src/frostfs_testlib/load/runners.py

@@ -3,7 +3,6 @@ import itertools
 import math
 import re
 import time
-from concurrent.futures import ThreadPoolExecutor
 from dataclasses import fields
 from typing import Optional
 from urllib.parse import urlparse
@@ -24,12 +23,14 @@ from frostfs_testlib.resources.load_params import (
     LOAD_NODE_SSH_USER,
     LOAD_NODES,
 )
+from frostfs_testlib.shell.command_inspectors import SuInspector
 from frostfs_testlib.shell.interfaces import CommandOptions, InteractiveInput
 from frostfs_testlib.storage.cluster import ClusterNode
 from frostfs_testlib.storage.controllers.cluster_state_controller import ClusterStateController
 from frostfs_testlib.storage.dataclasses.frostfs_services import S3Gate, StorageNode
 from frostfs_testlib.storage.dataclasses.wallet import WalletInfo
 from frostfs_testlib.testing import parallel, run_optionally
+from frostfs_testlib.testing.test_control import retry
 from frostfs_testlib.utils import FileKeeper, datetime_utils
 
 reporter = get_reporter()
@@ -296,18 +297,38 @@ class LocalRunner(RunnerBase):
         nodes_under_load: list[ClusterNode],
         k6_dir: str,
     ):
+        parallel(self.prepare_node, nodes_under_load, k6_dir, load_params)
+
+    @retry(3, 5, expected_result=True)
+    def allow_user_to_login_in_system(self, cluster_node: ClusterNode):
+        shell = cluster_node.host.get_shell()
+
+        result = None
+        try:
+            shell.exec(f"sudo chsh -s /bin/bash {STORAGE_USER_NAME}")
+            self.lock_passwd_on_node(cluster_node)
+            options = CommandOptions(check=False, extra_inspectors=[SuInspector(STORAGE_USER_NAME)])
+            result = shell.exec("whoami", options)
+        finally:
+            if not result or result.return_code:
+                self.restore_passwd_on_node(cluster_node)
+                return False
+
+        return True
+
     @reporter.step_deco("Prepare node {cluster_node}")
-        def prepare_node(cluster_node: ClusterNode):
+    def prepare_node(self, cluster_node: ClusterNode, k6_dir: str, load_params: LoadParams):
         shell = cluster_node.host.get_shell()
 
         with reporter.step("Allow storage user to login into system"):
-                shell.exec(f"sudo chsh -s /bin/bash {STORAGE_USER_NAME}")
+            self.allow_user_to_login_in_system(cluster_node)
-                shell.exec("sudo chattr +i /etc/passwd")
 
         with reporter.step("Update limits.conf"):
             limits_path = "/etc/security/limits.conf"
             self.file_keeper.add(cluster_node.storage_node, limits_path)
-                content = f"{STORAGE_USER_NAME} hard nofile 65536\n{STORAGE_USER_NAME} soft nofile 65536\n"
+            content = (
+                f"{STORAGE_USER_NAME} hard nofile 65536\n{STORAGE_USER_NAME} soft nofile 65536\n"
+            )
             shell.exec(f"echo '{content}' | sudo tee {limits_path}")
 
         with reporter.step("Download K6"):
@@ -324,13 +345,6 @@ class LocalRunner(RunnerBase):
             shell.exec(f'echo "{content}" | sudo tee {self.wallet.config_path}')
             shell.exec(f"sudo chmod -R 777 {self.wallet.config_path}")
 
-        with ThreadPoolExecutor(max_workers=len(nodes_under_load)) as executor:
-            result = executor.map(prepare_node, nodes_under_load)
-
-            # Check for exceptions
-            for _ in result:
-                pass
-
     @reporter.step_deco("Init k6 instances")
     def init_k6_instances(self, load_params: LoadParams, endpoints: list[str], k6_dir: str):
         self.k6_instances = []
@@ -379,16 +393,21 @@ class LocalRunner(RunnerBase):
         ):
             time.sleep(wait_after_start_time)
 
+    @reporter.step_deco("Restore passwd on {cluster_node}")
+    def restore_passwd_on_node(self, cluster_node: ClusterNode):
+        shell = cluster_node.host.get_shell()
+        shell.exec("sudo chattr -i /etc/passwd")
+
+    @reporter.step_deco("Lock passwd on {cluster_node}")
+    def lock_passwd_on_node(self, cluster_node: ClusterNode):
+        shell = cluster_node.host.get_shell()
+        shell.exec("sudo chattr +i /etc/passwd")
+
     def stop(self):
         for k6_instance in self.k6_instances:
             k6_instance.stop()
 
-        @reporter.step_deco("Restore passwd on {cluster_node}")
+        parallel(self.restore_passwd_on_node, self.nodes_under_load)
-        def restore_passwd_attr_on_node(cluster_node: ClusterNode):
-            shell = cluster_node.host.get_shell()
-            shell.exec("sudo chattr -i /etc/passwd")
-
-        parallel(restore_passwd_attr_on_node, self.nodes_under_load)
 
         self.cluster_state_controller.start_stopped_storage_services()
         self.cluster_state_controller.start_stopped_s3_gates()