Merge pull request #34 from aaomidi/amir/fix_race_condition

Fix race condition in the runner

.github/workflows/run-tests.yml

@@ -14,4 +14,4 @@ jobs:
 
       - name: Test
         working-directory: tests
-        run: bazel test //... 
+        run: bazel test --@rules_go//go/config:race --@rules_go//go/config:pure=false //...

runner/runner.go

@@ -272,7 +272,11 @@ func initializeServiceCmd(ctx context.Context, instance *ServiceInstance) error
 	instance.cmd = cmd
 	instance.killed = false
 	instance.startErrFn = sync.OnceValue(cmd.Start)
-	instance.waitErrFn = sync.OnceValue(cmd.Wait)
+	instance.waitErrFn = sync.OnceValue(func() error {
+		res := cmd.Wait()
+		instance.SetDone()
+		return res
+	})
 
 	if s.HotReloadable {
 		stdin, err := cmd.StdinPipe()

runner/service_instance.go

@@ -31,6 +31,7 @@ type ServiceInstance struct {
 	runErr               error
 	killed               bool
 	healthcheckAttempted bool
+	done                 bool
 }
 
 func (s *ServiceInstance) Start(ctx context.Context) error {
@@ -212,9 +213,10 @@ func (s *ServiceInstance) Stop(sig syscall.Signal) error {
 		return err
 	}
 
-	for s.cmd.ProcessState == nil {
+	for !s.isDone() {
 		time.Sleep(5 * time.Millisecond)
 	}
+
 	return nil
 }
 
@@ -236,3 +238,15 @@ func (s *ServiceInstance) Killed() bool {
 	defer s.mu.Unlock()
 	return s.killed
 }
+
+func (s *ServiceInstance) isDone() bool {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	return s.done && s.cmd.ProcessState != nil
+}
+
+func (s *ServiceInstance) SetDone() {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	s.done = true
+}