Source code for distributed.diagnostics.plugin

from __future__ import annotations

import abc
import contextlib
import functools
import logging
import os
import socket
import subprocess
import sys
import uuid
import zipfile
from collections.abc import Awaitable
from typing import TYPE_CHECKING, Any, ClassVar

from dask.utils import funcname, tmpfile

if TYPE_CHECKING:
    from distributed.scheduler import Scheduler, TaskStateState  # circular imports

logger = logging.getLogger(__name__)


[docs]class SchedulerPlugin: """Interface to extend the Scheduler The scheduler operates by triggering and responding to events like ``task_finished``, ``update_graph``, ``task_erred``, etc.. A plugin enables custom code to run at each of those same events. The scheduler will run the analogous methods on this class when each event is triggered. This runs user code within the scheduler thread that can perform arbitrary operations in synchrony with the scheduler itself. Plugins are often used for diagnostics and measurement, but have full access to the scheduler and could in principle affect core scheduling. To implement a plugin: 1. subclass this class 2. override some of its methods 3. add the plugin to the scheduler with ``Scheduler.add_plugin(myplugin)``. Examples -------- >>> class Counter(SchedulerPlugin): ... def __init__(self): ... self.counter = 0 ... ... def transition(self, key, start, finish, *args, **kwargs): ... if start == 'processing' and finish == 'memory': ... self.counter += 1 ... ... def restart(self, scheduler): ... self.counter = 0 >>> plugin = Counter() >>> scheduler.add_plugin(plugin) # doctest: +SKIP """
[docs] async def start(self, scheduler: Scheduler) -> None: """Run when the scheduler starts up This runs at the end of the Scheduler startup process """
[docs] async def before_close(self) -> None: """Runs prior to any Scheduler shutdown logic"""
[docs] async def close(self) -> None: """Run when the scheduler closes down This runs at the beginning of the Scheduler shutdown process, but after workers have been asked to shut down gracefully """
[docs] def update_graph( self, scheduler: Scheduler, keys: set[str], restrictions: dict[str, float], **kwargs: Any, ) -> None: """Run when a new graph / tasks enter the scheduler"""
[docs] def restart(self, scheduler: Scheduler) -> None: """Run when the scheduler restarts itself"""
[docs] def transition( self, key: str, start: TaskStateState, finish: TaskStateState, *args: Any, **kwargs: Any, ) -> None: """Run whenever a task changes state Parameters ---------- key : string start : string Start state of the transition. One of released, waiting, processing, memory, error. finish : string Final state of the transition. *args, **kwargs : More options passed when transitioning This may include worker ID, compute time, etc. """
[docs] def add_worker(self, scheduler: Scheduler, worker: str) -> None | Awaitable[None]: """Run when a new worker enters the cluster"""
[docs] def remove_worker( self, scheduler: Scheduler, worker: str ) -> None | Awaitable[None]: """Run when a worker leaves the cluster"""
[docs] def add_client(self, scheduler: Scheduler, client: str) -> None: """Run when a new client connects"""
[docs] def remove_client(self, scheduler: Scheduler, client: str) -> None: """Run when a client disconnects"""
[docs] def log_event(self, topic: str, msg: Any) -> None: """Run when an event is logged"""
[docs]class WorkerPlugin: """Interface to extend the Worker A worker plugin enables custom code to run at different stages of the Workers' lifecycle: at setup, during task state transitions, when a task or dependency is released, and at teardown. A plugin enables custom code to run at each of step of a Workers's life. Whenever such an event happens, the corresponding method on this class will be called. Note that the user code always runs within the Worker's main thread. To implement a plugin implement some of the methods of this class and register the plugin to your client in order to have it attached to every existing and future workers with ``Client.register_worker_plugin``. Examples -------- >>> class ErrorLogger(WorkerPlugin): ... def __init__(self, logger): ... self.logger = logger ... ... def setup(self, worker): ... self.worker = worker ... ... def transition(self, key, start, finish, *args, **kwargs): ... if finish == 'error': ... ts = self.worker.tasks[key] ... exc_info = (type(ts.exception), ts.exception, ts.traceback) ... self.logger.error( ... "Error during computation of '%s'.", key, ... exc_info=exc_info ... ) >>> import logging >>> plugin = ErrorLogger(logging) >>> client.register_worker_plugin(plugin) # doctest: +SKIP """
