fix: rerun dev port cleanup before server restarts

packages/browseros-agent/tools/dev/cmd/cleanup.go

@@ -2,6 +2,7 @@ package cmd
 
 import (
 	"fmt"
+	"time"
 
 	"browseros-dev/proc"
 
@@ -33,7 +34,9 @@ func runCleanup(cmd *cobra.Command, args []string) error {
 	if doPorts {
 		ports := proc.DefaultLocalPorts()
 		proc.LogMsgf(proc.TagInfo, "Killing processes on ports %d, %d, %d...", ports.CDP, ports.Server, ports.Extension)
-		proc.KillPorts(ports)
+		if err := proc.KillPortsAndWait(ports, 3*time.Second); err != nil {
+			return err
+		}
 		proc.LogMsg(proc.TagInfo, "Ports cleared")
 	}
 

packages/browseros-agent/tools/dev/cmd/watch.go

@@ -8,6 +8,7 @@ import (
 	"path/filepath"
 	"sync"
 	"syscall"
+	"time"
 
 	"browseros-dev/browser"
 	"browseros-dev/proc"
@@ -62,7 +63,9 @@ func runWatch(cmd *cobra.Command, args []string) error {
 			return fmt.Errorf("creating user-data dir: %w", err)
 		}
 		proc.LogMsg(proc.TagInfo, "Killing processes on preferred ports...")
-		proc.KillPorts(defaultPorts)
+		if err := proc.KillPortsAndWait(defaultPorts, 3*time.Second); err != nil {
+			return err
+		}
 		proc.LogMsg(proc.TagInfo, "Ports cleared")
 
 		p, reservations, err = proc.ResolveWatchPorts(false)
@@ -159,6 +162,9 @@ func runWatch(cmd *cobra.Command, args []string) error {
 		Env:     env,
 		Restart: true,
 		Cmd:     []string{"bun", "--watch", "--env-file=.env.development", "src/index.ts"},
+		BeforeStart: func() error {
+			return proc.KillPortAndWait(p.Server, 3*time.Second)
+		},
 	}))
 
 	<-sigCh

packages/browseros-agent/tools/dev/proc/managed.go

@@ -11,11 +11,12 @@ import (
 )
 
 type ProcConfig struct {
-	Tag     Tag
-	Dir     string
-	Env     []string
-	Restart bool
-	Cmd     []string
+	Tag         Tag
+	Dir         string
+	Env         []string
+	Restart     bool
+	Cmd         []string
+	BeforeStart func() error
 }
 
 type ManagedProc struct {
@@ -49,6 +50,17 @@ func (mp *ManagedProc) run(ctx context.Context) {
 			return
 		}
 
+		if mp.Cfg.BeforeStart != nil {
+			if err := mp.Cfg.BeforeStart(); err != nil {
+				LogMsg(mp.Cfg.Tag, ErrorColor.Sprintf("Pre-start failed: %v", err))
+				if !mp.Cfg.Restart || ctx.Err() != nil {
+					return
+				}
+				time.Sleep(time.Second)
+				continue
+			}
+		}
+
 		LogMsgf(mp.Cfg.Tag, "Starting: %s", DimColor.Sprint(strings.Join(mp.Cfg.Cmd, " ")))
 
 		cmd := exec.Command(mp.Cfg.Cmd[0], mp.Cfg.Cmd[1:]...)

packages/browseros-agent/tools/dev/proc/managed_test.go

@@ -0,0 +1,60 @@
+package proc
+
+import (
+	"context"
+	"os"
+	"path/filepath"
+	"sync"
+	"sync/atomic"
+	"testing"
+	"time"
+)
+
+func TestStartManagedRunsBeforeStartOnEachRetry(t *testing.T) {
+	ctx, cancel := context.WithTimeout(context.Background(), 2200*time.Millisecond)
+	defer cancel()
+
+	var count atomic.Int32
+	var wg sync.WaitGroup
+
+	StartManaged(ctx, &wg, ProcConfig{
+		Tag:     TagInfo,
+		Dir:     t.TempDir(),
+		Restart: true,
+		Cmd:     []string{"sh", "-c", "exit 1"},
+		BeforeStart: func() error {
+			count.Add(1)
+			return nil
+		},
+	})
+
+	wg.Wait()
+
+	if count.Load() < 2 {
+		t.Fatalf("expected BeforeStart to run on retries, got %d calls", count.Load())
+	}
+}
+
+func TestStartManagedSkipsLaunchWhenBeforeStartFails(t *testing.T) {
+	ctx, cancel := context.WithCancel(context.Background())
+	defer cancel()
+
+	sentinel := filepath.Join(t.TempDir(), "started")
+	var wg sync.WaitGroup
+
+	StartManaged(ctx, &wg, ProcConfig{
+		Tag:     TagInfo,
+		Dir:     t.TempDir(),
+		Restart: false,
+		Cmd:     []string{"sh", "-c", "touch " + sentinel},
+		BeforeStart: func() error {
+			return context.DeadlineExceeded
+		},
+	})
+
+	wg.Wait()
+
+	if _, err := os.Stat(sentinel); !os.IsNotExist(err) {
+		t.Fatalf("expected process launch to be skipped, stat err=%v", err)
+	}
+}

packages/browseros-agent/tools/dev/proc/ports.go

@@ -133,6 +133,29 @@ func KillPort(port int) {
 	exec.Command("sh", "-c", fmt.Sprintf("lsof -ti:%d | xargs kill -9 2>/dev/null || true", port)).Run()
 }
 
+func KillPortAndWait(port int, timeout time.Duration) error {
+	deadline := time.Now().Add(timeout)
+	for {
+		KillPort(port)
+		if IsPortAvailable(port) {
+			return nil
+		}
+		if time.Now().After(deadline) {
+			return fmt.Errorf("port %d is still in use after kill -9 cleanup", port)
+		}
+		time.Sleep(100 * time.Millisecond)
+	}
+}
+
+func KillPortsAndWait(p Ports, timeout time.Duration) error {
+	for _, port := range []int{p.CDP, p.Server, p.Extension} {
+		if err := KillPortAndWait(port, timeout); err != nil {
+			return err
+		}
+	}
+	return nil
+}
+
 func BuildEnv(p Ports, nodeEnv string) []string {
 	env := os.Environ()
 	env = append(env,