R1D2 Profiling Python
I spent some time fiddling with Lambda. The code now more or less works.
Moving on to optimization, I started reading about profiling in python.
I’ve been using time
to see how long the script takes to run. Using cProfile
allows you to see how long each specific method call in your application takes to run. Running this on old_posts.py produced the following output.
The following command executes the script under profiling, sorts the results by time and saves the output to a text file.
python -m cProfile -s time old_posts.py > profile.txt
289508 function calls (286008 primitive calls) in 6.616 seconds
Ordered by: internal time
ncalls tottime percall cumtime percall filename:lineno(function)
284 5.495 0.019 5.495 0.019 {method ‘read’ of ‘_ssl._SSLSocket’ objects}
24 0.514 0.021 0.514 0.021 {method ‘do_handshake’ of ‘_ssl._SSLSocket’ objects}
24 0.157 0.007 0.157 0.007 {method ‘connect’ of ‘_socket.socket’ objects}
24 0.123 0.005 0.123 0.005 {method ’load_verify_locations’ of ‘_ssl._SSLContext’ objects}
24 0.027 0.001 0.030 0.001 connectionpool.py:410(close)
24 0.015 0.001 0.015 0.001 {built-in method _socket.gethostbyname}
173 0.015 0.000 0.015 0.000 {built-in method marshal.loads}
22 0.013 0.001 0.013 0.001 decoder.py:345(raw_decode)
24 0.012 0.001 0.013 0.001 {built-in method _socket.getaddrinfo}
20 0.012 0.001 0.012 0.001 {built-in method _imp.create_dynamic}
486/483 0.009 0.000 0.017 0.000 {built-in method builtins.build_class}
1019 0.007 0.000 0.007 0.000 {built-in method posix.stat}
24 0.007 0.000 0.007 0.000 {built-in method _scproxy._get_proxy_settings}
189/1 0.005 0.000 6.616 6.616 {built-in method builtins.exec}
2472 0.005 0.000 0.008 0.000 parser.py:68(get_token)
1 0.004 0.004 0.004 0.004 {method ‘read’ of ‘_io.TextIOWrapper’ objects}
266/80 0.003 0.000 0.010 0.000 sre_parse.py:470(_parse)
206 0.003 0.000 0.023 0.000 parser.py:622(_parse)
460 0.003 0.000 0.015 0.000 <frozen importlib._bootstrap_external>:1233(find_spec)
14107 0.003 0.000 0.005 0.000 {built-in method builtins.isinstance}
601 0.003 0.000 0.003 0.000 {built-in method new of type object at 0x107136800}
173 0.003 0.000 0.004 0.000 <frozen importlib._bootstrap_external>:830(get_data)
24 0.003 0.000 0.003 0.000 {built-in method _scproxy._get_proxies}
24068/23811 0.002 0.000 0.004 0.000 {built-in method builtins.len}
3615 0.002 0.000 0.007 0.000 os.py:664(getitem)
3936 0.002 0.000 0.011 0.000 _collections_abc.py:742(iter)
495/71 0.002 0.000 0.010 0.000 sre_compile.py:64(_compile)
1962 0.002 0.000 0.005 0.000 {built-in method builtins.min}
3830 0.002 0.000 0.003 0.000 enum.py:515(new)
48 0.002 0.000 0.002 0.000 {method ‘set_ciphers’ of ‘_ssl._SSLContext’ objects}
148 0.002 0.000 0.002 0.000 {built-in method posix.getcwd}
10300 0.002 0.000 0.003 0.000 parser.py:320(<genexpr>)
3843 0.002 0.000 0.008 0.000 enum.py:265(call)
6335 0.002 0.000 0.002 0.000 {built-in method builtins.getattr}
3642/3195 0.002 0.000 0.021 0.000 {built-in method builtins.hasattr}
1563 0.002 0.000 0.005 0.000 enum.py:801(and)
225 0.002 0.000 0.002 0.000 sre_compile.py:250(_optimize_charset)
4681 0.002 0.000 0.002 0.000 sre_parse.py:232(__next)
6892 0.002 0.000 0.002 0.000 {method ‘decode’ of ‘bytes’ objects}
173 0.002 0.000 0.002 0.000 {method ‘read’ of ‘_io.FileIO’ objects}
By default the wordpress API returns 10 posts per page. Updating this value to 100 decreased the time spent making network calls and overall function calls by almost 50%!
Results after per_page=100
145834 function calls (142393 primitive calls) in 2.679 seconds
Ordered by: internal time
ncalls tottime percall cumtime percall filename:lineno(function)
186 2.364 0.013 2.364 0.013 {method ‘read’ of ‘_ssl._SSLSocket’ objects}
5 0.073 0.015 0.073 0.015 {method ‘do_handshake’ of ‘_ssl._SSLSocket’ objects}
5 0.043 0.009 0.043 0.009 {method ‘connect’ of ‘_socket.socket’ objects}
5 0.026 0.005 0.026 0.005 {method ’load_verify_locations’ of ‘_ssl._SSLContext’ objects}
Lessons
Profiling is great. Instead of guessing why my code is slow, I can ask python to tell me explicitly.Most importantly, for the real world. AWS Lambda bills by the nearest 100 milliseconds.
Duration is calculated from the time your code begins executing until it returns or otherwise terminates, rounded up to the nearest 100ms. The price depends on the amount of memory you allocate to your function.So in a model like this, reducing the total time for your code to run from 6 seconds to 3 seconds will also reduce your bill by 50%.
Thank you for reading! Share your thoughts with me on bluesky, mastodon, or via email.
Check out some more stuff to read down below.
Most popular posts this month
- 2024
- Reinstalling Windows at 1am
- SQLite DB Migrations with PRAGMA user_version
- My Custom Miniflux CSS Theme
- How to Disable Wayland in Debian Testing
Recent Favorite Blog Posts
This is a collection of the last 8 posts that I bookmarked.
- The Software Essays that Shaped Me from Refactoring English
- Give Your Spouse the Gift of a Couple's Email Domain from mtlynch.io
- Skip the Next iPhone from Articles on Jose M.
- Have smart glasses finally hit an inflection point? from The Torment Nexus
- The McPhee method from the jsomers.net blog
- Pluralistic: LLMs are slot-machines (16 Aug 2025) from Pluralistic: Daily links from Cory Doctorow
- Pluralistic: Bluesky creates the world's weirdest, hardest-to-understand binding arbitration clause (15 Aug 2025) from Pluralistic: Daily links from Cory Doctorow
- Just a Little More Context Bro, I Promise, and It’ll Fix Everything from Jim Nielsen’s Blog
Articles from blogs I follow around the net
On concrete examples
I had some great conversations via email over the past couple of weeks with a bunch of different people, discussing all sorts of things that I’ll for sure end up writing about. Today I wanted to briefly touch on the topic of examples, which was pa…
via Manuel Moreale — Everything Feed October 16, 2025Hacking Workshop for November 2025
For next month, I'm scheduling 2 or 3 discussions of Matthias van de Meent's talk, Improving scalability; Reducing overhead in shared memory, given at 2025.pgconf.dev (talk description here). If you're interested in joining us, please sign up …
via Robert Haas October 16, 2025Should we be afraid of AI? Maybe a little
Almost exactly a year ago, I wrote a piece for The Torment Nexus about the threat of AI, and more specifically what some call "artificial general intelligence" or AGI, which is a shorthand term for something that approaches human-like intelligence…
via The Torment Nexus October 16, 2025Generated by openring