usage_stats.py

from __future__ import annotations

import os
import typing as t
import logging
import secrets
import threading
import contextlib
from typing import TYPE_CHECKING
from datetime import datetime
from datetime import timezone
from functools import wraps
from functools import lru_cache

import attr
import requests
from simple_di import inject
from simple_di import Provide

from .schemas import EventMeta
from .schemas import ServeInitEvent
from .schemas import TrackingPayload
from .schemas import CommonProperties
from .schemas import ServeUpdateEvent
from ...configuration.containers import BentoMLContainer

if TYPE_CHECKING:
    P = t.ParamSpec("P")
    T = t.TypeVar("T")
    AsyncFunc = t.Callable[P, t.Coroutine[t.Any, t.Any, t.Any]]

    from bentoml import Service

    from ...server.metrics.prometheus import PrometheusClient

logger = logging.getLogger(__name__)

BENTOML_DO_NOT_TRACK = "BENTOML_DO_NOT_TRACK"
USAGE_TRACKING_URL = "https://t.bentoml.com"
SERVE_USAGE_TRACKING_INTERVAL_SECONDS = int(12 * 60 * 60)  # every 12 hours
USAGE_REQUEST_TIMEOUT_SECONDS = 1


@lru_cache(maxsize=1)
def do_not_track() -> bool:  # pragma: no cover
    # Returns True if and only if the environment variable is defined and has value True.
    # The function is cached for better performance.
    return os.environ.get(BENTOML_DO_NOT_TRACK, str(False)).lower() == "true"


@lru_cache(maxsize=1)
def _usage_event_debugging() -> bool:
    # For BentoML developers only - debug and print event payload if turned on
    return os.environ.get("__BENTOML_DEBUG_USAGE", str(False)).lower() == "true"


def silent(func: t.Callable[P, T]) -> t.Callable[P, T]:  # pragma: no cover
    # Silent errors when tracking
    @wraps(func)
    def wrapper(*args: P.args, **kwargs: P.kwargs) -> t.Any:
        try:
            return func(*args, **kwargs)
        except Exception as err:  # pylint: disable=broad-except
            if _usage_event_debugging():
                logger.info(f"Tracking Error: {err}")
            else:
                logger.debug(f"Tracking Error: {err}")

    return wrapper


@attr.define
class ServeInfo:
    serve_id: str
    serve_started_timestamp: datetime


def get_serve_info() -> ServeInfo:  # pragma: no cover
    # Returns a safe token for serve as well as timestamp of creating this token
    return ServeInfo(
        serve_id=secrets.token_urlsafe(32),
        serve_started_timestamp=datetime.now(timezone.utc),
    )


@inject
def get_payload(
    event_properties: EventMeta,
    session_id: str = Provide[BentoMLContainer.session_id],
) -> t.Dict[str, t.Any]:
    return TrackingPayload(
        session_id=session_id,
        common_properties=CommonProperties(),
        event_properties=event_properties,
        event_type=event_properties.event_name,
    ).to_dict()


@silent
def track(event_properties: EventMeta):
    if do_not_track():
        return
    payload = get_payload(event_properties=event_properties)

    if _usage_event_debugging():
        # For internal debugging purpose
        logger.info("Tracking Payload: %s", payload)
        return

    requests.post(
        USAGE_TRACKING_URL, json=payload, timeout=USAGE_REQUEST_TIMEOUT_SECONDS
    )


@inject
def _track_serve_init(
    svc: Service,
    production: bool,
    serve_info: ServeInfo = Provide[BentoMLContainer.serve_info],
):
    if svc.bento is not None:
        bento = svc.bento
        event_properties = ServeInitEvent(
            serve_id=serve_info.serve_id,
            serve_from_bento=True,
            production=production,
            bento_creation_timestamp=bento.info.creation_time,
            num_of_models=len(bento.info.models),
            num_of_runners=len(svc.runners),
            num_of_apis=len(bento.info.apis),
            model_types=[m.module for m in bento.info.models],
            runnable_types=[r.runnable_type for r in bento.info.runners],
            api_input_types=[api.input_type for api in bento.info.apis],
            api_output_types=[api.output_type for api in bento.info.apis],
        )
    else:
        event_properties = ServeInitEvent(
            serve_id=serve_info.serve_id,
            serve_from_bento=False,
            production=production,
            bento_creation_timestamp=None,
            num_of_models=len(
                set(
                    svc.models
                    + [model for runner in svc.runners for model in runner.models]
                )
            ),
            num_of_runners=len(svc.runners),
            num_of_apis=len(svc.apis.keys()),
            runnable_types=[r.runnable_class.__name__ for r in svc.runners],
            api_input_types=[api.input.__class__.__name__ for api in svc.apis.values()],
            api_output_types=[
                api.output.__class__.__name__ for api in svc.apis.values()
            ],
        )

    track(event_properties)


EXCLUDE_PATHS = {"/docs.json", "/livez", "/healthz", "/readyz"}


def get_metrics_report(
    metrics_client: PrometheusClient,
) -> list[dict[str, str | float]]:
    metrics_text = metrics_client.generate_latest().decode("utf-8")
    if not metrics_text:
        return []

    from prometheus_client.parser import (
        text_string_to_metric_families,  # type: ignore (unfinished prometheus types)
    )

    for metric in text_string_to_metric_families(metrics_text):
        # Searching for the metric BENTOML_{service_name}_request of type Counter
        if (
            metric.type == "counter"
            and metric.name.startswith("BENTOML_")
            and metric.name.endswith("_request")
        ):
            return [
                {**sample.labels, "value": sample.value}
                for sample in metric.samples
                if "endpoint" in sample.labels
                # exclude common infra paths
                and sample.labels["endpoint"] not in EXCLUDE_PATHS
                # exclude static_content prefix
                and not sample.labels["endpoint"].startswith("/static_content/")
            ]

    return []


@inject
@contextlib.contextmanager
def track_serve(
    svc: Service,
    production: bool,
    metrics_client: PrometheusClient = Provide[BentoMLContainer.metrics_client],
    serve_info: ServeInfo = Provide[BentoMLContainer.serve_info],
) -> t.Generator[None, None, None]:
    if do_not_track():
        yield
        return

    _track_serve_init(svc, production)

    if _usage_event_debugging():
        tracking_interval = 5
    else:
        tracking_interval = SERVE_USAGE_TRACKING_INTERVAL_SECONDS

    stop_event = threading.Event()

    @silent
    def loop() -> t.NoReturn:  # type: ignore
        last_tracked_timestamp: datetime = serve_info.serve_started_timestamp
        while not stop_event.wait(tracking_interval):  # pragma: no cover
            now = datetime.now(timezone.utc)
            event_properties = ServeUpdateEvent(
                serve_id=serve_info.serve_id,
                production=production,
                triggered_at=now,
                duration_in_seconds=int((now - last_tracked_timestamp).total_seconds()),
                metrics=get_metrics_report(metrics_client),
            )
            last_tracked_timestamp = now
            track(event_properties)

    tracking_thread = threading.Thread(target=loop, daemon=True)
    try:
        tracking_thread.start()
        yield
    finally:
        stop_event.set()
        tracking_thread.join()