import os import uuid from common import ASSETS_DIR, SIMPLE_OBJ_SIZE def create_file_with_content(file_path: str = None, content: str = None) -> str: mode = 'w+' if not content: content = os.urandom(SIMPLE_OBJ_SIZE) mode = 'wb' if not file_path: file_path = f"{os.getcwd()}/{ASSETS_DIR}/{str(uuid.uuid4())}" else: if not os.path.exists(os.path.dirname(file_path)): os.makedirs(os.path.dirname(file_path)) with open(file_path, mode) as out_file: out_file.write(content) return file_path def get_file_content(file_path: str, content_len: int = None, mode='r') -> str: with open(file_path, mode) as out_file: if content_len: content = out_file.read(content_len) else: content = out_file.read() return content def split_file(file_path: str, parts: int) -> list[str]: files = [] with open(file_path, 'rb') as in_file: data = in_file.read() content_size = len(data) chunk_size = int((content_size + parts) / parts) part_id = 1 for start_position in range(0, content_size + 1, chunk_size): part_file_name = f'{file_path}_part_{part_id}' files.append(part_file_name) with open(part_file_name, 'wb') as out_file: out_file.write(data[start_position:start_position + chunk_size]) part_id += 1 return files def robot_time_to_int(value: str) -> int: if value.endswith('s'): return int(value[:-1]) if value.endswith('m'): return int(value[:-1]) * 60 def placement_policy_from_container(container_info: str) -> str: """ Get placement policy from container info: container ID: j7k4auNHRmiPMSmnH2qENLECD2au2y675fvTX6csDwd version: 2.12 owner ID: NQ8HUxE5qEj7UUvADj7z9Z7pcvJdjtPwuw basic ACL: 0x0fbfbfff (eacl-public-read-write) attribute: Timestamp=1656340345 (2022-06-27 17:32:25 +0300 MSK) nonce: 1c511e88-efd7-4004-8dbf-14391a5d375a placement policy: REP 1 IN LOC_PLACE CBF 1 SELECT 1 FROM LOC_SW AS LOC_PLACE FILTER Country EQ Sweden AS LOC_SW Args: container_info: output from neofs-cli container get command Returns: placement policy as a string """ assert ':' in container_info, f'Could not find placement rule in the output {container_info}' return container_info.split(':')[-1].replace('\n', ' ').strip()