lmnr 0.6.20__py3-none-any.whl → 0.7.0__py3-none-any.whl

lmnr/opentelemetry_lib/opentelemetry/instrumentation/anthropic/__init__.py

@@ -0,0 +1,678 @@
+"""OpenTelemetry Anthropic instrumentation"""
+
+import logging
+import time
+from typing import Callable, Collection, Optional
+
+from opentelemetry import context as context_api
+from opentelemetry._events import EventLogger, get_event_logger
+from .config import Config
+from .event_emitter import (
+    emit_input_events,
+    emit_response_events,
+)
+from .span_utils import (
+    aset_input_attributes,
+    set_response_attributes,
+)
+from .streaming import (
+    abuild_from_streaming_response,
+    build_from_streaming_response,
+)
+from .utils import (
+    acount_prompt_tokens_from_request,
+    count_prompt_tokens_from_request,
+    dont_throw,
+    error_metrics_attributes,
+    run_async,
+    set_span_attribute,
+    shared_metrics_attributes,
+    should_emit_events,
+)
+from .version import __version__
+
+from lmnr.opentelemetry_lib.tracing.context import get_current_context
+from opentelemetry.instrumentation.instrumentor import BaseInstrumentor
+from opentelemetry.instrumentation.utils import _SUPPRESS_INSTRUMENTATION_KEY, unwrap
+from opentelemetry.metrics import Counter, Histogram, Meter, get_meter
+from opentelemetry.semconv_ai import (
+    SUPPRESS_LANGUAGE_MODEL_INSTRUMENTATION_KEY,
+    LLMRequestTypeValues,
+    Meters,
+    SpanAttributes,
+)
+from opentelemetry.trace import Span, SpanKind, Tracer, get_tracer
+from opentelemetry.trace.status import Status, StatusCode
+from typing_extensions import Coroutine
+from wrapt import wrap_function_wrapper
+
+from anthropic._streaming import AsyncStream, Stream
+
+logger = logging.getLogger(__name__)
+
+_instruments = ("anthropic >= 0.3.11",)
+
+WRAPPED_METHODS = [
+    {
+        "package": "anthropic.resources.completions",
+        "object": "Completions",
+        "method": "create",
+        "span_name": "anthropic.completion",
+    },
+    {
+        "package": "anthropic.resources.messages",
+        "object": "Messages",
+        "method": "create",
+        "span_name": "anthropic.chat",
+    },
+    {
+        "package": "anthropic.resources.messages",
+        "object": "Messages",
+        "method": "stream",
+        "span_name": "anthropic.chat",
+    },
+]
+
+WRAPPED_AMETHODS = [
+    {
+        "package": "anthropic.resources.completions",
+        "object": "AsyncCompletions",
+        "method": "create",
+        "span_name": "anthropic.completion",
+    },
+    {
+        "package": "anthropic.resources.messages",
+        "object": "AsyncMessages",
+        "method": "create",
+        "span_name": "anthropic.chat",
+    },
+    {
+        "package": "anthropic.resources.messages",
+        "object": "AsyncMessages",
+        "method": "stream",
+        "span_name": "anthropic.chat",
+    },
+]
+
+
+def is_streaming_response(response):
+    return isinstance(response, Stream) or isinstance(response, AsyncStream)
+
+
+@dont_throw
+async def _aset_token_usage(
+    span,
+    anthropic,
+    request,
+    response,
+    metric_attributes: dict = {},
+    token_histogram: Histogram = None,
+    choice_counter: Counter = None,
+):
+    if not isinstance(response, dict):
+        response = response.__dict__
+
+    if usage := response.get("usage"):
+        prompt_tokens = usage.input_tokens
+        cache_read_tokens = dict(usage).get("cache_read_input_tokens", 0) or 0
+        cache_creation_tokens = dict(usage).get("cache_creation_input_tokens", 0) or 0
+    else:
+        prompt_tokens = await acount_prompt_tokens_from_request(anthropic, request)
+        cache_read_tokens = 0
+        cache_creation_tokens = 0
+
+    input_tokens = prompt_tokens + cache_read_tokens + cache_creation_tokens
+
+    if token_histogram and isinstance(input_tokens, int) and input_tokens >= 0:
+        token_histogram.record(
+            input_tokens,
+            attributes={
+                **metric_attributes,
+                SpanAttributes.LLM_TOKEN_TYPE: "input",
+            },
+        )
+
+    if usage := response.get("usage"):
+        completion_tokens = usage.output_tokens
+    else:
+        completion_tokens = 0
+        if hasattr(anthropic, "count_tokens"):
+            if response.get("completion"):
+                completion_tokens = await anthropic.count_tokens(
+                    response.get("completion")
+                )
+            elif response.get("content"):
+                completion_tokens = await anthropic.count_tokens(
