Add staged kubernetes version 1.21.3
Multiple versions of kubernetes are required to support upgrade. This adds staged version of kubernetes 1.21.3, built with a specific version of golang. All subpackage versions are included in the iso image without collisions. The following patches are ported to specific kubernetes version: kubelet-cpumanager-disable-CFS-quota-throttling-for-.patch kubelet-cpumanager-keep-normal-containers-off-reserv.patch kubelet-cpumanager-infrastructure-pods-use-system-re.patch kubelet-cpumanager-introduce-concept-of-isolated-CPU.patch kubeadm-create-platform-pods-with-zero-CPU-resources.patch enable-support-for-kubernetes-to-ignore-isolcpus.patch The following changes were made for 1.21.3: - following upstream commit was reverted: Revert-use-subpath-for-coredns-only-for-default-repo.patch - kubelet-cpumanager-disable-CFS-quota-throttling-for-.patch was refactored due to new internal_container_lifecycle framework We leverage the same mechanism to set Linux resources as: cpu manager: specify the container CPU set during the creation (commit 38dc7509f862f081828e7d9167107b8c6e98ea23). - kubelet-cpumanager-introduce-concept-of-isolated-CPU.patch was refactored due to upstream API change: node: podresources: make GetDevices() consistent (commit ad68f9588c72d6477b5a290c548a9031063ac659). The routine podIsolCPUs() was refactored in 1.21.3 since the API p.deviceManager.GetDevices() is returning multiple devices with a device per cpu. The resultant cpuset needs to be the aggregate. Story: 2008972 Task: 43056 Signed-off-by: Jim Gauld <james.gauld@windriver.com> Change-Id: I5ba7ff2e6aebb744af265698c0f90256ac5e70f4
This commit is contained in:
Jim Gauld
parent
229a6b32af
commit
872dd513fc
|
|
@ -165,6 +165,9 @@ kubernetes-1.19.13-client
|
|||
kubernetes-1.20.9-node
|
||||
kubernetes-1.20.9-kubeadm
|
||||
kubernetes-1.20.9-client
|
||||
kubernetes-1.21.3-node
|
||||
kubernetes-1.21.3-kubeadm
|
||||
kubernetes-1.21.3-client
|
||||
containerd
|
||||
k8s-pod-recovery
|
||||
containernetworking-plugins
|
||||
|
|
|
|||
|
|
@ -57,6 +57,7 @@ kubernetes/cni/plugins
|
|||
kubernetes/kubernetes-1.18.1
|
||||
kubernetes/kubernetes-1.19.13
|
||||
kubernetes/kubernetes-1.20.9
|
||||
kubernetes/kubernetes-1.21.3
|
||||
kubernetes/kubernetes-unversioned
|
||||
kubernetes/docker-distribution
|
||||
kubernetes/etcd
|
||||
|
|
|
|||
|
|
@ -34,6 +34,7 @@ kubernetes-contrib-v1.18.1.tar.gz#kubernetes-contrib-1.18.1#https://github.com/k
|
|||
kubernetes-v1.18.1.tar.gz#kubernetes-1.18.1#https://github.com/kubernetes/kubernetes/archive/7879fc12a63337efff607952a323df90cdc7a335.tar.gz#http##
|
||||
kubernetes-v1.19.13.tar.gz#kubernetes-1.19.13#https://github.com/kubernetes/kubernetes/archive/refs/tags/v1.19.13.tar.gz#http##
|
||||
kubernetes-v1.20.9.tar.gz#kubernetes-1.20.9#https://github.com/kubernetes/kubernetes/archive/refs/tags/v1.20.9.tar.gz#http##
|
||||
kubernetes-v1.21.3.tar.gz#kubernetes-1.21.3#https://github.com/kubernetes/kubernetes/archive/refs/tags/v1.21.3.tar.gz#http##
|
||||
kvm-unit-tests.git-4ea7633.tar.bz2#kvm-unit-tests#https://git.kernel.org/pub/scm/virt/kvm/kvm-unit-tests.git/snapshot/kvm-unit-tests-20171020.tar.gz#http##
|
||||
ldapscripts-2.0.8.tgz#ldapscripts-2.0.8#https://sourceforge.net/projects/ldapscripts/files/ldapscripts/ldapscripts-2.0.8/ldapscripts-2.0.8.tgz/download#http##
|
||||
libtpms-0.6.0-4f0d59d.tar.gz#libtpms-0.6.0#https://github.com/stefanberger/libtpms/tarball/c421ca0f4d00c0caceeda8d62c1efb2b7e47db04#http##
|
||||
|
|
|
|||
|
|
@ -0,0 +1,9 @@
|
|||
The spec file used here was from the kubernetes 1.10.0 src rpm.
|
||||
The orig file is included to help show modifications made to that
|
||||
spec file, to help understand which changes were needed and to
|
||||
assist with future upversioning.
|
||||
|
||||
The contrib tarball does not have the same versioning as kubernetes and
|
||||
there is little activity in that repo.
|
||||
|
||||
The version for the contrib tarball is arbitrary.
|
||||
|
|
@ -0,0 +1,5 @@
|
|||
VERSION=1.21.3
|
||||
TAR_NAME=kubernetes
|
||||
TAR="$TAR_NAME-v$VERSION.tar.gz"
|
||||
COPY_LIST="${CGCS_BASE}/downloads/$TAR $FILES_BASE/*"
|
||||
TIS_PATCH_VER=PKG_GITREVCOUNT
|
||||
|
|
@ -0,0 +1,117 @@
|
|||
From b90e3858a8d319c7526dd8190ee05edce24ba072 Mon Sep 17 00:00:00 2001
|
||||
From: Jim Gauld <james.gauld@windriver.com>
|
||||
Date: Thu, 9 Sep 2021 13:16:26 -0400
|
||||
Subject: [PATCH 7/7] Revert "use subpath for coredns only for default
|
||||
repository"
|
||||
|
||||
This reverts commit 38a41e1557649a7cc763bf737779db9aa03ec75e.
|
||||
---
|
||||
cmd/kubeadm/app/constants/constants.go | 2 +-
|
||||
cmd/kubeadm/app/images/images.go | 5 ---
|
||||
cmd/kubeadm/app/images/images_test.go | 50 --------------------------
|
||||
3 files changed, 1 insertion(+), 56 deletions(-)
|
||||
|
||||
diff --git a/cmd/kubeadm/app/constants/constants.go b/cmd/kubeadm/app/constants/constants.go
|
||||
index aed3a713020..3cb2d11ad45 100644
|
||||
--- a/cmd/kubeadm/app/constants/constants.go
|
||||
+++ b/cmd/kubeadm/app/constants/constants.go
|
||||
@@ -328,7 +328,7 @@ const (
|
||||
CoreDNSDeploymentName = "coredns"
|
||||
|
||||
// CoreDNSImageName specifies the name of the image for CoreDNS add-on
|
||||
- CoreDNSImageName = "coredns"
|
||||
+ CoreDNSImageName = "coredns/coredns"
|
||||
|
||||
// CoreDNSVersion is the version of CoreDNS to be deployed if it is used
|
||||
CoreDNSVersion = "v1.8.0"
|
||||
diff --git a/cmd/kubeadm/app/images/images.go b/cmd/kubeadm/app/images/images.go
|
||||
index 7ada3b75018..edf087e9c46 100644
|
||||
--- a/cmd/kubeadm/app/images/images.go
|
||||
+++ b/cmd/kubeadm/app/images/images.go
|
||||
@@ -21,7 +21,6 @@ import (
|
||||
|
||||
"k8s.io/klog/v2"
|
||||
kubeadmapi "k8s.io/kubernetes/cmd/kubeadm/app/apis/kubeadm"
|
||||
- kubeadmapiv1beta2 "k8s.io/kubernetes/cmd/kubeadm/app/apis/kubeadm/v1beta2"
|
||||
"k8s.io/kubernetes/cmd/kubeadm/app/constants"
|
||||
kubeadmutil "k8s.io/kubernetes/cmd/kubeadm/app/util"
|
||||
)
|
||||
@@ -53,10 +52,6 @@ func GetDNSImage(cfg *kubeadmapi.ClusterConfiguration) string {
|
||||
if cfg.DNS.ImageRepository != "" {
|
||||
dnsImageRepository = cfg.DNS.ImageRepository
|
||||
}
|
||||
- // Handle the renaming of the official image from "k8s.gcr.io/coredns" to "k8s.gcr.io/coredns/coredns
|
||||
- if dnsImageRepository == kubeadmapiv1beta2.DefaultImageRepository {
|
||||
- dnsImageRepository = fmt.Sprintf("%s/coredns", dnsImageRepository)
|
||||
- }
|
||||
// DNS uses an imageTag that corresponds to the DNS version matching the Kubernetes version
|
||||
dnsImageTag := constants.GetDNSVersion(cfg.DNS.Type)
|
||||
|
||||
diff --git a/cmd/kubeadm/app/images/images_test.go b/cmd/kubeadm/app/images/images_test.go
|
||||
index e5b417bcdc7..f5165406e0f 100644
|
||||
--- a/cmd/kubeadm/app/images/images_test.go
|
||||
+++ b/cmd/kubeadm/app/images/images_test.go
|
||||
@@ -22,7 +22,6 @@ import (
|
||||
"testing"
|
||||
|
||||
kubeadmapi "k8s.io/kubernetes/cmd/kubeadm/app/apis/kubeadm"
|
||||
- kubeadmapiv1beta2 "k8s.io/kubernetes/cmd/kubeadm/app/apis/kubeadm/v1beta2"
|
||||
"k8s.io/kubernetes/cmd/kubeadm/app/constants"
|
||||
)
|
||||
|
||||
@@ -238,52 +237,3 @@ func TestGetAllImages(t *testing.T) {
|
||||
})
|
||||
}
|
||||
}
|
||||
-
|
||||
-func TestGetDNSImage(t *testing.T) {
|
||||
- var tests = []struct {
|
||||
- expected string
|
||||
- cfg *kubeadmapi.ClusterConfiguration
|
||||
- }{
|
||||
- {
|
||||
- expected: "foo.io/coredns:v1.8.0",
|
||||
- cfg: &kubeadmapi.ClusterConfiguration{
|
||||
- ImageRepository: "foo.io",
|
||||
- DNS: kubeadmapi.DNS{
|
||||
- Type: kubeadmapi.CoreDNS,
|
||||
- },
|
||||
- },
|
||||
- },
|
||||
- {
|
||||
- expected: kubeadmapiv1beta2.DefaultImageRepository + "/coredns/coredns:v1.8.0",
|
||||
- cfg: &kubeadmapi.ClusterConfiguration{
|
||||
- ImageRepository: kubeadmapiv1beta2.DefaultImageRepository,
|
||||
- DNS: kubeadmapi.DNS{
|
||||
- Type: kubeadmapi.CoreDNS,
|
||||
- },
|
||||
- },
|
||||
- },
|
||||
- {
|
||||
- expected: "foo.io/coredns/coredns:v1.8.0",
|
||||
- cfg: &kubeadmapi.ClusterConfiguration{
|
||||
- ImageRepository: "foo.io",
|
||||
- DNS: kubeadmapi.DNS{
|
||||
- Type: kubeadmapi.CoreDNS,
|
||||
- ImageMeta: kubeadmapi.ImageMeta{
|
||||
- ImageRepository: "foo.io/coredns",
|
||||
- },
|
||||
- },
|
||||
- },
|
||||
- },
|
||||
- }
|
||||
-
|
||||
- for _, test := range tests {
|
||||
- actual := GetDNSImage(test.cfg)
|
||||
- if actual != test.expected {
|
||||
- t.Errorf(
|
||||
- "failed to GetDNSImage:\n\texpected: %s\n\t actual: %s",
|
||||
- test.expected,
|
||||
- actual,
|
||||
- )
|
||||
- }
|
||||
- }
|
||||
-}
|
||||
--
|
||||
2.17.1
|
||||
|
||||
|
|
@ -0,0 +1,79 @@
|
|||
From a0011e7749f7e54d3f1a689e717ded88e284860f Mon Sep 17 00:00:00 2001
|
||||
From: Chris Friesen <chris.friesen@windriver.com>
|
||||
Date: Fri, 23 Oct 2020 17:46:10 -0600
|
||||
Subject: [PATCH 6/7] enable support for kubernetes to ignore isolcpus
|
||||
|
||||
The normal mechanisms for allocating isolated CPUs do not allow
|
||||
a mix of isolated and exclusive CPUs in the same container. In
|
||||
order to allow this in *very* limited cases where the pod spec
|
||||
is known in advance we will add the ability to disable the normal
|
||||
isolcpus behaviour.
|
||||
|
||||
If the file "/etc/kubernetes/ignore_isolcpus" exists, then kubelet
|
||||
will basically forget everything it knows about isolcpus and just
|
||||
treat them like regular CPUs.
|
||||
|
||||
The admin user can then rely on the fact that CPU allocation is
|
||||
deterministic to ensure that the isolcpus they configure end up being
|
||||
allocated to the correct pods.
|
||||
---
|
||||
pkg/kubelet/cm/cpumanager/cpu_manager.go | 9 +++++++++
|
||||
pkg/kubelet/cm/cpumanager/policy_static.go | 8 ++++++++
|
||||
2 files changed, 17 insertions(+)
|
||||
|
||||
diff --git a/pkg/kubelet/cm/cpumanager/cpu_manager.go b/pkg/kubelet/cm/cpumanager/cpu_manager.go
|
||||
index 2563f61e7b5..1b226187fef 100644
|
||||
--- a/pkg/kubelet/cm/cpumanager/cpu_manager.go
|
||||
+++ b/pkg/kubelet/cm/cpumanager/cpu_manager.go
|
||||
@@ -19,6 +19,7 @@ package cpumanager
|
||||
import (
|
||||
"fmt"
|
||||
"math"
|
||||
+ "os"
|
||||
"sync"
|
||||
"time"
|
||||
"strings"
|
||||
@@ -55,6 +56,14 @@ const cpuManagerStateFileName = "cpu_manager_state"
|
||||
|
||||
// get the system-level isolated CPUs
|
||||
func getIsolcpus() cpuset.CPUSet {
|
||||
+
|
||||
+ // This is a gross hack to basically turn off awareness of isolcpus to enable
|
||||
+ // isolated cpus to be allocated to pods the same way as non-isolated CPUs.
