def submit(
address: Optional[str],
job_id: Optional[str],
runtime_env: Optional[str],
runtime_env_json: Optional[str],
working_dir: Optional[str],
entrypoint: Tuple[str],
no_wait: bool,
):
"""Submits a job to be run on the cluster.
Example:
ray job submit -- python my_script.py --arg=val
"""
client = _get_sdk_client(address, create_cluster_if_needed=True)
final_runtime_env = parse_runtime_env_args(
runtime_env=runtime_env,
runtime_env_json=runtime_env_json,
working_dir=working_dir,
)
job_id = client.submit_job(
entrypoint=list2cmdline(entrypoint),
job_id=job_id,
runtime_env=final_runtime_env,
)
_log_big_success_msg(f"Job '{job_id}' submitted successfully")
with cli_logger.group("Next steps"):
cli_logger.print("Query the logs of the job:")
with cli_logger.indented():
cli_logger.print(cf.bold(f"ray job logs {job_id}"))
cli_logger.print("Query the status of the job:")
with cli_logger.indented():
cli_logger.print(cf.bold(f"ray job status {job_id}"))
cli_logger.print("Request the job to be stopped:")
with cli_logger.indented():
cli_logger.print(cf.bold(f"ray job stop {job_id}"))
cli_logger.newline()
sdk_version = client.get_version()
# sdk version 0 does not have log streaming
if not no_wait:
if int(sdk_version) > 0:
cli_logger.print("Tailing logs until the job exits "
"(disable with --no-wait):")
asyncio.get_event_loop().run_until_complete(
_tail_logs(client, job_id))
else:
cli_logger.warning(
"Tailing logs is not enabled for job sdk client version "
f"{sdk_version}. Please upgrade your ray to latest version "
"for this feature.")
def run(
self,
cmd,
timeout=120,
exit_on_fail=False,
port_forward=None,
with_output=False,
environment_variables: Dict[str, object] = None,
run_env="auto", # Unused argument.
ssh_options_override_ssh_key="",
shutdown_after_run=False,
silent=False):
if shutdown_after_run:
cmd += "; sudo shutdown -h now"
if ssh_options_override_ssh_key:
ssh_options = SSHOptions(ssh_options_override_ssh_key)
else:
ssh_options = self.ssh_options
assert isinstance(
ssh_options, SSHOptions
), "ssh_options must be of type SSHOptions, got {}".format(
type(ssh_options))
self._set_ssh_ip_if_required()
if is_using_login_shells():
ssh = ["ssh", "-tt"]
else:
ssh = ["ssh"]
if port_forward:
with cli_logger.group("Forwarding ports"):
if not isinstance(port_forward, list):
port_forward = [port_forward]
for local, remote in port_forward:
cli_logger.verbose(
"Forwarding port {} to port {} on localhost.",
cf.bold(local), cf.bold(remote)) # todo: msg
ssh += ["-L", "{}:localhost:{}".format(remote, local)]
final_cmd = ssh + ssh_options.to_ssh_options_list(
timeout=timeout) + ["{}@{}".format(self.ssh_user, self.ssh_ip)]
if cmd:
if environment_variables:
cmd = _with_environment_variables(cmd, environment_variables)
if is_using_login_shells():
final_cmd += _with_interactive(cmd)
else:
final_cmd += [cmd]
else:
# We do this because `-o ControlMaster` causes the `-N` flag to
# still create an interactive shell in some ssh versions.
final_cmd.append("while true; do sleep 86400; done")
cli_logger.verbose("Running `{}`", cf.bold(cmd))
with cli_logger.indented():
cli_logger.very_verbose("Full command is `{}`",
cf.bold(" ".join(final_cmd)))
if cli_logger.verbosity > 0:
with cli_logger.indented():
return self._run_helper(final_cmd,
with_output,
exit_on_fail,
silent=silent)
else:
return self._run_helper(final_cmd,
with_output,
exit_on_fail,
silent=silent)
def job_submit(
address: Optional[str],
job_id: Optional[str],
runtime_env: Optional[str],
runtime_env_json: Optional[str],
working_dir: Optional[str],
entrypoint: Tuple[str],
no_wait: bool,
):
"""Submits a job to be run on the cluster.
