def _run_elastic(args):
# construct host discovery component
if args.host_discovery_script:
discover_hosts = discovery.HostDiscoveryScript(
args.host_discovery_script, args.slots)
elif args.hosts:
_, available_host_slots = hosts.parse_hosts_and_slots(args.hosts)
if len(available_host_slots) < 2:
raise ValueError(
'Cannot run in fault tolerance mode with fewer than 2 hosts.')
discover_hosts = discovery.FixedHosts(available_host_slots)
else:
raise ValueError(
'One of --host-discovery-script, --hosts, or --hostnames must be provided'
)
# horovodrun has to finish all the checks before this timeout runs out.
if args.start_timeout:
start_timeout = args.start_timeout
else:
# Lookup default timeout from the environment variable.
start_timeout = int(os.getenv('HOROVOD_START_TIMEOUT', '30'))
tmout = timeout.Timeout(start_timeout,
message='Timed out waiting for {activity}. Please '
'check connectivity between servers. You '
'may need to increase the --start-timeout '
'parameter if you have too many servers.')
settings = elastic_settings.ElasticSettings(
discovery=discover_hosts,
min_num_proc=args.min_num_proc or args.num_proc,
max_num_proc=args.max_num_proc,
elastic_timeout=args.elastic_timeout,
reset_limit=args.reset_limit,
cooldown_range=args.cooldown_range,
num_proc=args.num_proc,
verbose=2 if args.verbose else 0,
ssh_port=args.ssh_port,
ssh_identity_file=args.ssh_identity_file,
extra_mpi_args=args.mpi_args,
key=secret.make_secret_key(),
start_timeout=tmout,
output_filename=args.output_filename,
run_func_mode=args.run_func is not None,
nics=args.nics,
prefix_output_with_timestamp=args.prefix_output_with_timestamp)
if not gloo_built(verbose=(settings.verbose >= 2)):
raise ValueError(
'Gloo support is required to use elastic training, but has not been built. Ensure CMake is '
'installed and reinstall Horovod with HOROVOD_WITH_GLOO=1 to debug the build error.'
)
env = os.environ.copy()
config_parser.set_env_from_args(env, args)
executable = args.executable or sys.executable
return gloo_run_elastic(settings, env,
args.run_func if args.run_func else args.command,
executable)
def elastic_driver_fn():
global_rendezv = RendezvousServer(verbose=1)
discover_hosts = discovery.HostDiscoveryScript(
"/Users/zuston/iqiyiDev/horovod-opal/dis.sh", 3)
driver = ElasticDriver(global_rendezv, discover_hosts, min_np=2, max_np=4)
handler = create_rendezvous_handler(driver)
global_rendezv_port = global_rendezv.start(handler)
print('port: ' + str(global_rendezv_port))
print('wait for available slots: {}'.format(2))
current_hosts = driver.wait_for_available_slots(2)
print("current hosts:" + str(current_hosts))
pending_slots = driver._update_host_assignments(current_hosts)
print("pending hosts:" + str(pending_slots))
driver._worker_registry.reset(driver.world_size())