import asyncio import heapq import inspect import itertools import json import logging import math import operator import os import random import sys import uuid import warnings import weakref from collections import defaultdict, deque from collections.abc import ( Callable, Collection, Hashable, Iterable, Iterator, Mapping, Set, ) from contextlib import suppress from datetime import timedelta from functools import partial from numbers import Number from typing import Any, ClassVar, Container from typing import cast as pep484_cast import psutil from sortedcontainers import SortedDict, SortedSet from tlz import ( compose, first, groupby, merge, merge_sorted, merge_with, pluck, second, valmap, ) from tornado.ioloop import IOLoop, PeriodicCallback import dask from dask.highlevelgraph import HighLevelGraph from dask.utils import format_bytes, format_time, parse_bytes, parse_timedelta, tmpfile from dask.widgets import get_template from distributed.utils import recursive_to_dict from . import preloading, profile from . import versions as version_module from .active_memory_manager import ActiveMemoryManagerExtension from .batched import BatchedSend from .comm import ( Comm, get_address_host, normalize_address, resolve_address, unparse_host_port, ) from .comm.addressing import addresses_from_user_args from .core import CommClosedError, Status, clean_exception, rpc, send_recv from .diagnostics.memory_sampler import MemorySamplerExtension from .diagnostics.plugin import SchedulerPlugin, _get_plugin_name from .event import EventExtension from .http import get_handlers from .lock import LockExtension from .metrics import time from .multi_lock import MultiLockExtension from .node import ServerNode from .proctitle import setproctitle from .protocol.pickle import loads from .publish import PublishExtension from .pubsub import PubSubSchedulerExtension from .queues import QueueExtension from .recreate_tasks import ReplayTaskScheduler from .security import Security from .semaphore import SemaphoreExtension from .stealing import WorkStealing from .utils import ( All, TimeoutError, empty_context, get_fileno_limit, key_split, key_split_group, log_errors, no_default, validate_key, ) from .utils_comm import gather_from_workers, retry_operation, scatter_to_workers from .utils_perf import disable_gc_diagnosis, enable_gc_diagnosis from .variable import VariableExtension try: from cython import compiled except ImportError: compiled = False if compiled: from cython import ( Py_hash_t, Py_ssize_t, bint, cast, ccall, cclass, cfunc, declare, double, exceptval, final, inline, nogil, ) else: from ctypes import c_double as double from ctypes import c_ssize_t as Py_hash_t from ctypes import c_ssize_t as Py_ssize_t bint = bool def cast(T, v, *a, **k): return v def ccall(func): return func def cclass(cls): return cls def cfunc(func): return func def declare(*a, **k): if len(a) == 2: return a[1] else: pass def exceptval(*a, **k): def wrapper(func): return func return wrapper def final(cls): return cls def inline(func): return func def nogil(func): return func if sys.version_info < (3, 8): try: import pickle5 as pickle except ImportError: import pickle else: import pickle logger = logging.getLogger(__name__) LOG_PDB = dask.config.get("distributed.admin.pdb-on-err") DEFAULT_DATA_SIZE = declare( Py_ssize_t, parse_bytes(dask.config.get("distributed.scheduler.default-data-size")) ) DEFAULT_EXTENSIONS = [ LockExtension, MultiLockExtension, PublishExtension, ReplayTaskScheduler, QueueExtension, VariableExtension, PubSubSchedulerExtension, SemaphoreExtension, EventExtension, ActiveMemoryManagerExtension, MemorySamplerExtension, ] ALL_TASK_STATES = declare( set, {"released", "waiting", "no-worker", "processing", "erred", "memory"} ) globals()["ALL_TASK_STATES"] = ALL_TASK_STATES COMPILED = declare(bint, compiled) globals()["COMPILED"] = COMPILED @final @cclass class ClientState: """ A simple object holding information about a client. .. attribute:: client_key: str A unique identifier for this client. This is generally an opaque string generated by the client itself. .. attribute:: wants_what: {TaskState} A set of tasks this client wants kept in memory, so that it can download its result when desired. This is the reverse mapping of :class:`TaskState.who_wants`. Tasks are typically removed from this set when the corresponding object in the client's space (for example a ``Future`` or a Dask collection) gets garbage-collected. """ _client_key: str _hash: Py_hash_t _wants_what: set _last_seen: double _versions: dict __slots__ = ("_client_key", "_hash", "_wants_what", "_last_seen", "_versions") def __init__(self, client: str, versions: dict = None): self._client_key = client self._hash = hash(client) self._wants_what = set() self._last_seen = time() self._versions = versions or {} def __hash__(self): return self._hash def __eq__(self, other): typ_self: type = type(self) typ_other: type = type(other) if typ_self == typ_other: other_cs: ClientState = other return self._client_key == other_cs._client_key else: return False def __repr__(self): return "" % self._client_key def __str__(self): return self._client_key @property def client_key(self): return self._client_key @property def wants_what(self): return self._wants_what @property def last_seen(self): return self._last_seen @property def versions(self): return self._versions @final @cclass class MemoryState: """Memory readings on a worker or on the whole cluster. managed Sum of the output of sizeof() for all dask keys held by the worker, both in memory and spilled to disk managed_in_memory Sum of the output of sizeof() for the dask keys held in RAM managed_spilled Sum of the output of sizeof() for the dask keys spilled to the hard drive. Note that this is the size in memory; serialized size may be different. process Total RSS memory measured by the OS on the worker process. This is always exactly equal to managed_in_memory + unmanaged. unmanaged process - managed_in_memory. This is the sum of - Python interpreter and modules - global variables - memory temporarily allocated by the dask tasks that are currently running - memory fragmentation - memory leaks - memory not yet garbage collected - memory not yet free()'d by the Python memory manager to the OS unmanaged_old Minimum of the 'unmanaged' measures over the last ``distributed.memory.recent-to-old-time`` seconds unmanaged_recent unmanaged - unmanaged_old; in other words process memory that has been recently allocated but is not accounted for by dask; hopefully it's mostly a temporary spike. optimistic managed_in_memory + unmanaged_old; in other words the memory held long-term by the process under the hopeful assumption that all unmanaged_recent memory is a temporary spike """ __slots__ = ("_process", "_managed_in_memory", "_managed_spilled", "_unmanaged_old") _process: Py_ssize_t _managed_in_memory: Py_ssize_t _managed_spilled: Py_ssize_t _unmanaged_old: Py_ssize_t def __init__( self, *, process: Py_ssize_t, unmanaged_old: Py_ssize_t, managed: Py_ssize_t, managed_spilled: Py_ssize_t, ): # Some data arrives with the heartbeat, some other arrives in realtime as the # tasks progress. Also, sizeof() is not guaranteed to return correct results. # This can cause glitches where a partial measure is larger than the whole, so # we need to force all numbers to add up exactly by definition. self._process = process self._managed_spilled = min(managed_spilled, managed) # Subtractions between unsigned ints guaranteed by construction to be >= 0 self._managed_in_memory = min(managed - self._managed_spilled, process) self._unmanaged_old = min(unmanaged_old, process - self._managed_in_memory) @property def process(self) -> Py_ssize_t: return self._process @property def managed_in_memory(self) -> Py_ssize_t: return self._managed_in_memory @property def managed_spilled(self) -> Py_ssize_t: return self._managed_spilled @property def unmanaged_old(self) -> Py_ssize_t: return self._unmanaged_old @classmethod def sum(cls, *infos: "MemoryState") -> "MemoryState": out = MemoryState(process=0, unmanaged_old=0, managed=0, managed_spilled=0) ms: MemoryState for ms in infos: out._process += ms._process out._managed_spilled += ms._managed_spilled out._managed_in_memory += ms._managed_in_memory out._unmanaged_old += ms._unmanaged_old return out @property def managed(self) -> Py_ssize_t: return self._managed_in_memory + self._managed_spilled @property def unmanaged(self) -> Py_ssize_t: # This is never negative thanks to __init__ return self._process - self._managed_in_memory @property def unmanaged_recent(self) -> Py_ssize_t: # This is never negative thanks to __init__ return self._process - self._managed_in_memory - self._unmanaged_old @property def optimistic(self) -> Py_ssize_t: return self._managed_in_memory + self._unmanaged_old def __repr__(self) -> str: return ( f"Process memory (RSS) : {format_bytes(self._process)}\n" f" - managed by Dask : {format_bytes(self._managed_in_memory)}\n" f" - unmanaged (old) : {format_bytes(self._unmanaged_old)}\n" f" - unmanaged (recent): {format_bytes(self.unmanaged_recent)}\n" f"Spilled to disk : {format_bytes(self._managed_spilled)}\n" ) @final @cclass class WorkerState: """ A simple object holding information about a worker. .. attribute:: address: str This worker's unique key. This can be its connected address (such as ``'tcp://127.0.0.1:8891'``) or an alias (such as ``'alice'``). .. attribute:: processing: {TaskState: cost} A dictionary of tasks that have been submitted to this worker. Each task state is associated with the expected cost in seconds of running that task, summing both the task's expected computation time and the expected communication time of its result. If a task is already executing on the worker and the excecution time is twice the learned average TaskGroup duration, this will be set to twice the current executing time. If the task is unknown, the default task duration is used instead of the TaskGroup average. Multiple tasks may be submitted to a worker in advance and the worker will run them eventually, depending on its execution resources (but see :doc:`work-stealing`). All the tasks here are in the "processing" state. This attribute is kept in sync with :attr:`TaskState.processing_on`. .. attribute:: executing: {TaskState: duration} A dictionary of tasks that are currently being run on this worker. Each task state is asssociated with the duration in seconds which the task has been running. .. attribute:: has_what: {TaskState} An insertion-sorted set-like of tasks which currently reside on this worker. All the tasks here are in the "memory" state. This is the reverse mapping of :class:`TaskState.who_has`. .. attribute:: nbytes: int The total memory size, in bytes, used by the tasks this worker holds in memory (i.e. the tasks in this worker's :attr:`has_what`). .. attribute:: nthreads: int The number of CPU threads made available on this worker. .. attribute:: resources: {str: Number} The available resources on this worker like ``{'gpu': 2}``. These are abstract quantities that constrain certain tasks from running at the same time on this worker. .. attribute:: used_resources: {str: Number} The sum of each resource used by all tasks allocated to this worker. The numbers in this dictionary can only be less or equal than those in this worker's :attr:`resources`. .. attribute:: occupancy: double The total expected runtime, in seconds, of all tasks currently processing on this worker. This is the sum of all the costs in this worker's :attr:`processing` dictionary. .. attribute:: status: Status Read-only worker status, synced one way from the remote Worker object .. attribute:: nanny: str Address of the associated Nanny, if present .. attribute:: last_seen: Py_ssize_t The last time we received a heartbeat from this worker, in local scheduler time. .. attribute:: actors: {TaskState} A set of all TaskStates on this worker that are actors. This only includes those actors whose state actually lives on this worker, not actors to which this worker has a reference. """ # XXX need a state field to signal active/removed? _actors: set _address: str _bandwidth: double _executing: dict _extra: dict # _has_what is a dict with all values set to None as rebalance() relies on the # property of Python >=3.7 dicts to be insertion-sorted. _has_what: dict _hash: Py_hash_t _last_seen: double _local_directory: str _memory_limit: Py_ssize_t _memory_other_history: "deque[tuple[float, Py_ssize_t]]" _memory_unmanaged_old: Py_ssize_t _metrics: dict _name: object _nanny: str _nbytes: Py_ssize_t _nthreads: Py_ssize_t _occupancy: double _pid: Py_ssize_t _processing: dict _resources: dict _services: dict _status: Status _time_delay: double _used_resources: dict _versions: dict __slots__ = ( "_actors", "_address", "_bandwidth", "_extra", "_executing", "_has_what", "_hash", "_last_seen", "_local_directory", "_memory_limit", "_memory_other_history", "_memory_unmanaged_old", "_metrics", "_name", "_nanny", "_nbytes", "_nthreads", "_occupancy", "_pid", "_processing", "_resources", "_services", "_status", "_time_delay", "_used_resources", "_versions", ) def __init__( self, *, address: str, status: Status, pid: Py_ssize_t, name: object, nthreads: Py_ssize_t = 0, memory_limit: Py_ssize_t, local_directory: str, nanny: str, services: "dict | None" = None, versions: "dict | None" = None, extra: "dict | None" = None, ): self._address = address self._pid = pid self._name = name self._nthreads = nthreads self._memory_limit = memory_limit self._local_directory = local_directory self._services = services or {} self._versions = versions or {} self._nanny = nanny self._status = status self._hash = hash(address) self._nbytes = 0 self._occupancy = 0 self._memory_unmanaged_old = 0 self._memory_other_history = deque() self._metrics = {} self._last_seen = 0 self._time_delay = 0 self._bandwidth = float( parse_bytes(dask.config.get("distributed.scheduler.bandwidth")) ) self._actors = set() self._has_what = {} self._processing = {} self._executing = {} self._resources = {} self._used_resources = {} self._extra = extra or {} def __hash__(self): return self._hash def __eq__(self, other): typ_self: type = type(self) typ_other: type = type(other) if typ_self == typ_other: other_ws: WorkerState = other return self._address == other_ws._address else: return False @property def actors(self): return self._actors @property def address(self) -> str: return self._address @property def bandwidth(self): return self._bandwidth @property def executing(self): return self._executing @property def extra(self): return self._extra @property def has_what(self) -> "Set[TaskState]": return self._has_what.keys() @property def host(self): return get_address_host(self._address) @property def last_seen(self): return self._last_seen @property def local_directory(self): return self._local_directory @property def memory_limit(self): return self._memory_limit @property def metrics(self): return self._metrics @property def memory(self) -> MemoryState: return MemoryState( # metrics["memory"] is None if the worker sent a heartbeat before its # SystemMonitor ever had a chance to run process=self._metrics["memory"] or 0, managed=self._nbytes, managed_spilled=self._metrics["spilled_nbytes"], unmanaged_old=self._memory_unmanaged_old, ) @property def name(self): return self._name @property def nanny(self): return self._nanny @property def nbytes(self): return self._nbytes @nbytes.setter def nbytes(self, v: Py_ssize_t): self._nbytes = v @property def nthreads(self): return self._nthreads @property def occupancy(self): return self._occupancy @occupancy.setter def occupancy(self, v: double): self._occupancy = v @property def pid(self): return self._pid @property def processing(self): return self._processing @property def resources(self): return self._resources @property def services(self): return self._services @property def status(self): return self._status @status.setter def status(self, new_status): if not isinstance(new_status, Status): raise TypeError(f"Expected Status; got {new_status!r}") self._status = new_status @property def time_delay(self): return self._time_delay @property def used_resources(self): return self._used_resources @property def versions(self): return self._versions @ccall def clean(self): """Return a version of this object that is appropriate for serialization""" ws: WorkerState = WorkerState( address=self._address, status=self._status, pid=self._pid, name=self._name, nthreads=self._nthreads, memory_limit=self._memory_limit, local_directory=self._local_directory, services=self._services, nanny=self._nanny, extra=self._extra, ) ts: TaskState ws._processing = {ts._key: cost for ts, cost in self._processing.items()} ws._executing = {ts._key: duration for ts, duration in self._executing.items()} return ws def __repr__(self): return "" % ( self._address, self._name, self._status.name, len(self._has_what), len(self._processing), ) def _repr_html_(self): return get_template("worker_state.html.j2").render( address=self.address, name=self.name, status=self.status.name, has_what=self._has_what, processing=self.processing, ) @ccall @exceptval(check=False) def identity(self) -> dict: return { "type": "Worker", "id": self._name, "host": self.host, "resources": self._resources, "local_directory": self._local_directory, "name": self._name, "nthreads": self._nthreads, "memory_limit": self._memory_limit, "last_seen": self._last_seen, "services": self._services, "metrics": self._metrics, "nanny": self._nanny, **self._extra, } @property def ncores(self): warnings.warn("WorkerState.ncores has moved to WorkerState.nthreads") return self._nthreads @final @cclass class Computation: """ Collection tracking a single compute or persist call See also -------- TaskPrefix TaskGroup TaskState """ _start: double _groups: set _code: object _id: object def __init__(self): self._start = time() self._groups = set() self._code = SortedSet() self._id = uuid.uuid4() @property def code(self): return self._code @property def start(self): return self._start @property def stop(self): if self.groups: return max(tg.stop for tg in self.groups) else: return -1 @property def states(self): tg: TaskGroup return merge_with(sum, [tg._states for tg in self._groups]) @property def groups(self): return self._groups def __repr__(self): return ( f"" ) def _repr_html_(self): return get_template("computation.html.j2").render( id=self._id, start=self.start, stop=self.stop, groups=self.groups, states=self.states, code=self.code, ) @final @cclass class TaskPrefix: """Collection tracking all tasks within a group Keys often have a structure like ``("x-123", 0)`` A group takes the first section, like ``"x"`` .. attribute:: name: str The name of a group of tasks. For a task like ``("x-123", 0)`` this is the text ``"x"`` .. attribute:: states: Dict[str, int] The number of tasks in each state, like ``{"memory": 10, "processing": 3, "released": 4, ...}`` .. attribute:: duration_average: float An exponentially weighted moving average duration of all tasks with this prefix .. attribute:: suspicious: int Numbers of times a task was marked as suspicious with this prefix See Also -------- TaskGroup """ _name: str _all_durations: "defaultdict[str, float]" _duration_average: double _suspicious: Py_ssize_t _groups: list def __init__(self, name: str): self._name = name self._groups = [] # store timings for each prefix-action self._all_durations = defaultdict(float) task_durations = dask.config.get("distributed.scheduler.default-task-durations") if self._name in task_durations: self._duration_average = parse_timedelta(task_durations[self._name]) else: self._duration_average = -1 self._suspicious = 0 @property def name(self) -> str: return self._name @property def all_durations(self) -> "defaultdict[str, float]": return self._all_durations @ccall @exceptval(check=False) def add_duration(self, action: str, start: double, stop: double): duration = stop - start self._all_durations[action] += duration if action == "compute": old = self._duration_average if old < 0: self._duration_average = duration else: self._duration_average = 0.5 * duration + 0.5 * old @property def duration_average(self) -> double: return self._duration_average @property def suspicious(self) -> Py_ssize_t: return self._suspicious @property def groups(self): return self._groups @property def states(self): tg: TaskGroup return merge_with(sum, [tg._states for tg in self._groups]) @property def active(self) -> "list[TaskGroup]": tg: TaskGroup return [ tg for tg in self._groups if any([v != 0 for k, v in tg._states.items() if k != "forgotten"]) ] @property def active_states(self): tg: TaskGroup return merge_with(sum, [tg._states for tg in self.active]) def __repr__(self): return ( "<" + self._name + ": " + ", ".join( "%s: %d" % (k, v) for (k, v) in sorted(self.states.items()) if v ) + ">" ) @property def nbytes_total(self): tg: TaskGroup return sum([tg._nbytes_total for tg in self._groups]) def __len__(self): return sum(map(len, self._groups)) @property def duration(self): tg: TaskGroup return sum([tg._duration for tg in self._groups]) @property def types(self): tg: TaskGroup return set().union(*[tg._types for tg in self._groups]) @final @cclass class TaskGroup: """Collection tracking all tasks within a group Keys often have a structure like ``("x-123", 0)`` A group takes the first section, like ``"x-123"`` .. attribute:: name: str The name of a group of tasks. For a task like ``("x-123", 0)`` this is the text ``"x-123"`` .. attribute:: states: Dict[str, int] The number of tasks in each state, like ``{"memory": 10, "processing": 3, "released": 4, ...}`` .. attribute:: dependencies: Set[TaskGroup] The other TaskGroups on which this one depends .. attribute:: nbytes_total: int The total number of bytes that this task group has produced .. attribute:: duration: float The total amount of time spent on all tasks in this TaskGroup .. attribute:: types: Set[str] The result types of this TaskGroup .. attribute:: last_worker: WorkerState The worker most recently assigned a task from this group, or None when the group is not identified to be root-like by `SchedulerState.decide_worker`. .. attribute:: last_worker_tasks_left: int If `last_worker` is not None, the number of times that worker should be assigned subsequent tasks until a new worker is chosen. See also -------- TaskPrefix """ _name: str _prefix: TaskPrefix # TaskPrefix | None _states: dict _dependencies: set _nbytes_total: Py_ssize_t _duration: double _types: set _start: double _stop: double _all_durations: "defaultdict[str, float]" _last_worker: WorkerState # WorkerState | None _last_worker_tasks_left: Py_ssize_t def __init__(self, name: str): self._name = name self._prefix = None # type: ignore self._states = {state: 0 for state in ALL_TASK_STATES} self._states["forgotten"] = 0 self._dependencies = set() self._nbytes_total = 0 self._duration = 0 self._types = set() self._start = 0.0 self._stop = 0.0 self._all_durations = defaultdict(float) self._last_worker = None # type: ignore self._last_worker_tasks_left = 0 @property def name(self) -> str: return self._name @property def prefix(self) -> "TaskPrefix | None": return self._prefix @property def states(self) -> dict: return self._states @property def dependencies(self) -> set: return self._dependencies @property def nbytes_total(self): return self._nbytes_total @property def duration(self) -> double: return self._duration @ccall @exceptval(check=False) def add_duration(self, action: str, start: double, stop: double): duration = stop - start self._all_durations[action] += duration if action == "compute": if self._stop < stop: self._stop = stop self._start = self._start or start self._duration += duration self._prefix.add_duration(action, start, stop) @property def types(self) -> set: return self._types @property def all_durations(self) -> "defaultdict[str, float]": return self._all_durations @property def start(self) -> double: return self._start @property def stop(self) -> double: return self._stop @property def last_worker(self) -> "WorkerState | None": return self._last_worker @property def last_worker_tasks_left(self) -> int: return self._last_worker_tasks_left @ccall def add(self, other: "TaskState"): self._states[other._state] += 1 other._group = self def __repr__(self): return ( "<" + (self._name or "no-group") + ": " + ", ".join( "%s: %d" % (k, v) for (k, v) in sorted(self._states.items()) if v ) + ">" ) def __len__(self): return sum(self._states.values()) @final @cclass class TaskState: """ A simple object holding information about a task. .. attribute:: key: str The key is the unique identifier of a task, generally formed from the name of the function, followed by a hash of the function and arguments, like ``'inc-ab31c010444977004d656610d2d421ec'``. .. attribute:: prefix: TaskPrefix The broad class of tasks to which this task belongs like "inc" or "read_csv" .. attribute:: run_spec: object A specification of how to run the task. The type and meaning of this value is opaque to the scheduler, as it is only interpreted by the worker to which the task is sent for executing. As a special case, this attribute may also be ``None``, in which case the task is "pure data" (such as, for example, a piece of data loaded in the scheduler using :meth:`Client.scatter`). A "pure data" task cannot be computed again if its value is lost. .. attribute:: priority: tuple The priority provides each task with a relative ranking which is used to break ties when many tasks are being considered for execution. This ranking is generally a 2-item tuple. The first (and dominant) item corresponds to when it was submitted. Generally, earlier tasks take precedence. The second item is determined by the client, and is a way to prioritize tasks within a large graph that may be important, such as if they are on the critical path, or good to run in order to release many dependencies. This is explained further in :doc:`Scheduling Policy `. .. attribute:: state: str This task's current state. Valid