[docs] def setup(self, worker): """ Run when the plugin is attached to a worker. This happens when the plugin is registered and attached to existing workers, or when a worker is created after the plugin has been registered. """
[docs] def teardown(self, worker): """Run when the worker to which the plugin is attached to is closed"""
[docs] def transition(self, key, start, finish, **kwargs): """ Throughout the lifecycle of a task (see :doc:`Worker <worker>`), Workers are instructed by the scheduler to compute certain tasks, resulting in transitions in the state of each task. The Worker owning the task is then notified of this state transition. Whenever a task changes its state, this method will be called. Parameters ---------- key : string start : string Start state of the transition. One of waiting, ready, executing, long-running, memory, error. finish : string Final state of the transition. kwargs : More options passed when transitioning """
[docs]class NannyPlugin: """Interface to extend the Nanny A worker plugin enables custom code to run at different stages of the Workers' lifecycle. A nanny plugin does the same thing, but benefits from being able to run code before the worker is started, or to restart the worker if necessary. To implement a plugin implement some of the methods of this class and register the plugin to your client in order to have it attached to every existing and future nanny by passing ``nanny=True`` to :meth:`Client.register_worker_plugin<distributed.Client.register_worker_plugin>`. The ``restart`` attribute is used to control whether or not a running ``Worker`` needs to be restarted when registering the plugin. See Also -------- WorkerPlugin SchedulerPlugin """ restart = False
[docs] def setup(self, nanny): """ Run when the plugin is attached to a nanny. This happens when the plugin is registered and attached to existing nannies, or when a nanny is created after the plugin has been registered. """
[docs] def teardown(self, nanny): """Run when the nanny to which the plugin is attached to is closed"""
def _get_plugin_name(plugin: SchedulerPlugin | WorkerPlugin | NannyPlugin) -> str: """Return plugin name. If plugin has no name attribute a random name is used. """ if hasattr(plugin, "name"): return plugin.name else: return funcname(type(plugin)) + "-" + str(uuid.uuid4()) class PackageInstall(WorkerPlugin, abc.ABC): """Abstract parent class for a worker plugin to install a set of packages This accepts a set of packages to install on all workers. You can also optionally ask for the worker to restart itself after performing this installation. .. note:: This will increase the time it takes to start up each worker. If possible, we recommend including the libraries in the worker environment or image. This is primarily intended for experimentation and debugging. Parameters ---------- packages A list of packages (with optional versions) to install restart Whether or not to restart the worker after installing the packages Only functions if the worker has an attached nanny process See Also -------- CondaInstall PipInstall """ INSTALLER: ClassVar[str] name: str packages: list[str] restart: bool def __init__( self, packages: list[str], restart: bool, ): self.packages = packages self.restart = restart self.name = f"{self.INSTALLER}-install-{uuid.uuid4()}" async def setup(self, worker): from distributed.semaphore import Semaphore async with ( await Semaphore(max_leases=1, name=socket.gethostname(), register=True) ): if not await self._is_installed(worker): logger.info( "%s installing the following packages: %s", self.INSTALLER, self.packages, ) await self._set_installed(worker) self.install() else: logger.info( "The following packages have already been installed: %s", self.packages, ) if self.restart and worker.nanny and not await self._is_restarted(worker): logger.info("Restarting worker to refresh interpreter.") await self._set_restarted(worker) worker.loop.add_callback( worker.close_gracefully, restart=True, reason=f"{self.name}-setup" ) @abc.abstractmethod def install(self) -> None: """Install the requested packages""" async def _is_installed(self, worker): return await worker.client.get_metadata( self._compose_installed_key(), default=False ) async def _set_installed(self, worker): await worker.client.set_metadata( self._compose_installed_key(), True, ) def _compose_installed_key(self): return [ self.name, "installed", socket.gethostname(), ] async def _is_restarted(self, worker): return await worker.client.get_metadata( self._compose_restarted_key(worker), default=False, ) async def _set_restarted(self, worker): await worker.client.set_metadata( self._compose_restarted_key(worker), True, ) def _compose_restarted_key(self, worker): return [self.name, "restarted", worker.nanny]
