test: fix flaky test TestForwardTestSuite in next-gen (#10203)

close #10200

Signed-off-by: okjiang <819421878@qq.com>

Author: okJiang

server/server.go

@@ -330,14 +330,42 @@ func CreateServer(ctx context.Context, cfg *config.Config, services []string, le
 	return s, nil
 }
 
-func (s *Server) startEtcd(ctx context.Context) error {
+func (s *Server) startEtcd(ctx context.Context) (retErr error) {
 	newCtx, cancel := context.WithTimeout(ctx, EtcdStartTimeout)
 	defer cancel()
 
 	etcd, err := embed.StartEtcd(s.etcdCfg)
 	if err != nil {
 		return errs.ErrStartEtcd.Wrap(err).GenWithStackByCause()
 	}
+	cleanup := func() {
+		// NOTE: `embed.Etcd.Close()` can block for a long time in some failure paths
+		// (e.g. when starting a removed member that can never become ready). Avoid
+		// blocking the caller (tests may wait for the start error) by stopping the
+		// server synchronously and closing the embedded etcd asynchronously.
+		if etcd.Server != nil {
+			etcd.Server.Stop()
+		}
+		go etcd.Close()
+		if s.client != nil {
+			if cerr := s.client.Close(); cerr != nil {
+				log.Error("close etcd client meet error", errs.ZapError(errs.ErrCloseEtcdClient, cerr))
+			}
+		}
+		if s.electionClient != nil {
+			if cerr := s.electionClient.Close(); cerr != nil {
+				log.Error("close election client meet error", errs.ZapError(errs.ErrCloseEtcdClient, cerr))
+			}
+		}
+		if s.httpClient != nil {
+			s.httpClient.CloseIdleConnections()
+		}
+	}
+	defer func() {
+		if retErr != nil {
+			cleanup()
+		}
+	}()
 
 	// Check cluster ID
 	urlMap, err := etcdtypes.NewURLsMap(s.cfg.InitialCluster)

tests/cluster.go

@@ -691,7 +691,10 @@ func (c *TestCluster) runInitialServersWithRetry(maxRetries int) error {
 
 		errMsg := lastErr.Error()
 		switch {
-		case strings.Contains(errMsg, "address already in use"):
+		case strings.Contains(errMsg, "address already in use") || strings.Contains(errMsg, "Etcd cluster ID mismatch"):
+			// `Etcd cluster ID mismatch` can happen when the allocated peer URL happens to
+			// connect to another test's etcd cluster (port reuse across concurrent `go test`
+			// processes). Treat it as a port conflict and recreate servers with new ports.
 			log.Warn("port conflict detected, recreating servers with new ports",
 				zap.Int("attempt", i+1),
 				zap.Int("maxRetries", maxRetries),