from __future__ import annotations import contextlib import inspect import json import sys import warnings from functools import cached_property from time import time from typing import ( TYPE_CHECKING, Any, Callable, ContextManager, Iterable, Literal, Sequence, TypeVar, Union, cast, overload, ) import opentelemetry.context as context_api import opentelemetry.trace as trace_api from opentelemetry.context import Context from opentelemetry.metrics import CallbackT, Counter, Histogram, UpDownCounter from opentelemetry.sdk.trace import ReadableSpan, Span from opentelemetry.trace import SpanContext, Tracer from opentelemetry.util import types as otel_types from typing_extensions import LiteralString, ParamSpec from ..version import VERSION from . import async_ from .auto_trace import AutoTraceModule, install_auto_tracing from .config import GLOBAL_CONFIG, LogfireConfig from .config_params import PydanticPluginRecordValues from .constants import ( ATTRIBUTES_JSON_SCHEMA_KEY, ATTRIBUTES_LOG_LEVEL_NUM_KEY, ATTRIBUTES_MESSAGE_KEY, ATTRIBUTES_MESSAGE_TEMPLATE_KEY, ATTRIBUTES_SAMPLE_RATE_KEY, ATTRIBUTES_SPAN_TYPE_KEY, ATTRIBUTES_TAGS_KEY, DISABLE_CONSOLE_KEY, LEVEL_NUMBERS, NULL_ARGS_KEY, OTLP_MAX_INT_SIZE, LevelName, log_level_attributes, ) from .formatter import logfire_format, logfire_format_with_magic from .instrument import instrument from .json_encoder import logfire_json_dumps from .json_schema import ( JsonSchemaProperties, attributes_json_schema, attributes_json_schema_properties, create_json_schema, ) from .metrics import ProxyMeterProvider from .stack_info import get_user_stack_info from .tracer import ProxyTracerProvider, record_exception, set_exception_status from .utils import get_version, handle_internal_errors, log_internal_error, uniquify_sequence if TYPE_CHECKING: from types import ModuleType from wsgiref.types import WSGIApplication import anthropic import httpx import openai import requests from django.http import HttpRequest, HttpResponse from fastapi import FastAPI from flask.app import Flask from opentelemetry.instrumentation.asgi.types import ClientRequestHook, ClientResponseHook, ServerRequestHook from opentelemetry.metrics import _Gauge as Gauge from pymongo.monitoring import CommandFailedEvent, CommandStartedEvent, CommandSucceededEvent from sqlalchemy import Engine from sqlalchemy.ext.asyncio import AsyncEngine from starlette.applications import Starlette from starlette.requests import Request from starlette.websockets import WebSocket from typing_extensions import Unpack from ..integrations.flask import ( CommenterOptions as FlaskCommenterOptions, RequestHook as FlaskRequestHook, ResponseHook as FlaskResponseHook, ) from ..integrations.httpx import ( AsyncRequestHook as HttpxAsyncRequestHook, AsyncResponseHook as HttpxAsyncResponseHook, RequestHook as HttpxRequestHook, ResponseHook as HttpxResponseHook, ) from ..integrations.psycopg import CommenterOptions as PsycopgCommenterOptions from ..integrations.redis import RequestHook as RedisRequestHook, ResponseHook as RedisResponseHook from ..integrations.sqlalchemy import CommenterOptions as SQLAlchemyCommenterOptions from ..integrations.wsgi import RequestHook as WSGIRequestHook, ResponseHook as WSGIResponseHook from .integrations.asgi import ASGIApp, ASGIInstrumentKwargs from .integrations.aws_lambda import LambdaEvent, LambdaHandler from .integrations.mysql import MySQLConnection from .integrations.psycopg import Psycopg2Connection, PsycopgConnection from .integrations.sqlite3 import SQLite3Connection from .integrations.system_metrics import Base as SystemMetricsBase, Config as SystemMetricsConfig from .utils import SysExcInfo # This is the type of the exc_info/_exc_info parameter of the log methods. # sys.exc_info() returns a tuple of (type, value, traceback) or (None, None, None). # We just need the exception, but we allow the user to pass the tuple because: # 1. It's convenient to pass the result of sys.exc_info() directly # 2. It mirrors the exc_info argument of the stdlib logging methods # 3. The argument name exc_info is very suggestive of the sys function. ExcInfo = Union[SysExcInfo, BaseException, bool, None] class Logfire: """The main logfire class.""" def __init__( self, *, config: LogfireConfig = GLOBAL_CONFIG, sample_rate: float | None = None, tags: Sequence[str] = (), console_log: bool = True, otel_scope: str = 'logfire', ) -> None: self._tags = tuple(tags) self._config = config self._sample_rate = sample_rate self._console_log = console_log self._otel_scope = otel_scope @property def config(self) -> LogfireConfig: return self._config @cached_property def _tracer_provider(self) -> ProxyTracerProvider: self._config.warn_if_not_initialized('No logs or spans will be created') return self._config.get_tracer_provider() @cached_property def _meter_provider(self) -> ProxyMeterProvider: # pragma: no cover return self._config.get_meter_provider() @cached_property def _meter(self): return self._meter_provider.get_meter(self._otel_scope, VERSION) @cached_property def _logs_tracer(self) -> Tracer: return self._get_tracer(is_span_tracer=False) @cached_property def _spans_tracer(self) -> Tracer: return self._get_tracer(is_span_tracer=True) def _get_tracer(self, *, is_span_tracer: bool) -> Tracer: # pragma: no cover return self._tracer_provider.get_tracer( self._otel_scope, VERSION, is_span_tracer=is_span_tracer, ) # If any changes are made to this method, they may need to be reflected in `_fast_span` as well. def _span( self, msg_template: str, attributes: dict[str, Any], *, _tags: Sequence[str] | None = None, _span_name: str | None = None, _level: LevelName | int | None = None, _links: Sequence[tuple[SpanContext, otel_types.Attributes]] = (), ) -> LogfireSpan: try: stack_info = get_user_stack_info() merged_attributes = {**stack_info, **attributes} if self._config.inspect_arguments: fstring_frame = inspect.currentframe().f_back # type: ignore else: fstring_frame = None log_message, extra_attrs, msg_template = logfire_format_with_magic( msg_template, merged_attributes, self._config.scrubber, fstring_frame=fstring_frame, ) merged_attributes.update(extra_attrs) attributes.update(extra_attrs) # for the JSON schema merged_attributes[ATTRIBUTES_MESSAGE_TEMPLATE_KEY] = msg_template merged_attributes[ATTRIBUTES_MESSAGE_KEY] = log_message otlp_attributes = prepare_otlp_attributes(merged_attributes) if json_schema_properties := attributes_json_schema_properties(attributes): otlp_attributes[ATTRIBUTES_JSON_SCHEMA_KEY] = attributes_json_schema(json_schema_properties) tags = (self._tags or ()) + tuple(_tags or ()) if tags: otlp_attributes[ATTRIBUTES_TAGS_KEY] = uniquify_sequence(tags) sample_rate = ( self._sample_rate if self._sample_rate is not None else otlp_attributes.pop(ATTRIBUTES_SAMPLE_RATE_KEY, None) ) if sample_rate is not None and sample_rate != 1: # pragma: no cover otlp_attributes[ATTRIBUTES_SAMPLE_RATE_KEY] = sample_rate if _level is not None: otlp_attributes.update(log_level_attributes(_level)) return LogfireSpan( _span_name or msg_template, otlp_attributes, self._spans_tracer, json_schema_properties, links=_links, ) except Exception: log_internal_error() return NoopSpan() # type: ignore def _fast_span(self, name: str, attributes: otel_types.Attributes) -> FastLogfireSpan: """A simple version of `_span` optimized for auto-tracing that doesn't support message formatting. Returns a similarly simplified version of `LogfireSpan` which must immediately be used as a context manager. """ try: span = self._spans_tracer.start_span(name=name, attributes=attributes) return FastLogfireSpan(span) except Exception: # pragma: no cover log_internal_error() return NoopSpan() # type: ignore def _instrument_span_with_args( self, name: str, attributes: dict[str, otel_types.AttributeValue], function_args: dict[str, Any] ) -> FastLogfireSpan: """A version of `_span` used by `@instrument` with `extract_args=True`. This is a bit faster than `_span` but not as fast as `_fast_span` because it supports message formatting and arbitrary types of attributes. """ try: msg_template: str = attributes[ATTRIBUTES_MESSAGE_TEMPLATE_KEY] # type: ignore attributes[ATTRIBUTES_MESSAGE_KEY] = logfire_format(msg_template, function_args, self._config.scrubber) if json_schema_properties := attributes_json_schema_properties(function_args): # pragma: no branch attributes[ATTRIBUTES_JSON_SCHEMA_KEY] = attributes_json_schema(json_schema_properties) attributes.update(prepare_otlp_attributes(function_args)) return self._fast_span(name, attributes) except Exception: # pragma: no cover log_internal_error() return NoopSpan() # type: ignore def trace( self, msg_template: str, /, *, _tags: Sequence[str] | None = None, _exc_info: ExcInfo = False, **attributes: Any, ) -> None: """Log a trace message. ```py import logfire logfire.configure() logfire.trace('This is a trace log') ``` Args: msg_template: The message to log. attributes: The attributes to bind to the log. _tags: An optional sequence of tags to include in the log. _exc_info: Set to an exception or a tuple as returned by [`sys.exc_info()`][sys.exc_info] to record a traceback with the log message. Set to `True` to use the currently handled exception. """ if any(k.startswith('_') for k in attributes): raise ValueError('Attribute keys cannot start with an underscore.') self.log('trace', msg_template, attributes, tags=_tags, exc_info=_exc_info) def debug( self, msg_template: str, /, *, _tags: Sequence[str] | None = None, _exc_info: ExcInfo = False, **attributes: Any, ) -> None: """Log a debug message. ```py import logfire logfire.configure() logfire.debug('This is a debug log') ``` Args: msg_template: The message to log. attributes: The attributes to bind to the log. _tags: An optional sequence of tags to include in the log. _exc_info: Set to an exception or a