Christian,

 

26032 is up on http://boinc.berkeley.edu/dl/.

 

It includes:

VBOX: Add logging in case of a trickle-up failure.

VBOX: Adjust the VM process priority right before a suspend command to
speed up how quickly the VM is suspended.

 

----- Rom

 

From: Christian Beer [mailto:[email protected]] 
Sent: Thursday, November 07, 2013 6:45 AM
To: Rom Walton
Cc: BOINCDev Mailing List; David Anderson (BOINC); Lammert van der Veen
Subject: Re: ongoing problems with vboxwrapper

 

Hi Rom,

can you please move the new trickle status printf down after
boinc_send_trickle_up() and check the return value of this? This is more
helpful in case the trickle didn't get send and the reason is in the
log. I will deploy this new version on RNAWorld and create some long
running tasks then.

@David: Is there an easy way for me to assign this results to specific
users? I want to focus on users that I can contact in our forum to check
on the stderr.txt during runtime.

Regards
Christian

Am 07.11.2013 00:53, schrieb Rom Walton:

        I've posted 26031 to http://boinc.berkeley.edu/dl/.

         

        It contains the following changes:

        VBOX: Use the same technique for calculating when to report a
trickle as we use for performing checkpoints.

        VBOX: Add a trickle-up status report entry to stderr.txt every
time we send a trickle event.

        VBOX: Add VirtualBox 4.3.0 to bad builds list.

        VBOX: We only need to filter the vboxmanage output in one place.

        VBOX: Add additional check to determine if the get VM log
command really failed.

         

        ----- Rom

         

        From: Christian Beer [mailto:[email protected]] 
        Sent: Wednesday, November 06, 2013 11:29 AM
        To: Rom Walton
        Cc: BOINCDev Mailing List; David Anderson (BOINC); Lammert van
der Veen
        Subject: Re: ongoing problems with vboxwrapper

         

        Possible explanation. The current control script is buggy and
does not redirect the scientific app's stdout and stderr to files so it
ends up in the VM log. But this is happening on all tasks not just on
some.
        
        Regards
        Christian
        
        Am 06.11.2013 17:24, schrieb Rom Walton:

                Ah, so we are tripping up on the new code to check for
EXIT_OUT_OF_MEMORY. Fun fun fun.

                 

                Okay, I'll commit a change for this.

                 

                I'm not sure why vboxmanage would be returning a
non-zero exit status in this situation.

                 

                ----- Rom

                 

                From: Christian Beer [mailto:[email protected]] 
                Sent: Wednesday, November 06, 2013 11:11 AM
                To: Rom Walton
                Cc: BOINCDev Mailing List; David Anderson (BOINC);
Lammert van der Veen
                Subject: Re: ongoing problems with vboxwrapper

                 

                It seems that the general scheme seems to be this (see
the User's post:
https://www.rechenkraft.net/forum/viewtopic.php?f=76&t=13059&start=180#p
143176):
                
                
                
                

                2013-10-17 11:21:40 (816): Creating new snapshot for VM.
                2013-10-17 11:21:48 (816): Deleting stale snapshot.
                2013-10-17 11:21:49 (816): Checkpoint completed.
                2013-10-17 11:26:42 (816): Error in get vm log for VM: 3
                Arguments:
                showvminfo "boinc_fb76a72fc6655131" --log 0 
                Output:
                VirtualBox VM 4.2.16 r86992 win.amd64 (Jul  4 2013
15:51:44) release log
                00:00:00.042649 Log opened
2013-10-16T18:39:34.043413700Z

                followed by the actual VM Log (rather long) and this
over and over:
                
                
                
                

