Fix cpu share issues on systems with large amounts of cpu(cpucore>256)
Signed-off-by: zhaoxiaohu <zhaoxiaohu@kuaishou.com>
This commit is contained in:
zhaoxiaohu
liuxu
parent
ef0ef3875b
commit
d8397b11a3
@ -0,0 +1,104 @@
|
||||
From 051b66b82d8cc76eca0da38e44ae0e7dd391ba09 Mon Sep 17 00:00:00 2001
|
||||
From: zhaoxiaohu <zhaoxiaohu@kuaishou.com>
|
||||
Date: Tue, 27 Aug 2024 16:04:40 +0800
|
||||
Subject: [PATCH] Fix cpu share issues on systems with large amounts of cpu
|
||||
|
||||
On systems where the calculated cpu shares results in a value above the
|
||||
max value in linux, containers getting that value are unable to start.
|
||||
This occur on systems with 300+ cpu cores, and where containers are
|
||||
given such a value.
|
||||
|
||||
This issue was fixed for the pod and qos control groups in the similar
|
||||
cm.MilliCPUToShares that also has tests verifying the behavior. Since
|
||||
this code already has an dependency on kubelet/cm, lets reuse that code
|
||||
instead.
|
||||
|
||||
Reference: https://github.com/kubernetes/kubernetes/pull/106570/commits/de0ece541c7fd885a32f1562342fe85535fc11f5
|
||||
|
||||
Signed-off-by: zhaoxiaohu <zhaoxiaohu@kuaishou.com>
|
||||
Signed-off-by: Odin Ugedal <odin@uged.al>
|
||||
Signed-off-by: yuwang@kuaishou.com
|
||||
---
|
||||
pkg/kubelet/kuberuntime/helpers_linux.go | 17 -----------------
|
||||
pkg/kubelet/kuberuntime/helpers_unsupported.go | 5 -----
|
||||
.../kuberuntime/kuberuntime_container_linux.go | 7 ++++---
|
||||
3 files changed, 4 insertions(+), 25 deletions(-)
|
||||
|
||||
diff --git a/pkg/kubelet/kuberuntime/helpers_linux.go b/pkg/kubelet/kuberuntime/helpers_linux.go
|
||||
index 204bc4e9..4257a014 100644
|
||||
--- a/pkg/kubelet/kuberuntime/helpers_linux.go
|
||||
+++ b/pkg/kubelet/kuberuntime/helpers_linux.go
|
||||
@@ -19,9 +19,6 @@ limitations under the License.
|
||||
package kuberuntime
|
||||
|
||||
const (
|
||||
- // Taken from lmctfy https://github.com/google/lmctfy/blob/master/lmctfy/controllers/cpu_controller.cc
|
||||
- minShares = 2
|
||||
- sharesPerCPU = 1024
|
||||
milliCPUToCPU = 1000
|
||||
|
||||
// 100000 is equivalent to 100ms
|
||||
@@ -29,20 +26,6 @@ const (
|
||||
minQuotaPeriod = 1000
|
||||
)
|
||||
|
||||
-// milliCPUToShares converts milliCPU to CPU shares
|
||||
-func milliCPUToShares(milliCPU int64) int64 {
|
||||
- if milliCPU == 0 {
|
||||
- // Return 2 here to really match kernel default for zero milliCPU.
|
||||
- return minShares
|
||||
- }
|
||||
- // Conceptually (milliCPU / milliCPUToCPU) * sharesPerCPU, but factored to improve rounding.
|
||||
- shares := (milliCPU * sharesPerCPU) / milliCPUToCPU
|
||||
- if shares < minShares {
|
||||
- return minShares
|
||||
- }
|
||||
- return shares
|
||||
-}
|
||||
-
|
||||
// milliCPUToQuota converts milliCPU to CFS quota and period values
|
||||
func milliCPUToQuota(milliCPU int64, period int64) (quota int64) {
|
||||
// CFS quota is measured in two values:
|
||||
diff --git a/pkg/kubelet/kuberuntime/helpers_unsupported.go b/pkg/kubelet/kuberuntime/helpers_unsupported.go
|
||||
index cc1e88a5..8f6da8f4 100644
|
||||
--- a/pkg/kubelet/kuberuntime/helpers_unsupported.go
|
||||
+++ b/pkg/kubelet/kuberuntime/helpers_unsupported.go
|
||||
@@ -17,8 +17,3 @@ limitations under the License.
