feat(guest): add cpu_iowait_ratio to Metrics RPC

Author: syscod3

internal/guest/metrics_linux.go

@@ -12,27 +12,12 @@ import (
 	"time"
 )
 
-// cpuUsage samples /proc/stat twice 100ms apart and returns the ratio 0.0–1.0.
-func cpuUsage() (float64, error) {
-	s1, err := readCPUStat()
-	if err != nil {
-		return 0, err
-	}
-	time.Sleep(100 * time.Millisecond)
-	s2, err := readCPUStat()
-	if err != nil {
-		return 0, err
-	}
-	total := float64(s2.total - s1.total)
-	idle := float64(s2.idle - s1.idle)
-	if total == 0 {
-		return 0, nil
-	}
-	return (total - idle) / total, nil
+type cpuStat struct {
+	total  uint64
+	idle   uint64
+	iowait uint64
 }
 
-type cpuStat struct{ total, idle uint64 }
-
 func readCPUStat() (cpuStat, error) {
 	f, err := os.Open("/proc/stat")
 	if err != nil {
@@ -59,11 +44,38 @@ func readCPUStat() (cpuStat, error) {
 		if len(vals) > 3 {
 			idle = vals[3]
 		}
-		return cpuStat{total: total, idle: idle}, nil
+		iowait := uint64(0)
+		if len(vals) > 4 {
+			iowait = vals[4]
+		}
+		return cpuStat{total: total, idle: idle, iowait: iowait}, nil
 	}
 	return cpuStat{}, fmt.Errorf("cpu line not found in /proc/stat")
 }
 
+// cpuAndIOWaitUsage samples /proc/stat twice 100ms apart and returns both
+// the CPU usage ratio and the iowait ratio (both 0.0–1.0).
+func cpuAndIOWaitUsage() (cpuRatio, iowaitRatio float64, err error) {
+	s1, err := readCPUStat()
+	if err != nil {
+		return
+	}
+	time.Sleep(100 * time.Millisecond)
+	s2, err := readCPUStat()
+	if err != nil {
+		return
+	}
+	total := float64(s2.total - s1.total)
+	if total == 0 {
+		return
+	}
+	idle := float64(s2.idle - s1.idle)
+	iowait := float64(s2.iowait - s1.iowait)
+	cpuRatio = (total - idle) / total
+	iowaitRatio = iowait / total
+	return
+}
+
 // memUsedBytes returns MemTotal - MemAvailable from /proc/meminfo.
 func memUsedBytes() (int64, error) {
 	f, err := os.Open("/proc/meminfo")

internal/guest/server.go

@@ -84,9 +84,10 @@ func (s *Server) HTTPCheck(_ context.Context, req *pb.HTTPCheckRequest) (*pb.HTT
 
 // Metrics reads CPU, memory and disk usage from /proc and syscall.
 func (s *Server) Metrics(_ context.Context, _ *pb.MetricsRequest) (*pb.MetricsResponse, error) {
-	cpu, err := cpuUsage()
+	cpu, iowait, err := cpuAndIOWaitUsage()
 	if err != nil {
 		cpu = 0
+		iowait = 0
 	}
 	mem, err := memUsedBytes()
 	if err != nil {
@@ -98,6 +99,7 @@ func (s *Server) Metrics(_ context.Context, _ *pb.MetricsRequest) (*pb.MetricsRe
 	}
 	return &pb.MetricsResponse{
 		CpuUsageRatio:   cpu,
+		CpuIowaitRatio:  iowait,
 		MemoryUsedBytes: mem,
 		DiskUsedBytes:   disk,
 	}, nil

internal/guest/server_test.go

@@ -121,6 +121,17 @@ func TestMetrics_returns(t *testing.T) {
 	}
 }
 
+func TestMetrics_iowait(t *testing.T) {
+	client := startTestServer(t)
+	resp, err := client.Metrics(context.Background(), &pb.MetricsRequest{})
+	if err != nil {
+		t.Fatal(err)
+	}
+	if resp.CpuIowaitRatio < 0 || resp.CpuIowaitRatio > 1 {
+		t.Errorf("cpu_iowait_ratio = %f, want 0.0–1.0", resp.CpuIowaitRatio)
+	}
+}
+
 func TestHTTPCheck_connectionRefused(t *testing.T) {
 	// Use a port we know has no listener (bind then close to get a free port)
 	l, err := net.Listen("tcp", "127.0.0.1:0")

internal/proto/guest/guest.pb.go

@@ -35,7 +35,8 @@ message HTTPCheckResponse {
 message MetricsRequest {}
 
 message MetricsResponse {
-  double cpu_usage_ratio   = 1;
-  int64  memory_used_bytes = 2;
-  int64  disk_used_bytes   = 3;
+  double cpu_usage_ratio    = 1;
+  int64  memory_used_bytes  = 2;
+  int64  disk_used_bytes    = 3;
+  double cpu_iowait_ratio   = 4;
 }