server, client: fix hanging problem when etcd failed to start (#1267)

tikv · Oct 12, 2018 · a6b3a6c · a6b3a6c
1 parent 36b88f9
commit a6b3a6c
Show file tree

Hide file tree

Showing 2 changed files with 7 additions and 1 deletion.
diff --git a/client/client.go b/client/client.go
@@ -163,7 +163,9 @@ func (c *client) initClusterID() error {
 	defer cancel()
 	for i := 0; i < maxInitClusterRetries; i++ {
 		for _, u := range c.urls {
-			members, err := c.getMembers(ctx, u)
+			timeoutCtx, timeoutCancel := context.WithTimeout(ctx, pdTimeout)
+			members, err := c.getMembers(timeoutCtx, u)
+			timeoutCancel()
 			if err != nil || members.GetHeader() == nil {
 				log.Errorf("[pd] failed to get cluster id: %v", err)
 				continue

diff --git a/server/server.go b/server/server.go
@@ -43,6 +43,7 @@ import (
 
 const (
 	etcdTimeout           = time.Second * 3
+	etcdStartTimeout      = time.Minute * 5
 	serverMetricsInterval = time.Minute
 	// pdRootPath for all pd servers.
 	pdRootPath      = "/pd"
@@ -134,6 +135,9 @@ func CreateServer(cfg *Config, apiRegister func(*Server) http.Handler) (*Server,
 
 func (s *Server) startEtcd(ctx context.Context) error {
 	log.Info("start embed etcd")
+	ctx, cancel := context.WithTimeout(ctx, etcdStartTimeout)
+	defer cancel()
+
 	etcd, err := embed.StartEtcd(s.etcdCfg)
 	if err != nil {
 		return errors.WithStack(err)