|
||||
*/
|
||||
|
||||
package kuberuntime
|
||||
-
|
||||
-// milliCPUToShares converts milliCPU to CPU shares
|
||||
-func milliCPUToShares(milliCPU int64) int64 {
|
||||
- return 0
|
||||
-}
|
||||
diff --git a/pkg/kubelet/kuberuntime/kuberuntime_container_linux.go b/pkg/kubelet/kuberuntime/kuberuntime_container_linux.go
|
||||
index d7c22c86..25fb68ad 100644
|
||||
--- a/pkg/kubelet/kuberuntime/kuberuntime_container_linux.go
|
||||
+++ b/pkg/kubelet/kuberuntime/kuberuntime_container_linux.go
|
||||
@@ -28,6 +28,7 @@ import (
|
||||
"k8s.io/klog/v2"
|
||||
v1helper "k8s.io/kubernetes/pkg/apis/core/v1/helper"
|
||||
kubefeatures "k8s.io/kubernetes/pkg/features"
|
||||
+ "k8s.io/kubernetes/pkg/kubelet/cm"
|
||||
kubecontainer "k8s.io/kubernetes/pkg/kubelet/container"
|
||||
"k8s.io/kubernetes/pkg/kubelet/qos"
|
||||
)
|
||||
@@ -69,11 +70,11 @@ func (m *kubeGenericRuntimeManager) generateLinuxContainerConfig(container *v1.C
|
||||
// API server does this for new containers, but we repeat this logic in Kubelet
|
||||
// for containers running on existing Kubernetes clusters.
|
||||
if cpuRequest.IsZero() && !cpuLimit.IsZero() {
|
||||
- cpuShares = milliCPUToShares(cpuLimit.MilliValue())
|
||||
+ cpuShares = int64(cm.MilliCPUToShares(cpuLimit.MilliValue()))
|
||||
} else {
|
||||
- // if cpuRequest.Amount is nil, then milliCPUToShares will return the minimal number
|
||||
+ // if cpuRequest.Amount is nil, then MilliCPUToShares will return the minimal number
|
||||
// of CPU shares.
|
||||
- cpuShares = milliCPUToShares(cpuRequest.MilliValue())
|
||||
+ cpuShares = int64(cm.MilliCPUToShares(cpuRequest.MilliValue()))
|
||||
}
|
||||
lc.Resources.CpuShares = cpuShares
|
||||
if memoryLimit != 0 {
|
||||
--
|
||||
2.33.0
|
||||
|
||||
@ -3,7 +3,7 @@
|
||||
|
||||
Name: kubernetes
|
||||
Version: 1.20.2
|
||||
Release: 24
|
||||
Release: 25
|
||||
Summary: Container cluster management
|
||||
License: ASL 2.0
|
||||
URL: https://k8s.io/kubernetes
|
||||
@ -43,6 +43,7 @@ Patch6015: 0016-Add-envFrom-to-serviceaccount-admission-plugin.patch
|
||||
Patch6016: 0017-backport-Fix-kubelet-panic-when-allocate-resource-for-pod.patch
|
||||
Patch6017: 0018-backport-reduce-configmap-and-secret-watch-of-kubelet.patch
|
||||
Patch6018: 0019-backport-Don-t-prematurely-close-reflectors-in-case-of-slow-i.patch
|
||||
Patch6019: 0020-backport-Fix-cpu-share-issues-on-systems-with-large-amounts-o.patch
|
||||
|
||||
%description
|
||||
Container cluster management.
|
||||
@ -274,6 +275,12 @@ getent passwd kube >/dev/null || useradd -r -g kube -d / -s /sbin/nologin \
|
||||
%systemd_postun kubelet kube-proxy
|
||||
|
||||
%changelog
|
||||
* Tue Aug 27 2024 zhaoxiaohu <zhaoxiaohu@kuaishou.com> - 1.20.2-25
|
||||
- Type:bugfix
|
||||
- CVE:NA
|
||||
- SUG:NA
|
||||
- DESC:Fix cpu share issues on systems with large amounts of cpu(when cpucore>256)
|
||||
|
||||
* Fri Aug 23 2024 zhaoxiaohu <zhaoxiaohu@kuaishou.com> - 1.20.2-24
|
||||
- DESC:Don't prematurely close reflectors in case of slow initialization in watch based manager
|
||||
|
||||
|
||||
Loading…
x
Reference in New Issue
Block a user