kvclient(ticdc): fix the workerpool data race (#10196) (#10201)

ref #10095

cdc/kv/region_worker.go

@@ -41,6 +41,7 @@ import (
 var (
 	regionWorkerPool workerpool.WorkerPool
 	workerPoolOnce   sync.Once
+	workerPoolLock   sync.Mutex
 	// The magic number here is keep the same with some magic numbers in some
 	// other components in TiCDC, including worker pool task chan size, mounter
 	// chan size etc.
@@ -412,6 +413,8 @@ func (w *regionWorker) processEvent(ctx context.Context, event *regionStatefulEv
 
 func (w *regionWorker) initPoolHandles() {
 	handles := make([]workerpool.EventHandle, 0, w.concurrency)
+	workerPoolLock.Lock()
+	defer workerPoolLock.Unlock()
 	for i := 0; i < w.concurrency; i++ {
 		poolHandle := regionWorkerPool.RegisterEvent(func(ctx context.Context, eventI interface{}) error {
 			event := eventI.(*regionStatefulEvent)
@@ -867,6 +870,8 @@ func getWorkerPoolSize() (size int) {
 func InitWorkerPool() {
 	workerPoolOnce.Do(func() {
 		size := getWorkerPoolSize()
+		workerPoolLock.Lock()
+		defer workerPoolLock.Unlock()
 		regionWorkerPool = workerpool.NewDefaultWorkerPool(size)
 	})
 }

cdc/server/server.go

@@ -342,10 +342,6 @@ func (s *server) run(ctx context.Context) (err error) {
 
 	eg, egCtx := errgroup.WithContext(ctx)
 
-	eg.Go(func() error {
-		return s.capture.Run(egCtx)
-	})
-
 	eg.Go(func() error {
 		return s.upstreamPDHealthChecker(egCtx)
 	})
@@ -371,6 +367,10 @@ func (s *server) run(ctx context.Context) (err error) {
 		return nil
 	})
 
+	eg.Go(func() error {
+		return s.capture.Run(egCtx)
+	})
+
 	return eg.Wait()
 }