|
||||
+ if _, err := os.Stat("/etc/kubernetes/ignore_isolcpus"); err == nil {
|
||||
+ klog.Infof("[cpumanager] turning off isolcpus awareness")
|
||||
+ return cpuset.NewCPUSet()
|
||||
+ }
|
||||
+
|
||||
dat, err := ioutil.ReadFile("/sys/devices/system/cpu/isolated")
|
||||
if err != nil {
|
||||
klog.Errorf("[cpumanager] unable to read sysfs isolcpus subdir")
|
||||
diff --git a/pkg/kubelet/cm/cpumanager/policy_static.go b/pkg/kubelet/cm/cpumanager/policy_static.go
|
||||
index 2ad14a98911..73b74d5c4cc 100644
|
||||
--- a/pkg/kubelet/cm/cpumanager/policy_static.go
|
||||
+++ b/pkg/kubelet/cm/cpumanager/policy_static.go
|
||||
@@ -18,6 +18,7 @@ package cpumanager
|
||||
|
||||
import (
|
||||
"fmt"
|
||||
+ "os"
|
||||
"strconv"
|
||||
|
||||
v1 "k8s.io/api/core/v1"
|
||||
@@ -613,6 +614,13 @@ func isKubeInfra(pod *v1.Pod) bool {
|
||||
|
||||
// get the isolated CPUs (if any) from the devices associated with a specific container
|
||||
func (p *staticPolicy) podIsolCPUs(pod *v1.Pod, container *v1.Container) cpuset.CPUSet {
|
||||
+
|
||||
+ // This is a gross hack to basically turn off awareness of isolcpus to enable
|
||||
+ // isolated cpus to be allocated to pods the same way as non-isolated CPUs.
|
||||
+ if _, err := os.Stat("/etc/kubernetes/ignore_isolcpus"); err == nil {
|
||||
+ return cpuset.NewCPUSet()
|
||||
+ }
|
||||
+
|
||||
// NOTE: This is required for TestStaticPolicyAdd() since makePod() does
|
||||
// not create UID. We also need a way to properly stub devicemanager.
|
||||
if len(string(pod.UID)) == 0 {
|
||||
--
|
||||
2.17.1
|
||||
|
||||
|
|
@ -0,0 +1,84 @@
|
|||
MDSFORMANPAGES="kube-apiserver.md kube-controller-manager.md kube-proxy.md kube-scheduler.md kubelet.md"
|
||||
|
||||
# remove comments from man pages
|
||||
for manpage in ${MDSFORMANPAGES}; do
|
||||
pos=$(grep -n "<\!-- END MUNGE: UNVERSIONED_WARNING -->" ${manpage} | cut -d':' -f1)
|
||||
if [ -n ${pos} ]; then
|
||||
sed -i "1,${pos}{/.*/d}" ${manpage}
|
||||
fi
|
||||
done
|
||||
|
||||
# for each man page add NAME and SYNOPSIS section
|
||||
# kube-apiserver
|
||||
sed -i -s "s/## kube-apiserver/# NAME\nkube-apiserver \- Provides the API for kubernetes orchestration.\n\n# SYNOPSIS\n**kube-apiserver** [OPTIONS]\n/" kube-apiserver.md
|
||||
|
||||
cat << 'EOF' >> kube-apiserver.md
|
||||
# EXAMPLES
|
||||
```
|
||||
/usr/bin/kube-apiserver --logtostderr=true --v=0 --etcd_servers=http://127.0.0.1:4001 --insecure_bind_address=127.0.0.1 --insecure_port=8080 --kubelet_port=10250 --service-cluster-ip-range=10.1.1.0/24 --allow_privileged=false
|
||||
```
|
||||
EOF
|
||||
# kube-controller-manager
|
||||
sed -i -s "s/## kube-controller-manager/# NAME\nkube-controller-manager \- Enforces kubernetes services.\n\n# SYNOPSIS\n**kube-controller-manager** [OPTIONS]\n/" kube-controller-manager.md
|
||||
|
||||
cat << 'EOF' >> kube-controller-manager.md
|
||||
# EXAMPLES
|
||||
```
|
||||
/usr/bin/kube-controller-manager --logtostderr=true --v=0 --master=127.0.0.1:8080
|
||||
```
|
||||
EOF
|
||||
# kube-proxy
|
||||
sed -i -s "s/## kube-proxy/# NAME\nkube-proxy \- Provides network proxy services.\n\n# SYNOPSIS\n**kube-proxy** [OPTIONS]\n/" kube-proxy.md
|
||||
|
||||
cat << 'EOF' >> kube-proxy.md
|
||||
# EXAMPLES
|
||||
```
|
||||
/usr/bin/kube-proxy --logtostderr=true --v=0 --master=http://127.0.0.1:8080
|
||||
```
|
||||
EOF
|
||||
# kube-scheduler
|
||||
sed -i -s "s/## kube-scheduler/# NAME\nkube-scheduler \- Schedules containers on hosts.\n\n# SYNOPSIS\n**kube-scheduler** [OPTIONS]\n/" kube-scheduler.md
|
||||
|
||||
cat << 'EOF' >> kube-scheduler.md
|
||||
# EXAMPLES
|
||||
```
|
||||
/usr/bin/kube-scheduler --logtostderr=true --v=0 --master=127.0.0.1:8080
|
||||
```
|
||||
EOF
|
||||
# kubelet
|
||||
sed -i -s "s/## kubelet/# NAME\nkubelet \- Processes a container manifest so the containers are launched according to how they are described.\n\n# SYNOPSIS\n**kubelet** [OPTIONS]\n/" kubelet.md
|
||||
|
||||
cat << 'EOF' >> kubelet.md
|
||||
# EXAMPLES
|
||||
```
|
||||
/usr/bin/kubelet --logtostderr=true --v=0 --api_servers=http://127.0.0.1:8080 --address=127.0.0.1 --port=10250 --hostname_override=127.0.0.1 --allow-privileged=false
|
||||
```
|
||||
EOF
|
||||
|
||||
# for all man-pages
|
||||
for md in $MDSFORMANPAGES; do
|
||||
# correct section names
|
||||
sed -i -s "s/### Synopsis/# DESCRIPTION/" $md
|
||||
sed -i -s "s/### Options/# OPTIONS/" $md
|
||||
# add header
|
||||
sed -i "s/# NAME/% KUBERNETES(1) kubernetes User Manuals\n# NAME/" $md
|
||||
# modify list of options
|
||||
# options with no value in ""
|
||||
sed -i -r 's/(^ )(-[^":][^":]*)(:)(.*)/\*\*\2\*\*\n\t\4\n/' $md
|
||||
# option with value in ""
|
||||
sed -i -r 's/(^ )(-[^":][^":]*)("[^"]*")(:)(.*)/\*\*\2\3\*\*\n\t\5\n/' $md
|
||||
# options in -s, --long
|
||||
sed -i -r 's/(^ )(-[a-z], -[^":][^":]*)(:)(.*)/\*\*\2\*\*\n\t\4\n/' $md
|
||||
sed -i -r 's/(^ )(-[a-z], -[^":][^":]*)("[^"]*")(:)(.*)/\*\*\2\3\*\*\n\t\5\n/' $md
|
||||
# remove ```
|
||||
sed -i 's/```//' $md
|
||||
# remove all lines starting with ######
|
||||
sed -i 's/^######.*//' $md
|
||||
# modify footer
|
||||
sed -i -r "s/^\[!\[Analytics\].*//" $md
|
||||
# md does not contain section => taking 1
|
||||
name="${md%.md}"
|
||||
go-md2man -in $md -out man/man1/$name.1
|
||||
done
|
||||
|
||||
|
||||
|
|
@ -0,0 +1,108 @@
|
|||
From de653bd0823b248d623a39c17a3872e85ce952b0 Mon Sep 17 00:00:00 2001
|
||||
From: Chris Friesen <chris.friesen@windriver.com>
|
||||
Date: Fri, 3 Sep 2021 18:05:15 -0400
|
||||
Subject: [PATCH 5/7] kubeadm: create platform pods with zero CPU resources
|
||||
|
||||
We want to specify zero CPU resources when creating the manifests
|
||||
for the static platform pods, as a workaround for the lack of
|
||||
separate resource tracking for platform resources.
|
||||
|
||||
We also specify zero CPU resources for the coredns deployment.
|
||||
manifests.go appears to be the main file for this, not sure if the
|
||||
others are used but I changed them just in case.
|
||||
|
||||
Signed-off-by: Jim Gauld <james.gauld@windriver.com>
|
||||
---
|
||||
cluster/addons/dns/coredns/coredns.yaml.base | 2 +-
|
||||
cluster/addons/dns/coredns/coredns.yaml.in | 2 +-
|
||||
cluster/addons/dns/coredns/coredns.yaml.sed | 2 +-
|
||||
cmd/kubeadm/app/phases/addons/dns/manifests.go | 2 +-
|
||||
cmd/kubeadm/app/phases/controlplane/manifests.go | 6 +++---
|
||||
5 files changed, 7 insertions(+), 7 deletions(-)
|
||||
|
||||
diff --git a/cluster/addons/dns/coredns/coredns.yaml.base b/cluster/addons/dns/coredns/coredns.yaml.base
|
||||
index 4ee054f8ba5..d2b58f4af0e 100644
|
||||
--- a/cluster/addons/dns/coredns/coredns.yaml.base
|
||||
+++ b/cluster/addons/dns/coredns/coredns.yaml.base
|
||||
@@ -138,7 +138,7 @@ spec:
|
||||
limits:
|
||||
memory: __DNS__MEMORY__LIMIT__
|
||||
requests:
|
||||
- cpu: 100m
|
||||
+ cpu: 0
|
||||
memory: 70Mi
|
||||
args: [ "-conf", "/etc/coredns/Corefile" ]
|
||||
volumeMounts:
|
||||
diff --git a/cluster/addons/dns/coredns/coredns.yaml.in b/cluster/addons/dns/coredns/coredns.yaml.in
|
||||
index 1f791e447c9..ff03a801646 100644
|
||||
--- a/cluster/addons/dns/coredns/coredns.yaml.in
|
||||
+++ b/cluster/addons/dns/coredns/coredns.yaml.in
|
||||
@@ -138,7 +138,7 @@ spec:
|
||||
limits:
|
||||
memory: 'dns_memory_limit'
|
||||
requests:
|
||||
- cpu: 100m
|
||||
+ cpu: 0
|
||||
memory: 70Mi
|
||||
args: [ "-conf", "/etc/coredns/Corefile" ]
|
||||
volumeMounts:
|
||||
diff --git a/cluster/addons/dns/coredns/coredns.yaml.sed b/cluster/addons/dns/coredns/coredns.yaml.sed
|
||||
index 4d64278aaa4..38fc9196b28 100644
|
||||
--- a/cluster/addons/dns/coredns/coredns.yaml.sed
|
||||
+++ b/cluster/addons/dns/coredns/coredns.yaml.sed
|
||||
@@ -138,7 +138,7 @@ spec:
|
||||
limits:
|
||||
memory: $DNS_MEMORY_LIMIT
|
||||
requests:
|
||||
- cpu: 100m
|
||||
+ cpu: 0
|
||||
memory: 70Mi
|
||||
args: [ "-conf", "/etc/coredns/Corefile" ]
|
||||
volumeMounts:
|
||||
diff --git a/cmd/kubeadm/app/phases/addons/dns/manifests.go b/cmd/kubeadm/app/phases/addons/dns/manifests.go
|
||||
index 3ac6856bfc6..0763b4c63db 100644
|
||||
--- a/cmd/kubeadm/app/phases/addons/dns/manifests.go
|
||||
+++ b/cmd/kubeadm/app/phases/addons/dns/manifests.go
|
||||
@@ -95,7 +95,7 @@ spec:
|
||||
limits:
|
||||
memory: 170Mi
|
||||
requests:
|
||||
- cpu: 100m
|
||||
+ cpu: 0
|
||||
memory: 70Mi
|
||||
args: [ "-conf", "/etc/coredns/Corefile" ]
|
||||
volumeMounts:
|
||||
diff --git a/cmd/kubeadm/app/phases/controlplane/manifests.go b/cmd/kubeadm/app/phases/controlplane/manifests.go
|
||||
index 8181bea63a4..4c4b4448dd4 100644
|
||||
--- a/cmd/kubeadm/app/phases/controlplane/manifests.go
|
||||
+++ b/cmd/kubeadm/app/phases/controlplane/manifests.go
|
||||
@@ -60,7 +60,7 @@ func GetStaticPodSpecs(cfg *kubeadmapi.ClusterConfiguration, endpoint *kubeadmap
|
||||
LivenessProbe: staticpodutil.LivenessProbe(staticpodutil.GetAPIServerProbeAddress(endpoint), "/livez", int(endpoint.BindPort), v1.URISchemeHTTPS),
|
||||
ReadinessProbe: staticpodutil.ReadinessProbe(staticpodutil.GetAPIServerProbeAddress(endpoint), "/readyz", int(endpoint.BindPort), v1.URISchemeHTTPS),
|
||||
StartupProbe: staticpodutil.StartupProbe(staticpodutil.GetAPIServerProbeAddress(endpoint), "/livez", int(endpoint.BindPort), v1.URISchemeHTTPS, cfg.APIServer.TimeoutForControlPlane),
|
||||
- Resources: staticpodutil.ComponentResources("250m"),
|
||||
+ Resources: staticpodutil.ComponentResources("0"),
|
||||
Env: kubeadmutil.GetProxyEnvVars(),
|
||||
}, mounts.GetVolumes(kubeadmconstants.KubeAPIServer),
|
||||
map[string]string{kubeadmconstants.KubeAPIServerAdvertiseAddressEndpointAnnotationKey: endpoint.String()}),
|
||||
@@ -72,7 +72,7 @@ func GetStaticPodSpecs(cfg *kubeadmapi.ClusterConfiguration, endpoint *kubeadmap
|
||||
VolumeMounts: staticpodutil.VolumeMountMapToSlice(mounts.GetVolumeMounts(kubeadmconstants.KubeControllerManager)),
|
||||
LivenessProbe: staticpodutil.LivenessProbe(staticpodutil.GetControllerManagerProbeAddress(cfg), "/healthz", kubeadmconstants.KubeControllerManagerPort, v1.URISchemeHTTPS),
|
||||
StartupProbe: staticpodutil.StartupProbe(staticpodutil.GetControllerManagerProbeAddress(cfg), "/healthz", kubeadmconstants.KubeControllerManagerPort, v1.URISchemeHTTPS, cfg.APIServer.TimeoutForControlPlane),
|
||||
- Resources: staticpodutil.ComponentResources("200m"),
|
||||
+ Resources: staticpodutil.ComponentResources("0"),
|
||||
Env: kubeadmutil.GetProxyEnvVars(),
|
||||
}, mounts.GetVolumes(kubeadmconstants.KubeControllerManager), nil),
|
||||
kubeadmconstants.KubeScheduler: staticpodutil.ComponentPod(v1.Container{
|
||||
@@ -83,7 +83,7 @@ func GetStaticPodSpecs(cfg *kubeadmapi.ClusterConfiguration, endpoint *kubeadmap
|
||||
VolumeMounts: staticpodutil.VolumeMountMapToSlice(mounts.GetVolumeMounts(kubeadmconstants.KubeScheduler)),
|
||||
LivenessProbe: staticpodutil.LivenessProbe(staticpodutil.GetSchedulerProbeAddress(cfg), "/healthz", kubeadmconstants.KubeSchedulerPort, v1.URISchemeHTTPS),
|
||||
StartupProbe: staticpodutil.StartupProbe(staticpodutil.GetSchedulerProbeAddress(cfg), "/healthz", kubeadmconstants.KubeSchedulerPort, v1.URISchemeHTTPS, cfg.APIServer.TimeoutForControlPlane),
|
||||
- Resources: staticpodutil.ComponentResources("100m"),
|
||||
+ Resources: staticpodutil.ComponentResources("0"),
|
||||
Env: kubeadmutil.GetProxyEnvVars(),
|
||||
}, mounts.GetVolumes(kubeadmconstants.KubeScheduler), nil),
|
||||
}
|
||||
--
|
||||
2.17.1
|
||||
|
||||
|
|
@ -0,0 +1,17 @@
|
|||
# Note: This dropin only works with kubeadm and kubelet v1.11+
|
||||
[Service]
|
||||
Environment="KUBELET_KUBECONFIG_ARGS=--bootstrap-kubeconfig=/etc/kubernetes/bootstrap-kubelet.conf --kubeconfig=/etc/kubernetes/kubelet.conf"
|
||||
Environment="KUBELET_CONFIG_ARGS=--config=/var/lib/kubelet/config.yaml"
|
||||
# This is a file that "kubeadm init" and "kubeadm join" generates at runtime, populating the KUBELET_KUBEADM_ARGS variable dynamically
|
||||
EnvironmentFile=-/var/lib/kubelet/kubeadm-flags.env
|
||||
# This is a file that the user can use for overrides of the kubelet args as a last resort. Preferably, the user should use
|
||||
# the .NodeRegistration.KubeletExtraArgs object in the configuration files instead. KUBELET_EXTRA_ARGS should be sourced from this file.
