Windows shut down order. | Tech Off

Massif
aim stupidly high, expect to fail often.

This may have been approached in a video somewhere, but I thought I ask actual people instead of trawling through hours of footage to find a possible answer.

When I log off / shut down - how does windows decide which order to shut down applications in? I originally assumed it just sent the same shutdown message to all applications. But is it smarter than that? Does it send large applications shutdown messages early? Does it only try to shut down N processes at a time. (Where N is the number of processors the machine has.)

Just idle curiosity really.

AndyC

Windows just starts sending every app a WM_QUERYENDSESSION message then, assuming nothing attempts to block shutdown, sends everything an WM_ENDSESSION message. Apps have to respond in a timely fashion to these messages or Windows will just kill them The end result of this is that there is no defined order of apps shutting down and you should never ever try to rely on one.

Massif
aim stupidly high, expect to fail often.

In reply to AndyC

#Aug 21st @ 1:49 AM

I wasn't even thinking of relying on one, I was just wondering if there were any measures to mitigate slow to shut down applications by starting them shutting down first. (Possibly leaving resources free for that app to shut down before kicking everything else off).

I imagine it's almost impossible to calculate which apps will be slow to shut down, so I don't suppose it's worth it.

Massif
aim stupidly high, expect to fail often.

In reply to AndyC

#Aug 21st @ 1:49 AM

Gah! Double post...

W3bbo
The Master of Baiters

In reply to AndyC

#Aug 21st @ 1:49 AM

How is it that when logging off or shutting down Windows can really kill a process, yet no amount of taskkill /f will stop some unrulely processes?

blowdart
Peek-a-boo

In reply to W3bbo

#Aug 21st @ 3:54 AM

Because Windows runs at a higher privilege level than any user ever can.

Sven Groot
My name has 9 letters. Coincidence? I think not...

In reply to W3bbo

#Aug 21st @ 3:54 AM

One of the "nice" things about Linux is that a process can get in a state where it truly can't be killed, even by the system. In this situation, you cannot even shut down because even that will get stuck trying to kill these processes. The only way out is a hard reset.

Dodo
I'm your creativity creator™ :)

In reply to Sven Groot

#Aug 21st @ 5:07 AM

That 'hard reset', can be triggered via software with sufficient privilegues at any given time.

Sven Groot
My name has 9 letters. Coincidence? I think not...

In reply to Dodo

#Aug 21st @ 5:30 AM

Yes, I know, the SysRq key sequences that you can also trigger using some file in /proc somewhere. It's still rather rediculous that a user-mode process can get the system in such a state that it cannot be cleanly shut down.

Larry Osterman

In reply to Sven Groot

#Aug 21st @ 7:00 AM

Windows has similar "you can't kill this process" issues. Typically they're related to buggy drivers that can't cancel I/O's properly (before a thread can be terminated all the I/O that's outstanding on that thread has to be canceled and if a driver hangs on cancelation (or fails to handle cancelation properly) the thread can't be stopped (or the machine would bluescreen when the I/O finally did complete).

There are WHQL tests to verify that 3rd party WHQL signed drivers properly handle cancelation but there are buggy non WHQL signed drivers that can cause this to happen.

blowdart
Peek-a-boo

In reply to Larry Osterman

#Aug 21st @ 6:56 PM

What do the WHQL tests do? It's nice to find out something, but having dealt with Nvidia's drivers which are WHQL certified, and yet as buggy as sin I've often wondered what is actually tested.

AndyC

In reply to blowdart

#Aug 22nd @ 1:36 AM

The problem with WHQL tests is that it's very easy to submit a driver and have it pass all the quality tests, then package that signed driver in an installer that adds a bunch of registry keys that cause the driver to switch into an "optimized" mode that would never have passed the tests.

blowdart
Peek-a-boo

In reply to AndyC

#Aug 22nd @ 1:58 AM

Which raises the question "What's the point?"

ManipUni
Proving QQ for 5 years!

In reply to blowdart

#Aug 22nd @ 9:54 AM

Better quality drivers.

AndyC

In reply to blowdart

#Aug 22nd @ 9:54 AM

Exactly, it doesn't really amount to much, but there's no real way around that. If manufacturers are prepared to game the system, they'll always find away to get around the quality testing. The hope, I guess, is that the mere existence of the tests is enough to encourage some developers to raise their standards.

Sven Groot
My name has 9 letters. Coincidence? I think not...

In reply to Larry Osterman

#Aug 21st @ 6:56 PM

But when that happens, does it also become impossible to reboot or shut down the system? I've had unkillable processes on Windows, but I've never had them so unkillable that the system couldn't be shut down anymore. That's what happened to me on Linux.