fix: use allowed cpus from target container

extcontainer/action_stress.go

@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// SPDX-FileCopyrightText: 2024 Steadybit GmbH
+// SPDX-FileCopyrightText: 2025 Steadybit GmbH
 
 package extcontainer
 
@@ -92,7 +92,7 @@ func (a *stressAction) Prepare(ctx context.Context, state *StressActionState, re
 		return nil, err
 	}
 
-	readAndAdaptToContainerLimits(ctx, processInfo.CGroupPath, &opts)
+	readAndAdaptToContainerLimits(ctx, processInfo, &opts)
 
 	state.StressOpts = opts
 	state.ExecutionId = request.ExecutionId

extcontainer/action_stress_limit_helpers.go

@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-// SPDX-FileCopyrightText: 2024 Steadybit GmbH
+// SPDX-FileCopyrightText: 2025 Steadybit GmbH
 
 package extcontainer
 
@@ -8,39 +8,56 @@ import (
 	"fmt"
 	"github.com/kataras/iris/v12/x/mathx"
 	"github.com/rs/zerolog/log"
+	"github.com/steadybit/action-kit/go/action_kit_commons/runc"
 	"github.com/steadybit/action-kit/go/action_kit_commons/stress"
+	"github.com/steadybit/action-kit/go/action_kit_commons/utils"
 	"github.com/steadybit/extension-kit/extutil"
 	"math"
 	"os"
 	"path/filepath"
-	"runtime"
 	"strconv"
 	"strings"
 )
 
 var cgroupV1MemUnlimited = (math.MaxInt64 / os.Getpagesize()) * os.Getpagesize()
 var osFs = osFileSystem{}
 
-func readAndAdaptToContainerLimits(_ context.Context, cGroupPath string, opts *stress.Opts) {
+func readAndAdaptToContainerLimits(_ context.Context, p runc.LinuxProcessInfo, opts *stress.Opts) {
 	cpuLimitInMilliCpu := -1
 	memLimitInBytes := -1
 
 	if isCGroupV1() {
-		cpuLimitInMilliCpu = readCGroupV1CpuLimit(cGroupPath, osFs)
-		memLimitInBytes = readCGroupV1MemLimit(cGroupPath, osFs)
+		cpuLimitInMilliCpu = readCGroupV1CpuLimit(p.CGroupPath, osFs)
+		memLimitInBytes = readCGroupV1MemLimit(p.CGroupPath, osFs)
 	} else {
-		cpuLimitInMilliCpu = readCGroupV2CpuLimit(cGroupPath, osFs)
-		memLimitInBytes = readCGroupV2MemLimit(cGroupPath, osFs)
+		cpuLimitInMilliCpu = readCGroupV2CpuLimit(p.CGroupPath, osFs)
+		memLimitInBytes = readCGroupV2MemLimit(p.CGroupPath, osFs)
 	}
 
-	if opts.CpuWorkers != nil && cpuLimitInMilliCpu >= 0 {
-		adaptToCpuContainerLimits(cpuLimitInMilliCpu, runtime.NumCPU(), opts)
+	if opts.CpuWorkers != nil {
+		if cpuLimitInMilliCpu >= 0 {
+			adaptToCpuContainerLimits(cpuLimitInMilliCpu, opts)
+		} else if *opts.CpuWorkers == 0 {
+			//there might be no limit set but the process to be restricted to certain CPUs or some CPUs programmatically turned off.
+			//In this case we need to read for the allowed list of CPUs for the process and pass this to the stress command as stress-ng
+			//always uses configured CPUs and not online CPUs
+			adaptToAllowedCpus(p.Pid, opts)
+		}
 	}
+
 	if opts.VmWorkers != nil && memLimitInBytes >= 0 {
 		adaptToMemContainerLimits(memLimitInBytes, opts)
 	}
 }
 
+func adaptToAllowedCpus(pid int, opts *stress.Opts) {
+	if cpuCount, err := utils.ReadCpusAllowedCount(fmt.Sprintf("/proc/%d/status", pid)); err != nil {
+		opts.CpuWorkers = extutil.Ptr(cpuCount)
+	} else {
+		log.Debug().Err(err).Msg("failed to read cpus_allowed count.")
+	}
+}
+
 func adaptToMemContainerLimits(memLimitInBytes int, opts *stress.Opts) {
 	memConsumptionInPercent := 0
 	if _, err := fmt.Sscanf(opts.VmBytes, "%d%%", &memConsumptionInPercent); err != nil {
@@ -55,9 +72,9 @@ func adaptToMemContainerLimits(memLimitInBytes int, opts *stress.Opts) {
 	log.Info().Msgf("container memory limit is %dK. Starting %d workers with memory consumption of %s each", memLimitInBytes/1024, *opts.VmWorkers, opts.VmBytes)
 }
 
