Kube-scheduler is not choosing prioritized node by Telemetry Aware Scheduling about platform-aware-scheduling HOT 3 CLOSED

kube-scheduler logs

I1027 19:48:52.759571 1 trace.go:205] Trace[1755174988]: "Scheduling" namespace:default,name:demo-app-55b7cffd5d-hn476 (27-Oct-2022 19:48:52.597) (total time: 161ms): Trace[1755174988]: ---"Computing predicates done" 91ms (19:48:52.689) Trace[1755174988]: ---"Prioritizing done" 68ms (19:48:52.757) Trace[1755174988]: [161.160364ms] [161.160364ms] END

from platform-aware-scheduling.

The prioritization is not guaranteed to result in the Pod ending up scheduled into the node which TAS finds best suited.

The kubernetes scheduler takes into account many factors while considering the best suited node, and the priority score from TAS can be characterized as weak. One can fiddle with the weight in the scheduler configuration, to increase the impact of TAS.

You can trace the scoring decisions in the kube-scheduler by increasing the verbosity of kube-scheduler to at least 10 (-v=10 to the command line) . See: https://github.com/kubernetes/kubernetes/blob/5de293c425ec0cde9a2cf8917a69329bebcd0bbd/pkg/scheduler/schedule_one.go#L732

from platform-aware-scheduling.

Thank you for your answer.

from platform-aware-scheduling.