Files
ray-project--ray/python/ray/autoscaler/node_provider.py
T
2026-07-13 13:17:40 +08:00

305 lines
12 KiB
Python

import logging
from types import ModuleType
from typing import Any, Dict, List, Optional
from ray.autoscaler._private.command_runner import DockerCommandRunner, SSHCommandRunner
from ray.autoscaler.command_runner import CommandRunnerInterface
from ray.util.annotations import DeveloperAPI
logger = logging.getLogger(__name__)
@DeveloperAPI
class NodeProvider:
"""Interface for getting and returning nodes from a Cloud.
**Important**: This is an INTERNAL API that is only exposed for the purpose
of implementing custom node providers. It is not allowed to call into
NodeProvider methods from any Ray package outside the autoscaler, only to
define new implementations of NodeProvider for use with the "external" node
provider option.
NodeProviders are namespaced by the `cluster_name` parameter; they only
operate on nodes within that namespace.
Nodes may be in one of three states: {pending, running, terminated}. Nodes
appear immediately once started by `create_node`, and transition
immediately to terminated when `terminate_node` is called.
Threading and concurrency:
- The autoscaler calls the following methods from multiple threads
(NodeLauncher, NodeUpdaterThread, autoscaler main loop, and
NodeProviderAdapter executors).
- These methods MUST be thread-safe:
non_terminated_nodes, is_running, is_terminated, node_tags, internal_ip,
external_ip, get_node_id, create_node/create_node_with_resources_and_labels,
set_node_tags, terminate_node/terminate_nodes.
TODO (rueian): make sure all the existing implementations are thread-safe.
"""
def __init__(self, provider_config: Dict[str, Any], cluster_name: str) -> None:
self.provider_config = provider_config
self.cluster_name = cluster_name
self._internal_ip_cache: Dict[str, str] = {}
self._external_ip_cache: Dict[str, str] = {}
def is_readonly(self) -> bool:
"""Returns whether this provider is readonly.
Readonly node providers do not allow nodes to be created or terminated.
"""
return False
def non_terminated_nodes(self, tag_filters: Dict[str, str]) -> List[str]:
"""Return a list of node ids filtered by the specified tags dict.
This list must not include terminated nodes. For performance reasons,
providers are allowed to cache the result of a call to
non_terminated_nodes() to serve single-node queries
(e.g. is_running(node_id)). This means that non_terminate_nodes() must
be called again to refresh results.
Args:
tag_filters: Tag key/value pairs that nodes must match to be
included in the result.
Returns:
A list of node ids matching the given tag filters.
Examples:
>>> from ray.autoscaler.node_provider import NodeProvider
>>> from ray.autoscaler.tags import TAG_RAY_NODE_KIND
>>> provider = NodeProvider(...) # doctest: +SKIP
>>> provider.non_terminated_nodes( # doctest: +SKIP
... {TAG_RAY_NODE_KIND: "worker"})
["node-1", "node-2"]
"""
raise NotImplementedError
def nodes_for_teardown(self, tag_filters: Dict[str, str]) -> List[str]:
"""Return all node ids matching tag_filters, including terminated nodes.
Used during teardown to ensure cleanup of external resources (e.g.
Docker containers) on nodes whose state may not be accurately tracked
by this provider instance. For example, LocalNodeProvider on the
machine invoking ``ray down`` may show workers as terminated even
though the head node's autoscaler started them and their Docker
containers are still running.
The default delegates to non_terminated_nodes(), which is correct for
cloud providers that always query live infrastructure state. Providers
that maintain state locally should override this to include all known
nodes regardless of recorded state.
"""
return self.non_terminated_nodes(tag_filters)
def is_running(self, node_id: str) -> bool:
"""Return whether the specified node is running."""
raise NotImplementedError
def is_terminated(self, node_id: str) -> bool:
"""Return whether the specified node is terminated."""
raise NotImplementedError
def node_tags(self, node_id: str) -> Dict[str, str]:
"""Returns the tags of the given node (string dict)."""
raise NotImplementedError
def external_ip(self, node_id: str) -> str:
"""Returns the external ip of the given node."""
raise NotImplementedError
def internal_ip(self, node_id: str) -> str:
"""Returns the internal ip (Ray ip) of the given node."""
raise NotImplementedError
def get_node_id(self, ip_address: str, use_internal_ip: bool = False) -> str:
"""Returns the node_id given an IP address.
Assumes ip-address is unique per node.
Args:
ip_address: Address of node.
use_internal_ip: Whether the ip address is
public or private.
Returns:
The node id corresponding to the given IP address.
Raises:
ValueError: If not found.
