mirror of
https://github.com/vale981/ray
synced 2025-03-05 10:01:43 -05:00

Signed-off-by: Alan Guo <aguo@anyscale.com> ## Why are these changes needed? Reduces memory footprint of the dashboard. Also adds some cleanup to the errors data. Also cleans up actor cache by removing dead actors from the cache. Dashboard UI no longer allows you to see logs for all workers in a node. You must click into each worker's logs individually. <img width="1739" alt="Screen Shot 2022-07-20 at 9 13 00 PM" src="https://user-images.githubusercontent.com/711935/180128633-1633c187-39c9-493e-b694-009fbb27f73b.png"> ## Related issue number fixes #23680 fixes #22027 fixes #24272
344 lines
13 KiB
Python
344 lines
13 KiB
Python
import asyncio
|
|
import logging
|
|
import ray.dashboard.consts as dashboard_consts
|
|
import ray.dashboard.memory_utils as memory_utils
|
|
|
|
# TODO(fyrestone): Not import from dashboard module.
|
|
from ray.dashboard.modules.actor.actor_utils import actor_classname_from_task_spec
|
|
from ray.dashboard.utils import Dict, Signal, async_loop_forever
|
|
|
|
logger = logging.getLogger(__name__)
|
|
|
|
|
|
class GlobalSignals:
|
|
node_info_fetched = Signal(dashboard_consts.SIGNAL_NODE_INFO_FETCHED)
|
|
node_summary_fetched = Signal(dashboard_consts.SIGNAL_NODE_SUMMARY_FETCHED)
|
|
job_info_fetched = Signal(dashboard_consts.SIGNAL_JOB_INFO_FETCHED)
|
|
worker_info_fetched = Signal(dashboard_consts.SIGNAL_WORKER_INFO_FETCHED)
|
|
|
|
|
|
class DataSource:
|
|
# {node id hex(str): node stats(dict of GetNodeStatsReply
|
|
# in node_manager.proto)}
|
|
node_stats = Dict()
|
|
# {node id hex(str): node physical stats(dict from reporter_agent.py)}
|
|
node_physical_stats = Dict()
|
|
# {actor id hex(str): actor table data(dict of ActorTableData
|
|
# in gcs.proto)}
|
|
actors = Dict()
|
|
# {job id hex(str): job table data(dict of JobTableData in gcs.proto)}
|
|
jobs = Dict()
|
|
# {node id hex(str): dashboard agent [http port(int), grpc port(int)]}
|
|
agents = Dict()
|
|
# {node id hex(str): gcs node info(dict of GcsNodeInfo in gcs.proto)}
|
|
nodes = Dict()
|
|
# {node id hex(str): ip address(str)}
|
|
node_id_to_ip = Dict()
|
|
# {node id hex(str): hostname(str)}
|
|
node_id_to_hostname = Dict()
|
|
# {node id hex(str): worker list}
|
|
node_workers = Dict()
|
|
# {node id hex(str): {actor id hex(str): actor table data}}
|
|
node_actors = Dict()
|
|
# {job id hex(str): worker list}
|
|
job_workers = Dict()
|
|
# {job id hex(str): {actor id hex(str): actor table data}}
|
|
job_actors = Dict()
|
|
# {worker id(str): core worker stats}
|
|
core_worker_stats = Dict()
|
|
# {job id hex(str): {event id(str): event dict}}
|
|
events = Dict()
|
|
# {node ip (str): log counts by pid
|
|
# (dict from pid to count of logs for that pid)}
|
|
ip_and_pid_to_log_counts = Dict()
|
|
# {node ip (str): error entries by pid
|
|
# (dict from pid to list of latest err entries)}
|
|
ip_and_pid_to_errors = Dict()
|
|
|
|
|
|
class DataOrganizer:
|
|
@staticmethod
|
|
@async_loop_forever(dashboard_consts.PURGE_DATA_INTERVAL_SECONDS)
|
|
async def purge():
|
|
# Purge data that is out of date.
|
|
# These data sources are maintained by DashboardHead,
|
|
# we do not needs to purge them:
|
|
# * agents
|
|
# * nodes
|
|
# * node_id_to_ip
|
|
# * node_id_to_hostname
|
|
logger.info("Purge data.")