states include ``released``, ``waiting``, ``no-worker``, ``processing``, ``memory``, ``erred`` and ``forgotten``. If it is ``forgotten``, the task isn't stored in the ``tasks`` dictionary anymore and will probably disappear soon from memory. .. attribute:: dependencies: {TaskState} The set of tasks this task depends on for proper execution. Only tasks still alive are listed in this set. If, for whatever reason, this task also depends on a forgotten task, the :attr:`has_lost_dependencies` flag is set. A task can only be executed once all its dependencies have already been successfully executed and have their result stored on at least one worker. This is tracked by progressively draining the :attr:`waiting_on` set. .. attribute:: dependents: {TaskState} The set of tasks which depend on this task. Only tasks still alive are listed in this set. This is the reverse mapping of :attr:`dependencies`. .. attribute:: has_lost_dependencies: bool Whether any of the dependencies of this task has been forgotten. For memory consumption reasons, forgotten tasks are not kept in memory even though they may have dependent tasks. When a task is forgotten, therefore, each of its dependents has their :attr:`has_lost_dependencies` attribute set to ``True``. If :attr:`has_lost_dependencies` is true, this task cannot go into the "processing" state anymore. .. attribute:: waiting_on: {TaskState} The set of tasks this task is waiting on *before* it can be executed. This is always a subset of :attr:`dependencies`. Each time one of the dependencies has finished processing, it is removed from the :attr:`waiting_on` set. Once :attr:`waiting_on` becomes empty, this task can move from the "waiting" state to the "processing" state (unless one of the dependencies errored out, in which case this task is instead marked "erred"). .. attribute:: waiters: {TaskState} The set of tasks which need this task to remain alive. This is always a subset of :attr:`dependents`. Each time one of the dependents has finished processing, it is removed from the :attr:`waiters` set. Once both :attr:`waiters` and :attr:`who_wants` become empty, this task can be released (if it has a non-empty :attr:`run_spec`) or forgotten (otherwise) by the scheduler, and by any workers in :attr:`who_has`. .. note:: Counter-intuitively, :attr:`waiting_on` and :attr:`waiters` are not reverse mappings of each other. .. attribute:: who_wants: {ClientState} The set of clients who want this task's result to remain alive. This is the reverse mapping of :attr:`ClientState.wants_what`. When a client submits a graph to the scheduler it also specifies which output tasks it desires, such that their results are not released from memory. Once a task has finished executing (i.e. moves into the "memory" or "erred" state), the clients in :attr:`who_wants` are notified. Once both :attr:`waiters` and :attr:`who_wants` become empty, this task can be released (if it has a non-empty :attr:`run_spec`) or forgotten (otherwise) by the scheduler, and by any workers in :attr:`who_has`. .. attribute:: who_has: {WorkerState} The set of workers who have this task's result in memory. It is non-empty iff the task is in the "memory" state. There can be more than one worker in this set if, for example, :meth:`Client.scatter` or :meth:`Client.replicate` was used. This is the reverse mapping of :attr:`WorkerState.has_what`. .. attribute:: processing_on: WorkerState (or None) If this task is in the "processing" state, which worker is currently processing it. Otherwise this is ``None``. This attribute is kept in sync with :attr:`WorkerState.processing`. .. attribute:: retries: int The number of times this task can automatically be retried in case of failure. If a task fails executing (the worker returns with an error), its :attr:`retries` attribute is checked. If it is equal to 0, the task is marked "erred". If it is greater than 0, the :attr:`retries` attribute is decremented and execution is attempted again. .. attribute:: nbytes: int (or None) The number of bytes, as determined by ``sizeof``, of the result of a finished task. This number is used for diagnostics and to help prioritize work. .. attribute:: type: str The type of the object as a string. Only present for tasks that have been computed. .. attribute:: exception: object If this task failed executing, the exception object is stored here. Otherwise this is ``None``. .. attribute:: traceback: object If this task failed executing, the traceback object is stored here. Otherwise this is ``None``. .. attribute:: exception_blame: TaskState (or None) If this task or one of its dependencies failed executing, the failed task is stored here (possibly itself). Otherwise this is ``None``. .. attribute:: erred_on: set(str) Worker addresses on which errors appeared causing this task to be in an error state. .. attribute:: suspicious: int The number of times this task has been involved in a worker death. Some tasks may cause workers to die (such as calling ``os._exit(0)``). When a worker dies, all of the tasks on that worker are reassigned to others. This combination of behaviors can cause a bad task to catastrophically destroy all workers on the cluster, one after another. Whenever a worker dies, we mark each task currently processing on that worker (as recorded by :attr:`WorkerState.processing`) as suspicious. If a task is involved in three deaths (or some other fixed constant) then we mark the task as ``erred``. .. attribute:: host_restrictions: {hostnames} A set of hostnames where this task can be run (or ``None`` if empty). Usually this is empty unless the task has been specifically restricted to only run on certain hosts. A hostname may correspond to one or several connected workers. .. attribute:: worker_restrictions: {worker addresses} A set of complete worker addresses where this can be run (or ``None`` if empty). Usually this is empty unless the task has been specifically restricted to only run on certain workers. Note this is tracking worker addresses, not worker states, since the specific workers may not be connected at this time. .. attribute:: resource_restrictions: {resource: quantity} Resources required by this task, such as ``{'gpu': 1}`` or ``{'memory': 1e9}`` (or ``None`` if empty). These are user-defined names and are matched against the contents of each :attr:`WorkerState.resources` dictionary. .. attribute:: loose_restrictions: bool If ``False``, each of :attr:`host_restrictions`, :attr:`worker_restrictions` and :attr:`resource_restrictions` is a hard constraint: if no worker is available satisfying those restrictions, the task cannot go into the "processing" state and will instead go into the "no-worker" state. If ``True``, the above restrictions are mere preferences: if no worker is available satisfying those restrictions, the task can still go into the "processing" state and be sent for execution to another connected worker. .. attribute:: metadata: dict Metadata related to task. .. attribute:: actor: bool Whether or not this task is an Actor. .. attribute:: group: TaskGroup The group of tasks to which this one belongs. .. attribute:: annotations: dict Task annotations """ _key: str _hash: Py_hash_t _prefix: TaskPrefix _run_spec: object _priority: tuple # tuple | None _state: str # str | None _dependencies: set # set[TaskState] _dependents: set # set[TaskState] _has_lost_dependencies: bint _waiting_on: set # set[TaskState] _waiters: set # set[TaskState] _who_wants: set # set[ClientState] _who_has: set # set[WorkerState] _processing_on: WorkerState # WorkerState | None _retries: Py_ssize_t _nbytes: Py_ssize_t _type: str # str | None _exception: object _exception_text: str _traceback: object _traceback_text: str _exception_blame: "TaskState" # TaskState | None" _erred_on: set _suspicious: Py_ssize_t _host_restrictions: set # set[str] | None _worker_restrictions: set # set[str] | None _resource_restrictions: dict # dict | None _loose_restrictions: bint _metadata: dict _annotations: dict _actor: bint _group: TaskGroup # TaskGroup | None _group_key: str __slots__ = ( # === General description === "_actor", # Key name "_key", # Hash of the key name "_hash", # Key prefix (see key_split()) "_prefix", # How to run the task (None if pure data) "_run_spec", # Alive dependents and dependencies "_dependencies", "_dependents", # Compute priority "_priority", # Restrictions "_host_restrictions", "_worker_restrictions", # not WorkerStates but addresses "_resource_restrictions", "_loose_restrictions", # === Task state === "_state", # Whether some dependencies were forgotten "_has_lost_dependencies", # If in 'waiting' state, which tasks need to complete # before we can run "_waiting_on", # If in 'waiting' or 'processing' state, which tasks needs us # to complete before they can run "_waiters", # In in 'processing' state, which worker we are processing on "_processing_on", # If in 'memory' state, Which workers have us "_who_has", # Which clients want us "_who_wants", "_exception", "_exception_text", "_traceback", "_traceback_text", "_erred_on", "_exception_blame", "_suspicious", "_retries", "_nbytes", "_type", "_group_key", "_group", "_metadata", "_annotations", ) def __init__(self, key: str, run_spec: object): self._key = key self._hash = hash(key) self._run_spec = run_spec self._state = None # type: ignore self._exception = None self._exception_blame = None # type: ignore self._traceback = None self._exception_text = "" self._traceback_text = "" self._suspicious = 0 self._retries = 0 self._nbytes = -1 self._priority = None # type: ignore self._who_wants = set() self._dependencies = set() self._dependents = set() self._waiting_on = set() self._waiters = set() self._who_has = set() self._processing_on = None # type: ignore self._has_lost_dependencies = False self._host_restrictions = None # type: ignore self._worker_restrictions = None # type: ignore self._resource_restrictions = None # type: ignore self._loose_restrictions = False self._actor = False self._type = None # type: ignore self._group_key = key_split_group(key) self._group = None # type: ignore self._metadata = {} self._annotations = {} self._erred_on = set() def __hash__(self): return self._hash def __eq__(self, other): typ_self: type = type(self) typ_other: type = type(other) if typ_self == typ_other: other_ts: TaskState = other return self._key == other_ts._key else: return False @property def key(self): return self._key @property def prefix(self): return self._prefix @property def run_spec(self): return self._run_spec @property def priority(self) -> "tuple | None": return self._priority @property def state(self) -> "str | None": return self._state @state.setter def state(self, value: str): self._group._states[self._state] -= 1 self._group._states[value] += 1 self._state = value @property def dependencies(self) -> "set[TaskState]": return self._dependencies @property def dependents(self) -> "set[TaskState]": return self._dependents @property def has_lost_dependencies(self): return self._has_lost_dependencies @property def waiting_on(self) -> "set[TaskState]": return self._waiting_on @property def waiters(self) -> "set[TaskState]": return self._waiters @property def who_wants(self) -> "set[ClientState]": return self._who_wants @property def who_has(self) -> "set[WorkerState]": return self._who_has @property def processing_on(self) -> "WorkerState | None": return self._processing_on @processing_on.setter def processing_on(self, v: WorkerState) -> None: self._processing_on = v @property def retries(self): return self._retries @property def nbytes(self): return self._nbytes @nbytes.setter def nbytes(self, v: Py_ssize_t): self._nbytes = v @property def type(self) -> "str | None": return self._type @property def exception(self): return self._exception @property def exception_text(self): return self._exception_text @property def traceback(self): return self._traceback @property def traceback_text(self): return self._traceback_text @property def exception_blame(self) -> "TaskState | None": return self._exception_blame @property def suspicious(self): return self._suspicious @property def host_restrictions(self) -> "set[str] | None": return self._host_restrictions @property def worker_restrictions(self) -> "set[str] | None": return self._worker_restrictions @property def resource_restrictions(self) -> "dict | None": return self._resource_restrictions @property def loose_restrictions(self): return self._loose_restrictions @property def metadata(self): return self._metadata @property def annotations(self): return self._annotations @property def actor(self): return self._actor @property def group(self) -> "TaskGroup | None": return self._group @property def group_key(self) -> str: return self._group_key @property def prefix_key(self): return self._prefix._name @property def erred_on(self): return self._erred_on @ccall def add_dependency(self, other: "TaskState"): """Add another task as a dependency of this task""" self._dependencies.add(other) self._group._dependencies.add(other._group) other._dependents.add(self) @ccall @inline @nogil def get_nbytes(self) -> Py_ssize_t: return self._nbytes if self._nbytes >= 0 else DEFAULT_DATA_SIZE @ccall def set_nbytes(self, nbytes: Py_ssize_t): diff: Py_ssize_t = nbytes old_nbytes: Py_ssize_t = self._nbytes if old_nbytes >= 0: diff -= old_nbytes self._group._nbytes_total += diff ws: WorkerState for ws in self._who_has: ws._nbytes += diff self._nbytes = nbytes def __repr__(self): return f"" def _repr_html_(self): return get_template("task_state.html.j2").render( state=self._state, nbytes=self._nbytes, key=self._key, ) @ccall def validate(self): try: for cs in self._who_wants: assert isinstance(cs, ClientState), (repr(cs), self._who_wants) for ws in self._who_has: assert isinstance(ws, WorkerState), (repr(ws), self._who_has) for ts in self._dependencies: assert isinstance(ts, TaskState), (repr(ts), self._dependencies) for ts in self._dependents: assert isinstance(ts, TaskState), (repr(ts), self._dependents) validate_task_state(self) except Exception as e: logger.exception(e) if LOG_PDB: import pdb pdb.set_trace() def get_nbytes_deps(self): nbytes: Py_ssize_t = 0 ts: TaskState for ts in self._dependencies: nbytes += ts.get_nbytes() return nbytes @ccall def _to_dict(self, *, exclude: Container[str] = None): """ A very verbose dictionary representation for debugging purposes. Not type stable and not inteded for roundtrips. Parameters ---------- exclude: A list of attributes which must not be present in the output. See also -------- Client.dump_cluster_state """ if not exclude: exclude = set() members = inspect.getmembers(self) return recursive_to_dict( {k: v for k, v in members if k not in exclude and not callable(v)}, exclude=exclude, ) class _StateLegacyMapping(Mapping): """ A mapping interface mimicking the former Scheduler state dictionaries. """ def __init__(self, states, accessor): self._states = states self._accessor = accessor def __iter__(self): return iter(self._states) def __len__(self): return len(self._states) def __getitem__(self, key): return self._accessor(self._states[key]) def __repr__(self): return f"{self.__class__}({dict(self)})" class _OptionalStateLegacyMapping(_StateLegacyMapping): """ Similar to _StateLegacyMapping, but a false-y value is interpreted as a missing key. """ # For tasks etc. def __iter__(self): accessor = self._accessor for k, v in self._states.items(): if accessor(v): yield k def __len__(self): accessor = self._accessor return sum(bool(accessor(v)) for v in self._states.values()) def __getitem__(self, key): v = self._accessor(self._states[key]) if v: return v else: raise KeyError class _StateLegacySet(Set): """ Similar to _StateLegacyMapping, but exposes a set containing all values with a true value. """ # For loose_restrictions def __init__(self, states, accessor): self._states = states self._accessor = accessor def __iter__(self): return (k for k, v in self._states.items() if self._accessor(v)) def __len__(self): return sum(map(bool, map(self._accessor, self._states.values()))) def __contains__(self, k): st = self._states.get(k) return st is not None and bool(self._accessor(st)) def __repr__(self): return f"{self.__class__}({set(self)})" def _legacy_task_key_set(tasks): """ Transform a set of task states into a set of task keys. """ ts: TaskState return {ts._key for ts in tasks} def _legacy_client_key_set(clients): """ Transform a set of client states into a set of client keys. """ cs: ClientState return {cs._client_key for cs in clients} def _legacy_worker_key_set(workers): """ Transform a set of worker states into a set of worker keys. """ ws: WorkerState return {ws._address for ws in workers} def _legacy_task_key_dict(task_dict: dict): """ Transform a dict of {task state: value} into a dict of {task key: value}. """ ts: TaskState return {ts._key: value for ts, value in task_dict.items()} def _task_key_or_none(task: TaskState): return task._key if task is not None else None @cclass class SchedulerState: """Underlying task state of dynamic scheduler Tracks the current state of workers, data, and computations. Handles transitions between different task states. Notifies the Scheduler of changes by messaging passing through Queues, which the Scheduler listens to responds accordingly. All events are handled quickly, in linear time with respect to their input (which is often of constant size) and generally within a millisecond. Additionally when Cythonized, this can be faster still. To accomplish this the scheduler tracks a lot of state. Every operation maintains the consistency of this state. Users typically do not interact with ``Transitions`` directly. Instead users interact with the ``Client``, which in turn engages the ``Scheduler`` affecting different transitions here under-the-hood. In the background ``Worker``s also engage with the ``Scheduler`` affecting these state transitions as well. **State** The ``Transitions`` object contains the following state variables. Each variable is listed along with what it stores and a brief description. * **tasks:** ``{task key: TaskState}`` Tasks currently known to the scheduler * **unrunnable:** ``{TaskState}`` Tasks in the "no-worker" state * **workers:** ``{worker key: WorkerState}`` Workers currently connected to the scheduler * **idle:** ``{WorkerState}``: Set of workers that are not fully utilized * **saturated:** ``{WorkerState}``: Set of workers that are not over-utilized * **running:** ``{WorkerState}``: Set of workers that are currently in running state * **clients:** ``{client key: ClientState}`` Clients currently connected to the scheduler * **task_duration:** ``{key-prefix: time}`` Time we expect certain functions to take, e.g. ``{'sum': 0.25}`` """ _aliases: dict _bandwidth: double _clients: dict # dict[str, ClientState] _computations: object _extensions: dict _host_info: dict _idle: "SortedDict[str, WorkerState]" _idle_dv: dict # dict[str, WorkerState] _n_tasks: Py_ssize_t _resources: dict _saturated: set # set[WorkerState] _running: set # set[WorkerState] _tasks: dict _task_groups: dict _task_prefixes: dict _task_metadata: dict _replicated_tasks: set _total_nthreads: Py_ssize_t _total_occupancy: double _transitions_table: dict _unknown_durations: dict _unrunnable: set _validate: bint _workers: "SortedDict[str, WorkerState]" _workers_dv: dict # dict[str, WorkerState] _transition_counter: Py_ssize_t _plugins: dict # dict[str, SchedulerPlugin] # Variables from dask.config, cached by __init__ for performance UNKNOWN_TASK_DURATION: double MEMORY_RECENT_TO_OLD_TIME: double MEMORY_REBALANCE_MEASURE: str MEMORY_REBALANCE_SENDER_MIN: double MEMORY_REBALANCE_RECIPIENT_MAX: double MEMORY_REBALANCE_HALF_GAP: double def __init__( self, aliases: dict, clients: "dict[str, ClientState]", workers: "SortedDict[str, WorkerState]", host_info: dict, resources: dict, tasks: dict, unrunnable: set, validate: bint, plugins: "Iterable[SchedulerPlugin]" = (), **kwargs, # Passed verbatim to Server.__init__() ): self._aliases = aliases self._bandwidth = parse_bytes( dask.config.get("distributed.scheduler.bandwidth") ) self._clients = clients self._clients["fire-and-forget"] = ClientState("fire-and-forget") self._extensions = {} self._host_info = host_info self._idle = SortedDict() # Note: cython.cast, not typing.cast! self._idle_dv = cast(dict, self._idle) self._n_tasks = 0 self._resources = resources self._saturated = set() self._tasks = tasks self._replicated_tasks = { ts for ts in self._tasks.values() if len(ts._who_has) > 1 } self._computations = deque( maxlen=dask.config.get("distributed.diagnostics.computations.max-history") ) self._task_groups = {} self._task_prefixes = {} self._task_metadata = {} self._total_nthreads = 0 self._total_occupancy = 0 self._transitions_table = { ("released", "waiting"): self.transition_released_waiting, ("waiting", "released"): self.transition_waiting_released, ("waiting", "processing"): self.transition_waiting_processing, ("waiting", "memory"): self.transition_waiting_memory, ("processing", "released"): self.transition_processing_released, ("processing", "memory"): self.transition_processing_memory, ("processing", "erred"): self.transition_processing_erred, ("no-worker", "released"): self.transition_no_worker_released, ("no-worker", "waiting"): self.transition_no_worker_waiting, ("no-worker", "memory"): self.transition_no_worker_memory, ("released", "forgotten"): self.transition_released_forgotten, ("memory", "forgotten"): self.transition_memory_forgotten, ("erred", "released"): self.transition_erred_released, ("memory", "released"): self.transition_memory_released, ("released", "erred"): self.transition_released_erred, } self._unknown_durations = {} self._unrunnable = unrunnable self._validate = validate self._workers = workers # Note: cython.cast, not typing.cast! self._workers_dv = cast(dict, self._workers) self._running = { ws for ws in self._workers.values() if ws.status == Status.running } self._plugins = {} if not plugins else {_get_plugin_name(p): p for p in plugins} # Variables from dask.config, cached by __init__ for performance self.UNKNOWN_TASK_DURATION = parse_timedelta( dask.config.get("distributed.scheduler.unknown-task-duration") ) self.MEMORY_RECENT_TO_OLD_TIME = parse_timedelta( dask.config.get("distributed.worker.memory.recent-to-old-time") ) self.MEMORY_REBALANCE_MEASURE = dask.config.get( "distributed.worker.memory.rebalance.measure" ) self.MEMORY_REBALANCE_SENDER_MIN = dask.config.get( "distributed.worker.memory.rebalance.sender-min" ) self.MEMORY_REBALANCE_RECIPIENT_MAX = dask.config.get( "distributed.worker.memory.rebalance.recipient-max" ) self.MEMORY_REBALANCE_HALF_GAP = ( dask.config.get("distributed.worker.memory.rebalance.sender-recipient-gap") / 2.0 ) self._transition_counter = 0 # Call Server.__init__() super().