1
0
mirror of https://github.com/openshift/openshift-docs.git synced 2026-02-05 12:46:18 +01:00

Merge pull request #98796 from kquinn1204/TELCODOCS-2269-1

Telcodocs 2269 1: Disable RPS - pods should "pay" for networking
This commit is contained in:
Alexandra Molnar
2025-09-12 13:26:20 +01:00
committed by GitHub

View File

@@ -563,6 +563,30 @@ In {product-title} {product-version}, the NUMA Resources Operator automatically
For more information, see xref:../scalability_and_performance/cnf-numa-aware-scheduling.adoc#cnf-managing-ha-nrop_numa-aware[Managing high availability (HA) for the NUMA-aware scheduler].
[id="ocp-4-20-receive-packet-steering-disabled_{context}"]
==== Receive Packet Steering (RPS) is now disabled by default
With this release, Receive Packet Steering (RPS) is no longer configured when Performance Profile is applied. The RPS configuration affects containers that perform networking system calls, such as send, directly within latency-sensitive threads. To avoid latency impacts when RPS is not configured, move networking calls to helper threads or processes.
The previous RPS configuration resolved latency issues at the expense of overall pod kernel networking performance. The current default configuration promotes transparency by requiring developers to address the underlying application design instead of obscuring performance impacts.
To revert to the previous behavior, add the `performance.openshift.io/enable-rps` annotation to the PerformanceProfile manifest:
[source,yaml]
----
apiVersion: performance.openshift.io/v2
kind: PerformanceProfile
metadata:
name: example-performanceprofile
annotations:
performance.openshift.io/enable-rps: "enable"
----
[NOTE]
====
This action restores the prior functionality at the cost of globally reducing networking performance for all pods.
====
[id="ocp-release-notes-security_{context}"]
=== Security