Thread Profiling in Python

@amjithr THREAD PROFILING DJANGOCON EU ▪ Wednesday, May 15, 13

@amjithr Hi  Wednesday, May 15, 13

@amjithr ♥ Wednesday, May 15, 13

Wednesday, May 15, 13

@amjithr PERFORMANCE! Wednesday, May 15, 13

@amjithr WHAT’S PROFILING? Wednesday, May 15, 13

@amjithr WHAT’S PROFILING? Measuring Performance Wednesday, May 15, 13

@amjithr cPROFILE Wednesday, May 15, 13

@amjithr $ python -m cProfile lcm.py 7780242 function calls in
3.822 seconds Ordered by: standard name ncalls tottime percall cumtime percall filename:lineno(function) 1 0.000 0.000 3.822 3.822 lcm.py:12(<module>) 1 2.383 2.383 3.822 3.822 lcm.py:12(lcm) 3890120 0.724 0.000 0.724 0.000 {max} 1 0.000 0.000 0.000 0.000 {method 'disable' of 3890119 0.716 0.000 0.716 0.000 {min} Wednesday, May 15, 13

@amjithr $ python -m cProfile shorten.py 88466 function calls (86374
primitive calls) in 0.220 seconds Ordered by: standard name ncalls tottime percall cumtime percall filename:lineno(function) 1 0.000 0.000 0.000 0.000 <string>:1(<lambda>) 1 0.000 0.000 0.000 0.000 <string>:1(<module>) 1 0.000 0.000 0.000 0.000 <string>:1(AnnotatedCTE) 1 0.000 0.000 0.000 0.000 <string>:1(AnnotatedCase) 1 0.000 0.000 0.000 0.000 <string>:1(AnnotatedCast) 1 0.000 0.000 0.000 0.000 <string>:1(AnnotatedClauseElement) 1 0.000 0.000 0.000 0.000 <string>:1(AnnotatedClauseList) 1 0.000 0.000 0.000 0.000 <string>:1(AnnotatedColumn) 1 0.000 0.000 0.000 0.000 <string>:1(AnnotatedColumnClause) 1 0.000 0.000 0.000 0.000 <string>:1(AnnotatedColumnElement) 1 0.000 0.000 0.000 0.000 <string>:1(AnnotatedCompoundSelect) 1 0.000 0.000 0.000 0.000 <string>:1(AnnotatedDelete) .... .... ..... ..... .... ...... .... .... ..... ..... .... ...... Wednesday, May 15, 13

@amjithr  RunSnakeRun Wednesday, May 15, 13

@amjithr Wednesday, May 15, 13

@amjithr $ time python fib.py python fib.py 5.43s user 0.01s
system 99% cpu 5.451 total $ time python -m cProfile fib.py python -m cProfile fib.py 9.63s user 0.04s system 99% cpu 9.670 total Wednesday, May 15, 13

@amjithr $ time python fib.py python fib.py 5.43s user 0.01s
system 99% cpu 5.451 total $ time python -m cProfile fib.py python -m cProfile fib.py 9.63s user 0.04s system 99% cpu 9.670 total ~ % OVERHEAD Wednesday, May 15, 13

@amjithr You can’t run it in production Wednesday, May 15,
13

@amjithr PROFILER IN PRODUCTION Wednesday, May 15, 13

@amjithr SLOW PRODUCTION NOT REALISTIC cProﬁle Wednesday, May 15, 13

@amjithr CRITICAL FUNCTIONS HYBRID PROFILING TARGETED PROFILING Wednesday, May 15,
13

@amjithr WEB FRAMEWORKS VIEW HANDLERS SQL CALLS TARGET Wednesday, May
15, 13

@amjithr TRANSACTION TRACE Wednesday, May 15, 13

@amjithr NON-INSTRUMENTED CODE Wednesday, May 15, 13

@amjithr Cool Story, Bro! Can we talk about statistical proﬁlers?

@amjithr STATISTICAL THREAD PROFILER Wednesday, May 15, 13

@amjithr WAT? Wednesday, May 15, 13

@amjithr OVERLY ATTACHED Wednesday, May 15, 13

@amjithr OVERLY ATTACHED Interrupt Inquire Collate Wednesday, May 15, 13

@amjithr OVERLY ATTACHED  Interrupt Inquire Collate Wednesday, May 15,
13

@amjithr OVERLY ATTACHED  ◴ Interrupt Inquire Collate Wednesday, May
15, 13

@amjithr  ◴ Interrupt Inquire Collate STATISTICAL PROFILER Wednesday, May
15, 13

@amjithr SIGNALS THREADS INTERRUPT Wednesday, May 15, 13

@amjithr HTTP://FLIC.KR/P/bE NXH SIGNALS Wednesday, May 15, 13

@amjithr import signal # Use the virtual timer sig =
signal.ITIMER_VIRTUAL # register a handler signal.signal(sig, handler) # 100ms repeating timer signal.setitimer(sig, 0.1, 0.1) SIGNAL Wednesday, May 15, 13

@amjithr PROs CONs Accurate timer Unix only - Ignored in
mod_wsgi - Can’t interrupt C-extensions SIGNAL Wednesday, May 15, 13

@amjithr HTTP://FLIC.KR/P/db ZS THREADS Wednesday, May 15, 13

@amjithr import threading # Create a background thread t =
threading.Thread(target=handler) t.setDaemon(True) # Start the thread t.start() # Sleep for 100ms in handler time.sleep(0.1) THREAD Wednesday, May 15, 13

