Sampling profiler for Python programs
py-spy is a sampling profiler for Python programs. It lets you visualize what your Python program is spending time on without restarting the program or modifying the code in any way. py-spy is extremely low overhead: it is written in Rust for speed and doesn't run in the same process as the profiled Python program. This means py-spy is safe to use against production Python code.
py-spy works on Linux, OSX, Windows and FreeBSD, and supports profiling all recent versions of the CPython interpreter (versions 2.3-2.7 and 3.3-3.8).
Prebuilt binary wheels can be installed from PyPI with:
pip install py-spy
You can also download prebuilt binaries from the GitHub Releases Page. This includes binaries for ARM and FreeBSD, which can't be installed using pip. If you're a Rust user, py-spy can also be installed with:
cargo install py-spy. On Arch Linux, py-spy is in AUR and can be installed with
yay -S py-spy.
py-spy works from the command line and takes either the PID of the program you want to sample from or the command line of the python program you want to run. py-spy has three subcommands
py-spy supports recording profiles to a file using the
recordcommand. For example, you can generate a flame graph of your python process by going:
py-spy record -o profile.svg --pid 12345 # OR py-spy record -o profile.svg -- python myprogram.py
Which will generate an interactive SVG file looking like:
You can change the file format to generate speedscope profiles or raw data with the
py-spy record --helpfor information on other options including changing the sampling rate, filtering to only include threads that hold the GIL, profiling native C extensions, showing thread-ids, profiling subprocesses and more.
Top shows a live view of what functions are taking the most time in your python program, similar to the Unix top command. Running py-spy with:
py-spy top --pid 12345 # OR py-spy top -- python myprogram.py
will bring up a live updating high level view of your python program:
py-spy can also display the current call stack for each python thread with the
py-spy dump --pid 12345
This will dump out the call stacks for each thread, and some other basic process info to the console:
This is useful for the case where you just need a single call stack to figure out where your python program is hung on. This command also has the ability to print out the local variables associated with each stack frame by setting the
This project aims to let you profile and debug any running Python program, even if the program is serving production traffic.
While there are many other python profiling projects, almost all of them require modifying the profiled program in some way. Usually, the profiling code runs inside of the target python process, which will slow down and change how the program operates. This means it's not generally safe to use these profilers for debugging issues in production services since they will usually have a noticeable impact on performance.
Figuring out the call stack of the Python program is done by looking at the global PyInterpreterState variable to get all the Python threads running in the interpreter, and then iterating over each PyFrameObject in each thread to get the call stack. Since the Python ABI changes between versions, we use rust's bindgen to generate different rust structures for each Python interpreter class we care about and use these generated structs to figure out the memory layout in the Python program.
Getting the memory address of the Python Interpreter can be a little tricky due to Address Space Layout Randomization. If the target python interpreter ships with symbols it is pretty easy to figure out the memory address of the interpreter by dereferencing the
_PyRuntimevariables depending on the Python version. However, many Python versions are shipped with either stripped binaries or shipped without the corresponding PDB symbol files on Windows. In these cases we scan through the BSS section for addresses that look like they may point to a valid PyInterpreterState and check if the layout of that address is what we expect.
Yes! py-spy supports profiling native python extensions written in languages like C/C++ or Cython, on x86_64 Linux and Windows. You can enable this mode by passing
--nativeon the command line. For best results, you should compile your Python extension with symbols. Also worth noting for Cython programs is that py-spy needs the generated C or C++ file in order to return line numbers of the original .pyx file. Read the blog post for more information.
By passing in the
--subprocessesflag to either the record or top view, py-spy will also include the output from any python process that is a child process of the target program. This is useful for profiling applications that use multiprocessing or gunicorn worker pools. py-spy will monitor for new processes being created, and automatically attach to them and include samples from them in the output. The record view will include the PID and cmdline of each program in the callstack, with subprocesses appearing as children of their parent processes.
py-spy works by reading memory from a different python process, and this might not be allowed for security reasons depending on your OS and system settings. In many cases, running as a root user (with sudo or similar) gets around these security restrictions. OSX always requires running as root, but on Linux it depends on how you are launching py-spy and the system security settings.
On Linux the default configuration is to require root permissions when attaching to a process that isn't a child. For py-spy this means you can profile without root access by getting py-spy to create the process (
py-spy -- python myprogram.py) but attaching to an existing process by specifying a PID will usually require root (
sudo py-spy --pid 123456). You can remove this restriction on Linux by setting the ptrace_scope sysctl variable.
py-spy attempts to only include stack traces from threads that are actively running code, and exclude threads that are sleeping or otherwise idle. When possible, py-spy attempts to get this thread activity information from the OS: by reading in
/proc/PID/staton Linux, by using the mach threadbasicinfo call on OSX, and by looking if the current SysCall is known to be idle on Windows.