                05:44:49.794013 ********************* End of CFGM dump
**********************
                05:44:50.074438 Changing the VM state from 'SUSPENDED'
to 'RESUMING'.
                05:44:50.074495 Changing the VM state from 'RESUMING' to
'RUNNING'.
                05:44:55.318224 Changing the VM state from 'RUNNING' to
'SUSPENDING'.
                05:44:55.774983 PDMR3Suspend: 456 736 124 ns run time
                05:44:55.775003 Changing the VM state from 'SUSPENDING'
to 'SUSPENDED'.
                05:44:55.775847 DrvBlock: Flushes will be ignored
                05:44:55.775855 DrvBlock: Async flushes will be passed
to the disk
                05:44:55.776116 VD: Opening the disk took 236410 ns
                05:44:55.776131 PIIX3 ATA: LUN#0: disk, PCHS=4161/16/63,
total number of sectors 4194304
                05:44:55.776139 ************************* CFGM dump
*************************
                05:44:55.776140 [/Devices/piix3ide/0/] (level 0)
                05:44:55.776142   PCIBusNo      <integer> =
0x0000000000000000 (0)
                05:44:55.776145   PCIDeviceNo   <integer> =
0x0000000000000001 (1)
                05:44:55.776146   PCIFunctionNo <integer> =
0x0000000000000001 (1)
                05:44:55.776147   Trusted       <integer> =
0x0000000000000001 (1)
                05:44:55.776148 
                05:44:55.776149 [/Devices/piix3ide/0/Config/] (level 1)
(restricted root)
                05:44:55.776151   Type <string>  = "PIIX4" (cb=6)
                05:44:55.776152 
                05:44:55.776153
[/Devices/piix3ide/0/Config/PrimaryMaster/] (level 2)
                05:44:55.776155   NonRotationalMedium <integer> =
0x0000000000000000 (0)
                05:44:55.776156 
                05:44:55.776156 [/Devices/piix3ide/0/LUN#0/] (level 1)
                05:44:55.776158   Driver <string>  = "Block" (cb=6)
                05:44:55.776159 
                05:44:55.776159
[/Devices/piix3ide/0/LUN#0/AttachedDriver/] (level 2)
                05:44:55.776161   Driver <string>  = "VD" (cb=3)
                05:44:55.776162 
                05:44:55.776163
[/Devices/piix3ide/0/LUN#0/AttachedDriver/Config/] (level 3) (restricted
root)
                05:44:55.776165   Format     <string>  = "VDI" (cb=4)
                05:44:55.776166   Path       <string>  =
"D:\ProgramData\BOINC\slots\9\boinc_fb76a72fc6655131\Snapshots\{650bac36
-f84e-43c7-b30c-c8a078244a51}.vdi" (cb=105)
                05:44:55.776167   SetupMerge <integer> =
0x0000000000000001 (1)
                05:44:55.776168   Type       <string>  = "HardDisk"
(cb=9)
                05:44:55.776169 
                05:44:55.776170
[/Devices/piix3ide/0/LUN#0/AttachedDriver/Config/Parent/] (level 4)
                05:44:55.776172   Format      <string>  = "VDI" (cb=4)
                05:44:55.776173   MergeSource <integer> =
0x0000000000000001 (1)
                05:44:55.776174   Path        <string>  =
"D:\ProgramData\BOINC\slots\9\boinc_fb76a72fc6655131\Snapshots\{ed98fbf7
-ab54-4f4c-97d9-ec954d59d419}.vdi" (cb=105)
                05:44:55.776176 
                05:44:55.776176
[/Devices/piix3ide/0/LUN#0/AttachedDriver/Config/Parent/Parent/] (level
5)
                05:44:55.776178   Format      <string>  = "VDI" (cb=4)
                05:44:55.776179   MergeTarget <integer> =
0x0000000000000001 (1)
                05:44:55.776180   Path        <string>  =
"D:\ProgramData\BOINC\slots\9\vm_image.vdi" (cb=42)
                05:44:55.776182 
                05:44:55.776182 [/Devices/piix3ide/0/LUN#0/Config/]
(level 2) (restricted root)
                05:44:55.776184   Mountable <integer> =
0x0000000000000000 (0)
                05:44:55.776185   Type      <string>  = "HardDisk"
(cb=9)
                05:44:55.776186 
                05:44:55.776187 [/Devices/piix3ide/0/LUN#999/] (level 1)
                05:44:55.776188   Driver <string>  = "MainStatus"
(cb=11)
                05:44:55.776189 
                05:44:55.776190 [/Devices/piix3ide/0/LUN#999/Config/]
(level 2) (restricted root)
                05:44:55.776192   DeviceInstance        <string>  =
"piix3ide/0" (cb=11)
                05:44:55.776193   First                 <integer> =
0x0000000000000000 (0)
                05:44:55.776194   Last                  <integer> =
0x0000000000000003 (3)
                05:44:55.776196   pConsole              <integer> =
0x0000000001cc8280 (30 179 968)
                05:44:55.776198   papLeds               <integer> =
0x0000000001cc8598 (30 180 760)
                05:44:55.776200   pmapMediumAttachments <integer> =
0x0000000001cc88a0 (30 181 536)
                05:44:55.776201 
                05:44:55.776202 ********************* End of CFGM dump
**********************
                05:44:55.776213 Changing the VM state from 'SUSPENDED'
to 'RESUMING'.

