(nworker, hostfile, port, sshport, cmd)
| 94 | |
| 95 | |
| 96 | def submit(nworker, hostfile, port, sshport, cmd): |
| 97 | if nworker <= 0: |
| 98 | return |
| 99 | ip_ports = [] |
| 100 | if hostfile is not None: |
| 101 | hosts = parse_hostfile(hostfile) |
| 102 | ip_ports = [h + ":" + str(port) for h in hosts] |
| 103 | else: |
| 104 | assert nworker == 1 |
| 105 | local_host = "127.0.0.1" |
| 106 | hosts = [local_host] * nworker |
| 107 | ports = find_available_ports(local_host, nworker) |
| 108 | for p in ports: |
| 109 | ip_ports.append(local_host + ":" + str(p)) |
| 110 | |
| 111 | local_dir = os.getcwd() + "/" |
| 112 | working_dir = local_dir |
| 113 | |
| 114 | # thread func to run the job |
| 115 | def run(prog): |
| 116 | print("launch prog ", prog) |
| 117 | try: |
| 118 | subprocess.check_call(prog, shell=True) |
| 119 | except subprocess.CalledProcessError as e: |
| 120 | logging.info("subprocess({}) failed({})! {}".format( |
| 121 | e.cmd, e.returncode, e.output)) |
| 122 | os._exit(-1) |
| 123 | |
| 124 | pass_envs = os.environ.copy() |
| 125 | if "HOST_NODE_ADDR" not in os.environ: |
| 126 | pass_envs["HOST_NODE_ADDR"] = hosts[0] |
| 127 | |
| 128 | thread_list = [] |
| 129 | for i in range(nworker): |
| 130 | node = hosts[i % len(hosts)] |
| 131 | |
| 132 | # ssh_port_arg = " -p " + str(sshport) + " " |
| 133 | prog = get_env(pass_envs) + " cd " + working_dir + "; " + cmd |
| 134 | prog = ( |
| 135 | "ssh -o StrictHostKeyChecking=no " |
| 136 | # + ssh_port_arg # no port available in aidi |
| 137 | + node + " '" + prog + "'") |
| 138 | thread = Thread(target=run, args=(prog, )) |
| 139 | thread.setDaemon(True) |
| 140 | thread.start() |
| 141 | thread_list.append(thread) |
| 142 | |
| 143 | for t in thread_list: |
| 144 | t.join() |
| 145 | print("thread join success") |
| 146 | print("process end success") |
| 147 | |
| 148 | |
| 149 | if __name__ == "__main__": |
no test coverage detected