fix: proper job cancellation

Commit `a9e9d52`

Eric Bower · 2026-08-09 10:47:31 -0400 EDT

parent 234c991

fix: proper job cancellation

2 files changed, +26, -4

M main.go

M main_test.go

M main.go

+23, -4

......@@ -829,9 +829,9 @@ func waitAndReport(cfg *Cfg, log *slog.Logger, name, jobID, eventType string) er
	ticker := time.NewTicker(interval)
	defer ticker.Stop()

	// Handle ^C gracefully
	// Handle ^C and termination signals gracefully
	sigCh := make(chan os.Signal, 1)
	signal.Notify(sigCh, syscall.SIGINT)
	signal.Notify(sigCh, syscall.SIGINT, syscall.SIGTERM)
	defer signal.Stop(sigCh)

	fmt.Fprintln(os.Stdout)                                //nolint:errcheck

......@@ -855,8 +855,9 @@ func waitAndReport(cfg *Cfg, log *slog.Logger, name, jobID, eventType string) er
			fmt.Fprintln(os.Stdout, "\n⏹ job cancelled") //nolint:errcheck
			return cfg.Ctx.Err()
		case <-sigCh:
			fmt.Fprintln(os.Stdout, "\n⏹ cancelled") //nolint:errcheck
			return nil
			cancelJobSessions(prefix)
			fmt.Fprintln(os.Stdout, "\n⏹ job cancelled") //nolint:errcheck
			return fmt.Errorf("job cancelled by signal")
		case <-ticker.C:
		}

......@@ -1842,6 +1843,24 @@ func killSessions(names []string) error {
	return nil
}

// cancelJobSessions finds and kills all active zmx sessions with the given prefix.
func cancelJobSessions(prefix string) {
	listOutput, err := exec.Command("zmx", "list").CombinedOutput()
	if err != nil {
		return
	}
	sessions := parseZMXList(string(listOutput))
	var toKill []string
	for _, s := range sessions {
		if strings.HasPrefix(s.Name, prefix) && s.Ended == "" {
			toKill = append(toKill, s.Name)
		}
	}
	if len(toKill) > 0 {
		_ = killSessions(toKill)
	}
}

// runGC kills all ci. zmx sessions older than 3 hours, regardless of status.
func runGC(cfg *Cfg) error {
	log := cfg.Logger.With("cmd", "gc")

M main_test.go

+3, -0

......@@ -285,6 +285,9 @@ zmx run fast-step echo "done"
	artifactDir := t.TempDir()
	ctx, cancel := context.WithCancel(context.Background())
	defer cancel()
	t.Cleanup(func() {
		_ = exec.Command("zmx", "kill", "-f", "local.dup-repo").Run()
	})

	makeCfg := func(eventSource io.ReadCloser) *Cfg {
		return &Cfg{

...	...		@@ -829,9 +829,9 @@ func waitAndReport(cfg Cfg, log slog.Logger, name, jobID, eventType string) er
829	829		ticker := time.NewTicker(interval)
830	830		defer ticker.Stop()
831	831
832		-	// Handle ^C gracefully
	832	+	// Handle ^C and termination signals gracefully
833	833		sigCh := make(chan os.Signal, 1)
834		-	signal.Notify(sigCh, syscall.SIGINT)
	834	+	signal.Notify(sigCh, syscall.SIGINT, syscall.SIGTERM)
835	835		defer signal.Stop(sigCh)
836	836
837	837		fmt.Fprintln(os.Stdout) //nolint:errcheck

...	...		@@ -855,8 +855,9 @@ func waitAndReport(cfg Cfg, log slog.Logger, name, jobID, eventType string) er
855	855		fmt.Fprintln(os.Stdout, "\n⏹ job cancelled") //nolint:errcheck
856	856		return cfg.Ctx.Err()
857	857		case <-sigCh:
858		-	fmt.Fprintln(os.Stdout, "\n⏹ cancelled") //nolint:errcheck
859		-	return nil
	858	+	cancelJobSessions(prefix)
	859	+	fmt.Fprintln(os.Stdout, "\n⏹ job cancelled") //nolint:errcheck
	860	+	return fmt.Errorf("job cancelled by signal")
860	861		case <-ticker.C:
861	862		}
862	863

...	...		@@ -1842,6 +1843,24 @@ func killSessions(names []string) error {
1842	1843		return nil
1843	1844		}
1844	1845
	1846	+	// cancelJobSessions finds and kills all active zmx sessions with the given prefix.
	1847	+	func cancelJobSessions(prefix string) {
	1848	+	listOutput, err := exec.Command("zmx", "list").CombinedOutput()
	1849	+	if err != nil {
	1850	+	return
	1851	+	}
	1852	+	sessions := parseZMXList(string(listOutput))
	1853	+	var toKill []string
	1854	+	for _, s := range sessions {
	1855	+	if strings.HasPrefix(s.Name, prefix) && s.Ended == "" {
	1856	+	toKill = append(toKill, s.Name)
	1857	+	}
	1858	+	}
	1859	+	if len(toKill) > 0 {
	1860	+	_ = killSessions(toKill)
	1861	+	}
	1862	+	}
	1863	+
1845	1864		// runGC kills all ci. zmx sessions older than 3 hours, regardless of status.
1846	1865		func runGC(cfg *Cfg) error {
1847	1866		log := cfg.Logger.With("cmd", "gc")

...	...		@@ -285,6 +285,9 @@ zmx run fast-step echo "done"
285	285		artifactDir := t.TempDir()
286	286		ctx, cancel := context.WithCancel(context.Background())
287	287		defer cancel()
	288	+	t.Cleanup(func() {
	289	+	_ = exec.Command("zmx", "kill", "-f", "local.dup-repo").Run()
	290	+	})
288	291
289	292		makeCfg := func(eventSource io.ReadCloser) *Cfg {
290	293		return &Cfg{

Commit a9e9d52

Commit `a9e9d52`