-func adaptToCpuContainerLimits(cpuLimitInMilliCpu int, cpuCount int, opts *stress.Opts) {
+func adaptToCpuContainerLimits(cpuLimitInMilliCpu int, opts *stress.Opts) {
 	cpuLoadInMillis := cpuLimitInMilliCpu * opts.CpuLoad / 100
-	log.Debug().Int("cpuCount", cpuCount).Int("cpuLoad", opts.CpuLoad).Int("cpuLoadInMillis", cpuLoadInMillis).Msg("adapting to container cpu limit")
+	log.Debug().Int("cpuLoad", opts.CpuLoad).Int("cpuLoadInMillis", cpuLoadInMillis).Msg("adapting to container cpu limit")
 
 	if *opts.CpuWorkers == 0 {
 		// user didn't specify the number of workers. we start as many workers as we need to reach the desired cpu consumption

extcontainer/action_stress_limit_helpers_test.go

@@ -1,3 +1,6 @@
+// SPDX-License-Identifier: MIT
+// SPDX-FileCopyrightText: 2025 Steadybit GmbH
+
 package extcontainer
 
 import (
@@ -10,7 +13,6 @@ import (
 func Test_adaptToCpuContainerLimits(t *testing.T) {
 	type args struct {
 		cpuLimitInMilliCpu int
-		cpuCount           int
 		givenCpuWorkers    int
 		givenCpuLoad       int
 	}
@@ -27,7 +29,6 @@ func Test_adaptToCpuContainerLimits(t *testing.T) {
 			name: "worker-count not specified, desired cpu load can be handled by one worker",
 			args: args{
 				cpuLimitInMilliCpu: 200,
-				cpuCount:           4,
 				givenCpuLoad:       100,
 				givenCpuWorkers:    0,
 			},
@@ -40,7 +41,6 @@ func Test_adaptToCpuContainerLimits(t *testing.T) {
 			name: "worker-count not specified, desired cpu load needs multiple workers",
 			args: args{
 				cpuLimitInMilliCpu: 1500,
-				cpuCount:           4,
 				givenCpuLoad:       100,
 				givenCpuWorkers:    0,
 			},
@@ -53,7 +53,6 @@ func Test_adaptToCpuContainerLimits(t *testing.T) {
 			name: "worker-count not specified, desired 60% cpu fits to single worker",
 			args: args{
 				cpuLimitInMilliCpu: 1500,
-				cpuCount:           4,
 				givenCpuLoad:       60,
 				givenCpuWorkers:    0,
 			},
@@ -66,7 +65,6 @@ func Test_adaptToCpuContainerLimits(t *testing.T) {
 			name: "worker-count specified, desired 60% cpu is spread across workers",
 			args: args{
 				cpuLimitInMilliCpu: 1500,
-				cpuCount:           4,
 				givenCpuLoad:       60,
 				givenCpuWorkers:    3,
 			},
@@ -82,7 +80,7 @@ func Test_adaptToCpuContainerLimits(t *testing.T) {
 				CpuWorkers: &tt.args.givenCpuWorkers,
 				CpuLoad:    tt.args.givenCpuLoad,
 			}
-			adaptToCpuContainerLimits(tt.args.cpuLimitInMilliCpu, tt.args.cpuCount, &opts)
+			adaptToCpuContainerLimits(tt.args.cpuLimitInMilliCpu, &opts)
 			assert.Equal(t, tt.expected.adaptedCpuWorkers, *opts.CpuWorkers)
 			assert.Equal(t, tt.expected.adaptedCpuLoad, opts.CpuLoad)
 		})

go.mod

@@ -19,7 +19,7 @@ require (
 	github.com/opencontainers/runtime-spec v1.2.0
 	github.com/rs/zerolog v1.33.0
 	github.com/steadybit/action-kit/go/action_kit_api/v2 v2.9.6
-	github.com/steadybit/action-kit/go/action_kit_commons v1.2.17
+	github.com/steadybit/action-kit/go/action_kit_commons v1.2.18
 	github.com/steadybit/action-kit/go/action_kit_sdk v1.1.14
 	github.com/steadybit/action-kit/go/action_kit_test v1.3.2
 	github.com/steadybit/discovery-kit/go/discovery_kit_api v1.6.0

go.sum

@@ -257,8 +257,8 @@ github.com/spf13/pflag v1.0.5/go.mod h1:McXfInJRrz4CZXVZOBLb0bTZqETkiAhM9Iw0y3An
 github.com/spkg/bom v0.0.0-20160624110644-59b7046e48ad/go.mod h1:qLr4V1qq6nMqFKkMo8ZTx3f+BZEkzsRUY10Xsm2mwU0=
 github.com/steadybit/action-kit/go/action_kit_api/v2 v2.9.6 h1:Qci7Numf66mjCIRo7KDwHUimIxUZzq+GBfyv/1f/QCU=
 github.com/steadybit/action-kit/go/action_kit_api/v2 v2.9.6/go.mod h1:ycF2RLgRsB8I/jD52aE+dKZKVru1GIEtmkcRcIR3vXk=
-github.com/steadybit/action-kit/go/action_kit_commons v1.2.17 h1:NOKvnUA/iZo7mlcgSK+c8EHjxAEYCLSdyk8KVQSXmv8=
-github.com/steadybit/action-kit/go/action_kit_commons v1.2.17/go.mod h1:GFLcaf/WluBIqnq+iSCEGkEqxnfB9h1I16IBR1T7kNQ=
+github.com/steadybit/action-kit/go/action_kit_commons v1.2.18 h1:zg3c7b/Dl0M/odtNMbbhwVn/EbnvPZO14IPwbFy9WbY=
+github.com/steadybit/action-kit/go/action_kit_commons v1.2.18/go.mod h1:GFLcaf/WluBIqnq+iSCEGkEqxnfB9h1I16IBR1T7kNQ=
 github.com/steadybit/action-kit/go/action_kit_sdk v1.1.14 h1:x94RX+vh9Iyc0tS6BhiSpvknj+xE36AV0Nc3D5Yuub0=
 github.com/steadybit/action-kit/go/action_kit_sdk v1.1.14/go.mod h1:Tp/klK5b7k+BCvb3JTSDZSNcnpFBaHauhndzOarnMW4=
 github.com/steadybit/action-kit/go/action_kit_test v1.3.2 h1:DFDznoWEbTGv+fiGYiRaq7tq5es9VTScjrWusRAbS08=