Fix bug: delete permissions after deleting all resources. (#97)

* Fix bug: delete permissions after delete all resoruces.

coordinator/gscoordinator/cluster.py

@@ -52,7 +52,6 @@
 from graphscope.deploy.kubernetes.utils import delete_kubernetes_object
 from graphscope.deploy.kubernetes.utils import get_kubernetes_object_info
 from graphscope.deploy.kubernetes.utils import get_service_endpoints
-from graphscope.framework.utils import random_string
 from graphscope.proto import types_pb2
 
 from gscoordinator.io_utils import PipeWatcher
@@ -111,6 +110,7 @@ def __init__(
         image_pull_policy=None,
         image_pull_secrets=None,
         num_workers=None,
+        instance_id=None,
         log_level=None,
         timeout_seconds=None,
         waiting_for_delete=None,
@@ -125,19 +125,21 @@ def __init__(
         self._core_api = kube_client.CoreV1Api(self._api_client)
         self._app_api = kube_client.AppsV1Api(self._api_client)
 
+        self._instance_id = instance_id
+
         # random for multiple k8s cluster in the same namespace
-        self._engine_name = self._engine_name_prefix + random_string(6)
-        self._etcd_name = self._etcd_name_prefix + random_string(6)
-        self._etcd_service_name = self._etcd_service_name_prefix + random_string(6)
+        self._engine_name = self._engine_name_prefix + self._instance_id
+        self._etcd_name = self._etcd_name_prefix + self._instance_id
+        self._etcd_service_name = self._etcd_service_name_prefix + self._instance_id
 
         self._gie_graph_manager_name = (
-            self._gie_graph_manager_name_prefix + random_string(6)
+            self._gie_graph_manager_name_prefix + self._instance_id
         )
         self._gie_graph_manager_service_name = (
-            self._gie_graph_manager_service_name_prefix + random_string(6)
+            self._gie_graph_manager_service_name_prefix + self._instance_id
         )
         self._vineyard_service_name = (
-            self._vineyard_service_name_prefix + random_string(6)
+            self._vineyard_service_name_prefix + self._instance_id
         )
 
         self._namespace = namespace

coordinator/gscoordinator/coordinator.py

@@ -643,6 +643,11 @@ def parse_sys_args():
         default=4,
         help="The number of engine workers.",
     )
+    parser.add_argument(
+        "--instance_id",
+        type=str,
+        help="Unique id for each GraphScope instance.",
+    )
     parser.add_argument(
         "--port",
         type=int,
@@ -816,6 +821,7 @@ def launch_graphscope():
             image_pull_policy=args.k8s_image_pull_policy,
             image_pull_secrets=args.k8s_image_pull_secrets,
             num_workers=args.num_workers,
+            instance_id=args.instance_id,
             log_level=args.log_level,
             timeout_seconds=args.timeout_seconds,
             waiting_for_delete=args.waiting_for_delete,

python/graphscope/deploy/kubernetes/cluster.py

@@ -124,8 +124,8 @@ class KubernetesCluster(object):
     _coordinator_service_name_prefix = "coordinator-service-"
     _coordinator_container_name = "coordinator"
 
-    _role_name = "gs-reader"
-    _role_binding_name = "gs-reader-binding"
+    _role_name_prefix = "gs-reader-"
+    _role_binding_name_prefix = "gs-reader-binding-"
     _cluster_role_name_prefix = "gs-cluster-reader-"
     _cluster_role_binding_name_prefix = "gs-cluster-reader-binding-"
 
@@ -186,15 +186,18 @@ def __init__(
         self._engine_mem = engine_mem
         self._waiting_for_delete = waiting_for_delete
 
+        self._instance_id = random_string(6)
+        self._role_name = self._role_name_prefix + self._instance_id
+        self._role_binding_name = self._role_binding_name_prefix + self._instance_id
         self._cluster_role_name = ""
         self._cluster_role_binding_name = ""
 
         # all resource created inside namsapce
         self._resource_object = []
 
-        self._coordinator_name = self._coordinator_name_prefix + random_string(6)
+        self._coordinator_name = self._coordinator_name_prefix + self._instance_id
         self._coordinator_service_name = (
-            self._coordinator_service_name_prefix + random_string(6)
+            self._coordinator_service_name_prefix + self._instance_id
         )
         self._coordinator_cpu = coordinator_cpu
         self._coordinator_mem = coordinator_mem
@@ -306,7 +309,7 @@ def _create_role_and_binding(self):
                 name=self._role_name,
                 namespace=self._namespace,
                 api_groups="apps,",
-                resources="configmaps,deployments,endpoints,events,pods,pods/log,pods/exec,pods/status,services,replicasets",
+                resources="configmaps,deployments,deployments/status,endpoints,events,pods,pods/log,pods/exec,pods/status,services,replicasets",  # noqa: E501
                 verbs="create,delete,get,update,watch,list",
             )
             targets.append(
@@ -406,6 +409,7 @@ def _create_coordinator(self):
             name=self._coordinator_container_name,
             port=self._random_coordinator_service_port,
             num_workers=self._num_workers,
+            instance_id=self._instance_id,
             log_level=gs_config.log_level,
             namespace=self._namespace,
             service_type=self._service_type,
@@ -586,7 +590,8 @@ def stop(self, wait=False):
         """
         if not self._closed:
             # delete resources created by graphscope inside namespace
-            for target in self._resource_object:
+            # make sure delete permission resouces in the end
+            for target in reversed(self._resource_object):
                 delete_kubernetes_object(
                     api_client=self._api_client,
                     target=target,

python/graphscope/deploy/kubernetes/resource_builder.py

@@ -861,6 +861,7 @@ def add_coordinator_container(
         name,
         port,
         num_workers,
+        instance_id,
         log_level,
         namespace,
         service_type,
@@ -885,6 +886,7 @@ def add_coordinator_container(
     ):
         self._port = port
         self._num_workers = num_workers
+        self._instance_id = instance_id
         self._log_level = log_level
         self._namespace = namespace
         self._service_type = service_type
@@ -965,6 +967,8 @@ def build_container_command(self):
             str(self._port),
             "--num_workers",
             str(self._num_workers),
+            "--instance_id",
+            self._instance_id,
             "--log_level",
             self._log_level,
             "--k8s_namespace",