+                    response.get("content")[0].text
+                )
+
+    if (
+        token_histogram
+        and isinstance(completion_tokens, int)
+        and completion_tokens >= 0
+    ):
+        token_histogram.record(
+            completion_tokens,
+            attributes={
+                **metric_attributes,
+                SpanAttributes.LLM_TOKEN_TYPE: "output",
+            },
+        )
+
+    total_tokens = input_tokens + completion_tokens
+
+    choices = 0
+    if isinstance(response.get("content"), list):
+        choices = len(response.get("content"))
+    elif response.get("completion"):
+        choices = 1
+
+    if choices > 0 and choice_counter:
+        choice_counter.add(
+            choices,
+            attributes={
+                **metric_attributes,
+                SpanAttributes.LLM_RESPONSE_STOP_REASON: response.get("stop_reason"),
+            },
+        )
+
+    set_span_attribute(span, SpanAttributes.LLM_USAGE_PROMPT_TOKENS, input_tokens)
+    set_span_attribute(
+        span, SpanAttributes.LLM_USAGE_COMPLETION_TOKENS, completion_tokens
+    )
+    set_span_attribute(span, SpanAttributes.LLM_USAGE_TOTAL_TOKENS, total_tokens)
+
+    set_span_attribute(
+        span, SpanAttributes.LLM_USAGE_CACHE_READ_INPUT_TOKENS, cache_read_tokens
+    )
+    set_span_attribute(
+        span,
+        SpanAttributes.LLM_USAGE_CACHE_CREATION_INPUT_TOKENS,
+        cache_creation_tokens,
+    )
+
+
+@dont_throw
+def _set_token_usage(
+    span,
+    anthropic,
+    request,
+    response,
+    metric_attributes: dict = {},
+    token_histogram: Histogram = None,
+    choice_counter: Counter = None,
+):
+    if not isinstance(response, dict):
+        response = response.__dict__
+
+    if usage := response.get("usage"):
+        prompt_tokens = usage.input_tokens
+        cache_read_tokens = dict(usage).get("cache_read_input_tokens", 0) or 0
+        cache_creation_tokens = dict(usage).get("cache_creation_input_tokens", 0) or 0
+    else:
+        prompt_tokens = count_prompt_tokens_from_request(anthropic, request)
+        cache_read_tokens = 0
+        cache_creation_tokens = 0
+
+    input_tokens = prompt_tokens + cache_read_tokens + cache_creation_tokens
+
+    if token_histogram and isinstance(input_tokens, int) and input_tokens >= 0:
+        token_histogram.record(
+            input_tokens,
+            attributes={
+                **metric_attributes,
+                SpanAttributes.LLM_TOKEN_TYPE: "input",
+            },
+        )
+
+    if usage := response.get("usage"):
+        completion_tokens = usage.output_tokens
+    else:
+        completion_tokens = 0
+        if hasattr(anthropic, "count_tokens"):
+            if response.get("completion"):
+                completion_tokens = anthropic.count_tokens(response.get("completion"))
+            elif response.get("content"):
+                completion_tokens = anthropic.count_tokens(
+                    response.get("content")[0].text
+                )
+
+    if (
+        token_histogram
+        and isinstance(completion_tokens, int)
+        and completion_tokens >= 0
+    ):
+        token_histogram.record(
+            completion_tokens,
+            attributes={
+                **metric_attributes,
+                SpanAttributes.LLM_TOKEN_TYPE: "output",
+            },
+        )
+
+    total_tokens = input_tokens + completion_tokens
+
+    choices = 0
+    if isinstance(response.get("content"), list):
+        choices = len(response.get("content"))
+    elif response.get("completion"):
+        choices = 1
+
+    if choices > 0 and choice_counter:
+        choice_counter.add(
+            choices,
+            attributes={
+                **metric_attributes,
+                SpanAttributes.LLM_RESPONSE_STOP_REASON: response.get("stop_reason"),
+            },
+        )
+
+    set_span_attribute(span, SpanAttributes.LLM_USAGE_PROMPT_TOKENS, input_tokens)
+    set_span_attribute(
+        span, SpanAttributes.LLM_USAGE_COMPLETION_TOKENS, completion_tokens
+    )
+    set_span_attribute(span, SpanAttributes.LLM_USAGE_TOTAL_TOKENS, total_tokens)
+
+    set_span_attribute(
+        span, SpanAttributes.LLM_USAGE_CACHE_READ_INPUT_TOKENS, cache_read_tokens
+    )
+    set_span_attribute(
+        span,
+        SpanAttributes.LLM_USAGE_CACHE_CREATION_INPUT_TOKENS,
+        cache_creation_tokens,
+    )
+
+
+def _with_chat_telemetry_wrapper(func):
+    """Helper for providing tracer for wrapper functions. Includes metric collectors."""