|
||||
EnvironmentFile=-/etc/sysconfig/kubelet
|
||||
ExecStart=
|
||||
ExecStart=/usr/bin/kubelet $KUBELET_KUBECONFIG_ARGS $KUBELET_CONFIG_ARGS $KUBELET_KUBEADM_ARGS $KUBELET_EXTRA_ARGS
|
||||
ExecStartPre=-/usr/bin/kubelet-cgroup-setup.sh
|
||||
ExecStartPost=/bin/bash -c 'echo $MAINPID > /var/run/kubelet.pid;'
|
||||
ExecStopPost=/bin/rm -f /var/run/kubelet.pid
|
||||
Restart=always
|
||||
StartLimitInterval=0
|
||||
RestartSec=10
|
||||
|
|
@ -0,0 +1,132 @@
|
|||
#!/bin/bash
|
||||
#
|
||||
# Copyright (c) 2019 Wind River Systems, Inc.
|
||||
#
|
||||
# SPDX-License-Identifier: Apache-2.0
|
||||
#
|
||||
# This script does minimal cgroup setup for kubelet. This creates k8s-infra
|
||||
# cgroup for a minimal set of resource controllers, and configures cpuset
|
||||
# attributes to span all online cpus and nodes. This will do nothing if
|
||||
# the k8s-infra cgroup already exists (i.e., assume already configured).
|
||||
# NOTE: The creation of directories under /sys/fs/cgroup is volatile, and
|
||||
# does not persist reboots. The cpuset.mems and cpuset.cpus is later updated
|
||||
# by puppet kubernetes.pp manifest.
|
||||
#
|
||||
|
||||
# Define minimal path
|
||||
PATH=/bin:/usr/bin:/usr/local/bin
|
||||
|
||||
# Log info message to /var/log/daemon.log
|
||||
function LOG {
|
||||
logger -p daemon.info "$0($$): $@"
|
||||
}
|
||||
|
||||
# Log error message to /var/log/daemon.log
|
||||
function ERROR {
|
||||
logger -s -p daemon.error "$0($$): ERROR: $@"
|
||||
}
|
||||
|
||||
# Create minimal cgroup directories and configure cpuset attributes if required
|
||||
function create_cgroup {
|
||||
local cg_name=$1
|
||||
local cg_nodeset=$2
|
||||
local cg_cpuset=$3
|
||||
|
||||
local CGROUP=/sys/fs/cgroup
|
||||
local CONTROLLERS_AUTO_DELETED=("pids" "hugetlb")
|
||||
local CONTROLLERS_PRESERVED=("cpuset" "memory" "cpu,cpuacct" "systemd")
|
||||
local cnt=''
|
||||
local CGDIR=''
|
||||
local RC=0
|
||||
|
||||
# Ensure that these cgroups are created every time as they are auto deleted
|
||||
for cnt in ${CONTROLLERS_AUTO_DELETED[@]}; do
|
||||
CGDIR=${CGROUP}/${cnt}/${cg_name}
|
||||
if [ -d ${CGDIR} ]; then
|
||||
LOG "Nothing to do, already configured: ${CGDIR}."
|
||||
continue
|
||||
fi
|
||||
LOG "Creating: ${CGDIR}"
|
||||
mkdir -p ${CGDIR}
|
||||
RC=$?
|
||||
if [ ${RC} -ne 0 ]; then
|
||||
ERROR "Creating: ${CGDIR}, rc=${RC}"
|
||||
exit ${RC}
|
||||
fi
|
||||
done
|
||||
|
||||
# These cgroups are preserved so if any of these are encountered additional
|
||||
# cgroup setup is not required
|
||||
for cnt in ${CONTROLLERS_PRESERVED[@]}; do
|
||||
CGDIR=${CGROUP}/${cnt}/${cg_name}
|
||||
if [ -d ${CGDIR} ]; then
|
||||
LOG "Nothing to do, already configured: ${CGDIR}."
|
||||
exit ${RC}
|
||||
fi
|
||||
LOG "Creating: ${CGDIR}"
|
||||
mkdir -p ${CGDIR}
|
||||
RC=$?
|
||||
if [ ${RC} -ne 0 ]; then
|
||||
ERROR "Creating: ${CGDIR}, rc=${RC}"
|
||||
exit ${RC}
|
||||
fi
|
||||
done
|
||||
|
||||
# Customize cpuset attributes
|
||||
LOG "Configuring cgroup: ${cg_name}, nodeset: ${cg_nodeset}, cpuset: ${cg_cpuset}"
|
||||
CGDIR=${CGROUP}/cpuset/${cg_name}
|
||||
local CGMEMS=${CGDIR}/cpuset.mems
|
||||
local CGCPUS=${CGDIR}/cpuset.cpus
|
||||
local CGTASKS=${CGDIR}/tasks
|
||||
|
||||
# Assign cgroup memory nodeset
|
||||
LOG "Assign nodeset ${cg_nodeset} to ${CGMEMS}"
|
||||
/bin/echo ${cg_nodeset} > ${CGMEMS}
|
||||
RC=$?
|
||||
if [ ${RC} -ne 0 ]; then
|
||||
ERROR "Unable to write to: ${CGMEMS}, rc=${RC}"
|
||||
exit ${RC}
|
||||
fi
|
||||
|
||||
# Assign cgroup cpus
|
||||
LOG "Assign cpuset ${cg_cpuset} to ${CGCPUS}"
|
||||
/bin/echo ${cg_cpuset} > ${CGCPUS}
|
||||
RC=$?
|
||||
if [ ${RC} -ne 0 ]; then
|
||||
ERROR "Assigning: ${cg_cpuset} to ${CGCPUS}, rc=${RC}"
|
||||
exit ${RC}
|
||||
fi
|
||||
|
||||
# Set file ownership
|
||||
chown root:root ${CGMEMS} ${CGCPUS} ${CGTASKS}
|
||||
RC=$?
|
||||
if [ ${RC} -ne 0 ]; then
|
||||
ERROR "Setting owner for: ${CGMEMS}, ${CGCPUS}, ${CGTASKS}, rc=${RC}"
|
||||
exit ${RC}
|
||||
fi
|
||||
|
||||
# Set file mode permissions
|
||||
chmod 644 ${CGMEMS} ${CGCPUS} ${CGTASKS}
|
||||
RC=$?
|
||||
if [ ${RC} -ne 0 ]; then
|
||||
ERROR "Setting mode for: ${CGMEMS}, ${CGCPUS}, ${CGTASKS}, rc=${RC}"
|
||||
exit ${RC}
|
||||
fi
|
||||
|
||||
return ${RC}
|
||||
}
|
||||
|
||||
if [ $UID -ne 0 ]; then
|
||||
ERROR "Require sudo/root."
|
||||
exit 1
|
||||
fi
|
||||
|
||||
# Configure default kubepods cpuset to span all online cpus and nodes.
|
||||
ONLINE_NODESET=$(/bin/cat /sys/devices/system/node/online)
|
||||
ONLINE_CPUSET=$(/bin/cat /sys/devices/system/cpu/online)
|
||||
|
||||
# Configure kubelet cgroup to match cgroupRoot.
|
||||
create_cgroup 'k8s-infra' ${ONLINE_NODESET} ${ONLINE_CPUSET}
|
||||
|
||||
exit $?
|
||||
|
||||
|
|
@ -0,0 +1,256 @@
|
|||
From 3f69868f7bca99f6875dd4d197b3a974d1b558ed Mon Sep 17 00:00:00 2001
|
||||
From: Jim Gauld <james.gauld@windriver.com>
|
||||
Date: Wed, 22 Sep 2021 10:09:06 -0400
|
||||
Subject: [PATCH 1/7] kubelet cpumanager disable CFS quota throttling for
|
||||
Guaranteed pods
|
||||
|
||||
This disables CFS CPU quota to avoid performance degradation due to
|
||||
Linux kernel CFS quota implementation. Note that 4.18 kernel attempts
|
||||
to solve the CFS throttling problem, but there are reports that it is
|
||||
not completely effective.
|
||||
|
||||
This disables CFS quota throttling for Guaranteed pods for both
|
||||
parent and container cgroups by writing -1 to cgroup cpu.cfs_quota_us.
|
||||
Disabling has a dramatic latency improvement for HTTP response times.
|
||||
|
||||
This patch is refactored in 1.21.3 due to new internal_container_lifecycle
|
||||
framework. We leverage the same mechanism to set Linux resources as:
|
||||
cpu manager: specify the container CPU set during the creation
|
||||
(commit 38dc7509f862f081828e7d9167107b8c6e98ea23).
|
||||
|
||||
Signed-off-by: Jim Gauld <james.gauld@windriver.com>
|
||||
---
|
||||
pkg/kubelet/cm/cpumanager/cpu_manager.go | 7 ++++
|
||||
pkg/kubelet/cm/cpumanager/fake_cpu_manager.go | 6 +++
|
||||
pkg/kubelet/cm/helpers_linux.go | 10 +++++
|
||||
pkg/kubelet/cm/helpers_linux_test.go | 42 ++++++++++---------
|
||||
.../cm/internal_container_lifecycle_linux.go | 9 ++++
|
||||
5 files changed, 54 insertions(+), 20 deletions(-)
|
||||
|
||||
diff --git a/pkg/kubelet/cm/cpumanager/cpu_manager.go b/pkg/kubelet/cm/cpumanager/cpu_manager.go
|
||||
index 5a6e5082f15..f7b9c8d07bf 100644
|
||||
--- a/pkg/kubelet/cm/cpumanager/cpu_manager.go
|
||||
+++ b/pkg/kubelet/cm/cpumanager/cpu_manager.go
|
||||
@@ -72,6 +72,9 @@ type Manager interface {
|
||||
// State returns a read-only interface to the internal CPU manager state.
|
||||
State() state.Reader
|
||||
|
||||
+ // GetCPUPolicy returns the assigned CPU manager policy
|
||||
+ GetCPUPolicy() string
|
||||
+
|
||||
// GetTopologyHints implements the topologymanager.HintProvider Interface
|
||||
// and is consulted to achieve NUMA aware resource alignment among this
|
||||
// and other resource controllers.
|
||||
@@ -291,6 +294,10 @@ func (m *manager) State() state.Reader {
|
||||
return m.state
|
||||
}
|
||||
|
||||
+func (m *manager) GetCPUPolicy() string {
|
||||
+ return m.policy.Name()
|
||||
+}
|
||||
+
|
||||
func (m *manager) GetTopologyHints(pod *v1.Pod, container *v1.Container) map[string][]topologymanager.TopologyHint {
|
||||
// Garbage collect any stranded resources before providing TopologyHints
|
||||
m.removeStaleState()
|
||||
diff --git a/pkg/kubelet/cm/cpumanager/fake_cpu_manager.go b/pkg/kubelet/cm/cpumanager/fake_cpu_manager.go
|
||||
index 2c38b52b374..1cb0ea10923 100644
|
||||
--- a/pkg/kubelet/cm/cpumanager/fake_cpu_manager.go
|
||||
+++ b/pkg/kubelet/cm/cpumanager/fake_cpu_manager.go
|
||||
@@ -28,6 +28,7 @@ import (
|
||||
)
|
||||
|
||||
type fakeManager struct {
|
||||
+ policy Policy
|
||||
state state.State
|
||||
}
|
||||
|
||||
@@ -69,6 +70,10 @@ func (m *fakeManager) State() state.Reader {
|
||||
return m.state
|
||||
}
|
||||
|
||||
+func (m *fakeManager) GetCPUPolicy() string {
|
||||
+ return m.policy.Name()
|
||||
+}
|
||||
+
|
||||
func (m *fakeManager) GetCPUs(podUID, containerName string) cpuset.CPUSet {
|
||||
klog.InfoS("GetCPUs", "podUID", podUID, "containerName", containerName)
|
||||
return cpuset.CPUSet{}
|
||||
@@ -82,6 +87,7 @@ func (m *fakeManager) GetAllocatableCPUs() cpuset.CPUSet {
|
||||
// NewFakeManager creates empty/fake cpu manager
|
||||
func NewFakeManager() Manager {
|
||||
return &fakeManager{
|
||||
+ policy: &nonePolicy{},
|
||||
state: state.NewMemoryState(),
|
||||
}
|
||||
}
|
||||
diff --git a/pkg/kubelet/cm/helpers_linux.go b/pkg/kubelet/cm/helpers_linux.go
|
||||
index aa5c37639dc..302284ef408 100644
|
||||
--- a/pkg/kubelet/cm/helpers_linux.go
|
||||
+++ b/pkg/kubelet/cm/helpers_linux.go
|
||||
@@ -169,6 +169,16 @@ func ResourceConfigForPod(pod *v1.Pod, enforceCPULimits bool, cpuPeriod uint64)
|
||||
// build the result
|
||||
result := &ResourceConfig{}
|
||||
if qosClass == v1.PodQOSGuaranteed {
|
||||
+ // Disable CFS CPU quota to avoid performance degradation due to
|
||||
+ // Linux kernel CFS throttle implementation.
|
||||
+ // NOTE: 4.18 kernel attempts to solve CFS throttling problem,
|
||||
+ // but there are reports that it is not completely effective.
