rapl-daemon/README.md

@@ -11,3 +11,29 @@ on all worker nodes.
   --data '{"percentage":75}' \
   http://localhost:9090/powercap
  ```
+
+### Payload
+
+```json
+{
+    "percentage":75
+}
+```
+
+### Response
+
+The daemon will respond with a json payload containing zones that were
+successfully capped as well as the zones that were not capped.
+
+```json
+{
+    "cappedZones": null,
+    "failedZones": [
+        "intel-rapl:0",
+        "intel-rapl:1"
+    ],
+    "error": "some zones were not able to be powercapped"
+}
+```
+
+Field error will not exist if failed zones is empty.

rapl-daemon/main.go

@@ -15,6 +15,13 @@ type Cap struct {
 	Percentage int
 }
 
+// CapResponse is the payload sent with information about the capping call
+type CapResponse struct {
+	CappedZones []string `json:"cappedZones"`
+	FailedZones []string `json:"failedZones"`
+	Error       *string  `json:"error"`
+}
+
 func main() {
 	http.HandleFunc("/", func(w http.ResponseWriter, r *http.Request) {
 		fmt.Fprintf(w, "Unsupported endpoint %s", html.EscapeString(r.URL.Path))
@@ -26,19 +33,28 @@ func main() {
 
 // Handler for the powercapping HTTP API endpoint.
 func powercapEndpoint(w http.ResponseWriter, r *http.Request) {
+	w.Header().Set("Content-Type", "application/json")
+
 	var payload Cap
+	var response CapResponse
+
 	decoder := json.NewDecoder(r.Body)
 	err := decoder.Decode(&payload)
 	if err != nil {
-		http.Error(w, "error parsing payload: "+err.Error(), 400)
+		errorMsg := "error parsing payload: " + err.Error()
+		response.Error = &errorMsg
+		json.NewEncoder(w).Encode(response)
 		return
 	}
 
-	err = capNode(powercapDir, payload.Percentage)
+	cappedZones, failedZones, err := capNode(powercapDir, payload.Percentage)
 	if err != nil {
-		http.Error(w, err.Error(), 400)
-		return
+		errorMsg := err.Error()
+		response.Error = &errorMsg
 	}
 
-	fmt.Fprintf(w, "capped node at %d percent", payload.Percentage)
+	response.CappedZones = cappedZones
+	response.FailedZones = failedZones
+
+	json.NewEncoder(w).Encode(response)
 }

rapl-daemon/util.go

@@ -19,19 +19,19 @@ const powerLimitFileLongWindow = "constraint_0_power_limit_uw"
 // capNode uses pseudo files made available by the Linux kernel
 // in order to capNode CPU power. More information is available at:
 // https://www.kernel.org/doc/html/latest/power/powercap/powercap.html
-func capNode(base string, percentage int) error {
+func capNode(base string, percentage int) ([]string, []string, error) {
 
 	if percentage <= 0 || percentage > 100 {
-		return fmt.Errorf("cap percentage must be between (0, 100]: %d", percentage)
+		return nil, nil, fmt.Errorf("cap percentage must be between 0 (non-inclusive) and 100 (inclusive): %d", percentage)
 	}
 
 	files, err := ioutil.ReadDir(base)
 	if err != nil {
-		return err
+		return nil, nil, err
 	}
 
+	var capped, failed []string
 	for _, file := range files {
-
 		fields := strings.Split(file.Name(), ":")
 
 		// Fields should be in the form intel-rapl:X where X is the power zone
@@ -43,22 +43,28 @@ func capNode(base string, percentage int) error {
 		if fields[0] == raplPrefixCPU {
 			maxPower, err := maxPower(filepath.Join(base, file.Name(), maxPowerFileLongWindow))
 			if err != nil {
+				failed = append(failed, file.Name())
 				fmt.Println("unable to retreive max power for zone ", err)
 				continue
 			}
 
-			// We use floats to mitigate the possibility of an integer overflows.
+			// We use floats to mitigate the possibility of an integer overflow.
 			powercap := uint64(math.Ceil(float64(maxPower) * (float64(percentage) / 100)))
 
-			err = capZone(filepath.Join(base, file.Name(), powerLimitFileLongWindow), powercap)
-			if err != nil {
+			if err := capZone(filepath.Join(base, file.Name(), powerLimitFileLongWindow), powercap); err != nil {
+				failed = append(failed, file.Name())
 				fmt.Println("unable to write powercap value: ", err)
 				continue
 			}
+			capped = append(capped, file.Name())
 		}
 	}
 
-	return nil
+	if len(failed) > 0 {
+		return capped, failed, fmt.Errorf("some zones were not able to be powercapped")
+	}
+
+	return capped, nil, nil
 }
 
 // maxPower returns the value in float of the maximum watts a power zone can use.
@@ -88,3 +94,17 @@ func capZone(limitFile string, value uint64) error {
 	}
 	return nil
 }
+
+func currentCap(limit string) (uint64, error) {
+	powercap, err := ioutil.ReadFile(limit)
+	if err != nil {
+		return 0, err
+	}
+
+	powercapuW, err := strconv.ParseUint(strings.TrimSpace(string(powercap)), 10, 64)
+	if err != nil {
+		return 0, err
+	}
+
+	return powercapuW, nil
+}

rapl-daemon/util_test.go

@@ -7,7 +7,6 @@ import (
 	"os"
 	"path/filepath"
 	"strconv"
-	"strings"
 	"testing"
 
 	"github.com/stretchr/testify/assert"
@@ -48,18 +47,25 @@ func TestMain(m *testing.M) {
 	os.Exit(m.Run())
 }
 
+// TODO(rdelvalle): Add tests where capping fails
 func TestCapNode(t *testing.T) {
-	err := capNode(raplDir, 95)
+	capped, failed, err := capNode(raplDir, 95)
 	assert.NoError(t, err)
+	assert.Len(t, capped, 1)
+	assert.Nil(t, failed)
 
 	t.Run("bad-percentage", func(t *testing.T) {
-		err := capNode(raplDir, 1000)
+		capped, failed, err := capNode(raplDir, 1000)
 		assert.Error(t, err)
+		assert.Nil(t, capped)
+		assert.Nil(t, failed)
 	})
 
 	t.Run("zero-percent", func(t *testing.T) {
-		err := capNode(raplDir, 0)
+		capped, failed, err := capNode(raplDir, 0)
 		assert.Error(t, err)
+		assert.Nil(t, capped)
+		assert.Nil(t, failed)
 	})
 }
 
@@ -84,10 +90,7 @@ func TestCapZone(t *testing.T) {
 	err := capZone(limitFile, powercap)
 	assert.NoError(t, err)
 
-	newCapBytes, err := ioutil.ReadFile(limitFile)
-	assert.NoError(t, err)
-
-	newCap, err := strconv.ParseUint(strings.TrimSpace(string(newCapBytes)), 10, 64)
+	newCap, err := currentCap(limitFile)
 	assert.NoError(t, err)
 	assert.Equal(t, powercap, newCap)