loongsuite-instrumentation-dashscope 0.4.0

LoongSuite DashScope instrumentation

This library allows tracing calls to Alibaba Cloud DashScope APIs.

Installation

pip install loongsuite-instrumentation-dashscope

Usage

from opentelemetry import trace
from opentelemetry.sdk.trace import TracerProvider
from opentelemetry.sdk.trace.export import ConsoleSpanExporter, SimpleSpanProcessor

from opentelemetry.instrumentation.dashscope import DashScopeInstrumentor

# Initialize tracing
trace.set_tracer_provider(TracerProvider())
trace.get_tracer_provider().add_span_processor(
    SimpleSpanProcessor(ConsoleSpanExporter())
)

# Instrument DashScope
DashScopeInstrumentor().instrument()

# Now use DashScope as normal
from dashscope import Generation

response = Generation.call(
    model="qwen-turbo",
    prompt="Hello!"
)

Supported APIs

• Text Generation

  • Generation.call (sync)

  • AioGeneration.call (async)

  • Streaming support for both sync and async

• Text Embedding

  • TextEmbedding.call

• Text Rerank

  • TextReRank.call

• Image Synthesis

  • ImageSynthesis.call (sync)

  • ImageSynthesis.async_call (async task submission)

  • ImageSynthesis.wait (async task waiting)

• Speech Synthesis V1

  • SpeechSynthesizer.call (V1)

• Speech Synthesis V2

  • SpeechSynthesizer.call (V2)

• Video Synthesis

  • VideoSynthesis.call (sync)

  • VideoSynthesis.async_call (async task submission)

  • VideoSynthesis.wait (async task waiting)

Captured Attributes

This instrumentation follows the OpenTelemetry GenAI Semantic Conventions and uses opentelemetry-util-genai for standardized telemetry collection.

For Text Generation (Chat Completion) operations, the following attributes are captured:

Required Attributes: * gen_ai.operation.name - Operation type (“chat”) * gen_ai.provider.name - Provider name (“dashscope”)

Conditionally Required Attributes: * gen_ai.request.model - Requested model name (if available) * gen_ai.response.model - Actual model used (if different from request) * gen_ai.response.id - Request ID from DashScope (if available) * error.type - Error type (if operation ended in error)

Recommended Attributes: * gen_ai.response.finish_reasons - Finish reasons array (e.g., [“stop”], [“length”]) * gen_ai.usage.input_tokens - Input token count * gen_ai.usage.output_tokens - Output token count * gen_ai.request.temperature - Temperature parameter (if provided) * gen_ai.request.top_p - Top-p parameter (if provided) * gen_ai.request.top_k - Top-k parameter (if provided) * gen_ai.request.max_tokens - Max tokens parameter (if provided)

Opt-In Attributes (require environment variable configuration): * gen_ai.input.messages - Chat history provided to the model * gen_ai.output.messages - Messages returned by the model

To enable Opt-In attributes, set the environment variable:

export OTEL_SEMCONV_STABILITY_OPT_IN=gen_ai_latest_experimental
export OTEL_INSTRUMENTATION_GENAI_CAPTURE_MESSAGE_CONTENT=SPAN_ONLY

For Text Embedding operations:

Required Attributes: * gen_ai.operation.name - Operation type (“embeddings”, note: plural form) * gen_ai.provider.name - Provider name (“dashscope”)

Conditionally Required Attributes: * gen_ai.request.model - Requested model name (if available) * error.type - Error type (if operation ended in error)

Recommended Attributes: * gen_ai.usage.input_tokens - Input token count (if available) * gen_ai.embeddings.dimension.count - Number of dimensions (if available) * gen_ai.request.encoding_formats - Encoding formats (if specified)

For Text Rerank operations:

Note: Rerank operations are not yet explicitly defined in GenAI semantic conventions. This instrumentation follows the pattern of other GenAI operations.

Basic Attributes: * gen_ai.operation.name - Operation type (“rerank”) * gen_ai.provider.name - Provider name (“dashscope”) * gen_ai.request.model - Requested model name (if available) * error.type - Error type (if operation ended in error)

Configuration

API Key

Set your DashScope API key via environment variable:

export DASHSCOPE_API_KEY="your-api-key-here"

Message Content Capture

By default, message content (gen_ai.input.messages and gen_ai.output.messages) is not captured to protect sensitive data. To enable content capture, set the following environment variables:

export OTEL_SEMCONV_STABILITY_OPT_IN=gen_ai_latest_experimental
export OTEL_INSTRUMENTATION_GENAI_CAPTURE_MESSAGE_CONTENT=SPAN_ONLY

Options for OTEL_INSTRUMENTATION_GENAI_CAPTURE_MESSAGE_CONTENT: * SPAN_ONLY - Capture content in span attributes only * SPAN_AND_EVENT - Capture content in both spans and events * EVENT_ONLY - Capture content in events only (not yet supported) * NO_CONTENT - Do not capture content (default)

References

• OpenTelemetry Project

• DashScope SDK