Jeremy Kloth added the comment: On Mon, Jun 19, 2017 at 6:24 AM, STINNER Victor <rep...@bugs.python.org> wrote: > > STINNER Victor added the comment: > >> Pushed utility to github: https://github.com/jkloth/loadavg > > Hum, I see two implementations: > > * LOADAVG_USE_REG: use the registry, RegQueryValueExW(HKEY_PERFORMANCE_DATA, > L"2", ...) > * LOADAVG_USE_PDH: PdhOpenQueryW() > > Are these API accessible by everyone? On Linux, the data is public to > everybody:
Yes. > Why do you prefer registry over PDH? A PDH query adds an additional 4MB to the running process. > Should CalculateLoad be called every 5 seconds in a thread? If yes, IMHO it's > a blocker issue for regrtest since some tests fail if other threads are > running. Yes, as that is exactly what Linux does (not in a thread, but as part of the scheduler). This is the reason for using the Thread Pool. They are kernel provided worker threads for the running process. As a matter of fact, Windows 10 now uses them to speed up DLL loading on startup so every Python process is running with 2+ threads. However, regtest when run with -jX already starts an add'l thread so that can no longer be an issue. > I'm not sure that such code belongs to Python stdlib. What do you think of > putting it on PyPI? Sure, but packaging it up as a proper extension is quite low on my todo list. I am keen, however, on getting the buildbot fleet (well, mine in particular) in good shape. > For Python 2, sometimes I would like to use faulthandler on our buildbots > since it's really hard to guess what crashed Python when a random test crash. > But faulthandler was only added to Python 3.3. So using it on Python 2 would > require to install it using pip. Right now, Python only requires itself to > run tests. I'm not sure that it's a good idea to have dependencies, even if > there are externals. Agreed, which is why I offered to implement as a patch to the test suite. > Again, I'm not sure that this specific issue is the best place to discuss > implementing "os.getloadavg()", since your implementation is more complex > than a single function call, it seems like it requires to run a thread "in > background", so maybe an object to start/stop the thread, etc. Don't get too caught up with `os.getloadavg()`, it was just a possibility. I was leaning more towards a private utility function just for regrtest, but realized it may be useful outside of just testing. However, as you note, it does require a thread, which is why it is best implemented using the Thread Pool API to minimize any impact on Python proper. ---------- nosy: +jeremy.kloth _______________________________________ Python tracker <rep...@bugs.python.org> <http://bugs.python.org/issue30263> _______________________________________ _______________________________________________ Python-bugs-list mailing list Unsubscribe: https://mail.python.org/mailman/options/python-bugs-list/archive%40mail-archive.com