|
||||
+ // This will configure cgroup CFS parameters at pod level:
|
||||
+ // /sys/fs/cgroup/cpu/k8s-infra/kubepods/<pod>/cpu.cfs_quota_us
|
||||
+ // /sys/fs/cgroup/cpu/k8s-infra/kubepods/<pod>/cpu.cfs_period_us
|
||||
+ cpuQuota = int64(-1)
|
||||
+ cpuPeriod = uint64(100000)
|
||||
+
|
||||
result.CpuShares = &cpuShares
|
||||
result.CpuQuota = &cpuQuota
|
||||
result.CpuPeriod = &cpuPeriod
|
||||
diff --git a/pkg/kubelet/cm/helpers_linux_test.go b/pkg/kubelet/cm/helpers_linux_test.go
|
||||
index 56d765fbc22..8c7309937dd 100644
|
||||
--- a/pkg/kubelet/cm/helpers_linux_test.go
|
||||
+++ b/pkg/kubelet/cm/helpers_linux_test.go
|
||||
@@ -63,8 +63,9 @@ func TestResourceConfigForPod(t *testing.T) {
|
||||
burstablePartialShares := MilliCPUToShares(200)
|
||||
burstableQuota := MilliCPUToQuota(200, int64(defaultQuotaPeriod))
|
||||
guaranteedShares := MilliCPUToShares(100)
|
||||
- guaranteedQuota := MilliCPUToQuota(100, int64(defaultQuotaPeriod))
|
||||
- guaranteedTunedQuota := MilliCPUToQuota(100, int64(tunedQuotaPeriod))
|
||||
+ guaranteedQuotaPeriod := uint64(100000)
|
||||
+ guaranteedQuota := int64(-1)
|
||||
+ guaranteedTunedQuota := int64(-1)
|
||||
memoryQuantity = resource.MustParse("100Mi")
|
||||
cpuNoLimit := int64(-1)
|
||||
guaranteedMemory := memoryQuantity.Value()
|
||||
@@ -203,8 +204,8 @@ func TestResourceConfigForPod(t *testing.T) {
|
||||
},
|
||||
},
|
||||
enforceCPULimits: true,
|
||||
- quotaPeriod: defaultQuotaPeriod,
|
||||
- expected: &ResourceConfig{CpuShares: &guaranteedShares, CpuQuota: &guaranteedQuota, CpuPeriod: &defaultQuotaPeriod, Memory: &guaranteedMemory},
|
||||
+ quotaPeriod: guaranteedQuotaPeriod,
|
||||
+ expected: &ResourceConfig{CpuShares: &guaranteedShares, CpuQuota: &guaranteedQuota, CpuPeriod: &guaranteedQuotaPeriod, Memory: &guaranteedMemory},
|
||||
},
|
||||
"guaranteed-no-cpu-enforcement": {
|
||||
pod: &v1.Pod{
|
||||
@@ -217,8 +218,8 @@ func TestResourceConfigForPod(t *testing.T) {
|
||||
},
|
||||
},
|
||||
enforceCPULimits: false,
|
||||
- quotaPeriod: defaultQuotaPeriod,
|
||||
- expected: &ResourceConfig{CpuShares: &guaranteedShares, CpuQuota: &cpuNoLimit, CpuPeriod: &defaultQuotaPeriod, Memory: &guaranteedMemory},
|
||||
+ quotaPeriod: guaranteedQuotaPeriod,
|
||||
+ expected: &ResourceConfig{CpuShares: &guaranteedShares, CpuQuota: &cpuNoLimit, CpuPeriod: &guaranteedQuotaPeriod, Memory: &guaranteedMemory},
|
||||
},
|
||||
"guaranteed-with-tuned-quota": {
|
||||
pod: &v1.Pod{
|
||||
@@ -231,8 +232,8 @@ func TestResourceConfigForPod(t *testing.T) {
|
||||
},
|
||||
},
|
||||
enforceCPULimits: true,
|
||||
- quotaPeriod: tunedQuotaPeriod,
|
||||
- expected: &ResourceConfig{CpuShares: &guaranteedShares, CpuQuota: &guaranteedTunedQuota, CpuPeriod: &tunedQuotaPeriod, Memory: &guaranteedMemory},
|
||||
+ quotaPeriod: guaranteedQuotaPeriod,
|
||||
+ expected: &ResourceConfig{CpuShares: &guaranteedShares, CpuQuota: &guaranteedTunedQuota, CpuPeriod: &guaranteedQuotaPeriod, Memory: &guaranteedMemory},
|
||||
},
|
||||
"guaranteed-no-cpu-enforcement-with-tuned-quota": {
|
||||
pod: &v1.Pod{
|
||||
@@ -245,8 +246,8 @@ func TestResourceConfigForPod(t *testing.T) {
|
||||
},
|
||||
},
|
||||
enforceCPULimits: false,
|
||||
- quotaPeriod: tunedQuotaPeriod,
|
||||
- expected: &ResourceConfig{CpuShares: &guaranteedShares, CpuQuota: &cpuNoLimit, CpuPeriod: &tunedQuotaPeriod, Memory: &guaranteedMemory},
|
||||
+ quotaPeriod: guaranteedQuotaPeriod,
|
||||
+ expected: &ResourceConfig{CpuShares: &guaranteedShares, CpuQuota: &cpuNoLimit, CpuPeriod: &guaranteedQuotaPeriod, Memory: &guaranteedMemory},
|
||||
},
|
||||
}
|
||||
|
||||
@@ -283,8 +284,9 @@ func TestResourceConfigForPodWithCustomCPUCFSQuotaPeriod(t *testing.T) {
|
||||
burstablePartialShares := MilliCPUToShares(200)
|
||||
burstableQuota := MilliCPUToQuota(200, int64(defaultQuotaPeriod))
|
||||
guaranteedShares := MilliCPUToShares(100)
|
||||
- guaranteedQuota := MilliCPUToQuota(100, int64(defaultQuotaPeriod))
|
||||
- guaranteedTunedQuota := MilliCPUToQuota(100, int64(tunedQuotaPeriod))
|
||||
+ guaranteedQuotaPeriod := uint64(100000)
|
||||
+ guaranteedQuota := int64(-1)
|
||||
+ guaranteedTunedQuota := int64(-1)
|
||||
memoryQuantity = resource.MustParse("100Mi")
|
||||
cpuNoLimit := int64(-1)
|
||||
guaranteedMemory := memoryQuantity.Value()
|
||||
@@ -423,8 +425,8 @@ func TestResourceConfigForPodWithCustomCPUCFSQuotaPeriod(t *testing.T) {
|
||||
},
|
||||
},
|
||||
enforceCPULimits: true,
|
||||
- quotaPeriod: defaultQuotaPeriod,
|
||||
- expected: &ResourceConfig{CpuShares: &guaranteedShares, CpuQuota: &guaranteedQuota, CpuPeriod: &defaultQuotaPeriod, Memory: &guaranteedMemory},
|
||||
+ quotaPeriod: guaranteedQuotaPeriod,
|
||||
+ expected: &ResourceConfig{CpuShares: &guaranteedShares, CpuQuota: &guaranteedQuota, CpuPeriod: &guaranteedQuotaPeriod, Memory: &guaranteedMemory},
|
||||
},
|
||||
"guaranteed-no-cpu-enforcement": {
|
||||
pod: &v1.Pod{
|
||||
@@ -437,8 +439,8 @@ func TestResourceConfigForPodWithCustomCPUCFSQuotaPeriod(t *testing.T) {
|
||||
},
|
||||
},
|
||||
enforceCPULimits: false,
|
||||
- quotaPeriod: defaultQuotaPeriod,
|
||||
- expected: &ResourceConfig{CpuShares: &guaranteedShares, CpuQuota: &cpuNoLimit, CpuPeriod: &defaultQuotaPeriod, Memory: &guaranteedMemory},
|
||||
+ quotaPeriod: guaranteedQuotaPeriod,
|
||||
+ expected: &ResourceConfig{CpuShares: &guaranteedShares, CpuQuota: &cpuNoLimit, CpuPeriod: &guaranteedQuotaPeriod, Memory: &guaranteedMemory},
|
||||
},
|
||||
"guaranteed-with-tuned-quota": {
|
||||
pod: &v1.Pod{
|
||||
@@ -451,8 +453,8 @@ func TestResourceConfigForPodWithCustomCPUCFSQuotaPeriod(t *testing.T) {
|
||||
},
|
||||
},
|
||||
enforceCPULimits: true,
|
||||
- quotaPeriod: tunedQuotaPeriod,
|
||||
- expected: &ResourceConfig{CpuShares: &guaranteedShares, CpuQuota: &guaranteedTunedQuota, CpuPeriod: &tunedQuotaPeriod, Memory: &guaranteedMemory},
|
||||
+ quotaPeriod: guaranteedQuotaPeriod,
|
||||
+ expected: &ResourceConfig{CpuShares: &guaranteedShares, CpuQuota: &guaranteedTunedQuota, CpuPeriod: &guaranteedQuotaPeriod, Memory: &guaranteedMemory},
|
||||
},
|
||||
"guaranteed-no-cpu-enforcement-with-tuned-quota": {
|
||||
pod: &v1.Pod{
|
||||
@@ -465,8 +467,8 @@ func TestResourceConfigForPodWithCustomCPUCFSQuotaPeriod(t *testing.T) {
|
||||
},
|
||||
},
|
||||
enforceCPULimits: false,
|
||||
- quotaPeriod: tunedQuotaPeriod,
|
||||
- expected: &ResourceConfig{CpuShares: &guaranteedShares, CpuQuota: &cpuNoLimit, CpuPeriod: &tunedQuotaPeriod, Memory: &guaranteedMemory},
|
||||
+ quotaPeriod: guaranteedQuotaPeriod,
|
||||
+ expected: &ResourceConfig{CpuShares: &guaranteedShares, CpuQuota: &cpuNoLimit, CpuPeriod: &guaranteedQuotaPeriod, Memory: &guaranteedMemory},
|
||||
},
|
||||
}
|
||||
|
||||
diff --git a/pkg/kubelet/cm/internal_container_lifecycle_linux.go b/pkg/kubelet/cm/internal_container_lifecycle_linux.go
|
||||
index 9cf41620b8c..fa15dbe1671 100644
|
||||
--- a/pkg/kubelet/cm/internal_container_lifecycle_linux.go
|
||||
+++ b/pkg/kubelet/cm/internal_container_lifecycle_linux.go
|
||||
@@ -19,10 +19,12 @@ limitations under the License.
|
||||
package cm
|
||||
|
||||
import (
|
||||
+ //"fmt"
|
||||
"strconv"
|
||||
"strings"
|
||||
|
||||
"k8s.io/api/core/v1"
|
||||
+ v1qos "k8s.io/kubernetes/pkg/apis/core/v1/helper/qos"
|
||||
runtimeapi "k8s.io/cri-api/pkg/apis/runtime/v1alpha2"
|
||||
)
|
||||
|
||||
@@ -32,6 +34,13 @@ func (i *internalContainerLifecycleImpl) PreCreateContainer(pod *v1.Pod, contain
|
||||
if !allocatedCPUs.IsEmpty() {
|
||||
containerConfig.Linux.Resources.CpusetCpus = allocatedCPUs.String()
|
||||
}
|
||||
+ // Disable cgroup CFS throttle at the container level.
|
||||
+ // /sys/fs/cgroup/cpu/k8s-infra/kubepods/<pod>/<container>/cpu.cfs_quota_us
|
||||
+ // /sys/fs/cgroup/cpu/k8s-infra/kubepods/<pod>/<container>/cpu.cfs_period_us
|
||||
+ if i.cpuManager.GetCPUPolicy() == "static" && v1qos.GetPodQOS(pod) == v1.PodQOSGuaranteed {
|
||||
+ containerConfig.Linux.Resources.CpuPeriod = int64(100000)
|
||||
+ containerConfig.Linux.Resources.CpuQuota = int64(-1)
|
||||
+ }
|
||||
}
|
||||
|
||||
if i.memoryManager != nil {
|
||||
--
|
||||
2.17.1
|
||||
|
||||
|
|
@ -0,0 +1,139 @@
|
|||
From 66f4a61ce77624dcc0b520e01fccacec61dfea37 Mon Sep 17 00:00:00 2001
|
||||
From: Jim Gauld <james.gauld@windriver.com>
|
||||
Date: Tue, 7 Sep 2021 14:21:03 -0400
|
||||
Subject: [PATCH 3/7] kubelet cpumanager infrastructure pods use system
|
||||
reserved CPUs
|
||||
|
||||
This assigns system infrastructure pods to the "reserved" cpuset
|
||||
to isolate them from the shared pool of CPUs.
|
||||
|
||||
Infrastructure pods include any pods that belong to the kube-system,
|
||||
armada, cert-manager, vault, platform-deployment-manager, portieris,
|
||||
or notification namespaces.
|
||||
|
||||
The implementation is a bit simplistic, it is assumed that the
|
||||
"reserved" cpuset is large enough to handle all infrastructure pods
|
||||
CPU allocations.
|
||||
|
||||
This also prevents infrastucture pods from using Guaranteed resources.
|
||||
|
||||
Signed-off-by: Jim Gauld <james.gauld@windriver.com>
|
||||
---
|
||||
pkg/kubelet/cm/cpumanager/policy_static.go | 44 +++++++++++++++++++
|
||||
.../cm/cpumanager/policy_static_test.go | 19 +++++++-
|
||||
2 files changed, 62 insertions(+), 1 deletion(-)
|
||||
|
||||
diff --git a/pkg/kubelet/cm/cpumanager/policy_static.go b/pkg/kubelet/cm/cpumanager/policy_static.go
|
||||
index 94d81796132..b48a5b997a3 100644
|
||||
--- a/pkg/kubelet/cm/cpumanager/policy_static.go
|
||||
+++ b/pkg/kubelet/cm/cpumanager/policy_static.go
|
||||
@@ -32,6 +32,11 @@ import (
|
||||
// PolicyStatic is the name of the static policy
|
||||
const PolicyStatic policyName = "static"
|
||||
|
||||
+// Define namespaces used by platform infrastructure pods
|
||||
+var infraNamespaces = [...]string{
|
||||
+ "kube-system", "armada", "cert-manager", "platform-deployment-manager", "portieris", "vault", "notification",
|
||||
+}
|
||||
+
|
||||
// staticPolicy is a CPU manager policy that does not change CPU
|
||||
// assignments for exclusively pinned guaranteed containers after the main
|
||||
// container process starts.
|
||||
@@ -232,6 +237,31 @@ func (p *staticPolicy) updateCPUsToReuse(pod *v1.Pod, container *v1.Container, c
|
||||
}
|
||||
|
||||
func (p *staticPolicy) Allocate(s state.State, pod *v1.Pod, container *v1.Container) error {
|
||||
+ // Process infra pods before guaranteed pods
|
||||
+ if isKubeInfra(pod) {
|
||||
+ // Container belongs in reserved pool.
|
||||
+ // We don't want to fall through to the p.guaranteedCPUs() clause below so return either nil or error.
|
||||
+ if _, ok := s.GetCPUSet(string(pod.UID), container.Name); ok {
|
||||
+ klog.Infof("[cpumanager] static policy: reserved container already present in state, skipping " +
|
||||
+ "(namespace: %s, pod UID: %s, pod: %s, container: %s)",
|
||||
+ pod.Namespace, string(pod.UID), pod.Name, container.Name)
|
||||
+ return nil
|
||||
+ }
|
||||
+
|
||||
+ cpuset := p.reserved
|
||||
+ if cpuset.IsEmpty() {
|
||||
+ // If this happens then someone messed up.