__init__(**kwargs) # type: ignore @property def aliases(self): return self._aliases @property def bandwidth(self): return self._bandwidth @property def clients(self): return self._clients @property def computations(self): return self._computations @property def extensions(self): return self._extensions @property def host_info(self): return self._host_info @property def idle(self): return self._idle @property def n_tasks(self): return self._n_tasks @property def resources(self): return self._resources @property def saturated(self) -> "set[WorkerState]": return self._saturated @property def running(self) -> "set[WorkerState]": return self._running @property def tasks(self): return self._tasks @property def task_groups(self): return self._task_groups @property def task_prefixes(self): return self._task_prefixes @property def task_metadata(self): return self._task_metadata @property def replicated_tasks(self): return self._replicated_tasks @property def total_nthreads(self): return self._total_nthreads @property def total_occupancy(self): return self._total_occupancy @total_occupancy.setter def total_occupancy(self, v: double): self._total_occupancy = v @property def transition_counter(self): return self._transition_counter @property def unknown_durations(self): return self._unknown_durations @property def unrunnable(self): return self._unrunnable @property def validate(self): return self._validate @validate.setter def validate(self, v: bint): self._validate = v @property def workers(self): return self._workers @property def plugins(self) -> "dict[str, SchedulerPlugin]": return self._plugins @property def memory(self) -> MemoryState: return MemoryState.sum(*(w.memory for w in self.workers.values())) @property def __pdict__(self): return { "bandwidth": self._bandwidth, "resources": self._resources, "saturated": self._saturated, "unrunnable": self._unrunnable, "n_tasks": self._n_tasks, "unknown_durations": self._unknown_durations, "validate": self._validate, "tasks": self._tasks, "task_groups": self._task_groups, "task_prefixes": self._task_prefixes, "total_nthreads": self._total_nthreads, "total_occupancy": self._total_occupancy, "extensions": self._extensions, "clients": self._clients, "workers": self._workers, "idle": self._idle, "host_info": self._host_info, } @ccall @exceptval(check=False) def new_task( self, key: str, spec: object, state: str, computation: Computation = None ) -> TaskState: """Create a new task, and associated states""" ts: TaskState = TaskState(key, spec) ts._state = state tp: TaskPrefix prefix_key = key_split(key) tp = self._task_prefixes.get(prefix_key) # type: ignore if tp is None: self._task_prefixes[prefix_key] = tp = TaskPrefix(prefix_key) ts._prefix = tp group_key = ts._group_key tg: TaskGroup = self._task_groups.get(group_key) # type: ignore if tg is None: self._task_groups[group_key] = tg = TaskGroup(group_key) if computation: computation.groups.add(tg) tg._prefix = tp tp._groups.append(tg) tg.add(ts) self._tasks[key] = ts return ts ##################### # State Transitions # ##################### def _transition(self, key, finish: str, *args, **kwargs): """Transition a key from its current state to the finish state Examples -------- >>> self._transition('x', 'waiting') {'x': 'processing'} Returns ------- Dictionary of recommendations for future transitions See Also -------- Scheduler.transitions : transitive version of this function """ parent: SchedulerState = cast(SchedulerState, self) ts: TaskState start: str start_finish: tuple finish2: str recommendations: dict worker_msgs: dict client_msgs: dict msgs: list new_msgs: list dependents: set dependencies: set try: recommendations = {} worker_msgs = {} client_msgs = {} ts = parent._tasks.get(key) # type: ignore if ts is None: return recommendations, client_msgs, worker_msgs start = ts._state if start == finish: return recommendations, client_msgs, worker_msgs if self.plugins: dependents = set(ts._dependents) dependencies = set(ts._dependencies) start_finish = (start, finish) func = self._transitions_table.get(start_finish) if func is not None: recommendations, client_msgs, worker_msgs = func(key, *args, **kwargs) self._transition_counter += 1 elif "released" not in start_finish: assert not args and not kwargs, (args, kwargs, start_finish) a_recs: dict a_cmsgs: dict a_wmsgs: dict a: tuple = self._transition(key, "released") a_recs, a_cmsgs, a_wmsgs = a v = a_recs.get(key, finish) func = self._transitions_table["released", v] b_recs: dict b_cmsgs: dict b_wmsgs: dict b: tuple = func(key) b_recs, b_cmsgs, b_wmsgs = b recommendations.update(a_recs) for c, new_msgs in a_cmsgs.items(): msgs = client_msgs.get(c) # type: ignore if msgs is not None: msgs.extend(new_msgs) else: client_msgs[c] = new_msgs for w, new_msgs in a_wmsgs.items(): msgs = worker_msgs.get(w) # type: ignore if msgs is not None: msgs.extend(new_msgs) else: worker_msgs[w] = new_msgs recommendations.update(b_recs) for c, new_msgs in b_cmsgs.items(): msgs = client_msgs.get(c) # type: ignore if msgs is not None: msgs.extend(new_msgs) else: client_msgs[c] = new_msgs for w, new_msgs in b_wmsgs.items(): msgs = worker_msgs.get(w) # type: ignore if msgs is not None: msgs.extend(new_msgs) else: worker_msgs[w] = new_msgs start = "released" else: raise RuntimeError("Impossible transition from %r to %r" % start_finish) finish2 = ts._state # FIXME downcast antipattern scheduler = pep484_cast(Scheduler, self) scheduler.transition_log.append( (key, start, finish2, recommendations, time()) ) if parent._validate: logger.debug( "Transitioned %r %s->%s (actual: %s). Consequence: %s", key, start, finish2, ts._state, dict(recommendations), ) if self.plugins: # Temporarily put back forgotten key for plugin to retrieve it if ts._state == "forgotten": ts._dependents = dependents ts._dependencies = dependencies parent._tasks[ts._key] = ts for plugin in list(self.plugins.values()): try: plugin.transition(key, start, finish2, *args, **kwargs) except Exception: logger.info("Plugin failed with exception", exc_info=True) if ts._state == "forgotten": del parent._tasks[ts._key] tg: TaskGroup = ts._group if ts._state == "forgotten" and tg._name in parent._task_groups: # Remove TaskGroup if all tasks are in the forgotten state all_forgotten: bint = True for s in ALL_TASK_STATES: if tg._states.get(s): all_forgotten = False break if all_forgotten: ts._prefix._groups.remove(tg) del parent._task_groups[tg._name] return recommendations, client_msgs, worker_msgs except Exception: logger.exception("Error transitioning %r from %r to %r", key, start, finish) if LOG_PDB: import pdb pdb.set_trace() raise def _transitions(self, recommendations: dict, client_msgs: dict, worker_msgs: dict): """Process transitions until none are left This includes feedback from previous transitions and continues until we reach a steady state """ keys: set = set() recommendations = recommendations.copy() msgs: list new_msgs: list new: tuple new_recs: dict new_cmsgs: dict new_wmsgs: dict while recommendations: key, finish = recommendations.popitem() keys.add(key) new = self._transition(key, finish) new_recs, new_cmsgs, new_wmsgs = new recommendations.update(new_recs) for c, new_msgs in new_cmsgs.items(): msgs = client_msgs.get(c) # type: ignore if msgs is not None: msgs.extend(new_msgs) else: client_msgs[c] = new_msgs for w, new_msgs in new_wmsgs.items(): msgs = worker_msgs.get(w) # type: ignore if msgs is not None: msgs.extend(new_msgs) else: worker_msgs[w] = new_msgs if self._validate: # FIXME downcast antipattern scheduler = pep484_cast(Scheduler, self) for key in keys: scheduler.validate_key(key) def transition_released_waiting(self, key): try: ts: TaskState = self._tasks[key] dts: TaskState recommendations: dict = {} client_msgs: dict = {} worker_msgs: dict = {} if self._validate: assert ts._run_spec assert not ts._waiting_on assert not ts._who_has assert not ts._processing_on assert not any([dts._state == "forgotten" for dts in ts._dependencies]) if ts._has_lost_dependencies: recommendations[key] = "forgotten" return recommendations, client_msgs, worker_msgs ts.state = "waiting" dts: TaskState for dts in ts._dependencies: if dts._exception_blame: ts._exception_blame = dts._exception_blame recommendations[key] = "erred" return recommendations, client_msgs, worker_msgs for dts in ts._dependencies: dep = dts._key if not dts._who_has: ts._waiting_on.add(dts) if dts._state == "released": recommendations[dep] = "waiting" else: dts._waiters.add(ts) ts._waiters = {dts for dts in ts._dependents if dts._state == "waiting"} if not ts._waiting_on: if self._workers_dv: recommendations[key] = "processing" else: self._unrunnable.add(ts) ts.state = "no-worker" return recommendations, client_msgs, worker_msgs except Exception as e: logger.exception(e) if LOG_PDB: import pdb pdb.set_trace() raise def transition_no_worker_waiting(self, key): try: ts: TaskState = self._tasks[key] dts: TaskState recommendations: dict = {} client_msgs: dict = {} worker_msgs: dict = {} if self._validate: assert ts in self._unrunnable assert not ts._waiting_on assert not ts._who_has assert not ts._processing_on self._unrunnable.remove(ts) if ts._has_lost_dependencies: recommendations[key] = "forgotten" return recommendations, client_msgs, worker_msgs for dts in ts._dependencies: dep = dts._key if not dts._who_has: ts._waiting_on.add(dts) if dts._state == "released": recommendations[dep] = "waiting" else: dts._waiters.add(ts) ts.state = "waiting" if not ts._waiting_on: if self._workers_dv: recommendations[key] = "processing" else: self._unrunnable.add(ts) ts.state = "no-worker" return recommendations, client_msgs, worker_msgs except Exception as e: logger.exception(e) if LOG_PDB: import pdb pdb.set_trace() raise def transition_no_worker_memory( self, key, nbytes=None, type=None, typename: str = None, worker=None ): try: ws: WorkerState = self._workers_dv[worker] ts: TaskState = self._tasks[key] recommendations: dict = {} client_msgs: dict = {} worker_msgs: dict = {} if self._validate: assert not ts._processing_on assert not ts._waiting_on assert ts._state == "no-worker" self._unrunnable.remove(ts) if nbytes is not None: ts.set_nbytes(nbytes) self.check_idle_saturated(ws) _add_to_memory( self, ts, ws, recommendations, client_msgs, type=type, typename=typename ) ts.state = "memory" return recommendations, client_msgs, worker_msgs except Exception as e: logger.exception(e) if LOG_PDB: import pdb pdb.set_trace() raise @ccall @exceptval(check=False) def decide_worker(self, ts: TaskState) -> WorkerState: # -> WorkerState | None """ Decide on a worker for task *ts*. Return a WorkerState. If it's a root or root-like task, we place it with its relatives to reduce future data tansfer. If it has dependencies or restrictions, we use `decide_worker_from_deps_and_restrictions`. Otherwise, we pick the least occupied worker, or pick from all workers in a round-robin fashion. """ if not self._workers_dv: return None # type: ignore ws: WorkerState tg: TaskGroup = ts._group valid_workers: set = self.valid_workers(ts) if ( valid_workers is not None and not valid_workers and not ts._loose_restrictions ): self._unrunnable.add(ts) ts.state = "no-worker" return None # type: ignore # Group is larger than cluster with few dependencies? # Minimize future data transfers. if ( valid_workers is None and len(tg) > self._total_nthreads * 2 and len(tg._dependencies) < 5 and sum(map(len, tg._dependencies)) < 5 ): ws = tg._last_worker if not ( ws and tg._last_worker_tasks_left and ws._address in self._workers_dv ): # Last-used worker is full or unknown; pick a new worker for the next few tasks ws = min( (self._idle_dv or self._workers_dv).values(), key=partial(self.worker_objective, ts), ) tg._last_worker_tasks_left = math.floor( (len(tg) / self._total_nthreads) * ws._nthreads ) # Record `last_worker`, or clear it on the final task tg._last_worker = ( ws if tg.states["released"] + tg.states["waiting"] > 1 else None ) tg._last_worker_tasks_left -= 1 return ws if ts._dependencies or valid_workers is not None: ws = decide_worker( ts, self._workers_dv.values(), valid_workers, partial(self.worker_objective, ts), ) else: # Fastpath when there are no related tasks or restrictions worker_pool = self._idle or self._workers # Note: cython.cast, not typing.cast! worker_pool_dv = cast(dict, worker_pool) wp_vals = worker_pool.values() n_workers: Py_ssize_t = len(worker_pool_dv) if n_workers < 20: # smart but linear in small case ws = min(wp_vals, key=operator.attrgetter("occupancy")) if ws._occupancy == 0: # special case to use round-robin; linear search # for next worker with zero occupancy (or just # land back where we started). wp_i: WorkerState start: Py_ssize_t = self._n_tasks % n_workers i: Py_ssize_t for i in range(n_workers): wp_i = wp_vals[(i + start) % n_workers] if wp_i._occupancy == 0: ws = wp_i break else: # dumb but fast in large case ws = wp_vals[self._n_tasks % n_workers] if self._validate: assert ws is None or isinstance(ws, WorkerState), ( type(ws), ws, ) assert ws._address in self._workers_dv return ws @ccall def set_duration_estimate(self, ts: TaskState, ws: WorkerState) -> double: """Estimate task duration using worker state and task state. If a task takes longer than twice the current average duration we estimate the task duration to be 2x current-runtime, otherwise we set it to be the average duration. See also ``_remove_from_processing`` """ exec_time: double = ws._executing.get(ts, 0) duration: double = self.get_task_duration(ts) total_duration: double if exec_time > 2 * duration: total_duration = 2 * exec_time else: comm: double = self.get_comm_cost(ts, ws) total_duration = duration + comm old = ws._processing.get(ts, 0) ws._processing[ts] = total_duration self._total_occupancy += total_duration - old ws._occupancy += total_duration - old return total_duration def transition_waiting_processing(self, key): try: ts: TaskState = self._tasks[key] dts: TaskState recommendations: dict = {} client_msgs: dict = {} worker_msgs: dict = {} if self._validate: assert not ts._waiting_on assert not ts._who_has assert not ts._exception_blame assert not ts._processing_on assert not ts._has_lost_dependencies assert ts not in self._unrunnable assert all([dts._who_has for dts in ts._dependencies]) ws: WorkerState = self.decide_worker(ts) if ws is None: return recommendations, client_msgs, worker_msgs worker = ws._address self.set_duration_estimate(ts, ws) ts._processing_on = ws ts.state = "processing" self.consume_resources(ts, ws) self.check_idle_saturated(ws) self._n_tasks += 1 if ts._actor: ws._actors.add(ts) # logger.debug("Send job to worker: %s, %s", worker, key) worker_msgs[worker] = [_task_to_msg(self, ts)] return recommendations, client_msgs, worker_msgs except Exception as e: logger.exception(e) if LOG_PDB: import pdb pdb.set_trace() raise def transition_waiting_memory( self, key, nbytes=None, type=None, typename: str = None, worker=None, **kwargs ): try: ws: WorkerState = self._workers_dv[worker] ts: TaskState = self._tasks[key] recommendations: dict = {} client_msgs: dict = {} worker_msgs: dict = {} if self._validate: assert not ts._processing_on assert ts._waiting_on assert ts._state == "waiting" ts._waiting_on.clear() if nbytes is not None: ts.set_nbytes(nbytes) self.check_idle_saturated(ws) _add_to_memory( self, ts, ws, recommendations, client_msgs, type=type, typename=typename ) if self._validate: assert not ts._processing_on assert not ts._waiting_on assert ts._who_has return recommendations, client_msgs, worker_msgs except Exception as e: logger.exception(e) if LOG_PDB: import pdb pdb.set_trace() raise def transition_processing_memory( self, key, nbytes=None, type=None, typename: str = None, worker=None, startstops=None, **kwargs, ): ws: WorkerState wws: WorkerState recommendations: dict = {} client_msgs: dict = {} worker_msgs: dict = {} try: ts: TaskState = self._tasks[key] assert worker assert isinstance(worker, str) if self._validate: assert ts._processing_on ws = ts._processing_on assert ts in ws._processing assert not ts._waiting_on assert not ts._who_has, (ts, ts._who_has) assert not ts._exception_blame assert ts._state == "processing" ws = self._workers_dv.get(worker) # type: ignore if ws is None: recommendations[key] = "released" return recommendations, client_msgs, worker_msgs if ws != ts._processing_on: # someone else has this task logger.info( "Unexpected worker completed task. Expected: %s, Got: %s, Key: %s", ts._processing_on, ws, key, ) worker_msgs[ts._processing_on.address] = [ { "op": "cancel-compute", "key": key, "reason": "Finished on different worker", } ] ############################# # Update Timing Information # ############################# if startstops: startstop: dict for startstop in startstops: ts._group.add_duration( stop=startstop["stop"], start=startstop["start"], action=startstop["action"], ) s: set = self._unknown_durations.pop(ts._prefix._name, set()) tts: TaskState steal = self.extensions.get("stealing") for tts in s: if tts._processing_on: self.set_duration_estimate(tts, tts._processing_on) if steal: steal.put_key_in_stealable(tts) ############################ # Update State Information # ############################ if nbytes is not None: ts.set_nbytes(nbytes) _remove_from_processing(self, ts) _add_to_memory( self, ts, ws, recommendations, client_msgs, type=type, typename=typename ) if self._validate: assert not ts._processing_on assert not ts._waiting_on return recommendations, client_msgs, worker_msgs except Exception as e: logger.exception(e) if LOG_PDB: import pdb pdb.set_trace() raise def transition_memory_released(self, key, safe: bint = False): ws: WorkerState try: ts: TaskState = self._tasks[key] dts: TaskState recommendations: dict = {} client_msgs: dict = {} worker_msgs: dict = {} if self._validate: assert not ts._waiting_on assert not ts._processing_on if safe: assert not ts._waiters if ts._actor: for ws in ts._who_has: ws._actors.discard(ts) if ts._who_wants: ts._exception_blame = ts ts._exception = "Worker holding Actor was lost" recommendations[ts._key] = "erred" return ( recommendations, client_msgs, worker_msgs, ) # don't try to recreate for dts in ts._waiters: if dts._state in ("no-worker", "processing"): recommendations[dts._key] = "waiting" elif dts._state == "waiting": dts._waiting_on.add(ts) # XXX factor this out? worker_msg = { "op": "free-keys", "keys": [key], "stimulus_id": f"memory-released-{time()}", } for ws in ts._who_has: worker_msgs[ws._address] = [worker_msg] self.remove_all_replicas(ts) ts.state = "released" report_msg = {"op": "lost-data", "key": key} cs: ClientState for cs in ts._who_wants: client_msgs[cs._client_key] = [report_msg] if not ts._run_spec: # pure data recommendations[key] = "forgotten" elif ts._has_lost_dependencies: recommendations[key] = "forgotten" elif ts._who_wants or ts._waiters: recommendations[key] = "waiting" if self._validate: assert not ts._waiting_on return recommendations, client_msgs, worker_msgs except Exception as e: logger.exception(e) if LOG_PDB: import pdb pdb.set_trace() raise def transition_released_erred(self, key): try: ts: TaskState = self._tasks[key] dts: TaskState failing_ts: TaskState recommendations: dict = {} client_msgs: dict = {} worker_msgs: dict = {} if self._validate: with log_errors(pdb=LOG_PDB): assert ts._exception_blame assert not ts._who_has assert not ts._waiting_on assert not ts._waiters failing_ts = ts._exception_blame for dts in ts._dependents: dts._exception_blame = failing_ts if not dts._who_has: recommendations[dts._key] = "erred" report_msg = { "op": "task-erred", "key": key, "exception": failing_ts._exception, "traceback": failing_ts._traceback, } cs: ClientState for cs in ts._who_wants: client_msgs[cs._client_key] = [report_msg] ts.state = "erred" # TODO: waiting data? return recommendations, client_msgs, worker_msgs except Exception as e: logger.exception(e) if LOG_PDB: import pdb pdb.set_trace() raise def transition_erred_released(self, key): try: ts: TaskState = self._tasks[key] dts: TaskState recommendations: dict = {} client_msgs: dict = {} worker_msgs: dict = {} if self._validate: with log_errors(pdb=LOG_PDB): assert ts._exception_blame assert not ts._who_has assert not ts._waiting_on assert not ts._waiters ts._exception = None ts._exception_blame = None ts._traceback = None for dts in ts._dependents: if dts._state == "erred": recommendations[dts._key] = "waiting" w_msg = { "op": "free-keys", "keys": [key], "stimulus_id": f"erred-released-{time()}", } for ws_addr in ts._erred_on: worker_msgs[ws_addr] = [w_msg] ts._erred_on.clear() report_msg = {"op": "task-retried", "key": key} cs: ClientState for cs in ts._who_wants: client_msgs[cs._client_key] = [report_msg] ts.state = "released" return recommendations, client_msgs, worker_msgs except Exception as e: logger.exception(e) if LOG_PDB: import pdb pdb.set_trace() raise def transition_waiting_released(self, key): try: ts: TaskState = self._tasks[key] recommendations: dict = {} client_msgs: dict = {} worker_msgs: dict = {} if self._validate: assert not ts._who_has assert not ts._processing_on dts: TaskState for dts in ts._dependencies: if ts in dts._waiters: dts._waiters.discard(ts) if not dts._waiters and not dts._who_wants: recommendations[dts._key] = "released" ts._waiting_on.clear() ts.state = "released" if ts._has_lost_dependencies: recommendations[key] = "forgotten" elif not ts._exception_blame and (ts._who_wants or ts._waiters): recommendations[key] = "waiting" else: ts._waiters.clear() return recommendations, client_msgs, worker_msgs except Exception as e: logger.exception(e) if LOG_PDB: import pdb pdb.set_trace() raise def transition_processing_released(self, key): try: ts: TaskState = self._tasks[key] dts: TaskState recommendations: dict = {} client_msgs: dict = {} worker_msgs: dict = {} if self._validate: assert ts._processing_on assert not ts._who_has assert not ts._waiting_on assert self._tasks[key].state == "processing" w: str = _remove_from_processing(self, ts) if w: worker_msgs[w] = [ { "op": "free-keys", "keys": [key], "stimulus_id": f"processing-released-{time()}", } ] ts.state = "released" if ts._has_lost_dependencies: recommendations[key] = "forgotten" elif ts._waiters or ts._who_wants: recommendations[key] = "waiting" if recommendations.get(key) != "waiting": for dts in ts._dependencies: if dts._state != "released": dts._waiters.discard(ts) if not dts._waiters and not dts._who_wants: recommendations[dts._key] = "released" ts._waiters.clear() if self._validate: assert not ts._processing_on return recommendations, client_msgs, worker_msgs except Exception as e: logger.exception(e) if LOG_PDB: import pdb pdb.set_trace() raise def transition_processing_erred( self, key: str, cause: str = None, exception=None, traceback=None, exception_text: str = None, traceback_text: str = None, worker: str = None, **kwargs, ): ws: WorkerState try: ts: TaskState = self._tasks[key] dts: TaskState failing_ts: TaskState recommendations: dict = {} client_msgs: dict = {} worker_msgs: dict = {} if self._validate: assert cause or ts._exception_blame assert ts._processing_on assert not ts._who_has assert not ts._waiting_on if ts._actor: ws = ts._processing_on ws._actors.remove(ts) w = _remove_from_processing(self, ts) ts._erred_on.add(w or worker) if exception is not None: ts._exception = exception ts._exception_text = exception_text # type: ignore if traceback is not None: ts._traceback = traceback ts._traceback_text = traceback_text # type: ignore if cause is not None: failing_ts = self._tasks[cause] ts._exception_blame = failing_ts else: failing_ts = ts._exception_blame # type: ignore for dts in ts._dependents: dts._exception_blame = failing_ts recommendations[dts._key] = "erred" for dts in ts._dependencies: dts._waiters.discard(ts) if not dts._waiters and not dts._who_wants: recommendations[dts._key] = "released" ts._waiters.clear() # do anything with this? ts.state = "erred" report_msg = { "op": "task-erred", "key": key, "exception": failing_ts._exception, "traceback": failing_ts._traceback, } cs: ClientState for cs in ts._who_wants: client_msgs[cs._client_key] = [report_msg] cs = self._clients["fire-and-forget"] if ts in cs._wants_what: _client_releases_keys( self, cs=cs, keys=[key], recommendations=recommendations, ) if self._validate: assert not ts._processing_on return recommendations, client_msgs, worker_msgs except Exception as e: logger.exception(e) if LOG_PDB: import pdb pdb.set_trace() raise def transition_no_worker_released(self, key): try: ts: TaskState = self._tasks[key] dts: TaskState recommendations: dict = {} client_msgs: dict = {} worker_msgs: dict = {} if self._validate: assert self._tasks[key].state == "no-worker" assert not ts._who_has assert not ts._waiting_on self._unrunnable.remove(ts) ts.state = "released" for dts in ts._dependencies: dts._waiters.discard(ts) ts._waiters.clear() return recommendations, client_msgs, worker_msgs except Exception as e: logger.exception(e) if LOG_PDB: import pdb pdb.set_trace() raise @ccall def remove_key(self, key): ts: TaskState = self._tasks.pop(key) assert ts._state == "forgotten" self._unrunnable.discard(ts) cs: ClientState for cs in ts._who_wants: cs._wants_what.remove(ts) ts._who_wants.clear() ts._processing_on = None ts._exception_blame = ts._exception = ts._traceback = None self._task_metadata.pop(key, None) def transition_memory_forgotten(self, key): ws: WorkerState try: ts: TaskState = self._tasks[key] recommendations: dict = {} client_msgs: dict = {} worker_msgs: dict = {} if self._validate: assert ts._state == "memory" assert not ts._processing_on assert not ts._waiting_on if not ts._run_spec: # It's ok to forget a pure data task pass elif ts._has_lost_dependencies: # It's ok to forget a task with forgotten dependencies pass elif not ts._who_wants and not ts._waiters and not ts._dependents: # It's ok to forget a task that nobody needs pass else: assert 0, (ts,) if ts._actor: for ws in ts._who_has: ws._actors.discard(ts) _propagate_forgotten(self, ts, recommendations, worker_msgs) client_msgs = _task_to_client_msgs(self, ts) self.remove_key(key) return recommendations, client_msgs, worker_msgs except Exception as e: logger.exception(e) if LOG_PDB: import pdb pdb.set_trace() raise def transition_released_forgotten(self, key): try: ts: TaskState = self._tasks[key] recommendations: dict = {} client_msgs: dict = {} worker_msgs: dict = {} if self._validate: assert ts._state in ("released", "erred") assert not ts._who_has assert not ts._processing_on assert not ts._waiting_on, (ts, ts._waiting_on) if not ts._run_spec: # It's ok to forget a pure data task pass elif ts._has_lost_dependencies: # It's ok to forget a task with forgotten dependencies pass elif not ts._who_wants and not ts._waiters and not ts._dependents: # It's ok to forget a task that nobody needs pass else: assert 0, (ts,) _propagate_forgotten(self, ts, recommendations, worker_msgs) client_msgs = _task_to_client_msgs(self, ts) self.remove_key(key) return recommendations, client_msgs, worker_msgs except Exception as e: logger.exception(e) if LOG_PDB: import pdb pdb.set_trace() raise ############################## # Assigning Tasks to Workers # ############################## @ccall @exceptval(check=False) def check_idle_saturated(self, ws: WorkerState, occ: double = -1.0): """Update the status of the idle and saturated state The scheduler keeps track of workers that are .. - Saturated: have enough work to stay busy - Idle: do not have enough work to stay busy They are considered saturated if they both have enough tasks to occupy all of their threads, and if the expected runtime of those tasks is large enough. This is useful for load balancing and adaptivity. """ if self._total_nthreads == 0 or ws.status == Status.closed: return if occ < 0: occ = ws._occupancy nc: Py_ssize_t = ws._nthreads p: Py_ssize_t = len(ws._processing) avg: double = self._total_occupancy / self._total_nthreads idle = self._idle saturated: set = self._saturated if p < nc or occ < nc * avg / 2: idle[ws._address] = ws saturated.discard(ws) else: idle.pop(ws._address, None) if p > nc: pending: double = occ * (p - nc) / (p * nc) if 0.4 < pending > 1.9 * avg: saturated.add(ws) return saturated.discard(ws) @ccall def get_comm_cost(self, ts: TaskState, ws: WorkerState) -> double: """ Get the estimated communication cost (in s.) to compute the task on the given worker. """ dts: TaskState deps: set = ts._dependencies.difference(ws._has_what) nbytes: Py_ssize_t = 0 for dts in deps: nbytes += dts._nbytes return nbytes / self._bandwidth @ccall def get_task_duration(self, ts: TaskState) -> double: """Get the estimated computation cost of the given task (not including any communication cost). If no data has been observed, value of `distributed.scheduler.default-task-durations` are used. If none is set for this task, `distributed.scheduler.unknown-task-duration` is used instead. """ duration: double = ts._prefix._duration_average if duration >= 0: return duration s: set = self._unknown_durations.get(ts._prefix._name) # type: ignore if s is None: self._unknown_durations[ts._prefix._name] = s = set() s.add(ts) return self.UNKNOWN_TASK_DURATION @ccall @exceptval(check=False) def valid_workers(self, ts: TaskState) -> set: # set[WorkerState] | None """Return