|
|
alive_nodes = {
|
|
node_id
|
|
for node_id, node_info in DataSource.nodes.items()
|
|
if node_info["state"] == "ALIVE"
|
|
}
|
|
for key in DataSource.node_stats.keys() - alive_nodes:
|
|
DataSource.node_stats.pop(key)
|
|
|
|
for key in DataSource.node_physical_stats.keys() - alive_nodes:
|
|
DataSource.node_physical_stats.pop(key)
|
|
|
|
@classmethod
|
|
@async_loop_forever(dashboard_consts.ORGANIZE_DATA_INTERVAL_SECONDS)
|
|
async def organize(cls):
|
|
job_workers = {}
|
|
node_workers = {}
|
|
core_worker_stats = {}
|
|
# await inside for loop, so we create a copy of keys().
|
|
for node_id in list(DataSource.nodes.keys()):
|
|
workers = await cls.get_node_workers(node_id)
|
|
for worker in workers:
|
|
job_id = worker["jobId"]
|
|
job_workers.setdefault(job_id, []).append(worker)
|
|
for stats in worker.get("coreWorkerStats", []):
|
|
worker_id = stats["workerId"]
|
|
core_worker_stats[worker_id] = stats
|
|
node_workers[node_id] = workers
|
|
DataSource.job_workers.reset(job_workers)
|
|
DataSource.node_workers.reset(node_workers)
|
|
DataSource.core_worker_stats.reset(core_worker_stats)
|
|
|
|
@classmethod
|
|
async def get_node_workers(cls, node_id):
|
|
workers = []
|
|
node_ip = DataSource.node_id_to_ip[node_id]
|
|
node_log_counts = DataSource.ip_and_pid_to_log_counts.get(node_ip, {})
|
|
node_errs = DataSource.ip_and_pid_to_errors.get(node_ip, {})
|
|
node_physical_stats = DataSource.node_physical_stats.get(node_id, {})
|
|
node_stats = DataSource.node_stats.get(node_id, {})
|
|
# Merge coreWorkerStats (node stats) to workers (node physical stats)
|
|
pid_to_worker_stats = {}
|
|
pid_to_language = {}
|
|
pid_to_job_id = {}
|
|
pids_on_node = set()
|
|
for core_worker_stats in node_stats.get("coreWorkersStats", []):
|
|
pid = core_worker_stats["pid"]
|
|
pids_on_node.add(pid)
|
|
pid_to_worker_stats.setdefault(pid, []).append(core_worker_stats)
|
|
pid_to_language[pid] = core_worker_stats["language"]
|
|
pid_to_job_id[pid] = core_worker_stats["jobId"]
|
|
|
|
# Clean up logs from a dead pid.
|
|
dead_pids = set(node_log_counts.keys()) - pids_on_node
|
|
for dead_pid in dead_pids:
|
|
if dead_pid in node_log_counts:
|
|
node_log_counts.mutable().pop(dead_pid)
|
|
|
|
for worker in node_physical_stats.get("workers", []):
|
|
worker = dict(worker)
|
|
pid = worker["pid"]
|
|
worker["logCount"] = node_log_counts.get(str(pid), 0)
|
|
worker["errorCount"] = len(node_errs.get(str(pid), []))
|
|
worker["coreWorkerStats"] = pid_to_worker_stats.get(pid, [])
|
|
worker["language"] = pid_to_language.get(
|
|
pid, dashboard_consts.DEFAULT_LANGUAGE
|
|
)
|
|
worker["jobId"] = pid_to_job_id.get(pid, dashboard_consts.DEFAULT_JOB_ID)
|
|
|
|
await GlobalSignals.worker_info_fetched.send(node_id, worker)
|
|
|
|
workers.append(worker)
|
|
return workers
|
|
|
|
@classmethod
|
|
async def get_node_info(cls, node_id):
|
|
