make gpud port number configurable

cmd/gpud/command/command.go

@@ -176,7 +176,7 @@ sudo rm /etc/systemd/system/gpud.service
 				&cli.StringFlag{
 					Name:  "listen-address",
 					Usage: "set the listen address",
-					Value: fmt.Sprintf("0.0.0.0:%d", pkgconfig.DefaultGPUdPort),
+					Value: fmt.Sprintf("0.0.0.0:%d", pkgconfig.GPUdPortNumber()),
 				},
 				&cli.BoolFlag{
 					Name:  "pprof",

cmd/gpud/compact/command.go

@@ -47,9 +47,9 @@ func Command(cliContext *cli.Context) error {
 		}
 	}
 
-	portOpen := isPortOpen(config.DefaultGPUdPort)
+	portOpen := isPortOpen(config.GPUdPortNumber())
 	if portOpen {
-		return fmt.Errorf("gpud is running on port %d (must be stopped before running compact)", config.DefaultGPUdPort)
+		return fmt.Errorf("gpud is running on port %d (must be stopped before running compact)", config.GPUdPortNumber())
 	}
 
 	log.Logger.Infow("successfully checked gpud is not running")

cmd/gpud/compact/command_test.go

@@ -81,7 +81,7 @@ func TestCommand_ReturnsErrorWhenPortOpen(t *testing.T) {
 
 	err := Command(cliContext)
 	require.Error(t, err)
-	assert.Equal(t, fmt.Sprintf("gpud is running on port %d (must be stopped before running compact)", config.DefaultGPUdPort), err.Error())
+	assert.Equal(t, fmt.Sprintf("gpud is running on port %d (must be stopped before running compact)", config.GPUdPortNumber()), err.Error())
 }
 
 func TestCommand_Success(t *testing.T) {

cmd/gpud/list-plugins/command.go

@@ -30,7 +30,7 @@ func Command(cliContext *cli.Context) error {
 	// Get the server address from the flag, default to http://localhost:<Default GPUd port>
 	serverAddr := cliContext.String("server")
 	if serverAddr == "" {
-		serverAddr = fmt.Sprintf("https://localhost:%d", config.DefaultGPUdPort)
+		serverAddr = fmt.Sprintf("https://localhost:%d", config.GPUdPortNumber())
 	}
 
 	// Get custom plugins

cmd/gpud/run-plugin-group/command.go

@@ -33,7 +33,7 @@ func Command(cliContext *cli.Context) error {
 	// Get the server address from the flag, default to http://localhost:<Default GPUd port>
 	serverAddr := cliContext.String("server")
 	if serverAddr == "" {
-		serverAddr = fmt.Sprintf("https://localhost:%d", config.DefaultGPUdPort)
+		serverAddr = fmt.Sprintf("https://localhost:%d", config.GPUdPortNumber())
 	}
 
 	// Create a context with timeout

cmd/gpud/set-healthy/command.go

@@ -29,7 +29,7 @@ func CreateCommand() func(*cli.Context) error {
 		// Get the server address from the flag, default to https://localhost:<Default GPUd port>
 		serverAddr := cliContext.String("server")
 		if serverAddr == "" {
-			serverAddr = fmt.Sprintf("https://localhost:%d", config.DefaultGPUdPort)
+			serverAddr = fmt.Sprintf("https://localhost:%d", config.GPUdPortNumber())
 		}
 
 		// Get the components from the positional argument

cmd/gpud/status/command.go

@@ -106,7 +106,7 @@ func Command(cliContext *cli.Context) error {
 
 	if err := clientv1.BlockUntilServerReady(
 		rootCtx,
-		fmt.Sprintf("https://localhost:%d", config.DefaultGPUdPort),
+		fmt.Sprintf("https://localhost:%d", config.GPUdPortNumber()),
 	); err != nil {
 		return err
 	}
@@ -118,7 +118,7 @@ func Command(cliContext *cli.Context) error {
 	for {
 		var err error
 		cctx, ccancel := context.WithTimeout(rootCtx, 15*time.Second)
-		lastPackageStatus, err = clientv1.GetPackageStatus(cctx, fmt.Sprintf("https://localhost:%d%s", config.DefaultGPUdPort, server.URLPathAdminPackages))
+		lastPackageStatus, err = clientv1.GetPackageStatus(cctx, fmt.Sprintf("https://localhost:%d%s", config.GPUdPortNumber(), server.URLPathAdminPackages))
 		ccancel()
 		if err != nil {
 			fmt.Printf("%s failed to get package status: %v\n", cmdcommon.WarningSign, err)

pkg/config/default.go

@@ -5,6 +5,7 @@ import (
 	"fmt"
 	stdos "os"
 	"path/filepath"
+	"strconv"
 	"time"
 
 	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
@@ -14,8 +15,8 @@ import (
 
 const (
 	DefaultAPIVersion = "v1"
-	DefaultGPUdPort   = 15132
 	DefaultDataDir    = "/var/lib/gpud"
+	DefaultGPUdPort   = 15132
 )
 
 var (
@@ -43,7 +44,7 @@ func DefaultConfig(ctx context.Context, opts ...OpOption) (*Config, error) {
 
 	cfg := &Config{
 		APIVersion:       DefaultAPIVersion,
-		Address:          fmt.Sprintf(":%d", DefaultGPUdPort),
+		Address:          fmt.Sprintf(":%d", GPUdPortNumber()),
 		DataDir:          dataDir,
 		RetentionPeriod:  DefaultRetentionPeriod,
 		CompactPeriod:    DefaultCompactPeriod,
@@ -141,3 +142,17 @@ func PackagesDir(dataDir string) string {
 func VersionFilePath(dataDir string) string {
 	return filepath.Join(dataDir, "target_version")
 }
+
+func GPUdPortNumber() int {
+	portStr, found := stdos.LookupEnv("GPUD_PORT")
+	if !found {
+		return DefaultGPUdPort
+	}
+
+	port, err := strconv.Atoi(portStr)
+	if err != nil {
+		return DefaultGPUdPort
+	}
+
+	return port
+}

pkg/config/default_test.go

@@ -20,7 +20,7 @@ func TestDefaultConfig(t *testing.T) {
 
 		// Check basic properties that don't depend on the environment
 		assert.Equal(t, DefaultAPIVersion, cfg.APIVersion)
-		assert.Equal(t, fmt.Sprintf(":%d", DefaultGPUdPort), cfg.Address)
+		assert.Equal(t, fmt.Sprintf(":%d", GPUdPortNumber()), cfg.Address)
 		assert.Equal(t, DefaultRetentionPeriod, cfg.RetentionPeriod)
 		assert.Equal(t, DefaultCompactPeriod, cfg.CompactPeriod)
 		assert.False(t, cfg.Pprof)