def _bootstrap_config(config: Dict[str, Any],
no_config_cache: bool = False) -> Dict[str, Any]:
config = prepare_config(config)
hasher = hashlib.sha1()
hasher.update(json.dumps([config], sort_keys=True).encode("utf-8"))
cache_key = os.path.join(tempfile.gettempdir(),
"ray-config-{}".format(hasher.hexdigest()))
if os.path.exists(cache_key) and not no_config_cache:
config_cache = json.loads(open(cache_key).read())
if config_cache.get("_version", -1) == CONFIG_CACHE_VERSION:
# todo: is it fine to re-resolve? afaik it should be.
# we can have migrations otherwise or something
# but this seems overcomplicated given that resolving is
# relatively cheap
try_reload_log_state(config_cache["config"]["provider"],
config_cache.get("provider_log_info"))
if log_once("_printed_cached_config_warning"):
cli_logger.verbose_warning(
"Loaded cached provider configuration "
"from " + cf.bold("{}"), cache_key)
if cli_logger.verbosity == 0:
cli_logger.warning("Loaded cached provider configuration")
cli_logger.warning(
"If you experience issues with "
"the cloud provider, try re-running "
"the command with {}.", cf.bold("--no-config-cache"))
return config_cache["config"]
else:
cli_logger.warning(
"Found cached cluster config "
"but the version " + cf.bold("{}") + " "
"(expected " + cf.bold("{}") + ") does not match.\n"
"This is normal if cluster launcher was updated.\n"
"Config will be re-resolved.",
config_cache.get("_version", "none"), CONFIG_CACHE_VERSION)
validate_config(config)
importer = _NODE_PROVIDERS.get(config["provider"]["type"])
if not importer:
raise NotImplementedError("Unsupported provider {}".format(
config["provider"]))
provider_cls = importer(config["provider"])
cli_logger.print("Checking {} environment settings",
_PROVIDER_PRETTY_NAMES.get(config["provider"]["type"]))
resolved_config = provider_cls.bootstrap_config(config)
if not no_config_cache:
with open(cache_key, "w") as f:
config_cache = {
"_version": CONFIG_CACHE_VERSION,
"provider_log_info": try_get_log_state(config["provider"]),
"config": resolved_config
}
f.write(json.dumps(config_cache))
return resolved_config
def log_to_cli(config):
provider_name = _PROVIDER_PRETTY_NAMES.get("aws", None)
cli_logger.doassert(provider_name is not None,
"Could not find a pretty name for the AWS provider.")
with cli_logger.group("{} config", provider_name):
def same_everywhere(key):
return config["head_node"][key] == config["worker_nodes"][key]
def print_info(resource_string,
key,
head_src_key,
workers_src_key,
allowed_tags=None,
list_value=False):
if allowed_tags is None:
allowed_tags = ["default"]
head_tags = {}
workers_tags = {}
if _log_info[head_src_key] in allowed_tags:
head_tags[_log_info[head_src_key]] = True
if _log_info[workers_src_key] in allowed_tags:
workers_tags[_log_info[workers_src_key]] = True
head_value_str = config["head_node"][key]
if list_value:
head_value_str = cli_logger.render_list(head_value_str)
if same_everywhere(key):
cli_logger.labeled_value( # todo: handle plural vs singular?