@amjithr PROs CONs Cross Platform Inaccurate for CPU tasks mod_wsgi
compatible Can’t interrupt C-extensions THREAD Wednesday, May 15, 13

@amjithr INQUIRE Stack frame of every thread Wednesday, May 15,
13

@amjithr >>> import sys # Get stack frames of every
thread >>> frames = sys._current_frames() # Dictionary of { thread_id: frame_obj } >>> frames {4420: <frame object at 0x7fd96c04a650>, 1407: <frame object at 0x7fd96c301730>} INQUIRE Wednesday, May 15, 13

@amjithr function3 function2 function1 FRAMES Wednesday, May 15, 13

@amjithr function3 function2 function1 Fn Name File Name Line No
Arguments Globals Locals ... FRAMES Wednesday, May 15, 13

@amjithr >>> import traceback # Extract useful info from frame
>>> traceback.extract_stack(frame) [('code.py', 88, 'runsource', 'return'), ('code.py', 110, 'runcode', 'print')] EXTRACT Wednesday, May 15, 13

@amjithr COLLATE Wednesday, May 15, 13

@amjithr Default Dictionary ATTEMPT Wednesday, May 15, 13

@amjithr Default Dictionary Uniquely identify a function fdata = (f_name,
file, line) Build a dictionary calls = defaultdict(int) Increment the value when found in frames calls[fn_a] += 1 ATTEMPT Wednesday, May 15, 13

@amjithr Results fn_a=3 fn_b=4 fn_c=5 fn_d=1 { } ATTEMPT Wednesday,
May 15, 13

@amjithr Simple to implement Identiﬁes hot functions No context ATTEMPT

@amjithr Call Tree ATTEMPT Wednesday, May 15, 13

@amjithr Example T1 Wednesday, May 15, 13

@amjithr Example fn_a fn_b fn_c T1 Wednesday, May 15, 13

@amjithr Example fn_a fn_b fn_c T1 fn_a=1 Wednesday, May 15,
13

@amjithr Example fn_a fn_b fn_c T1 fn_a=1 fn_b=1 Wednesday, May
15, 13

@amjithr Example fn_a fn_b fn_c T1 fn_a=1 fn_b=1 fn_c=1 Wednesday,
May 15, 13

@amjithr T2 fn_a=1 fn_b=1 fn_c=1 Example Wednesday, May 15, 13

@amjithr fn_a fn_b fn_d T2 fn_a=1 fn_b=1 fn_c=1 Example Wednesday,
May 15, 13

@amjithr fn_a fn_b fn_d T2 fn_a=1 fn_b=1 fn_c=1 fn_a=2 Example

@amjithr fn_a fn_b fn_d T2 fn_a=1 fn_b=1 fn_c=1 fn_a=2 fn_b=2
Example Wednesday, May 15, 13

@amjithr fn_a fn_b fn_d T2 fn_a=1 fn_b=1 fn_a=2 fn_b=2 fn_c=1

@amjithr fn_a fn_b fn_d T2 fn_a=1 fn_b=1 fn_a=2 fn_b=2 fn_c=1
fn_d=1 Example Wednesday, May 15, 13

@amjithr fn_a=1 fn_a=2 fn_b=2 fn_c=1 fn_d=1 T3 Example Wednesday, May
15, 13

@amjithr fn_a fn_d fn_a=1 fn_a=2 fn_b=2 fn_c=1 fn_d=1 T3 Example

@amjithr fn_a fn_d fn_a=1 fn_a=2 fn_b=2 fn_c=1 fn_d=1 fn_a=3 T3

@amjithr fn_a fn_d fn_a=1 fn_a=2 fn_a=3 fn_b=2 fn_c=1 fn_d=1 T3

@amjithr fn_a fn_d fn_a=1 fn_a=2 fn_a=3 fn_b=2 fn_c=1 fn_d=1 fn_d=1
T3 Example Wednesday, May 15, 13

@amjithr EXISTING TOOLS Wednesday, May 15, 13

@amjithr http://pypi.python.org/pypi/statprof StatProf Signals CLI Wednesday, May 15, 13

@amjithr https://github.com/bdarnell/plop PLOP Signals D Call Graph Wednesday, May 15,
13

@amjithr NewRelic Threads GUI Wednesday, May 15, 13

@amjithr SUMMARY Features Limitations Low Overhead No green threads CPython
& PyPy No Jython support Wednesday, May 15, 13

@amjithr WRONG USE OF MEME. WHO LET THIS GUY ON
THE STAGE? PROFILER in PRODUCTION Wednesday, May 15, 13

@amjithr FINALE DETERMINISTIC + STATISTICAL Wednesday, May 15, 13

@amjithr X-Ray Sessions Wednesday, May 15, 13

@amjithr DETERMINISTIC Wednesday, May 15, 13

@amjithr STATISTICAL Wednesday, May 15, 13

@amjithr HISTOGRAM Wednesday, May 15, 13

@amjithr RESPONSE & THROUGHPUT Wednesday, May 15, 13

@amjithr THANK YOU! @amjithr Wednesday, May 15, 13

Questions? • Promo: newrelic.com/djangocon • 30 days of Pro. •
Free T-shirt. • In-depth data. Wednesday, May 15, 13

Thread Profiling in Python

Thread Profiling in Python

Transcript

More Decks by Amjith

Other Decks in Programming

Featured