                Another user reported that after a host restart the
growth of stderr.txt was normal again.
                
                Regards
                Christian
                
                Am 06.11.2013 16:59, schrieb Rom Walton:

                        Lammert, what did you discover?

                         

                        Christian, do you happen to know what kind of
messages stderr.txt was filled with?

                         

                        Vboxwrapper uses wall clock time internally.
I'll see what I can find about the trickle messages.

                         

                        ----- Rom

                         

                        From: Christian Beer [mailto:[email protected]]

                        Sent: Wednesday, November 06, 2013 10:30 AM
                        To: BOINCDev Mailing List
                        Cc: Rom Walton; David Anderson (BOINC); Lammert
van der Veen
                        Subject: ongoing problems with vboxwrapper

                         

                        Hello,
                        
                        we are running the 26028 version of the
vboxwrapper for some time now and I want to update you on some ongoing
problems.
                        
                        Some users reported that the stderr.txt is
filled with lots of error messages and file size increases to several
GB. The file was truncated by the user and I didn't see any unusual
disk_size_limit_reached errors. So either this was an isolated incident
or the file size doesn't matter.
                        
                        Many users reported that the VM is still running
after the BOINC Client was shut down. Lammert van der Veen did some
research to the cause and I hope this can be fixed by limiting one
concurrent VM per Host and the 26031 wrapper as soon as I upgrade our
application.
                        
                        Trickle messages were working fine when running
with short tasks. Now that we have some longer tasks the trickle up
messages stopped. We didn't receive any in over a month. I think I have
an explanation for this:
                        In the vboxwrapper the trickles are generated
every X seconds cpu_time and not wall_clock_time and as the vboxwrapper
is not doing much the cpu_time increases very slowly. What I want is a
trickle message every X hours of VM runtime! Please look into this asap
because without this feature I have to monitor the deadlines and extend
them by hand.
                        If it may be helpful:
                        A recent long running task reported back with
cpu_time=793517.4 and elapsed_time=839064.946827 but I also have a task
with cpu_time=4280.246 and elapsed_time=378984.324896 I can't see any
trickle messages for both of them.
                        first:
http://www.rnaworld.de/rnaworld/result.php?resultid=14920843 (Job
Duration is always 0, Elapsed time is increasing)
                        second:
http://www.rnaworld.de/rnaworld/result.php?resultid=14921349 (can't see
anything in stderr)
                        
                        Speaking of deadlines, it would also be great to
update the deadline on the client. I know of one user who updates his
client_state.xml by hand to prevent the Client from going into high
priority mode for RNAWorld when there is no need.
                        
                        Regards
                        Christian

                 

         

 

_______________________________________________
boinc_dev mailing list
[email protected]
http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev
To unsubscribe, visit the above URL and
(near bottom of page) enter your email address.

Reply via email to