"""
def find_node_id():
if use_internal_ip:
return self._internal_ip_cache.get(ip_address)
else:
return self._external_ip_cache.get(ip_address)
if not find_node_id():
all_nodes = self.non_terminated_nodes({})
ip_func = self.internal_ip if use_internal_ip else self.external_ip
ip_cache = (
self._internal_ip_cache if use_internal_ip else self._external_ip_cache
)
for node_id in all_nodes:
ip_cache[ip_func(node_id)] = node_id
if not find_node_id():
if use_internal_ip:
known_msg = f"Worker internal IPs: {list(self._internal_ip_cache)}"
else:
known_msg = f"Worker external IP: {list(self._external_ip_cache)}"
raise ValueError(f"ip {ip_address} not found. " + known_msg)
return find_node_id()
def create_node(
self, node_config: Dict[str, Any], tags: Dict[str, str], count: int
) -> Optional[Dict[str, Any]]:
"""Creates a number of nodes within the namespace.
Optionally returns a mapping from created node ids to node metadata.
Optionally may throw a
ray.autoscaler.node_launch_exception.NodeLaunchException which the
autoscaler may use to provide additional functionality such as
observability.
"""
raise NotImplementedError
def create_node_with_resources_and_labels(
self,
node_config: Dict[str, Any],
tags: Dict[str, str],
count: int,
resources: Dict[str, float],
labels: Dict[str, str],
) -> Optional[Dict[str, Any]]:
"""Create nodes with a given resource and label config.
This is the method actually called by the autoscaler. Prefer to
implement this when possible directly, otherwise it delegates to the
create_node() implementation.
Optionally may throw a ray.autoscaler.node_launch_exception.NodeLaunchException.
"""
return self.create_node(node_config, tags, count)
def set_node_tags(self, node_id: str, tags: Dict[str, str]) -> None:
"""Sets the tag values (string dict) for the specified node."""
raise NotImplementedError
def terminate_node(self, node_id: str) -> Optional[Dict[str, Any]]:
"""Terminates the specified node.
Optionally return a mapping from deleted node ids to node
metadata.
"""
raise NotImplementedError
def terminate_nodes(self, node_ids: List[str]) -> Optional[Dict[str, Any]]:
"""Terminates a set of nodes.
May be overridden with a batch method, which optionally may return a
mapping from deleted node ids to node metadata.
"""
for node_id in node_ids:
logger.info("NodeProvider: {}: Terminating node".format(node_id))
self.terminate_node(node_id)
return None
@property
def max_terminate_nodes(self) -> Optional[int]:
"""The maximum number of nodes which can be terminated in one single
API request. By default, this is "None", which means that the node
provider's underlying API allows infinite requests to be terminated
with one request.
For example, AWS only allows 1000 nodes to be terminated
at once; to terminate more, we must issue multiple separate API
requests. If the limit is infinity, then simply set this to None.
This may be overridden. The value may be useful when overriding the
"terminate_nodes" method.
"""
return None
@staticmethod
def bootstrap_config(cluster_config: Dict[str, Any]) -> Dict[str, Any]:
"""Bootstraps the cluster config by adding env defaults if needed."""
return cluster_config
def get_command_runner(
self,
log_prefix: str,
node_id: str,
auth_config: Dict[str, Any],
cluster_name: str,
process_runner: ModuleType,
use_internal_ip: bool,
docker_config: Optional[Dict[str, Any]] = None,
) -> CommandRunnerInterface:
"""Returns the CommandRunner class used to perform SSH commands.
Args:
log_prefix: stores "NodeUpdater: {}: ".format(<node_id>). Used
to print progress in the CommandRunner.
node_id: the node ID.
auth_config: the authentication configs from the autoscaler
yaml file.
cluster_name: the name of the cluster.
process_runner: the module to use to run the commands
in the CommandRunner. E.g., subprocess.
use_internal_ip: whether the node_id belongs to an internal ip
or external ip.
docker_config: If set, the docker information of the docker
container that commands should be run on.
Returns:
A CommandRunner instance for the node.
"""
common_args = {
"log_prefix": log_prefix,
"node_id": node_id,
"provider": self,
"auth_config": auth_config,
"cluster_name": cluster_name,
"process_runner": process_runner,
"use_internal_ip": use_internal_ip,
}
if docker_config and docker_config["container_name"] != "":
return DockerCommandRunner(docker_config, **common_args)
else:
return SSHCommandRunner(**common_args)
def prepare_for_head_node(self, cluster_config: Dict[str, Any]) -> Dict[str, Any]:
"""Returns a new cluster config with custom configs for head node."""
return cluster_config
@staticmethod
def fillout_available_node_types_resources(
cluster_config: Dict[str, Any]
) -> Dict[str, Any]:
"""Fills out missing "resources" field for available_node_types."""
return cluster_config
def safe_to_scale(self) -> bool:
"""Optional condition to determine if it's safe to proceed with an autoscaling
update. Can be used to wait for convergence of state managed by an external
cluster manager.
Called by the autoscaler immediately after non_terminated_nodes().
If False is returned, the autoscaler will abort the update.
"""
return True
def post_process(self) -> None:
"""This optional method is executed at the end of
StandardAutoscaler._update().
"""
pass