Example:
>>> ray job submit -- python my_script.py --arg=val
"""
client = _get_sdk_client(address, create_cluster_if_needed=True)
final_runtime_env = {}
if runtime_env is not None:
if runtime_env_json is not None:
raise ValueError("Only one of --runtime_env and "
"--runtime-env-json can be provided.")
with open(runtime_env, "r") as f:
final_runtime_env = yaml.safe_load(f)
elif runtime_env_json is not None:
final_runtime_env = json.loads(runtime_env_json)
if working_dir is not None:
if "working_dir" in final_runtime_env:
cli_logger.warning(
"Overriding runtime_env working_dir with --working-dir option")
final_runtime_env["working_dir"] = working_dir
job_id = client.submit_job(
entrypoint=list2cmdline(entrypoint),
job_id=job_id,
runtime_env=final_runtime_env,
)
_log_big_success_msg(f"Job '{job_id}' submitted successfully")
with cli_logger.group("Next steps"):
cli_logger.print("Query the logs of the job:")
with cli_logger.indented():
cli_logger.print(cf.bold(f"ray job logs {job_id}"))
cli_logger.print("Query the status of the job:")
with cli_logger.indented():
cli_logger.print(cf.bold(f"ray job status {job_id}"))
cli_logger.print("Request the job to be stopped:")
with cli_logger.indented():
cli_logger.print(cf.bold(f"ray job stop {job_id}"))
cli_logger.newline()
sdk_version = client.get_version()
# sdk version 0 does not have log streaming
if not no_wait:
if int(sdk_version) > 0:
cli_logger.print("Tailing logs until the job exits "
"(disable with --no-wait):")
asyncio.get_event_loop().run_until_complete(
_tail_logs(client, job_id))
else:
cli_logger.warning(
"Tailing logs is not enabled for job sdk client version "
f"{sdk_version}. Please upgrade your ray to latest version "
"for this feature.")
cli_logger.print("List: {}", cli_logger.render_list([1, 2, 3]))
cli_logger.newline()
cli_logger.very_verbose("Very verbose")
cli_logger.verbose("Verbose")
cli_logger.verbose_warning("Verbose warning")
cli_logger.verbose_error("Verbose error")
cli_logger.print("Info")
cli_logger.success("Success")
cli_logger.warning("Warning")
cli_logger.error("Error")
cli_logger.newline()
try:
cli_logger.abort("Abort")
except Exception:
pass
try:
cli_logger.doassert(False, "Assert")
except Exception:
pass
cli_logger.newline()
cli_logger.confirm(True, "example")
cli_logger.newline()
with cli_logger.indented():
cli_logger.print("Indented")
with cli_logger.group("Group"):
cli_logger.print("Group contents")
with cli_logger.timed("Timed (unimplemented)"):
cli_logger.print("Timed contents")
with cli_logger.verbatim_error_ctx("Verbtaim error"):
cli_logger.print("Error contents")
def start(node_ip_address, address, port, redis_password, redis_shard_ports,
object_manager_port, node_manager_port, gcs_server_port,
min_worker_port, max_worker_port, worker_port_list, memory,
object_store_memory, redis_max_memory, num_cpus, num_gpus, resources,
head, include_dashboard, dashboard_host, dashboard_port, block,
plasma_directory, autoscaling_config, no_redirect_worker_output,
no_redirect_output, plasma_store_socket_name, raylet_socket_name,
temp_dir, java_worker_options, load_code_from_local,
code_search_path, system_config, lru_evict,
enable_object_reconstruction, metrics_export_port, log_style,
log_color, verbose):
"""Start Ray processes manually on the local machine."""
cli_logger.configure(log_style, log_color, verbose)
if gcs_server_port and not head:
raise ValueError(
"gcs_server_port can be only assigned when you specify --head.")
