pytb 0.0.10

Utility for viewing stacktraces of running python processes

Utility for inspecting python stack of running process (x64 linux only)

Features:

• Python3 friendly (pytb itself requires python2 to run, but process being analyzed can be python3)
• Greenlet (gevent) friendly, can show stacktrace for each greenlet
• Works by just reading /proc/<pid>/mem, so doesn’t need to attach to target process (e.g. if you don’t want to stop the target process, or are debugging an unattachable process, like a D-state process)

Bugs:

• Linux only, 64-bit only.

Installation:

pip install pytb

Usage:

sudo pytb [pid]

will show you stacktrace of each python thread.

If you want to also see greenlet stacktraces, use

sudo pytb -g [pid]

this will also show you greenlets stacktraces. NOTE: in order to find all greenlets pytb looks for all gc tracked objects, so this may take a while if you have lots of objects.

Notes:

Supports viewing python2 and python3 tracebacks (it will try to guess, whether it’s 2 or 3, however you can help it by specifying -2 or -3 option)

The utility works by reading /proc/<pid>/mem (this is why sudo is needed). It uses nm and objdump for finding inner python structures in memory.

It’s also possible that python process is compiled with different flags that influence inner python’s structure fields alignment (on which pytb heavily relies) In this case you can run it with -d python-executable option where python-executable is an executable that contains debug symbols for the process you are interested in. pytb will launch gdb in this case and ask it for offsets of fields it’s interested in.

There is also a gotcha that frame->f_code->co_filename contains not an absolute path, but relative, and if your process has chdir’ed somewhere, those relative paths will become invalid (they are used when rendering stacktraces). For this situation there is --scriptdir dir argument which defaults to /proc/<pid>/cwd and serves as a root for co_filename.