From 723ebe76e787cfa6b08cc9587dd679f3234a1025 Mon Sep 17 00:00:00 2001 From: "Erlend E. Aasland" Date: Thu, 27 Oct 2022 15:06:48 +0200 Subject: gh-96143: Improve perf profiler docs (#96445) --- Doc/howto/perf_profiling.rst | 69 ++++++++++++++++++++++++-------------------- Doc/library/sys.rst | 32 ++++++++++++++++++++ Doc/using/cmdline.rst | 17 ++++++----- Doc/whatsnew/3.12.rst | 22 ++++++++++++++ Python/clinic/sysmodule.c.h | 10 ++++--- Python/sysmodule.c | 14 +++++---- 6 files changed, 116 insertions(+), 48 deletions(-) diff --git a/Doc/howto/perf_profiling.rst b/Doc/howto/perf_profiling.rst index 387fb3f..ad2eb7b 100644 --- a/Doc/howto/perf_profiling.rst +++ b/Doc/howto/perf_profiling.rst @@ -8,10 +8,11 @@ Python support for the Linux ``perf`` profiler :author: Pablo Galindo -The Linux ``perf`` profiler is a very powerful tool that allows you to profile and -obtain information about the performance of your application. ``perf`` also has -a very vibrant ecosystem of tools that aid with the analysis of the data that it -produces. +`The Linux perf profiler `_ +is a very powerful tool that allows you to profile and obtain +information about the performance of your application. +``perf`` also has a very vibrant ecosystem of tools +that aid with the analysis of the data that it produces. The main problem with using the ``perf`` profiler with Python applications is that ``perf`` only allows to get information about native symbols, this is, the names of @@ -25,7 +26,7 @@ fly before the execution of every Python function and it will teach ``perf`` the relationship between this piece of code and the associated Python function using `perf map files`_. -.. warning:: +.. note:: Support for the ``perf`` profiler is only currently available for Linux on selected architectures. Check the output of the configure build step or @@ -51,11 +52,11 @@ For example, consider the following script: if __name__ == "__main__": baz(1000000) -We can run perf to sample CPU stack traces at 9999 Hertz: +We can run ``perf`` to sample CPU stack traces at 9999 Hertz:: $ perf record -F 9999 -g -o perf.data python my_script.py -Then we can use perf report to analyze the data: +Then we can use ``perf`` report to analyze the data: .. code-block:: shell-session @@ -101,7 +102,7 @@ As you can see here, the Python functions are not shown in the output, only ``_P functions use the same C function to evaluate bytecode so we cannot know which Python function corresponds to which bytecode-evaluating function. -Instead, if we run the same experiment with perf support activated we get: +Instead, if we run the same experiment with ``perf`` support enabled we get: .. code-block:: shell-session @@ -147,52 +148,58 @@ Instead, if we run the same experiment with perf support activated we get: -Enabling perf profiling mode ----------------------------- +How to enable ``perf`` profiling support +---------------------------------------- -There are two main ways to activate the perf profiling mode. If you want it to be -active since the start of the Python interpreter, you can use the ``-Xperf`` option: +``perf`` profiling support can either be enabled from the start using +the environment variable :envvar:`PYTHONPERFSUPPORT` or the +:option:`-X perf <-X>` option, +or dynamically using :func:`sys.activate_stack_trampoline` and +:func:`sys.deactivate_stack_trampoline`. - $ python -Xperf my_script.py +The :mod:`!sys` functions take precedence over the :option:`!-X` option, +the :option:`!-X` option takes precedence over the environment variable. -You can also set the :envvar:`PYTHONPERFSUPPORT` to a nonzero value to actiavate perf -profiling mode globally. +Example, using the environment variable:: -There is also support for dynamically activating and deactivating the perf -profiling mode by using the APIs in the :mod:`sys` module: + $ PYTHONPERFSUPPORT=1 + $ python script.py + $ perf report -g -i perf.data -.. code-block:: python - - import sys - sys.activate_stack_trampoline("perf") +Example, using the :option:`!