+
+    def _with_chat_telemetry(
+        tracer,
+        token_histogram,
+        choice_counter,
+        duration_histogram,
+        exception_counter,
+        event_logger,
+        to_wrap,
+    ):
+        def wrapper(wrapped, instance, args, kwargs):
+            return func(
+                tracer,
+                token_histogram,
+                choice_counter,
+                duration_histogram,
+                exception_counter,
+                event_logger,
+                to_wrap,
+                wrapped,
+                instance,
+                args,
+                kwargs,
+            )
+
+        return wrapper
+
+    return _with_chat_telemetry
+
+
+def _create_metrics(meter: Meter):
+    token_histogram = meter.create_histogram(
+        name=Meters.LLM_TOKEN_USAGE,
+        unit="token",
+        description="Measures number of input and output tokens used",
+    )
+
+    choice_counter = meter.create_counter(
+        name=Meters.LLM_GENERATION_CHOICES,
+        unit="choice",
+        description="Number of choices returned by chat completions call",
+    )
+
+    duration_histogram = meter.create_histogram(
+        name=Meters.LLM_OPERATION_DURATION,
+        unit="s",
+        description="GenAI operation duration",
+    )
+
+    exception_counter = meter.create_counter(
+        name=Meters.LLM_ANTHROPIC_COMPLETION_EXCEPTIONS,
+        unit="time",
+        description="Number of exceptions occurred during chat completions",
+    )
+
+    return token_histogram, choice_counter, duration_histogram, exception_counter
+
+
+@dont_throw
+def _handle_input(span: Span, event_logger: Optional[EventLogger], kwargs):
+    if should_emit_events() and event_logger:
+        emit_input_events(event_logger, kwargs)
+    else:
+        if not span.is_recording():
+            return
+        run_async(aset_input_attributes(span, kwargs))
+
+
+@dont_throw
+async def _ahandle_input(span: Span, event_logger: Optional[EventLogger], kwargs):
+    if should_emit_events() and event_logger:
+        emit_input_events(event_logger, kwargs)
+    else:
+        if not span.is_recording():
+            return
+        await aset_input_attributes(span, kwargs)
+
+
+@dont_throw
+def _handle_response(span: Span, event_logger: Optional[EventLogger], response):
+    if should_emit_events():
+        emit_response_events(event_logger, response)
+    else:
+        if not span.is_recording():
+            return
+        set_response_attributes(span, response)
+
+
+@_with_chat_telemetry_wrapper
+def _wrap(
+    tracer: Tracer,
+    token_histogram: Histogram,
+    choice_counter: Counter,
+    duration_histogram: Histogram,
+    exception_counter: Counter,
+    event_logger: Optional[EventLogger],
+    to_wrap,
+    wrapped,
+    instance,
+    args,
+    kwargs,
+):
+    """Instruments and calls every function defined in TO_WRAP."""