set of currently valid workers for key If all workers are valid then this returns ``None``. This checks tracks the following state: * worker_restrictions * host_restrictions * resource_restrictions """ s: set = None # type: ignore if ts._worker_restrictions: s = {addr for addr in ts._worker_restrictions if addr in self._workers_dv} if ts._host_restrictions: # Resolve the alias here rather than early, for the worker # may not be connected when host_restrictions is populated hr: list = [self.coerce_hostname(h) for h in ts._host_restrictions] # XXX need HostState? sl: list = [] for h in hr: dh: dict = self._host_info.get(h) # type: ignore if dh is not None: sl.append(dh["addresses"]) ss: set = set.union(*sl) if sl else set() if s is None: s = ss else: s |= ss if ts._resource_restrictions: dw: dict = {} for resource, required in ts._resource_restrictions.items(): dr: dict = self._resources.get(resource) # type: ignore if dr is None: self._resources[resource] = dr = {} sw: set = set() for addr, supplied in dr.items(): if supplied >= required: sw.add(addr) dw[resource] = sw ww: set = set.intersection(*dw.values()) if s is None: s = ww else: s &= ww if s is None: if len(self._running) < len(self._workers_dv): return self._running.copy() else: s = {self._workers_dv[addr] for addr in s} if len(self._running) < len(self._workers_dv): s &= self._running return s @ccall def consume_resources(self, ts: TaskState, ws: WorkerState): if ts._resource_restrictions: for r, required in ts._resource_restrictions.items(): ws._used_resources[r] += required @ccall def release_resources(self, ts: TaskState, ws: WorkerState): if ts._resource_restrictions: for r, required in ts._resource_restrictions.items(): ws._used_resources[r] -= required @ccall def coerce_hostname(self, host): """ Coerce the hostname of a worker. """ alias = self._aliases.get(host) if alias is not None: ws: WorkerState = self._workers_dv[alias] return ws.host else: return host @ccall @exceptval(check=False) def worker_objective(self, ts: TaskState, ws: WorkerState) -> tuple: """ Objective function to determine which worker should get the task Minimize expected start time. If a tie then break with data storage. """ dts: TaskState nbytes: Py_ssize_t comm_bytes: Py_ssize_t = 0 for dts in ts._dependencies: if ws not in dts._who_has: nbytes = dts.get_nbytes() comm_bytes += nbytes stack_time: double = ws._occupancy / ws._nthreads start_time: double = stack_time + comm_bytes / self._bandwidth if ts._actor: return (len(ws._actors), start_time, ws._nbytes) else: return (start_time, ws._nbytes) @ccall def add_replica(self, ts: TaskState, ws: WorkerState): """Note that a worker holds a replica of a task with state='memory'""" if self._validate: assert ws not in ts._who_has assert ts not in ws._has_what ws._nbytes += ts.get_nbytes() ws._has_what[ts] = None ts._who_has.add(ws) if len(ts._who_has) == 2: self._replicated_tasks.add(ts) @ccall def remove_replica(self, ts: TaskState, ws: WorkerState): """Note that a worker no longer holds a replica of a task""" ws._nbytes -= ts.get_nbytes() del ws._has_what[ts] ts._who_has.remove(ws) if len(ts._who_has) == 1: self._replicated_tasks.remove(ts) @ccall def remove_all_replicas(self, ts: TaskState): """Remove all replicas of a task from all workers""" ws: WorkerState nbytes: Py_ssize_t = ts.get_nbytes() for ws in ts._who_has: ws._nbytes -= nbytes del ws._has_what[ts] if len(ts._who_has) > 1: self._replicated_tasks.remove(ts) ts._who_has.clear() class Scheduler(SchedulerState, ServerNode): """Dynamic distributed task scheduler The scheduler tracks the current state of workers, data, and computations. The scheduler listens for events and responds by controlling workers appropriately. It continuously tries to use the workers to execute an ever growing dask graph. All events are handled quickly, in linear time with respect to their input (which is often of constant size) and generally within a millisecond. To accomplish this the scheduler tracks a lot of state. Every operation maintains the consistency of this state. The scheduler communicates with the outside world through Comm objects. It maintains a consistent and valid view of the world even when listening to several clients at once. A Scheduler is typically started either with the ``dask-scheduler`` executable:: $ dask-scheduler Scheduler started at 127.0.0.1:8786 Or within a LocalCluster a Client starts up without connection information:: >>> c = Client() # doctest: +SKIP >>> c.cluster.scheduler # doctest: +SKIP Scheduler(...) Users typically do not interact with the scheduler directly but rather with the client object ``Client``. **State** The scheduler contains the following state variables. Each variable is listed along with what it stores and a brief description. * **tasks:** ``{task key: TaskState}`` Tasks currently known to the scheduler * **unrunnable:** ``{TaskState}`` Tasks in the "no-worker" state * **workers:** ``{worker key: WorkerState}`` Workers currently connected to the scheduler * **idle:** ``{WorkerState}``: Set of workers that are not fully utilized * **saturated:** ``{WorkerState}``: Set of workers that are not over-utilized * **host_info:** ``{hostname: dict}``: Information about each worker host * **clients:** ``{client key: ClientState}`` Clients currently connected to the scheduler * **services:** ``{str: port}``: Other services running on this scheduler, like Bokeh * **loop:** ``IOLoop``: The running Tornado IOLoop * **client_comms:** ``{client key: Comm}`` For each client, a Comm object used to receive task requests and report task status updates. * **stream_comms:** ``{worker key: Comm}`` For each worker, a Comm object from which we both accept stimuli and report results * **task_duration:** ``{key-prefix: time}`` Time we expect certain functions to take, e.g. ``{'sum': 0.25}`` """ default_port = 8786 _instances: "ClassVar[weakref.WeakSet[Scheduler]]" = weakref.WeakSet() def __init__( self, loop=None, delete_interval="500ms", synchronize_worker_interval="60s", services=None, service_kwargs=None, allowed_failures=None, extensions=None, validate=None, scheduler_file=None, security=None, worker_ttl=None, idle_timeout=None, interface=None, host=None, port=0, protocol=None, dashboard_address=None, dashboard=None, http_prefix="/", preload=None, preload_argv=(), plugins=(), **kwargs, ): self._setup_logging(logger) # Attributes if allowed_failures is None: allowed_failures = dask.config.get("distributed.scheduler.allowed-failures") self.allowed_failures = allowed_failures if validate is None: validate = dask.config.get("distributed.scheduler.validate") self.proc = psutil.Process() self.delete_interval = parse_timedelta(delete_interval, default="ms") self.synchronize_worker_interval = parse_timedelta( synchronize_worker_interval, default="ms" ) self.digests = None self.service_specs = services or {} self.service_kwargs = service_kwargs or {} self.services = {} self.scheduler_file = scheduler_file worker_ttl = worker_ttl or dask.config.get("distributed.scheduler.worker-ttl") self.worker_ttl = parse_timedelta(worker_ttl) if worker_ttl else None idle_timeout = idle_timeout or dask.config.get( "distributed.scheduler.idle-timeout" ) if idle_timeout: self.idle_timeout = parse_timedelta(idle_timeout) else: self.idle_timeout = None self.idle_since = time() self.time_started = self.idle_since # compatibility for dask-gateway self._lock = asyncio.Lock() self.bandwidth_workers = defaultdict(float) self.bandwidth_types = defaultdict(float) if not preload: preload = dask.config.get("distributed.scheduler.preload") if not preload_argv: preload_argv = dask.config.get("distributed.scheduler.preload-argv") self.preloads = preloading.process_preloads(self, preload, preload_argv) if isinstance(security, dict): security = Security(**security) self.security = security or Security() assert isinstance(self.security, Security) self.connection_args = self.security.get_connection_args("scheduler") self.connection_args["handshake_overrides"] = { # common denominator "pickle-protocol": 4 } self._start_address = addresses_from_user_args( host=host, port=port, interface=interface, protocol=protocol, security=security, default_port=self.default_port, ) http_server_modules = dask.config.get("distributed.scheduler.http.routes") show_dashboard = dashboard or (dashboard is None and dashboard_address) # install vanilla route if show_dashboard but bokeh is not installed if show_dashboard: try: import distributed.dashboard.scheduler except ImportError: show_dashboard = False http_server_modules.append("distributed.http.scheduler.missing_bokeh") routes = get_handlers( server=self, modules=http_server_modules, prefix=http_prefix ) self.start_http_server(routes, dashboard_address, default_port=8787) if show_dashboard: distributed.dashboard.scheduler.connect( self.http_application, self.http_server, self, prefix=http_prefix ) # Communication state self.loop = loop or IOLoop.current() self.client_comms = {} self.stream_comms = {} self._worker_coroutines = [] self._ipython_kernel = None # Task state tasks = {} for old_attr, new_attr, wrap in [ ("priority", "priority", None), ("dependencies", "dependencies", _legacy_task_key_set), ("dependents", "dependents", _legacy_task_key_set), ("retries", "retries", None), ]: func = operator.attrgetter(new_attr) if wrap is not None: func = compose(wrap, func) setattr(self, old_attr, _StateLegacyMapping(tasks, func)) for old_attr, new_attr, wrap in [ ("nbytes", "nbytes", None), ("who_wants", "who_wants", _legacy_client_key_set), ("who_has", "who_has", _legacy_worker_key_set), ("waiting", "waiting_on", _legacy_task_key_set), ("waiting_data", "waiters", _legacy_task_key_set), ("rprocessing", "processing_on", None), ("host_restrictions", "host_restrictions", None), ("worker_restrictions", "worker_restrictions", None), ("resource_restrictions", "resource_restrictions", None), ("suspicious_tasks", "suspicious", None), ("exceptions", "exception", None), ("tracebacks", "traceback", None), ("exceptions_blame", "exception_blame", _task_key_or_none), ]: func = operator.attrgetter(new_attr) if wrap is not None: func = compose(wrap, func) setattr(self, old_attr, _OptionalStateLegacyMapping(tasks, func)) for old_attr, new_attr, wrap in [ ("loose_restrictions", "loose_restrictions", None) ]: func = operator.attrgetter(new_attr) if wrap is not None: func = compose(wrap, func) setattr(self, old_attr, _StateLegacySet(tasks, func)) self.generation = 0 self._last_client = None self._last_time = 0 unrunnable = set() self.datasets = {} # Prefix-keyed containers # Client state clients = {} for old_attr, new_attr, wrap in [ ("wants_what", "wants_what", _legacy_task_key_set) ]: func = operator.attrgetter(new_attr) if wrap is not None: func = compose(wrap, func) setattr(self, old_attr, _StateLegacyMapping(clients, func)) # Worker state workers = SortedDict() for old_attr, new_attr, wrap in [ ("nthreads", "nthreads", None), ("worker_bytes", "nbytes", None), ("worker_resources", "resources", None), ("used_resources", "used_resources", None), ("occupancy", "occupancy", None), ("worker_info", "metrics", None), ("processing", "processing", _legacy_task_key_dict), ("has_what", "has_what", _legacy_task_key_set), ]: func = operator.attrgetter(new_attr) if wrap is not None: func = compose(wrap, func) setattr(self, old_attr, _StateLegacyMapping(workers, func)) host_info = {} resources = {} aliases = {} self._task_state_collections = [unrunnable] self._worker_collections = [ workers, host_info, resources, aliases, ] self.transition_log = deque( maxlen=dask.config.get("distributed.scheduler.transition-log-length") ) self.log = deque( maxlen=dask.config.get("distributed.scheduler.transition-log-length") ) self.events = defaultdict( partial( deque, maxlen=dask.config.get("distributed.scheduler.events-log-length") ) ) self.event_counts = defaultdict(int) self.event_subscriber = defaultdict(set) self.worker_plugins = {} self.nanny_plugins = {} worker_handlers = { "task-finished": self.handle_task_finished, "task-erred": self.handle_task_erred, "release-worker-data": self.release_worker_data, "add-keys": self.add_keys, "missing-data": self.handle_missing_data, "long-running": self.handle_long_running, "reschedule": self.reschedule, "keep-alive": lambda *args, **kwargs: None, "log-event": self.log_worker_event, "worker-status-change": self.handle_worker_status_change, } client_handlers = { "update-graph": self.update_graph, "update-graph-hlg": self.update_graph_hlg, "client-desires-keys": self.client_desires_keys, "update-data": self.update_data, "report-key": self.report_on_key, "client-releases-keys": self.client_releases_keys, "heartbeat-client": self.client_heartbeat, "close-client": self.remove_client, "restart": self.restart, "subscribe-topic": self.subscribe_topic, "unsubscribe-topic": self.unsubscribe_topic, } self.handlers = { "register-client": self.add_client, "scatter": self.scatter, "register-worker": self.add_worker, "register_nanny": self.add_nanny, "unregister": self.remove_worker, "gather": self.gather, "cancel": self.stimulus_cancel, "retry": self.stimulus_retry, "feed": self.feed, "terminate": self.close, "broadcast": self.broadcast, "proxy": self.proxy, "ncores": self.get_ncores, "ncores_running": self.get_ncores_running, "has_what": self.get_has_what, "who_has": self.get_who_has, "processing": self.get_processing, "call_stack": self.get_call_stack, "profile": self.get_profile, "performance_report": self.performance_report, "get_logs": self.get_logs, "logs": self.get_logs, "worker_logs": self.get_worker_logs, "log_event": self.log_worker_event, "events": self.get_events, "nbytes": self.get_nbytes, "versions": self.versions, "add_keys": self.add_keys, "rebalance": self.rebalance, "replicate": self.replicate, "start_ipython": self.start_ipython, "run_function": self.run_function, "update_data": self.update_data, "set_resources": self.add_resources, "retire_workers": self.retire_workers, "get_metadata": self.get_metadata, "set_metadata": self.set_metadata, "set_restrictions": self.set_restrictions, "heartbeat_worker": self.heartbeat_worker, "get_task_status": self.get_task_status, "get_task_stream": self.get_task_stream, "register_scheduler_plugin": self.register_scheduler_plugin, "register_worker_plugin": self.register_worker_plugin, "unregister_worker_plugin": self.unregister_worker_plugin, "register_nanny_plugin": self.register_nanny_plugin, "unregister_nanny_plugin": self.unregister_nanny_plugin, "adaptive_target": self.adaptive_target, "workers_to_close": self.workers_to_close, "subscribe_worker_status": self.subscribe_worker_status, "start_task_metadata": self.start_task_metadata, "stop_task_metadata": self.stop_task_metadata, } connection_limit = get_fileno_limit() / 2 super().__init__( # Arguments to SchedulerState aliases=aliases, clients=clients, workers=workers, host_info=host_info, resources=resources, tasks=tasks, unrunnable=unrunnable, validate=validate, plugins=plugins, # Arguments to ServerNode handlers=self.handlers, stream_handlers=merge(worker_handlers, client_handlers), io_loop=self.loop, connection_limit=connection_limit, deserialize=False, connection_args=self.connection_args, **kwargs, ) if self.worker_ttl: pc = PeriodicCallback(self.check_worker_ttl, self.worker_ttl) self.periodic_callbacks["worker-ttl"] = pc if self.idle_timeout: pc = PeriodicCallback(self.check_idle, self.idle_timeout / 4) self.periodic_callbacks["idle-timeout"] = pc if extensions is None: extensions = list(DEFAULT_EXTENSIONS) if dask.config.get("distributed.scheduler.work-stealing"): extensions.append(WorkStealing) for ext in extensions: ext(self) setproctitle("dask-scheduler [not started]") Scheduler._instances.add(self) self.rpc.allow_offload = False self.status = Status.undefined ################## # Administration # ################## def __repr__(self): parent: SchedulerState = cast(SchedulerState, self) return '' % ( self.address, len(parent._workers_dv), parent._total_nthreads, len(parent._tasks), ) def _repr_html_(self): parent: SchedulerState = cast(SchedulerState, self) return get_template("scheduler.html.j2").render( address=self.address, workers=parent._workers_dv, threads=parent._total_nthreads, tasks=parent._tasks, ) def identity(self, comm=None): """Basic information about ourselves and our cluster""" parent: SchedulerState = cast(SchedulerState, self) d = { "type": type(self).__name__, "id": str(self.id), "address": self.address, "services": {key: v.port for (key, v) in self.services.items()}, "started": self.time_started, "workers": { worker.address: worker.identity() for worker in parent._workers_dv.values() }, } return d def _to_dict( self, comm: Comm = None, *, exclude: Container[str] = None ) -> "dict[str, Any]": """ A very verbose dictionary representation for debugging purposes. Not type stable and not inteded for roundtrips. Parameters ---------- comm: exclude: A list of attributes which must not be present in the output. See also -------- Server.identity Client.dump_cluster_state """ info = super()._to_dict(exclude=exclude) extra = { "transition_log": self.transition_log, "log": self.log, "tasks": self.tasks, "events": self.events, } info.update(extra) extensions = {} for name, ex in self.extensions.items(): if hasattr(ex, "_to_dict"): extensions[name] = ex._to_dict() return recursive_to_dict(info, exclude=exclude) def get_worker_service_addr(self, worker, service_name, protocol=False): """ Get the (host, port) address of the named service on the *worker*. Returns None if the service doesn't exist. Parameters ---------- worker : address service_name : str Common services include 'bokeh' and 'nanny' protocol : boolean Whether or not to include a full address with protocol (True) or just a (host, port) pair """ parent: SchedulerState = cast(SchedulerState, self) ws: WorkerState = parent._workers_dv[worker] port = ws._services.get(service_name) if port is None: return None elif protocol: return "%(protocol)s://%(host)s:%(port)d" % { "protocol": ws._address.split("://")[0], "host": ws.host, "port": port, } else: return ws.host, port async def start(self): """Clear out old state and restart all running coroutines""" await super().start() assert self.status != Status.running enable_gc_diagnosis() self.clear_task_state() with suppress(AttributeError): for c in self._worker_coroutines: c.cancel() for addr in self._start_address: await self.listen( addr, allow_offload=False, handshake_overrides={"pickle-protocol": 4, "compression": None}, **self.security.get_listen_args("scheduler"), ) self.ip = get_address_host(self.listen_address) listen_ip = self.ip if listen_ip == "0.0.0.0": listen_ip = "" if self.address.startswith("inproc://"): listen_ip = "localhost" # Services listen on all addresses self.start_services(listen_ip) for listener in self.listeners: logger.info(" Scheduler at: %25s", listener.contact_address) for k, v in self.services.items(): logger.info("%11s at: %25s", k, "%s:%d" % (listen_ip, v.port)) self.loop.add_callback(self.reevaluate_occupancy) if self.scheduler_file: with open(self.scheduler_file, "w") as f: json.dump(self.identity(), f, indent=2) fn = self.scheduler_file # remove file when we close the process def del_scheduler_file(): if os.path.exists(fn): os.remove(fn) weakref.finalize(self, del_scheduler_file) for preload in self.preloads: await preload.start() await asyncio.gather( *[plugin.start(self) for plugin in list(self.plugins.values())] ) self.start_periodic_callbacks() setproctitle(f"dask-scheduler [{self.address}]") return self async def close(self, comm=None, fast=False, close_workers=False): """Send cleanup signal to all coroutines then wait until finished See Also -------- Scheduler.cleanup """ parent: SchedulerState = cast(SchedulerState, self) if self.status in (Status.closing, Status.closed): await self.finished() return self.status = Status.closing logger.info("Scheduler closing...") setproctitle("dask-scheduler [closing]") for preload in self.preloads: await preload.teardown() if close_workers: await self.broadcast(msg={"op": "close_gracefully"}, nanny=True) for worker in parent._workers_dv: # Report would require the worker to unregister with the # currently closing scheduler. This is not necessary and might # delay shutdown of the worker unnecessarily self.worker_send(worker, {"op": "close", "report": False}) for i in range(20): # wait a second for send signals to clear if parent._workers_dv: await asyncio.sleep(0.05) else: break await asyncio.gather( *[plugin.close() for plugin in list(self.plugins.values())] ) for pc in self.periodic_callbacks.values(): pc.stop() self.periodic_callbacks.clear() self.stop_services() for ext in parent._extensions.values(): with suppress(AttributeError): ext.teardown() logger.info("Scheduler closing all comms") futures = [] for w, comm in list(self.stream_comms.items()): if not comm.closed(): comm.send({"op": "close", "report": False}) comm.send({"op": "close-stream"}) with suppress(AttributeError): futures.append(comm.close()) for future in futures: # TODO: do all at once await future for comm in self.client_comms.values(): comm.abort() await self.rpc.close() self.status = Status.closed self.stop() await super().close() setproctitle("dask-scheduler [closed]") disable_gc_diagnosis() async def close_worker(self, comm=None, worker=None, safe=None): """Remove a worker from the cluster This both removes the worker from our local state and also sends a signal to the worker to shut down. This works regardless of whether or not the worker has a nanny process restarting it """ logger.info("Closing worker %s", worker) with log_errors(): self.log_event(worker, {"action": "close-worker"}) # FIXME: This does not handle nannies self.worker_send(worker, {"op": "close", "report": False}) await self.remove_worker(address=worker, safe=safe) ########### # Stimuli # ########### def heartbeat_worker( self, comm=None, *, address, resolve_address: bool = True, now: float = None, resources: dict = None, host_info: dict = None, metrics: dict, executing: dict = None, ): parent: SchedulerState = cast(SchedulerState, self) address = self.coerce_address(address, resolve_address) address = normalize_address(address) ws: WorkerState = parent._workers_dv.get(address) # type: ignore if ws is None: return {"status": "missing"} host = get_address_host(address) local_now = time() host_info = host_info or {} dh: dict = parent._host_info.setdefault(host, {}) dh["last-seen"] = local_now frac = 1 / len(parent._workers_dv) parent._bandwidth = ( parent._bandwidth * (1 - frac) + metrics["bandwidth"]["total"] * frac ) for other, (bw, count) in metrics["bandwidth"]["workers"].items(): if (address, other) not in self.bandwidth_workers: self.bandwidth_workers[address, other] = bw / count else: alpha = (1 - frac) ** count self.bandwidth_workers[address, other] = self.bandwidth_workers[ address, other ] * alpha + bw * (1 - alpha) for typ, (bw, count) in metrics["bandwidth"]["types"].items(): if typ not in self.bandwidth_types: self.bandwidth_types[typ] = bw / count else: alpha = (1 - frac) ** count self.bandwidth_types[typ] = self.bandwidth_types[typ] * alpha + bw * ( 1 - alpha ) ws._last_seen = local_now if executing is not None: ws._executing = { parent._tasks[key]: duration for key, duration in executing.items() if key in parent._tasks } ws._metrics = metrics # Calculate RSS - dask keys, separating "old" and "new" usage # See MemoryState for details max_memory_unmanaged_old_hist_age = local_now - parent.MEMORY_RECENT_TO_OLD_TIME memory_unmanaged_old = ws._memory_unmanaged_old while ws._memory_other_history: timestamp, size = ws._memory_other_history[0] if timestamp >= max_memory_unmanaged_old_hist_age: break ws._memory_other_history.popleft() if size == memory_unmanaged_old: memory_unmanaged_old = 0 # recalculate min() # metrics["memory"] is None if the worker sent a heartbeat before its # SystemMonitor ever had a chance to run. # ws._nbytes is updated at a different time and sizeof() may not be accurate, # so size may be (temporarily) negative; floor it to zero. size = max(0, (metrics["memory"] or 0) - ws._nbytes + metrics["spilled_nbytes"]) ws._memory_other_history.append((local_now, size)) if not memory_unmanaged_old: # The worker has just been started or the previous minimum has been expunged # because too old. # Note: this algorithm is capped to 200 * MEMORY_RECENT_TO_OLD_TIME elements # cluster-wide by heartbeat_interval(), regardless of the number of workers ws._memory_unmanaged_old = min(map(second, ws._memory_other_history)) elif size < memory_unmanaged_old: ws._memory_unmanaged_old = size if host_info: dh = parent._host_info.setdefault(host, {}) dh.update(host_info) if now: ws._time_delay = local_now - now if resources: self.add_resources(worker=address, resources=resources) self.log_event(address, merge({"action": "heartbeat"}, metrics)) return { "status": "OK", "time": local_now, "heartbeat-interval": heartbeat_interval(len(parent._workers_dv)), } async def add_worker( self, comm=None, *, address: str, status: str, keys=(), nthreads=None, name=None, resolve_address=True, nbytes=None, types=None, now=None, resources=None, host_info=None, memory_limit=None, metrics=None, pid=0, services=None, local_directory=None, versions=None, nanny=None, extra=None, ): """Add a new worker to the cluster""" parent: SchedulerState = cast(SchedulerState, self) with log_errors(): address = self.coerce_address(address, resolve_address) address = normalize_address(address) host = get_address_host(address) if address in parent._workers_dv: raise ValueError("Worker already exists %s" % address) if name in parent._aliases: logger.warning( "Worker tried to connect with a duplicate name: %s", name ) msg = { "status": "error", "message": "name taken, %s" % name, "time": time(), } if comm: await comm.write(msg) return ws: WorkerState parent._workers[address] = ws = WorkerState( address=address, status=Status.lookup[status], # type: ignore pid=pid, nthreads=nthreads, memory_limit=memory_limit or 0, name=name, local_directory=local_directory, services=services, versions=versions, nanny=nanny, extra=extra, ) if ws._status == Status.running: parent._running.add(ws) dh: dict = parent._host_info.get(host) # type: ignore if dh is None: parent._host_info[host] = dh = {} dh_addresses: set = dh.get("addresses") # type: ignore if dh_addresses is None: dh["addresses"] = dh_addresses = set() dh["nthreads"] = 0 dh_addresses.add(address) dh["nthreads"] += nthreads parent._total_nthreads += nthreads parent._aliases[name] = address self.heartbeat_worker( address=address, resolve_address=resolve_address, now=now, resources=resources, host_info=host_info, metrics=metrics, ) # Do not need to adjust parent._total_occupancy as self.occupancy[ws] cannot # exist before this. self.check_idle_saturated(ws) # for key in keys: # TODO # self.mark_key_in_memory(key, [address]) self.stream_comms[address] = BatchedSend(interval="5ms", loop=self.loop) if ws._nthreads > len(ws._processing): parent._idle[ws._address] = ws for plugin in list(self.plugins.values()): try: result = plugin.add_worker(scheduler=self, worker=address) if inspect.isawaitable(result): await result except Exception as e: logger.exception(e) recommendations: dict = {} client_msgs: dict = {} worker_msgs: dict = {} if nbytes: assert isinstance(nbytes, dict) already_released_keys = [] for key in nbytes: ts: TaskState = parent._tasks.get(key) # type: ignore if ts is not None and ts.state != "released": if ts.state == "memory": self.add_keys(worker=address, keys=[key]) else: t: tuple = parent._transition( key, "memory", worker=address, nbytes=nbytes[key], typename=types[key], ) recommendations, client_msgs, worker_msgs = t parent._transitions( recommendations, client_msgs, worker_msgs ) recommendations = {} else: already_released_keys.append(key) if already_released_keys: if address not in worker_msgs: worker_msgs[address] = [] worker_msgs[address].append( { "op": "remove-replicas", "keys": already_released_keys, "stimulus_id": f"reconnect-already-released-{time()}", } ) if ws._status == Status.running: for ts in parent._unrunnable: valid: set = self.valid_workers(ts) if valid is None or ws in valid: recommendations[ts._key] = "waiting" if recommendations: parent._transitions(recommendations, client_msgs, worker_msgs) self.send_all(client_msgs, worker_msgs) self.log_event(address, {"action": "add-worker"}) self.log_event("all", {"action": "add-worker", "worker": address}) logger.info("Register worker %s", ws) msg = { "status": "OK", "time": time(), "heartbeat-interval": heartbeat_interval(len(parent._workers_dv)), "worker-plugins": self.worker_plugins, } cs: ClientState version_warning = version_module.error_message( version_module.get_versions(), merge( {w: ws._versions for w, ws in parent._workers_dv.items()}, { c: cs._versions for c, cs in parent._clients.items() if cs._versions }, ), versions, client_name="This Worker", ) msg.update(version_warning) if comm: await comm.write(msg) await self.handle_worker(comm=comm, worker=address) async def add_nanny(self, comm): msg = { "status": "OK", "nanny-plugins": self.nanny_plugins, } return msg def update_graph_hlg( self, client=None, hlg=None, keys=None, dependencies=None, restrictions=None, priority=None, loose_restrictions=None, resources=None, submitting_task=None, retries=None, user_priority=0, actors=None, fifo_timeout=0, code=None, ): unpacked_graph = HighLevelGraph.