diff --git a/e2e/keywords/volume.resource b/e2e/keywords/volume.resource
index ce595a322..b322b369e 100644
--- a/e2e/keywords/volume.resource
+++ b/e2e/keywords/volume.resource
@@ -99,6 +99,10 @@ Delete volume ${volume_id} replica on ${replica_locality}
     ${volume_name} =    generate_name_with_suffix    volume    ${volume_id}
     delete_replica_on_node    ${volume_name}    ${replica_locality}
 
+Delete ${count} replicas of volume ${volume_id}
+    ${volume_name} =    generate_name_with_suffix    volume    ${volume_id}
+    delete_replicas    ${volume_name}    ${count}
+
 Wait for volume ${volume_id} healthy
     ${volume_name} =    generate_name_with_suffix    volume    ${volume_id}
     wait_for_volume_healthy    ${volume_name}
@@ -178,6 +182,10 @@ Wait until volume ${volume_id} replicas rebuilding completed
     ${volume_name} =    generate_name_with_suffix    volume    ${volume_id}
     wait_for_replica_rebuilding_to_complete   ${volume_name}
 
+Monitor only one replica rebuilding will start at a time for volume ${volume_id}
+    ${volume_name} =    generate_name_with_suffix    volume    ${volume_id}
+    wait_for_replica_rebuilding_to_complete   ${volume_name}
+
 Wait until volume ${volume_id} replica rebuilding stopped on ${replica_locality}
     ${volume_name} =    generate_name_with_suffix    volume    ${volume_id}
     wait_for_replica_rebuilding_to_stop_on_node    ${volume_name}    ${replica_locality}
@@ -192,6 +200,10 @@ Both volume ${volume_id_0} and volume ${volume_id_1} replica rebuilding on ${rep
     ${volume_name_1} =    generate_name_with_suffix    volume    ${volume_id_1}
     both_replica_rebuildings_will_start_at_the_same_time_on_node    ${volume_name_0}    ${volume_name_1}    ${replica_locality}
 
+Only one replica rebuilding will start at a time for volume ${volume_id}
+    ${volume_name} =    generate_name_with_suffix    volume    ${volume_id}
+    only_one_replica_rebuilding_will_start_at_a_time    ${volume_name}
+
 Crash volume ${volume_id} replica processes
     ${volume_name} =    generate_name_with_suffix    volume    ${volume_id}
     crash_replica_processes    ${volume_name}
diff --git a/e2e/libs/keywords/volume_keywords.py b/e2e/libs/keywords/volume_keywords.py
index 69f096b0f..2ef9c77ac 100644
--- a/e2e/libs/keywords/volume_keywords.py
+++ b/e2e/libs/keywords/volume_keywords.py
@@ -128,6 +128,13 @@ def delete_replica_on_nodes(self, volume_name, replica_locality):
             logging(f"Deleting volume {volume_name}'s replica on node {node_id}")
             self.volume.delete_replica(volume_name, node_id)
 
+    def delete_replicas(self, volume_name, count):
+        replica_list = self.replica.get(volume_name, node_name="")
+        replica_names = [replica['metadata']['name'] for replica in replica_list]
+        for i in range(int(count)):
+            logging(f"Deleting volume {volume_name} replica volume {replica_names[i]}")
+            self.volume.delete_replica_by_name(volume_name, replica_names[i])
+
     def set_annotation(self, volume_name, annotation_key, annotation_value):
         self.volume.set_annotation(volume_name, annotation_key, annotation_value)
 
@@ -152,11 +159,7 @@ def wait_for_replica_rebuilding_to_complete_on_node(self, volume_name, replica_l
         self.volume.wait_for_replica_rebuilding_complete(volume_name, node_name)
 
     def wait_for_replica_rebuilding_to_complete(self, volume_name):
-        for node_name in self.node.list_node_names_by_role("worker"):
-            if self.node.is_node_schedulable(node_name) == "False":
-                continue
-            logging(f"Waiting for volume {volume_name}'s replica on node {node_name} rebuilding completed")
-            self.volume.wait_for_replica_rebuilding_complete(volume_name, node_name)
+        self.volume.wait_for_replica_rebuilding_complete(volume_name)
 
     async def only_one_replica_rebuilding_will_start_at_a_time_on_node(self, volume_name_0, volume_name_1, replica_locality):
 