# Convert hostnames to numerical IP address.
if node_ip_address is not None:
node_ip_address = services.address_to_ip(node_ip_address)
redis_address = None
if address is not None:
(redis_address, redis_address_ip,
redis_address_port) = services.validate_redis_address(address)
try:
resources = json.loads(resources)
except Exception:
cli_logger.error("`{}` is not a valid JSON string.",
cf.bold("--resources"))
cli_logger.abort(
"Valid values look like this: `{}`",
cf.bold("--resources='\"CustomResource3\": 1, "
"\"CustomResource2\": 2}'"))
raise Exception("Unable to parse the --resources argument using "
"json.loads. Try using a format like\n\n"
" --resources='{\"CustomResource1\": 3, "
"\"CustomReseource2\": 2}'")
redirect_worker_output = None if not no_redirect_worker_output else True
redirect_output = None if not no_redirect_output else True
ray_params = ray.parameter.RayParams(
node_ip_address=node_ip_address,
min_worker_port=min_worker_port,
max_worker_port=max_worker_port,
worker_port_list=worker_port_list,
object_manager_port=object_manager_port,
node_manager_port=node_manager_port,
gcs_server_port=gcs_server_port,
memory=memory,
object_store_memory=object_store_memory,
redis_password=redis_password,
redirect_worker_output=redirect_worker_output,
redirect_output=redirect_output,
num_cpus=num_cpus,
num_gpus=num_gpus,
resources=resources,
plasma_directory=plasma_directory,
huge_pages=False,
plasma_store_socket_name=plasma_store_socket_name,
raylet_socket_name=raylet_socket_name,
temp_dir=temp_dir,
include_dashboard=include_dashboard,
dashboard_host=dashboard_host,
dashboard_port=dashboard_port,
java_worker_options=java_worker_options,
load_code_from_local=load_code_from_local,
code_search_path=code_search_path,
_system_config=system_config,
lru_evict=lru_evict,
enable_object_reconstruction=enable_object_reconstruction,
metrics_export_port=metrics_export_port)
if head:
# Use default if port is none, allocate an available port if port is 0
if port is None:
port = ray_constants.DEFAULT_PORT
if port == 0:
with socket() as s:
s.bind(("", 0))
port = s.getsockname()[1]
num_redis_shards = None
# Start Ray on the head node.
if redis_shard_ports is not None:
redis_shard_ports = redis_shard_ports.split(",")
# Infer the number of Redis shards from the ports if the number is
# not provided.
num_redis_shards = len(redis_shard_ports)
if redis_address is not None:
cli_logger.abort(
"`{}` starts a new Redis server, `{}` should not be set.",
cf.bold("--head"), cf.bold("--address"))
raise Exception("If --head is passed in, a Redis server will be "
"started, so a Redis address should not be "
"provided.")
node_ip_address = services.get_node_ip_address()
# Get the node IP address if one is not provided.
ray_params.update_if_absent(node_ip_address=node_ip_address)
cli_logger.labeled_value("Local node IP", ray_params.node_ip_address)
ray_params.update_if_absent(
redis_port=port,
redis_shard_ports=redis_shard_ports,
redis_max_memory=redis_max_memory,
num_redis_shards=num_redis_shards,
redis_max_clients=None,
autoscaling_config=autoscaling_config,
)
# Fail early when starting a new cluster when one is already running
if address is None:
default_address = f"{node_ip_address}:{port}"
redis_addresses = services.find_redis_address(default_address)
if len(redis_addresses) > 0:
raise ConnectionError(
f"Ray is already running at {default_address}. "
f"Please specify a different port using the `--port`"
f" command to `ray start`.")
node = ray.node.Node(
ray_params, head=True, shutdown_at_exit=block, spawn_reaper=block)
redis_address = node.redis_address
# this is a noop if new-style is not set, so the old logger calls
# are still in place
cli_logger.newline()
startup_msg = "Ray runtime started."
