9ec39a1da0
Pods with namespace 'kube-system', or labeled with 'app.starlingx.io/component=platform' are identified as 'platform'. These have isolated cpu affinity cpuset when kubelet cpu-manager 'static' policy is configured. This decouples performance of application pods from the platform. The new patch Identify-platform-pods-based-on-pod-or-namespace-labels patch is used to identify platform pods using above criteria for Kubernetes 1.25 and later. The hard-coded list of platform namespaces is no longer required. The original hard-coded list is still kept for Kubernetes 1.24 so we can phase in this change. This will also fix the Kubernetes make test fail issue after merging the review https://review.opendev.org/c/starlingx/integ/+/907637 This change also include the missed out patch kubeadm-reduce-UpgradeManifestTimeout.patch for k8s 1.29. Test Plan: PASS: Run all Kubelet, kubeadm, kubectl make tests for affected code. PASS: All affected versions of kubernetes package build successfully. PASS: Create a pod with the platform label. Pod is classified as a platform pod. PASS: Create a pod without the platform label but in a namespace with the platform label. Pod is classified as a platform pod. PASS: Create a pod without the platform label and in a namespace without the platform label. Pod is not classified as a platform pod. PASS: Verify correct cpuset affinity of platform vs application pods after host lock/unlock PASS: Verify kube-system namespace pods are correctly identified as Platform Closes-Bug: 2058042 Story: 2010612 Task: 49910 Change-Id: I0cc05c2e9057b64f9191317e72a1bf4d7b8d0771 Signed-off-by: Boovan Rajendran <boovan.rajendran@windriver.com>
34 lines
1.3 KiB
Diff
34 lines
1.3 KiB
Diff
From 341756eb57cea280d96e74f56d02033402cee133 Mon Sep 17 00:00:00 2001
|
|
From: Ramesh Kumar Sivanandam <rameshkumar.sivanandam@windriver.com>
|
|
Date: Fri, 15 Mar 2024 03:49:15 -0400
|
|
Subject: [PATCH] kubeadm: reduce UpgradeManifestTimeout
|
|
|
|
This modifies kubeadm UpgradeManifestTimeout from 5 minutes default
|
|
to 3 minutes to reduce the unnecessary delay in retries during
|
|
kubeadm-upgrade-apply failures.
|
|
|
|
The typical control-plane upgrade of static pods is 75 to 85 seconds,
|
|
so 3 minutes gives adequate buffer to complete the operation.
|
|
|
|
Signed-off-by: Ramesh Kumar Sivanandam <rameshkumar.sivanandam@windriver.com>
|
|
---
|
|
cmd/kubeadm/app/phases/upgrade/staticpods.go | 2 +-
|
|
1 file changed, 1 insertion(+), 1 deletion(-)
|
|
|
|
diff --git a/cmd/kubeadm/app/phases/upgrade/staticpods.go b/cmd/kubeadm/app/phases/upgrade/staticpods.go
|
|
index 540c1549fff..b40bc76f0fc 100644
|
|
--- a/cmd/kubeadm/app/phases/upgrade/staticpods.go
|
|
+++ b/cmd/kubeadm/app/phases/upgrade/staticpods.go
|
|
@@ -46,7 +46,7 @@ import (
|
|
|
|
const (
|
|
// UpgradeManifestTimeout is timeout of upgrading the static pod manifest
|
|
- UpgradeManifestTimeout = 5 * time.Minute
|
|
+ UpgradeManifestTimeout = 3 * time.Minute
|
|
)
|
|
|
|
// StaticPodPathManager is responsible for tracking the directories used in the static pod upgrade transition
|
|
--
|
|
2.25.1
|
|
|