__dask_distributed_unpack__(hlg) dsk = unpacked_graph["dsk"] dependencies = unpacked_graph["deps"] annotations = unpacked_graph["annotations"] # Remove any self-dependencies (happens on test_publish_bag() and others) for k, v in dependencies.items(): deps = set(v) if k in deps: deps.remove(k) dependencies[k] = deps if priority is None: # Removing all non-local keys before calling order() dsk_keys = set(dsk) # intersection() of sets is much faster than dict_keys stripped_deps = { k: v.intersection(dsk_keys) for k, v in dependencies.items() if k in dsk_keys } priority = dask.order.order(dsk, dependencies=stripped_deps) return self.update_graph( client, dsk, keys, dependencies, restrictions, priority, loose_restrictions, resources, submitting_task, retries, user_priority, actors, fifo_timeout, annotations, code=code, ) def update_graph( self, client=None, tasks=None, keys=None, dependencies=None, restrictions=None, priority=None, loose_restrictions=None, resources=None, submitting_task=None, retries=None, user_priority=0, actors=None, fifo_timeout=0, annotations=None, code=None, ): """ Add new computations to the internal dask graph This happens whenever the Client calls submit, map, get, or compute. """ parent: SchedulerState = cast(SchedulerState, self) start = time() fifo_timeout = parse_timedelta(fifo_timeout) keys = set(keys) if len(tasks) > 1: self.log_event( ["all", client], {"action": "update_graph", "count": len(tasks)} ) # Remove aliases for k in list(tasks): if tasks[k] is k: del tasks[k] dependencies = dependencies or {} if parent._total_occupancy > 1e-9 and parent._computations: # Still working on something. Assign new tasks to same computation computation = cast(Computation, parent._computations[-1]) else: computation = Computation() parent._computations.append(computation) if code and code not in computation._code: # add new code blocks computation._code.add(code) n = 0 while len(tasks) != n: # walk through new tasks, cancel any bad deps n = len(tasks) for k, deps in list(dependencies.items()): if any( dep not in parent._tasks and dep not in tasks for dep in deps ): # bad key logger.info("User asked for computation on lost data, %s", k) del tasks[k] del dependencies[k] if k in keys: keys.remove(k) self.report({"op": "cancelled-key", "key": k}, client=client) self.client_releases_keys(keys=[k], client=client) # Avoid computation that is already finished ts: TaskState already_in_memory = set() # tasks that are already done for k, v in dependencies.items(): if v and k in parent._tasks: ts = parent._tasks[k] if ts._state in ("memory", "erred"): already_in_memory.add(k) dts: TaskState if already_in_memory: dependents = dask.core.reverse_dict(dependencies) stack = list(already_in_memory) done = set(already_in_memory) while stack: # remove unnecessary dependencies key = stack.pop() ts = parent._tasks[key] try: deps = dependencies[key] except KeyError: deps = self.dependencies[key] for dep in deps: if dep in dependents: child_deps = dependents[dep] else: child_deps = self.dependencies[dep] if all(d in done for d in child_deps): if dep in parent._tasks and dep not in done: done.add(dep) stack.append(dep) for d in done: tasks.pop(d, None) dependencies.pop(d, None) # Get or create task states stack = list(keys) touched_keys = set() touched_tasks = [] while stack: k = stack.pop() if k in touched_keys: continue # XXX Have a method get_task_state(self, k) ? ts = parent._tasks.get(k) if ts is None: ts = parent.new_task( k, tasks.get(k), "released", computation=computation ) elif not ts._run_spec: ts._run_spec = tasks.get(k) touched_keys.add(k) touched_tasks.append(ts) stack.extend(dependencies.get(k, ())) self.client_desires_keys(keys=keys, client=client) # Add dependencies for key, deps in dependencies.items(): ts = parent._tasks.get(key) if ts is None or ts._dependencies: continue for dep in deps: dts = parent._tasks[dep] ts.add_dependency(dts) # Compute priorities if isinstance(user_priority, Number): user_priority = {k: user_priority for k in tasks} annotations = annotations or {} restrictions = restrictions or {} loose_restrictions = loose_restrictions or [] resources = resources or {} retries = retries or {} # Override existing taxonomy with per task annotations if annotations: if "priority" in annotations: user_priority.update(annotations["priority"]) if "workers" in annotations: restrictions.update(annotations["workers"]) if "allow_other_workers" in annotations: loose_restrictions.extend( k for k, v in annotations["allow_other_workers"].items() if v ) if "retries" in annotations: retries.update(annotations["retries"]) if "resources" in annotations: resources.update(annotations["resources"]) for a, kv in annotations.items(): for k, v in kv.items(): # Tasks might have been culled, in which case # we have nothing to annotate. ts = parent._tasks.get(k) if ts is not None: ts._annotations[a] = v # Add actors if actors is True: actors = list(keys) for actor in actors or []: ts = parent._tasks[actor] ts._actor = True priority = priority or dask.order.order( tasks ) # TODO: define order wrt old graph if submitting_task: # sub-tasks get better priority than parent tasks ts = parent._tasks.get(submitting_task) if ts is not None: generation = ts._priority[0] - 0.01 else: # super-task already cleaned up generation = self.generation elif self._last_time + fifo_timeout < start: self.generation += 1 # older graph generations take precedence generation = self.generation self._last_time = start else: generation = self.generation for key in set(priority) & touched_keys: ts = parent._tasks[key] if ts._priority is None: ts._priority = (-(user_priority.get(key, 0)), generation, priority[key]) # Ensure all runnables have a priority runnables = [ts for ts in touched_tasks if ts._run_spec] for ts in runnables: if ts._priority is None and ts._run_spec: ts._priority = (self.generation, 0) if restrictions: # *restrictions* is a dict keying task ids to lists of # restriction specifications (either worker names or addresses) for k, v in restrictions.items(): if v is None: continue ts = parent._tasks.get(k) if ts is None: continue ts._host_restrictions = set() ts._worker_restrictions = set() # Make sure `v` is a collection and not a single worker name / address if not isinstance(v, (list, tuple, set)): v = [v] for w in v: try: w = self.coerce_address(w) except ValueError: # Not a valid address, but perhaps it's a hostname ts._host_restrictions.add(w) else: ts._worker_restrictions.add(w) if loose_restrictions: for k in loose_restrictions: ts = parent._tasks[k] ts._loose_restrictions = True if resources: for k, v in resources.items(): if v is None: continue assert isinstance(v, dict) ts = parent._tasks.get(k) if ts is None: continue ts._resource_restrictions = v if retries: for k, v in retries.items(): assert isinstance(v, int) ts = parent._tasks.get(k) if ts is None: continue ts._retries = v # Compute recommendations recommendations: dict = {} for ts in sorted(runnables, key=operator.attrgetter("priority"), reverse=True): if ts._state == "released" and ts._run_spec: recommendations[ts._key] = "waiting" for ts in touched_tasks: for dts in ts._dependencies: if dts._exception_blame: ts._exception_blame = dts._exception_blame recommendations[ts._key] = "erred" break for plugin in list(self.plugins.values()): try: plugin.update_graph( self, client=client, tasks=tasks, keys=keys, restrictions=restrictions or {}, dependencies=dependencies, priority=priority, loose_restrictions=loose_restrictions, resources=resources, annotations=annotations, ) except Exception as e: logger.exception(e) self.transitions(recommendations) for ts in touched_tasks: if ts._state in ("memory", "erred"): self.report_on_key(ts=ts, client=client) end = time() if self.digests is not None: self.digests["update-graph-duration"].add(end - start) # TODO: balance workers def stimulus_task_finished(self, key=None, worker=None, **kwargs): """Mark that a task has finished execution on a particular worker""" parent: SchedulerState = cast(SchedulerState, self) logger.debug("Stimulus task finished %s, %s", key, worker) recommendations: dict = {} client_msgs: dict = {} worker_msgs: dict = {} ws: WorkerState = parent._workers_dv[worker] ts: TaskState = parent._tasks.get(key) if ts is None or ts._state == "released": logger.debug( "Received already computed task, worker: %s, state: %s" ", key: %s, who_has: %s", worker, ts._state if ts else "forgotten", key, ts._who_has if ts else {}, ) worker_msgs[worker] = [ { "op": "free-keys", "keys": [key], "stimulus_id": f"already-released-or-forgotten-{time()}", } ] elif ts._state == "memory": self.add_keys(worker=worker, keys=[key]) else: ts._metadata.update(kwargs["metadata"]) r: tuple = parent._transition(key, "memory", worker=worker, **kwargs) recommendations, client_msgs, worker_msgs = r if ts._state == "memory": assert ws in ts._who_has return recommendations, client_msgs, worker_msgs def stimulus_task_erred( self, key=None, worker=None, exception=None, traceback=None, **kwargs ): """Mark that a task has erred on a particular worker""" parent: SchedulerState = cast(SchedulerState, self) logger.debug("Stimulus task erred %s, %s", key, worker) ts: TaskState = parent._tasks.get(key) if ts is None or ts._state != "processing": return {}, {}, {} if ts._retries > 0: ts._retries -= 1 return parent._transition(key, "waiting") else: return parent._transition( key, "erred", cause=key, exception=exception, traceback=traceback, worker=worker, **kwargs, ) def stimulus_retry(self, comm=None, keys=None, client=None): parent: SchedulerState = cast(SchedulerState, self) logger.info("Client %s requests to retry %d keys", client, len(keys)) if client: self.log_event(client, {"action": "retry", "count": len(keys)}) stack = list(keys) seen = set() roots = [] ts: TaskState dts: TaskState while stack: key = stack.pop() seen.add(key) ts = parent._tasks[key] erred_deps = [dts._key for dts in ts._dependencies if dts._state == "erred"] if erred_deps: stack.extend(erred_deps) else: roots.append(key) recommendations: dict = {key: "waiting" for key in roots} self.transitions(recommendations) if parent._validate: for key in seen: assert not parent._tasks[key].exception_blame return tuple(seen) async def remove_worker(self, comm=None, address=None, safe=False, close=True): """ Remove worker from cluster We do this when a worker reports that it plans to leave or when it appears to be unresponsive. This may send its tasks back to a released state. """ parent: SchedulerState = cast(SchedulerState, self) with log_errors(): if self.status == Status.closed: return address = self.coerce_address(address) if address not in parent._workers_dv: return "already-removed" host = get_address_host(address) ws: WorkerState = parent._workers_dv[address] self.log_event( ["all", address], { "action": "remove-worker", "processing-tasks": dict(ws._processing), }, ) logger.info("Remove worker %s", ws) if close: with suppress(AttributeError, CommClosedError): self.stream_comms[address].send({"op": "close", "report": False}) self.remove_resources(address) dh: dict = parent._host_info.get(host) if dh is None: parent._host_info[host] = dh = {} dh_addresses: set = dh["addresses"] dh_addresses.remove(address) dh["nthreads"] -= ws._nthreads parent._total_nthreads -= ws._nthreads if not dh_addresses: dh = None dh_addresses = None del parent._host_info[host] self.rpc.remove(address) del self.stream_comms[address] del parent._aliases[ws._name] parent._idle.pop(ws._address, None) parent._saturated.discard(ws) del parent._workers[address] ws.status = Status.closed parent._running.discard(ws) parent._total_occupancy -= ws._occupancy recommendations: dict = {} ts: TaskState for ts in list(ws._processing): k = ts._key recommendations[k] = "released" if not safe: ts._suspicious += 1 ts._prefix._suspicious += 1 if ts._suspicious > self.allowed_failures: del recommendations[k] e = pickle.dumps( KilledWorker(task=k, last_worker=ws.clean()), protocol=4 ) r = self.transition(k, "erred", exception=e, cause=k) recommendations.update(r) logger.info( "Task %s marked as failed because %d workers died" " while trying to run it", ts._key, self.allowed_failures, ) for ts in list(ws._has_what): parent.remove_replica(ts, ws) if not ts._who_has: if ts._run_spec: recommendations[ts._key] = "released" else: # pure data recommendations[ts._key] = "forgotten" self.transitions(recommendations) for plugin in list(self.plugins.values()): try: result = plugin.remove_worker(scheduler=self, worker=address) if inspect.isawaitable(result): await result except Exception as e: logger.exception(e) if not parent._workers_dv: logger.info("Lost all workers") for w in parent._workers_dv: self.bandwidth_workers.pop((address, w), None) self.bandwidth_workers.pop((w, address), None) def remove_worker_from_events(): # If the worker isn't registered anymore after the delay, remove from events if address not in parent._workers_dv and address in self.events: del self.events[address] cleanup_delay = parse_timedelta( dask.config.get("distributed.scheduler.events-cleanup-delay") ) self.loop.call_later(cleanup_delay, remove_worker_from_events) logger.debug("Removed worker %s", ws) return "OK" def stimulus_cancel(self, comm, keys=None, client=None, force=False): """Stop execution on a list of keys""" logger.info("Client %s requests to cancel %d keys", client, len(keys)) if client: self.log_event( client, {"action": "cancel", "count": len(keys), "force": force} ) for key in keys: self.cancel_key(key, client, force=force) def cancel_key(self, key, client, retries=5, force=False): """Cancel a particular key and all dependents""" # TODO: this should be converted to use the transition mechanism parent: SchedulerState = cast(SchedulerState, self) ts: TaskState = parent._tasks.get(key) dts: TaskState try: cs: ClientState = parent._clients[client] except KeyError: return if ts is None or not ts._who_wants: # no key yet, lets try again in a moment if retries: self.loop.call_later( 0.2, lambda: self.cancel_key(key, client, retries - 1) ) return if force or ts._who_wants == {cs}: # no one else wants this key for dts in list(ts._dependents): self.cancel_key(dts._key, client, force=force) logger.info("Scheduler cancels key %s. Force=%s", key, force) self.report({"op": "cancelled-key", "key": key}) clients = list(ts._who_wants) if force else [cs] for cs in clients: self.client_releases_keys(keys=[key], client=cs._client_key) def client_desires_keys(self, keys=None, client=None): parent: SchedulerState = cast(SchedulerState, self) cs: ClientState = parent._clients.get(client) if cs is None: # For publish, queues etc. parent._clients[client] = cs = ClientState(client) ts: TaskState for k in keys: ts = parent._tasks.get(k) if ts is None: # For publish, queues etc. ts = parent.new_task(k, None, "released") ts._who_wants.add(cs) cs._wants_what.add(ts) if ts._state in ("memory", "erred"): self.report_on_key(ts=ts, client=client) def client_releases_keys(self, keys=None, client=None): """Remove keys from client desired list""" parent: SchedulerState = cast(SchedulerState, self) if not isinstance(keys, list): keys = list(keys) cs: ClientState = parent._clients[client] recommendations: dict = {} _client_releases_keys(parent, keys=keys, cs=cs, recommendations=recommendations) self.transitions(recommendations) def client_heartbeat(self, client=None): """Handle heartbeats from Client""" parent: SchedulerState = cast(SchedulerState, self) cs: ClientState = parent._clients[client] cs._last_seen = time() ################### # Task Validation # ################### def validate_released(self, key): parent: SchedulerState = cast(SchedulerState, self) ts: TaskState = parent._tasks[key] dts: TaskState assert ts._state == "released" assert not ts._waiters assert not ts._waiting_on assert not ts._who_has assert not ts._processing_on assert not any([ts in dts._waiters for dts in ts._dependencies]) assert ts not in parent._unrunnable def validate_waiting(self, key): parent: SchedulerState = cast(SchedulerState, self) ts: TaskState = parent._tasks[key] dts: TaskState assert ts._waiting_on assert not ts._who_has assert not ts._processing_on assert ts not in parent._unrunnable for dts in ts._dependencies: # We are waiting on a dependency iff it's not stored assert bool(dts._who_has) != (dts in ts._waiting_on) assert ts in dts._waiters # XXX even if dts._who_has? def validate_processing(self, key): parent: SchedulerState = cast(SchedulerState, self) ts: TaskState = parent._tasks[key] dts: TaskState assert not ts._waiting_on ws: WorkerState = ts._processing_on assert ws assert ts in ws._processing assert not ts._who_has for dts in ts._dependencies: assert dts._who_has assert ts in dts._waiters def validate_memory(self, key): parent: SchedulerState = cast(SchedulerState, self) ts: TaskState = parent._tasks[key] dts: TaskState assert ts._who_has assert bool(ts in parent._replicated_tasks) == (len(ts._who_has) > 1) assert not ts._processing_on assert not ts._waiting_on assert ts not in parent._unrunnable for dts in ts._dependents: assert (dts in ts._waiters) == (dts._state in ("waiting", "processing")) assert ts not in dts._waiting_on def validate_no_worker(self, key): parent: SchedulerState = cast(SchedulerState, self) ts: TaskState = parent._tasks[key] dts: TaskState assert ts in parent._unrunnable assert not ts._waiting_on assert ts in parent._unrunnable assert not ts._processing_on assert not ts._who_has for dts in ts._dependencies: assert dts._who_has def validate_erred(self, key): parent: SchedulerState = cast(SchedulerState, self) ts: TaskState = parent._tasks[key] assert ts._exception_blame assert not ts._who_has def validate_key(self, key, ts: TaskState = None): parent: SchedulerState = cast(SchedulerState, self) try: if ts is None: ts = parent._tasks.get(key) if ts is None: logger.debug("Key lost: %s", key) else: ts.validate() try: func = getattr(self, "validate_" + ts._state.replace("-", "_")) except AttributeError: logger.error( "self.validate_%s not found", ts._state.replace("-", "_") ) else: func(key) except Exception as e: logger.exception(e) if LOG_PDB: import pdb pdb.set_trace() raise def validate_state(self, allow_overlap=False): parent: SchedulerState = cast(SchedulerState, self) validate_state(parent._tasks, parent._workers, parent._clients) if not (set(parent._workers_dv) == set(self.stream_comms)): raise ValueError("Workers not the same in all collections") ws: WorkerState for w, ws in parent._workers_dv.items(): assert isinstance(w, str), (type(w), w) assert isinstance(ws, WorkerState), (type(ws), ws) assert ws._address == w if not ws._processing: assert not ws._occupancy assert ws._address in parent._idle_dv assert (ws._status == Status.running) == (ws in parent._running) for ws in parent._running: assert ws._status == Status.running assert ws._address in parent._workers_dv ts: TaskState for k, ts in parent._tasks.items(): assert isinstance(ts, TaskState), (type(ts), ts) assert ts._key == k assert bool(ts in parent._replicated_tasks) == (len(ts._who_has) > 1) self.validate_key(k, ts) for ts in parent._replicated_tasks: assert ts._state == "memory" assert ts._key in parent._tasks c: str cs: ClientState for c, cs in parent._clients.items(): # client=None is often used in tests... assert c is None or type(c) == str, (type(c), c) assert type(cs) == ClientState, (type(cs), cs) assert cs._client_key == c a = {w: ws._nbytes for w, ws in parent._workers_dv.items()} b = { w: sum(ts.get_nbytes() for ts in ws._has_what) for w, ws in parent._workers_dv.items() } assert a == b, (a, b) actual_total_occupancy = 0 for worker, ws in parent._workers_dv.items(): assert abs(sum(ws._processing.values()) - ws._occupancy) < 1e-8 actual_total_occupancy += ws._occupancy assert abs(actual_total_occupancy - parent._total_occupancy) < 1e-8, ( actual_total_occupancy, parent._total_occupancy, ) ################### # Manage Messages # ################### def report(self, msg: dict, ts: TaskState = None, client: str = None): """ Publish updates to all listening Queues and Comms If the message contains a key then we only send the message to those comms that care about the key. """ parent: SchedulerState = cast(SchedulerState, self) if ts is None: msg_key = msg.get("key") if msg_key is not None: tasks: dict = parent._tasks ts = tasks.get(msg_key) cs: ClientState client_comms: dict = self.client_comms client_keys: list if ts is None: # Notify all clients client_keys = list(client_comms) elif client is None: # Notify clients interested in key client_keys = [cs._client_key for cs in ts._who_wants] else: # Notify clients interested in key (including `client`) client_keys = [ cs._client_key for cs in ts._who_wants if cs._client_key != client ] client_keys.append(client) k: str for k in client_keys: c = client_comms.get(k) if c is None: continue try: c.send(msg) # logger.debug("Scheduler sends message to client %s", msg) except CommClosedError: if self.status == Status.running: logger.critical( "Closed comm %r while trying to write %s", c, msg, exc_info=True ) async def add_client(self, comm, client=None, versions=None): """Add client to network We listen to all future messages from this Comm. """ parent: SchedulerState = cast(SchedulerState, self) assert client is not None comm.name = "Scheduler->Client" logger.info("Receive client connection: %s", client) self.log_event(["all", client], {"action": "add-client", "client": client}) parent._clients[client] = ClientState(client, versions=versions) for plugin in list(self.plugins.values()): try: plugin.add_client(scheduler=self, client=client) except Exception as e: logger.exception(e) try: bcomm = BatchedSend(interval="2ms", loop=self.loop) bcomm.start(comm) self.client_comms[client] = bcomm msg = {"op": "stream-start"} ws: WorkerState version_warning = version_module.error_message( version_module.get_versions(), {w: ws._versions for w, ws in parent._workers_dv.items()}, versions, ) msg.update(version_warning) bcomm.send(msg) try: await self.handle_stream(comm=comm, extra={"client": client}) finally: self.remove_client(client=client) logger.debug("Finished handling client %s", client) finally: if not comm.closed(): self.client_comms[client].send({"op": "stream-closed"}) try: if not sys.is_finalizing(): await self.client_comms[client].close() del self.client_comms[client] if self.status == Status.running: logger.info("Close client connection: %s", client) except TypeError: # comm becomes None during GC pass def remove_client(self, client=None): """Remove client from network""" parent: SchedulerState = cast(SchedulerState, self) if self.status == Status.running: logger.info("Remove