Fix vm deletion caused by error string check

Author: Kwaizer

alts/shared/utils/log_utils.py

@@ -30,20 +30,19 @@ def read_and_cleanup_temp_log_files(
     return out, err
 
 
-def check_for_error_string(stage_data: dict) -> bool:
+def check_for_error_string(stderr: str) -> bool:
     """
     Checks if we encounter errors during testing that worth keepin VM alive.
 
     Parameters
     ----------
-    stage_data: dict
+    stderr: str
 
     Returns
     -------
     bool
         True if we encounter any error from the list. False otherwise
     """
-    err = stage_data.get('stderr', '')
-    if any(error_str in err for error_str in ERROR_STRINGS):
+    if stderr and any(error_str in stderr for error_str in ERROR_STRINGS):
         return True
     return False

alts/worker/runners/base.py

@@ -1555,11 +1555,6 @@ def setup(self, skip_provision: bool = False):
 
     def teardown(self, publish_artifacts: bool = True):
         try:
-            if not self.vm_alive:
-                for _, stage_data in self.artifacts.items():
-                    self.vm_alive = check_for_error_string(stage_data)
-                    for _, inner_data in stage_data.items():
-                        self.vm_alive = check_for_error_string(inner_data)
             self.stop_env()
         except Exception as e:
             self._logger.exception('Error while stop environment: %s', e)
@@ -1674,7 +1669,7 @@ def __init__(
         self._tests_dir = CONFIG.tests_base_dir
         self._ssh_client: Optional[Union[AsyncSSHClient, LongRunSSHClient]] = None
         self._vm_ip = None
-        self._vm_alive = vm_alive
+        self.vm_alive = vm_alive
         self.start_env_failed = False
 
     def _wait_for_ssh(self, retries=60):
@@ -1762,6 +1757,8 @@ def start_env(self):
         final_exit_code = exit_code or ssh_exit_code
         final_stdout = f'{stdout}\n\n{ssh_stdout}'
         final_stderr = f'{stderr}\n\n{ssh_stderr}'
+        if not self.vm_alive:
+            self.vm_alive = check_for_error_string(final_stderr)
         return final_exit_code, final_stdout, final_stderr
 
     def setup(self, skip_provision: bool = False):

alts/worker/runners/opennebula.py

@@ -177,7 +177,7 @@ def _stop_env(self):
             )
             self._logger.warning(err_msg)
             return 0, err_msg, ''
-        if self._vm_alive:
+        if self.vm_alive:
             return 0, "WARNING: VM won't be destroyed because vm_alive=True was given", ""
         stop_exit_code, stop_out, stop_err = super()._stop_env()
         if stop_exit_code == 0: