Initial work on rapl-daemon. Initial server set up. API to read max power per zone and API to write new power cap have both been written.

rapl-daemon/README.md

@@ -0,0 +1,13 @@
+# RAPL Daemon
+
+This runs a server that is capable changing the percentage at which
+a node is being throttled using RAPL. This daemon should be installed
+on all worker nodes.
+
+### Sample payload for testing:
+```
+ curl --header "Content-Type: application/json" \
+  --request POST \
+  --data '{"percentage":75}' \
+  http://localhost:9090/powercap
+ ```

rapl-daemon/main.go

@@ -0,0 +1,49 @@
+package main
+
+import (
+	"encoding/json"
+	"fmt"
+	"html"
+	"log"
+	"net/http"
+)
+
+const powercapDir = "/sys/class/powercap/"
+
+// Cap is a payload that is expected from Elektron to cap a node
+type Cap struct {
+	Percentage int
+}
+
+func main() {
+	http.HandleFunc("/", func(w http.ResponseWriter, r *http.Request) {
+		fmt.Fprintf(w, "Unsupported endpoint %s", html.EscapeString(r.URL.Path))
+	})
+
+	http.HandleFunc("/powercap", powercapEndpoint)
+	log.Fatal(http.ListenAndServe(":9090", nil))
+}
+
+// Handler powercapping HTTP API endpoint
+func powercapEndpoint(w http.ResponseWriter, r *http.Request) {
+	var payload Cap
+	decoder := json.NewDecoder(r.Body)
+	err := decoder.Decode(&payload)
+	if err != nil {
+		http.Error(w, "Error parsing payload: "+err.Error(), 400)
+		return
+	}
+
+	if payload.Percentage < 0 || payload.Percentage > 100 {
+		http.Error(w, "Bad payload: percentage must be between 0 and 100", 400)
+		return
+	}
+
+	err = capNode(powercapDir, payload.Percentage)
+	if err != nil {
+		http.Error(w, err.Error(), 400)
+		return
+	}
+
+	fmt.Fprintf(w, "Capped node at %d percent", payload.Percentage)
+}

rapl-daemon/util.go

@@ -0,0 +1,88 @@
+package main
+
+import (
+	"fmt"
+	"io/ioutil"
+	"math"
+	"path/filepath"
+	"strconv"
+	"strings"
+)
+
+const raplPrefixCPU = "intel-rapl"
+
+// constraint_0 is usually the longer window while constraint_1 is usually the longer window
+const maxPowerFileShortWindow = "constraint_0_max_power_uw"
+const powerLimitFileShortWindow = "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 {
+
+	if percentage <= 0 || percentage > 100 {
+		return 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
+	}
+
+	for _, file := range files {
+
+		fields := strings.Split(file.Name(), ":")
+
+		// Fields should be in the form intel-rapl:X where X is the power zone
+		// We ignore sub-zones which follow the form intel-rapl:X:Y
+		if len(fields) != 2 {
+			continue
+		}
+
+		if fields[0] == raplPrefixCPU {
+			maxPower, err := maxPower(filepath.Join(base, file.Name(), maxPowerFileShortWindow))
+			if err != nil {
+				fmt.Println("unable to retreive max power for zone ", err)
+				continue
+			}
+
+			// We use floats to mitigate the possibility of an integer overflows.
+			powercap := uint(math.Ceil(float64(maxPower) * (float64(percentage) / 100)))
+
+			err = capZone(filepath.Join(base, file.Name(), powerLimitFileShortWindow), powercap)
+			if err != nil {
+				fmt.Println("unable to write powercap value: ", err)
+				continue
+			}
+			fmt.Println(file.Name(), ": ", int(maxPower), ", ", int(powercap))
+		}
+	}
+
+	return nil
+}
+
+// maxPower returns the value in float of the maximum watts a power zone
+// can use.
+func maxPower(zone string) (uint64, error) {
+	maxPower, err := ioutil.ReadFile(zone)
+	if err != nil {
+		return 0.0, err
+	}
+
+	maxPoweruW, err := strconv.ParseUint(strings.TrimSpace(string(maxPower)), 10, 64)
+	if err != nil {
+		return 0.0, err
+	}
+
+	return maxPoweruW, nil
+
+}
+
+// capZone caps a power zone to a specific amount of watts specified by value
+func capZone(zone string, value uint) error {
+	err := ioutil.WriteFile(zone, []byte(string(value)), 0644)
+	if err != nil {
+		return err
+	}
+	return nil
+}

rapl-daemon/util_test.go

@@ -0,0 +1,14 @@
+package main
+
+import "testing"
+
+// TODO(rdelvalle): Add more thourough testing. Generate mock files
+// that mimic the powercap subsystem and create test to operate on it.
+func TestCap(t *testing.T) {
+
+	err := capNode("/sys/devices/virtual/powercap/intel-rapl", 95)
+
+	if err != nil {
+		t.Fail()
+	}
+}