[Test] Make the retrieval of flexible instance types more resilient to prevent test failures.

tests/integration-tests/conftest.py

@@ -698,7 +698,18 @@ def inject_placement_group_settings(vpc_stack, instance, region, kwargs):
 
 
 def inject_flexible_instance_types_settings(instance, region, kwargs):
-    kwargs["flexible_instance_types"] = list({instance, *get_similar_instance_types(instance, region, 5)})
+    flexible_instance_types = [instance]
+    try:
+        flexible_instance_types.extend(it for it in get_similar_instance_types(instance, region, 5) if it != instance)
+    except Exception:
+        logging.warning(
+            "Failed to retrieve instance types equivalent to %s in region %s. "
+            "Falling back to using only the original instance type %s.",
+            instance,
+            region,
+            instance,
+        )
+    kwargs["flexible_instance_types"] = flexible_instance_types
 
 
 def inject_additional_image_configs_settings(image_config, request):

tests/integration-tests/utils.py

@@ -1054,6 +1054,7 @@ def _get_gpu_spec(instance_type_data):
     return frozenset((gpu.get("Manufacturer", ""), gpu.get("Count", 0)) for gpu in gpu_info.get("Gpus", []))
 
 
+@retry(wait_fixed=seconds(10), stop_max_delay=minutes(1))
 def get_similar_instance_types(instance_type: str, region: str = None, max_items: int = None):
     ec2 = boto3.client("ec2", region_name=region)
 
@@ -1096,8 +1097,14 @@ def get_similar_instance_types(instance_type: str, region: str = None, max_items
                 and instance_inference_accelerators == target_inference_accelerators
             ):
                 similar_instances.append(instance["InstanceType"])
-                if max_items and len(similar_instances) >= max_items:
-                    return similar_instances
+
+    # Sort before truncating so that multiple calls always return the same instance types in the same order,
+    # regardless of the order in which the API returns them.
+    similar_instances = sorted(similar_instances)
+    if max_items:
+        similar_instances = similar_instances[:max_items]
+
+    logging.info(f"Retrieved instance types equivalent to {instance_type} in {region}: {similar_instances}")
 
     return similar_instances