Added CPU Profiling Support via pprof

This adds an option to enable cpu profiling using pprof at startup. The profiling data is dumped to a file when the daemon is stopped. Signed-off-by: Blake Devcich <[email protected]>
NearNodeFlash · Dec 13, 2023 · ff6a3ac · ff6a3ac
1 parent 9c7d43c
commit ff6a3ac
Show file tree

Hide file tree

Showing 2 changed files with 39 additions and 12 deletions.
diff --git a/daemons/compute/server/main.go b/daemons/compute/server/main.go
@@ -27,7 +27,9 @@ import (
 	"os"
 	"os/signal"
 	"runtime"
+	"runtime/pprof"
 	"syscall"
+	"time"
 
 	"github.com/takama/daemon"
 	"google.golang.org/grpc"
@@ -49,7 +51,7 @@ type Service struct {
 	daemon.Daemon
 }
 
-func (service *Service) Manage() (string, error) {
+func (service *Service) Manage() (msg string, err error) {
 
 	if len(os.Args) > 1 {
 		command := os.Args[1]
@@ -99,6 +101,27 @@ func (service *Service) Manage() (string, error) {
 		return fmt.Sprintf("Failed to set permissions on socket %s", *socketAddr), err
 	}
 
+	// Enable CPU profiling with pprof
+	if len(options.CpuProfile) > 0 {
+		filename := options.CpuProfile + "-" + time.Now().UTC().Format(time.RFC3339)
+		f, err := os.Create(filename)
+		if err != nil {
+			return fmt.Sprintf("could not create CPU profile"), err
+		}
+		defer func() {
+			if cerr := f.Close(); cerr != nil {
+				err = cerr
+				msg = "could not close CPU profile"
+			}
+		}()
+
+		if err := pprof.StartCPUProfile(f); err != nil {
+			return fmt.Sprintf("could not start CPU profile"), err
+		}
+		errlog.Printf("CPU profiling enabled (%s)\n", filename)
+		defer pprof.StopCPUProfile()
+	}
+
 	go service.Run(server, listener)
 
 	for {

diff --git a/daemons/compute/server/servers/server.go b/daemons/compute/server/servers/server.go
@@ -32,24 +32,26 @@ type ServerOptions struct {
 	tokenFile string
 	certFile  string
 
-	name      string
-	nodeName  string
-	sysConfig string
-	simulated bool
+	name       string
+	nodeName   string
+	sysConfig  string
+	CpuProfile string
+	simulated  bool
 
 	k8sQPS   int
 	k8sBurst int
 }
 
 func GetOptions() (*ServerOptions, error) {
 	opts := ServerOptions{
-		host:      os.Getenv("KUBERNETES_SERVICE_HOST"),
-		port:      os.Getenv("KUBERNETES_SERVICE_PORT"),
-		name:      os.Getenv("NODE_NAME"),
-		nodeName:  os.Getenv("NNF_NODE_NAME"),
-		tokenFile: os.Getenv("NNF_DATA_MOVEMENT_SERVICE_TOKEN_FILE"),
-		certFile:  os.Getenv("NNF_DATA_MOVEMENT_SERVICE_CERT_FILE"),
-		simulated: false,
+		host:       os.Getenv("KUBERNETES_SERVICE_HOST"),
+		port:       os.Getenv("KUBERNETES_SERVICE_PORT"),
+		name:       os.Getenv("NODE_NAME"),
+		nodeName:   os.Getenv("NNF_NODE_NAME"),
+		tokenFile:  os.Getenv("NNF_DATA_MOVEMENT_SERVICE_TOKEN_FILE"),
+		certFile:   os.Getenv("NNF_DATA_MOVEMENT_SERVICE_CERT_FILE"),
+		CpuProfile: os.Getenv("NNF_DATA_MOVEMENT_SERVICE_CPU_PROFILE"),
+		simulated:  false,
 
 		// These options adjust the client-side rate-limiting for k8s. The new defaults are 50 and
 		// 100 (rather than 5, 10). See more info https://github.com/kubernetes/kubernetes/pull/116121
@@ -68,6 +70,8 @@ func GetOptions() (*ServerOptions, error) {
 	flag.BoolVar(&opts.simulated, "simulated", opts.simulated, "Run in simulation mode where no requests are sent to the server")
 	flag.IntVar(&opts.k8sQPS, "kubernetes-qps", opts.k8sQPS, "Kubernetes client queries per second (QPS)")
 	flag.IntVar(&opts.k8sBurst, "kubernetes-burst", opts.k8sBurst, "Kubernetes client additional concurrent calls above QPS")
+	flag.StringVar(&opts.CpuProfile, "cpu-profile", opts.CpuProfile,
+		"Enable and dump CPU profiling data to this file after daemon is stopped. Timestamp is added to end of filename.")
 	flag.Parse()
 	return &opts, nil
 }