@@ -203,6 +206,21 @@ async def wait_for_both_replica_rebuildings():
         assert self.volume.is_replica_rebuilding_in_progress(volume_name_0, node_id) and self.volume.is_replica_rebuilding_in_progress(volume_name_1, node_id), \
             f"Expect {volume_name_0} and {volume_name_1} replica rebuilding at the same time"
 
+    async def only_one_replica_rebuilding_will_start_at_a_time(self, volume_name):
+
+        async def wait_for_replica_rebuilding():
+            tasks = [
+                asyncio.create_task(self.volume.wait_for_replica_rebuilding_start(volume_name), name=volume_name),
+            ]
+
+            done, pending = await asyncio.wait(tasks, return_when=asyncio.ALL_COMPLETED)
+            logging(f"Observed {done.pop().get_name()} started replica rebuilding")
+
+        await wait_for_replica_rebuilding()
+
+        assert self.volume.is_replica_rebuilding_in_progress(volume_name), \
+            f"Expect {volume_name} replica rebuilding in progress"
+
     def crash_replica_processes(self, volume_name):
         self.volume.crash_replica_processes(volume_name)
 
diff --git a/e2e/libs/volume/crd.py b/e2e/libs/volume/crd.py
index b8ff66f58..9598a99b8 100644
--- a/e2e/libs/volume/crd.py
+++ b/e2e/libs/volume/crd.py
@@ -420,6 +420,23 @@ def delete_replica(self, volume_name, node_name):
             name=replica_list['items'][0]['metadata']['name']
         )
 
+    def delete_replica_by_name(self, volume_name, replica_name):
+        replica = self.obj_api.get_namespaced_custom_object(
+            group="longhorn.io",
+            version="v1beta2",
+            namespace="longhorn-system",
+            plural="replicas",
+            name=replica_name
+        )
+        logging(f"Deleting replica {replica['metadata']['name']}")
+        self.obj_api.delete_namespaced_custom_object(
+            group="longhorn.io",
+            version="v1beta2",
+            namespace="longhorn-system",
+            plural="replicas",
+            name=replica['metadata']['name']
+        )
+
     def wait_for_replica_rebuilding_start(self, volume_name, node_name):
         return Rest().wait_for_replica_rebuilding_start(volume_name, node_name)
 
@@ -432,7 +449,7 @@ def crash_replica_processes(self, volume_name):
     def crash_node_replica_process(self, volume_name, node_name):
         return Rest().crash_node_replica_process(volume_name, node_name)
 
-    def wait_for_replica_rebuilding_complete(self, volume_name, node_name):
+    def wait_for_replica_rebuilding_complete(self, volume_name, node_name=None):
         return Rest().wait_for_replica_rebuilding_complete(volume_name, node_name)
 
     def check_data_checksum(self, volume_name, data_id):
diff --git a/e2e/libs/volume/rest.py b/e2e/libs/volume/rest.py
index 9cce306a4..3d6a4225a 100644
--- a/e2e/libs/volume/rest.py
+++ b/e2e/libs/volume/rest.py
@@ -118,22 +118,26 @@ def keep_writing_data(self, volume_name, size):
     def delete_replica(self, volume_name, node_name):
         return NotImplemented
 
-    async def wait_for_replica_rebuilding_start(self, volume_name, node_name):
+    async def wait_for_replica_rebuilding_start(self, volume_name, node_name=None):
         rebuilding_replica_name = None
         for i in range(self.retry_count):
             try:
                 v = get_longhorn_client().by_id_volume(volume_name)
                 logging(f"Trying to get volume {volume_name} rebuilding replicas ... ({i})")
                 for replica in v.replicas:
-                    if replica.hostId == node_name:
+                    if node_name and replica.hostId == node_name and replica.mode == "WO":
                         rebuilding_replica_name = replica.name
                         break
+                    elif replica.mode == "WO":
+                        rebuilding_replica_name = replica.name
+                        node_name = replica.hostId
+                        break
                 if rebuilding_replica_name:
                     break
             except Exception as e:
                 logging(f"Failed to get volume {volume_name} with error: {e}")
             await asyncio.sleep(self.retry_interval)
-        assert rebuilding_replica_name != None
+        assert rebuilding_replica_name != None, f"Waiting for replica rebuilding start for volume {volume_name} on node {node_name} failed: replicas = {v.replicas}"
         logging(f"Got volume {volume_name} rebuilding replica = {rebuilding_replica_name} on node {node_name}")
 
