fix: jobs-v2 program runs can leave background children running after a successful command exits

cmd/serve_jobs_v2.go

@@ -265,18 +265,20 @@ func (r *jobsV2ProgramRunner) Run(ctx context.Context, job jobsV2Job, pw progres
 	} else {
 		cmd = exec.CommandContext(ctx, cfg.Command, cfg.Args...)
 	}
-	cmd.WaitDelay = jobsV2ProgramWaitDelay
 	if strings.TrimSpace(cfg.Cwd) != "" {
 		cmd.Dir = cfg.Cwd
 	}
 	if len(cfg.Env) > 0 {
 		cmd.Env = append(os.Environ(), cfg.Env...)
 	}
 
-	cleanup, prepErr := procutil.PrepareCommand(cmd)
+	cleanup, prepErr := tools.PrepareCommand(cmd)
 	if prepErr != nil {
 		return jobsV2RunResult{}, fmt.Errorf("program setup failed: %w", prepErr)
 	}
+	// tools.PrepareCommand installs shell-tool defaults while adding robust
+	// descendant cleanup; keep the program runner's existing wait-delay behavior.
+	cmd.WaitDelay = jobsV2ProgramWaitDelay
 	defer cleanup()
 
 	stdout := procutil.NewLimitedBuffer(jobsV2ProgramOutputLimit)

cmd/serve_jobs_v2_program_test.go

@@ -3,6 +3,14 @@ package cmd
 import (
 	"context"
 	"encoding/json"
+	"errors"
+	"fmt"
+	"os"
+	"path/filepath"
+	"runtime"
+	"strconv"
+	"strings"
+	"syscall"
 	"testing"
 	"time"
 )
@@ -55,6 +63,33 @@ func TestJobsV2ProgramRunner_TimeoutKillsBackgroundChildrenPromptly(t *testing.T
 	}
 }
 
+func TestJobsV2ProgramRunner_CleansUpBackgroundChildOnSuccess(t *testing.T) {
+	if runtime.GOOS == "windows" {
+		t.Skip("process-group cleanup test is unix-specific")
+	}
+
+	pidPath := filepath.Join(t.TempDir(), "child.pid")
+	runner := &jobsV2ProgramRunner{}
+	job := testJobsV2ProgramJob(t, jobsV2ProgramConfig{
+		Command: fmt.Sprintf(`sleep 30 >/dev/null 2>&1 & echo $! > %s`, jobsV2ProgramTestShellQuote(pidPath)),
+		Shell:   true,
+	})
+
+	result, err := runner.Run(context.Background(), job, nil)
+	if err != nil {
+		t.Fatalf("Run failed: %v (stdout=%q stderr=%q)", err, result.Stdout, result.Stderr)
+	}
+
+	pid := readJobsV2ProgramPID(t, pidPath)
+	defer func() {
+		if !jobsV2ProcessHasExited(pid) {
+			_ = syscall.Kill(pid, syscall.SIGKILL)
+		}
+	}()
+
+	waitForJobsV2ProcessExit(t, pid)
+}
+
 func TestJobsV2ProgramRunner_TruncatesCapturedOutput(t *testing.T) {
 	origLimit := jobsV2ProgramOutputLimit
 	jobsV2ProgramOutputLimit = 32
@@ -88,3 +123,65 @@ func TestJobsV2ProgramRunner_TruncatesCapturedOutput(t *testing.T) {
 		t.Fatalf("expected classifyRunError to preserve truncation")
 	}
 }
+
+func readJobsV2ProgramPID(t *testing.T, path string) int {
+	t.Helper()
+
+	data, err := os.ReadFile(path)
+	if err != nil {
+		t.Fatalf("read child pid file: %v", err)
+	}
+	pidText := strings.TrimSpace(string(data))
+	pid, err := strconv.Atoi(pidText)
+	if err != nil {
+		t.Fatalf("parse child pid %q: %v", pidText, err)
+	}
+	return pid
+}
+
+func waitForJobsV2ProcessExit(t *testing.T, pid int) {
+	t.Helper()
+
+	deadline := time.Now().Add(2 * time.Second)
+	for time.Now().Before(deadline) {
+		if jobsV2ProcessHasExited(pid) {
+			return
+		}
+		time.Sleep(20 * time.Millisecond)
+	}
+
+	t.Fatalf("timed out waiting for background child process %d to exit", pid)
+}
+
+func jobsV2ProcessHasExited(pid int) bool {
+	err := syscall.Kill(pid, 0)
+	if err != nil {
+		return errors.Is(err, syscall.ESRCH)
+	}
+	if runtime.GOOS == "linux" {
+		state, ok := jobsV2LinuxProcState(pid)
+		return ok && state == 'Z'
+	}
+	return false
+}
+
+func jobsV2LinuxProcState(pid int) (byte, bool) {
+	data, err := os.ReadFile(filepath.Join("/proc", strconv.Itoa(pid), "stat"))
+	if err != nil {
+		return 0, false
+	}
+	stat := string(data)
+	end := strings.LastIndex(stat, ")")
+	if end == -1 {
+		return 0, false
+	}
+	rest := strings.TrimSpace(stat[end+1:])
+	if rest == "" {
+		return 0, false
+	}
+	return rest[0], true
+}
+
+func jobsV2ProgramTestShellQuote(s string) string {
+	return "'" + strings.ReplaceAll(s, "'", "'\\''") + "'"
+}