+    if context_api.get_value(_SUPPRESS_INSTRUMENTATION_KEY) or context_api.get_value(
+        SUPPRESS_LANGUAGE_MODEL_INSTRUMENTATION_KEY
+    ):
+        return wrapped(*args, **kwargs)
+
+    name = to_wrap.get("span_name")
+    span = tracer.start_span(
+        name,
+        kind=SpanKind.CLIENT,
+        attributes={
+            SpanAttributes.LLM_SYSTEM: "anthropic",
+            SpanAttributes.LLM_REQUEST_TYPE: LLMRequestTypeValues.COMPLETION.value,
+        },
+        context=get_current_context(),
+    )
+
+    _handle_input(span, event_logger, kwargs)
+
+    start_time = time.time()
+    try:
+        response = wrapped(*args, **kwargs)
+    except Exception as e:  # pylint: disable=broad-except
+        end_time = time.time()
+        attributes = error_metrics_attributes(e)
+
+        if duration_histogram:
+            duration = end_time - start_time
+            duration_histogram.record(duration, attributes=attributes)
+
+        if exception_counter:
+            exception_counter.add(1, attributes=attributes)
+
+        raise e
+
+    end_time = time.time()
+
+    if is_streaming_response(response):
+        return build_from_streaming_response(
+            span,
+            response,
+            instance._client,
+            start_time,
+            token_histogram,
+            choice_counter,
+            duration_histogram,
+            exception_counter,
+            event_logger,
+            kwargs,
+        )
+    elif response:
+        try:
+            metric_attributes = shared_metrics_attributes(response)
+
+            if duration_histogram:
+                duration = time.time() - start_time
+                duration_histogram.record(
+                    duration,
+                    attributes=metric_attributes,
+                )
+
+            _handle_response(span, event_logger, response)
+            if span.is_recording():
+                _set_token_usage(
+                    span,
+                    instance._client,
+                    kwargs,
+                    response,
+                    metric_attributes,
+                    token_histogram,
+                    choice_counter,
+                )
+        except Exception as ex:  # pylint: disable=broad-except
+            logger.warning(
+                "Failed to set response attributes for anthropic span, error: %s",
+                str(ex),
+            )
+
+        if span.is_recording():
+            span.set_status(Status(StatusCode.OK))
+    span.end()
+    return response
+
+
+@_with_chat_telemetry_wrapper
+async def _awrap(
+    tracer,
+    token_histogram: Histogram,
+    choice_counter: Counter,
+    duration_histogram: Histogram,
+    exception_counter: Counter,
+    event_logger: Optional[EventLogger],
+    to_wrap,
+    wrapped,
+    instance,
+    args,
+    kwargs,
+):
+    """Instruments and calls every function defined in TO_WRAP."""
+    if context_api.get_value(_SUPPRESS_INSTRUMENTATION_KEY) or context_api.get_value(
+        SUPPRESS_LANGUAGE_MODEL_INSTRUMENTATION_KEY
+    ):
+        return await wrapped(*args, **kwargs)
+
+    name = to_wrap.get("span_name")
+    span = tracer.start_span(
+        name,
+        kind=SpanKind.CLIENT,
+        attributes={
+            SpanAttributes.LLM_SYSTEM: "anthropic",
+            SpanAttributes.LLM_REQUEST_TYPE: LLMRequestTypeValues.COMPLETION.value,
+        },
+        context=get_current_context(),
+    )
+    await _ahandle_input(span, event_logger, kwargs)
+
+    start_time = time.time()
+    try:
+        response = await wrapped(*args, **kwargs)
+    except Exception as e:  # pylint: disable=broad-except
+        end_time = time.time()
+        attributes = error_metrics_attributes(e)
+
+        if duration_histogram:
+            duration = end_time - start_time
+            duration_histogram.record(duration, attributes=attributes)
+
+        if exception_counter:
+            exception_counter.add(1, attributes=attributes)
+
+        raise e
+
+    if is_streaming_response(response):
+        return abuild_from_streaming_response(
+            span,
+            response,
+            instance._client,
+            start_time,
+            token_histogram,
+            choice_counter,
+            duration_histogram,
+            exception_counter,
+            event_logger,
+            kwargs,
+        )
+    elif response:
+        metric_attributes = shared_metrics_attributes(response)
+
+        if duration_histogram:
+            duration = time.time() - start_time
+            duration_histogram.record(
+                duration,
+                attributes=metric_attributes,
+            )
+
+        _handle_response(span, event_logger, response)
+
+        if span.is_recording():
+            await _aset_token_usage(
+                span,
+                instance._client,
+                kwargs,
+                response,
+                metric_attributes,
+                token_histogram,
+                choice_counter,
+            )
+            span.set_status(Status(StatusCode.OK))
+    span.end()
+    return response
+
+
+def is_metrics_enabled() -> bool:
+    return False
+
+
+class AnthropicInstrumentor(BaseInstrumentor):
+    """An instrumentor for Anthropic's client library."""