tuple as returned by [`sys.exc_info()`][sys.exc_info] to record a traceback with the log message. Set to `True` to use the currently handled exception. """ if any(k.startswith('_') for k in attributes): raise ValueError('Attribute keys cannot start with an underscore.') self.log('debug', msg_template, attributes, tags=_tags, exc_info=_exc_info) def info( self, msg_template: str, /, *, _tags: Sequence[str] | None = None, _exc_info: ExcInfo = False, **attributes: Any, ) -> None: """Log an info message. ```py import logfire logfire.configure() logfire.info('This is an info log') ``` Args: msg_template: The message to log. attributes: The attributes to bind to the log. _tags: An optional sequence of tags to include in the log. _exc_info: Set to an exception or a tuple as returned by [`sys.exc_info()`][sys.exc_info] to record a traceback with the log message. Set to `True` to use the currently handled exception. """ if any(k.startswith('_') for k in attributes): raise ValueError('Attribute keys cannot start with an underscore.') self.log('info', msg_template, attributes, tags=_tags, exc_info=_exc_info) def notice( self, msg_template: str, /, *, _tags: Sequence[str] | None = None, _exc_info: ExcInfo = False, **attributes: Any, ) -> None: """Log a notice message. ```py import logfire logfire.configure() logfire.notice('This is a notice log') ``` Args: msg_template: The message to log. attributes: The attributes to bind to the log. _tags: An optional sequence of tags to include in the log. _exc_info: Set to an exception or a tuple as returned by [`sys.exc_info()`][sys.exc_info] to record a traceback with the log message. Set to `True` to use the currently handled exception. """ if any(k.startswith('_') for k in attributes): raise ValueError('Attribute keys cannot start with an underscore.') self.log('notice', msg_template, attributes, tags=_tags, exc_info=_exc_info) def warn( self, msg_template: str, /, *, _tags: Sequence[str] | None = None, _exc_info: ExcInfo = False, **attributes: Any, ) -> None: """Log a warning message. ```py import logfire logfire.configure() logfire.warn('This is a warning log') ``` Args: msg_template: The message to log. attributes: The attributes to bind to the log. _tags: An optional sequence of tags to include in the log. _exc_info: Set to an exception or a tuple as returned by [`sys.exc_info()`][sys.exc_info] to record a traceback with the log message. Set to `True` to use the currently handled exception. """ if any(k.startswith('_') for k in attributes): raise ValueError('Attribute keys cannot start with an underscore.') self.log('warn', msg_template, attributes, tags=_tags, exc_info=_exc_info) def error( self, msg_template: str, /, *, _tags: Sequence[str] | None = None, _exc_info: ExcInfo = False, **attributes: Any, ) -> None: """Log an error message. ```py import logfire logfire.configure() logfire.error('This is an error log') ``` Args: msg_template: The message to log. attributes: The attributes to bind to the log. _tags: An optional sequence of tags to include in the log. _exc_info: Set to an exception or a tuple as returned by [`sys.exc_info()`][sys.exc_info] to record a traceback with the log message. Set to `True` to use the currently handled exception. """ if any(k.startswith('_') for k in attributes): raise ValueError('Attribute keys cannot start with an underscore.') self.log('error', msg_template, attributes, tags=_tags, exc_info=_exc_info) def fatal( self, msg_template: str, /, *, _tags: Sequence[str] | None = None, _exc_info: ExcInfo = False, **attributes: Any, ) -> None: """Log a fatal message. ```py import logfire logfire.configure() logfire.fatal('This is a fatal log') ``` Args: msg_template: The message to log. attributes: The attributes to bind to the log. _tags: An optional sequence of tags to include in the log. _exc_info: Set to an exception or a tuple as returned by [`sys.exc_info()`][sys.exc_info] to record a traceback with the log message. Set to `True` to use the currently handled exception. """ if any(k.startswith('_') for k in attributes): raise ValueError('Attribute keys cannot start with an underscore.') self.log('fatal', msg_template, attributes, tags=_tags, exc_info=_exc_info) def exception( self, msg_template: str, /, *, _tags: Sequence[str] | None = None, _exc_info: ExcInfo = True, **attributes: Any, ) -> None: """The same as `error` but with `_exc_info=True` by default. This means that a traceback will be logged for any currently handled exception. Args: msg_template: The message to log. attributes: The attributes to bind to the log. _tags: An optional sequence of tags to include in the log. _exc_info: Set to an exception or a tuple as returned by [`sys.exc_info()`][sys.exc_info] to record a traceback with the log message. """ if any(k.startswith('_') for k in attributes): # pragma: no cover raise ValueError('Attribute keys cannot start with an underscore.') self.log('error', msg_template, attributes, tags=_tags, exc_info=_exc_info) def span( self, msg_template: str, /, *, _tags: Sequence[str] | None = None, _span_name: str | None = None, _level: LevelName | None = None, _links: Sequence[tuple[SpanContext, otel_types.Attributes]] = (), **attributes: Any, ) -> LogfireSpan: """Context manager for creating a span. ```py import logfire logfire.configure() with logfire.span('This is a span {a=}', a='data'): logfire.info('new log 1') ``` Args: msg_template: The template for the span message. _span_name: The span name. If not provided, the `msg_template` will be used. _tags: An optional sequence of tags to include in the span. _level: An optional log level name. _links: An optional sequence of links to other spans. Each link is a tuple of a span context and attributes. attributes: The arguments to include in the span and format the message template with. Attributes starting with an underscore are not allowed. """ if any(k.startswith('_') for k in attributes): raise ValueError('Attribute keys cannot start with an underscore.') return self._span( msg_template, attributes, _tags=_tags, _span_name=_span_name, _level=_level, _links=_links, ) @overload def instrument( self, msg_template: LiteralString | None = None, *, span_name: str | None = None, extract_args: bool | Iterable[str] = True, allow_generator: bool = False, ) -> Callable[[Callable[P, R]], Callable[P, R]]: """Decorator for instrumenting a function as a span. ```py import logfire logfire.configure() @logfire.instrument('This is a span {a=}') def my_function(a: int): logfire.info('new log {a=}', a=a) ``` Args: msg_template: The template for the span message. If not provided, the module and function name will be used. span_name: The span name. If not provided, the `msg_template` will be used. extract_args: By default, all function call arguments are logged as span attributes. Set to `False` to disable this, or pass an iterable of argument names to include. allow_generator: Set to `True` to prevent a warning when instrumenting a generator function. Read https://logfire.pydantic.dev/docs/guides/advanced/generators/#using-logfireinstrument first. """ @overload def instrument(self, func: Callable[P, R]) -> Callable[P, R]: """Decorator for instrumenting a function as a span, with default configuration. ```py import logfire logfire.configure() @logfire.instrument def my_function(a: int): logfire.info('new log {a=}', a=a) ``` """ def instrument( # type: ignore[reportInconsistentOverload] self, msg_template: Callable[P, R] | LiteralString | None = None, *, span_name: str | None = None, extract_args: bool | Iterable[str] = True, allow_generator: bool = False, ) -> Callable[[Callable[P, R]], Callable[P, R]] | Callable[P, R]: """Decorator for instrumenting a function as a span. ```py import logfire logfire.configure() @logfire.instrument('This is a span {a=}') def my_function(a: int): logfire.info('new log {a=}', a=a) ``` Args: msg_template: The template for the span message. If not provided, the module and function name will be used. span_name: The span name. If not provided, the `msg_template` will be used. extract_args: By default, all function call arguments are logged as span attributes. Set to `False` to disable this, or pass an iterable of argument names to include. allow_generator: Set to `True` to prevent a warning when instrumenting a generator function. Read https://logfire.pydantic.dev/docs/guides/advanced/generators/#using-logfireinstrument first. """ if callable(msg_template): return self.instrument()(msg_template) return instrument(self, tuple(self._tags), msg_template, span_name, extract_args, allow_generator) def log( self, level: LevelName | int, msg_template: str, attributes: dict[str, Any] | None = None, tags: Sequence[str] | None = None, exc_info: ExcInfo = False, console_log: bool | None = None, ) -> None: """Log a message. ```py import logfire logfire.configure() logfire.log('info', 'This is a log {a}', {'a': 'Apple'}) ``` Args: level: The level of the log. msg_template: The message to log. attributes: The attributes to bind to the log. tags: An optional sequence of tags to include in the log. exc_info: Set to an exception or a tuple as returned by [`sys.exc_info()`][sys.exc_info] to record a traceback with the log message. Set to `True` to use the currently handled exception. console_log: Whether to log to the console, defaults to `True`. """ with handle_internal_errors(): stack_info = get_user_stack_info() attributes = attributes or {} merged_attributes = {**stack_info, **attributes} if (msg := attributes.pop(ATTRIBUTES_MESSAGE_KEY, None)) is None: fstring_frame = None if self._config.inspect_arguments: fstring_frame = inspect.currentframe() if fstring_frame.f_back.f_code.co_filename == Logfire.log.