To manually install the instrumentation use
patch_all as early as possible
in the application:
from ddtrace import patch_all patch_all()
To toggle instrumentation for a particular module:
from ddtrace import patch_all patch_all(redis=False, cassandra=False)
By default all supported libraries will be instrumented when
Note: To ensure that the supported libraries are instrumented properly in
the application, they must be patched prior to being imported. So make sure
patch_all before importing libraries that are to be instrumented.
More information about
patch_all is available in the
To extend the functionality of the
ddtrace library several APIs are
ddtrace provides a decorator that can be used to trace a particular method
in your application:
@tracer.wrap() def business_logic(): """A method that would be of interest to trace.""" # ... # ...
API documentation can be found here
# trace some interesting operation with tracer.trace('interesting.operations'): # do some interesting operation(s) # ... # ...
API documentation can be found here
Using the API¶
If the above methods are still not enough to satisfy your tracing needs, a manual API to provide complete control over starting and stopping spans is available:
span = tracer.trace('operations.of.interest') # span is started once created # do some operation(s) of interest in between # NOTE: be sure to call span.finish() or the trace will not be sent to # Datadog span.finish()
API details for creating and finishing spans can be found here:
To automatically profile your code, you can import the ddtrace.profiling.auto module. As soon as it is imported, it will start capturing CPU profiling information on your behalf:
If you want to control which part of your code should be profiled, you can use the ddtrace.profiling.Profiler object:
from ddtrace.profiling import Profiler prof = Profiler() prof.start() # At shutdown prof.stop()
The profiler has been designed to be always-on. The
methods are provided in case you need a fine-grained control over the
profiler lifecycle. They are not provided for starting and stopping the
profiler many times during your application lifecycle. Do not use them for
e.g. building a context manager.
When your process forks using os.fork, the profiler is stopped in the child process.
For Python 3.7 and later on POSIX platforms, a new profiler will be started if you enabled the profiler via ddtrace-run or ddtrace.profiling.auto.
If you manually instrument the profiler, or if you rely on Python 3.6 or a non-POSIX platform and earlier version, you’ll have to manually restart the profiler in your child.
The global profiler instrumented by ddtrace-run and ddtrace.profiling.auto can be started by calling ddtrace.profiling.auto.start_profiler.
The profiler supports the
asyncio library and retrieves the
asyncio.Task names to tag along the profiled data.
For this to work, the profiler replaces the default event loop policy with a custom policy that tracks threads to loop mapping.
The custom asyncio loop policy is installed by default at profiler startup. You
can disable this behavior by using the
asyncio_loop_policy parameter and
from ddtrace.profiling import Profiler prof = Profiler(asyncio_loop_policy=None)
You can also pass a custom class that implements the interface from
from ddtrace.profiling import Profiler prof = Profiler(asyncio_loop_policy=MyLoopPolicy())
If the loop policy has been overridden after the profiler has started, you can
always restore the profiler asyncio loop policy by calling
from ddtrace.profiling import Profiler prof = Profiler() prof.set_asyncio_event_loop_policy()