-
Bug
-
Resolution: Done
-
Major
-
None
-
4.19.z
-
Quality / Stability / Reliability
-
False
-
-
None
-
None
-
None
-
None
-
None
-
None
-
None
-
None
-
None
-
None
-
None
-
None
-
None
-
None
Description of problem:
When reading CPU node consumption using "oc adm top node" on an idle cluster, it shows high CPU consumption sometimes because of steal time values on IBM s390x clusters. Although steal time is not the time when CPUs are being used to run workload. It confuses user as to which pod is consuming the CPU, although none of the pod is doing so.
Version-Release number of selected component (if applicable):
How reproducible:
Simulate steal time on bastion: # vmcp set share <user> abs 5% limithard Read CPU consumption: # oc adm top node worker-1 NAME CPU(cores) CPU% MEMORY(bytes) MEMORY% worker-1 1879m 125% 4553Mi 66% Although workers were idle we see high CPU consumption due to high steal time.
Actual results:
steal time is added to CPU usage.
Expected results:
steal time should not be added to CPU usage.
Additional info:
This problem has been fixed in upstream kernel: https://git.kernel.org/pub/scm/linux/kernel/git/tj/cgroup.git/commit/?h=for-6.14-fixes This bug has been created to document the problem and do a backport of the above patch in RHCOS kernel.
