-
Task
-
Resolution: Unresolved
-
Undefined
-
None
-
None
-
None
-
None
-
Quality / Stability / Reliability
-
False
-
-
False
-
None
-
None
-
None
-
None
-
None
Plan a Scale/Pref/Bechmark test for PSI with OpenShift 4.21 with k8s 1.34.
rhn-support-qili We'll discuss and update this. For now its created for tracking
High level plan by rhn-support-qili on 2025-10-22
- Understand and evaluate on the Google's performance test on GKE
- Generate our own automation assets for OpenShift
- Create a cluster with equivalent resource as the upstream test on GKE
- Run test under CPU, memory, and I/O stress with and without PSI to
- Verify that enabling PSI on nodes does not introduce excessive CPU or memory usage in the Linux kernel.
- Verify that the collection of PSI metrics by the kubelet does not cause a significant increase in its own CPU or memory consumption.
- Analyze results and generate benchmark report on OpenShift
- Document for users to know what they will get by using PSI
Reference:
Google's performance test on GKE https://docs.google.com/document/d/15yfzC6eD4faylW7Vcd_rYGL7DXtCjXFO
slack context https://redhat-internal.slack.com/archives/CK1AE4ZCK/p1760369400160509?thread_ts=1758543815.463189&cid=CK1AE4ZCK
- is related to
-
OCPNODE-3806 Node Team to reconsider enabling PSI Metrics to help CNV Descheduler
-
- Closed
-