diff --git a/ch_backup/clickhouse/metadata_cleaner.py b/ch_backup/clickhouse/metadata_cleaner.py
index f44f15c..0b7a293 100644
--- a/ch_backup/clickhouse/metadata_cleaner.py
+++ b/ch_backup/clickhouse/metadata_cleaner.py
@@ -77,7 +77,12 @@ def clean_tables_metadata(self, replicated_tables: List[Table]) -> None:
 
             with self._zk_ctl.zk_client as zk_client:
                 # Both paths are already abs.
-                full_table_zk_path = self._zk_ctl.zk_root_path + path_resolved
+                full_table_zk_path = (
+                    self._zk_ctl.zk_root_path  # type: ignore
+                    + path_resolved
+                    + "/replicas/"
+                    + self._replica_to_drop
+                )
 
                 if not zk_client.exists(full_table_zk_path):
                     logging.debug(
@@ -89,9 +94,7 @@ def clean_tables_metadata(self, replicated_tables: List[Table]) -> None:
 
                 # We are sure that we want to  drop the table from zk.
                 # To force it we will remove it active flag.
-                active_flag_path = os.path.join(
-                    full_table_zk_path, "replicas", self._replica_to_drop, "is_active"  # type: ignore
-                )
+                active_flag_path = os.path.join(full_table_zk_path, "is_active")
                 try:
                     zk_client.delete(active_flag_path)
                 except NoNodeError:
@@ -149,7 +152,12 @@ def clean_database_metadata(self, replicated_databases: List[Database]) -> None:
 
             with self._zk_ctl.zk_client as zk_client:
                 # Both paths are already abs.
-                full_database_zk_path = self._zk_ctl.zk_root_path + path_resolved
+                full_database_zk_path = (
+                    self._zk_ctl.zk_root_path
+                    + path_resolved
+                    + "/replicas/"
+                    + full_replica_name
+                )
 
                 if not zk_client.exists(full_database_zk_path):
                     logging.debug(
@@ -159,9 +167,7 @@ def clean_database_metadata(self, replicated_databases: List[Database]) -> None:
                     )
                     continue
 
-                active_flag_path = os.path.join(
-                    full_database_zk_path, "replicas", full_replica_name, "active"
-                )
+                active_flag_path = os.path.join(full_database_zk_path, "active")
                 try:
                     zk_client.delete(active_flag_path)
                 except NoNodeError:
diff --git a/tests/integration/features/backup_replicated.feature b/tests/integration/features/backup_replicated.feature
index d692c28..0434aee 100644
--- a/tests/integration/features/backup_replicated.feature
+++ b/tests/integration/features/backup_replicated.feature
@@ -859,6 +859,48 @@ Feature: Backup replicated merge tree table
     | zookeeper_path          |
     |/databases/replicated/db_repl|
 
+    Scenario Outline: Add new host with replicated
+    Given we have enabled shared zookeeper for clickhouse01
+    And we have enabled shared zookeeper for clickhouse02
+    Given ClickHouse settings
+    """
+      allow_experimental_database_replicated: 1
+    """
+    And we have executed queries on clickhouse01
+    """
+    DROP DATABASE IF EXISTS db_repl ON CLUSTER 'default' SYNC;
+    CREATE DATABASE db_repl ENGINE = Replicated('<zookeeper_path>', 'shard_01', '{replica}');
+
+    DROP TABLE IF EXISTS table_01 ON CLUSTER 'default' SYNC;
+    CREATE TABLE table_01 (
+        EventDate DateTime,
+        CounterID UInt32,
+        UserID UInt32
+    )
+    ENGINE = ReplicatedMergeTree('/clickhouse/tables/shard01/test_db.table_01', '{replica}')
+    PARTITION BY CounterID % 10
+    ORDER BY (CounterID, EventDate, intHash32(UserID))
+    SAMPLE BY intHash32(UserID);
+    INSERT INTO table_01 SELECT now(), number, rand() FROM system.numbers LIMIT 10;
+    """
+    When we create clickhouse01 clickhouse backup
+
+    When we restore clickhouse backup #0 to clickhouse02
+    """
+    schema_only: true
+    """
+    Then we got same clickhouse data at clickhouse01 clickhouse02
+    
+    @require_version_24.8
+    Examples:
+    | zookeeper_path          |
+    |/databases/{uuid}/db_repl|
+    
+    @require_version_22.8
+    Examples:
+    | zookeeper_path          |
+    |/databases/replicated/db_repl|
+
   ## Note: Sometimes we can have active orphaned table in the zookeeper.
   ## Here we are imitating such situation by creating objects with static replica name.
   @require_version_23.3