Added Job.HasStarted()

Author: David Grieser

pkg/proc/basejob.go

@@ -31,7 +31,7 @@ func (job *baseJob) SignalAll(sig syscall.Signal) {
 		}
 	}
 
-	if job.cmd == nil || job.cmd.Process == nil {
+	if !job.HasStarted() {
 		errFunc(
 			fmt.Errorf("job is not running"),
 		)
@@ -43,6 +43,9 @@ func (job *baseJob) SignalAll(sig syscall.Signal) {
 }
 
 func (job *baseJob) signalAll(sig syscall.Signal) error {
+	if !job.HasStarted() {
+		return nil
+	}
 	if job.cmd.Process.Pid <= 0 { // Ensure PID is positive before negating
 		return syscall.Errno(syscall.ESRCH) // No such process or invalid PID
 	}
@@ -56,7 +59,7 @@ func (job *baseJob) Signal(sig os.Signal) {
 		}
 	}
 
-	if job.cmd == nil || job.cmd.Process == nil {
+	if !job.HasStarted() {
 		errFunc(
 			fmt.Errorf("job is not running"),
 		)
@@ -89,6 +92,10 @@ func (job *baseJob) IsControllable() bool {
 	return job.Config.Controllable
 }
 
+func (job *baseJob) HasStarted() bool {
+	return job.cmd != nil && job.cmd.Process != nil
+}
+
 func (job *baseJob) GetPhase() *JobPhase {
 	return &job.phase
 }
@@ -183,13 +190,11 @@ func (job *baseJob) startOnce(ctx context.Context, process chan<- *os.Process) e
 	select {
 	// job errChan or failed
 	case err := <-errChan:
-		if job.cmd != nil && job.cmd.Process != nil { // Check if process actually started
-			if termErr := job.signalAll(syscall.SIGTERM); termErr != nil {
-				if e, ok := termErr.(syscall.Errno); ok && e == syscall.ESRCH {
-					// ESRCH (Error No Such Process) is fine, means process group already gone
-				} else {
-					l.WithError(termErr).Error("failed to send SIGTERM to job's process group")
-				}
+		if termErr := job.signalAll(syscall.SIGTERM); termErr != nil {
+			if e, ok := termErr.(syscall.Errno); ok && e == syscall.ESRCH {
+				// ESRCH (Error No Such Process) is fine, means process group already gone
+			} else {
+				l.WithError(termErr).Error("failed to send SIGTERM to job's process group")
 			}
 		}
 
@@ -209,7 +214,7 @@ func (job *baseJob) startOnce(ctx context.Context, process chan<- *os.Process) e
 		}
 		return err
 	case <-ctx.Done():
-		if job.cmd != nil && job.cmd.Process != nil { // Check if process actually started
+		if job.HasStarted() {
 			// ctx canceled, try to terminate job
 			_ = job.signalAll(syscall.SIGTERM)
 			l.WithField("job.name", job.Config.Name).Info("sent SIGTERM to job's process group on ctx.Done")
@@ -226,9 +231,8 @@ func (job *baseJob) startOnce(ctx context.Context, process chan<- *os.Process) e
 				return err
 			}
 		}
-		// If job.cmd or job.cmd.Process is nil, it means the process didn't start or already cleaned up.
 		l.WithField("job.name", job.Config.Name).Info("context done, but process was not running or already cleaned up.")
-		return ctx.Err() // Return context error
+		return ctx.Err()
 	}
 }
 

pkg/proc/job_common.go

@@ -177,7 +177,7 @@ func (job *CommonJob) Watch() {
 }
 
 func (job *CommonJob) IsRunning() bool {
-	if job.cmd == nil || job.cmd.Process == nil || job.cmd.Process.Pid <= 0 {
+	if !job.HasStarted() || job.cmd.Process.Pid <= 0 {
 		return false
 	}
 	err := job.signal(syscall.Signal(0))
@@ -186,31 +186,24 @@ func (job *CommonJob) IsRunning() bool {
 
 func (job *CommonJob) Restart() {
 	job.restart = true
-	// Ensure job.cmd and job.cmd.Process are not nil before trying to signal
-	if job.cmd != nil && job.cmd.Process != nil {
-		job.SignalAll(syscall.SIGTERM)
-	}
-	if job.interrupt != nil { // Check if interrupt function is set
+	job.SignalAll(syscall.SIGTERM)
+	if job.interrupt != nil {
 		job.interrupt()
 	}
 }
 
 func (job *CommonJob) Stop() {
 	job.stop = true
-	// Ensure job.cmd and job.cmd.Process are not nil before trying to signal
-	if job.cmd != nil && job.cmd.Process != nil {
-		job.SignalAll(syscall.SIGTERM)
-	}
-	if job.interrupt != nil { // Check if interrupt function is set
+	job.SignalAll(syscall.SIGTERM)
+	if job.interrupt != nil {
 		job.interrupt()
 	}
 }
 
 func (job *CommonJob) Status() *CommonJobStatus {
-	running := job.IsRunning()
+	running := job.HasStarted() && job.IsRunning()
 	var pid int
-	// Ensure job.cmd and job.cmd.Process are not nil before trying to access Pid
-	if running && job.cmd != nil && job.cmd.Process != nil {
+	if running {
 		pid = job.cmd.Process.Pid
 	}
 	return &CommonJobStatus{

pkg/proc/job_lazy.go

@@ -125,8 +125,7 @@ func (job *LazyJob) startProcessReaper(ctx context.Context) {
 					continue
 				}
 
-				// Ensure job.cmd and job.cmd.Process are not nil before accessing PID
-				if job.cmd == nil || job.cmd.Process == nil {
+				if !job.HasStarted() {
 					l.Warn("job.process is not nil, but job.cmd or job.cmd.Process is nil; skipping reap cycle")
 					job.lazyStartLock.Unlock()
 					continue
@@ -148,16 +147,16 @@ func (job *LazyJob) startProcessReaper(ctx context.Context) {
 				case <-graceTimer.C:
 					job.lazyStartLock.Lock()
 					// Check if the process we sent SIGTERM to is still running
-					if job.process != nil && job.cmd != nil && job.cmd.Process != nil && job.cmd.Process.Pid == pidToReap {
+					if job.process != nil && job.HasStarted() && job.cmd.Process.Pid == pidToReap {
 						l.Warnf("process PID %d did not exit after SIGTERM and grace period; sending SIGKILL", pidToReap)
 						if err := job.signalAll(syscall.SIGKILL); err != nil {
-							l.WithError(err).Errorf("failed to send SIGKILL to PID %d", pidToReap)
+							l.WithError(err).Errorf("failed to send1 SIGKILL to PID %d", pidToReap)
 						}
 					} else if job.process != nil {
 						// Process is not nil, but it's not the one we targeted.
 						// This could happen if the job was quickly restarted.
 						currentPid := -1
-						if job.cmd != nil && job.cmd.Process != nil {
+						if job.HasStarted() {
 							currentPid = job.cmd.Process.Pid
 						}
 						l.Warnf("original process PID %d seems to have exited or changed; current PID is %d. Skipping SIGKILL.", pidToReap, currentPid)