cli_logger.success("-" * len(startup_msg))
cli_logger.success(startup_msg)
cli_logger.success("-" * len(startup_msg))
cli_logger.newline()
with cli_logger.group("Next steps"):
cli_logger.print(
"To connect to this Ray runtime from another node, run")
cli_logger.print(
cf.bold(" ray start --address='{}'{}"), redis_address,
f" --redis-password='******'"
if redis_password else "")
cli_logger.newline()
cli_logger.print("Alternatively, use the following Python code:")
with cli_logger.indented():
with cf.with_style("monokai") as c:
cli_logger.print("{} ray", c.magenta("import"))
cli_logger.print(
"ray{}init(address{}{}{})", c.magenta("."),
c.magenta("="), c.yellow("'auto'"),
", _redis_password{}{}".format(
c.magenta("="),
c.yellow("'" + redis_password + "'"))
if redis_password else "")
cli_logger.newline()
cli_logger.print(
cf.underlined("If connection fails, check your "
"firewall settings and "
"network configuration."))
cli_logger.newline()
cli_logger.print("To terminate the Ray runtime, run")
cli_logger.print(cf.bold(" ray stop"))
else:
# Start Ray on a non-head node.
if not (port is None):
cli_logger.abort("`{}` should not be specified without `{}`.",
cf.bold("--port"), cf.bold("--head"))
raise Exception("If --head is not passed in, --port is not "
"allowed.")
if redis_shard_ports is not None:
cli_logger.abort("`{}` should not be specified without `{}`.",
cf.bold("--redis-shard-ports"), cf.bold("--head"))
raise Exception("If --head is not passed in, --redis-shard-ports "
"is not allowed.")
if redis_address is None:
cli_logger.abort("`{}` is required unless starting with `{}`.",
cf.bold("--address"), cf.bold("--head"))
raise Exception("If --head is not passed in, --address must "
"be provided.")
if include_dashboard:
cli_logger.abort("`{}` should not be specified without `{}`.",
cf.bold("--include-dashboard"), cf.bold("--head"))
raise ValueError(
"If --head is not passed in, the --include-dashboard"
"flag is not relevant.")
# Wait for the Redis server to be started. And throw an exception if we
# can't connect to it.
services.wait_for_redis_to_start(
redis_address_ip, redis_address_port, password=redis_password)
# Create a Redis client.
redis_client = services.create_redis_client(
redis_address, password=redis_password)
# Check that the version information on this node matches the version
# information that the cluster was started with.
services.check_version_info(redis_client)
# Get the node IP address if one is not provided.
ray_params.update_if_absent(
node_ip_address=services.get_node_ip_address(redis_address))
cli_logger.labeled_value("Local node IP", ray_params.node_ip_address)
# Check that there aren't already Redis clients with the same IP
# address connected with this Redis instance. This raises an exception
# if the Redis server already has clients on this node.
check_no_existing_redis_clients(ray_params.node_ip_address,
redis_client)
ray_params.update(redis_address=redis_address)
node = ray.node.Node(
ray_params, head=False, shutdown_at_exit=block, spawn_reaper=block)
cli_logger.newline()
startup_msg = "Ray runtime started."
cli_logger.success("-" * len(startup_msg))
cli_logger.success(startup_msg)
cli_logger.success("-" * len(startup_msg))
cli_logger.newline()
cli_logger.print("To terminate the Ray runtime, run")
cli_logger.print(cf.bold(" ray stop"))
if block:
cli_logger.newline()
with cli_logger.group(cf.bold("--block")):
cli_logger.print(
"This command will now block until terminated by a signal.")
cli_logger.print(
"Runing subprocesses are monitored and a message will be "
"printed if any of them terminate unexpectedly.")
while True:
time.sleep(1)
deceased = node.dead_processes()
if len(deceased) > 0:
cli_logger.newline()
cli_logger.error("Some Ray subprcesses exited unexpectedly:")
with cli_logger.indented():
for process_type, process in deceased:
cli_logger.error(
"{}",
cf.bold(str(process_type)),
_tags={"exit code": str(process.returncode)})
# shutdown_at_exit will handle cleanup.
cli_logger.newline()
cli_logger.error("Remaining processes will be killed.")
sys.exit(1)