resource_string + " (head & workers)",
"{}",
head_value_str,
_tags=head_tags)
else:
workers_value_str = config["worker_nodes"][key]
if list_value:
workers_value_str = cli_logger.render_list(
workers_value_str)
cli_logger.labeled_value(
resource_string + " (head)",
"{}",
head_value_str,
_tags=head_tags)
cli_logger.labeled_value(
resource_string + " (workers)",
"{}",
workers_value_str,
_tags=workers_tags)
tags = {"default": _log_info["head_instance_profile_src"] == "default"}
profile_arn = config["head_node"]["IamInstanceProfile"].get("Arn")
profile_name = _arn_to_name(profile_arn) \
if profile_arn \
else config["head_node"]["IamInstanceProfile"]["Name"]
cli_logger.labeled_value("IAM Profile", "{}", profile_name, _tags=tags)
if ("KeyName" in config["head_node"]
and "KeyName" in config["worker_nodes"]):
print_info("EC2 Key pair", "KeyName", "keypair_src", "keypair_src")
print_info(
"VPC Subnets",
"SubnetIds",
"head_subnet_src",
"workers_subnet_src",
list_value=True)
print_info(
"EC2 Security groups",
"SecurityGroupIds",
"head_security_group_src",
"workers_security_group_src",
list_value=True)
print_info(
"EC2 AMI",
"ImageId",
"head_ami_src",
"workers_ami_src",
allowed_tags=["dlami"])
cli_logger.newline()
"(expected " + cf.bold("{}") + ") does not match.\n"
"This is normal if cluster launcher was updated.\n"
"Config will be re-resolved.",
config_cache.get("_version", "none"), CONFIG_CACHE_VERSION)
validate_config(config)
importer = _NODE_PROVIDERS.get(config["provider"]["type"])
if not importer:
raise NotImplementedError("Unsupported provider {}".format(
config["provider"]))
provider_cls = importer(config["provider"])
<<<<<<< HEAD:python/ray/autoscaler/_private/commands.py
cli_logger.print("Checking {} environment settings",
_PROVIDER_PRETTY_NAMES.get(config["provider"]["type"]))
resolved_config = provider_cls.bootstrap_config(config)
=======
with cli_logger.timed(
"Checking {} environment settings",
PROVIDER_PRETTY_NAMES.get(config["provider"]["type"])):
resolved_config = provider_cls.bootstrap_config(config)
>>>>>>> upstream/releases/1.0.0:python/ray/autoscaler/commands.py
if not no_config_cache:
with open(cache_key, "w") as f:
config_cache = {
"_version": CONFIG_CACHE_VERSION,
"provider_log_info": try_get_log_state(config["provider"]),
"config": resolved_config
}
def _bootstrap_config(config: Dict[str, Any],
no_config_cache: bool = False) -> Dict[str, Any]:
config = prepare_config(config)
hasher = hashlib.sha1()
hasher.update(json.dumps([config], sort_keys=True).encode("utf-8"))
cache_key = os.path.join(tempfile.gettempdir(),
"ray-config-{}".format(hasher.hexdigest()))
if os.path.exists(cache_key) and not no_config_cache:
config_cache = json.loads(open(cache_key).read())
if config_cache.get("_version", -1) == CONFIG_CACHE_VERSION:
# todo: is it fine to re-resolve? afaik it should be.
# we can have migrations otherwise or something
# but this seems overcomplicated given that resolving is
# relatively cheap
try_reload_log_state(config_cache["config"]["provider"],
config_cache.get("provider_log_info"))
if log_once("_printed_cached_config_warning"):
cli_logger.verbose_warning(
"Loaded cached provider configuration "
"from " + cf.bold("{}"), cache_key)
if cli_logger.verbosity == 0:
cli_logger.warning("Loaded cached provider configuration")
cli_logger.warning(
"If you experience issues with "
"the cloud provider, try re-running "
"the command with {}.", cf.bold("--no-config-cache"))
return config_cache["config"]
else:
cli_logger.warning(
"Found cached cluster config "
"but the version " + cf.bold("{}") + " "
"(expected " + cf.bold("{}") + ") does not match.\n"
"This is normal if cluster launcher was updated.\n"
"Config will be re-resolved.",
config_cache.get("_version", "none"), CONFIG_CACHE_VERSION)
importer = _NODE_PROVIDERS.get(config["provider"]["type"])
if not importer:
raise NotImplementedError("Unsupported provider {}".format(
config["provider"]))
provider_cls = importer(config["provider"])
cli_logger.print("Checking {} environment settings",
_PROVIDER_PRETTY_NAMES.get(config["provider"]["type"]))
try:
config = provider_cls.fillout_available_node_types_resources(config)
except Exception as exc:
if cli_logger.verbosity > 2:
logger.exception("Failed to autodetect node resources.")
else:
cli_logger.warning(
f"Failed to autodetect node resources: {str(exc)}. "
"You can see full stack trace with higher verbosity.")