__code__.co_filename: # type: ignore # fstring_frame.f_back should be the user's frame. # The user called logfire.info or a similar method rather than calling logfire.log directly. fstring_frame = fstring_frame.f_back # type: ignore msg, extra_attrs, msg_template = logfire_format_with_magic( msg_template, merged_attributes, self._config.scrubber, fstring_frame=fstring_frame, ) if extra_attrs: merged_attributes.update(extra_attrs) # Only do this if extra_attrs is not empty since the copy of `attributes` might be expensive. # We update both because attributes_json_schema_properties looks at `attributes`. attributes = {**attributes, **extra_attrs} else: # The message has already been filled in, presumably by a logging integration. # Make sure it's a string. msg = merged_attributes[ATTRIBUTES_MESSAGE_KEY] = str(msg) msg_template = str(msg_template) otlp_attributes = prepare_otlp_attributes(merged_attributes) otlp_attributes = { ATTRIBUTES_SPAN_TYPE_KEY: 'log', **log_level_attributes(level), ATTRIBUTES_MESSAGE_TEMPLATE_KEY: msg_template, ATTRIBUTES_MESSAGE_KEY: msg, **otlp_attributes, } if json_schema_properties := attributes_json_schema_properties(attributes): otlp_attributes[ATTRIBUTES_JSON_SCHEMA_KEY] = attributes_json_schema(json_schema_properties) tags = self._tags + tuple(tags or ()) if tags: otlp_attributes[ATTRIBUTES_TAGS_KEY] = uniquify_sequence(tags) sample_rate = ( self._sample_rate if self._sample_rate is not None else otlp_attributes.pop(ATTRIBUTES_SAMPLE_RATE_KEY, None) ) if sample_rate is not None and sample_rate != 1: # pragma: no cover otlp_attributes[ATTRIBUTES_SAMPLE_RATE_KEY] = sample_rate if not (self._console_log if console_log is None else console_log): otlp_attributes[DISABLE_CONSOLE_KEY] = True start_time = self._config.advanced.ns_timestamp_generator() span = self._logs_tracer.start_span( msg_template, attributes=otlp_attributes, start_time=start_time, ) if exc_info: if exc_info is True: exc_info = sys.exc_info() if isinstance(exc_info, tuple): exc_info = exc_info[1] if isinstance(exc_info, BaseException): record_exception(span, exc_info) if otlp_attributes[ATTRIBUTES_LOG_LEVEL_NUM_KEY] >= LEVEL_NUMBERS['error']: # type: ignore # Set the status description to the exception message. # OTEL only lets us set the description when the status code is ERROR, # which we only want to do when the log level is error. set_exception_status(span, exc_info) elif exc_info is not None: # pragma: no cover raise TypeError(f'Invalid type for exc_info: {exc_info.__class__.__name__}') span.end(start_time) def with_tags(self, *tags: str) -> Logfire: """A new Logfire instance which always uses the given tags. ```py import logfire logfire.configure() local_logfire = logfire.with_tags('tag1') local_logfire.info('a log message', _tags=['tag2']) # This is equivalent to: logfire.info('a log message', _tags=['tag1', 'tag2']) ``` Args: tags: The tags to add. Returns: A new Logfire instance with the `tags` added to any existing tags. """ return self.with_settings(tags=tags) def with_trace_sample_rate(self, sample_rate: float) -> Logfire: # pragma: no cover """A new Logfire instance with the given sampling ratio applied. Args: sample_rate: The sampling ratio to use. Returns: A new Logfire instance with the sampling ratio applied. """ if sample_rate > 1 or sample_rate < 0: raise ValueError('sample_rate must be between 0 and 1') return Logfire( config=self._config, tags=self._tags, sample_rate=sample_rate, ) def with_settings( self, *, tags: Sequence[str] = (), stack_offset: int | None = None, console_log: bool | None = None, custom_scope_suffix: str | None = None, ) -> Logfire: """A new Logfire instance which uses the given settings. Args: tags: Sequence of tags to include in the log. stack_offset: The stack level offset to use when collecting stack info, also affects the warning which message formatting might emit, defaults to `0` which means the stack info will be collected from the position where [`logfire.log`][logfire.Logfire.log] was called. console_log: Whether to log to the console, defaults to `True`. custom_scope_suffix: A custom suffix to append to `logfire.` e.g. `logfire.loguru`. It should only be used when instrumenting another library with Logfire, such as structlog or loguru. See the `instrumenting_module_name` parameter on [TracerProvider.get_tracer][opentelemetry.sdk.trace.TracerProvider.get_tracer] for more info. Returns: A new Logfire instance with the given settings applied. """ # TODO add sample_rate once it's more stable return Logfire( config=self._config, tags=self._tags + tuple(tags), sample_rate=self._sample_rate, console_log=self._console_log if console_log is None else console_log, otel_scope=self._otel_scope if custom_scope_suffix is None else f'logfire.{custom_scope_suffix}', ) def force_flush(self, timeout_millis: int = 3_000) -> bool: # pragma: no cover """Force flush all spans and metrics. Args: timeout_millis: The timeout in milliseconds. Returns: Whether the flush of spans was successful. """ return self._config.force_flush(timeout_millis) def log_slow_async_callbacks(self, slow_duration: float = 0.1) -> ContextManager[None]: """Log a warning whenever a function running in the asyncio event loop blocks for too long. This works by patching the `asyncio.events.Handle._run` method. Args: slow_duration: the threshold in seconds for when a callback is considered slow. Returns: A context manager that will revert the patch when exited. This context manager doesn't take into account threads or other concurrency. Calling this method will immediately apply the patch without waiting for the context manager to be opened, i.e. it's not necessary to use this as a context manager. """ return async_.log_slow_callbacks(self, slow_duration) def install_auto_tracing( self, modules: Sequence[str] | Callable[[AutoTraceModule], bool], *, min_duration: float, check_imported_modules: Literal['error', 'warn', 'ignore'] = 'error', ) -> None: """Install automatic tracing. See the [Auto-Tracing guide](https://logfire.pydantic.dev/docs/guides/onboarding_checklist/add_auto_tracing/) for more info. This will trace all non-generator function calls in the modules specified by the modules argument. It's equivalent to wrapping the body of every function in matching modules in `with logfire.span(...):`. !!! note This function MUST be called before any of the modules to be traced are imported. Generator functions will not be traced for reasons explained [here](https://logfire.pydantic.dev/docs/guides/advanced/generators/). This works by inserting a new meta path finder into `sys.meta_path`, so inserting another finder before it may prevent it from working. It relies on being able to retrieve the source code via at least one other existing finder in the meta path, so it may not work if standard finders are not present or if the source code is not available. A modified version of the source code is then compiled and executed in place of the original module. Args: modules: List of module names to trace, or a function which returns True for modules that should be traced. If a list is provided, any submodules within a given module will also be traced. min_duration: A minimum duration in seconds for which a function must run before it's traced. Setting to `0` causes all functions to be traced from the beginning. Otherwise, the first time(s) each function is called, it will be timed but not traced. Only after the function has run for at least `min_duration` will it be traced in subsequent calls. check_imported_modules: If this is `'error'` (the default), then an exception will be raised if any of the modules in `sys.modules` (i.e. modules that have already been imported) match the modules to trace. Set to `'warn'` to issue a warning instead, or `'ignore'` to skip the check. """ install_auto_tracing(self, modules, check_imported_modules=check_imported_modules, min_duration=min_duration) def _warn_if_not_initialized_for_instrumentation(self): self.config.warn_if_not_initialized('Instrumentation will have no effect') def instrument_pydantic( self, record: PydanticPluginRecordValues = 'all', include: Iterable[str] = (), exclude: Iterable[str] = (), ) -> None: """Instrument Pydantic model validations. This must be called before defining and importing the model classes you want to instrument. See the [Pydantic integration guide](https://logfire.pydantic.dev/docs/integrations/pydantic/) for more info. Args: record: The record mode for the Pydantic plugin. It can be one of the following values: - `all`: Send traces and metrics for all events. This is default value. - `failure`: Send metrics for all validations and traces only for validation failures. - `metrics`: Send only metrics. - `off`: Disable instrumentation. include: By default, third party modules are not instrumented. This option allows you to include specific modules. exclude: Exclude specific modules from instrumentation. """ # Note that unlike most instrument_* methods, we intentionally don't call # _warn_if_not_initialized_for_instrumentation, because this method needs to be called early. if record != 'off': import pydantic if get_version(pydantic.