client %s", client) self.log_event(["all", client], {"action": "remove-client", "client": client}) try: cs: ClientState = parent._clients[client] except KeyError: # XXX is this a legitimate condition? pass else: ts: TaskState self.client_releases_keys( keys=[ts._key for ts in cs._wants_what], client=cs._client_key ) del parent._clients[client] for plugin in list(self.plugins.values()): try: plugin.remove_client(scheduler=self, client=client) except Exception as e: logger.exception(e) def remove_client_from_events(): # If the client isn't registered anymore after the delay, remove from events if client not in parent._clients and client in self.events: del self.events[client] cleanup_delay = parse_timedelta( dask.config.get("distributed.scheduler.events-cleanup-delay") ) self.loop.call_later(cleanup_delay, remove_client_from_events) def send_task_to_worker(self, worker, ts: TaskState, duration: double = -1): """Send a single computational task to a worker""" parent: SchedulerState = cast(SchedulerState, self) try: msg: dict = _task_to_msg(parent, ts, duration) self.worker_send(worker, msg) except Exception as e: logger.exception(e) if LOG_PDB: import pdb pdb.set_trace() raise def handle_uncaught_error(self, **msg): logger.exception(clean_exception(**msg)[1]) def handle_task_finished(self, key=None, worker=None, **msg): parent: SchedulerState = cast(SchedulerState, self) if worker not in parent._workers_dv: return validate_key(key) recommendations: dict client_msgs: dict worker_msgs: dict r: tuple = self.stimulus_task_finished(key=key, worker=worker, **msg) recommendations, client_msgs, worker_msgs = r parent._transitions(recommendations, client_msgs, worker_msgs) self.send_all(client_msgs, worker_msgs) def handle_task_erred(self, key=None, **msg): parent: SchedulerState = cast(SchedulerState, self) recommendations: dict client_msgs: dict worker_msgs: dict r: tuple = self.stimulus_task_erred(key=key, **msg) recommendations, client_msgs, worker_msgs = r parent._transitions(recommendations, client_msgs, worker_msgs) self.send_all(client_msgs, worker_msgs) def handle_missing_data(self, key=None, errant_worker=None, **kwargs): parent: SchedulerState = cast(SchedulerState, self) logger.debug("handle missing data key=%s worker=%s", key, errant_worker) self.log.append(("missing", key, errant_worker)) ts: TaskState = parent._tasks.get(key) if ts is None: return ws: WorkerState = parent._workers_dv.get(errant_worker) if ws is not None and ws in ts._who_has: parent.remove_replica(ts, ws) if not ts._who_has: if ts._run_spec: self.transitions({key: "released"}) else: self.transitions({key: "forgotten"}) def release_worker_data(self, comm=None, key=None, worker=None): parent: SchedulerState = cast(SchedulerState, self) ws: WorkerState = parent._workers_dv.get(worker) ts: TaskState = parent._tasks.get(key) if not ws or not ts: return recommendations: dict = {} if ws in ts._who_has: parent.remove_replica(ts, ws) if not ts._who_has: recommendations[ts._key] = "released" if recommendations: self.transitions(recommendations) def handle_long_running(self, key=None, worker=None, compute_duration=None): """A task has seceded from the thread pool We stop the task from being stolen in the future, and change task duration accounting as if the task has stopped. """ parent: SchedulerState = cast(SchedulerState, self) if key not in parent._tasks: logger.debug("Skipping long_running since key %s was already released", key) return ts: TaskState = parent._tasks[key] steal = parent._extensions.get("stealing") if steal is not None: steal.remove_key_from_stealable(ts) ws: WorkerState = ts._processing_on if ws is None: logger.debug("Received long-running signal from duplicate task. Ignoring.") return if compute_duration: old_duration: double = ts._prefix._duration_average new_duration: double = compute_duration avg_duration: double if old_duration < 0: avg_duration = new_duration else: avg_duration = 0.5 * old_duration + 0.5 * new_duration ts._prefix._duration_average = avg_duration occ: double = ws._processing[ts] ws._occupancy -= occ parent._total_occupancy -= occ ws._processing[ts] = 0 self.check_idle_saturated(ws) def handle_worker_status_change(self, status: str, worker: str) -> None: parent: SchedulerState = cast(SchedulerState, self) ws: WorkerState = parent._workers_dv.get(worker) # type: ignore if not ws: return prev_status = ws._status ws._status = Status.lookup[status] # type: ignore if ws._status == prev_status: return self.log_event( ws._address, { "action": "worker-status-change", "prev-status": prev_status.name, "status": status, }, ) if ws._status == Status.running: parent._running.add(ws) recs = {} ts: TaskState for ts in parent._unrunnable: valid: set = self.valid_workers(ts) if valid is None or ws in valid: recs[ts._key] = "waiting" if recs: client_msgs: dict = {} worker_msgs: dict = {} parent._transitions(recs, client_msgs, worker_msgs) self.send_all(client_msgs, worker_msgs) else: parent._running.discard(ws) async def handle_worker(self, comm=None, worker=None): """ Listen to responses from a single worker This is the main loop for scheduler-worker interaction See Also -------- Scheduler.handle_client: Equivalent coroutine for clients """ comm.name = "Scheduler connection to worker" worker_comm = self.stream_comms[worker] worker_comm.start(comm) logger.info("Starting worker compute stream, %s", worker) try: await self.handle_stream(comm=comm, extra={"worker": worker}) finally: if worker in self.stream_comms: worker_comm.abort() await self.remove_worker(address=worker) def add_plugin( self, plugin: SchedulerPlugin, *, idempotent: bool = False, name: "str | None" = None, **kwargs, ): """Add external plugin to scheduler. See https://distributed.readthedocs.io/en/latest/plugins.html Parameters ---------- plugin : SchedulerPlugin SchedulerPlugin instance to add idempotent : bool If true, the plugin is assumed to already exist and no action is taken. name : str A name for the plugin, if None, the name attribute is checked on the Plugin instance and generated if not discovered. **kwargs Deprecated; additional arguments passed to the `plugin` class if it is not already an instance """ if isinstance(plugin, type): warnings.warn( "Adding plugins by class is deprecated and will be disabled in a " "future release. Please add plugins by instance instead.", category=FutureWarning, ) plugin = plugin(self, **kwargs) # type: ignore elif kwargs: raise ValueError("kwargs provided but plugin is already an instance") if name is None: name = _get_plugin_name(plugin) if name in self.plugins: if idempotent: return warnings.warn( f"Scheduler already contains a plugin with name {name}; overwriting.", category=UserWarning, ) self.plugins[name] = plugin def remove_plugin( self, name: "str | None" = None, plugin: "SchedulerPlugin | None" = None, ) -> None: """Remove external plugin from scheduler Parameters ---------- name : str Name of the plugin to remove plugin : SchedulerPlugin Deprecated; use `name` argument instead. Instance of a SchedulerPlugin class to remove; """ # TODO: Remove this block of code once removing plugins by value is disabled if bool(name) == bool(plugin): raise ValueError("Must provide plugin or name (mutually exclusive)") if isinstance(name, SchedulerPlugin): # Backwards compatibility - the sig used to be (plugin, name) plugin = name name = None if plugin is not None: warnings.warn( "Removing scheduler plugins by value is deprecated and will be disabled " "in a future release. Please remove scheduler plugins by name instead.", category=FutureWarning, ) if hasattr(plugin, "name"): name = plugin.name # type: ignore else: names = [k for k, v in self.plugins.items() if v is plugin] if not names: raise ValueError( f"Could not find {plugin} among the current scheduler plugins" ) if len(names) > 1: raise ValueError( f"Multiple instances of {plugin} were found in the current " "scheduler plugins; we cannot remove this plugin." ) name = names[0] assert name is not None # End deprecated code try: del self.plugins[name] except KeyError: raise ValueError( f"Could not find plugin {name!r} among the current scheduler plugins" ) async def register_scheduler_plugin(self, comm=None, plugin=None, name=None): """Register a plugin on the scheduler.""" if not dask.config.get("distributed.scheduler.pickle"): raise ValueError( "Cannot register a scheduler plugin as the scheduler " "has been explicitly disallowed from deserializing " "arbitrary bytestrings using pickle via the " "'distributed.scheduler.pickle' configuration setting." ) plugin = loads(plugin) if hasattr(plugin, "start"): result = plugin.start(self) if inspect.isawaitable(result): await result self.add_plugin(plugin, name=name) def worker_send(self, worker, msg): """Send message to worker This also handles connection failures by adding a callback to remove the worker on the next cycle. """ stream_comms: dict = self.stream_comms try: stream_comms[worker].send(msg) except (CommClosedError, AttributeError): self.loop.add_callback(self.remove_worker, address=worker) def client_send(self, client, msg): """Send message to client""" client_comms: dict = self.client_comms c = client_comms.get(client) if c is None: return try: c.send(msg) except CommClosedError: if self.status == Status.running: logger.critical( "Closed comm %r while trying to write %s", c, msg, exc_info=True ) def send_all(self, client_msgs: dict, worker_msgs: dict): """Send messages to client and workers""" client_comms: dict = self.client_comms stream_comms: dict = self.stream_comms msgs: list for client, msgs in client_msgs.items(): c = client_comms.get(client) if c is None: continue try: c.send(*msgs) except CommClosedError: if self.status == Status.running: logger.critical( "Closed comm %r while trying to write %s", c, msgs, exc_info=True, ) for worker, msgs in worker_msgs.items(): try: w = stream_comms[worker] w.send(*msgs) except KeyError: # worker already gone pass except (CommClosedError, AttributeError): self.loop.add_callback(self.remove_worker, address=worker) ############################ # Less common interactions # ############################ async def scatter( self, comm=None, data=None, workers=None, client=None, broadcast=False, timeout=2, ): """Send data out to workers See also -------- Scheduler.broadcast: """ parent: SchedulerState = cast(SchedulerState, self) ws: WorkerState start = time() while True: if workers is None: wss = parent._running else: workers = [self.coerce_address(w) for w in workers] wss = {parent._workers_dv[w] for w in workers} wss = {ws for ws in wss if ws._status == Status.running} if wss: break if time() > start + timeout: raise TimeoutError("No valid workers found") await asyncio.sleep(0.1) nthreads = {ws._address: ws.nthreads for ws in wss} assert isinstance(data, dict) keys, who_has, nbytes = await scatter_to_workers( nthreads, data, rpc=self.rpc, report=False ) self.update_data(who_has=who_has, nbytes=nbytes, client=client) if broadcast: n = len(nthreads) if broadcast is True else broadcast await self.replicate(keys=keys, workers=workers, n=n) self.log_event( [client, "all"], {"action": "scatter", "client": client, "count": len(data)} ) return keys async def gather(self, comm=None, keys=None, serializers=None): """Collect data from workers to the scheduler""" parent: SchedulerState = cast(SchedulerState, self) ws: WorkerState keys = list(keys) who_has = {} for key in keys: ts: TaskState = parent._tasks.get(key) if ts is not None: who_has[key] = [ws._address for ws in ts._who_has] else: who_has[key] = [] data, missing_keys, missing_workers = await gather_from_workers( who_has, rpc=self.rpc, close=False, serializers=serializers ) if not missing_keys: result = {"status": "OK", "data": data} else: missing_states = [ (parent._tasks[key].state if key in parent._tasks else None) for key in missing_keys ] logger.exception( "Couldn't gather keys %s state: %s workers: %s", missing_keys, missing_states, missing_workers, ) result = {"status": "error", "keys": missing_keys} with log_errors(): # Remove suspicious workers from the scheduler but allow them to # reconnect. await asyncio.gather( *( self.remove_worker(address=worker, close=False) for worker in missing_workers ) ) recommendations: dict client_msgs: dict = {} worker_msgs: dict = {} for key, workers in missing_keys.items(): # Task may already be gone if it was held by a # `missing_worker` ts: TaskState = parent._tasks.get(key) logger.exception( "Workers don't have promised key: %s, %s", str(workers), str(key), ) if not workers or ts is None: continue recommendations: dict = {key: "released"} for worker in workers: ws = parent._workers_dv.get(worker) if ws is not None and ws in ts._who_has: parent.remove_replica(ts, ws) parent._transitions( recommendations, client_msgs, worker_msgs ) self.send_all(client_msgs, worker_msgs) self.log_event("all", {"action": "gather", "count": len(keys)}) return result def clear_task_state(self): # XXX what about nested state such as ClientState.wants_what # (see also fire-and-forget...) logger.info("Clear task state") for collection in self._task_state_collections: collection.clear() async def restart(self, client=None, timeout=30): """Restart all workers. Reset local state.""" parent: SchedulerState = cast(SchedulerState, self) with log_errors(): n_workers = len(parent._workers_dv) logger.info("Send lost future signal to clients") cs: ClientState ts: TaskState for cs in parent._clients.values(): self.client_releases_keys( keys=[ts._key for ts in cs._wants_what], client=cs._client_key ) ws: WorkerState nannies = {addr: ws._nanny for addr, ws in parent._workers_dv.items()} for addr in list(parent._workers_dv): try: # Ask the worker to close if it doesn't have a nanny, # otherwise the nanny will kill it anyway await self.remove_worker(address=addr, close=addr not in nannies) except Exception: logger.info( "Exception while restarting. This is normal", exc_info=True ) self.clear_task_state() for plugin in list(self.plugins.values()): try: plugin.restart(self) except Exception as e: logger.exception(e) logger.debug("Send kill signal to nannies: %s", nannies) nannies = [ rpc(nanny_address, connection_args=self.connection_args) for nanny_address in nannies.values() if nanny_address is not None ] resps = All( [ nanny.restart( close=True, timeout=timeout * 0.8, executor_wait=False ) for nanny in nannies ] ) try: resps = await asyncio.wait_for(resps, timeout) except TimeoutError: logger.error( "Nannies didn't report back restarted within " "timeout. Continuuing with restart process" ) else: if not all(resp == "OK" for resp in resps): logger.error( "Not all workers responded positively: %s", resps, exc_info=True ) finally: await asyncio.gather(*[nanny.close_rpc() for nanny in nannies]) self.clear_task_state() with suppress(AttributeError): for c in self._worker_coroutines: c.cancel() self.log_event([client, "all"], {"action": "restart", "client": client}) start = time() while time() < start + 10 and len(parent._workers_dv) < n_workers: await asyncio.sleep(0.01) self.report({"op": "restart"}) async def broadcast( self, comm=None, msg=None, workers=None, hosts=None, nanny=False, serializers=None, ): """Broadcast message to workers, return all results""" parent: SchedulerState = cast(SchedulerState, self) if workers is None or workers is True: if hosts is None: workers = list(parent._workers_dv) else: workers = [] if hosts is not None: for host in hosts: dh: dict = parent._host_info.get(host) if dh is not None: workers.extend(dh["addresses"]) # TODO replace with worker_list if nanny: addresses = [parent._workers_dv[w].nanny for w in workers] else: addresses = workers async def send_message(addr): comm = await self.rpc.connect(addr) comm.name = "Scheduler Broadcast" try: resp = await send_recv(comm, close=True, serializers=serializers, **msg) finally: self.rpc.reuse(addr, comm) return resp results = await All( [send_message(address) for address in addresses if address is not None] ) return dict(zip(workers, results)) async def proxy(self, comm=None, msg=None, worker=None, serializers=None): """Proxy a communication through the scheduler to some other worker""" d = await self.broadcast( comm=comm, msg=msg, workers=[worker], serializers=serializers ) return d[worker] async def gather_on_worker( self, worker_address: str, who_has: "dict[str, list[str]]" ) -> set: """Peer-to-peer copy of keys from multiple workers to a single worker Parameters ---------- worker_address: str Recipient worker address to copy keys to who_has: dict[Hashable, list[str]] {key: [sender address, sender address, ...], key: ...} Returns ------- returns: set of keys that failed to be copied """ try: result = await retry_operation( self.rpc(addr=worker_address).gather, who_has=who_has ) except OSError as e: # This can happen e.g. if the worker is going through controlled shutdown; # it doesn't necessarily mean that it went unexpectedly missing logger.warning( f"Communication with worker {worker_address} failed during " f"replication: {e.__class__.__name__}: {e}" ) return set(who_has) parent: SchedulerState = cast(SchedulerState, self) ws: WorkerState = parent._workers_dv.get(worker_address) # type: ignore if ws is None: logger.warning(f"Worker {worker_address} lost during replication") return set(who_has) elif result["status"] == "OK": keys_failed = set() keys_ok: Set = who_has.keys() elif result["status"] == "partial-fail": keys_failed = set(result["keys"]) keys_ok = who_has.keys() - keys_failed logger.warning( f"Worker {worker_address} failed to acquire keys: {result['keys']}" ) else: # pragma: nocover raise ValueError(f"Unexpected message from {worker_address}: {result}") for key in keys_ok: ts: TaskState = parent._tasks.get(key) # type: ignore if ts is None or ts._state != "memory": logger.warning(f"Key lost during replication: {key}") continue if ws not in ts._who_has: parent.add_replica(ts, ws) return keys_failed async def delete_worker_data( self, worker_address: str, keys: "Collection[str]" ) -> None: """Delete data from a worker and update the corresponding worker/task states Parameters ---------- worker_address: str Worker address to delete keys from keys: list[str] List of keys to delete on the specified worker """ parent: SchedulerState = cast(SchedulerState, self) try: await retry_operation( self.rpc(addr=worker_address).free_keys, keys=list(keys), stimulus_id=f"delete-data-{time()}", ) except OSError as e: # This can happen e.g. if the worker is going through controlled shutdown; # it doesn't necessarily mean that it went unexpectedly missing logger.warning( f"Communication with worker {worker_address} failed during " f"replication: {e.__class__.__name__}: {e}" ) return ws: WorkerState = parent._workers_dv.get(worker_address) # type: ignore if ws is None: return for key in keys: ts: TaskState = parent._tasks.get(key) # type: ignore if ts is not None and ws in ts._who_has: assert ts._state == "memory" parent.remove_replica(ts, ws) if not ts._who_has: # Last copy deleted self.transitions({key: "released"}) self.log_event(ws._address, {"action": "remove-worker-data", "keys": keys}) async def rebalance( self, comm=None, keys: "Iterable[Hashable]" = None, workers: "Iterable[str]" = None, ) -> dict: """Rebalance keys so that each worker ends up with roughly the same process memory (managed+unmanaged). .. warning:: This operation is generally not well tested against normal operation of the scheduler. It is not recommended to use it while waiting on computations. **Algorithm** #. Find the mean occupancy of the cluster, defined as data managed by dask + unmanaged process memory that has been there for at least 30 seconds (``distributed.worker.memory.recent-to-old-time``). This lets us ignore temporary spikes caused by task heap usage. Alternatively, you may change how memory is measured both for the individual workers as well as to calculate the mean through ``distributed.worker.memory.rebalance.measure``. Namely, this can be useful to disregard inaccurate OS memory measurements. #. Discard workers whose occupancy is within 5% of the mean cluster occupancy (``distributed.worker.memory.rebalance.sender-recipient-gap`` / 2). This helps avoid data from bouncing around the cluster repeatedly. #. Workers above the mean are senders; those below are recipients. #. Discard senders whose absolute occupancy is below 30% (``distributed.worker.memory.rebalance.sender-min``). In other words, no data is moved regardless of imbalancing as long as all workers are below 30%. #. Discard recipients whose absolute occupancy is above 60% (``distributed.worker.memory.rebalance.recipient-max``). Note that this threshold by default is the same as ``distributed.worker.memory.target`` to prevent workers from accepting data and immediately spilling it out to disk. #. Iteratively pick the sender and recipient that are farthest from the mean and move the *least recently inserted* key between the two, until either all senders or all recipients fall within 5% of the mean. A recipient will be skipped if it already has a copy of the data. In other words, this method does not degrade replication. A key will be skipped if there are no recipients available with enough memory to accept the key and that don't already hold a copy. The least recently insertd (LRI) policy is a greedy choice with the advantage of being O(1), trivial to implement (it relies on python dict insertion-sorting) and hopefully good enough in most cases. Discarded alternative policies were: - Largest first. O(n*log(n)) save for non-trivial additional data structures and risks causing the largest chunks of data to repeatedly move around the cluster like pinballs. - Least recently used (LRU). This information is currently available on the workers only and not trivial to replicate on the scheduler; transmitting it over the network would be very expensive. Also, note that dask will go out of its way to minimise the amount of time intermediate keys are held in memory, so in such a case LRI is a close approximation of LRU. Parameters ---------- keys: optional whitelist of dask keys that should be considered for moving. All other keys will be ignored. Note that this offers no guarantee that a key will actually be moved (e.g. because it is unnecessary or because there are no viable recipient workers for it). workers: optional whitelist of workers addresses to be considered as senders or recipients. All other workers will be ignored. The mean cluster occupancy will be calculated only using the whitelisted workers. """ parent: SchedulerState = cast(SchedulerState, self) with log_errors(): wss: "Collection[WorkerState]" if workers is not None: wss = [parent._workers_dv[w] for w in workers] else: wss = parent._workers_dv.values() if not wss: return {"status": "OK"} if keys is not None: if not isinstance(keys, Set): keys = set(keys) # unless already a set-like if not keys: return {"status": "OK"} missing_data = [ k for k in keys if k not in parent._tasks or not parent._tasks[k].who_has ] if missing_data: return {"status": "partial-fail", "keys": missing_data} msgs = self._rebalance_find_msgs(keys, wss) if not msgs: return {"status": "OK"} async with self._lock: result = await self._rebalance_move_data(msgs) if result["status"] == "partial-fail" and keys is None: # Only return failed keys if the client explicitly asked for them result = {"status": "OK"} return result def _rebalance_find_msgs( self, keys: "Set[Hashable] | None", workers: "Iterable[WorkerState]", ) -> "list[tuple[WorkerState, WorkerState, TaskState]]": """Identify workers that need to lose keys and those that can receive them, together with how many bytes each needs to lose/receive. Then, pair a sender worker with a recipient worker for each key, until the cluster is rebalanced. This method only defines the work to be performed; it does not start any network transfers itself. The big-O complexity is O(wt + ke*log(we)), where - wt is the total number of workers on the cluster (or the number of whitelisted workers, if explicitly stated by the user) - we is the number of workers that are eligible to be senders or recipients - kt is the total number of keys on the cluster (or on the whitelisted workers) - ke is the number of keys that need to be moved in order to achieve a balanced cluster There is a degenerate edge case O(wt + kt*log(we)) when kt is much greater than the number of whitelisted keys, or when most keys are replicated or cannot be moved for some other reason. Returns list of tuples to feed into _rebalance_move_data: - sender worker - recipient worker - task to be transferred """ parent: SchedulerState = cast(SchedulerState, self) ts: TaskState ws: WorkerState # Heaps of workers, managed by the heapq module, that need to send/receive data, # with how many bytes each needs to send/receive. # # Each element of the heap is a tuple constructed as follows: # - snd_bytes_max/rec_bytes_max: maximum number of bytes to send or receive. # This number is negative, so that the workers farthest from the cluster mean # are at the top of the smallest-first heaps. # - snd_bytes_min/rec_bytes_min: minimum number of bytes after sending/receiving # which the worker should not be considered anymore. This is also negative. # - arbitrary unique number, there just to to make sure that WorkerState objects # are never used for sorting in the unlikely event that two processes have # exactly the same number of bytes allocated. # - WorkerState # - iterator of all tasks in memory on the worker (senders only), insertion # sorted (least recently inserted first). # Note that this iterator will typically *not* be exhausted. It will only be # exhausted if, after moving away from the worker all keys that can be moved, # is insufficient to drop snd_bytes_min above 0. senders: "list[tuple[int, int, int, WorkerState, Iterator[TaskState]]]" = [] recipients: "list[tuple[int, int, int, WorkerState]]" = [] # Output: [(sender, recipient, task), ...] msgs: "list[tuple[WorkerState, WorkerState, TaskState]]" = [] # By default, this is