import sys ROOT_DIR = __file__.rsplit("/", 3)[0] if ROOT_DIR not in sys.path: sys.path.append(ROOT_DIR) import uvicorn import os import requests import json from utils.server_tool import get_ip, check_port from fastapi import FastAPI app = FastAPI() BACKEND_DIR = f"{ROOT_DIR}/demo/backend" # Map the function name to the server directory FUNCTION_MAP = { "search": f"{BACKEND_DIR}/servers/retrieval/server_list", "compute": f"{BACKEND_DIR}/servers/retrieval/server_list", "generate_embedding": f"{BACKEND_DIR}/servers/embedding_generation/server_list", } def get_idle_node(server_dir: str, filter_func=None) -> str: """ Find an idle node in the server list to perform the request Returns: server_dir: The directory that contains the server list filter_func: A function to select suitable servers based on different tasks. If it returns True, the server is suitable. """ # Find the first idle node server_list = list(filter(lambda x: x.endswith(".flag"), os.listdir(server_dir))) # Sort by the last modified time server_list.sort(key=lambda file: os.path.getmtime(f"{server_dir}/{file}")) for ip_port in server_list: ip, port = ip_port.split(".flag")[0].split(":") ip_info = f"{server_dir}/{ip_port}" # Remove inaccessible server if not check_port(ip, int(port)): os.remove(ip_info) continue with open(ip_info, "r") as r: try: state_dict = json.load(r) except Exception: continue if state_dict["state"] == "idle": # If a filter function is provided, check if the server meets the criteria if filter_func is not None and not filter_func(state_dict): continue return ip_port.split(".flag")[0] # No idle node raise Exception("No idle node available") @app.get("/search") def search(input: str, topk: int, input_type: str, query_type: str, subsection_type: str, db: str): """ This function is used for multi-modal search Args: input: Input query topk: Number of results to return input_type: Type of input, e.g., "sequence", "structure", "text" query_type: Type of database to search, e.g., "sequence", "structure", "text" subsection_type: If db_type is text, search in this subsection db: Database name for a specific db_type, e.g., "uniprot", "pdb" in sequence databases """ def filter_func(state_dict): # Check if the server contains the required database db_list = state_dict[query_type] if db not in db_list: return False else: return True ip = get_idle_node(FUNCTION_MAP["search"], filter_func) print(ip) # Send request to the idle node url = f"http://{ip}/search" params = { "manager_ip_port": f"{get_ip()}:7861", "input": input, "topk": topk, "input_type": input_type, "query_type": query_type, "subsection_type": subsection_type, "db": db, } response = requests.get(url=url, params=params).json() return response @app.get("/compute") def compute_score(input_type_1: str, input_1: str, input_type_2: str, input_2: str): """ This function is used to compute the similarity score between two inputs Args: input_type_1: Type of input 1, e.g., "sequence", "structure", "text" input_1: Input query 1 input_type_2: Type of input 2, e.g., "sequence", "structure", "text" input_2: Input query 2 """ ip = get_idle_node(FUNCTION_MAP["compute"]) # Send request to the idle node url = f"http://{ip}/compute" params = { "manager_ip_port": f"{get_ip()}:7861", "input_type_1": input_type_1, "input_1": input_1, "input_type_2": input_type_2, "input_2": input_2, } response = requests.get(url=url, params=params).json() return response @app.get("/generate_embedding") def generate_embedding(input: str, input_type: str): """ This function is used for generating embeddings Args: input: Input query input_type: Type of input, e.g., "sequence", "structure", "text" """ ip = get_idle_node(FUNCTION_MAP["generate_embedding"]) # Send request to the idle node url = f"http://{ip}/generate_embedding" params = { "input": input, "input_type": input_type, } response = requests.get(url=url, params=params).json() return response PORT = 7861 if __name__ == "__main__": uvicorn.run("server_manager:app", host="0.0.0.0", port=7861)