__version__) < get_version('2.5.0'): raise RuntimeError('The Pydantic plugin requires Pydantic 2.5.0 or newer.') from logfire.integrations.pydantic import PydanticPlugin, set_pydantic_plugin_config if isinstance(include, str): include = {include} if isinstance(exclude, str): exclude = {exclude} # TODO instrument using this instance, i.e. pass `self` somewhere, rather than always using the global instance set_pydantic_plugin_config( PydanticPlugin( record=record, include=set(include), exclude=set(exclude), ) ) def instrument_fastapi( self, app: FastAPI, *, capture_headers: bool = False, request_attributes_mapper: Callable[ [ Request | WebSocket, dict[str, Any], ], dict[str, Any] | None, ] | None = None, excluded_urls: str | Iterable[str] | None = None, record_send_receive: bool = False, **opentelemetry_kwargs: Any, ) -> ContextManager[None]: """Instrument a FastAPI app so that spans and logs are automatically created for each request. Uses the [OpenTelemetry FastAPI Instrumentation](https://opentelemetry-python-contrib.readthedocs.io/en/latest/instrumentation/fastapi/fastapi.html) under the hood, with some additional features. Args: app: The FastAPI app to instrument. capture_headers: Set to `True` to capture all request and response headers. request_attributes_mapper: A function that takes a [`Request`][fastapi.Request] or [`WebSocket`][fastapi.WebSocket] and a dictionary of attributes and returns a new dictionary of attributes. The input dictionary will contain: - `values`: A dictionary mapping argument names of the endpoint function to parsed and validated values. - `errors`: A list of validation errors for any invalid inputs. The returned dictionary will be used as the attributes for a log message. If `None` is returned, no log message will be created. You can use this to e.g. only log validation errors, or nothing at all. You can also add custom attributes. The default implementation will return the input dictionary unchanged. The function mustn't modify the contents of `values` or `errors`. excluded_urls: A string of comma-separated regexes which will exclude a request from tracing if the full URL matches any of the regexes. This applies to both the Logfire and OpenTelemetry instrumentation. If not provided, the environment variables `OTEL_PYTHON_FASTAPI_EXCLUDED_URLS` and `OTEL_PYTHON_EXCLUDED_URLS` will be checked. record_send_receive: Set to `True` to allow the OpenTelemetry ASGI middleware to create send/receive spans. These are disabled by default to reduce overhead and the number of spans created, since many can be created for a single request, and they are not often useful. If enabled, they will be set to debug level, meaning they will usually still be hidden in the UI. opentelemetry_kwargs: Additional keyword arguments to pass to the OpenTelemetry FastAPI instrumentation. Returns: A context manager that will revert the instrumentation when exited. This context manager doesn't take into account threads or other concurrency. Calling this method will immediately apply the instrumentation without waiting for the context manager to be opened, i.e. it's not necessary to use this as a context manager. """ from .integrations.fastapi import instrument_fastapi self._warn_if_not_initialized_for_instrumentation() return instrument_fastapi( self, app, capture_headers=capture_headers, request_attributes_mapper=request_attributes_mapper, excluded_urls=excluded_urls, record_send_receive=record_send_receive, **opentelemetry_kwargs, ) def instrument_openai( self, openai_client: openai.OpenAI | openai.AsyncOpenAI | type[openai.OpenAI] | type[openai.AsyncOpenAI] | None = None, *, suppress_other_instrumentation: bool = True, ) -> ContextManager[None]: """Instrument an OpenAI client so that spans are automatically created for each request. The following methods are instrumented for both the sync and the async clients: - [`client.chat.completions.create`](https://platform.openai.com/docs/guides/text-generation/chat-completions-api) — with and without `stream=True` - [`client.completions.create`](https://platform.openai.com/docs/guides/text-generation/completions-api) — with and without `stream=True` - [`client.embeddings.create`](https://platform.openai.com/docs/guides/embeddings/how-to-get-embeddings) - [`client.images.generate`](https://platform.openai.com/docs/guides/images/generations) When `stream=True` a second span is created to instrument the streamed response. Example usage: ```python import logfire import openai client = openai.OpenAI() logfire.configure() logfire.instrument_openai(client) response = client.chat.completions.create( model='gpt-4', messages=[ {'role': 'system', 'content': 'You are a helpful assistant.'}, {'role': 'user', 'content': 'What is four plus five?'}, ], ) print('answer:', response.choices[0].message.content) ``` Args: openai_client: The OpenAI client or class to instrument: - `None` (the default) to instrument both the `openai.OpenAI` and `openai.AsyncOpenAI` classes. - The `openai.OpenAI` class or a subclass - The `openai.AsyncOpenAI` class or a subclass - An instance of `openai.OpenAI` - An instance of `openai.AsyncOpenAI` suppress_other_instrumentation: If True, suppress any other OTEL instrumentation that may be otherwise enabled. In reality, this means the HTTPX instrumentation, which could otherwise be called since OpenAI uses HTTPX to make HTTP requests. Returns: A context manager that will revert the instrumentation when exited. Use of this context manager is optional. """ import openai from .integrations.llm_providers.llm_provider import instrument_llm_provider from .integrations.llm_providers.openai import get_endpoint_config, is_async_client, on_response self._warn_if_not_initialized_for_instrumentation() return instrument_llm_provider( self, openai_client or (openai.OpenAI, openai.AsyncOpenAI), suppress_other_instrumentation, 'OpenAI', get_endpoint_config, on_response, is_async_client, ) def instrument_anthropic( self, anthropic_client: ( anthropic.Anthropic | anthropic.AsyncAnthropic | anthropic.AnthropicBedrock | anthropic.AsyncAnthropicBedrock | type[anthropic.Anthropic] | type[anthropic.AsyncAnthropic] | type[anthropic.AnthropicBedrock] | type[anthropic.AsyncAnthropicBedrock] | None ) = None, *, suppress_other_instrumentation: bool = True, ) -> ContextManager[None]: """Instrument an Anthropic client so that spans are automatically created for each request. The following methods are instrumented for both the sync and async clients: - [`client.messages.create`](https://docs.anthropic.com/en/api/messages) - [`client.messages.stream`](https://docs.anthropic.com/en/api/messages-streaming) - [`client.beta.tools.messages.create`](https://docs.anthropic.com/en/docs/tool-use) When `stream=True` a second span is created to instrument the streamed response. Example usage: ```python import logfire import anthropic client = anthropic.Anthropic() logfire.configure() logfire.instrument_anthropic(client) response = client.messages.create( model='claude-3-haiku-20240307', system='You are a helpful assistant.', messages=[ {'role': 'user', 'content': 'What is four plus five?'}, ], ) print('answer:', response.content[0].text) ``` Args: anthropic_client: The Anthropic client or class to instrument: - `None` (the default) to instrument all Anthropic client types - The `anthropic.Anthropic` or `anthropic.AnthropicBedrock` class or subclass - The `anthropic.AsyncAnthropic` or `anthropic.AsyncAnthropicBedrock` class or subclass - An instance of any of the above classes suppress_other_instrumentation: If True, suppress any other OTEL instrumentation that may be otherwise enabled. In reality, this means the HTTPX instrumentation, which could otherwise be called since OpenAI uses HTTPX to make HTTP requests. Returns: A context manager that will revert the instrumentation when exited. Use of this context manager is optional. """ import anthropic from .integrations.llm_providers.anthropic import get_endpoint_config, is_async_client, on_response from .integrations.llm_providers.llm_provider import instrument_llm_provider self._warn_if_not_initialized_for_instrumentation() return instrument_llm_provider( self, anthropic_client or ( anthropic.Anthropic, anthropic.AsyncAnthropic, anthropic.AnthropicBedrock, anthropic.AsyncAnthropicBedrock, ), suppress_other_instrumentation, 'Anthropic', get_endpoint_config, on_response, is_async_client, ) def instrument_asyncpg(self, **kwargs: Any) -> None: """Instrument the `asyncpg` module so that spans are automatically created for each query.""" from .integrations.asyncpg import instrument_asyncpg self._warn_if_not_initialized_for_instrumentation() return instrument_asyncpg( **{ 'tracer_provider': self._config.get_tracer_provider(), 'meter_provider': self._config.get_meter_provider(), **kwargs, }, ) @overload def instrument_httpx( self, client: httpx.Client, *, capture_all: bool = False, capture_headers: bool = False, capture_request_body: bool = False, capture_response_body: bool = False, request_hook: HttpxRequestHook | None = None, response_hook: HttpxResponseHook | None = None, **kwargs: Any, ) -> None: ... @overload def instrument_httpx( self, client: httpx.AsyncClient, *, capture_all: bool = False, capture_headers: bool = False, capture_request_body: bool = False, capture_response_body: bool = False, request_hook: HttpxRequestHook | HttpxAsyncRequestHook | None = None, response_hook: HttpxResponseHook | HttpxAsyncResponseHook | None = None, **kwargs: Any, ) -> None: ... @overload def instrument_httpx( self, client: None = None, *, capture_all: bool = False, capture_headers: bool = False, capture_request_body: bool = False, capture_response_body: bool = False, request_hook: HttpxRequestHook | None = None, response_hook: HttpxResponseHook | None = None, async_request_hook: HttpxAsyncRequestHook | None = None, async_response_hook: HttpxAsyncResponseHook | None = None, **kwargs: Any, ) -> None: ... def instrument_httpx( self, client: httpx.Client | httpx.AsyncClient | None = None, *, capture_all: bool = False, capture_headers: bool = False, capture_request_body: bool = False, capture_response_body: bool = False, request_hook: HttpxRequestHook | HttpxAsyncRequestHook | None = None, response_hook: HttpxResponseHook | HttpxAsyncResponseHook | None = None, async_request_hook: HttpxAsyncRequestHook | None = None, async_response_hook: HttpxAsyncResponseHook | None = None, **kwargs: Any, ) -> None: """Instrument the `httpx` module so that spans are automatically created for each request. Optionally, pass an `httpx.Client` instance to instrument