|
||||
+ return fmt.Errorf("[cpumanager] static policy: reserved container unable to allocate cpus " +
|
||||
+ "(namespace: %s, pod UID: %s, pod: %s, container: %s); cpuset=%v, reserved:%v",
|
||||
+ pod.Namespace, string(pod.UID), pod.Name, container.Name, cpuset, p.reserved)
|
||||
+ }
|
||||
+ s.SetCPUSet(string(pod.UID), container.Name, cpuset)
|
||||
+ klog.Infof("[cpumanager] static policy: reserved: AddContainer " +
|
||||
+ "(namespace: %s, pod UID: %s, pod: %s, container: %s); cpuset=%v",
|
||||
+ pod.Namespace, string(pod.UID), pod.Name, container.Name, cpuset)
|
||||
+ return nil
|
||||
+ }
|
||||
+
|
||||
if numCPUs := p.guaranteedCPUs(pod, container); numCPUs != 0 {
|
||||
klog.InfoS("Static policy: Allocate", "pod", klog.KObj(pod), "containerName", container.Name)
|
||||
// container belongs in an exclusively allocated pool
|
||||
@@ -321,6 +351,10 @@ func (p *staticPolicy) guaranteedCPUs(pod *v1.Pod, container *v1.Container) int
|
||||
if cpuQuantity.Value()*1000 != cpuQuantity.MilliValue() {
|
||||
return 0
|
||||
}
|
||||
+ // Infrastructure pods use reserved CPUs even if they're in the Guaranteed QoS class
|
||||
+ if isKubeInfra(pod) {
|
||||
+ return 0
|
||||
+ }
|
||||
// Safe downcast to do for all systems with < 2.1 billion CPUs.
|
||||
// Per the language spec, `int` is guaranteed to be at least 32 bits wide.
|
||||
// https://golang.org/ref/spec#Numeric_types
|
||||
@@ -523,3 +557,13 @@ func (p *staticPolicy) generateCPUTopologyHints(availableCPUs cpuset.CPUSet, reu
|
||||
|
||||
return hints
|
||||
}
|
||||
+
|
||||
+// check if a given pod is in a platform infrastructure namespace
|
||||
+func isKubeInfra(pod *v1.Pod) bool {
|
||||
+ for _, namespace := range infraNamespaces {
|
||||
+ if namespace == pod.Namespace {
|
||||
+ return true
|
||||
+ }
|
||||
+ }
|
||||
+ return false
|
||||
+}
|
||||
\ No newline at end of file
|
||||
diff --git a/pkg/kubelet/cm/cpumanager/policy_static_test.go b/pkg/kubelet/cm/cpumanager/policy_static_test.go
|
||||
index 93a2870532a..90ad98e71bc 100644
|
||||
--- a/pkg/kubelet/cm/cpumanager/policy_static_test.go
|
||||
+++ b/pkg/kubelet/cm/cpumanager/policy_static_test.go
|
||||
@@ -747,7 +747,8 @@ func TestStaticPolicyStartWithResvList(t *testing.T) {
|
||||
}
|
||||
|
||||
func TestStaticPolicyAddWithResvList(t *testing.T) {
|
||||
-
|
||||
+ infraPod := makePod("fakePod", "fakeContainer2", "200m", "200m")
|
||||
+ infraPod.Namespace = "kube-system"
|
||||
testCases := []staticPolicyTestWithResvList{
|
||||
{
|
||||
description: "GuPodSingleCore, SingleSocketHT, ExpectError",
|
||||
@@ -789,6 +790,22 @@ func TestStaticPolicyAddWithResvList(t *testing.T) {
|
||||
expCPUAlloc: true,
|
||||
expCSet: cpuset.NewCPUSet(4, 5),
|
||||
},
|
||||
+ {
|
||||
+ description: "InfraPod, SingleSocketHT, ExpectAllocReserved",
|
||||
+ topo: topoSingleSocketHT,
|
||||
+ numReservedCPUs: 2,
|
||||
+ reserved: cpuset.NewCPUSet(0, 1),
|
||||
+ stAssignments: state.ContainerCPUAssignments{
|
||||
+ "fakePod": map[string]cpuset.CPUSet{
|
||||
+ "fakeContainer100": cpuset.NewCPUSet(2, 3, 6, 7),
|
||||
+ },
|
||||
+ },
|
||||
+ stDefaultCPUSet: cpuset.NewCPUSet(4, 5),
|
||||
+ pod: infraPod,
|
||||
+ expErr: nil,
|
||||
+ expCPUAlloc: true,
|
||||
+ expCSet: cpuset.NewCPUSet(0, 1),
|
||||
+ },
|
||||
}
|
||||
|
||||
testExcl := true
|
||||
--
|
||||
2.17.1
|
||||
|
||||
|
|
@ -0,0 +1,535 @@
|
|||
From 81a6d41690c45b168034df04df0199cd5abe60e9 Mon Sep 17 00:00:00 2001
|
||||
From: Jim Gauld <james.gauld@windriver.com>
|
||||
Date: Wed, 8 Sep 2021 09:28:40 -0400
|
||||
Subject: [PATCH 4/7] kubelet cpumanager introduce concept of isolated CPUs
|
||||
|
||||
This introduces the concept of "isolated CPUs", which are CPUs that
|
||||
have been isolated at the kernel level via the "isolcpus" kernel boot
|
||||
parameter.
|
||||
|
||||
When starting the kubelet process, two separate sets of reserved CPUs
|
||||
may be specified. With this change CPUs reserved via
|
||||
'--system-reserved=cpu' will be used for infrastructure pods while the
|
||||
isolated CPUs should be reserved via '--kube-reserved=cpu' to cause
|
||||
kubelet to skip over them for "normal" CPU resource tracking. The
|
||||
kubelet code will double-check that the specified isolated CPUs match
|
||||
what the kernel exposes in "/sys/devices/system/cpu/isolated".
|
||||
|
||||
A plugin (outside the scope of this commit) will expose the isolated
|
||||
CPUs to kubelet via the device plugin API.
|
||||
|
||||
If a pod specifies some number of "isolcpus" resources, the device
|
||||
manager will allocate them. In this code we check whether such
|
||||
resources have been allocated, and if so we set the container cpuset to
|
||||
the isolated CPUs. This does mean that it really only makes sense to
|
||||
specify "isolcpus" resources for best-effort or burstable pods, not for
|
||||
guaranteed ones since that would throw off the accounting code. In
|
||||
order to ensure the accounting still works as designed, if "isolcpus"
|
||||
are specified for guaranteed pods, the affinity will be set to the
|
||||
non-isolated CPUs.
|
||||
|
||||
This patch was refactored in 1.21.3 due to upstream API change
|
||||
node: podresources: make GetDevices() consistent
|
||||
(commit ad68f9588c72d6477b5a290c548a9031063ac659).
|
||||
|
||||
The routine podIsolCPUs() was refactored in 1.21.3 since the API
|
||||
p.deviceManager.GetDevices() is returning multiple devices with
|
||||
a device per cpu. The resultant cpuset needs to be the aggregate.
|
||||
|
||||
Signed-off-by: Jim Gauld <james.gauld@windriver.com>
|
||||
Co-authored-by: Chris Friesen <chris.friesen@windriver.com>
|
||||
---
|
||||
pkg/kubelet/cm/container_manager_linux.go | 1 +
|
||||
pkg/kubelet/cm/cpumanager/cpu_manager.go | 31 ++++++-
|
||||
pkg/kubelet/cm/cpumanager/cpu_manager_test.go | 13 ++-
|
||||
pkg/kubelet/cm/cpumanager/policy_static.go | 85 +++++++++++++++++--
|
||||
.../cm/cpumanager/policy_static_test.go | 44 ++++++++--
|
||||
5 files changed, 156 insertions(+), 18 deletions(-)
|
||||
|
||||
diff --git a/pkg/kubelet/cm/container_manager_linux.go b/pkg/kubelet/cm/container_manager_linux.go
|
||||
index e7bed73fc69..d74a6339939 100644
|
||||
--- a/pkg/kubelet/cm/container_manager_linux.go
|
||||
+++ b/pkg/kubelet/cm/container_manager_linux.go
|
||||
@@ -337,6 +337,7 @@ func NewContainerManager(mountUtil mount.Interface, cadvisorInterface cadvisor.I
|
||||
cm.GetNodeAllocatableReservation(),
|
||||
nodeConfig.KubeletRootDir,
|
||||
cm.topologyManager,
|
||||
+ cm.deviceManager,
|
||||
)
|
||||
if err != nil {
|
||||
klog.ErrorS(err, "Failed to initialize cpu manager")
|
||||
diff --git a/pkg/kubelet/cm/cpumanager/cpu_manager.go b/pkg/kubelet/cm/cpumanager/cpu_manager.go
|
||||
index 5b82e5b02ac..2563f61e7b5 100644
|
||||
--- a/pkg/kubelet/cm/cpumanager/cpu_manager.go
|
||||
+++ b/pkg/kubelet/cm/cpumanager/cpu_manager.go
|
||||
@@ -21,6 +21,8 @@ import (
|
||||
"math"
|
||||
"sync"
|
||||
"time"
|
||||
+ "strings"
|
||||
+ "io/ioutil"
|
||||
|
||||
cadvisorapi "github.com/google/cadvisor/info/v1"
|
||||
v1 "k8s.io/api/core/v1"
|
||||
@@ -34,6 +36,7 @@ import (
|
||||
"k8s.io/kubernetes/pkg/kubelet/cm/cpuset"
|
||||
"k8s.io/kubernetes/pkg/kubelet/cm/topologymanager"
|
||||
"k8s.io/kubernetes/pkg/kubelet/config"
|
||||
+ "k8s.io/kubernetes/pkg/kubelet/cm/devicemanager"
|
||||
kubecontainer "k8s.io/kubernetes/pkg/kubelet/container"
|
||||
"k8s.io/kubernetes/pkg/kubelet/status"
|
||||
)
|
||||
@@ -50,6 +53,25 @@ type policyName string
|
||||
// cpuManagerStateFileName is the file name where cpu manager stores its state
|
||||
const cpuManagerStateFileName = "cpu_manager_state"
|
||||
|
||||
+// get the system-level isolated CPUs
|
||||
+func getIsolcpus() cpuset.CPUSet {
|
||||
+ dat, err := ioutil.ReadFile("/sys/devices/system/cpu/isolated")
|
||||
+ if err != nil {
|
||||
+ klog.Errorf("[cpumanager] unable to read sysfs isolcpus subdir")
|
||||
+ return cpuset.NewCPUSet()
|
||||
+ }
|
||||
+
|
||||
+ // The isolated cpus string ends in a newline
|
||||
+ cpustring := strings.TrimSuffix(string(dat), "\n")
|
||||
+ cset, err := cpuset.Parse(cpustring)
|
||||
+ if err != nil {
|
||||
+ klog.Errorf("[cpumanager] unable to parse sysfs isolcpus string to cpuset")
|
||||
+ return cpuset.NewCPUSet()
|
||||
+ }
|
||||
+
|
||||
+ return cset
|
||||
+}
|
||||
+
|
||||
// Manager interface provides methods for Kubelet to manage pod cpus.
|
||||
type Manager interface {
|
||||
// Start is called during Kubelet initialization.
|
||||
@@ -143,7 +165,7 @@ func (s *sourcesReadyStub) AddSource(source string) {}
|
||||
func (s *sourcesReadyStub) AllReady() bool { return true }
|
||||
|
||||
// NewManager creates new cpu manager based on provided policy
|
||||
-func NewManager(cpuPolicyName string, reconcilePeriod time.Duration, machineInfo *cadvisorapi.MachineInfo, specificCPUs cpuset.CPUSet, nodeAllocatableReservation v1.ResourceList, stateFileDirectory string, affinity topologymanager.Store) (Manager, error) {
|
||||
+func NewManager(cpuPolicyName string, reconcilePeriod time.Duration, machineInfo *cadvisorapi.MachineInfo, specificCPUs cpuset.CPUSet, nodeAllocatableReservation v1.ResourceList, stateFileDirectory string, affinity topologymanager.Store, deviceManager devicemanager.Manager) (Manager, error) {
|
||||
var topo *topology.CPUTopology
|
||||
var policy Policy
|
||||
|
||||
@@ -181,8 +203,11 @@ func NewManager(cpuPolicyName string, reconcilePeriod time.Duration, machineInfo
|
||||
// NOTE: Set excludeReserved unconditionally to exclude reserved CPUs from default cpuset.
|
||||
// This variable is primarily to make testing easier.
|
||||
excludeReserved := true
|
||||
- policy, err = NewStaticPolicy(topo, numReservedCPUs, specificCPUs, affinity, excludeReserved)
|
||||
-
|
||||
+ // isolCPUs is the set of kernel-isolated CPUs. They should be a subset of specificCPUs or
|
||||
+ // of the CPUs that NewStaticPolicy() will pick if numReservedCPUs is set. It's only in the
|
||||
+ // argument list here for ease of testing, it's really internal to the policy.