[docs]class CondaInstall(PackageInstall): """A Worker Plugin to conda install a set of packages This accepts a set of packages to install on all workers as well as options to use when installing. You can also optionally ask for the worker to restart itself after performing this installation. .. note:: This will increase the time it takes to start up each worker. If possible, we recommend including the libraries in the worker environment or image. This is primarily intended for experimentation and debugging. Parameters ---------- packages A list of packages (with optional versions) to install using conda conda_options Additional options to pass to conda restart Whether or not to restart the worker after installing the packages Only functions if the worker has an attached nanny process Examples -------- >>> from dask.distributed import CondaInstall >>> plugin = CondaInstall(packages=["scikit-learn"], conda_options=["--update-deps"]) >>> client.register_worker_plugin(plugin) See Also -------- PackageInstall PipInstall """ INSTALLER = "conda" conda_options: list[str] def __init__( self, packages: list[str], conda_options: list[str] | None = None, restart: bool = False, ): super().__init__(packages, restart=restart) self.conda_options = conda_options or [] def install(self) -> None: try: from conda.cli.python_api import Commands, run_command except ModuleNotFoundError as e: # pragma: nocover msg = ( "conda install failed because conda could not be found. " "Please make sure that conda is installed." ) logger.error(msg) raise RuntimeError(msg) from e try: _, stderr, returncode = run_command( Commands.INSTALL, self.conda_options + self.packages ) except Exception as e: msg = "conda install failed" logger.error(msg) raise RuntimeError(msg) from e if returncode != 0: msg = f"conda install failed with '{stderr.decode().strip()}'" logger.error(msg) raise RuntimeError(msg)
[docs]class PipInstall(PackageInstall): """A Worker Plugin to pip install a set of packages This accepts a set of packages to install on all workers as well as options to use when installing. You can also optionally ask for the worker to restart itself after performing this installation. .. note:: This will increase the time it takes to start up each worker. If possible, we recommend including the libraries in the worker environment or image. This is primarily intended for experimentation and debugging. Parameters ---------- packages A list of packages (with optional versions) to install using pip pip_options Additional options to pass to pip restart Whether or not to restart the worker after installing the packages Only functions if the worker has an attached nanny process Examples -------- >>> from dask.distributed import PipInstall >>> plugin = PipInstall(packages=["scikit-learn"], pip_options=["--upgrade"]) >>> client.register_worker_plugin(plugin) See Also -------- PackageInstall CondaInstall """ INSTALLER = "pip" pip_options: list[str] def __init__( self, packages: list[str], pip_options: list[str] | None = None, restart: bool = False, ): super().__init__(packages, restart=restart) self.pip_options = pip_options or [] def install(self) -> None: proc = subprocess.Popen( [sys.executable, "-m", "pip", "install"] + self.pip_options + self.packages, stdout=subprocess.PIPE, stderr=subprocess.PIPE, ) _, stderr = proc.communicate() returncode = proc.wait() if returncode != 0: msg = f"pip install failed with '{stderr.decode().strip()}'" logger.error(msg) raise RuntimeError(msg)
# Adapted from https://github.com/dask/distributed/issues/3560#issuecomment-596138522
[docs]class UploadFile(WorkerPlugin): """A WorkerPlugin to upload a local file to workers. Parameters ---------- filepath: str A path to the file (.py, egg, or zip) to upload Examples -------- >>> from distributed.diagnostics.plugin import UploadFile >>> client.register_worker_plugin(UploadFile("/path/to/file.py")) # doctest: +SKIP """ name = "upload_file" def __init__(self, filepath): """ Initialize the plugin by reading in the data from the given file. """ self.filename = os.path.basename(filepath) with open(filepath, "rb") as f: self.data = f.read() async def setup(self, worker): response = await worker.upload_file( filename=self.filename, data=self.data, load=True ) assert len(self.data) == response["nbytes"]