         started = False
@@ -141,6 +145,15 @@ async def wait_for_replica_rebuilding_start(self, volume_name, node_name):
             try:
                 v = get_longhorn_client().by_id_volume(volume_name)
                 logging(f"Got volume {volume_name} rebuild status = {v.rebuildStatus}")
+
+                # During monitoring replica rebuilding
+                # at the same time monitoring if there are unexpected concurrent replica rebuilding
+                rebuilding_count = 0
+                for replica in v.replicas:
+                    if replica.mode == "WO":
+                        rebuilding_count +=1
+                assert rebuilding_count <= 1, f"Unexpected concurrent replica rebuilding = {rebuilding_count}, replicas = {v.replicas}"
+
                 for status in v.rebuildStatus:
                     for replica in v.replicas:
                         if status.replica == replica.name and \
@@ -156,7 +169,7 @@ async def wait_for_replica_rebuilding_start(self, volume_name, node_name):
             await asyncio.sleep(self.retry_interval)
         assert started, f"wait for replica on node {node_name} rebuilding timeout: {v}"
 
-    def is_replica_rebuilding_in_progress(self, volume_name, node_name):
+    def is_replica_rebuilding_in_progress(self, volume_name, node_name=None):
         in_progress = False
         for i in range(self.retry_count):
             try:
@@ -165,8 +178,9 @@ def is_replica_rebuilding_in_progress(self, volume_name, node_name):
                 for status in v.rebuildStatus:
                     for replica in v.replicas:
                         if status.replica == replica.name and \
-                           replica.hostId == node_name and \
+                           (node_name is None or replica.hostId == node_name) and \
                            status.state == "in_progress":
+                            node_name = replica.hostId if not node_name else node_name
                             logging(f"Volume {volume_name} replica rebuilding {replica.name} in progress on {node_name}")
                             in_progress = True
                 break
@@ -217,31 +231,48 @@ def get_replica_name_on_node(self, volume_name, node_name):
                 if r.hostId == node_name:
                     return r.name
 
-    def wait_for_replica_rebuilding_complete(self, volume_name, node_name):
+    def wait_for_replica_rebuilding_complete(self, volume_name, node_name=None):
         completed = False
         for i in range(self.retry_count):
-            logging(f"wait for {volume_name} replica rebuilding completed on {node_name} ... ({i})")
+            logging(f"wait for {volume_name} replica rebuilding completed on {'all nodes' if not node_name else node_name} ... ({i})")
             try:
                 v = get_longhorn_client().by_id_volume(volume_name)
+
+                # During monitoring replica rebuilding
+                # at the same time monitoring if there are unexpected concurrent replica rebuilding
+                rebuilding_count = 0
                 for replica in v.replicas:
-                    # use replica.mode is RW or RO to check if this replica
-                    # has been rebuilt or not
-                    # because rebuildStatus is not reliable
-                    # when the rebuild progress reaches 100%
-                    # it will be removed from rebuildStatus immediately
-                    # and you will just get an empty rebuildStatus []
-                    # so it's no way to distinguish "rebuilding not started yet"
-                    # or "rebuilding already completed" using rebuildStatus
-                    if replica.hostId == node_name and replica.mode == "RW":
+                    if replica.mode == "WO":
+                        rebuilding_count +=1
+                assert rebuilding_count <= 1, f"Unexpected concurrent replica rebuilding = {rebuilding_count}, replicas = {v.replicas}"
+
+                if node_name:
+                    for replica in v.replicas:
+                        # use replica.mode is RW or RO to check if this replica
+                        # has been rebuilt or not
+                        # because rebuildStatus is not reliable
+                        # when the rebuild progress reaches 100%
+                        # it will be removed from rebuildStatus immediately
+                        # and you will just get an empty rebuildStatus []
+                        # so it's no way to distinguish "rebuilding not started yet"
+                        # or "rebuilding already completed" using rebuildStatus
+                        if replica.hostId == node_name and replica.mode == "RW":
+                            completed = True
+                            break
+                else:
+                    rw_replica_count = 0
+                    for replica in v.replicas:
+                        if replica.mode == "RW":
+                            rw_replica_count += 1
+                    if rw_replica_count == v.numberOfReplicas:
                         completed = True
-                        break
                 if completed:
                     break
             except Exception as e:
                 logging(f"Failed to get volume {volume_name} with error: {e}")
             time.sleep(self.retry_interval)
-        logging(f"Completed volume {volume_name} replica rebuilding on {node_name}")
-        assert completed, f"Expect volume {volume_name} replica rebuilding completed on {node_name}"
+        logging(f"Completed volume {volume_name} replica rebuilding on {'all nodes' if not node_name else node_name}")
+        assert completed, f"Expect volume {volume_name} replica rebuilding completed on {'all nodes' if not node_name else node_name}"
 
