Skip to content

LangChain

Use DDTrace to obtain cost estimates, prompt and completion sampling, error tracking, performance metrics, etc., from requests made using the LangChain Python library.

Configuration

Install DDTrace

pip install ddtrace>=1.17

DataKit Configuration

  • Enable the DDTrace collector

The DDTrace collector is used to collect trace information. Go to the DataKit installation directory and execute conf.d/ddtrace/, copy ddtrace.conf.sample and rename it to ddtrace.conf.

  • Enable the StatsD collector

The StatsD collector is used to collect metrics information. The default port is 8125.

  • Restart DataKit
systemctl restart datakit

Run the Application

DD_SERVICE="my-langchain" DD_ENV="dev" DD_AGENT_HOST="localhost" DD_AGENT_PORT="9529" ddtrace-run python <your-app>.py

If you need to enable debug mode, add the --debug parameter when starting:

DD_SERVICE="my-langchain" DD_ENV="dev" DD_AGENT_HOST="localhost" DD_AGENT_PORT="9529" ddtrace-run --debug python <your-app>.py

Metrics

langchain

Metrics Units Description
request_duration nanoseconds Distribution of request duration.
request_error errors Number of request exceptions.
tokens_completion tokens/request Number of tokens used in the completion response.
tokens_prompt tokens/request Number of tokens used in the request prompt.
tokens_total tokens/request Total number of tokens used in both request and response.
tokens_total_cost dollars Estimated cost (in dollars) based on token usage.

Feedback

Is this page helpful? ×