162 lines
5.0 KiB
Python

"""Basic definitions for structuring experiments."""
import logging
import random
from abc import ABC, abstractmethod
from collections.abc import Iterable
from typing import List, Optional
from typing_extensions import Generic, TypeVar
from benchmarks.core.config import Builder
from benchmarks.core.utils import await_predicate
logger = logging.getLogger(__name__)
class Experiment(ABC):
"""Base interface for an executable :class:`Experiment`."""
@abstractmethod
def run(self):
"""Synchronously runs the experiment, blocking the current thread until it's done."""
pass
TExperiment = TypeVar("TExperiment", bound=Experiment)
ExperimentBuilder = Builder[TExperiment]
class ExperimentWithLifecycle(Experiment):
"""An :class:`ExperimentWithLifecycle` is a basic implementation of an :class:`Experiment` with overridable
lifecycle hooks."""
def setup(self):
"""Hook that runs before the experiment."""
pass
def run(self):
try:
self.setup()
self.do_run()
self.teardown()
except Exception as ex:
self.teardown(ex)
raise ex
def do_run(self):
"""The main body of the experiment."""
pass
def teardown(self, exception: Optional[Exception] = None):
"""Hook that runs after the experiment."""
pass
class ExperimentComponent(ABC):
"""An :class:`ExperimentComponent` is a part of the environment for an experiment. These could be databases,
network nodes, etc."""
@abstractmethod
def is_ready(self) -> bool:
"""Returns whether this component is ready or not."""
pass
class ExperimentEnvironment(ExperimentComponent):
"""An :class:`ExperimentEnvironment` is a collection of :class:`ExperimentComponent`s that must be ready before
an :class:`Experiment` can execute. Note that we assume that readiness is stable; i.e., if a component is ready
at some point, then it will remain ready for the duration of the experiment."""
def __init__(
self,
components: Iterable[ExperimentComponent],
ping_max: int = 10,
polling_interval: float = 0,
):
self.components = components
self.polling_interval = polling_interval
self.ping_max = ping_max
self.not_ready = list(components)
def await_ready(self, timeout: float = 0) -> bool:
"""Awaits for all components to be ready, or until a timeout is reached."""
logging.info(
f"Awaiting for components to be ready: {self._component_names(self.not_ready)}"
)
if not await_predicate(self.is_ready, timeout, self.polling_interval):
logger.info(
f"Some components timed out: {self._component_names(self.not_ready)}"
)
return False
return True
def is_ready(self) -> bool:
for component in self._draw(self.not_ready):
if component.is_ready():
logger.info(f"Component {str(component)} is ready.")
self.not_ready.remove(component)
return len(self.not_ready) == 0
def _draw(self, components: List[ExperimentComponent]) -> List[ExperimentComponent]:
if len(components) <= self.ping_max:
return components
random.shuffle(components)
return components[: self.ping_max]
@staticmethod
def _component_names(components: List[ExperimentComponent]) -> str:
return ", ".join(str(component) for component in components)
def run(self, experiment: Experiment):
"""Runs the :class:`Experiment` within this :class:`ExperimentEnvironment`."""
if not self.await_ready():
raise RuntimeError(
"One or more environment components were not get ready in time"
)
experiment.run()
def bind(self, experiment: TExperiment) -> "BoundExperiment[TExperiment]":
return BoundExperiment(experiment, self)
class BoundExperiment(Experiment, Generic[TExperiment]):
def __init__(self, experiment: Experiment, env: ExperimentEnvironment):
self.experiment = experiment
self.env = env
def run(self):
self.env.run(self.experiment)
class IteratedExperiment(Experiment, Generic[TExperiment]):
"""An :class:`IteratedExperiment` will run a sequence of :class:`Experiment`s."""
def __init__(
self, experiments: Iterable[TExperiment], raise_when_failures: bool = True
):
self.successful_runs = 0
self.failed_runs = 0
self.raise_when_failures = raise_when_failures
self.experiments = experiments
def run(self):
for experiment in self.experiments:
try:
experiment.run()
self.successful_runs += 1
except Exception:
self.failed_runs += 1
logger.exception("Error running experiment repetition")
if self.failed_runs > 0 and self.raise_when_failures:
raise RuntimeError(
"One or more experiments with an iterated experiment have failed."
)