|
||||
+ isolCPUs := getIsolcpus()
|
||||
+ policy, err = NewStaticPolicy(topo, numReservedCPUs, specificCPUs, isolCPUs, affinity, deviceManager, excludeReserved)
|
||||
if err != nil {
|
||||
return nil, fmt.Errorf("new static policy error: %v", err)
|
||||
}
|
||||
diff --git a/pkg/kubelet/cm/cpumanager/cpu_manager_test.go b/pkg/kubelet/cm/cpumanager/cpu_manager_test.go
|
||||
index a2e73de6875..4c021634d6f 100644
|
||||
--- a/pkg/kubelet/cm/cpumanager/cpu_manager_test.go
|
||||
+++ b/pkg/kubelet/cm/cpumanager/cpu_manager_test.go
|
||||
@@ -38,6 +38,7 @@ import (
|
||||
"k8s.io/kubernetes/pkg/kubelet/cm/cpumanager/topology"
|
||||
"k8s.io/kubernetes/pkg/kubelet/cm/cpuset"
|
||||
"k8s.io/kubernetes/pkg/kubelet/cm/topologymanager"
|
||||
+ "k8s.io/kubernetes/pkg/kubelet/cm/devicemanager"
|
||||
)
|
||||
|
||||
type mockState struct {
|
||||
@@ -215,6 +216,7 @@ func makeMultiContainerPod(initCPUs, appCPUs []struct{ request, limit string })
|
||||
}
|
||||
|
||||
func TestCPUManagerAdd(t *testing.T) {
|
||||
+ testDM, _ := devicemanager.NewManagerStub()
|
||||
testExcl := false
|
||||
testPolicy, _ := NewStaticPolicy(
|
||||
&topology.CPUTopology{
|
||||
@@ -230,7 +232,9 @@ func TestCPUManagerAdd(t *testing.T) {
|
||||
},
|
||||
0,
|
||||
cpuset.NewCPUSet(),
|
||||
+ cpuset.NewCPUSet(),
|
||||
topologymanager.NewFakeManager(),
|
||||
+ testDM,
|
||||
testExcl)
|
||||
testCases := []struct {
|
||||
description string
|
||||
@@ -480,8 +484,9 @@ func TestCPUManagerAddWithInitContainers(t *testing.T) {
|
||||
}
|
||||
|
||||
testExcl := false
|
||||
+ testDM, _ := devicemanager.NewManagerStub()
|
||||
for _, testCase := range testCases {
|
||||
- policy, _ := NewStaticPolicy(testCase.topo, testCase.numReservedCPUs, cpuset.NewCPUSet(), topologymanager.NewFakeManager(), testExcl)
|
||||
+ policy, _ := NewStaticPolicy(testCase.topo, testCase.numReservedCPUs, cpuset.NewCPUSet(), cpuset.NewCPUSet(), topologymanager.NewFakeManager(), testDM, testExcl)
|
||||
|
||||
mockState := &mockState{
|
||||
assignments: testCase.stAssignments,
|
||||
@@ -635,7 +640,8 @@ func TestCPUManagerGenerate(t *testing.T) {
|
||||
}
|
||||
defer os.RemoveAll(sDir)
|
||||
|
||||
- mgr, err := NewManager(testCase.cpuPolicyName, 5*time.Second, machineInfo, cpuset.NewCPUSet(), testCase.nodeAllocatableReservation, sDir, topologymanager.NewFakeManager())
|
||||
+ testDM, err := devicemanager.NewManagerStub()
|
||||
+ mgr, err := NewManager(testCase.cpuPolicyName, 5*time.Second, machineInfo, cpuset.NewCPUSet(), testCase.nodeAllocatableReservation, sDir, topologymanager.NewFakeManager(), testDM)
|
||||
if testCase.expectedError != nil {
|
||||
if !strings.Contains(err.Error(), testCase.expectedError.Error()) {
|
||||
t.Errorf("Unexpected error message. Have: %s wants %s", err.Error(), testCase.expectedError.Error())
|
||||
@@ -990,6 +996,7 @@ func TestReconcileState(t *testing.T) {
|
||||
// the following tests are with --reserved-cpus configured
|
||||
func TestCPUManagerAddWithResvList(t *testing.T) {
|
||||
testExcl := false
|
||||
+ testDM, _ := devicemanager.NewManagerStub()
|
||||
testPolicy, _ := NewStaticPolicy(
|
||||
&topology.CPUTopology{
|
||||
NumCPUs: 4,
|
||||
@@ -1004,7 +1011,9 @@ func TestCPUManagerAddWithResvList(t *testing.T) {
|
||||
},
|
||||
1,
|
||||
cpuset.NewCPUSet(0),
|
||||
+ cpuset.NewCPUSet(),
|
||||
topologymanager.NewFakeManager(),
|
||||
+ testDM,
|
||||
testExcl)
|
||||
testCases := []struct {
|
||||
description string
|
||||
diff --git a/pkg/kubelet/cm/cpumanager/policy_static.go b/pkg/kubelet/cm/cpumanager/policy_static.go
|
||||
index b48a5b997a3..2ad14a98911 100644
|
||||
--- a/pkg/kubelet/cm/cpumanager/policy_static.go
|
||||
+++ b/pkg/kubelet/cm/cpumanager/policy_static.go
|
||||
@@ -18,6 +18,7 @@ package cpumanager
|
||||
|
||||
import (
|
||||
"fmt"
|
||||
+ "strconv"
|
||||
|
||||
v1 "k8s.io/api/core/v1"
|
||||
"k8s.io/klog/v2"
|
||||
@@ -27,6 +28,7 @@ import (
|
||||
"k8s.io/kubernetes/pkg/kubelet/cm/cpuset"
|
||||
"k8s.io/kubernetes/pkg/kubelet/cm/topologymanager"
|
||||
"k8s.io/kubernetes/pkg/kubelet/cm/topologymanager/bitmask"
|
||||
+ "k8s.io/kubernetes/pkg/kubelet/cm/devicemanager"
|
||||
)
|
||||
|
||||
// PolicyStatic is the name of the static policy
|
||||
@@ -80,6 +82,10 @@ type staticPolicy struct {
|
||||
topology *topology.CPUTopology
|
||||
// set of CPUs that is not available for exclusive assignment
|
||||
reserved cpuset.CPUSet
|
||||
+ // subset of reserved CPUs with isolcpus attribute
|
||||
+ isolcpus cpuset.CPUSet
|
||||
+ // parent containerManager, used to get device list
|
||||
+ deviceManager devicemanager.Manager
|
||||
// If true, default CPUSet should exclude reserved CPUs
|
||||
excludeReserved bool
|
||||
// topology manager reference to get container Topology affinity
|
||||
@@ -94,7 +100,7 @@ var _ Policy = &staticPolicy{}
|
||||
// NewStaticPolicy returns a CPU manager policy that does not change CPU
|
||||
// assignments for exclusively pinned guaranteed containers after the main
|
||||
// container process starts.
|
||||
-func NewStaticPolicy(topology *topology.CPUTopology, numReservedCPUs int, reservedCPUs cpuset.CPUSet, affinity topologymanager.Store, excludeReserved bool) (Policy, error) {
|
||||
+func NewStaticPolicy(topology *topology.CPUTopology, numReservedCPUs int, reservedCPUs cpuset.CPUSet, isolCPUs cpuset.CPUSet, affinity topologymanager.Store, deviceManager devicemanager.Manager, excludeReserved bool) (Policy, error) {
|
||||
allCPUs := topology.CPUDetails.CPUs()
|
||||
var reserved cpuset.CPUSet
|
||||
if reservedCPUs.Size() > 0 {
|
||||
@@ -115,9 +121,17 @@ func NewStaticPolicy(topology *topology.CPUTopology, numReservedCPUs int, reserv
|
||||
|
||||
klog.InfoS("Reserved CPUs not available for exclusive assignment", "reservedSize", reserved.Size(), "reserved", reserved)
|
||||
|
||||
+ if !isolCPUs.IsSubsetOf(reserved) {
|
||||
+ klog.Errorf("[cpumanager] isolCPUs %v is not a subset of reserved %v", isolCPUs, reserved)
|
||||
+ reserved = reserved.Union(isolCPUs)
|
||||
+ klog.Warningf("[cpumanager] mismatch isolCPUs %v, force reserved %v", isolCPUs, reserved)
|
||||
+ }
|
||||
+
|
||||
return &staticPolicy{
|
||||
topology: topology,
|
||||
reserved: reserved,
|
||||
+ isolcpus: isolCPUs,
|
||||
+ deviceManager: deviceManager,
|
||||
excludeReserved: excludeReserved,
|
||||
affinity: affinity,
|
||||
cpusToReuse: make(map[string]cpuset.CPUSet),
|
||||
@@ -154,8 +168,8 @@ func (p *staticPolicy) validateState(s state.State) error {
|
||||
} else {
|
||||
s.SetDefaultCPUSet(allCPUs)
|
||||
}
|
||||
- klog.Infof("[cpumanager] static policy: CPUSet: allCPUs:%v, reserved:%v, default:%v\n",
|
||||
- allCPUs, p.reserved, s.GetDefaultCPUSet())
|
||||
+ klog.Infof("[cpumanager] static policy: CPUSet: allCPUs:%v, reserved:%v, isolcpus:%v, default:%v\n",
|
||||
+ allCPUs, p.reserved, p.isolcpus, s.GetDefaultCPUSet())
|
||||
return nil
|
||||
}
|
||||
|
||||
@@ -248,12 +262,12 @@ func (p *staticPolicy) Allocate(s state.State, pod *v1.Pod, container *v1.Contai
|
||||
return nil
|
||||
}
|
||||
|
||||
- cpuset := p.reserved
|
||||
+ cpuset := p.reserved.Clone().Difference(p.isolcpus)
|
||||
if cpuset.IsEmpty() {
|
||||
// If this happens then someone messed up.
|
||||
return fmt.Errorf("[cpumanager] static policy: reserved container unable to allocate cpus " +
|
||||
- "(namespace: %s, pod UID: %s, pod: %s, container: %s); cpuset=%v, reserved:%v",
|
||||
- pod.Namespace, string(pod.UID), pod.Name, container.Name, cpuset, p.reserved)
|
||||
+ "(namespace: %s, pod UID: %s, pod: %s, container: %s); cpuset=%v, reserved:%v, isolcpus:%v",
|
||||
+ pod.Namespace, string(pod.UID), pod.Name, container.Name, cpuset, p.reserved, p.isolcpus)
|
||||
}
|
||||
s.SetCPUSet(string(pod.UID), container.Name, cpuset)
|
||||
klog.Infof("[cpumanager] static policy: reserved: AddContainer " +
|
||||
@@ -284,8 +298,37 @@ func (p *staticPolicy) Allocate(s state.State, pod *v1.Pod, container *v1.Contai
|
||||
}
|
||||
s.SetCPUSet(string(pod.UID), container.Name, cpuset)
|
||||
p.updateCPUsToReuse(pod, container, cpuset)
|
||||
+ klog.Infof("[cpumanager] guaranteed: AddContainer " +
|
||||
+ "(namespace: %s, pod UID: %s, pod: %s, container: %s); numCPUS=%d, cpuset=%v",
|
||||
+ pod.Namespace, string(pod.UID), pod.Name, container.Name, numCPUs, cpuset)
|
||||
+ return nil
|
||||
+ }
|
||||
|
||||
+ if isolcpus := p.podIsolCPUs(pod, container); isolcpus.Size() > 0 {
|
||||
+ // container has requested isolated CPUs
|
||||
+ if set, ok := s.GetCPUSet(string(pod.UID), container.Name); ok {
|
||||
+ if set.Equals(isolcpus) {
|
||||
+ klog.Infof("[cpumanager] isolcpus container already present in state, skipping " +
|
||||
+ "(namespace: %s, pod UID: %s, pod: %s, container: %s)",
|
||||
+ pod.Namespace, string(pod.UID), pod.Name, container.Name)
|
||||
+ return nil
|
||||
+ } else {
|
||||
+ klog.Infof("[cpumanager] isolcpus container state has cpus %v, should be %v" +
|
||||
+ "(namespace: %s, pod UID: %s, pod: %s, container: %s)",
|
||||
+ isolcpus, set, pod.Namespace, string(pod.UID), pod.Name, container.Name)
|
||||
+ }
|
||||
+ }
|
||||
+ // Note that we do not do anything about init containers here.
|
||||
+ // It looks like devices are allocated per-pod based on effective requests/limits
|
||||
+ // and extra devices from initContainers are not freed up when the regular containers start.
|
||||
+ // TODO: confirm this is still true for 1.20
|
||||
+ s.SetCPUSet(string(pod.UID), container.Name, isolcpus)
|
||||
+ klog.Infof("[cpumanager] isolcpus: AddContainer " +
|
||||
+ "(namespace: %s, pod UID: %s, pod: %s, container: %s); cpuset=%v",
|
||||
+ pod.Namespace, string(pod.UID), pod.Name, container.Name, isolcpus)
|
||||
+ return nil
|
||||
}
|
||||
+
|
||||
// container belongs in the shared pool (nothing to do; use default cpuset)
|
||||
return nil
|
||||
}
|
||||
@@ -566,4 +609,34 @@ func isKubeInfra(pod *v1.Pod) bool {
|
||||
}
|
||||
}
|
||||
return false
|
||||
+}
|
||||
+
|
||||
+// get the isolated CPUs (if any) from the devices associated with a specific container
|
||||
+func (p *staticPolicy) podIsolCPUs(pod *v1.Pod, container *v1.Container) cpuset.CPUSet {
|
||||
+ // NOTE: This is required for TestStaticPolicyAdd() since makePod() does
|
||||
+ // not create UID. We also need a way to properly stub devicemanager.