# NOTE: if `resources` field is missing, validate_config for providers
# other than AWS and Kubernetes will fail (the schema error will ask the
# user to manually fill the resources) as we currently support autofilling
# resources for AWS and Kubernetes only.
validate_config(config)
resolved_config = provider_cls.bootstrap_config(config)
if not no_config_cache:
with open(cache_key, "w") as f:
config_cache = {
"_version": CONFIG_CACHE_VERSION,
"provider_log_info": try_get_log_state(config["provider"]),
"config": resolved_config
}
f.write(json.dumps(config_cache))
return resolved_config
def log_to_cli(config: Dict[str, Any]) -> None:
provider_name = _PROVIDER_PRETTY_NAMES.get("aws", None)
cli_logger.doassert(
provider_name is not None, "Could not find a pretty name for the AWS provider."
)
head_node_type = config["head_node_type"]
head_node_config = config["available_node_types"][head_node_type]["node_config"]
with cli_logger.group("{} config", provider_name):
def print_info(
resource_string: str,
key: str,
src_key: str,
allowed_tags: Optional[List[str]] = None,
list_value: bool = False,
) -> None:
if allowed_tags is None:
allowed_tags = ["default"]
node_tags = {}
# set of configurations corresponding to `key`
unique_settings = set()
for node_type_key, node_type in config["available_node_types"].items():
node_tags[node_type_key] = {}
tag = _log_info[src_key][node_type_key]
if tag in allowed_tags:
node_tags[node_type_key][tag] = True
setting = node_type["node_config"].get(key)
if list_value:
unique_settings.add(tuple(setting))
else:
unique_settings.add(setting)
head_value_str = head_node_config[key]
if list_value:
head_value_str = cli_logger.render_list(head_value_str)
if len(unique_settings) == 1:
# all node types are configured the same, condense
# log output
cli_logger.labeled_value(
resource_string + " (all available node types)",
"{}",
head_value_str,
_tags=node_tags[config["head_node_type"]],
)
else:
# do head node type first
cli_logger.labeled_value(
resource_string + f" ({head_node_type})",
"{}",
head_value_str,
_tags=node_tags[head_node_type],
)
# go through remaining types
for node_type_key, node_type in config["available_node_types"].items():
if node_type_key == head_node_type:
continue
workers_value_str = node_type["node_config"][key]
if list_value:
workers_value_str = cli_logger.render_list(workers_value_str)
cli_logger.labeled_value(
resource_string + f" ({node_type_key})",
"{}",
workers_value_str,
_tags=node_tags[node_type_key],
)
tags = {"default": _log_info["head_instance_profile_src"] == "default"}
# head_node_config is the head_node_type's config,
# config["head_node"] is a field that gets applied only to the actual
# head node (and not workers of the head's node_type)
assert (
"IamInstanceProfile" in head_node_config
or "IamInstanceProfile" in config["head_node"]
)
if "IamInstanceProfile" in head_node_config:
# If the user manually configured the role we're here.
IamProfile = head_node_config["IamInstanceProfile"]
elif "IamInstanceProfile" in config["head_node"]:
# If we filled the default IAM role, we're here.
IamProfile = config["head_node"]["IamInstanceProfile"]
profile_arn = IamProfile.get("Arn")
profile_name = _arn_to_name(profile_arn) if profile_arn else IamProfile["Name"]
cli_logger.labeled_value("IAM Profile", "{}", profile_name, _tags=tags)
if all(
"KeyName" in node_type["node_config"]
for node_type in config["available_node_types"].values()
):
print_info("EC2 Key pair", "KeyName", "keypair_src")
print_info("VPC Subnets", "SubnetIds", "subnet_src", list_value=True)
print_info(
"EC2 Security groups",
"SecurityGroupIds",
"security_group_src",
list_value=True,
)
print_info("EC2 AMI", "ImageId", "ami_src", allowed_tags=["dlami"])
cli_logger.newline()