[docs]class Environ(NannyPlugin): restart = True def __init__(self, environ: dict | None = None): environ = environ or {} self.environ = {k: str(v) for k, v in environ.items()} async def setup(self, nanny): nanny.env.update(self.environ)
[docs]class UploadDirectory(NannyPlugin): """A NannyPlugin to upload a local file to workers. Parameters ---------- path: str A path to the directory to upload Examples -------- >>> from distributed.diagnostics.plugin import UploadDirectory >>> client.register_worker_plugin(UploadDirectory("/path/to/directory"), nanny=True) # doctest: +SKIP """ def __init__( self, path, restart=False, update_path=False, skip_words=(".git", ".github", ".pytest_cache", "tests", "docs"), skip=(lambda fn: os.path.splitext(fn)[1] == ".pyc",), ): """ Initialize the plugin by reading in the data from the given file. """ path = os.path.expanduser(path) self.path = os.path.split(path)[-1] self.restart = restart self.update_path = update_path self.name = "upload-directory-" + os.path.split(path)[-1] with tmpfile(extension="zip") as fn: with zipfile.ZipFile(fn, "w", zipfile.ZIP_DEFLATED) as z: for root, dirs, files in os.walk(path): for file in files: filename = os.path.join(root, file) if any(predicate(filename) for predicate in skip): continue dirs = filename.split(os.sep) if any(word in dirs for word in skip_words): continue archive_name = os.path.relpath( os.path.join(root, file), os.path.join(path, "..") ) z.write(filename, archive_name) with open(fn, "rb") as f: self.data = f.read() async def setup(self, nanny): fn = os.path.join(nanny.local_directory, f"tmp-{uuid.uuid4()}.zip") with open(fn, "wb") as f: f.write(self.data) import zipfile with zipfile.ZipFile(fn) as z: z.extractall(path=nanny.local_directory) if self.update_path: path = os.path.join(nanny.local_directory, self.path) if path not in sys.path: sys.path.insert(0, path) os.remove(fn)
class forward_stream: def __init__(self, stream, worker): self._worker = worker self._original_methods = {} self._stream = getattr(sys, stream) if stream == "stdout": self._file = 1 elif stream == "stderr": self._file = 2 else: raise ValueError( f"Expected stream to be 'stdout' or 'stderr'; got '{stream}'" ) self._file = 1 if stream == "stdout" else 2 self._buffer = [] def _write(self, write_fn, data): self._forward(data) write_fn(data) def _forward(self, data): self._buffer.append(data) # Mimic line buffering if "\n" in data or "\r" in data: self._send() def _send(self): msg = {"args": self._buffer, "file": self._file, "sep": "", "end": ""} self._worker.log_event("print", msg) self._buffer = [] def _flush(self, flush_fn): self._send() flush_fn() def _close(self, close_fn): self._send() close_fn() def _intercept(self, method_name, interceptor): original_method = getattr(self._stream, method_name) self._original_methods[method_name] = original_method setattr( self._stream, method_name, functools.partial(interceptor, original_method) ) def __enter__(self): self._intercept("write", self._write) self._intercept("flush", self._flush) self._intercept("close", self._close) return self._stream def __exit__(self, exc_type, exc_value, traceback): self._stream.flush() for attr, original in self._original_methods.items(): setattr(self._stream, attr, original) self._original_methods = {} class ForwardOutput(WorkerPlugin): """A Worker Plugin that forwards ``stdout`` and ``stderr`` from workers to clients This plugin forwards all output sent to ``stdout`` and ``stderr` on all workers to all clients where it is written to the respective streams. Analogous to the terminal, this plugin uses line buffering. To ensure that an output is written without a newline, make sure to flush the stream. .. warning:: Using this plugin will forward **all** output in ``stdout`` and ``stderr`` from every worker to every client. If the output is very chatty, this will add significant strain on the scheduler. Proceed with caution! Examples -------- >>> from dask.distributed import ForwardOutput >>> plugin = ForwardOutput() >>> client.register_worker_plugin(plugin) """ def setup(self, worker): self._exit_stack = contextlib.ExitStack() self._exit_stack.enter_context(forward_stream("stdout", worker=worker)) self._exit_stack.enter_context(forward_stream("stderr", worker=worker)) def teardown(self, worker): self._exit_stack.close()