the optimistic memory, meaning total process memory minus # unmanaged memory that appeared over the last 30 seconds # (distributed.worker.memory.recent-to-old-time). # This lets us ignore temporary spikes caused by task heap usage. memory_by_worker = [ (ws, getattr(ws.memory, parent.MEMORY_REBALANCE_MEASURE)) for ws in workers ] mean_memory = sum(m for _, m in memory_by_worker) // len(memory_by_worker) for ws, ws_memory in memory_by_worker: if ws.memory_limit: half_gap = int(parent.MEMORY_REBALANCE_HALF_GAP * ws.memory_limit) sender_min = parent.MEMORY_REBALANCE_SENDER_MIN * ws.memory_limit recipient_max = parent.MEMORY_REBALANCE_RECIPIENT_MAX * ws.memory_limit else: half_gap = 0 sender_min = 0.0 recipient_max = math.inf if ( ws._has_what and ws_memory >= mean_memory + half_gap and ws_memory >= sender_min ): # This may send the worker below sender_min (by design) snd_bytes_max = mean_memory - ws_memory # negative snd_bytes_min = snd_bytes_max + half_gap # negative # See definition of senders above senders.append( (snd_bytes_max, snd_bytes_min, id(ws), ws, iter(ws._has_what)) ) elif ws_memory < mean_memory - half_gap and ws_memory < recipient_max: # This may send the worker above recipient_max (by design) rec_bytes_max = ws_memory - mean_memory # negative rec_bytes_min = rec_bytes_max + half_gap # negative # See definition of recipients above recipients.append((rec_bytes_max, rec_bytes_min, id(ws), ws)) # Fast exit in case no transfers are necessary or possible if not senders or not recipients: self.log_event( "all", { "action": "rebalance", "senders": len(senders), "recipients": len(recipients), "moved_keys": 0, }, ) return [] heapq.heapify(senders) heapq.heapify(recipients) snd_ws: WorkerState rec_ws: WorkerState while senders and recipients: snd_bytes_max, snd_bytes_min, _, snd_ws, ts_iter = senders[0] # Iterate through tasks in memory, least recently inserted first for ts in ts_iter: if keys is not None and ts.key not in keys: continue nbytes = ts.nbytes if nbytes + snd_bytes_max > 0: # Moving this task would cause the sender to go below mean and # potentially risk becoming a recipient, which would cause tasks to # bounce around. Move on to the next task of the same sender. continue # Find the recipient, farthest from the mean, which # 1. has enough available RAM for this task, and # 2. doesn't hold a copy of this task already # There may not be any that satisfies these conditions; in this case # this task won't be moved. skipped_recipients = [] use_recipient = False while recipients and not use_recipient: rec_bytes_max, rec_bytes_min, _, rec_ws = recipients[0] if nbytes + rec_bytes_max > 0: # recipients are sorted by rec_bytes_max. # The next ones will be worse; no reason to continue iterating break use_recipient = ts not in rec_ws._has_what if not use_recipient: skipped_recipients.append(heapq.heappop(recipients)) for recipient in skipped_recipients: heapq.heappush(recipients, recipient) if not use_recipient: # This task has no recipients available. Leave it on the sender and # move on to the next task of the same sender. continue # Schedule task for transfer from sender to recipient msgs.append((snd_ws, rec_ws, ts)) # *_bytes_max/min are all negative for heap sorting snd_bytes_max += nbytes snd_bytes_min += nbytes rec_bytes_max += nbytes rec_bytes_min += nbytes # Stop iterating on the tasks of this sender for now and, if it still # has bytes to lose, push it back into the senders heap; it may or may # not come back on top again. if snd_bytes_min < 0: # See definition of senders above heapq.heapreplace( senders, (snd_bytes_max, snd_bytes_min, id(snd_ws), snd_ws, ts_iter), ) else: heapq.heappop(senders) # If recipient still has bytes to gain, push it back into the recipients # heap; it may or may not come back on top again. if rec_bytes_min < 0: # See definition of recipients above heapq.heapreplace( recipients, (rec_bytes_max, rec_bytes_min, id(rec_ws), rec_ws), ) else: heapq.heappop(recipients) # Move to next sender with the most data to lose. # It may or may not be the same sender again. break else: # for ts in ts_iter # Exhausted tasks on this sender heapq.heappop(senders) return msgs async def _rebalance_move_data( self, msgs: "list[tuple[WorkerState, WorkerState, TaskState]]" ) -> dict: """Perform the actual transfer of data across the network in rebalance(). Takes in input the output of _rebalance_find_msgs(), that is a list of tuples: - sender worker - recipient worker - task to be transferred FIXME this method is not robust when the cluster is not idle. """ snd_ws: WorkerState rec_ws: WorkerState ts: TaskState to_recipients: "defaultdict[str, defaultdict[str, list[str]]]" = defaultdict( lambda: defaultdict(list) ) for snd_ws, rec_ws, ts in msgs: to_recipients[rec_ws.address][ts._key].append(snd_ws.address) failed_keys_by_recipient = dict( zip( to_recipients, await asyncio.gather( *( # Note: this never raises exceptions self.gather_on_worker(w, who_has) for w, who_has in to_recipients.items() ) ), ) ) to_senders = defaultdict(list) for snd_ws, rec_ws, ts in msgs: if ts._key not in failed_keys_by_recipient[rec_ws.address]: to_senders[snd_ws.address].append(ts._key) # Note: this never raises exceptions await asyncio.gather( *(self.delete_worker_data(r, v) for r, v in to_senders.items()) ) for r, v in to_recipients.items(): self.log_event(r, {"action": "rebalance", "who_has": v}) self.log_event( "all", { "action": "rebalance", "senders": valmap(len, to_senders), "recipients": valmap(len, to_recipients), "moved_keys": len(msgs), }, ) missing_keys = {k for r in failed_keys_by_recipient.values() for k in r} if missing_keys: return {"status": "partial-fail", "keys": list(missing_keys)} else: return {"status": "OK"} async def replicate( self, comm=None, keys=None, n=None, workers=None, branching_factor=2, delete=True, lock=True, ): """Replicate data throughout cluster This performs a tree copy of the data throughout the network individually on each piece of data. Parameters ---------- keys: Iterable list of keys to replicate n: int Number of replications we expect to see within the cluster branching_factor: int, optional The number of workers that can copy data in each generation. The larger the branching factor, the more data we copy in a single step, but the more a given worker risks being swamped by data requests. See also -------- Scheduler.rebalance """ parent: SchedulerState = cast(SchedulerState, self) ws: WorkerState wws: WorkerState ts: TaskState assert branching_factor > 0 async with self._lock if lock else empty_context: if workers is not None: workers = {parent._workers_dv[w] for w in self.workers_list(workers)} workers = {ws for ws in workers if ws._status == Status.running} else: workers = parent._running if n is None: n = len(workers) else: n = min(n, len(workers)) if n == 0: raise ValueError("Can not use replicate to delete data") tasks = {parent._tasks[k] for k in keys} missing_data = [ts._key for ts in tasks if not ts._who_has] if missing_data: return {"status": "partial-fail", "keys": missing_data} # Delete extraneous data if delete: del_worker_tasks = defaultdict(set) for ts in tasks: del_candidates = tuple(ts._who_has & workers) if len(del_candidates) > n: for ws in random.sample( del_candidates, len(del_candidates) - n ): del_worker_tasks[ws].add(ts) # Note: this never raises exceptions await asyncio.gather( *[ self.delete_worker_data(ws._address, [t.key for t in tasks]) for ws, tasks in del_worker_tasks.items() ] ) # Copy not-yet-filled data while tasks: gathers = defaultdict(dict) for ts in list(tasks): if ts._state == "forgotten": # task is no longer needed by any client or dependant task tasks.remove(ts) continue n_missing = n - len(ts._who_has & workers) if n_missing <= 0: # Already replicated enough tasks.remove(ts) continue count = min(n_missing, branching_factor * len(ts._who_has)) assert count > 0 for ws in random.sample(tuple(workers - ts._who_has), count): gathers[ws._address][ts._key] = [ wws._address for wws in ts._who_has ] await asyncio.gather( *( # Note: this never raises exceptions self.gather_on_worker(w, who_has) for w, who_has in gathers.items() ) ) for r, v in gathers.items(): self.log_event(r, {"action": "replicate-add", "who_has": v}) self.log_event( "all", { "action": "replicate", "workers": list(workers), "key-count": len(keys), "branching-factor": branching_factor, }, ) def workers_to_close( self, comm=None, memory_ratio: "int | float | None" = None, n: "int | None" = None, key: "Callable[[WorkerState], Hashable] | None" = None, minimum: "int | None" = None, target: "int | None" = None, attribute: str = "address", ) -> "list[str]": """ Find workers that we can close with low cost This returns a list of workers that are good candidates to retire. These workers are not running anything and are storing relatively little data relative to their peers. If all workers are idle then we still maintain enough workers to have enough RAM to store our data, with a comfortable buffer. This is for use with systems like ``distributed.deploy.adaptive``. Parameters ---------- memory_ratio : Number Amount of extra space we want to have for our stored data. Defaults to 2, or that we want to have twice as much memory as we currently have data. n : int Number of workers to close minimum : int Minimum number of workers to keep around key : Callable(WorkerState) An optional callable mapping a WorkerState object to a group affiliation. Groups will be closed together. This is useful when closing workers must be done collectively, such as by hostname. target : int Target number of workers to have after we close attribute : str The attribute of the WorkerState object to return, like "address" or "name". Defaults to "address". Examples -------- >>> scheduler.workers_to_close() ['tcp://192.168.0.1:1234', 'tcp://192.168.0.2:1234'] Group workers by hostname prior to closing >>> scheduler.workers_to_close(key=lambda ws: ws.host) ['tcp://192.168.0.1:1234', 'tcp://192.168.0.1:4567'] Remove two workers >>> scheduler.workers_to_close(n=2) Keep enough workers to have twice as much memory as we we need. >>> scheduler.workers_to_close(memory_ratio=2) Returns ------- to_close: list of worker addresses that are OK to close See Also -------- Scheduler.retire_workers """ parent: SchedulerState = cast(SchedulerState, self) if target is not None and n is None: n = len(parent._workers_dv) - target if n is not None: if n < 0: n = 0 target = len(parent._workers_dv) - n if n is None and memory_ratio is None: memory_ratio = 2 ws: WorkerState with log_errors(): if not n and all([ws._processing for ws in parent._workers_dv.values()]): return [] if key is None: key = operator.attrgetter("address") if isinstance(key, bytes) and dask.config.get( "distributed.scheduler.pickle" ): key = pickle.loads(key) groups = groupby(key, parent._workers.values()) limit_bytes = { k: sum([ws._memory_limit for ws in v]) for k, v in groups.items() } group_bytes = {k: sum([ws._nbytes for ws in v]) for k, v in groups.items()} limit = sum(limit_bytes.values()) total = sum(group_bytes.values()) def _key(group): wws: WorkerState is_idle = not any([wws._processing for wws in groups[group]]) bytes = -group_bytes[group] return (is_idle, bytes) idle = sorted(groups, key=_key) to_close = [] n_remain = len(parent._workers_dv) while idle: group = idle.pop() if n is None and any([ws._processing for ws in groups[group]]): break if minimum and n_remain - len(groups[group]) < minimum: break limit -= limit_bytes[group] if ( n is not None and n_remain - len(groups[group]) >= cast(int, target) ) or (memory_ratio is not None and limit >= memory_ratio * total): to_close.append(group) n_remain -= len(groups[group]) else: break result = [getattr(ws, attribute) for g in to_close for ws in groups[g]] if result: logger.debug("Suggest closing workers: %s", result) return result async def retire_workers( self, comm=None, workers=None, remove=True, close_workers=False, names=None, lock=True, **kwargs, ) -> dict: """Gracefully retire workers from cluster Parameters ---------- workers: list (optional) List of worker addresses to retire. If not provided we call ``workers_to_close`` which finds a good set names: list (optional) List of worker names to retire. remove: bool (defaults to True) Whether or not to remove the worker metadata immediately or else wait for the worker to contact us close_workers: bool (defaults to False) Whether or not to actually close the worker explicitly from here. Otherwise we expect some external job scheduler to finish off the worker. **kwargs: dict Extra options to pass to workers_to_close to determine which workers we should drop Returns ------- Dictionary mapping worker ID/address to dictionary of information about that worker for each retired worker. See Also -------- Scheduler.workers_to_close """ parent: SchedulerState = cast(SchedulerState, self) ws: WorkerState ts: TaskState with log_errors(): async with self._lock if lock else empty_context: if names is not None: if workers is not None: raise TypeError("names and workers are mutually exclusive") if names: logger.info("Retire worker names %s", names) names = set(map(str, names)) workers = { ws._address for ws in parent._workers_dv.values() if str(ws._name) in names } elif workers is None: while True: try: workers = self.workers_to_close(**kwargs) if not workers: return {} return await self.retire_workers( workers=workers, remove=remove, close_workers=close_workers, lock=False, ) except KeyError: # keys left during replicate pass workers = { parent._workers_dv[w] for w in workers if w in parent._workers_dv } if not workers: return {} logger.info("Retire workers %s", workers) # Keys orphaned by retiring those workers keys = {k for w in workers for k in w.has_what} keys = {ts._key for ts in keys if ts._who_has.issubset(workers)} if keys: other_workers = set(parent._workers_dv.values()) - workers if not other_workers: return {} logger.info("Moving %d keys to other workers", len(keys)) await self.replicate( keys=keys, workers=[ws._address for ws in other_workers], n=1, delete=False, lock=False, ) worker_keys = {ws._address: ws.identity() for ws in workers} if close_workers: await asyncio.gather( *[self.close_worker(worker=w, safe=True) for w in worker_keys] ) if remove: await asyncio.gather( *[self.remove_worker(address=w, safe=True) for w in worker_keys] ) self.log_event( "all", { "action": "retire-workers", "workers": worker_keys, "moved-keys": len(keys), }, ) self.log_event(list(worker_keys), {"action": "retired"}) return worker_keys def add_keys(self, comm=None, worker=None, keys=(), stimulus_id=None): """ Learn that a worker has certain keys This should not be used in practice and is mostly here for legacy reasons. However, it is sent by workers from time to time. """ parent: SchedulerState = cast(SchedulerState, self) if worker not in parent._workers_dv: return "not found" ws: WorkerState = parent._workers_dv[worker] redundant_replicas = [] for key in keys: ts: TaskState = parent._tasks.get(key) if ts is not None and ts._state == "memory": if ws not in ts._who_has: parent.add_replica(ts, ws) else: redundant_replicas.append(key) if redundant_replicas: if not stimulus_id: stimulus_id = f"redundant-replicas-{time()}" self.worker_send( worker, { "op": "remove-replicas", "keys": redundant_replicas, "stimulus_id": stimulus_id, }, ) return "OK" def update_data( self, comm=None, *, who_has: dict, nbytes: dict, client=None, serializers=None, ): """ Learn that new data has entered the network from an external source See Also -------- Scheduler.mark_key_in_memory """ parent: SchedulerState = cast(SchedulerState, self) with log_errors(): who_has = { k: [self.coerce_address(vv) for vv in v] for k, v in who_has.items() } logger.debug("Update data %s", who_has) for key, workers in who_has.items(): ts: TaskState = parent._tasks.get(key) # type: ignore if ts is None: ts = parent.new_task(key, None, "memory") ts.state = "memory" ts_nbytes = nbytes.get(key, -1) if ts_nbytes >= 0: ts.set_nbytes(ts_nbytes) for w in workers: ws: WorkerState = parent._workers_dv[w] if ws not in ts._who_has: parent.add_replica(ts, ws) self.report( {"op": "key-in-memory", "key": key, "workers": list(workers)} ) if client: self.client_desires_keys(keys=list(who_has), client=client) def report_on_key(self, key: str = None, ts: TaskState = None, client: str = None): parent: SchedulerState = cast(SchedulerState, self) if ts is None: ts = parent._tasks.get(key) elif key is None: key = ts._key else: assert False, (key, ts) return report_msg: dict if ts is None: report_msg = {"op": "cancelled-key", "key": key} else: report_msg = _task_to_report_msg(parent, ts) if report_msg is not None: self.report(report_msg, ts=ts, client=client) async def feed( self, comm, function=None, setup=None, teardown=None, interval="1s", **kwargs ): """ Provides a data Comm to external requester Caution: this runs arbitrary Python code on the scheduler. This should eventually be phased out. It is mostly used by diagnostics. """ if not dask.config.get("distributed.scheduler.pickle"): logger.warn( "Tried to call 'feed' route with custom functions, but " "pickle is disallowed. Set the 'distributed.scheduler.pickle'" "config value to True to use the 'feed' route (this is mostly " "commonly used with progress bars)" ) return interval = parse_timedelta(interval) with log_errors(): if function: function = pickle.loads(function) if setup: setup = pickle.loads(setup) if teardown: teardown = pickle.loads(teardown) state = setup(self) if setup else None if inspect.isawaitable(state): state = await state try: while self.status == Status.running: if state is None: response = function(self) else: response = function(self, state) await comm.write(response) await asyncio.sleep(interval) except OSError: pass finally: if teardown: teardown(self, state) def log_worker_event(self, worker=None, topic=None, msg=None): self.log_event(topic, msg) def subscribe_worker_status(self, comm=None): WorkerStatusPlugin(self, comm) ident = self.identity() for v in ident["workers"].values(): del v["metrics"] del v["last_seen"] return ident def get_processing(self, comm=None, workers=None): parent: SchedulerState = cast(SchedulerState, self) ws: WorkerState ts: TaskState if workers is not None: workers = set(map(self.coerce_address, workers)) return { w: [ts._key for ts in parent._workers_dv[w].processing] for w in workers } else: return { w: [ts._key for ts in ws._processing] for w, ws in parent._workers_dv.items() } def get_who_has(self, comm=None, keys=None): parent: SchedulerState = cast(SchedulerState, self) ws: WorkerState ts: TaskState if keys is not None: return { k: [ws._address for ws in parent._tasks[k].who_has] if k in parent._tasks else [] for k in keys } else: return { key: [ws._address for ws in ts._who_has] for key, ts in parent._tasks.items() } def get_has_what(self, comm=None, workers=None): parent: SchedulerState = cast(SchedulerState, self) ws: WorkerState ts: TaskState if workers is not None: workers = map(self.coerce_address, workers) return { w: [ts._key for ts in parent._workers_dv[w].has_what] if w in parent._workers_dv else [] for w in workers } else: return { w: [ts._key for ts in ws.has_what] for w, ws in parent._workers_dv.items() } def get_ncores(self, comm=None, workers=None): parent: SchedulerState = cast(SchedulerState, self) ws: WorkerState if workers is not None: workers = map(self.coerce_address, workers) return { w: parent._workers_dv[w].nthreads for w in workers if w in parent._workers_dv } else: return {w: ws._nthreads for w, ws in parent._workers_dv.items()} def get_ncores_running(self, comm=None, workers=None): parent: SchedulerState = cast(SchedulerState, self) ncores = self.get_ncores(workers=workers) return { w: n for w, n in ncores.items() if parent._workers_dv[w].status == Status.running } async def get_call_stack(self, comm=None, keys=None): parent: SchedulerState = cast(SchedulerState, self) ts: TaskState dts: TaskState if keys is not None: stack = list(keys) processing = set() while stack: key = stack.pop() ts = parent._tasks[key] if ts._state == "waiting": stack.extend([dts._key for dts in ts._dependencies]) elif ts._state == "processing": processing.add(ts) workers = defaultdict(list) for ts in processing: if ts._processing_on: workers[ts._processing_on.address].append(ts._key) else: workers = {w: None for w in parent._workers_dv} if not workers: return {} results = await asyncio.gather( *(self.rpc(w).call_stack(keys=v) for w, v in workers.items()) ) response = {w: r for w, r in zip(workers, results) if r} return response def get_nbytes(self, comm=None, keys=None, summary=True): parent: SchedulerState = cast(SchedulerState, self) ts: TaskState with log_errors(): if keys is not None: result = {k: parent._tasks[k].nbytes for k in keys} else: result = { k: ts._nbytes for k, ts in parent._tasks.items() if ts._nbytes >= 0 } if summary: out = defaultdict(lambda: 0) for k, v in result.items(): out[key_split(k)] += v result = dict(out) return result def run_function(self, stream, function, args=(), kwargs={}, wait=True): """Run a function within this process See Also -------- Client.run_on_scheduler """ from .worker import run if not dask.config.get("distributed.scheduler.pickle"): raise ValueError( "Cannot run function as the scheduler has been explicitly disallowed from " "deserializing arbitrary bytestrings using pickle via the " "'distributed.scheduler.pickle' configuration setting." ) self.log_event("all", {"action": "run-function", "function": function}) return run(self, stream, function=function, args=args, kwargs=kwargs, wait=wait) def set_metadata(self, comm=None, keys=None, value=None): parent: SchedulerState = cast(SchedulerState, self) metadata = parent._task_metadata for key in keys[:-1]: if key not in metadata or not isinstance(metadata[key], (dict, list)): metadata[key] = {} metadata = metadata[key] metadata[keys[-1]] = value def get_metadata(self, comm=None, keys=None, default=no_default): parent: SchedulerState = cast(SchedulerState, self) metadata = parent._task_metadata for key in keys[:-1]: metadata = metadata[key] try: return metadata[keys[-1]] except KeyError: if default != no_default: return default else: raise def set_restrictions(self, comm=None, worker=None): ts: TaskState for key, restrictions in worker.items(): ts = self.tasks[key] if isinstance(restrictions, str): restrictions = {restrictions} ts._worker_restrictions = set(restrictions) def get_task_status(self, comm=None, keys=None): parent: SchedulerState = cast(SchedulerState, self) return { key: (parent._tasks[key].state if key in parent._tasks else None) for key in keys } def get_task_stream(self, comm=None, start=None, stop=None, count=None): from distributed.diagnostics.task_stream import TaskStreamPlugin if TaskStreamPlugin.name not in self.plugins: self.add_plugin(TaskStreamPlugin(self)) plugin = self.plugins[TaskStreamPlugin.name] return plugin.collect(start=start, stop=stop, count=count) def start_task_metadata(self, comm=None, name=None): plugin = CollectTaskMetaDataPlugin(scheduler=self, name=name) self.add_plugin(plugin) def stop_task_metadata(self, comm=None, name=None): plugins = [ p for p in list(self.plugins.values()) if isinstance(p, CollectTaskMetaDataPlugin) and p.name == name ] if len(plugins) != 1: raise ValueError( "Expected to find exactly one CollectTaskMetaDataPlugin " f"with name {name} but found {len(plugins)}." ) plugin = plugins[0] self.remove_plugin(name=plugin.name) return {"metadata": plugin.metadata, "state": plugin.state} async def register_worker_plugin(self, comm, plugin, name=None): """Registers a setup function, and call it on every worker""" self.worker_plugins[name] = plugin responses = await self.broadcast( msg=dict(op="plugin-add", plugin=plugin, name=name) ) return responses async def unregister_worker_plugin(self, comm, name): """Unregisters a worker plugin""" try: self.worker_plugins.pop(name) except KeyError: raise ValueError(f"The worker plugin {name} does not exists") responses = await self.broadcast(msg=dict(op="plugin-remove", name=name)) return responses async def register_nanny_plugin(self, comm, plugin, name=None): """Registers a setup function, and call it on every worker""" self.nanny_plugins[name] = plugin responses = await self.broadcast( msg=dict(op="plugin_add", plugin=plugin, name=name), nanny=True, ) return responses async def unregister_nanny_plugin(self, comm, name): """Unregisters a worker plugin""" try: self.nanny_plugins.pop(name) except KeyError: raise ValueError(f"The nanny plugin {name} does not exists") responses = await self.broadcast( msg=dict(op="plugin_remove", name=name), nanny=True ) return responses def transition(self, key, finish: str, *args, **kwargs): """Transition a key from its current state to the finish state Examples -------- >>> self.transition('x', 'waiting') {'x': 'processing'} Returns ------- Dictionary of recommendations for future transitions See Also -------- Scheduler.transitions: transitive version of this function """ parent: SchedulerState = cast(SchedulerState, self) recommendations: dict worker_msgs: dict client_msgs: dict a: tuple = parent._transition(key, finish, *args, **kwargs) recommendations, client_msgs, worker_msgs = a self.send_all(client_msgs, worker_msgs) return recommendations def transitions(self, recommendations: dict): """Process transitions until none are left This includes feedback from previous transitions and continues until we reach a steady state """ parent: SchedulerState = cast(SchedulerState, self) client_msgs: dict = {} worker_msgs: dict = {} parent._transitions(recommendations, client_msgs, worker_msgs) self.send_all(client_msgs, worker_msgs) def story(self, *keys): """Get all transitions that touch one of the input keys""" keys = {key.key if isinstance(key, TaskState) else key for key in keys} return [ t for t in self.transition_log if t[0] in keys or keys.intersection(t[3]) ] transition_story = story def reschedule(self, key=None, worker=None): """Reschedule a task Things