Profiling em Python

globo.com Profiling em Python

Friday, October 4, 13

porque profiling é importante?




então vamos otimizar tudo!



“We should forget about small efficiencies, say about 97% of the time: premature optimization is the root of all evil” - Donald Knuth


from timeit import timeit

if __name__ == "__main__": setup = "from htmlmin.minify import html_minify;" setup += "from data import raw_html" t = timeit( stmt="html_minify(raw_html)", setup=setup, number=100) print(t)

benchmark.py


$ python benchmark.py

25.8121981621


conheça seu código


‣ cProfile

‣ Profile

‣ hotshot (deprecated)

‣ trace

‣ line profiler

‣ memory profiler


from data import raw_htmlfrom htmlmin.minify import html_minify

if __name__ == "__main__": html_minify(raw_html)

profile.py


$ python -m cProfile profile.py












$ python -m cProfile -o out profile.py


$ python -m cProfile -o out profile.py


>>> import pstats>>> p = pstats.Stats("out")>>> p.sort_stats("cumulative").print_stats(10)

315165 function calls (311828 primitive calls) in 1.334 seconds

Ordered by: cumulative time List reduced from 675 to 10 due to restriction <10>

ncalls tottime percall cumtime percall filename:lineno(function) 1 0.092 0.092 1.358 1.358 profile.py:2(<module>) 1 0.038 0.038 0.879 0.879 /(...)/htmlmin/minify.py:7(<module>) 1 0.001 0.001 0.840 0.840 /(...)/bs4/__init__.py:17(<module>) 1 0.073 0.073 0.839 0.839 /(...)/bs4/builder/__init__.py:1(<module>) 1 0.027 0.027 0.511 0.511 /(...)/bs4/builder/_html5lib.py:2(<module>) 1 0.028 0.028 0.483 0.483 /(...)/html5lib/__init__.py:12(<module>) 1 0.001 0.001 0.387 0.387 /(...)/htmlmin/minify.py:26(html_minify) 2 0.000 0.000 0.297 0.148 /(...)/bs4/__init__.py:80(__init__) 2 0.000 0.000 0.296 0.148 /(...)/bs4/__init__.py:193(_feed) 2 0.000 0.000 0.296 0.148 /(...)/bs4/builder/_html5lib.py:33(feed)
































$ kernprof.py -l -v minify.py









$ python -m memory_profiler minify.py







gui?




outras ferramentas‣ meliae

‣ heapy (guppy)

‣ benchy

‣ valgrind

‣ python object graphs (objgraph)

‣ plop

‣ pycounters


bônus


django profiling


algumas ferramentas‣ django-debug-toolbar

‣ django-profiler

‣ new relic



$ newrelic-admin run-program gunicorn -w 3 wsgi:application



globo.com Estamos contratando!


globo.com obrigada!@flaviamissi


Profiling em Python

Technology

import html

pstats p

function calls

statsout p

primitive calls

html t

cumulative time list

timeit import timeit