[Windows] time.sleep() should use CREATE_WAITABLE_TIMER_HIGH_RESOLUTION

[vstinner]

BPO	45429
Nosy	@pfmoore, @vstinner, @tjguk, @ambv, @zware, @eryksun, @zooba, @corona10, @Livius90
PRs	bpo-45429: Support CREATE_WAITABLE_TIMER_HIGH_RESOLUTION if possible #29203 bpo-45429: Merge whatsnew about time.sleep #29589
Files	bpo-45429.py

^{Note: these values reflect the state of the issue at the time it was migrated and might not reflect the current state.}

Show more details

GitHub fields:

assignee = None
closed_at = <Date 2021-11-18.06:26:34.313>
created_at = <Date 2021-10-11.08:28:27.791>
labels = ['type-feature', 'library', 'OS-windows', '3.11']
title = '[Windows] time.sleep() should use CREATE_WAITABLE_TIMER_HIGH_RESOLUTION'
updated_at = <Date 2021-11-18.06:26:34.312>
user = 'https://github.com/vstinner'

bugs.python.org fields:

activity = <Date 2021-11-18.06:26:34.312>
actor = 'corona10'
assignee = 'none'
closed = True
closed_date = <Date 2021-11-18.06:26:34.313>
closer = 'corona10'
components = ['Library (Lib)', 'Windows']
creation = <Date 2021-10-11.08:28:27.791>
creator = 'vstinner'
dependencies = []
files = ['50392']
hgrepos = []
issue_num = 45429
keywords = ['patch']
message_count = 9.0
messages = ['403631', '403634', '403635', '403659', '403703', '404868', '404936', '406402', '406512']
nosy_count = 9.0
nosy_names = ['paul.moore', 'vstinner', 'tim.golden', 'lukasz.langa', 'zach.ware', 'eryksun', 'steve.dower', 'corona10', 'Livius']
pr_nums = ['29203', '29589']
priority = 'normal'
resolution = None
stage = 'resolved'
status = 'closed'
superseder = None
type = 'enhancement'
url = 'https://bugs.python.org/issue45429'
versions = ['Python 3.11']

[vstinner]

In bpo-21302, the Windows implementation of time.sleep() was modified to use a waitable timer:

New changeset 58f8adf by Victor Stinner in branch 'main':
bpo-21302: time.sleep() uses waitable timer on Windows (GH-28483)
58f8adf

It now calls the functions:

• CreateWaitableTimerW()
• SetWaitableTimer()
• WaitForMultipleObjects()

While SetWaitableTimer() has a resolution of 100 ns, the timer has a resolution of 15.6 ms in practice.

We could use the undocumented CREATE_WAITABLE_TIMER_HIGH_RESOLUTION flag with CreateWaitableTimerEx(). See: https://bugs.python.org/issue21302#msg403550

See also:

• https://randomascii.wordpress.com/2013/07/08/windows-timer-resolution-megawatts-wasted/
• https://vstinner.readthedocs.io/windows.html#time

[vstinner]

See also: https://groups.google.com/a/chromium.org/g/scheduler-dev/c/0GlSPYreJeY

[vstinner]

The Go programming language called timeBeginPeriod(1) to get more accurate timers. With the following change, it can now use a high resolution timer (CREATE_WAITABLE_TIMER_HIGH_RESOLUTION) to sleep:
https://go-review.googlesource.com/c/go/+/248699/

[vstinner]

See also bpo-19007: "precise time.time() under Windows 8: use GetSystemTimePreciseAsFileTime".

[eryksun]

It's up to the core devs whether or not Python should try to use a high-resolution timer, which is currently undocumented in the Windows API and implemented only in recent releases of Windows 10 and 11. But if this does get supported, the code should fall back on creating a normal timer if CREATE_WAITABLE_TIMER_HIGH_RESOLUTION makes the call fail. For example:

    #ifndef CREATE_WAITABLE_TIMER_HIGH_RESOLUTION
    #define CREATE_WAITABLE_TIMER_HIGH_RESOLUTION 0x00000002
    #endif

    LARGE_INTEGER relative_timeout;
    // No need to check for integer overflow, both types are signed
    assert(sizeof(relative_timeout) == sizeof(timeout_100ns));
    // SetWaitableTimerEx(): a negative due time is relative
    relative_timeout.QuadPart = -timeout_100ns;
    DWORD flags = CREATE_WAITABLE_TIMER_HIGH_RESOLUTION;

create_timer:

        HANDLE timer = CreateWaitableTimerExW(NULL, NULL, flags, TIMER_ALL_ACCESS);
        if (timer == NULL)
        {
            if (flags && GetLastError() == ERROR_INVALID_PARAMETER) {
                // CREATE_WAITABLE_TIMER_HIGH_RESOLUTION is not supported.
                flags = 0;
                goto create_timer;
            }
            PyErr_SetFromWindowsErr(0);
            return -1;
        }

        if (!SetWaitableTimerEx(timer, &relative_timeout,
              0,          // no period; the timer is signaled once
              NULL, NULL, // no completion routine
              NULL,       // no wake context; do not resume from suspend
              0))         // no tolerable delay for timer coalescing
        {
            PyErr_SetFromWindowsErr(0);
            goto error;
        }

[Livius90]

A similar solution was introduced in VirtualBox some months ago. Soon, i could get back my Windows 10 developing PC and i can try this things.

https://www.virtualbox.org/browser/vbox/trunk/src/VBox/Runtime/r3/win/timer-win.cpp#L312

[corona10]

AS-IS:
average: 0.015609736680984497

TO-BE:
average: 2.7387380599975585e-05

Impressive result :)

[corona10]

New changeset 55868f1 by Dong-hee Na in branch 'main':
bpo-45429: Support CREATE_WAITABLE_TIMER_HIGH_RESOLUTION if possible (GH-29203)
55868f1

[ambv]

New changeset fc4474e by Dong-hee Na in branch 'main':
bpo-45429: Merge whatsnew about time.sleep (GH-29589)
fc4474e