-X` option:: - # Run some code with Perf profiling active + $ python -X perf script.py + $ perf report -g -i perf.data - sys.deactivate_stack_trampoline() +Example, using the :mod:`sys` APIs in file :file:`example.py`: - # Perf profiling is not active anymore +.. code-block:: python -These APIs can be handy if you want to activate/deactivate profiling mode in -response to a signal or other communication mechanism with your process. + import sys + sys.activate_stack_trampoline("perf") + do_profiled_stuff() + sys.deactivate_stack_trampoline() + non_profiled_stuff() -Now we can analyze the data with ``perf report``: +...then:: - $ perf report -g -i perf.data + $ python ./example.py + $ perf report -g -i perf.data How to obtain the best results -------------------------------- +------------------------------ For the best results, Python should be compiled with ``CFLAGS="-fno-omit-frame-pointer -mno-omit-leaf-frame-pointer"`` as this allows profilers to unwind using only the frame pointer and not on DWARF debug -information. This is because as the code that is interposed to allow perf +information. This is because as the code that is interposed to allow ``perf`` support is dynamically generated it doesn't have any DWARF debugging information available. -You can check if you system has been compiled with this flag by running: +You can check if your system has been compiled with this flag by running:: $ python -m sysconfig | grep 'no-omit-frame-pointer' diff --git a/Doc/library/sys.rst b/Doc/library/sys.rst index 542b08b..17bbfa5 100644 --- a/Doc/library/sys.rst +++ b/Doc/library/sys.rst @@ -1555,6 +1555,38 @@ always available. This function has been added on a provisional basis (see :pep:`411` for details.) Use it only for debugging purposes. +.. function:: activate_stack_trampoline(backend, /) + + Activate the stack profiler trampoline *backend*. + The only supported backend is ``"perf"``. + + .. availability:: Linux. + + .. versionadded:: 3.12 + + .. seealso:: + + * :ref:`perf_profiling` + * https://perf.wiki.kernel.org + +.. function:: deactivate_stack_trampoline() + + Deactivate the current stack profiler trampoline backend. + + If no stack profiler is activated, this function has no effect. + + .. availability:: Linux. + + .. versionadded:: 3.12 + +.. function:: is_stack_trampoline_active() + + Return ``True`` if a stack profiler trampoline is active. + + .. availability:: Linux. + + .. versionadded:: 3.12 + .. function:: _enablelegacywindowsfsencoding() Changes the :term:`filesystem encoding and error handler` to 'mbcs' and diff --git a/Doc/using/cmdline.rst b/Doc/using/cmdline.rst index 02c9f30..2a4d070 100644 --- a/Doc/using/cmdline.rst +++ b/Doc/using/cmdline.rst @@ -538,12 +538,11 @@ Miscellaneous options development (running from the source tree) then the default is "off". Note that the "importlib_bootstrap" and "importlib_bootstrap_external" frozen modules are always used, even if this flag is set to "off". - * ``-X perf`` to activate compatibility mode with the ``perf`` profiler. - When this option is activated, the Linux ``perf`` profiler will be able to + * ``-X perf`` enables support for the Linux ``perf`` profiler. + When this option is provided, the ``perf`` profiler will be able to report Python calls. This option is only available on some platforms and will do nothing if is not supported on the current system. The default value - is "off". See also :envvar:`PYTHONPERFSUPPORT` and :ref:`perf_profiling` - for more information. + is "off". See also :envvar:`PYTHONPERFSUPPORT` and :ref:`perf_profiling`. It also allows passing arbitrary values and retrieving them through the :data:`sys._xoptions` dictionary. @@ -1048,9 +1047,13 @@ conflict. .. envvar:: PYTHONPERFSUPPORT - If this variable is set to a nonzero value, it activates compatibility mode - with the ``perf`` profiler so Python calls can be detected by it. See the - :ref:`perf_profiling` section for more information. + If this variable is set to a nonzero value, it enables support for + the Linux ``perf`` profiler so Python calls can be detected by it. + + If set to ``0``, disable Linux ``perf`` profiler support. + + See also the :option:`-X perf <-X>` command-line option + and :ref:`perf_profiling`. .. versionadded:: 3.12 diff --git a/Doc/whatsnew/3.12.rst b/Doc/whatsnew/3.12.rst index 8f8a994..39ad3ac 100644 --- a/Doc/whatsnew/3.12.rst +++ b/Doc/whatsnew/3.12.rst @@ -74,6 +74,15 @@ Important deprecations, removals or restrictions: New Features ============ +* Add :ref:`perf_profiling` through the new + environment variable :envvar:`PYTHONPERFSUPPORT`, + the new command-line option :option:`-X perf <-X>`, + as well as the new :func:`sys.activate_stack_trampoline`, + :func:`sys.deactivate_stack_trampoline`, + and :func:`sys.is_stack_trampoline_active` APIs. + (Design by Pablo Galindo. Contributed by Pablo Galindo and Christian Heimes + with contributions