XML

Word

Printable

Type: Epic
Resolution: Unresolved
Priority: Critical
Fix Version/s: None
Affects Version/s: None
Component/s: CNV Documentation, CNV Install, Upgrade and Operators, CNV Virt-Node
Labels:

Epic Name:
improve-contention-related-metrics
Story Points:
77
Acceptance Criteria:
Hide

Enable schedstats=enable by default (to be confirmed that we want this)

Fix kubevirt_vmi_vcpu_wait_seconds_total documentation to be IO specific

Add documentation for kubevirt_vmi_vcpu_delay_seconds_total to be CPU scheduling specific

Make sure the metric is displayed in a dashboard

Explore if Pressure Stall Informations are valueable to detect contended workloads

Comparison CPU steal (guest) VS vCPU wait (qemu) VS PSI (cgroup)
Show
Enable schedstats=enable by default (to be confirmed that we want this) Fix kubevirt_vmi_vcpu_wait_seconds_total documentation to be IO specific Add documentation for kubevirt_vmi_vcpu_delay_seconds_total to be CPU scheduling specific Make sure the metric is displayed in a dashboard Explore if Pressure Stall Informations are valueable to detect contended workloads Comparison CPU steal (guest) VS vCPU wait (qemu) VS PSI (cgroup)
Epic Status:
In Progress
Hierarchy Progress Bar:

67% To Do, 33% In Progress, 0% Done
Ready-Ready:

doc-ready
Target Version:

CNV v4.19.0, CNV v4.20.0

Goal

Improve vCPU contention related metrics by

make them work out of the box
add documentation for existing metric kubevirt_vmi_vcpu_wait_seconds_total to https://docs.redhat.com/en/documentation/openshift_container_platform/4.16/html/virtualization/monitoring#virt-promql-vcpu-metrics_virt-prometheus-queries
kubevirt_vmi_vcpu_wait_seconds_total is documented, but the docs lack, that the metric is IO specific (_delay metric is CPU scheduling specific)

In addition: A spike to explore how valueable PSI metrics are for VMs, if they complement vCPU ready.

User Stories

As a cluster administrator, I want know when there are vCPU performance issues with my VM, so that I can take action

List of things not included in this epic, to alleviate any doubt raised during the grooming process.

is documented by

CNV-45694 Add documentation for kubevirt_vmi_vcpu_delay_seconds_total

is related to

CNV-46221 Add to the Red Hat docs a link to the CNV metrics documentation

relates to

VIRTSTRAT-65 Load Aware balancing

1.	upstream roadmap issue	New	Unassigned
2.	upstream design	New	Unassigned
3.	upstream documentation	New	Unassigned
4.	upgrade consideration	New	Unassigned
5.	CEE/PX summary presentation	New	Unassigned
6.	test plans in polarion	New	Unassigned
7.	automated tests	New	Unassigned
8.	downstream documentation merged	New	Unassigned