|
||||
+ if len(string(pod.UID)) == 0 {
|
||||
+ return cpuset.NewCPUSet()
|
||||
+ }
|
||||
+ resContDevices := p.deviceManager.GetDevices(string(pod.UID), container.Name)
|
||||
+ cpuSet := cpuset.NewCPUSet()
|
||||
+ for resourceName, resourceDevs := range resContDevices {
|
||||
+ // this resource name needs to match the isolcpus device plugin
|
||||
+ if resourceName == "windriver.com/isolcpus" {
|
||||
+ for devID, _ := range resourceDevs {
|
||||
+ cpuStrList := []string{devID}
|
||||
+ if len(cpuStrList) > 0 {
|
||||
+ // loop over the list of strings, convert each one to int, add to cpuset
|
||||
+ for _, cpuStr := range cpuStrList {
|
||||
+ cpu, err := strconv.Atoi(cpuStr)
|
||||
+ if err != nil {
|
||||
+ panic(err)
|
||||
+ }
|
||||
+ cpuSet = cpuSet.Union(cpuset.NewCPUSet(cpu))
|
||||
+ }
|
||||
+ }
|
||||
+ }
|
||||
+ }
|
||||
+ }
|
||||
+ return cpuSet
|
||||
}
|
||||
\ No newline at end of file
|
||||
diff --git a/pkg/kubelet/cm/cpumanager/policy_static_test.go b/pkg/kubelet/cm/cpumanager/policy_static_test.go
|
||||
index 90ad98e71bc..8511d5635bd 100644
|
||||
--- a/pkg/kubelet/cm/cpumanager/policy_static_test.go
|
||||
+++ b/pkg/kubelet/cm/cpumanager/policy_static_test.go
|
||||
@@ -27,6 +27,7 @@ import (
|
||||
"k8s.io/kubernetes/pkg/kubelet/cm/cpuset"
|
||||
"k8s.io/kubernetes/pkg/kubelet/cm/topologymanager"
|
||||
"k8s.io/kubernetes/pkg/kubelet/cm/topologymanager/bitmask"
|
||||
+ "k8s.io/kubernetes/pkg/kubelet/cm/devicemanager"
|
||||
)
|
||||
|
||||
type staticPolicyTest struct {
|
||||
@@ -45,8 +46,9 @@ type staticPolicyTest struct {
|
||||
}
|
||||
|
||||
func TestStaticPolicyName(t *testing.T) {
|
||||
+ testDM, _ := devicemanager.NewManagerStub()
|
||||
testExcl := false
|
||||
- policy, _ := NewStaticPolicy(topoSingleSocketHT, 1, cpuset.NewCPUSet(), topologymanager.NewFakeManager(), testExcl)
|
||||
+ policy, _ := NewStaticPolicy(topoSingleSocketHT, 1, cpuset.NewCPUSet(), cpuset.NewCPUSet(), topologymanager.NewFakeManager(), testDM, testExcl)
|
||||
|
||||
policyName := policy.Name()
|
||||
if policyName != "static" {
|
||||
@@ -56,6 +58,7 @@ func TestStaticPolicyName(t *testing.T) {
|
||||
}
|
||||
|
||||
func TestStaticPolicyStart(t *testing.T) {
|
||||
+ testDM, _ := devicemanager.NewManagerStub()
|
||||
testCases := []staticPolicyTest{
|
||||
{
|
||||
description: "non-corrupted state",
|
||||
@@ -131,7 +134,7 @@ func TestStaticPolicyStart(t *testing.T) {
|
||||
}
|
||||
for _, testCase := range testCases {
|
||||
t.Run(testCase.description, func(t *testing.T) {
|
||||
- p, _ := NewStaticPolicy(testCase.topo, testCase.numReservedCPUs, cpuset.NewCPUSet(), topologymanager.NewFakeManager(), testCase.excludeReserved)
|
||||
+ p, _ := NewStaticPolicy(testCase.topo, testCase.numReservedCPUs, cpuset.NewCPUSet(), cpuset.NewCPUSet(), topologymanager.NewFakeManager(), testDM, testCase.excludeReserved)
|
||||
policy := p.(*staticPolicy)
|
||||
st := &mockState{
|
||||
assignments: testCase.stAssignments,
|
||||
@@ -179,6 +182,7 @@ func TestStaticPolicyAdd(t *testing.T) {
|
||||
largeTopoSock0CPUSet := largeTopoSock0Builder.Result()
|
||||
largeTopoSock1CPUSet := largeTopoSock1Builder.Result()
|
||||
|
||||
+ testDM, _ := devicemanager.NewManagerStub()
|
||||
testCases := []staticPolicyTest{
|
||||
{
|
||||
description: "GuPodSingleCore, SingleSocketHT, ExpectError",
|
||||
@@ -447,7 +451,7 @@ func TestStaticPolicyAdd(t *testing.T) {
|
||||
}
|
||||
|
||||
for _, testCase := range testCases {
|
||||
- policy, _ := NewStaticPolicy(testCase.topo, testCase.numReservedCPUs, cpuset.NewCPUSet(), topologymanager.NewFakeManager(), testCase.excludeReserved)
|
||||
+ policy, _ := NewStaticPolicy(testCase.topo, testCase.numReservedCPUs, cpuset.NewCPUSet(), cpuset.NewCPUSet(), topologymanager.NewFakeManager(), testDM, testCase.excludeReserved)
|
||||
|
||||
st := &mockState{
|
||||
assignments: testCase.stAssignments,
|
||||
@@ -490,6 +494,7 @@ func TestStaticPolicyAdd(t *testing.T) {
|
||||
}
|
||||
|
||||
func TestStaticPolicyRemove(t *testing.T) {
|
||||
+ testDM, _ := devicemanager.NewManagerStub()
|
||||
excludeReserved := false
|
||||
testCases := []staticPolicyTest{
|
||||
{
|
||||
@@ -549,7 +554,7 @@ func TestStaticPolicyRemove(t *testing.T) {
|
||||
}
|
||||
|
||||
for _, testCase := range testCases {
|
||||
- policy, _ := NewStaticPolicy(testCase.topo, testCase.numReservedCPUs, cpuset.NewCPUSet(), topologymanager.NewFakeManager(), excludeReserved)
|
||||
+ policy, _ := NewStaticPolicy(testCase.topo, testCase.numReservedCPUs, cpuset.NewCPUSet(), cpuset.NewCPUSet(), topologymanager.NewFakeManager(), testDM, excludeReserved)
|
||||
|
||||
st := &mockState{
|
||||
assignments: testCase.stAssignments,
|
||||
@@ -571,6 +576,7 @@ func TestStaticPolicyRemove(t *testing.T) {
|
||||
}
|
||||
|
||||
func TestTopologyAwareAllocateCPUs(t *testing.T) {
|
||||
+ testDM, _ := devicemanager.NewManagerStub()
|
||||
excludeReserved := false
|
||||
testCases := []struct {
|
||||
description string
|
||||
@@ -640,7 +646,7 @@ func TestTopologyAwareAllocateCPUs(t *testing.T) {
|
||||
},
|
||||
}
|
||||
for _, tc := range testCases {
|
||||
- p, _ := NewStaticPolicy(tc.topo, 0, cpuset.NewCPUSet(), topologymanager.NewFakeManager(), excludeReserved)
|
||||
+ p, _ := NewStaticPolicy(tc.topo, 0, cpuset.NewCPUSet(), cpuset.NewCPUSet(), topologymanager.NewFakeManager(), testDM, excludeReserved)
|
||||
policy := p.(*staticPolicy)
|
||||
st := &mockState{
|
||||
assignments: tc.stAssignments,
|
||||
@@ -673,6 +679,7 @@ type staticPolicyTestWithResvList struct {
|
||||
topo *topology.CPUTopology
|
||||
numReservedCPUs int
|
||||
reserved cpuset.CPUSet
|
||||
+ isolcpus cpuset.CPUSet
|
||||
stAssignments state.ContainerCPUAssignments
|
||||
stDefaultCPUSet cpuset.CPUSet
|
||||
pod *v1.Pod
|
||||
@@ -713,9 +720,10 @@ func TestStaticPolicyStartWithResvList(t *testing.T) {
|
||||
},
|
||||
}
|
||||
testExcl := false
|
||||
+ testDM, _ := devicemanager.NewManagerStub()
|
||||
for _, testCase := range testCases {
|
||||
t.Run(testCase.description, func(t *testing.T) {
|
||||
- p, err := NewStaticPolicy(testCase.topo, testCase.numReservedCPUs, testCase.reserved, topologymanager.NewFakeManager(), testExcl)
|
||||
+ p, err := NewStaticPolicy(testCase.topo, testCase.numReservedCPUs, testCase.reserved, cpuset.NewCPUSet(), topologymanager.NewFakeManager(), testDM, testExcl)
|
||||
if !reflect.DeepEqual(err, testCase.expNewErr) {
|
||||
t.Errorf("StaticPolicy Start() error (%v). expected error: %v but got: %v",
|
||||
testCase.description, testCase.expNewErr, err)
|
||||
@@ -755,6 +763,7 @@ func TestStaticPolicyAddWithResvList(t *testing.T) {
|
||||
topo: topoSingleSocketHT,
|
||||
numReservedCPUs: 1,
|
||||
reserved: cpuset.NewCPUSet(0),
|
||||
+ isolcpus: cpuset.NewCPUSet(),
|
||||
stAssignments: state.ContainerCPUAssignments{},
|
||||
stDefaultCPUSet: cpuset.NewCPUSet(1, 2, 3, 4, 5, 6, 7),
|
||||
pod: makePod("fakePod", "fakeContainer2", "8000m", "8000m"),
|
||||
@@ -767,6 +776,7 @@ func TestStaticPolicyAddWithResvList(t *testing.T) {
|
||||
topo: topoSingleSocketHT,
|
||||
numReservedCPUs: 2,
|
||||
reserved: cpuset.NewCPUSet(0, 1),
|
||||
+ isolcpus: cpuset.NewCPUSet(),
|
||||
stAssignments: state.ContainerCPUAssignments{},
|
||||
stDefaultCPUSet: cpuset.NewCPUSet(2, 3, 4, 5, 6, 7),
|
||||
pod: makePod("fakePod", "fakeContainer2", "1000m", "1000m"),
|
||||
@@ -779,6 +789,7 @@ func TestStaticPolicyAddWithResvList(t *testing.T) {
|
||||
topo: topoSingleSocketHT,
|
||||
numReservedCPUs: 2,
|
||||
reserved: cpuset.NewCPUSet(0, 1),
|
||||
+ isolcpus: cpuset.NewCPUSet(),
|
||||
stAssignments: state.ContainerCPUAssignments{
|
||||
"fakePod": map[string]cpuset.CPUSet{
|
||||
"fakeContainer100": cpuset.NewCPUSet(2, 3, 6, 7),
|
||||
@@ -795,6 +806,7 @@ func TestStaticPolicyAddWithResvList(t *testing.T) {
|
||||
topo: topoSingleSocketHT,
|
||||
numReservedCPUs: 2,
|
||||
reserved: cpuset.NewCPUSet(0, 1),
|
||||
+ isolcpus: cpuset.NewCPUSet(),
|
||||
stAssignments: state.ContainerCPUAssignments{
|
||||
"fakePod": map[string]cpuset.CPUSet{
|
||||
"fakeContainer100": cpuset.NewCPUSet(2, 3, 6, 7),
|
||||
@@ -806,11 +818,29 @@ func TestStaticPolicyAddWithResvList(t *testing.T) {
|
||||
expCPUAlloc: true,
|
||||
expCSet: cpuset.NewCPUSet(0, 1),
|
||||
},
|
||||
+ {
|
||||
+ description: "InfraPod, SingleSocketHT, Isolcpus, ExpectAllocReserved",
|
||||
+ topo: topoSingleSocketHT,
|
||||
+ numReservedCPUs: 2,
|
||||
+ reserved: cpuset.NewCPUSet(0, 1),
|
||||
+ isolcpus: cpuset.NewCPUSet(1),
|
||||
+ stAssignments: state.ContainerCPUAssignments{
|
||||
+ "fakePod": map[string]cpuset.CPUSet{
|
||||
+ "fakeContainer100": cpuset.NewCPUSet(2, 3, 6, 7),
|
||||
+ },
|
||||
+ },
|
||||
+ stDefaultCPUSet: cpuset.NewCPUSet(4, 5),
|
||||
+ pod: infraPod,
|
||||
+ expErr: nil,
|
||||
+ expCPUAlloc: true,
|
||||
+ expCSet: cpuset.NewCPUSet(0),
|
||||
+ },
|
||||
}
|
||||
|
||||
testExcl := true
|
||||
+ testDM, _ := devicemanager.NewManagerStub()
|
||||
for _, testCase := range testCases {
|
||||
- policy, _ := NewStaticPolicy(testCase.topo, testCase.numReservedCPUs, testCase.reserved, topologymanager.NewFakeManager(), testExcl)
|
||||
+ policy, _ := NewStaticPolicy(testCase.topo, testCase.numReservedCPUs, testCase.reserved, testCase.isolcpus, topologymanager.NewFakeManager(), testDM, testExcl)
|
||||
|
||||
st := &mockState{
|
||||
assignments: testCase.stAssignments,
|
||||
--
|
||||
2.17.1
|
||||
|
||||
|
|
@ -0,0 +1,313 @@
|
|||
From 38ecddc618a62bc1b73bafd530b45e9bd8cac19e Mon Sep 17 00:00:00 2001
|
||||
From: Jim Gauld <james.gauld@windriver.com>
|
||||
Date: Tue, 7 Sep 2021 13:22:41 -0400
|
||||
Subject: [PATCH 2/7] kubelet cpumanager keep normal containers off reserved
|
||||
CPUs
|
||||
|
||||
When starting the kubelet process, two separate sets of reserved CPUs
|
||||
may be specified. With this change CPUs reserved via '--system-reserved=cpu'
|
||||
or '--kube-reserved=cpu' will be ignored by kubernetes itself. A small
|
||||
tweak to the default CPU affinity ensures that "normal" Kubernetes
|
||||
pods won't run on the reserved CPUs.
|
||||
|
||||
Signed-off-by: Jim Gauld <james.gauld@windriver.com>
|
||||
---
|
||||
pkg/kubelet/cm/cpumanager/cpu_manager.go | 6 +++-
|
||||
pkg/kubelet/cm/cpumanager/cpu_manager_test.go | 11 ++++--
|
||||
pkg/kubelet/cm/cpumanager/policy_static.go | 29 ++++++++++++---
|
||||
.../cm/cpumanager/policy_static_test.go | 35 +++++++++++++------
|
||||
4 files changed, 62 insertions(+), 19 deletions(-)
|
||||
|
||||
diff --git a/pkg/kubelet/cm/cpumanager/cpu_manager.go b/pkg/kubelet/cm/cpumanager/cpu_manager.go
|
||||
index f7b9c8d07bf..5b82e5b02ac 100644
|
||||
--- a/pkg/kubelet/cm/cpumanager/cpu_manager.go
|
||||
+++ b/pkg/kubelet/cm/cpumanager/cpu_manager.go
|
||||
@@ -178,7 +178,11 @@ func NewManager(cpuPolicyName string, reconcilePeriod time.Duration, machineInfo
|
||||
// exclusively allocated.
|
||||
reservedCPUsFloat := float64(reservedCPUs.MilliValue()) / 1000
|
||||
numReservedCPUs := int(math.Ceil(reservedCPUsFloat))
|
||||
- policy, err = NewStaticPolicy(topo, numReservedCPUs, specificCPUs, affinity)
|
||||
+ // NOTE: Set excludeReserved unconditionally to exclude reserved CPUs from default cpuset.
|
||||
+ // This variable is primarily to make testing easier.
|
||||
+ excludeReserved := true
|
||||
+ policy, err = NewStaticPolicy(topo, numReservedCPUs, specificCPUs, affinity, excludeReserved)
|
||||
+
|
||||
if err != nil {
|
||||
return nil, fmt.Errorf("new static policy error: %v", err)
|
||||
}
|
||||
diff --git a/pkg/kubelet/cm/cpumanager/cpu_manager_test.go b/pkg/kubelet/cm/cpumanager/cpu_manager_test.go
|
||||
index 51c6ad99251..a2e73de6875 100644
|
||||
--- a/pkg/kubelet/cm/cpumanager/cpu_manager_test.go
|
||||
+++ b/pkg/kubelet/cm/cpumanager/cpu_manager_test.go
|
||||
@@ -215,6 +215,7 @@ func makeMultiContainerPod(initCPUs, appCPUs []struct{ request, limit string })
|
||||
}
|
||||
|
||||
func TestCPUManagerAdd(t *testing.T) {
|
||||
+ testExcl := false
|
||||
testPolicy, _ := NewStaticPolicy(
|
||||
&topology.CPUTopology{
|
||||
NumCPUs: 4,
|
||||
@@ -229,7 +230,8 @@ func TestCPUManagerAdd(t *testing.T) {
|
||||
},
|
||||
0,
|
||||
cpuset.NewCPUSet(),
|
||||
- topologymanager.NewFakeManager())
|
||||
+ topologymanager.NewFakeManager(),
|
||||
+ testExcl)
|
||||
testCases := []struct {
|
||||
description string
|
||||
updateErr error
|
||||
@@ -477,8 +479,9 @@ func TestCPUManagerAddWithInitContainers(t *testing.T) {
|
||||
},
|
||||
}
|
||||
|
||||
+ testExcl := false
|
||||
for _, testCase := range testCases {
|
||||
- policy, _ := NewStaticPolicy(testCase.topo, testCase.numReservedCPUs, cpuset.NewCPUSet(), topologymanager.NewFakeManager())
|
||||
+ policy, _ := NewStaticPolicy(testCase.topo, testCase.numReservedCPUs, cpuset.NewCPUSet(), topologymanager.NewFakeManager(), testExcl)
|
||||
|
||||
mockState := &mockState{
|
||||
assignments: testCase.stAssignments,
|
||||
@@ -986,6 +989,7 @@ func TestReconcileState(t *testing.T) {
|
||||
// above test cases are without kubelet --reserved-cpus cmd option
|
||||
// the following tests are with --reserved-cpus configured
|
||||
func TestCPUManagerAddWithResvList(t *testing.T) {
|
||||
+ testExcl := false
|
||||
testPolicy, _ := NewStaticPolicy(
|
||||
&topology.CPUTopology{
|
||||
NumCPUs: 4,
|
||||
@@ -1000,7 +1004,8 @@ func TestCPUManagerAddWithResvList(t *testing.T) {
|
||||
},
|
||||
1,
|
||||
cpuset.NewCPUSet(0),
|
||||
- topologymanager.NewFakeManager())
|
||||
+ topologymanager.NewFakeManager(),
|
||||
+ testExcl)
|
||||
testCases := []struct {
|
||||
description string
|
||||
updateErr error
|
||||
diff --git a/pkg/kubelet/cm/cpumanager/policy_static.go b/pkg/kubelet/cm/cpumanager/policy_static.go
|
||||
index ec25a15a3c2..94d81796132 100644
|
||||
--- a/pkg/kubelet/cm/cpumanager/policy_static.go
|
||||
+++ b/pkg/kubelet/cm/cpumanager/policy_static.go
|
||||
@@ -75,6 +75,8 @@ type staticPolicy struct {
|
||||
topology *topology.CPUTopology
|
||||
// set of CPUs that is not available for exclusive assignment
|
||||
reserved cpuset.CPUSet
|
||||
+ // If true, default CPUSet should exclude reserved CPUs
|
||||
+ excludeReserved bool
|
||||
// topology manager reference to get container Topology affinity
|
||||
affinity topologymanager.Store
|
||||
// set of CPUs to reuse across allocations in a pod
|
||||
@@ -87,7 +89,7 @@ var _ Policy = &staticPolicy{}
|
||||
// NewStaticPolicy returns a CPU manager policy that does not change CPU
|
||||
// assignments for exclusively pinned guaranteed containers after the main
|
||||
// container process starts.
