Attend the Women in ML Symposium on December 7 Register now

Module: tff.profiler

Utility functions for instrumenting code with timing and tracing data.

This module provides several functions for preserving trace context across various boundaries, namely between asyncio and regular python code:

  • wrap_coroutine_in_trace_context wraps a coroutine such that it inherits the ambient trace context. It should be used when executing a coroutine that should inherit trace context from the current thread or task.
  • EventLoops should use the Task factory provided by propagate_trace_context_task_factory by calling set_task_factory(propagate_trace_context_task_factory).


class Generator

class LoggingTracingProvider: Implements TracingProvider and outputs the results via logging.

class ThreadLocalSpanYields: The span set for the current thread.

class TracedFunctionReturned: The traced function returned successfully.

class TracedFunctionThrew: The traced function threw an exception.

class TracedSpan: The trace was wrapping a non-function span.

class TracingProvider: Abstract base class for tracers.


add_tracing_provider(...): Add to the global list of tracing providers.

propagate_trace_context_task_factory(...): Creates a new task on loop to run coro, inheriting current spans.

set_tracing_providers(...): Set the global list of tracing providers, replacing any existing.

span(...): Creates a ContextManager that wraps the code in question with a span.

trace(...): Delegates to the current global TracingProvider.

with_trace_context_from_rpc(...): Attempts to pick up the trace context from the receiving RPC call.

wrap_coroutine_in_current_trace_context(...): Wraps the coroutine in the currently active span.

wrap_rpc_in_trace_context(...): Attempts to record the trace context into the enclosed RPC call.

Type Aliases



T Instance of typing.TypeVar