from Gregory P. Smith [Google] and Mark Shannon + in :gh:`96123`.) Other Language Changes @@ -194,6 +203,19 @@ tempfile The :class:`tempfile.NamedTemporaryFile` function has a new optional parameter *delete_on_close* (Contributed by Evgeny Zorin in :gh:`58451`.) +sys +--- + +* Add :func:`sys.activate_stack_trampoline` and + :func:`sys.deactivate_stack_trampoline` for activating and deactivating + stack profiler trampolines, + and :func:`sys.is_stack_trampoline_active` for querying if stack profiler + trampolines are active. + (Contributed by Pablo Galindo and Christian Heimes + with contributions from Gregory P. Smith [Google] and Mark Shannon + in :gh:`96123`.) + + Optimizations ============= diff --git a/Python/clinic/sysmodule.c.h b/Python/clinic/sysmodule.c.h index 6864b8b..3dc7aa8 100644 --- a/Python/clinic/sysmodule.c.h +++ b/Python/clinic/sysmodule.c.h @@ -1231,7 +1231,7 @@ PyDoc_STRVAR(sys_activate_stack_trampoline__doc__, "activate_stack_trampoline($module, backend, /)\n" "--\n" "\n" -"Activate the perf profiler trampoline."); +"Activate stack profiler trampoline *backend*."); #define SYS_ACTIVATE_STACK_TRAMPOLINE_METHODDEF \ {"activate_stack_trampoline", (PyCFunction)sys_activate_stack_trampoline, METH_O, sys_activate_stack_trampoline__doc__}, @@ -1268,7 +1268,9 @@ PyDoc_STRVAR(sys_deactivate_stack_trampoline__doc__, "deactivate_stack_trampoline($module, /)\n" "--\n" "\n" -"Dectivate the perf profiler trampoline."); +"Deactivate the current stack profiler trampoline backend.\n" +"\n" +"If no stack profiler is activated, this function has no effect."); #define SYS_DEACTIVATE_STACK_TRAMPOLINE_METHODDEF \ {"deactivate_stack_trampoline", (PyCFunction)sys_deactivate_stack_trampoline, METH_NOARGS, sys_deactivate_stack_trampoline__doc__}, @@ -1286,7 +1288,7 @@ PyDoc_STRVAR(sys_is_stack_trampoline_active__doc__, "is_stack_trampoline_active($module, /)\n" "--\n" "\n" -"Returns *True* if the perf profiler trampoline is active."); +"Return *True* if a stack profiler trampoline is active."); #define SYS_IS_STACK_TRAMPOLINE_ACTIVE_METHODDEF \ {"is_stack_trampoline_active", (PyCFunction)sys_is_stack_trampoline_active, METH_NOARGS, sys_is_stack_trampoline_active__doc__}, @@ -1343,4 +1345,4 @@ sys_is_stack_trampoline_active(PyObject *module, PyObject *Py_UNUSED(ignored)) #ifndef SYS_GETANDROIDAPILEVEL_METHODDEF #define SYS_GETANDROIDAPILEVEL_METHODDEF #endif /* !defined(SYS_GETANDROIDAPILEVEL_METHODDEF) */ -/*[clinic end generated code: output=15318cdd96b62b06 input=a9049054013a1b77]*/ +/*[clinic end generated code: output=2b5e1bc24a3348bd input=a9049054013a1b77]*/ diff --git a/Python/sysmodule.c b/Python/sysmodule.c index 7081456..f73d332 100644 --- a/Python/sysmodule.c +++ b/Python/sysmodule.c @@ -2127,12 +2127,12 @@ sys.activate_stack_trampoline backend: str / -Activate the perf profiler trampoline. +Activate stack profiler trampoline *backend*. [clinic start generated code]*/ static PyObject * sys_activate_stack_trampoline_impl(PyObject *module, const char *backend) -/*[clinic end generated code: output=5783cdeb51874b43 input=b09020e3a17c78c5]*/ +/*[clinic end generated code: output=5783cdeb51874b43 input=a12df928758a82b4]*/ { #ifdef PY_HAVE_PERF_TRAMPOLINE if (strcmp(backend, "perf") == 0) { @@ -2163,12 +2163,14 @@ sys_activate_stack_trampoline_impl(PyObject *module, const char *backend) /*[clinic input] sys.deactivate_stack_trampoline -Dectivate the perf profiler trampoline. +Deactivate the current stack profiler trampoline backend. + +If no stack profiler is activated, this function has no effect. [clinic start generated code]*/ static PyObject * sys_deactivate_stack_trampoline_impl(PyObject *module) -/*[clinic end generated code: output=b50da25465df0ef1 input=491f4fc1ed615736]*/ +/*[clinic end generated code: output=b50da25465df0ef1 input=9f629a6be9fe7fc8]*/ { if (_PyPerfTrampoline_Init(0) < 0) { return NULL; @@ -2179,12 +2181,12 @@ sys_deactivate_stack_trampoline_impl(PyObject *module) /*[clinic input] sys.is_stack_trampoline_active -Returns *True* if the perf profiler trampoline is active. +Return *True* if a stack profiler trampoline is active. [clinic start generated code]*/ static PyObject * sys_is_stack_trampoline_active_impl(PyObject *module) -/*[clinic end generated code: output=ab2746de0ad9d293 input=061fa5776ac9dd59]*/ +/*[clinic end generated code: output=ab2746de0ad9d293 input=29616b7bf6a0b703]*/ { #ifdef PY_HAVE_PERF_TRAMPOLINE if (_PyIsPerfTrampolineActive()) { -- cgit v0.12