refactor(injector): mutualise stop and wait procedure.

Author: Zenithar

injector/cpu_pressure.go

@@ -7,11 +7,8 @@ package injector
 
 import (
 	"context"
-	"errors"
 	"fmt"
 	"math"
-	"os"
-	"time"
 
 	"github.com/DataDog/chaos-controller/api/v1beta1"
 	"github.com/DataDog/chaos-controller/command"
@@ -114,22 +111,5 @@ func (i *cpuPressureInjector) UpdateConfig(config Config) {
 }
 
 func (i *cpuPressureInjector) Clean() error {
-	if i.backgroundCmd == nil {
-		return nil
-	}
-
-	defer i.cancel()
-
-	if err := i.backgroundCmd.Stop(); err != nil && !errors.Is(err, os.ErrProcessDone) {
-		return fmt.Errorf("unable to stop background process: %w", err)
-	}
-
-	// Wait for the child process to fully exit before a potential re-inject during pulse mode.
-	select {
-	case <-i.backgroundCmd.Done():
-	case <-time.After(5 * time.Second):
-		i.config.Log.Warnw("timed out waiting for background process to exit")
-	}
-
-	return nil
+	return stopAndWaitForBackgroundCmd(i.config.Log, i.backgroundCmd, i.cancel)
 }

injector/injector.go

@@ -7,10 +7,14 @@ package injector
 
 import (
 	"context"
+	"errors"
+	"fmt"
+	"os"
 	"time"
 
 	chaosapi "github.com/DataDog/chaos-controller/api"
 	"github.com/DataDog/chaos-controller/cgroup"
+	"github.com/DataDog/chaos-controller/command"
 	"github.com/DataDog/chaos-controller/container"
 	"github.com/DataDog/chaos-controller/netns"
 	"github.com/DataDog/chaos-controller/o11y/metrics"
@@ -65,3 +69,25 @@ func (c Config) TargetName() string {
 
 	return UnknownTargetName
 }
+
+// stopAndWaitForBackgroundCmd stops a background command and waits for the process to fully exit
+// before returning. This prevents cgroup race conditions during pulse mode re-injection.
+func stopAndWaitForBackgroundCmd(log *zap.SugaredLogger, backgroundCmd command.BackgroundCmd, cancel context.CancelFunc) error {
+	if backgroundCmd == nil {
+		return nil
+	}
+
+	defer cancel()
+
+	if err := backgroundCmd.Stop(); err != nil && !errors.Is(err, os.ErrProcessDone) {
+		return fmt.Errorf("unable to stop background process: %w", err)
+	}
+
+	select {
+	case <-backgroundCmd.Done():
+	case <-time.After(5 * time.Second):
+		log.Warnw("timed out waiting for background process to exit")
+	}
+
+	return nil
+}

injector/memory_pressure.go

@@ -7,9 +7,7 @@ package injector
 
 import (
 	"context"
-	"errors"
 	"fmt"
-	"os"
 	"time"
 
 	"github.com/DataDog/chaos-controller/api/v1beta1"
@@ -97,23 +95,5 @@ func (i *memoryPressureInjector) UpdateConfig(config Config) {
 }
 
 func (i *memoryPressureInjector) Clean() error {
-	if i.backgroundCmd == nil {
-		return nil
-	}
-
-	defer i.cancel()
-
-	if err := i.backgroundCmd.Stop(); err != nil && !errors.Is(err, os.ErrProcessDone) {
-		return fmt.Errorf("unable to stop background process: %w", err)
-	}
-
-	// Wait for the child process to fully exit so cgroup memory accounting is updated
-	// before a potential re-inject during pulse mode.
-	select {
-	case <-i.backgroundCmd.Done():
-	case <-time.After(5 * time.Second):
-		i.config.Log.Warnw("timed out waiting for background process to exit")
-	}
-
-	return nil
+	return stopAndWaitForBackgroundCmd(i.config.Log, i.backgroundCmd, i.cancel)
 }