balloons: add support for cpu frequency governor tuning

This commit introduces a new field in the balloons configuration CR, enabling users to specify CPU frequency governor mode preferences. Signed-off-by: Feruzjon Muyassarov <[email protected]>
containers · Oct 7, 2024 · 8a8d920 · 8a8d920
1 parent 7399adb
commit 8a8d920
Show file tree

Hide file tree

Showing 9 changed files with 43 additions and 0 deletions.
diff --git a/config/crd/bases/config.nri_balloonspolicies.yaml b/config/crd/bases/config.nri_balloonspolicies.yaml
@@ -265,6 +265,9 @@ spec:
                               description: EnergyPerformancePreference for CPUs in
                                 this class.
                               type: integer
+                            freqGovernor:
+                              description: CPUFreq Governor for this class.
+                              type: string
                             maxFreq:
                               description: MaxFreq is the maximum frequency for this
                                 class.

diff --git a/config/crd/bases/config.nri_templatepolicies.yaml b/config/crd/bases/config.nri_templatepolicies.yaml
@@ -55,6 +55,9 @@ spec:
                               description: EnergyPerformancePreference for CPUs in
                                 this class.
                               type: integer
+                            freqGovernor:
+                              description: CPUFreq Governor for this class.
+                              type: string
                             maxFreq:
                               description: MaxFreq is the maximum frequency for this
                                 class.

diff --git a/config/crd/bases/config.nri_topologyawarepolicies.yaml b/config/crd/bases/config.nri_topologyawarepolicies.yaml
@@ -69,6 +69,9 @@ spec:
                               description: EnergyPerformancePreference for CPUs in
                                 this class.
                               type: integer
+                            freqGovernor:
+                              description: CPUFreq Governor for this class.
+                              type: string
                             maxFreq:
                               description: MaxFreq is the maximum frequency for this
                                 class.

diff --git a/deployment/helm/balloons/crds/config.nri_balloonspolicies.yaml b/deployment/helm/balloons/crds/config.nri_balloonspolicies.yaml
@@ -265,6 +265,9 @@ spec:
                               description: EnergyPerformancePreference for CPUs in
                                 this class.
                               type: integer
+                            freqGovernor:
+                              description: CPUFreq Governor for this class.
+                              type: string
                             maxFreq:
                               description: MaxFreq is the maximum frequency for this
                                 class.

diff --git a/deployment/helm/template/crds/config.nri_templatepolicies.yaml b/deployment/helm/template/crds/config.nri_templatepolicies.yaml
@@ -55,6 +55,9 @@ spec:
                               description: EnergyPerformancePreference for CPUs in
                                 this class.
                               type: integer
+                            freqGovernor:
+                              description: CPUFreq Governor for this class.
+                              type: string
                             maxFreq:
                               description: MaxFreq is the maximum frequency for this
                                 class.

diff --git a/deployment/helm/topology-aware/crds/config.nri_topologyawarepolicies.yaml b/deployment/helm/topology-aware/crds/config.nri_topologyawarepolicies.yaml
@@ -69,6 +69,9 @@ spec:
                               description: EnergyPerformancePreference for CPUs in
                                 this class.
                               type: integer
+                            freqGovernor:
+                              description: CPUFreq Governor for this class.
+                              type: string
                             maxFreq:
                               description: MaxFreq is the maximum frequency for this
                                 class.

diff --git a/pkg/apis/config/v1alpha1/resmgr/control/cpu/config.go b/pkg/apis/config/v1alpha1/resmgr/control/cpu/config.go
@@ -30,4 +30,6 @@ type Class struct {
 	UncoreMinFreq uint `json:"uncoreMinFreq,omitempty"`
 	// UncoreMaxFreq is the maximum uncore frequency for this class.
 	UncoreMaxFreq uint `json:"uncoreMaxFreq,omitempty"`
+	// CPUFreq Governor for this class.
+	FreqGovernor string `json:"freqGovernor,omitempty"`
 }
diff --git a/pkg/resmgr/control/cpu/api.go b/pkg/resmgr/control/cpu/api.go
@@ -61,6 +61,9 @@ func Assign(c cache.Cache, class string, cpus ...int) error {
 		if err := ctl.enforceCpufreq(class, cpus...); err != nil {
 			log.Error("cpufreq enforcement failed: %v", err)
 		}
+		if err := ctl.enforceCpufreqGovernor(class, cpus...); err != nil {
+			log.Error("cpufreq governor enforcement failed: %v", err)
+		}
 		if err := ctl.enforceUncore(assignments, cpus...); err != nil {
 			log.Error("uncore frequency enforcement failed: %v", err)
 		}

diff --git a/pkg/resmgr/control/cpu/cpu.go b/pkg/resmgr/control/cpu/cpu.go
@@ -16,6 +16,7 @@ package cpu
 
 import (
 	"fmt"
+	"strconv"
 
 	"github.com/containers/nri-plugins/pkg/utils/cpuset"
 
@@ -122,6 +123,22 @@ func (ctl *cpuctl) PostStopHook(c cache.Container) error {
 	return nil
 }
 
+// enforceCpufreqGovernor enforces a class-specific cpufreq governor to a cpuset.
+func (ctl *cpuctl) enforceCpufreqGovernor(class string, cpusIds ...int) error {
+	if _, ok := ctl.classes[class]; !ok {
+		return fmt.Errorf("non-existent cpu class %q", class)
+	}
+	governor := ctl.classes[class].FreqGovernor
+	for cpu := range cpusIds {
+		log.Info(strconv.Itoa(cpu), governor)
+	}
+	log.Debug("enforcing cpu frequency governor %q on %v", governor, cpusIds)
+	if err := utils.SetScalingGovernorForCPUs(cpusIds, governor); err != nil {
+		return fmt.Errorf("Cannot set cpufreq governor %d: %w", governor, err)
+	}
+	return nil
+}
+
 // enforceCpufreq enforces a class-specific cpufreq configuration to a cpuset
 func (ctl *cpuctl) enforceCpufreq(class string, cpus ...int) error {
 	if _, ok := ctl.classes[class]; !ok {
@@ -259,6 +276,9 @@ func (ctl *cpuctl) configure(cfg *cfgapi.Config) error {
 			if err := ctl.enforceCpufreq(class, cpus.SortedMembers()...); err != nil {
 				log.Error("cpufreq enforcement on re-configure failed: %v", err)
 			}
+			if err := ctl.enforceCpufreqGovernor(class, cpus.SortedMembers()...); err != nil {
+				log.Error("cpufreq enforcement on re-configure failed: %v", err)
+			}
 		} else {
 			// TODO: what should we really do with classes that do not exist in
 			// the configuration anymore? Now we remember the CPUs assigned to