which is more interesting in most cases anyways). Could you adjust your Grafana chart to use the taskmanager instead of the jobmanager and report back? Note that CPU is at 1 only if all cores are fully saturated on that machine. So if you have 10 cores in your machine, you'd get at most 0.1 if you run with parallelism 1.

Upvotes: 1

Related Questions