Properly propagate zk errors in request retry

internal/dcs/zk.go

@@ -274,10 +274,10 @@ func (z *zkDCS) Initialize() {
 func (z *zkDCS) retryRequestInternal(code func() error) error {
 	err := code()
 	if err != zk.ErrConnectionClosed {
-		return nil
+		return backoff.Permanent(err)
 	}
 	if !z.IsConnected() {
-		return nil
+		return backoff.Permanent(err)
 	}
 	return err
 }

tests/features/CLI.feature

@@ -100,7 +100,11 @@ Feature: CLI
         Then command return code should be "2"
         And command output should match regexp
         """
-        .*replica can be set cascade.*[[:space:]].*node already has priority 5 set.*
+        .*replica can be set cascade.*
+        """
+        And command output should match regexp
+        """
+        .*node already has priority 5 set.*
         """
         When I run command on host "mysql3"
         """
@@ -114,7 +118,11 @@ Feature: CLI
         Then command return code should be "1"
         And command output should match regexp
         """
-        .*node is already streaming from mysql3[[:space:]].*node mysql2 is not HA node, priority cannot be set.*
+        .*node is already streaming from mysql3.*
+        """
+        And command output should match regexp
+        """
+        .*node mysql2 is not HA node, priority cannot be set.*
         """
         Then zookeeper node "/test/cascade_nodes/mysql2" should exist within "5" seconds
         When I run command on host "mysql3"