may have shifted and this task may now be better suited to run elsewhere """ parent: SchedulerState = cast(SchedulerState, self) ts: TaskState try: ts = parent._tasks[key] except KeyError: logger.warning( "Attempting to reschedule task {}, which was not " "found on the scheduler. Aborting reschedule.".format(key) ) return if ts._state != "processing": return if worker and ts._processing_on.address != worker: return self.transitions({key: "released"}) ##################### # Utility functions # ##################### def add_resources(self, comm=None, worker=None, resources=None): parent: SchedulerState = cast(SchedulerState, self) ws: WorkerState = parent._workers_dv[worker] if resources: ws._resources.update(resources) ws._used_resources = {} for resource, quantity in ws._resources.items(): ws._used_resources[resource] = 0 dr: dict = parent._resources.get(resource, None) if dr is None: parent._resources[resource] = dr = {} dr[worker] = quantity return "OK" def remove_resources(self, worker): parent: SchedulerState = cast(SchedulerState, self) ws: WorkerState = parent._workers_dv[worker] for resource, quantity in ws._resources.items(): dr: dict = parent._resources.get(resource, None) if dr is None: parent._resources[resource] = dr = {} del dr[worker] def coerce_address(self, addr, resolve=True): """ Coerce possible input addresses to canonical form. *resolve* can be disabled for testing with fake hostnames. Handles strings, tuples, or aliases. """ # XXX how many address-parsing routines do we have? parent: SchedulerState = cast(SchedulerState, self) if addr in parent._aliases: addr = parent._aliases[addr] if isinstance(addr, tuple): addr = unparse_host_port(*addr) if not isinstance(addr, str): raise TypeError(f"addresses should be strings or tuples, got {addr!r}") if resolve: addr = resolve_address(addr) else: addr = normalize_address(addr) return addr def workers_list(self, workers): """ List of qualifying workers Takes a list of worker addresses or hostnames. Returns a list of all worker addresses that match """ parent: SchedulerState = cast(SchedulerState, self) if workers is None: return list(parent._workers) out = set() for w in workers: if ":" in w: out.add(w) else: out.update({ww for ww in parent._workers if w in ww}) # TODO: quadratic return list(out) def start_ipython(self, comm=None): """Start an IPython kernel Returns Jupyter connection info dictionary. """ from ._ipython_utils import start_ipython if self._ipython_kernel is None: self._ipython_kernel = start_ipython( ip=self.ip, ns={"scheduler": self}, log=logger ) return self._ipython_kernel.get_connection_info() async def get_profile( self, comm=None, workers=None, scheduler=False, server=False, merge_workers=True, start=None, stop=None, key=None, ): parent: SchedulerState = cast(SchedulerState, self) if workers is None: workers = parent._workers_dv else: workers = set(parent._workers_dv) & set(workers) if scheduler: return profile.get_profile(self.io_loop.profile, start=start, stop=stop) results = await asyncio.gather( *( self.rpc(w).profile(start=start, stop=stop, key=key, server=server) for w in workers ), return_exceptions=True, ) results = [r for r in results if not isinstance(r, Exception)] if merge_workers: response = profile.merge(*results) else: response = dict(zip(workers, results)) return response async def get_profile_metadata( self, comm=None, workers=None, merge_workers=True, start=None, stop=None, profile_cycle_interval=None, ): parent: SchedulerState = cast(SchedulerState, self) dt = profile_cycle_interval or dask.config.get( "distributed.worker.profile.cycle" ) dt = parse_timedelta(dt, default="ms") if workers is None: workers = parent._workers_dv else: workers = set(parent._workers_dv) & set(workers) results = await asyncio.gather( *(self.rpc(w).profile_metadata(start=start, stop=stop) for w in workers), return_exceptions=True, ) results = [r for r in results if not isinstance(r, Exception)] counts = [v["counts"] for v in results] counts = itertools.groupby(merge_sorted(*counts), lambda t: t[0] // dt * dt) counts = [(time, sum(pluck(1, group))) for time, group in counts] keys = set() for v in results: for t, d in v["keys"]: for k in d: keys.add(k) keys = {k: [] for k in keys} groups1 = [v["keys"] for v in results] groups2 = list(merge_sorted(*groups1, key=first)) last = 0 for t, d in groups2: tt = t // dt * dt if tt > last: last = tt for k, v in keys.items(): v.append([tt, 0]) for k, v in d.items(): keys[k][-1][1] += v return {"counts": counts, "keys": keys} async def performance_report( self, comm=None, start=None, last_count=None, code="", mode=None ): parent: SchedulerState = cast(SchedulerState, self) stop = time() # Profiles compute, scheduler, workers = await asyncio.gather( *[ self.get_profile(start=start), self.get_profile(scheduler=True, start=start), self.get_profile(server=True, start=start), ] ) from . import profile def profile_to_figure(state): data = profile.plot_data(state) figure, source = profile.plot_figure(data, sizing_mode="stretch_both") return figure compute, scheduler, workers = map( profile_to_figure, (compute, scheduler, workers) ) # Task stream task_stream = self.get_task_stream(start=start) total_tasks = len(task_stream) timespent = defaultdict(int) for d in task_stream: for x in d.get("startstops", []): timespent[x["action"]] += x["stop"] - x["start"] tasks_timings = "" for k in sorted(timespent.keys()): tasks_timings += f"\n
  • {k} time: {format_time(timespent[k])}
    • " from .dashboard.components.scheduler import task_stream_figure from .diagnostics.task_stream import rectangles rects = rectangles(task_stream) source, task_stream = task_stream_figure(sizing_mode="stretch_both") source.data.update(rects) # Bandwidth from distributed.dashboard.components.scheduler import ( BandwidthTypes, BandwidthWorkers, ) bandwidth_workers = BandwidthWorkers(self, sizing_mode="stretch_both") bandwidth_workers.update() bandwidth_types = BandwidthTypes(self, sizing_mode="stretch_both") bandwidth_types.update() # System monitor from distributed.dashboard.components.shared import SystemMonitor sysmon = SystemMonitor(self, last_count=last_count, sizing_mode="stretch_both") sysmon.update() # Scheduler logs from distributed.dashboard.components.scheduler import SchedulerLogs logs = SchedulerLogs(self) from bokeh.models import Div, Panel, Tabs import distributed # HTML ws: WorkerState html = """

    Dask Performance Report

    Select different tabs on the top for additional information

    Duration: {time}

    Tasks Information

    • number of tasks: {ntasks}
      • {tasks_timings}

    Scheduler Information

    • Address: {address}
    • Workers: {nworkers}
    • Threads: {threads}
    • Memory: {memory}
    • Dask Version: {dask_version}
    • Dask.Distributed Version: {distributed_version}

    Calling Code 

    {code}
    """.format( time=format_time(stop - start), ntasks=total_tasks, tasks_timings=tasks_timings, address=self.address, nworkers=len(parent._workers_dv), threads=sum([ws._nthreads for ws in parent._workers_dv.values()]), memory=format_bytes( sum([ws._memory_limit for ws in parent._workers_dv.values()]) ), code=code, dask_version=dask.__version__, distributed_version=distributed.__version__, ) html = Div( text=html, style={ "width": "100%", "height": "100%", "max-width": "1920px", "max-height": "1080px", "padding": "12px", "border": "1px solid lightgray", "box-shadow": "inset 1px 0 8px 0 lightgray", "overflow": "auto", }, ) html = Panel(child=html, title="Summary") compute = Panel(child=compute, title="Worker Profile (compute)") workers = Panel(child=workers, title="Worker Profile (administrative)") scheduler = Panel(child=scheduler, title="Scheduler Profile (administrative)") task_stream = Panel(child=task_stream, title="Task Stream") bandwidth_workers = Panel( child=bandwidth_workers.root, title="Bandwidth (Workers)" ) bandwidth_types = Panel(child=bandwidth_types.root, title="Bandwidth (Types)") system = Panel(child=sysmon.root, title="System") logs = Panel(child=logs.root, title="Scheduler Logs") tabs = Tabs( tabs=[ html, task_stream, system, logs, compute, workers, scheduler, bandwidth_workers, bandwidth_types, ] ) from bokeh.core.templates import get_env from bokeh.plotting import output_file, save with tmpfile(extension=".html") as fn: output_file(filename=fn, title="Dask Performance Report", mode=mode) template_directory = os.path.join( os.path.dirname(os.path.abspath(__file__)), "dashboard", "templates" ) template_environment = get_env() template_environment.loader.searchpath.append(template_directory) template = template_environment.get_template("performance_report.html") save(tabs, filename=fn, template=template) with open(fn) as f: data = f.read() return data async def get_worker_logs(self, comm=None, n=None, workers=None, nanny=False): results = await self.broadcast( msg={"op": "get_logs", "n": n}, workers=workers, nanny=nanny ) return results def log_event(self, name, msg): event = (time(), msg) if isinstance(name, (list, tuple)): for n in name: self.events[n].append(event) self.event_counts[n] += 1 self._report_event(n, event) else: self.events[name].append(event) self.event_counts[name] += 1 self._report_event(name, event) def _report_event(self, name, event): for client in self.event_subscriber[name]: self.report( { "op": "event", "topic": name, "event": event, }, client=client, ) def subscribe_topic(self, topic, client): self.event_subscriber[topic].add(client) def unsubscribe_topic(self, topic, client): self.event_subscriber[topic].discard(client) def get_events(self, comm=None, topic=None): if topic is not None: return tuple(self.events[topic]) else: return valmap(tuple, self.events) async def get_worker_monitor_info(self, recent=False, starts=None): parent: SchedulerState = cast(SchedulerState, self) if starts is None: starts = {} results = await asyncio.gather( *( self.rpc(w).get_monitor_info(recent=recent, start=starts.get(w, 0)) for w in parent._workers_dv ) ) return dict(zip(parent._workers_dv, results)) ########### # Cleanup # ########### def reevaluate_occupancy(self, worker_index: Py_ssize_t = 0): """Periodically reassess task duration time The expected duration of a task can change over time. Unfortunately we don't have a good constant-time way to propagate the effects of these changes out to the summaries that they affect, like the total expected runtime of each of the workers, or what tasks are stealable. In this coroutine we walk through all of the workers and re-align their estimates with the current state of tasks. We do this periodically rather than at every transition, and we only do it if the scheduler process isn't under load (using psutil.Process.cpu_percent()). This lets us avoid this fringe optimization when we have better things to think about. """ parent: SchedulerState = cast(SchedulerState, self) try: if self.status == Status.closed: return last = time() next_time = timedelta(seconds=0.1) if self.proc.cpu_percent() < 50: workers: list = list(parent._workers.values()) nworkers: Py_ssize_t = len(workers) i: Py_ssize_t for i in range(nworkers): ws: WorkerState = workers[worker_index % nworkers] worker_index += 1 try: if ws is None or not ws._processing: continue _reevaluate_occupancy_worker(parent, ws) finally: del ws # lose ref duration = time() - last if duration > 0.005: # 5ms since last release next_time = timedelta(seconds=duration * 5) # 25ms gap break self.loop.add_timeout( next_time, self.reevaluate_occupancy, worker_index=worker_index ) except Exception: logger.error("Error in reevaluate occupancy", exc_info=True) raise async def check_worker_ttl(self): parent: SchedulerState = cast(SchedulerState, self) ws: WorkerState now = time() for ws in parent._workers_dv.values(): if (ws._last_seen < now - self.worker_ttl) and ( ws._last_seen < now - 10 * heartbeat_interval(len(parent._workers_dv)) ): logger.warning( "Worker failed to heartbeat within %s seconds. Closing: %s", self.worker_ttl, ws, ) await self.remove_worker(address=ws._address) def check_idle(self): parent: SchedulerState = cast(SchedulerState, self) ws: WorkerState if ( any([ws._processing for ws in parent._workers_dv.values()]) or parent._unrunnable ): self.idle_since = None return elif not self.idle_since: self.idle_since = time() if time() > self.idle_since + self.idle_timeout: logger.info( "Scheduler closing after being idle for %s", format_time(self.idle_timeout), ) self.loop.add_callback(self.close) def adaptive_target(self, comm=None, target_duration=None): """Desired number of workers based on the current workload This looks at the current running tasks and memory use, and returns a number of desired workers. This is often used by adaptive scheduling. Parameters ---------- target_duration : str A desired duration of time for computations to take. This affects how rapidly the scheduler will ask to scale. See Also -------- distributed.deploy.Adaptive """ parent: SchedulerState = cast(SchedulerState, self) if target_duration is None: target_duration = dask.config.get("distributed.adaptive.target-duration") target_duration = parse_timedelta(target_duration) # CPU cpu = math.ceil( parent._total_occupancy / target_duration ) # TODO: threads per worker # Avoid a few long tasks from asking for many cores ws: WorkerState tasks_processing = 0 for ws in parent._workers_dv.values(): tasks_processing += len(ws._processing) if tasks_processing > cpu: break else: cpu = min(tasks_processing, cpu) if parent._unrunnable and not parent._workers_dv: cpu = max(1, cpu) # add more workers if more than 60% of memory is used limit = sum([ws._memory_limit for ws in parent._workers_dv.values()]) used = sum([ws._nbytes for ws in parent._workers_dv.values()]) memory = 0 if used > 0.6 * limit and limit > 0: memory = 2 * len(parent._workers_dv) target = max(memory, cpu) if target >= len(parent._workers_dv): return target else: # Scale down? to_close = self.workers_to_close() return len(parent._workers_dv) - len(to_close) @cfunc @exceptval(check=False) def _remove_from_processing( state: SchedulerState, ts: TaskState ) -> str: # -> str | None """ Remove *ts* from the set of processing tasks. See also ``Scheduler.set_duration_estimate`` """ ws: WorkerState = ts._processing_on ts._processing_on = None # type: ignore w: str = ws._address if w not in state._workers_dv: # may have been removed return None # type: ignore duration: double = ws._processing.pop(ts) if not ws._processing: state._total_occupancy -= ws._occupancy ws._occupancy = 0 else: state._total_occupancy -= duration ws._occupancy -= duration state.check_idle_saturated(ws) state.release_resources(ts, ws) return w @cfunc @exceptval(check=False) def _add_to_memory( state: SchedulerState, ts: TaskState, ws: WorkerState, recommendations: dict, client_msgs: dict, type=None, typename: str = None, ): """ Add *ts* to the set of in-memory tasks. """ if state._validate: assert ts not in ws._has_what state.add_replica(ts, ws) deps: list = list(ts._dependents) if len(deps) > 1: deps.sort(key=operator.attrgetter("priority"), reverse=True) dts: TaskState s: set for dts in deps: s = dts._waiting_on if ts in s: s.discard(ts) if not s: # new task ready to run recommendations[dts._key] = "processing" for dts in ts._dependencies: s = dts._waiters s.discard(ts) if not s and not dts._who_wants: recommendations[dts._key] = "released" report_msg: dict = {} cs: ClientState if not ts._waiters and not ts._who_wants: recommendations[ts._key] = "released" else: report_msg["op"] = "key-in-memory" report_msg["key"] = ts._key if type is not None: report_msg["type"] = type for cs in ts._who_wants: client_msgs[cs._client_key] = [report_msg] ts.state = "memory" ts._type = typename # type: ignore ts._group._types.add(typename) cs = state._clients["fire-and-forget"] if ts in cs._wants_what: _client_releases_keys( state, cs=cs, keys=[ts._key], recommendations=recommendations, ) @cfunc @exceptval(check=False) def _propagate_forgotten( state: SchedulerState, ts: TaskState, recommendations: dict, worker_msgs: dict ): ts.state = "forgotten" key: str = ts._key dts: TaskState for dts in ts._dependents: dts._has_lost_dependencies = True dts._dependencies.remove(ts) dts._waiting_on.discard(ts) if dts._state not in ("memory", "erred"): # Cannot compute task anymore recommendations[dts._key] = "forgotten" ts._dependents.clear() ts._waiters.clear() for dts in ts._dependencies: dts._dependents.remove(ts) dts._waiters.discard(ts) if not dts._dependents and not dts._who_wants: # Task not needed anymore assert dts is not ts recommendations[dts._key] = "forgotten" ts._dependencies.clear() ts._waiting_on.clear() ws: WorkerState for ws in ts._who_has: w: str = ws._address if w in state._workers_dv: # in case worker has died worker_msgs[w] = [ { "op": "free-keys", "keys": [key], "stimulus_id": f"propagate-forgotten-{time()}", } ] state.remove_all_replicas(ts) @cfunc @exceptval(check=False) def _client_releases_keys( state: SchedulerState, keys: list, cs: ClientState, recommendations: dict ): """Remove keys from client desired list""" logger.debug("Client %s releases keys: %s", cs._client_key, keys) ts: TaskState for key in keys: ts = state._tasks.get(key) # type: ignore if ts is not None and ts in cs._wants_what: cs._wants_what.remove(ts) ts._who_wants.remove(cs) if not ts._who_wants: if not ts._dependents: # No live dependents, can forget recommendations[ts._key] = "forgotten" elif ts._state != "erred" and not ts._waiters: recommendations[ts._key] = "released" @cfunc @exceptval(check=False) def _task_to_msg(state: SchedulerState, ts: TaskState, duration: double = -1) -> dict: """Convert a single computational task to a message""" ws: WorkerState dts: TaskState # FIXME: The duration attribute is not used on worker. We could safe ourselves the time to compute and submit this if duration < 0: duration = state.get_task_duration(ts) msg: dict = { "op": "compute-task", "key": ts._key, "priority": ts._priority, "duration": duration, "stimulus_id": f"compute-task-{time()}", "who_has": {}, } if ts._resource_restrictions: msg["resource_restrictions"] = ts._resource_restrictions if ts._actor: msg["actor"] = True deps: set = ts._dependencies if deps: msg["who_has"] = { dts._key: [ws._address for ws in dts._who_has] for dts in deps } msg["nbytes"] = {dts._key: dts._nbytes for dts in deps} if state._validate: assert all(msg["who_has"].values()) task = ts._run_spec if type(task) is dict: msg.update(task) else: msg["task"] = task if ts._annotations: msg["annotations"] = ts._annotations return msg @cfunc @exceptval(check=False) def _task_to_report_msg(state: SchedulerState, ts: TaskState) -> dict: # -> dict | None if ts._state == "forgotten": return {"op": "cancelled-key", "key": ts._key} elif ts._state == "memory": return {"op": "key-in-memory", "key": ts._key} elif ts._state == "erred": failing_ts: TaskState = ts._exception_blame return { "op": "task-erred", "key": ts._key, "exception": failing_ts._exception, "traceback": failing_ts._traceback, } else: return None # type: ignore @cfunc @exceptval(check=False) def _task_to_client_msgs(state: SchedulerState, ts: TaskState) -> dict: if ts._who_wants: report_msg: dict = _task_to_report_msg(state, ts) if report_msg is not None: cs: ClientState return {cs._client_key: [report_msg] for cs in ts._who_wants} return {} @cfunc @exceptval(check=False) def _reevaluate_occupancy_worker(state: SchedulerState, ws: WorkerState): """See reevaluate_occupancy""" ts: TaskState old = ws._occupancy for ts in ws._processing: state.set_duration_estimate(ts, ws) state.check_idle_saturated(ws) steal = state.extensions.get("stealing") if not steal: return if ws._occupancy > old * 1.3 or old > ws._occupancy * 1.3: for ts in ws._processing: steal.recalculate_cost(ts) @cfunc @exceptval(check=False) def decide_worker( ts: TaskState, all_workers, valid_workers: set, objective ) -> WorkerState: # -> WorkerState | None """ Decide which worker should take task *ts*. We choose the worker that has the data on which *ts* depends. If several workers have dependencies then we choose the less-busy worker. Optionally provide *valid_workers* of where jobs are allowed to occur (if all workers are allowed to take the task, pass None instead). If the task requires data communication because no eligible worker has all the dependencies already, then we choose to minimize the number of bytes sent between workers. This is determined by calling the *objective* function. """ ws: WorkerState = None # type: ignore wws: WorkerState dts: TaskState deps: set = ts._dependencies candidates: set assert all([dts._who_has for dts in deps]) if ts._actor: candidates = set(all_workers) else: candidates = {wws for dts in deps for wws in dts._who_has} if valid_workers is None: if not candidates: candidates = set(all_workers) else: candidates &= valid_workers if not candidates: candidates = valid_workers if not candidates: if ts._loose_restrictions: ws = decide_worker(ts, all_workers, None, objective) return ws ncandidates: Py_ssize_t = len(candidates) if ncandidates == 0: pass elif ncandidates == 1: for ws in candidates: break else: ws = min(candidates, key=objective) return ws def validate_task_state(ts: TaskState): """ Validate the given TaskState. """ ws: WorkerState dts: TaskState assert ts._state in ALL_TASK_STATES or ts._state == "forgotten", ts if ts._waiting_on: assert ts._waiting_on.issubset(ts._dependencies), ( "waiting not subset of dependencies", str(ts._waiting_on), str(ts._dependencies), ) if ts._waiters: assert ts._waiters.issubset(ts._dependents), ( "waiters not subset of dependents", str(ts._waiters), str(ts._dependents), ) for dts in ts._waiting_on: assert not dts._who_has, ("waiting on in-memory dep", str(ts), str(dts)) assert dts._state != "released", ("waiting on released dep", str(ts), str(dts)) for dts in ts._dependencies: assert ts in dts._dependents, ( "not in dependency's dependents", str(ts), str(dts), str(dts._dependents), ) if ts._state in ("waiting", "processing"): assert dts in ts._waiting_on or dts._who_has, ( "dep missing", str(ts), str(dts), ) assert dts._state != "forgotten" for dts in ts._waiters: assert dts._state in ("waiting", "processing"), ( "waiter not in play", str(ts), str(dts), ) for dts in ts._dependents: assert ts in dts._dependencies, ( "not in dependent's dependencies", str(ts), str(dts), str(dts._dependencies), ) assert dts._state != "forgotten" assert (ts._processing_on is not None) == (ts._state == "processing") assert bool(ts._who_has) == (ts._state == "memory"), (ts, ts._who_has, ts._state) if ts._state == "processing": assert all([dts._who_has for dts in ts._dependencies]), ( "task processing without all deps", str(ts), str(ts._dependencies), ) assert not ts._waiting_on if ts._who_has: assert ts._waiters or ts._who_wants, ( "unneeded task in memory", str(ts), str(ts._who_has), ) if ts._run_spec: # was computed assert ts._type assert isinstance(ts._type, str) assert not any([ts in dts._waiting_on for dts in ts._dependents]) for ws in ts._who_has: assert ts in ws._has_what, ( "not in who_has' has_what", str(ts), str(ws), str(ws._has_what), ) if ts._who_wants: cs: ClientState for cs in ts._who_wants: assert ts in cs._wants_what, ( "not in who_wants' wants_what", str(ts), str(cs), str(cs._wants_what), ) if ts._actor: if ts._state == "memory": assert sum([ts in ws._actors for ws in ts._who_has]) == 1 if ts._state == "processing": assert ts in ts._processing_on.actors def validate_worker_state(ws: WorkerState): ts: TaskState for ts in ws._has_what: assert ws in ts._who_has, ( "not in has_what' who_has", str(ws), str(ts), str(ts._who_has), ) for ts in ws._actors: assert ts._state in ("memory", "processing") def validate_state(tasks, workers, clients): """ Validate a current runtime state This performs a sequence of checks on the entire graph, running in about linear time. This raises assert errors if anything doesn't check out. """ ts: TaskState for ts in tasks.values(): validate_task_state(ts) ws: WorkerState for ws in workers.values(): validate_worker_state(ws) cs: ClientState for cs in clients.values(): for ts in cs._wants_what: assert cs in ts._who_wants, ( "not in wants_what' who_wants", str(cs), str(ts), str(ts._who_wants), ) _round_robin = [0] def heartbeat_interval(n): """ Interval in seconds that we desire heartbeats based on number of workers """ if n <= 10: return 0.5 elif n < 50: return 1 elif n < 200: return 2 else: # no more than 200 hearbeats a second scaled by workers return n / 200 + 1 class KilledWorker(Exception): def __init__(self, task, last_worker): super().__init__(task, last_worker) self.task = task self.last_worker = last_worker class WorkerStatusPlugin(SchedulerPlugin): """ An plugin to share worker status with a remote observer This is used in cluster managers to keep updated about the status of the scheduler. """ name = "worker-status" def __init__(self, scheduler, comm): self.bcomm = BatchedSend(interval="5ms") self.bcomm.start(comm) self.scheduler = scheduler self.scheduler.add_plugin(self) def add_worker(self, worker=None, **kwargs): ident = self.scheduler.workers[worker].identity() del ident["metrics"] del ident["last_seen"] try: self.bcomm.send(["add", {"workers": {worker: ident}}]) except CommClosedError: self.scheduler.remove_plugin(name=self.name) def remove_worker(self, worker=None, **kwargs): try: self.bcomm.send(["remove", worker]) except CommClosedError: self.scheduler.remove_plugin(name=self.name) def teardown(self): self.bcomm.close() class CollectTaskMetaDataPlugin(SchedulerPlugin): def __init__(self, scheduler, name): self.scheduler = scheduler self.name = name self.keys = set() self.metadata = {} self.state = {} def update_graph(self, scheduler, dsk=None, keys=None, restrictions=None, **kwargs): self.keys.update(keys) def transition(self, key, start, finish, *args, **kwargs): if finish == "memory" or finish == "erred": ts: TaskState = self.scheduler.tasks.get(key) if ts is not None and ts._key in self.keys: self.metadata[key] = ts._metadata self.state[key] = finish self.keys.discard(key)