node_physical_stats = dict(DataSource.node_physical_stats.get(node_id, {}))
|
|
node_stats = dict(DataSource.node_stats.get(node_id, {}))
|
|
node = DataSource.nodes.get(node_id, {})
|
|
node_ip = DataSource.node_id_to_ip.get(node_id)
|
|
# Merge node log count information into the payload
|
|
log_counts = DataSource.ip_and_pid_to_log_counts.get(node_ip, {})
|
|
node_log_count = 0
|
|
for entries in log_counts.values():
|
|
node_log_count += entries
|
|
error_info = DataSource.ip_and_pid_to_errors.get(node_ip, {})
|
|
node_err_count = 0
|
|
for entries in error_info.values():
|
|
node_err_count += len(entries)
|
|
|
|
node_stats.pop("coreWorkersStats", None)
|
|
|
|
view_data = node_stats.get("viewData", [])
|
|
ray_stats = cls._extract_view_data(
|
|
view_data, {"object_store_used_memory", "object_store_available_memory"}
|
|
)
|
|
|
|
node_info = node_physical_stats
|
|
# Merge node stats to node physical stats under raylet
|
|
node_info["raylet"] = node_stats
|
|
node_info["raylet"].update(ray_stats)
|
|
|
|
# Merge GcsNodeInfo to node physical stats
|
|
node_info["raylet"].update(node)
|
|
# Merge actors to node physical stats
|
|
node_info["actors"] = DataSource.node_actors.get(node_id, {})
|
|
# Update workers to node physical stats
|
|
node_info["workers"] = DataSource.node_workers.get(node_id, [])
|
|
node_info["logCount"] = node_log_count
|
|
node_info["errorCount"] = node_err_count
|
|
await GlobalSignals.node_info_fetched.send(node_info)
|
|
|
|
return node_info
|
|
|
|
@classmethod
|
|
async def get_node_summary(cls, node_id):
|
|
node_physical_stats = dict(DataSource.node_physical_stats.get(node_id, {}))
|
|
node_stats = dict(DataSource.node_stats.get(node_id, {}))
|
|
node = DataSource.nodes.get(node_id, {})
|
|
|
|
node_physical_stats.pop("workers", None)
|
|
node_stats.pop("workersStats", None)
|
|
view_data = node_stats.get("viewData", [])
|
|
ray_stats = cls._extract_view_data(
|
|
view_data, {"object_store_used_memory", "object_store_available_memory"}
|
|
)
|
|
node_stats.pop("viewData", None)
|
|
|
|
node_summary = node_physical_stats
|
|
# Merge node stats to node physical stats
|
|
node_summary["raylet"] = node_stats
|
|
node_summary["raylet"].update(ray_stats)
|
|
# Merge GcsNodeInfo to node physical stats
|
|
node_summary["raylet"].update(node)
|
|
|
|
await GlobalSignals.node_summary_fetched.send(node_summary)
|
|
|
|
return node_summary
|
|
|
|
@classmethod
|
|
async def get_all_node_summary(cls):
|
|
return [
|
|
await DataOrganizer.get_node_summary(node_id)
|
|
for node_id in DataSource.nodes.keys()
|
|
]
|
|
|
|
@classmethod
|
|
async def get_all_node_details(cls):
|
|
return [
|
|
await DataOrganizer.get_node_info(node_id)
|
|
for node_id in DataSource.nodes.keys()
|
|
]
|
|
|
|
@classmethod
|
|
async def get_all_actors(cls):
|
|
result = {}
|
|
for index, (actor_id, actor) in enumerate(DataSource.actors.items()):
|
|
result[actor_id] = await cls._get_actor(actor)