+
+    def __init__(
+        self,
+        enrich_token_usage: bool = False,
+        exception_logger=None,
+        use_legacy_attributes: bool = True,
+        get_common_metrics_attributes: Callable[[], dict] = lambda: {},
+        upload_base64_image: Optional[
+            Callable[[str, str, str, str], Coroutine[None, None, str]]
+        ] = None,
+    ):
+        super().__init__()
+        Config.exception_logger = exception_logger
+        Config.enrich_token_usage = enrich_token_usage
+        Config.get_common_metrics_attributes = get_common_metrics_attributes
+        Config.upload_base64_image = upload_base64_image
+        Config.use_legacy_attributes = use_legacy_attributes
+
+    def instrumentation_dependencies(self) -> Collection[str]:
+        return _instruments
+
+    def _instrument(self, **kwargs):
+        tracer_provider = kwargs.get("tracer_provider")
+        tracer = get_tracer(__name__, __version__, tracer_provider)
+
+        # meter and counters are inited here
+        meter_provider = kwargs.get("meter_provider")
+        meter = get_meter(__name__, __version__, meter_provider)
+
+        if is_metrics_enabled():
+            (
+                token_histogram,
+                choice_counter,
+                duration_histogram,
+                exception_counter,
+            ) = _create_metrics(meter)
+        else:
+            (
+                token_histogram,
+                choice_counter,
+                duration_histogram,
+                exception_counter,
+            ) = (None, None, None, None)
+
+        # event_logger is inited here
+        event_logger = None
+
+        if not Config.use_legacy_attributes:
+            event_logger_provider = kwargs.get("event_logger_provider")
+            event_logger = get_event_logger(
+                __name__, __version__, event_logger_provider=event_logger_provider
+            )
+
+        for wrapped_method in WRAPPED_METHODS:
+            wrap_package = wrapped_method.get("package")
+            wrap_object = wrapped_method.get("object")
+            wrap_method = wrapped_method.get("method")
+
+            try:
+                wrap_function_wrapper(
+                    wrap_package,
+                    f"{wrap_object}.{wrap_method}",
+                    _wrap(
+                        tracer,
+                        token_histogram,
+                        choice_counter,
+                        duration_histogram,
+                        exception_counter,
+                        event_logger,
+                        wrapped_method,
+                    ),
+                )
+            except ModuleNotFoundError:
+                pass  # that's ok, we don't want to fail if some methods do not exist
+
+        for wrapped_method in WRAPPED_AMETHODS:
+            wrap_package = wrapped_method.get("package")
+            wrap_object = wrapped_method.get("object")
+            wrap_method = wrapped_method.get("method")
+            try:
+                wrap_function_wrapper(
+                    wrap_package,
+                    f"{wrap_object}.{wrap_method}",
+                    _awrap(
+                        tracer,
+                        token_histogram,
+                        choice_counter,
+                        duration_histogram,
+                        exception_counter,
+                        event_logger,
+                        wrapped_method,
+                    ),
+                )
+            except ModuleNotFoundError:
+                pass  # that's ok, we don't want to fail if some methods do not exist
+
+    def _uninstrument(self, **kwargs):
+        for wrapped_method in WRAPPED_METHODS:
+            wrap_package = wrapped_method.get("package")
+            wrap_object = wrapped_method.get("object")
+            unwrap(
+                f"{wrap_package}.{wrap_object}",
+                wrapped_method.get("method"),
+            )
+        for wrapped_method in WRAPPED_AMETHODS:
+            wrap_package = wrapped_method.get("package")
+            wrap_object = wrapped_method.get("object")
+            unwrap(
+                f"{wrap_package}.{wrap_object}",
+                wrapped_method.get("method"),
+            )

lmnr/opentelemetry_lib/opentelemetry/instrumentation/anthropic/config.py

@@ -0,0 +1,13 @@
+from typing import Callable, Optional
+
+from typing_extensions import Coroutine
+
+
+class Config:
+    enrich_token_usage = False
+    exception_logger = None
+    get_common_metrics_attributes: Callable[[], dict] = lambda: {}
+    upload_base64_image: Optional[
+        Callable[[str, str, str, str], Coroutine[None, None, str]]
+    ] = None
+    use_legacy_attributes = True