only that client. Uses the [OpenTelemetry HTTPX Instrumentation](https://opentelemetry-python-contrib.readthedocs.io/en/latest/instrumentation/httpx/httpx.html) library, specifically `HTTPXClientInstrumentor().instrument()`, to which it passes `**kwargs`. Args: client: The `httpx.Client` or `httpx.AsyncClient` instance to instrument. If `None`, the default, all clients will be instrumented. capture_all: Set to `True` to capture all HTTP headers, request and response bodies. capture_headers: Set to `True` to capture all HTTP headers. If you don't want to capture all headers, you can customize the headers captured. See the [Capture Headers](https://logfire.pydantic.dev/docs/guides/advanced/capture_headers/) section for more info. capture_request_body: Set to `True` to capture the request body. capture_response_body: Set to `True` to capture the response body. request_hook: A function called right after a span is created for a request. response_hook: A function called right before a span is finished for the response. async_request_hook: A function called right after a span is created for an async request. async_response_hook: A function called right before a span is finished for an async response. **kwargs: Additional keyword arguments to pass to the OpenTelemetry `instrument` method, for future compatibility. """ from .integrations.httpx import instrument_httpx self._warn_if_not_initialized_for_instrumentation() return instrument_httpx( self, client, capture_all=capture_all, capture_headers=capture_headers, capture_request_body=capture_request_body, capture_response_body=capture_response_body, request_hook=request_hook, response_hook=response_hook, async_request_hook=async_request_hook, async_response_hook=async_response_hook, **kwargs, ) def instrument_celery(self, **kwargs: Any) -> None: """Instrument `celery` so that spans are automatically created for each task. Uses the [OpenTelemetry Celery Instrumentation](https://opentelemetry-python-contrib.readthedocs.io/en/latest/instrumentation/celery/celery.html) library. Args: **kwargs: Additional keyword arguments to pass to the OpenTelemetry `instrument` method, for future compatibility. """ from .integrations.celery import instrument_celery self._warn_if_not_initialized_for_instrumentation() return instrument_celery( **{ 'tracer_provider': self._config.get_tracer_provider(), 'meter_provider': self._config.get_meter_provider(), **kwargs, }, ) def instrument_django( self, capture_headers: bool = False, is_sql_commentor_enabled: bool | None = None, request_hook: Callable[[trace_api.Span, HttpRequest], None] | None = None, response_hook: Callable[[trace_api.Span, HttpRequest, HttpResponse], None] | None = None, excluded_urls: str | None = None, **kwargs: Any, ) -> None: """Instrument `django` so that spans are automatically created for each web request. Uses the [OpenTelemetry Django Instrumentation](https://opentelemetry-python-contrib.readthedocs.io/en/latest/instrumentation/django/django.html) library. Args: capture_headers: Set to `True` to capture all request and response headers. is_sql_commentor_enabled: Adds comments to SQL queries performed by Django, so that database logs have additional context. This does NOT create spans/logs for the queries themselves. For that you need to instrument the database driver, e.g. with `logfire.instrument_psycopg()`. To configure the SQL Commentor, see the OpenTelemetry documentation for the values that need to be added to `settings.py`. request_hook: A function called right after a span is created for a request. The function should accept two arguments: the span and the Django `Request` object. response_hook: A function called right before a span is finished for the response. The function should accept three arguments: the span, the Django `Request` object, and the Django `Response` object. excluded_urls: A string containing a comma-delimited list of regexes used to exclude URLs from tracking. **kwargs: Additional keyword arguments to pass to the OpenTelemetry `instrument` method, for future compatibility. """ from .integrations.django import instrument_django self._warn_if_not_initialized_for_instrumentation() return instrument_django( capture_headers=capture_headers, is_sql_commentor_enabled=is_sql_commentor_enabled, request_hook=request_hook, response_hook=response_hook, excluded_urls=excluded_urls, **{ 'tracer_provider': self._config.get_tracer_provider(), 'meter_provider': self._config.get_meter_provider(), **kwargs, }, ) def instrument_requests( self, excluded_urls: str | None = None, request_hook: Callable[[Span, requests.PreparedRequest], None] | None = None, response_hook: Callable[[Span, requests.PreparedRequest, requests.Response], None] | None = None, **kwargs: Any, ) -> None: """Instrument the `requests` module so that spans are automatically created for each request. Args: excluded_urls: A string containing a comma-delimited list of regexes used to exclude URLs from tracking request_hook: A function called right after a span is created for a request. response_hook: A function called right before a span is finished for the response. **kwargs: Additional keyword arguments to pass to the OpenTelemetry `instrument` methods, for future compatibility. """ from .integrations.requests import instrument_requests self._warn_if_not_initialized_for_instrumentation() return instrument_requests( excluded_urls=excluded_urls, request_hook=request_hook, response_hook=response_hook, **{ 'tracer_provider': self._config.get_tracer_provider(), 'meter_provider': self._config.get_meter_provider(), **kwargs, }, ) @overload def instrument_psycopg(self, conn_or_module: PsycopgConnection | Psycopg2Connection, **kwargs: Any) -> None: ... @overload def instrument_psycopg( self, conn_or_module: None | Literal['psycopg', 'psycopg2'] | ModuleType = None, enable_commenter: bool = False, commenter_options: PsycopgCommenterOptions | None = None, **kwargs: Any, ) -> None: ... def instrument_psycopg( self, conn_or_module: Any = None, enable_commenter: bool = False, commenter_options: PsycopgCommenterOptions | None = None, **kwargs: Any, ) -> None: """Instrument a `psycopg` connection or module so that spans are automatically created for each query. Uses the OpenTelemetry instrumentation libraries for [`psycopg`](https://opentelemetry-python-contrib.readthedocs.io/en/latest/instrumentation/psycopg/psycopg.html) and [`psycopg2`](https://opentelemetry-python-contrib.readthedocs.io/en/latest/instrumentation/psycopg2/psycopg2.html). Args: conn_or_module: Can be: - The `psycopg` (version 3) or `psycopg2` module. - The string `'psycopg'` or `'psycopg2'` to instrument the module. - `None` (the default) to instrument whichever module(s) are installed. - A `psycopg` or `psycopg2` connection. enable_commenter: Adds comments to SQL queries performed by Psycopg, so that database logs have additional context. commenter_options: Configure the tags to be added to the SQL comments. **kwargs: Additional keyword arguments to pass to the OpenTelemetry `instrument` methods, for future compatibility. """ from .integrations.psycopg import instrument_psycopg self._warn_if_not_initialized_for_instrumentation() if enable_commenter: kwargs.update({'enable_commenter': True, 'commenter_options': commenter_options or {}}) return instrument_psycopg(self, conn_or_module=conn_or_module, **kwargs) def instrument_flask( self, app: Flask, *, capture_headers: bool = False, enable_commenter: bool = True, commenter_options: FlaskCommenterOptions | None = None, exclude_urls: str | None = None, request_hook: FlaskRequestHook | None = None, response_hook: FlaskResponseHook | None = None, **kwargs: Any, ) -> None: """Instrument `app` so that spans are automatically created for each request. Uses the [OpenTelemetry Flask Instrumentation](https://opentelemetry-python-contrib.readthedocs.io/en/latest/instrumentation/flask/flask.html) library, specifically `FlaskInstrumentor().instrument_app()`, to which it passes `**kwargs`. Args: app: The Flask app to instrument. capture_headers: Set to `True` to capture all request and response headers. enable_commenter: Adds comments to SQL queries performed by Flask, so that database logs have additional context. commenter_options: Configure the tags to be added to the SQL comments. See more about it on the [SQLCommenter Configurations](https://opentelemetry-python-contrib.readthedocs.io/en/latest/instrumentation/flask/flask.html#sqlcommenter-configurations). exclude_urls: A string containing a comma-delimited list of regexes used to exclude URLs from tracking. request_hook: A function called right after a span is created for a request. response_hook: A function called right before a span is finished for the response. **kwargs: Additional keyword arguments to pass to the OpenTelemetry Flask instrumentation. """ from .integrations.flask import instrument_flask self._warn_if_not_initialized_for_instrumentation() return instrument_flask( app, capture_headers=capture_headers, enable_commenter=enable_commenter, commenter_options=commenter_options, exclude_urls=exclude_urls, request_hook=request_hook, response_hook=response_hook, **{ 'tracer_provider': self._config.get_tracer_provider(), 'meter_provider': self._config.get_meter_provider(), **kwargs, }, ) def instrument_starlette( self, app: Starlette, *, capture_headers: bool = False, record_send_receive: bool = False, server_request_hook: ServerRequestHook | None = None, client_request_hook: ClientRequestHook | None = None, client_response_hook: ClientResponseHook | None = None, **kwargs: Any, ) -> None: """Instrument `app` so that spans are automatically created for each request. Uses the [OpenTelemetry Starlette Instrumentation](https://opentelemetry-python-contrib.readthedocs.io/en/latest/instrumentation/starlette/starlette.html) library, specifically `StarletteInstrumentor.instrument_app()`, to which it passes `**kwargs`. Args: app: The Starlette app to instrument. capture_headers: Set to `True` to capture all request and response headers. record_send_receive: Set to `True` to allow the OpenTelemetry ASGI middleware to create send/receive spans. These are disabled by default to reduce overhead and the number of spans created, since many can be created for a single request, and they are not often useful. If enabled, they will be set to debug level, meaning they will usually still be hidden in the UI. server_request_hook: A function that receives a server span and the ASGI scope for every incoming request. client_request_hook: A function that receives a span, the ASGI scope and the receive ASGI message for every ASGI receive event. client_response_hook: A function that receives a span, the ASGI scope and the send ASGI message for every ASGI send event. **kwargs: Additional keyword arguments to pass to the OpenTelemetry Starlette instrumentation. """ from .integrations.starlette import instrument_starlette self._warn_if_not_initialized_for_instrumentation() return instrument_starlette( self, app, record_send_receive=record_send_receive, capture_headers=capture_headers, server_request_hook=server_request_hook, client_request_hook=client_request_hook, client_response_hook=client_response_hook, **kwargs, ) def instrument_asgi( self, app: ASGIApp, capture_headers: bool = False, record_send_receive: bool = False, **kwargs: Unpack[ASGIInstrumentKwargs], ) -> ASGIApp: """Instrument `app` so that spans are automatically created for each request. Uses the ASGI [`OpenTelemetryMiddleware`][opentelemetry.instrumentation.asgi.OpenTelemetryMiddleware] under the hood, to which it passes `**kwargs`. Warning: Instead of modifying the app in place, this method returns the instrumented ASGI application. Args: app: The ASGI application to instrument. capture_headers: Set to `True` to capture all request and response headers. record_send_receive: Set to `True` to allow the OpenTelemetry ASGI middleware to create send/receive spans. These are disabled by default to reduce overhead and the number of spans created, since many can be created for a single request, and they are not often useful. If enabled, they will be set to debug level, meaning they will usually still be hidden in the UI. **kwargs: Additional keyword arguments to pass to the OpenTelemetry ASGI middleware. Returns: The instrumented ASGI application. """ from .integrations.asgi import instrument_asgi self._warn_if_not_initialized_for_instrumentation() return instrument_asgi( self, app, record_send_receive=record_send_receive, capture_headers=capture_headers, **kwargs, ) def instrument_wsgi( self, app: WSGIApplication, capture_headers: bool = False, request_hook: WSGIRequestHook | None = None, response_hook: WSGIResponseHook | None = None, **kwargs: Any, ) -> WSGIApplication: """Instrument `app` so that spans are automatically created for each request. Uses the WSGI [`OpenTelemetryMiddleware`][opentelemetry.instrumentation.wsgi.OpenTelemetryMiddleware] under the hood, to which it passes `**kwargs`. Warning: Instead of modifying the app in place, this method returns the instrumented WSGI application. Args: app: The WSGI application to instrument. capture_headers: Set to `True` to capture all request and response headers. request_hook: A function called right after a span is created for a request. response_hook: A function called right before a span is finished for the response. **kwargs: Additional keyword arguments to pass to the OpenTelemetry WSGI middleware. Returns: The instrumented WSGI application. """ from .integrations.wsgi import instrument_wsgi self._warn_if_not_initialized_for_instrumentation() return instrument_wsgi( app, capture_headers=capture_headers, request_hook=request_hook, response_hook=response_hook, **{ 'tracer_provider': self._config.get_tracer_provider(), 'meter_provider': self._config.get_meter_provider(), **kwargs, }, ) def instrument_aiohttp_client(self, **kwargs: Any) -> None: """Instrument the `aiohttp` module so that spans are automatically created for each client request. Uses the [OpenTelemetry aiohttp client Instrumentation](https://opentelemetry-python-contrib.readthedocs.io/en/latest/instrumentation/aiohttp_client/aiohttp_client.html) library, specifically `AioHttpClientInstrumentor().instrument()`, to which it passes `**kwargs`. """ from .integrations.aiohttp_client import instrument_aiohttp_client self._warn_if_not_initialized_for_instrumentation() return instrument_aiohttp_client(self, **kwargs) def instrument_sqlalchemy( self, engine: AsyncEngine | Engine | None = None, enable_commenter: bool = False, commenter_options: SQLAlchemyCommenterOptions | None = None, **kwargs: Any, ) -> None: """Instrument the `sqlalchemy` module so that spans are automatically created for each query. Uses the [OpenTelemetry SQLAlchemy Instrumentation](https://opentelemetry-python-contrib.readthedocs.io/en/latest/instrumentation/sqlalchemy/sqlalchemy.html) library, specifically `SQLAlchemyInstrumentor().instrument()`, to which it passes `**kwargs`. Args: engine: The `sqlalchemy` engine to instrument, or `None` to instrument all engines. enable_commenter: Adds comments to SQL queries performed by SQLAlchemy, so that database logs have additional context. commenter_options: Configure the tags to be added to the SQL comments. **kwargs: Additional keyword arguments to pass to the OpenTelemetry `instrument` methods. """ from .integrations.sqlalchemy import instrument_sqlalchemy self._warn_if_not_initialized_for_instrumentation() return instrument_sqlalchemy( engine=engine, enable_commenter=enable_commenter, commenter_options=commenter_options or {}, **{ 'tracer_provider': self._config.get_tracer_provider(), 'meter_provider': self._config.get_meter_provider(), **kwargs, }, ) def instrument_sqlite3(self, conn: SQLite3Connection = None, **kwargs: Any) -> SQLite3Connection: """Instrument the `sqlite3` module or a specific connection so that spans are automatically created for each operation. Uses the [OpenTelemetry SQLite3 Instrumentation](https://opentelemetry-python-contrib.readthedocs.io/en/latest/instrumentation/sqlite3/sqlite3.html) library. Args: conn: The `sqlite3` connection to instrument, or `None` to instrument all connections. **kwargs: Additional keyword arguments to pass to the OpenTelemetry `instrument` methods. Returns: If a connection is provided, returns the instrumented connection. If no connection is provided, returns `None`. """ from .integrations.sqlite3 import instrument_sqlite3 self._warn_if_not_initialized_for_instrumentation() return instrument_sqlite3(conn=conn, **{'tracer_provider': self._config.get_tracer_provider(), **kwargs}) def instrument_aws_lambda( self, lambda_handler: LambdaHandler, event_context_extractor: Callable[[LambdaEvent], Context] | None = None, **kwargs: Any, ) -> None: """Instrument AWS Lambda so that spans are automatically created for each invocation. Uses the [OpenTelemetry AWS Lambda Instrumentation](https://opentelemetry-python-contrib.readthedocs.io/en/latest/instrumentation/aws_lambda/aws_lambda.html) library, specifically `AwsLambdaInstrumentor().instrument()`, to which it passes `**kwargs`. Args: lambda_handler: The lambda handler function to instrument. event_context_extractor: A function that returns an OTel Trace Context given the Lambda Event the AWS. **kwargs: Additional keyword arguments to pass to the OpenTelemetry `instrument` methods for future compatibility. """ from .integrations.aws_lambda import instrument_aws_lambda self._warn_if_not_initialized_for_instrumentation() return instrument_aws_lambda( lambda_handler=lambda_handler, event_context_extractor=event_context_extractor, **{ # type: ignore 'tracer_provider': self._config.get_tracer_provider(), 'meter_provider': self._config.get_meter_provider(), **kwargs, }, ) def instrument_pymongo( self, capture_statement: bool = False, request_hook: Callable[[Span, CommandStartedEvent], None] | None = None, response_hook: Callable[[Span, CommandSucceededEvent], None] | None = None, failed_hook: Callable[[Span, CommandFailedEvent], None] | None = None, **kwargs: Any, ) -> None: """Instrument the `pymongo` module so that spans are automatically created for each operation. Uses the [OpenTelemetry pymongo Instrumentation](https://opentelemetry-python-contrib.readthedocs.io/en/latest/instrumentation/pymongo/pymongo.html) library, specifically `PymongoInstrumentor().instrument()`, to which it passes `**kwargs`. Args: capture_statement: Set to `True` to capture the statement in the span attributes. request_hook: A function called when a command is sent to the server. response_hook: A function that is called when a command is successfully completed. failed_hook: A function that is called when a command fails. **kwargs: Additional keyword arguments to pass to the OpenTelemetry `instrument` methods for future compatibility. """ from .integrations.pymongo import instrument_pymongo self._warn_if_not_initialized_for_instrumentation() return instrument_pymongo( capture_statement=capture_statement, request_hook=request_hook, response_hook=response_hook, failed_hook=failed_hook, **{ 'tracer_provider': self._config.get_tracer_provider(), 'meter_provider': self._config.get_meter_provider(), **kwargs, }, ) def instrument_redis( self, capture_statement: bool = False, request_hook: RedisRequestHook | None = None, response_hook: RedisResponseHook | None = None, **kwargs: Any, ) -> None: """Instrument the `redis` module so that spans are automatically created for each operation. Uses the [OpenTelemetry Redis Instrumentation](https://opentelemetry-python-contrib.readthedocs.io/en/latest/instrumentation/redis/redis.html) library, specifically `RedisInstrumentor().instrument()`, to which it passes `**kwargs`. Args: capture_statement: Set to `True` to capture the statement in the span attributes. request_hook: A function that is called before performing the request. response_hook: A function that is called after receiving the response. **kwargs: Additional keyword arguments to pass to the OpenTelemetry `instrument` methods for future compatibility. """ from .integrations.redis import instrument_redis self._warn_if_not_initialized_for_instrumentation() return instrument_redis( capture_statement=capture_statement, request_hook=request_hook, response_hook=response_hook, **{ 'tracer_provider': self._config.get_tracer_provider(), 'meter_provider': self._config.get_meter_provider(), **kwargs, }, ) def instrument_mysql(self, conn: MySQLConnection = None, **kwargs: Any) -> MySQLConnection: """Instrument the `mysql` module or a specific MySQL connection so that spans are automatically created for each operation. Uses the [OpenTelemetry MySQL Instrumentation](https://opentelemetry-python-contrib.readthedocs.io/en/latest/instrumentation/mysql/mysql.html) library. Args: conn: The `mysql` connection to instrument, or `None` to instrument all connections. **kwargs: Additional keyword arguments to pass to the OpenTelemetry `instrument` methods. Returns: If a connection is provided, returns the instrumented connection. If no connection is provided, returns None. """ from .integrations.mysql import instrument_mysql self._warn_if_not_initialized_for_instrumentation() return instrument_mysql( conn=conn, **{ # type: ignore 'tracer_provider': self._config.get_tracer_provider(), 'meter_provider': self._config.get_meter_provider(), **kwargs, }, ) def instrument_system_metrics( self, config: SystemMetricsConfig | None = None, base: SystemMetricsBase = 'basic' ) -> None: """Collect system metrics. See [the guide](https://logfire.pydantic.dev/docs/integrations/system-metrics/) for more information. Args: config: A dictionary where the keys are metric names and the values are optional further configuration for that metric. base: A string indicating the base config dictionary which `config` will be merged with, or `None` for an empty base config. """ from .integrations.system_metrics import instrument_system_metrics self._warn_if_not_initialized_for_instrumentation() return instrument_system_metrics(self, config, base) def metric_counter(self, name: str, *, unit: str = '', description: str = '') -> Counter: """Create a counter metric. A counter is a cumulative metric that represents a single numerical value that only ever goes up. ```py import logfire logfire.configure() counter = logfire.metric_counter('exceptions', unit='1', description='Number of exceptions caught') try: raise Exception('oops') except Exception: counter.add(1) ``` See the [Opentelemetry documentation](https://opentelemetry.io/docs/specs/otel/metrics/api/#counter) about counters. Args: name: The name of the metric. unit: The unit of the metric. description: The description of the metric. Returns: The counter metric. """ return self._meter.create_counter(name, unit, description) def metric_histogram(self, name: str, *, unit: str = '', description: str = '') -> Histogram: """Create a histogram metric. A histogram is a metric that samples observations (usually things like request durations or response sizes). ```py import logfire logfire.configure() histogram = logfire.metric_histogram('bank.amount_transferred', unit='$', description='Amount transferred') def transfer(amount: int): histogram.record(amount) ``` See the [Opentelemetry documentation](https://opentelemetry.io/docs/specs/otel/metrics/api/#histogram) about Args: name: The name of the metric. unit: The unit of the metric. description: The description of the metric. Returns: The histogram metric. """ return self._meter.create_histogram(name, unit, description) def metric_gauge(self, name: str, *, unit: str = '', description: str = '') -> Gauge: """Create a gauge metric. Gauge is a synchronous instrument which can be used to record non-additive measurements. ```py import logfire logfire.configure() gauge = logfire.metric_gauge('system.cpu_usage', unit='%', description='CPU usage') def update_cpu_usage(cpu_percent): gauge.set(cpu_percent) ``` See the [Opentelemetry documentation](https://opentelemetry.io/docs/specs/otel/metrics/api/#gauge) about gauges. Args: name: The name of the metric. unit: The unit of the metric. description: The description of the metric. Returns: The gauge metric. """ return self._meter.create_gauge(name, unit, description) def metric_up_down_counter(self, name: str, *, unit: str = '', description: str = '') -> UpDownCounter: """Create an up-down counter metric. An up-down counter is a cumulative metric that represents a single numerical value that can be adjusted up or down. ```py import logfire logfire.configure() up_down_counter = logfire.metric_up_down_counter('users.logged_in', unit='1', description='Users logged in') def on_login(user): up_down_counter.add(1) def on_logout(user): up_down_counter.add(-1) ``` See the [Opentelemetry documentation](https://opentelemetry.io/docs/specs/otel/metrics/api/#updowncounter) about up-down counters. Args: name: The name of the metric. unit: The unit of the metric. description: The description of the metric. Returns: The up-down counter metric. """ return self._meter.create_up_down_counter(name, unit, description) def metric_counter_callback( self, name: str, *, callbacks: Sequence[CallbackT], unit: str = '', description: str = '', ) -> None: """Create a counter metric that uses a callback to collect observations. The counter metric is a cumulative metric that represents a single numerical value that only ever goes up. ```py import logfire import psutil from opentelemetry.metrics import CallbackOptions, Observation logfire.configure() def cpu_usage_callback(options: CallbackOptions): cpu_percents = psutil.cpu_percent(percpu=True) for i, cpu_percent in enumerate(cpu_percents): yield Observation(cpu_percent, {'cpu': i}) cpu_usage_counter = logfire.metric_counter_callback( 'system.cpu.usage', callbacks=[cpu_usage_callback], unit='%', description='CPU usage', ) ``` See the [Opentelemetry documentation](https://opentelemetry.io/docs/specs/otel/metrics/api/#asynchronous-counter) about asynchronous counter. Args: name: The name of the metric. callbacks: A sequence of callbacks that return an iterable of [Observation](https://opentelemetry-python.readthedocs.io/en/latest/api/metrics.html#opentelemetry.metrics.Observation). unit: The unit of the metric. description: The description of the metric. """ self._meter.create_observable_counter(name, callbacks, unit, description) def metric_gauge_callback( self, name: str, callbacks: Sequence[CallbackT], *, unit: str = '', description: str = '' ) -> None: """Create a gauge metric that uses a callback to collect observations. The gauge metric is a metric that represents a single numerical value that can arbitrarily go up and down. ```py import threading import logfire from opentelemetry.metrics import CallbackOptions, Observation logfire.configure() def thread_count_callback(options: CallbackOptions): yield Observation(threading.active_count()) logfire.metric_gauge_callback( 'system.thread_count', callbacks=[thread_count_callback], unit='1', description='Number of threads', ) ``` See the [Opentelemetry documentation](https://opentelemetry.io/docs/specs/otel/metrics/api/#asynchronous-gauge) about asynchronous gauge. Args: name: The name of the metric. callbacks: A sequence of callbacks that return an iterable of [Observation](https://opentelemetry-python.readthedocs.io/en/latest/api/metrics.html#opentelemetry.metrics.Observation). unit: The unit of the metric. description: The description of the metric. """ self._meter.create_observable_gauge(name, callbacks, unit, description) def metric_up_down_counter_callback( self, name: str, callbacks: Sequence[CallbackT], *, unit: str = '', description: str = '' ) -> None: """Create an up-down counter metric that uses a callback to collect observations. The up-down counter is a cumulative metric that represents a single numerical value that can be adjusted up or down. ```py import logfire from opentelemetry.metrics import CallbackOptions, Observation logfire.configure() items = [] def inventory_callback(options: CallbackOptions): yield Observation(len(items)) logfire.metric_up_down_counter_callback( name='store.inventory', description='Number of items in the inventory', callbacks=[inventory_callback], ) ``` See the [Opentelemetry documentation](https://opentelemetry.io/docs/specs/otel/metrics/api/#asynchronous-updowncounter) about asynchronous up-down counters. Args: name: The name of the metric. callbacks: A sequence of callbacks that return an iterable of [Observation](https://opentelemetry-python.readthedocs.io/en/latest/api/metrics.html#opentelemetry.metrics.Observation). unit: The unit of the metric. description: The description of the metric. """ self._meter.create_observable_up_down_counter(name, callbacks, unit, description) def suppress_scopes(self, *scopes: str) -> None: """Prevent spans and metrics from being created for the given OpenTelemetry scope names. To get the scope name of a span/metric, check the value of the `otel_scope_name` column in the Logfire database. """ self._config.suppress_scopes(*scopes) def shutdown(self, timeout_millis: int = 30_000, flush: bool = True) -> bool: # pragma: no cover """Shut down all tracers and meters. This will clean up any resources used by the tracers and meters and flush any remaining spans and metrics. Args: timeout_millis: The timeout in milliseconds. flush: Whether to flush remaining spans and metrics before shutting down. Returns: `False` if the timeout was reached before the shutdown was completed, `True` otherwise. """ start = time() if flush: # pragma: no branch self._tracer_provider.force_flush(timeout_millis) remaining = max(0, timeout_millis - (time() - start)) if not remaining: # pragma: no cover return False self._tracer_provider.shutdown() remaining = max(0, timeout_millis - (time() - start)) if not remaining: # pragma: no cover return False if flush: # pragma: no branch self._meter_provider.force_flush(remaining) remaining = max(0, timeout_millis - (time() - start)) if not remaining: # pragma: no cover return False self._meter_provider.shutdown(remaining) return (start - time()) < timeout_millis class FastLogfireSpan: """A simple version of `LogfireSpan` optimized for auto-tracing.""" __slots__ = ('_span', '_token') def __init__(self, span: trace_api.Span) -> None: self._span = span self._token = context_api.attach(trace_api.set_span_in_context(self._span)) def __enter__(self) -> FastLogfireSpan: return self @handle_internal_errors() def __exit__(self, exc_type: type[BaseException] | None, exc_value: BaseException | None, traceback: Any) -> None: context_api.detach(self._token) self._span.