|
|
# There can be thousands of actors including dead ones. Processing
|
|
# them all can take many seconds, which blocks all other requests
|
|
# to the dashboard. The ideal solution might be to implement
|
|
# pagination. For now, use a workaround to yield to the event loop
|
|
# periodically, so other request handlers have a chance to run and
|
|
# avoid long latencies.
|
|
if index % 1000 == 0 and index > 0:
|
|
# Canonical way to yield to the event loop:
|
|
# https://github.com/python/asyncio/issues/284
|
|
await asyncio.sleep(0)
|
|
return result
|
|
|
|
@staticmethod
|
|
async def _get_actor(actor):
|
|
actor = dict(actor)
|
|
worker_id = actor["address"]["workerId"]
|
|
core_worker_stats = DataSource.core_worker_stats.get(worker_id, {})
|
|
actor_constructor = core_worker_stats.get(
|
|
"actorTitle", "Unknown actor constructor"
|
|
)
|
|
actor["actorConstructor"] = actor_constructor
|
|
actor.update(core_worker_stats)
|
|
|
|
# TODO(fyrestone): remove this, give a link from actor
|
|
# info to worker info in front-end.
|
|
node_id = actor["address"]["rayletId"]
|
|
pid = core_worker_stats.get("pid")
|
|
node_physical_stats = DataSource.node_physical_stats.get(node_id, {})
|
|
actor_process_stats = None
|
|
actor_process_gpu_stats = []
|
|
if pid:
|
|
for process_stats in node_physical_stats.get("workers", []):
|
|
if process_stats["pid"] == pid:
|
|
actor_process_stats = process_stats
|
|
break
|
|
|
|
for gpu_stats in node_physical_stats.get("gpus", []):
|
|
# gpu_stats.get("processes") can be None, an empty list or a
|
|
# list of dictionaries.
|
|
for process in gpu_stats.get("processes") or []:
|
|
if process["pid"] == pid:
|
|
actor_process_gpu_stats.append(gpu_stats)
|
|
break
|
|
|
|
actor["gpus"] = actor_process_gpu_stats
|
|
actor["processStats"] = actor_process_stats
|
|
return actor
|
|
|
|
@classmethod
|
|
async def get_actor_creation_tasks(cls):
|
|
infeasible_tasks = sum(
|
|
(
|
|
list(node_stats.get("infeasibleTasks", []))
|
|
for node_stats in DataSource.node_stats.values()
|
|
),
|
|
[],
|
|
)
|
|
new_infeasible_tasks = []
|
|
for task in infeasible_tasks:
|
|
task = dict(task)
|
|
task["actorClass"] = actor_classname_from_task_spec(task)
|
|
task["state"] = "INFEASIBLE"
|
|
new_infeasible_tasks.append(task)
|
|
|
|
resource_pending_tasks = sum(
|
|
(
|
|
list(data.get("readyTasks", []))
|
|
for data in DataSource.node_stats.values()
|
|
),
|
|
[],
|
|
)
|
|
new_resource_pending_tasks = []
|
|
for task in resource_pending_tasks:
|
|
task = dict(task)
|
|
task["actorClass"] = actor_classname_from_task_spec(task)
|
|
task["state"] = "PENDING_RESOURCES"
|
|
new_resource_pending_tasks.append(task)
|
|
|
|
results = {
|
|
task["actorCreationTaskSpec"]["actorId"]: task
|
|
for task in new_resource_pending_tasks + new_infeasible_tasks
|
|
}
|
|
return results
|
|
|
|
@classmethod
|
|
async def get_memory_table(
|
|
cls,
|
|
sort_by=memory_utils.SortingType.OBJECT_SIZE,
|
|
group_by=memory_utils.GroupByType.STACK_TRACE,
|
|
):
|
|
all_worker_stats = []
|
|
for node_stats in DataSource.node_stats.values():
|
|
all_worker_stats.extend(node_stats.get("coreWorkersStats", []))
|
|
memory_information = memory_utils.construct_memory_table(
|
|
all_worker_stats, group_by=group_by, sort_by=sort_by
|
|
)
|
|
return memory_information
|
|
|
|
@staticmethod
|
|
def _extract_view_data(views, data_keys):
|
|
view_data = {}
|
|
for view in views:
|
|
view_name = view["viewName"]
|
|
if view_name in data_keys:
|
|
if not view.get("measures"):
|
|
view_data[view_name] = 0
|
|
continue
|
|
measure = view["measures"][0]
|
|
if "doubleValue" in measure:
|
|
measure_value = measure["doubleValue"]
|
|
elif "intValue" in measure:
|
|
measure_value = measure["intValue"]
|
|
else:
|
|
measure_value = 0
|
|
view_data[view_name] = measure_value
|
|
|
|
return view_data
|