PDX Python lightning talk

Transcript

2. Kevin McConnell @kevinmcconnell
3. A tiny thread pool class
4. A tiny thread pool class or The impact on performance of a Service Oriented Architecture
5. A tiny thread pool class or The impact on performance of a Service Oriented Architecture or That time we had that really slow website and we made it much faster with only about 10 lines of code
6. A typical web application • Is written using a framework • Has some business logic • Talks to other systems • Databases • Caches • APIs
7. parsing business logic rendering time
8. parsing business logic rendering time Some of this time spent waiting on other systems! (hopefully not too much)
9. time
10. time (╯°□°）╯︵ ┻━┻
11. Our application • Connects to ~6 internal APIs • ...most of which are slow • Hardly any state of its own • Relatively simple business logic
12. parsing rendering time call API call API call API call API business logic business logic business logic business logic
13. parsing business logic rendering time call API call API call API call API
14. parsing business logic rendering time call API call API call API call API ~500ms (mean)
15. parsing business logic rendering time call API call API call API call API ~500ms (mean) (mostly just waiting! zzzzz.....)
16. parsing business logic rendering time call API call API call API call API ~ 125ms
17. from threading import Thread! ! ! class ThreadQueue:! def __init__(self):! self._tasks = {}! self._results = {}! ! def run(self, name, fn, *args, **kwargs):! thread = Thread(target=self._perform,! args=[name, fn, args, kwargs])! self._tasks[name] = thread! thread.start()! ! def get(self, name):! self._tasks[name].join()! return self._results[name]! ! def _perform(self, name, fn, args, kwargs):! self._results[name] = fn(*args, **kwargs)!
18. from threading import Thread! ! ! class ThreadQueue:! def __init__(self):! self._tasks = {}! self._results = {}! ! def run(self, name, fn, *args, **kwargs):! thread = Thread(target=self._perform,! args=[name, fn, args, kwargs])! self._tasks[name] = thread! thread.start()! ! def get(self, name):! self._tasks[name].join()! return self._results[name]! ! def _perform(self, name, fn, args, kwargs):! self._results[name] = fn(*args, **kwargs)!
19. from threading import Thread! ! ! class ThreadQueue:! def __init__(self):! self._tasks = {}! self._results = {}! ! def run(self, name, fn, *args, **kwargs):! thread = Thread(target=self._perform,! args=[name, fn, args, kwargs])! self._tasks[name] = thread! thread.start()! ! def get(self, name):! self._tasks[name].join()! return self._results[name]! ! def _perform(self, name, fn, args, kwargs):! self._results[name] = fn(*args, **kwargs)!
20. from threadqueue import ThreadQueue! ! tq = ThreadQueue()! ! tq.run('notifications', get_user_notifications)! tq.run('billing', get_billing_status)! tq.run('privileges', get_user_privileges)! ! # ...things happen...! # ...! ! if tq.get('privileges')['can_create_projects']:! create_project(...)!
21. from threadqueue import ThreadQueue! ! tq = ThreadQueue()! ! tq.run('notifications', get_user_notifications)! tq.run('billing', get_billing_status)! tq.run('privileges', get_user_privileges)! ! # ...things happen...! # ...! ! if tq.get('privileges')['can_create_projects']:! create_project(...)! returns! immediately
22. from threadqueue import ThreadQueue! ! tq = ThreadQueue()! ! tq.run('notifications', get_user_notifications)! tq.run('billing', get_billing_status)! tq.run('privileges', get_user_privileges)! ! # ...things happen...! # ...! ! if tq.get('privileges')['can_create_projects']:! create_project(...)! returns! immediately blocks if! results! pending
23. What we learned • Apps that use many external services can spend a lot of time blocked on I/O • One way to minimize the waiting time is to do all your waiting at once • There's great support for all sorts of nice concurrency models nowadays, but often a Thread is all you need
24. @kevinmcconnell Questions? Thanks!