__exit__(exc_type, exc_value, traceback) # Changes to this class may need to be reflected in `FastLogfireSpan` and `NoopSpan` as well. class LogfireSpan(ReadableSpan): def __init__( self, span_name: str, otlp_attributes: dict[str, otel_types.AttributeValue], tracer: Tracer, json_schema_properties: JsonSchemaProperties, links: Sequence[tuple[SpanContext, otel_types.Attributes]], ) -> None: self._span_name = span_name self._otlp_attributes = otlp_attributes self._tracer = tracer self._json_schema_properties = json_schema_properties self._links = list(trace_api.Link(context=context, attributes=attributes) for context, attributes in links) self._added_attributes = False self._token: None | object = None self._span: None | trace_api.Span = None if not TYPE_CHECKING: # pragma: no branch def __getattr__(self, name: str) -> Any: return getattr(self._span, name) def __enter__(self) -> LogfireSpan: with handle_internal_errors(): if self._span is None: # pragma: no branch self._span = self._tracer.start_span( name=self._span_name, attributes=self._otlp_attributes, links=self._links, ) self._span.__enter__() if self._token is None: # pragma: no branch self._token = context_api.attach(trace_api.set_span_in_context(self._span)) return self @handle_internal_errors() def __exit__(self, exc_type: type[BaseException] | None, exc_value: BaseException | None, traceback: Any) -> None: if self._token is None: # pragma: no cover return assert self._span is not None context_api.detach(self._token) self._token = None if self._span.is_recording(): with handle_internal_errors(): if self._added_attributes: self._span.set_attribute( ATTRIBUTES_JSON_SCHEMA_KEY, attributes_json_schema(self._json_schema_properties) ) self._span.__exit__(exc_type, exc_value, traceback) @property def message_template(self) -> str | None: # pragma: no cover return self._get_attribute(ATTRIBUTES_MESSAGE_TEMPLATE_KEY, None) @property def tags(self) -> tuple[str, ...]: return self._get_attribute(ATTRIBUTES_TAGS_KEY, ()) @tags.setter @handle_internal_errors() def tags(self, new_tags: Sequence[str]) -> None: """Set or add tags to the span.""" if isinstance(new_tags, str): new_tags = (new_tags,) self._set_attribute(ATTRIBUTES_TAGS_KEY, tuple(uniquify_sequence(new_tags))) @property def message(self) -> str: return self._get_attribute(ATTRIBUTES_MESSAGE_KEY, self._span_name) @message.setter def message(self, message: str): self._set_attribute(ATTRIBUTES_MESSAGE_KEY, message) @handle_internal_errors() def set_attribute(self, key: str, value: Any) -> None: """Sets an attribute on the span. Args: key: The key of the attribute. value: The value of the attribute. """ self._added_attributes = True self._json_schema_properties[key] = create_json_schema(value, set()) key, otel_value = set_user_attribute(self._otlp_attributes, key, value) if self._span is not None: # pragma: no branch self._span.set_attribute(key, otel_value) def set_attributes(self, attributes: dict[str, Any]) -> None: """Sets the given attributes on the span.""" for key, value in attributes.items(): self.set_attribute(key, value) def add_link(self, context: SpanContext, attributes: otel_types.Attributes = None) -> None: if self._span is None: self._links += [trace_api.Link(context=context, attributes=attributes)] else: self._span.add_link(context, attributes) # TODO(Marcelo): We should add a test for `record_exception`. def record_exception( self, exception: BaseException, attributes: otel_types.Attributes = None, timestamp: int | None = None, escaped: bool = False, ) -> None: # pragma: no cover """Records an exception as a span event. Delegates to the OpenTelemetry SDK `Span.record_exception` method. """ if self._span is None: raise RuntimeError('Span has not been started') # Check if the span has been sampled out first, since _record_exception is somewhat expensive. if not self._span.is_recording(): return record_exception( self._span, exception, attributes=attributes, timestamp=timestamp, escaped=escaped, ) def is_recording(self) -> bool: return self._span is not None and self._span.is_recording() @handle_internal_errors() def set_level(self, level: LevelName | int): """Set the log level of this span.""" attributes = log_level_attributes(level) if self._span is None: self._otlp_attributes.update(attributes) else: self._span.set_attributes(attributes) def _get_attribute(self, key: str, default: Any) -> Any: attributes = getattr(self._span, 'attributes', self._otlp_attributes) return attributes.get(key, default) def _set_attribute(self, key: str, value: Any) -> None: """Set an attribute on the span or in the _otlp_attributes if span is not yet created.""" if self._span is None: self._otlp_attributes[key] = value else: self._span.set_attribute(key, value) class NoopSpan: """Implements the same methods as `LogfireSpan` but does nothing. Used in place of `LogfireSpan` and `FastLogfireSpan` when an exception occurs during span creation. This way code like: with logfire.span(...) as span: span.set_attribute(...) doesn't raise an error even if `logfire.span` fails internally. If `logfire.span` just returned `None` then the `with` block and the `span.set_attribute` call would raise an error. TODO this should also be used when tracing is disabled, e.g. before `logfire.configure()` has been called. """ def __init__(self, *_args: Any, **__kwargs: Any) -> None: pass def __getattr__(self, _name: str) -> Any: # Handle methods of LogfireSpan which return nothing return lambda *_args, **__kwargs: None # type: ignore def __enter__(self) -> NoopSpan: return self def __exit__(self, exc_type: type[BaseException] | None, exc_value: BaseException | None, traceback: Any) -> None: pass # Implement methods/properties that return something to get the type right. @property def message_template(self) -> str: # pragma: no cover return '' @property def tags(self) -> tuple[str, ...]: return () # This is required to make `span.tags = ` not raise an error. @tags.setter def tags(self, new_tags: Sequence[str]) -> None: pass @property def message(self) -> str: # pragma: no cover return '' # This is required to make `span.message = ` not raise an error. @message.setter def message(self, message: str): pass def is_recording(self) -> bool: return False AttributesValueType = TypeVar('AttributesValueType', bound=Union[Any, otel_types.AttributeValue]) def prepare_otlp_attributes(attributes: dict[str, Any]) -> dict[str, otel_types.AttributeValue]: """Prepare attributes for sending to OpenTelemetry. This will convert any non-OpenTelemetry compatible types to JSON. """ otlp_attributes: dict[str, otel_types.AttributeValue] = {} for key, value in attributes.items(): set_user_attribute(otlp_attributes, key, value) return otlp_attributes def set_user_attribute( otlp_attributes: dict[str, otel_types.AttributeValue], key: str, value: Any ) -> tuple[str, otel_types.AttributeValue]: """Convert a user attribute to an OpenTelemetry compatible type and add it to the given dictionary. Returns the final key and value that was added to the dictionary. The key will be the original key unless the value was `None`, in which case it will be `NULL_ARGS_KEY`. """ otel_value: otel_types.AttributeValue if value is None: otel_value = cast('list[str]', otlp_attributes.get(NULL_ARGS_KEY, [])) + [key] key = NULL_ARGS_KEY elif isinstance(value, int): if value > OTLP_MAX_INT_SIZE: warnings.warn( f'Integer value {value} is larger than the maximum OTLP integer size of {OTLP_MAX_INT_SIZE} (64-bits), ' ' if you need support for sending larger integers, please open a feature request', UserWarning, ) otel_value = str(value) else: otel_value = value elif isinstance(value, (str, bool, float)): otel_value = value else: otel_value = logfire_json_dumps(value) otlp_attributes[key] = otel_value return key, otel_value def set_user_attributes_on_raw_span(span: Span, attributes: dict[str, Any]) -> None: if not span.is_recording(): return otlp_attributes = prepare_otlp_attributes(attributes) if json_schema_properties := attributes_json_schema_properties(attributes): # pragma: no branch existing_properties = JsonSchemaProperties({}) existing_json_schema_str = (span.attributes or {}).get(ATTRIBUTES_JSON_SCHEMA_KEY) if existing_json_schema_str and isinstance(existing_json_schema_str, str): with contextlib.suppress(json.JSONDecodeError): existing_json_schema = json.loads(existing_json_schema_str) existing_properties = existing_json_schema.get('properties', {}) existing_properties.update(json_schema_properties) otlp_attributes[ATTRIBUTES_JSON_SCHEMA_KEY] = attributes_json_schema(existing_properties) span.set_attributes(otlp_attributes) P = ParamSpec('P') R = TypeVar('R')