|
||||
-func NewStaticPolicy(topology *topology.CPUTopology, numReservedCPUs int, reservedCPUs cpuset.CPUSet, affinity topologymanager.Store) (Policy, error) {
|
||||
+func NewStaticPolicy(topology *topology.CPUTopology, numReservedCPUs int, reservedCPUs cpuset.CPUSet, affinity topologymanager.Store, excludeReserved bool) (Policy, error) {
|
||||
allCPUs := topology.CPUDetails.CPUs()
|
||||
var reserved cpuset.CPUSet
|
||||
if reservedCPUs.Size() > 0 {
|
||||
@@ -111,6 +113,7 @@ func NewStaticPolicy(topology *topology.CPUTopology, numReservedCPUs int, reserv
|
||||
return &staticPolicy{
|
||||
topology: topology,
|
||||
reserved: reserved,
|
||||
+ excludeReserved: excludeReserved,
|
||||
affinity: affinity,
|
||||
cpusToReuse: make(map[string]cpuset.CPUSet),
|
||||
}, nil
|
||||
@@ -139,7 +142,15 @@ func (p *staticPolicy) validateState(s state.State) error {
|
||||
}
|
||||
// state is empty initialize
|
||||
allCPUs := p.topology.CPUDetails.CPUs()
|
||||
- s.SetDefaultCPUSet(allCPUs)
|
||||
+ if p.excludeReserved {
|
||||
+ // Exclude reserved CPUs from the default CPUSet to keep containers off them
|
||||
+ // unless explicitly affined.
|
||||
+ s.SetDefaultCPUSet(allCPUs.Difference(p.reserved))
|
||||
+ } else {
|
||||
+ s.SetDefaultCPUSet(allCPUs)
|
||||
+ }
|
||||
+ klog.Infof("[cpumanager] static policy: CPUSet: allCPUs:%v, reserved:%v, default:%v\n",
|
||||
+ allCPUs, p.reserved, s.GetDefaultCPUSet())
|
||||
return nil
|
||||
}
|
||||
|
||||
@@ -147,9 +158,11 @@ func (p *staticPolicy) validateState(s state.State) error {
|
||||
// 1. Check if the reserved cpuset is not part of default cpuset because:
|
||||
// - kube/system reserved have changed (increased) - may lead to some containers not being able to start
|
||||
// - user tampered with file
|
||||
- if !p.reserved.Intersection(tmpDefaultCPUset).Equals(p.reserved) {
|
||||
- return fmt.Errorf("not all reserved cpus: \"%s\" are present in defaultCpuSet: \"%s\"",
|
||||
- p.reserved.String(), tmpDefaultCPUset.String())
|
||||
+ if !p.excludeReserved {
|
||||
+ if !p.reserved.Intersection(tmpDefaultCPUset).Equals(p.reserved) {
|
||||
+ return fmt.Errorf("not all reserved cpus: \"%s\" are present in defaultCpuSet: \"%s\"",
|
||||
+ p.reserved.String(), tmpDefaultCPUset.String())
|
||||
+ }
|
||||
}
|
||||
|
||||
// 2. Check if state for static policy is consistent
|
||||
@@ -178,6 +191,9 @@ func (p *staticPolicy) validateState(s state.State) error {
|
||||
}
|
||||
}
|
||||
totalKnownCPUs = totalKnownCPUs.UnionAll(tmpCPUSets)
|
||||
+ if p.excludeReserved {
|
||||
+ totalKnownCPUs = totalKnownCPUs.Union(p.reserved)
|
||||
+ }
|
||||
if !totalKnownCPUs.Equals(p.topology.CPUDetails.CPUs()) {
|
||||
return fmt.Errorf("current set of available CPUs \"%s\" doesn't match with CPUs in state \"%s\"",
|
||||
p.topology.CPUDetails.CPUs().String(), totalKnownCPUs.String())
|
||||
@@ -248,6 +264,9 @@ func (p *staticPolicy) RemoveContainer(s state.State, podUID string, containerNa
|
||||
klog.InfoS("Static policy: RemoveContainer", "podUID", podUID, "containerName", containerName)
|
||||
if toRelease, ok := s.GetCPUSet(podUID, containerName); ok {
|
||||
s.Delete(podUID, containerName)
|
||||
+ if p.excludeReserved {
|
||||
+ toRelease = toRelease.Difference(p.reserved)
|
||||
+ }
|
||||
// Mutate the shared pool, adding released cpus.
|
||||
s.SetDefaultCPUSet(s.GetDefaultCPUSet().Union(toRelease))
|
||||
}
|
||||
diff --git a/pkg/kubelet/cm/cpumanager/policy_static_test.go b/pkg/kubelet/cm/cpumanager/policy_static_test.go
|
||||
index c54997787b4..93a2870532a 100644
|
||||
--- a/pkg/kubelet/cm/cpumanager/policy_static_test.go
|
||||
+++ b/pkg/kubelet/cm/cpumanager/policy_static_test.go
|
||||
@@ -33,6 +33,7 @@ type staticPolicyTest struct {
|
||||
description string
|
||||
topo *topology.CPUTopology
|
||||
numReservedCPUs int
|
||||
+ excludeReserved bool
|
||||
podUID string
|
||||
containerName string
|
||||
stAssignments state.ContainerCPUAssignments
|
||||
@@ -44,7 +45,8 @@ type staticPolicyTest struct {
|
||||
}
|
||||
|
||||
func TestStaticPolicyName(t *testing.T) {
|
||||
- policy, _ := NewStaticPolicy(topoSingleSocketHT, 1, cpuset.NewCPUSet(), topologymanager.NewFakeManager())
|
||||
+ testExcl := false
|
||||
+ policy, _ := NewStaticPolicy(topoSingleSocketHT, 1, cpuset.NewCPUSet(), topologymanager.NewFakeManager(), testExcl)
|
||||
|
||||
policyName := policy.Name()
|
||||
if policyName != "static" {
|
||||
@@ -74,6 +76,15 @@ func TestStaticPolicyStart(t *testing.T) {
|
||||
stDefaultCPUSet: cpuset.NewCPUSet(),
|
||||
expCSet: cpuset.NewCPUSet(0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11),
|
||||
},
|
||||
+ {
|
||||
+ description: "empty cpuset exclude reserved",
|
||||
+ topo: topoDualSocketHT,
|
||||
+ numReservedCPUs: 2,
|
||||
+ excludeReserved: true,
|
||||
+ stAssignments: state.ContainerCPUAssignments{},
|
||||
+ stDefaultCPUSet: cpuset.NewCPUSet(),
|
||||
+ expCSet: cpuset.NewCPUSet(1, 2, 3, 4, 5, 7, 8, 9, 10, 11),
|
||||
+ },
|
||||
{
|
||||
description: "reserved cores 0 & 6 are not present in available cpuset",
|
||||
topo: topoDualSocketHT,
|
||||
@@ -120,7 +131,7 @@ func TestStaticPolicyStart(t *testing.T) {
|
||||
}
|
||||
for _, testCase := range testCases {
|
||||
t.Run(testCase.description, func(t *testing.T) {
|
||||
- p, _ := NewStaticPolicy(testCase.topo, testCase.numReservedCPUs, cpuset.NewCPUSet(), topologymanager.NewFakeManager())
|
||||
+ p, _ := NewStaticPolicy(testCase.topo, testCase.numReservedCPUs, cpuset.NewCPUSet(), topologymanager.NewFakeManager(), testCase.excludeReserved)
|
||||
policy := p.(*staticPolicy)
|
||||
st := &mockState{
|
||||
assignments: testCase.stAssignments,
|
||||
@@ -436,7 +447,7 @@ func TestStaticPolicyAdd(t *testing.T) {
|
||||
}
|
||||
|
||||
for _, testCase := range testCases {
|
||||
- policy, _ := NewStaticPolicy(testCase.topo, testCase.numReservedCPUs, cpuset.NewCPUSet(), topologymanager.NewFakeManager())
|
||||
+ policy, _ := NewStaticPolicy(testCase.topo, testCase.numReservedCPUs, cpuset.NewCPUSet(), topologymanager.NewFakeManager(), testCase.excludeReserved)
|
||||
|
||||
st := &mockState{
|
||||
assignments: testCase.stAssignments,
|
||||
@@ -479,6 +490,7 @@ func TestStaticPolicyAdd(t *testing.T) {
|
||||
}
|
||||
|
||||
func TestStaticPolicyRemove(t *testing.T) {
|
||||
+ excludeReserved := false
|
||||
testCases := []staticPolicyTest{
|
||||
{
|
||||
description: "SingleSocketHT, DeAllocOneContainer",
|
||||
@@ -537,7 +549,7 @@ func TestStaticPolicyRemove(t *testing.T) {
|
||||
}
|
||||
|
||||
for _, testCase := range testCases {
|
||||
- policy, _ := NewStaticPolicy(testCase.topo, testCase.numReservedCPUs, cpuset.NewCPUSet(), topologymanager.NewFakeManager())
|
||||
+ policy, _ := NewStaticPolicy(testCase.topo, testCase.numReservedCPUs, cpuset.NewCPUSet(), topologymanager.NewFakeManager(), excludeReserved)
|
||||
|
||||
st := &mockState{
|
||||
assignments: testCase.stAssignments,
|
||||
@@ -559,6 +571,7 @@ func TestStaticPolicyRemove(t *testing.T) {
|
||||
}
|
||||
|
||||
func TestTopologyAwareAllocateCPUs(t *testing.T) {
|
||||
+ excludeReserved := false
|
||||
testCases := []struct {
|
||||
description string
|
||||
topo *topology.CPUTopology
|
||||
@@ -627,7 +640,7 @@ func TestTopologyAwareAllocateCPUs(t *testing.T) {
|
||||
},
|
||||
}
|
||||
for _, tc := range testCases {
|
||||
- p, _ := NewStaticPolicy(tc.topo, 0, cpuset.NewCPUSet(), topologymanager.NewFakeManager())
|
||||
+ p, _ := NewStaticPolicy(tc.topo, 0, cpuset.NewCPUSet(), topologymanager.NewFakeManager(), excludeReserved)
|
||||
policy := p.(*staticPolicy)
|
||||
st := &mockState{
|
||||
assignments: tc.stAssignments,
|
||||
@@ -699,9 +712,10 @@ func TestStaticPolicyStartWithResvList(t *testing.T) {
|
||||
expNewErr: fmt.Errorf("[cpumanager] unable to reserve the required amount of CPUs (size of 0-1 did not equal 1)"),
|
||||
},
|
||||
}
|
||||
+ testExcl := false
|
||||
for _, testCase := range testCases {
|
||||
t.Run(testCase.description, func(t *testing.T) {
|
||||
- p, err := NewStaticPolicy(testCase.topo, testCase.numReservedCPUs, testCase.reserved, topologymanager.NewFakeManager())
|
||||
+ p, err := NewStaticPolicy(testCase.topo, testCase.numReservedCPUs, testCase.reserved, topologymanager.NewFakeManager(), testExcl)
|
||||
if !reflect.DeepEqual(err, testCase.expNewErr) {
|
||||
t.Errorf("StaticPolicy Start() error (%v). expected error: %v but got: %v",
|
||||
testCase.description, testCase.expNewErr, err)
|
||||
@@ -741,7 +755,7 @@ func TestStaticPolicyAddWithResvList(t *testing.T) {
|
||||
numReservedCPUs: 1,
|
||||
reserved: cpuset.NewCPUSet(0),
|
||||
stAssignments: state.ContainerCPUAssignments{},
|
||||
- stDefaultCPUSet: cpuset.NewCPUSet(0, 1, 2, 3, 4, 5, 6, 7),
|
||||
+ stDefaultCPUSet: cpuset.NewCPUSet(1, 2, 3, 4, 5, 6, 7),
|
||||
pod: makePod("fakePod", "fakeContainer2", "8000m", "8000m"),
|
||||
expErr: fmt.Errorf("not enough cpus available to satisfy request"),
|
||||
expCPUAlloc: false,
|
||||
@@ -753,7 +767,7 @@ func TestStaticPolicyAddWithResvList(t *testing.T) {
|
||||
numReservedCPUs: 2,
|
||||
reserved: cpuset.NewCPUSet(0, 1),
|
||||
stAssignments: state.ContainerCPUAssignments{},
|
||||
- stDefaultCPUSet: cpuset.NewCPUSet(0, 1, 2, 3, 4, 5, 6, 7),
|
||||
+ stDefaultCPUSet: cpuset.NewCPUSet(2, 3, 4, 5, 6, 7),
|
||||
pod: makePod("fakePod", "fakeContainer2", "1000m", "1000m"),
|
||||
expErr: nil,
|
||||
expCPUAlloc: true,
|
||||
@@ -769,7 +783,7 @@ func TestStaticPolicyAddWithResvList(t *testing.T) {
|
||||
"fakeContainer100": cpuset.NewCPUSet(2, 3, 6, 7),
|
||||
},
|
||||
},
|
||||
- stDefaultCPUSet: cpuset.NewCPUSet(0, 1, 4, 5),
|
||||
+ stDefaultCPUSet: cpuset.NewCPUSet(4, 5),
|
||||
pod: makePod("fakePod", "fakeContainer3", "2000m", "2000m"),
|
||||
expErr: nil,
|
||||
expCPUAlloc: true,
|
||||
@@ -777,8 +791,9 @@ func TestStaticPolicyAddWithResvList(t *testing.T) {
|
||||
},
|
||||
}
|
||||
|
||||
+ testExcl := true
|
||||
for _, testCase := range testCases {
|
||||
- policy, _ := NewStaticPolicy(testCase.topo, testCase.numReservedCPUs, testCase.reserved, topologymanager.NewFakeManager())
|
||||
+ policy, _ := NewStaticPolicy(testCase.topo, testCase.numReservedCPUs, testCase.reserved, topologymanager.NewFakeManager(), testExcl)
|
||||
|
||||
st := &mockState{
|
||||
assignments: testCase.stAssignments,
|
||||
--
|
||||
2.17.1
|
||||
|
||||
File diff suppressed because it is too large
Load Diff
File diff suppressed because it is too large
Load Diff
Loading…
Reference in New Issue