fix

lisa/runners/lisa_runner.py

@@ -290,30 +290,51 @@ def _prepare_environments(self) -> None:
     def _deploy_environment_task(
         self, environment: Environment, test_results: List[TestResult]
     ) -> None:
-        try:
+        test_result = test_results[0]
+        retry_deployment = getattr(test_result.runtime_data, "retry_deployment", 0)
+        max_retries = max(retry_deployment, 1)  # Ensure at least one attempt
+
+        for attempt in range(max_retries):
             try:
+                # if throw_exception:
+                #     raise LisaException("test")
+                # Attempt to deploy the environment
                 self.platform.deploy_environment(environment)
-                assert (
-                    environment.status == EnvironmentStatus.Deployed
-                ), f"actual: {environment.status}"
+                if environment.status != EnvironmentStatus.Deployed:
+                    raise AssertionError(
+                        f"Expected status: {EnvironmentStatus.Deployed}, actual: {environment.status}"
+                    )
+
+                # Reset the awaitable timer and exit loop on success
                 self._reset_awaitable_timer("deploy")
+                break
+
             except ResourceAwaitableException as identifier:
                 if self._is_awaitable_timeout("deploy"):
                     self._log.info(
-                        f"[{environment.name}] timeout on waiting for more resource: "
-                        f"{identifier}, skip assigning case."
+                        f"[{environment.name}] Timeout while waiting for more resources: {identifier}. Skipping case."
                     )
                     raise SkippedException(identifier)
                 else:
-                    # rerun prepare to calculate resource again.
+                    # Recalculate resources and retry
                     environment.status = EnvironmentStatus.New
-        except Exception as identifier:
-            self._attach_failed_environment_to_result(
-                environment=environment,
-                result=test_results[0],
-                exception=identifier,
-            )
-            self._delete_environment_task(environment=environment, test_results=[])
+
+            except Exception as identifier:
+                if attempt >= max_retries - 1:
+                    # Final attempt failed; attach result and delete environment
+                    self._attach_failed_environment_to_result(
+                        environment=environment,
+                        result=test_result,
+                        exception=identifier,
+                    )
+                    self._delete_environment_task(
+                        environment=environment, test_results=[]
+                    )
+                else:
+                    # Log retry information and continue to the next attempt
+                    self._log.info(
+                        f"[{environment.name}] Deployment failed, retrying... (Attempt {attempt + 1}/{max_retries})"
+                    )
 
     def _initialize_environment_task(
         self, environment: Environment, test_results: List[TestResult]

lisa/testsuite.py

@@ -522,6 +522,7 @@ def __init__(self, metadata: TestCaseMetadata):
         self.select_action: str = ""
         self.times: int = 1
         self.retry: int = 0
+        self.retry_deployment: int = 3
         self.use_new_environment: bool = metadata.use_new_environment
         self.ignore_failure: bool = False
         self.environment_name: str = ""