o H h @sddlZddlZddlZddlZddlZddlZddlZddlZddlZddl m Z m Z m Z zddl Z Wn ey=YnweeZddZd de efddZGdd d ZdS) N)OptionalTextIOUnionc Cstjddtjtjd}|D]<}|\}}}}}zt|||}|d|d|WStyI}z|td|WYd}~q d}~wwt d)a Find a free port and binds a temporary socket to it so that the port can be "reserved" until used. .. note:: the returned socket must be closed before using the port, otherwise a ``address already in use`` error will happen. The socket should be held and closed as close to the consumer of the port as possible since otherwise, there is a greater chance of race-condition where a different process may see the port as being free and take it. Returns: a socket binded to the reserved free port Usage:: sock = find_free_port() port = sock.getsockname()[1] sock.close() use_port(port) localhostN)hostportfamilytype)rrrz Socket creation attempt failed: zFailed to create a socket) socket getaddrinfo AF_UNSPEC SOCK_STREAMbindlistenOSErrorcloseprint RuntimeError)addrsaddrrr proto_serd/var/www/vscode/kcb/lib/python3.10/site-packages/torch/distributed/elastic/rendezvous/etcd_server.pyfind_free_ports     rdata_dircCsP|r|durtd|||r&td|tj|dddSdS)Nzstopping etcd serverzdeleting etcd data dir: %sT ignore_errors)pollloggerinfo terminatewaitshutilrmtree) subprocessrrrr stop_etcdCs  r(c @seZdZdZddeefddZdejfddZ de fd d Z defd d Z defd dZ   dde de dee edfddfddZ d dede dee edfddfddZddZd!de ddfddZd"ddZdS)# EtcdServera .. note:: tested on etcd server v3.4.3. Starts and stops a local standalone etcd server on a random free port. Useful for single node, multi-worker launches or testing, where a sidecar etcd server is more convenient than having to separately setup an etcd server. This class registers a termination handler to shutdown the etcd subprocess on exit. This termination handler is NOT a substitute for calling the ``stop()`` method. The following fallback mechanism is used to find the etcd binary: 1. Uses env var TORCHELASTIC_ETCD_BINARY_PATH 2. Uses ``/bin/etcd`` if one exists 3. Uses ``etcd`` from ``PATH`` Usage :: server = EtcdServer("/usr/bin/etcd", 2379, "/tmp/default.etcd") server.start() client = server.get_client() # use client server.stop() Args: etcd_binary_path: path of etcd server binary (see above for fallback path) NrcCspd|_d|_tjt}tj|d}tjd||_ tj |j s%d|_ |r)|nt j dd|_ d|_d|_dS)Nrzbin/etcdTORCHELASTIC_ETCD_BINARY_PATHetcdtorchelastic_etcd_data)prefix)_port_hostospathdirname__file__joinenvironget_etcd_binary_pathisfiletempfilemkdtemp_base_data_dir _etcd_cmd _etcd_proc)selfrrootdefault_etcd_binrrr__init__ns  zEtcdServer.__init__returncCs|jstd|jS)Nz>No etcd server process started. Call etcd_server.start() first)r>rr?rrr_get_etcd_server_processs z#EtcdServer._get_etcd_server_processcC|jS)z)Return the port the server is running on.)r/rDrrrget_portzEtcdServer.get_portcCrF)z)Return the host the server is running on.)r0rDrrrget_hostrHzEtcdServer.get_hostcCs|jd|jS)z,Return the etcd server endpoint (host:port).:)r0r/rDrrr get_endpointszEtcdServer.get_endpoint<timeout num_retriesstderrc Csd} ztj|jt|}tj|dd||||WStyL}z$|d7}t|j t dt|||krBt j |jddWYd}~nd}~wwq)a Start the server, and waits for it to be ready. When this function returns the sever is ready to take requests. Args: timeout: time (in seconds) to wait for the server to be ready before giving up. num_retries: number of retries to start the server. Each retry will wait for max ``timeout`` before considering it as failed. stderr: the standard error file handle. Valid values are `subprocess.PIPE`, `subprocess.DEVNULL`, an existing file descriptor (a positive integer), an existing file object, and `None`. Raises: TimeoutError: if the server is not ready within the specified timeout rT)exist_okz4Failed to start etcd server, got error: %s, retryingrN)r1r2r5r<strmakedirs_start Exceptionr(r>r!warningr%r&atexitregister)r?rNrOrP curr_retriesrrrrrstarts&  zEtcdServer.startcCst}t}|d|_|d}td|jdd|dd|jd|jdd|jd|jd d|jd|g }t d || | t j |d |d |_ ||dS) NrR z --enable-v2z --data-dirz--listen-client-urlszhttp://rJz--advertise-client-urlsz--listen-peer-urlszStarting etcd server: [%s]T) close_fdsrP)r getsocknamer/shlexsplitr5r8r0r!r"rr'Popenr>_wait_for_ready)r?rrNrPsock sock_peer peer_portetcd_cmdrrrrUs0  zEtcdServer._startcCstj|j|jdddS)zNReturn an etcd client object that can be used to make requests to this server./v2 rrversion_prefix read_timeout)r,Clientr0r/rDrrr get_clients zEtcdServer.get_clientcCstj|j|jddd}t|}t|krK|dur,|j}td|z t d|j WdSt yDt dYnwt|kstd)Nrgriz*Etcd server process exited with the code: zetcd server ready. version: %srRz.Timed out waiting for etcd server to be ready!)r,rlr0r/timerEr returncoderr!r"versionrVsleep TimeoutError)r?rNclientmax_timeexitcoderrrrbs$      zEtcdServer._wait_for_readycCstdt|j|jdS)zGStop the server and cleans up auto generated resources (e.g. data dir).zEtcdServer stop method calledN)r!r"r(r>r<rDrrrstops zEtcdServer.stopN)rLrMN)rLN)rL)rCN)__name__ __module__ __qualname____doc__rrSrBr'rarEintrGrIrKrrr[rUrmrbrwrrrrr)Ns@  (   r)rx)rXloggingr1r_r%r r'r:rotypingrrrr,ModuleNotFoundError getLoggerryr!rrSr(r)rrrrs&   %