Development for LHC@home

Hi, thanks for waiting. Unfortunately I got side-tracked for a while due to CERN's changing the rules since the last time I renewed my contract...
I've changed the vboxwrapper to V.26158, so you can resume tasks if you want. It seems to run fine on Windows and Linux. Please run a test WU, esp. if you're on a Mac. (Maybe I can leverage this project to get a Mac myself!)
I've added the new feature to the vbox_job.xml file which will attempt to use 'savestate' instead of 'poweroff' when gracefully shutting down.
I haven't added another new feature , which will prevent snapshots from being created by vboxwrapper, because it wasn't immediately obvious to me how the volunteer overrides this, or sets his own checkpoint schedule -- no doubt this will be clarified to me in the very near future. I'm wary of just disabling checkpointing at this stage.

---

Hi Ivan (and Rom),

Just tested the new feature on my Mac but it didn't work. The boot of CMS code worked fine. After a BOINC suspend, the VM went into Paused state (and could be resumed OK), but after an exit of the BOINC manager the VM was left in PowerOff state, (not "Saved" state) and hence rebooted when BOINC was restarted.

Ben

Hi Ivan (and Rom),

Just tested the new feature on my Mac but it didn't work. The boot of CMS code worked fine. After a BOINC suspend, the VM went into Paused state (and could be resumed OK), but after an exit of the BOINC manager the VM was left in PowerOff state, (not "Saved" state) and hence rebooted when BOINC was restarted.

Ben

Can you abort the task so I can look at the stderr text?

----- Rom

Hi Ivan (and Rom),

Just tested the new feature on my Mac but it didn't work. The boot of CMS code worked fine. After a BOINC suspend, the VM went into Paused state (and could be resumed OK), but after an exit of the BOINC manager the VM was left in PowerOff state, (not "Saved" state) and hence rebooted when BOINC was restarted.

Ben

Can you abort the task so I can look at the stderr text?

----- Rom

OK, done!

......
I've added the new feature <enable_vm_savestate_usage/> to the vbox_job.xml file which will attempt to use 'savestate' instead of 'poweroff' when gracefully shutting down.
....

This new tag is not in the new CMS_26_03_2015.xml I got with a new task.

Pse have a closer look, ivan.

I said I'd screw up...
I'd put it in the wrong file at first, edited it out, and then forgot to put it in the right file!
Watch this space.

---

New version now available. Added both flags (at Ben's request):

enable_vm_savestate_usage/
disable_automatic_checkpoints/

As I understand it, this means you won't get the periodic checkpointing of the VM.

[Edit] Had to remove the angle brackets as otherwise the tags disappeared in my browser![/Edit]

---

..
As I understand it, this means you won't get the periodic checkpointing of the VM.
..

Rom choose for the name disable_automatic_checkpoints.
It's a bit confusion, because checkpoints are still made, but during a checkpoint no VM-snapshot is made.

In fact the name disable_automatic_snapshots would have been better.

New version now available. Added both flags (at Ben's request):

enable_vm_savestate_usage/
disable_automatic_checkpoints/

As I understand it, this means you won't get the periodic checkpointing of the VM.

[Edit] Had to remove the angle brackets as otherwise the tags disappeared in my browser![/Edit]

Bravo Ivan, bravo Rom !!

It works well on Mac. I will leave it running and bash it some more….

Ben

Even more good news: a more vicious test also succeeded. I exited the BOINC manager while the task was in Running state (as was the VM of course). The VM was saved correctly and upon relaunching BOINC both the task and the VM returned to Running state.

I like that!

So far no snapshots have appeared, by the way, after 20 minutes or so of running. Good riddance...

I'm curious what happens during random power failures, reboots, and vboxsvc going nuts.

----- Rom

I'm curious what happens during random power failures, reboots, and vboxsvc going nuts.

----- Rom

For your curiosity with running BOINC and running VM's:

After soft reset and sudden power off: VM's are in PowerOff mode and after starting BOINC the VM's are booted.

Killing VBoxSVC: VBoxHeadless.exe's are starving after a while and the VM's are in PowerOff mode.
The wrappers keeps on running, but after suspend and resume the VM's are booted.
But wrappers pausing and resuming longer leads to computation errors.

Win7 VBox 4.3.26-98988.
VM pauses and restarts OK when tasks switch and when
manually suspended.
BUT rebooting the host after ca 13 minutes running
(VM appeared to be running OK) caused VM to reboot.
Is this OK?. BOINC checkpoint interval is set to 10
mins in mgr.. Perhaps I didn't wait long enough.
This throws up a couple of questions:-
1. Is checkpointing now done? If so can someone tell me how, please.
Edit:- I, for one am going to need checkpointing to get
much useful work done.
2. Where can I find release notes for the wrappers?

Edit:-Rebooted host again after ca45 mins. VM rebooted but elapsed time
shown by mgr didn't reset.

BUT rebooting the host after ca 13 minutes running
(VM appeared to be running OK) caused VM to reboot.
Is this OK?

That's expected when the machine is rebooted without stopping BOINC first.

1. Is checkpointing now done? If so can someone tell me how, please.

Checkpointing is done for the task progress. Every 10 minutes. The first time the wrapper is started and after a wrapper restart it could be up to 20 minutes.
The task state is saved every time a task is suspended ("Leave applications in memory" ticked off) and when BOINC is stopped.

Thanks, CP.

That's expected when the machine is rebooted without stopping BOINC first.

1. Is checkpointing now done? If so can someone tell me how, please.

Checkpointing is done for the task progress. Every 10 minutes. The first time the wrapper is started and after a wrapper restart it could be up to 20 minutes.
The task state is saved every time a task is suspended ("Leave applications in memory" ticked off) and when BOINC is stopped.

Does this mean that, in order to save the VM job in progress (as opposed to the BOINC task), BOINC must be stopped before the host is shut down? My hosts run overnight and are normally shut down each day by a Win scheduled task (or Linux cron job) so I could stop BOINC first, but a normal user wouldn't do this.

Edit. How do I find out what changes/bugfixes are in each wrapper release?

Does this mean that, in order to save the VM job in progress (as opposed to the BOINC task), BOINC must be stopped before the host is shut down? My hosts run overnight and are normally shut down each day by a Win scheduled task (or Linux cron job) so I could stop BOINC first, but a normal user wouldn't do this.

1. A BOINC cruncher is not a normal user. ;-)
2. A BOINC cruncher using VBox is not a normal user at all. ;-)
3. It's always good to stop all user processes/save your work before shutdown.

You could add stopping BOINC to your scheduled task with the command "path"\boinccmd --quit