     def check_data_checksum(self, volume_name, data_id):
         return NotImplemented
diff --git a/e2e/libs/volume/volume.py b/e2e/libs/volume/volume.py
index a6f5da7a8..b039545a4 100644
--- a/e2e/libs/volume/volume.py
+++ b/e2e/libs/volume/volume.py
@@ -101,10 +101,13 @@ def keep_writing_data(self, volume_name):
     def delete_replica(self, volume_name, node_name):
         return self.volume.delete_replica(volume_name, node_name)
 
-    def wait_for_replica_rebuilding_start(self, volume_name, node_name):
+    def delete_replica_by_name(self, volume_name, replica_name):
+        return self.volume.delete_replica_by_name(volume_name, replica_name)
+
+    def wait_for_replica_rebuilding_start(self, volume_name, node_name=None):
         return self.volume.wait_for_replica_rebuilding_start(volume_name, node_name)
 
-    def is_replica_rebuilding_in_progress(self, volume_name, node_name):
+    def is_replica_rebuilding_in_progress(self, volume_name, node_name=None):
         return self.volume.is_replica_rebuilding_in_progress(volume_name, node_name)
 
     def crash_replica_processes(self, volume_name):
@@ -122,7 +125,7 @@ def wait_for_replica_running(self, volume_name, node_name):
     def get_replica_name_on_node(self, volume_name, node_name):
         return self.volume.get_replica_name_on_node(volume_name, node_name)
 
-    def wait_for_replica_rebuilding_complete(self, volume_name, node_name):
+    def wait_for_replica_rebuilding_complete(self, volume_name, node_name=None):
         return self.volume.wait_for_replica_rebuilding_complete(volume_name, node_name)
 
     def check_data_checksum(self, volume_name, data_id):
diff --git a/e2e/tests/regression/test_replica.robot b/e2e/tests/regression/test_replica.robot
new file mode 100644
index 000000000..1555d3e5b
--- /dev/null
+++ b/e2e/tests/regression/test_replica.robot
@@ -0,0 +1,51 @@
+*** Settings ***
+Documentation    Replica Test Cases
+
+Test Tags    regression
+
+Resource    ../keywords/common.resource
+Resource    ../keywords/volume.resource
+Resource    ../keywords/setting.resource
+Resource    ../keywords/deployment.resource
+Resource    ../keywords/persistentvolumeclaim.resource
+Resource    ../keywords/workload.resource
+
+Test Setup    Set test environment
+Test Teardown    Cleanup test resources
+
+*** Variables ***
+${LOOP_COUNT}    1
+${RETRY_COUNT}    300
+${RETRY_INTERVAL}    1
+${DATA_ENGINE}    v1
+
+*** Test Cases ***
+Test Replica Rebuilding Per Volume Limit
+    [Tags]    coretest
+    [Documentation]    Test the volume always only have one replica scheduled for rebuild
+    ...
+    ...    1. Set soft anti-affinity to `true`.
+    ...    2. Create a volume with 1 replica.
+    ...    3. Attach the volume and write a few hundreds MB data to it.
+    ...    4. Scale the volume replica to 5.
+    ...    5. Monitor the volume replica list to make sure there should be only 1 replica in WO state.
+    ...    6. Wait for the volume to complete rebuilding. Then remove 4 of the 5 replicas.
+    ...    7. Monitoring the volume replica list again.
+    ...    8. Once the rebuild was completed again, verify the data checksum.
+    Given Set setting replica-soft-anti-affinity to true
+    And Create volume 0 with    numberOfReplicas=3    dataEngine=${DATA_ENGINE}
+    And Attach volume 0
+    And Wait for volume 0 healthy
+    And Write data to volume 0
+
+    When Update volume 0 replica count to 5
+    Then Only one replica rebuilding will start at a time for volume 0
+    And Monitor only one replica rebuilding will start at a time for volume 0
+    And Wait until volume 0 replicas rebuilding completed
+
+    When Delete 4 replicas of volume 0
+    Then Only one replica rebuilding will start at a time for volume 0
+    And Monitor only one replica rebuilding will start at a time for volume 0
+    And Wait until volume 0 replicas rebuilding completed
+    And Wait for volume 0 healthy
+    And Check volume 0 data is intact