Manual Tracing

To measure and monitor operations that are not automatically instrumented, e.g. tool calls, wrap the code with trace() method or use @trace_function decorator.

To record payloads and track usage metrics, use Span.set_payload().

with graphsignal.trace('my-tool') as span:
    ...
    span.set_payload('my-data', data, usage=dict(size=my_data_size))

You can also record sub-operations to visualize operation timeline and latency breakdown.

with graphsignal.trace('inference') as op_span:
    with span.trace('model-cold-boot') as subop_span:
        ...

@graphsignal.trace_function
def my_function():
    ...

with graphsignal.trace('generate-completion', tags=dict(model_type='chat')) as span:
    res_data = my_llm_call(req_data)
    ...
    span.set_payload('request', req_data)
    span.set_payload('response', res_data)
    ...
    span.set_usage('prompt_tokens', prompt_tokens)
    span.set_usage('completion_tokens', completion_tokens)

For auto-instrumented libraries, or when using @trace_function decorator, trace() method with with context manager or callbacks, exceptions are automatically recorded. For other cases, use Span.add_exception.