There are some limitations with this approach though that may cause idle threads to still be marked as active. First off, we have to get this thread activity information before pausing the program, because getting this from a paused program will cause it to always return that this is idle. This means there is a potential race condition, where we get the thread activity and then the thread is in a different state when we get the stack trace. Querying the OS for thread activity also isn't implemented yet for FreeBSD and i686/ARM processors on Linux. On Windows, calls that are blocked on IO also won't be marked as idle yet, for instance when reading input from stdin. Finally, on some Linux calls the ptrace attach that we are using may cause idle threads to wake up momentarily, causing false positives when reading from procfs. For these reasons, we also have a heuristic fallback that marks known certain known calls in python as being idle.
You can disable this functionality by setting the
--idleflag, which will include frames that py-spy considers idle.
We get GIL activity by looking at the threadid value pointed to by the
_PyThreadState_Currentsymbol for Python 3.6 and earlier and by figuring out the equivalent from the
_PyRuntimestruct in Python 3.7 and later. These symbols might not be included in your python distribution, which will cause resolving which thread holds on to the GIL to fail. Current GIL usage is also shown in the
topview as %GIL.
--gilflag will only include traces for threads that are holding on to the Global Interpreter Lock. In some cases this might be a more accurate view of how your python program is spending its time, though you should be aware that this will miss activity in extensions that release the GIL while still active.
OSX has a feature called System Integrity Protection that prevents even the root user from reading memory from any binary located in /usr/bin. Unfortunately, this includes the python interpreter that ships with OSX.
There are a couple of different ways to deal with this: * You can install a different Python distribution. The built-in Python will be removed in a future OSX, and you probably want to migrate away from Python 2 anyways =). * You can use virtualenv to run the system python in an environment where SIP doesn't apply. * You can disable System Integrity Protection.
Running py-spy inside of a docker container will also usually bring up a permissions denied error even when running as root.
Alternatively you can edit the docker-compose yaml file
your_service: cap_add: - SYS_PTRACE
Note that you'll need to restart the docker container in order for this setting to take effect.
You can also use py-spy from the Host OS to profile a running process running inside the docker container.
SYS_PTRACEto be able to read process memory. Kubernetes drops that capability by default, resulting in the error
Permission Denied: Try running again with elevated permissions by going 'sudo env "PATH=$PATH" !!'The recommended way to deal with this is to edit the spec and add that capability. For a deployment, this is done by adding this to
securityContext: capabilities: add: - SYS_PTRACEMore details on this here: https://kubernetes.io/docs/tasks/configure-pod-container/security-context/#set-capabilities-for-a-container Note that this will remove the existing pods and create those again.
Alpine python opts out of the
manylinuxwheels: pypa/pip#3969 (comment). You can override this behaviour to use pip to install py-spy on Alpine by going:
echo 'manylinux1_compatible = True' > /usr/local/lib/python3.7/site-packages/_manylinux.py
Alternatively you can download a musl binary from the GitHub releases page.
By setting the
--nonblockingoption, py-spy won't pause the target python you are profiling from. While the performance impact of sampling from a process with py-spy is usually extremely low, setting this option will totally avoid interrupting your running python program.
With this option set, py-spy will instead read the interpreter state from the python process as it is running. Since the calls we use to read memory from are not atomic, and we have to issue multiple calls to get a stack trace this means that occasionally we get errors when sampling. This can show up as an increased error rate when sampling, or as partial stack frames being included in the output.
Ok, so no-one has ever actually asked me this - but I wanted to share since it's a pretty terrible hack that might be useful to other people.
I really wanted to distribute this package over PyPI, since installing with pip will make this much easier for most Python programmers to get installed on their system. Unfortunately, installing executables as python scripts isn't something that setuptools supports.
To get around this I'm using setuptools_rust package to build the py-spy binary, and then overriding the distutils install command to copy the built binary into the python scripts folder. By doing this with prebuilt wheels for supported platforms means that we can install py-spy with pip, and not require a Rust compiler on the machine that this is being installed onto.
Not yet =).
If there are features you'd like to see in py-spy either thumb up the appropriate issue or create a new one that describes what functionality is missing.
py-spy is heavily inspired by Julia Evans excellent work on rbspy. In particular, the code to generate flamegraph and speedscope files is taken directly from rbspy, and this project uses the read-process-memory and proc-maps crates that were spun off from rbspy.
py-spy is released under the MIT License, see the LICENSE file for the full text.