Testing Version 3.31

fred · March 10, 2011, 08:11:54 PM

Fixed: Throttle didn't work on CPU's with one core.

Pepo · March 10, 2011, 09:53:01 PM

Quote from: Pepo on March 09, 2011, 11:45:54 PM
The main window - Programs tab, clicked on the Program list button, Details checkbox, again the button and TTh locked up and crashed - minidump uploaded to SK download section.
Reproducibly - Details + Program list.

On 3.31, clicking on Program list button with Details checkbox selected also reliably locks TTh.

fred · March 10, 2011, 09:55:20 PM

Quote from: Pepo on March 10, 2011, 09:53:01 PM
Quote from: Pepo on March 09, 2011, 11:45:54 PM
The main window - Programs tab, clicked on the Program list button, Details checkbox, again the button and TTh locked up and crashed - minidump uploaded to SK download section.
Reproducibly - Details + Program list.
On 3.31, clicking on Program list button with Details checkbox selected also reliably locks TTh.

I can't reproduce it.
And by updating to 3.31 I can't debug the older version you uploaded.

Pepo · March 10, 2011, 10:29:35 PM

Quote from: fred on March 10, 2011, 09:55:20 PM
Quote from: Pepo on March 10, 2011, 09:53:01 PM
Quote from: Pepo on March 09, 2011, 11:45:54 PM
The main window - Programs tab, clicked on the Program list button, Details checkbox, again the button and TTh locked up and crashed - minidump uploaded to SK download section.
Reproducibly - Details + Program list.
On 3.31, clicking on Program list button with Details checkbox selected also reliably locks TTh.
I can't reproduce it.
And by updating to 3.31 I can't debug the older version you uploaded.

The previous minidump was created by Win Error Reporting, but not anymore.
I've created a new minidump with Process Explorer and uploaded to the same location - please let me know if it will be sufficient.
If not ...

fred · March 10, 2011, 10:31:05 PM

Quote from: Pepo on March 10, 2011, 10:29:35 PM
I've created a new minidump with Process Explorer and uploaded to the same location - please let me know if it will be sufficient.
If not ...

But was there a crash?

Pepo · March 10, 2011, 10:40:09 PM

Quote from: fred on March 10, 2011, 10:31:05 PM
Quote from: Pepo on March 10, 2011, 10:29:35 PM
I've created a new minidump with Process Explorer and uploaded to the same location - please let me know if it will be sufficient.
If not ...
But was there a crash?

No, not anymore, just each time a lockup. The interface seem to stay, just is not functional but frozen, all threads are in Wait state but the one which is errored somewhere deep...

fred · March 10, 2011, 11:37:50 PM

Quote from: Pepo on March 10, 2011, 10:40:09 PM
No, not anymore, just each time a lockup. The interface seem to stay, just is not functional but frozen, all threads are in Wait state but the one which is errored somewhere deep...

OK, its on the Bug list.

Pepo · March 11, 2011, 06:28:35 PM

I've noticed

Code Select

CPU:1, GPU:0, PID:7332 (4) Child: WerFault.exe

A non-existent process. (I've still three other WerFaults from TTh in-memory

) Is it related to this bug?

QuoteBug: Once in a while a parent PID is invalid and a couple of programs are added as child PID's. (BOINC client mode only)

Does it mean that those processes' PPID get (temporarily?) invalid and they seem to belong to BOINC process tree?

(Where is the nice emoticon with question marks gone??)

It appeared tonight in the log:

Code Select


----------------------------------------------------------------------------------------------------------------------------------------------
11 March 2011 - 00:58:16  Počet zhodných programov (procesov): 3
CPU:1, GPU:0, PID:17108 (3)	Slot:10	http://milkyway.cs.rpi.edu/milkyway/	de_nbody_model6_4_238995_1299754890
CPU:1, GPU:0, PID:15996 (3)	Slot:7	http://registro.ibercivis.es/	Mr.Wilson_02_04_52_18_731868690_1797040079
CPU:1, GPU:0, PID:8176 (4)	Slot:13	http://spin.fh-bielefeld.de/	24_Fe30_map_642_638
----------------------------------------------------------------------------------------------------------------------------------------------
11 March 2011 - 01:00:19  Počet zhodných programov (procesov): 4
CPU:1, GPU:0, PID:17108 (3)	Slot:10	http://milkyway.cs.rpi.edu/milkyway/	de_nbody_model6_4_238995_1299754890
CPU:1, GPU:0, PID:15996 (3)	Slot:7	http://registro.ibercivis.es/	Mr.Wilson_02_04_52_18_731868690_1797040079
CPU:1, GPU:0, PID:18372 (3)	Slot:15	http://www.primegrid.com/	pps_sr2sieve_13617730
CPU:1, GPU:0, PID:7332 (4)	Child:	WerFault.exe
----------------------------------------------------------------------------------------------------------------------------------------------

and is not willing to disappear, even after switching back and forth between BOINC client connected mode and manual mode.

As a BOINC subprocess events, a PrimeGrid GPU task exited twice and has been removed from memory, around this time period (0:57:48 + 0:59:27). If it is related at all to BOINC process tree.

(I could try to run Process Monitor long-term and try to monitor and capture a list of existing processes for the moment it happens again?)

fred · March 11, 2011, 10:26:00 PM

Quote from: Pepo on March 11, 2011, 06:28:35 PM
I've noticed
Code Select Expand
CPU:1, GPU:0, PID:7332 (4) Child: WerFault.exe

A non-existent process. (I've still three other WerFaults from TTh in-memory ) Is it related to this bug?
QuoteBug: Once in a while a parent PID is invalid and a couple of programs are added as child PID's. (BOINC client mode only)
Does it mean that those processes' PPID get (temporarily?) invalid and they seem to belong to BOINC process tree? (Where is the nice emoticon with question marks gone??)

Window has a rather clumsy way of dealing with Processes and Threads.
Both are separate lists.

Seems the process list is sometimes corrupted and marks an other process as child. But in reality it's not the case. The process explorer can detect this as invalid.

So I have to find out a way to detect an invalid PID. The problem is, that it happens very infrequently.

Pepo · March 11, 2011, 10:47:58 PM

Quote from: fred on March 11, 2011, 10:26:00 PM
Window has a rather clumsy way of dealing with Processes and Threads.
Both are separate lists.
Seems the process list is sometimes corrupted and marks an other process as child. But in reality it's not the case. The process explorer can detect this as invalid.

So I have to find out a way to detect an invalid PID. The problem is, that it happens very infrequently.

I guess you cant. Except that if you would regularly get the list (what TTh does anyway), store and maintain the relationships (like Process Explorer does) (maybe long-term? or not necessary?) and then upon comparing a new list, you could be able to eventually notice - "Ah this process has previously had someone other marked as its parent!"

Otherwise (unless you would not notice any loops) I suspect the chances are darn low.

Pepo · March 20, 2011, 10:47:49 AM

Quote from: fred on March 11, 2011, 10:26:00 PM
So I have to find out a way to detect an invalid PID. The problem is, that it happens very infrequently.

I saw a note it has been fixed.

How?

fred · March 20, 2011, 11:16:39 PM

Quote from: Pepo on March 20, 2011, 10:47:49 AM
I saw a note it has been fixed.

How?

A bit of a work around, I get the full path of the child exe.
Then I check if there is BOINC somewhere in the path name.
So this only works properly when you have a path with BOINC in it. Like C:\ProgramData\BOINC\projects
Hope this works in XP as I only have an WIN 7 machine for testing right now.

News:

Testing Version 3.31