kubectl and descheduler CPU utilisation values are vastly different

[duronrulez]

We've notice that the values for CPU utilisation between kubectl top nodes and the descheduler are vastly different.

I0423 10:01:19.375155    7053 lownodeutilization.go:156] Node "XXX" is appropriately utilized with usage: api.ResourceThresholds{"cpu":44.05815423514539, "memory":38.10868466231932, "pods":18.96551724137931}

while kubectl top nodes reports

NAME                                        CPU(cores)   CPU%   MEMORY(bytes)   MEMORY%
XXX    7946m        100%   10365Mi         81%

The values are vastly different and the ones reported by kubectl seem to be the ones that match the node usage. Is there any reason why the 2 values will missmatch so much?

[seanmalloy]

/close

@duronrulez currently the LowNodeUtilization descheduler strategy uses container requests not the actual CPU and memory utilization. Yes, the word utilization is in the name of the strategy, and this is confusing. There is a one liner in the README that describes this, Currently, pods request resource requirements are considered for computing node resource utilization..

See #7 for a discussion around utilization vs allocation terminology confusion. See #225 for a discussion on adding a new strategy that is based on actual container utilization metrics.

[k8s-ci-robot]

@seanmalloy: Closing this issue.

In response to this:

/close

@duronrulez currently the LowNodeUtilization descheduler strategy uses container requests not the actual CPU and memory utilization. Yes, the word utilization is in the name of the strategy, and this is confusing. There is a one liner in the README that describes this, Currently, pods request resource requirements are considered for computing node resource utilization..

See #7 for a discussion around utilization vs allocation terminology confusion. See #225 for a discussion on adding a new strategy that is based on actual container utilization metrics.

Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes/test-infra repository.

[seanmalloy]

/kind feature