def main():
options = common.parse_args(use_isolate_server=True, use_swarming=True)
try:
common.note(
'Archiving directory \'payload\' to %s' % options.isolate_server)
payload_isolated_sha1 = common.capture(
[
'isolateserver.py',
'archive',
'--isolate-server', options.isolate_server,
'payload',
]).split()[0]
common.note(
'Archiving custom .isolated file to %s' % options.isolate_server)
handle, isolated = tempfile.mkstemp(
prefix=u'hello_world', suffix=u'.isolated')
os.close(handle)
try:
data = {
'algo': 'sha-1',
'command': ['python', 'hello_world.py', 'Custom'],
'includes': [payload_isolated_sha1],
'version': '1.0',
}
with open(isolated, 'wb') as f:
json.dump(data, f, sort_keys=True, separators=(',',':'))
isolated_sha1 = common.capture(
[
'isolateserver.py',
'archive',
'--isolate-server', options.isolate_server,
isolated,
]).split()[0]
finally:
common.note('Deleting temporary file, it is not necessary anymore.')
os.remove(isolated)
# Now trigger as usual. You could look at run_exmaple_swarming_involved for
# the involved way but use the short way here.
common.note('Running %s on %s' % (isolated_sha1, options.swarming))
cmd = [
'swarming.py',
'run',
'--swarming', options.swarming,
'--isolate-server', options.isolate_server,
'--dimension', 'os', options.swarming_os,
'--dimension', 'pool', 'default',
'--task-name', options.task_name,
isolated_sha1,
]
if options.idempotent:
cmd.append('--idempotent')
if options.priority is not None:
cmd.extend(('--priority', str(options.priority)))
common.run(cmd, options.verbose)
return 0
except subprocess.CalledProcessError as e:
return e.returncode
def main():
options = common.parse_args(use_isolate_server=True, use_swarming=True)
try:
common.note(
'Archiving directory \'payload\' to %s' % options.isolate_server)
payload_isolated_sha1 = common.capture(
[
'isolateserver.py',
'archive',
'--isolate-server', options.isolate_server,
'payload',
]).split()[0]
common.note(
'Archiving custom .isolated file to %s' % options.isolate_server)
handle, isolated = tempfile.mkstemp(
prefix=u'hello_world', suffix=u'.isolated')
os.close(handle)
try:
data = {
'algo': 'sha-1',
'command': ['python', 'hello_world.py', 'Custom'],
'includes': [payload_isolated_sha1],
'version': '1.0',
}
with open(isolated, 'wb') as f:
json.dump(data, f, sort_keys=True, separators=(',',':'))
isolated_sha1 = common.capture(
[
'isolateserver.py',
'archive',
'--isolate-server', options.isolate_server,
isolated,
]).split()[0]
finally:
common.note('Deleting temporary file, it is not necessary anymore.')
os.remove(isolated)
# Now trigger as usual. You could look at run_exmaple_swarming_involved for
# the involved way but use the short way here.
common.note('Running %s on %s' % (isolated_sha1, options.swarming))
cmd = [
'swarming.py',
'run',
'--swarming', options.swarming,
'--isolate-server', options.isolate_server,
'--dimension', 'os', options.swarming_os,
'--dimension', 'pool', 'default',
'--task-name', options.task_name,
isolated_sha1,
]
if options.idempotent:
cmd.append('--idempotent')
if options.priority is not None:
cmd.extend(('--priority', str(options.priority)))
common.run(cmd, options.verbose)
return 0
except subprocess.CalledProcessError as e:
return e.returncode
def horovod_args(self, mode, controller, run_func=None, command=None):
if mode == 'local':
local_hosts = ['localhost', '127.0.0.1']
remote_hosts = []
elif mode == 'remote':
local_hosts = []
remote_hosts = ['localhost', '127.0.0.1']
else:
local_hosts = ['localhost']
remote_hosts = ['127.0.0.1']
hosts = local_hosts.copy()
hosts.extend(remote_hosts)
if remote_hosts:
ssh_works = _check_all_hosts_ssh_successful(remote_hosts, fn_cache=None)
if not ssh_works:
self.skipTest('password-less ssh to {} is required for this test'
.format(' and '.join(remote_hosts)))
hargs = _HorovodArgs()
hargs.np = 4
hargs.hosts = ','.join(['{}:2'.format(host) for host in hosts])
hargs.use_gloo = controller == 'gloo'
hargs.use_mpi = controller == 'mpi'
hargs.run_func = run_func
hargs.command = [command] if command else None
hargs.nics = []
hargs.verbose = 2
hargs.disable_cache = True
stdout = io.StringIO()
try:
with capture(stdout=stdout):
with mock.patch('horovod.runner.launch.network.filter_local_addresses',
side_effect=lambda hosts: [host for host in hosts if host not in local_hosts]), \
mock.patch('horovod.runner.gloo_run.network.get_local_host_addresses',
return_value=local_hosts), \
mock.patch('horovod.runner.gloo_run.network.resolve_host_address',
side_effect=lambda host: host), \
mock.patch('horovod.runner.mpi_run.os.execve') as exec:
yield hargs, exec
finally:
stdout = stdout.readlines()
print(''.join(stdout), file=sys.stdout)
if mode == 'local':
self.assertIn('Remote host found: \n', stdout)
self.assertIn('All hosts are local, finding the interfaces with address 127.0.0.1\n', stdout)
self.assertEqual(1, len([line for line in stdout if line.startswith('Local interface found ')]))
elif mode == 'mixed':
self.assertIn('Remote host found: 127.0.0.1\n', stdout)
elif mode == 'remote':
self.assertIn('Remote host found: localhost 127.0.0.1\n', stdout)
else:
raise RuntimeError('unknown mode {}'.format(mode))
if mode in ['mixed', 'remote']:
self.assertIn('Checking ssh on all remote hosts.\n', stdout)
self.assertIn('SSH was successful into all the remote hosts.\n', stdout)
self.assertIn('Testing interfaces on all the hosts.\n', stdout)
self.assertIn('Launched horovod server.\n', stdout)
self.assertEqual(2, len([line for line in stdout if line.startswith('Launching horovod task function: ')]))
self.assertEqual(1, len([line for line in stdout if line.startswith('Launching horovod task function: ssh -o PasswordAuthentication=no -o StrictHostKeyChecking=no 127.0.0.1 ')]), stdout)
if mode == 'remote':
self.assertEqual(1, len([line for line in stdout if line.startswith('Launching horovod task function: ssh -o PasswordAuthentication=no -o StrictHostKeyChecking=no localhost ')]))
else:
self.assertEqual(1, len([line for line in stdout if line.startswith('Launching horovod task function: ') and not line.startswith('Launching horovod task function: ssh')]))
self.assertIn('Attempted to launch horovod task servers.\n', stdout)
self.assertIn('Waiting for the hosts to acknowledge.\n', stdout)
self.assertIn('Notified all the hosts that the registration is complete.\n', stdout)
self.assertIn('Waiting for hosts to perform host-to-host interface checking.\n', stdout)
self.assertIn('Host-to-host interface checking successful.\n', stdout)
self.assertIn('Interfaces on all the hosts were successfully checked.\n', stdout)
self.assertEqual(1, len([line for line in stdout if line.startswith('Common interface found: ')]))