Profiling on live Django server?

问题:

I've never done code coverage in Python, but I'm looking for something like GCC's gcov, which tells me how many times each line executes, or Apple's Shark which gives a hierarchial breakdown of how long each function is taking.

My problem is that I have a live server which is experiencing high load, and I can't tell from the logs what's causing it. I would like to attach something to my Django instance to monitor which lines are the hottest and/or which functions are taking the longest time.

This is something like, but not exactly, code coverage. I would like to introduce it to a live running server, preferably without modifying too much.

Ideas?

回答1:

cProfile + RunSnakeRun: http://www.vrplumber.com/programming/runsnakerun/

回答2:

Django-live-profiler is a drop-in Django app that lets you profile your running application using statprof and visualize the results.

回答3:

I use this pattern: A monitor threads writes stacktraces to a log file every 0.3 seconds. After some hours I can see where the interpreter spends the most time. It does not slow down the my server:

Live Profiling of Python Server