Re: [Dmtcp-forum] checkpoint only certain processes

[Kapil A. <ka...@cc...>]

Hristo,

I forgot to mention this earlier, there is a simpler and modular way to
achieve what we are trying to here. We have started support for third party
modules for DMTCP. What that means for you is that you can now write your
own small module (should be very few lines of code) which will disconnect
the raw socket for process B before checkpoint and will create a dead socket
at restart. To get you started, I can provide you with a skeleton for that
module and you can then fill in the details. How does this sound? This would
be easier for us and hopefully easier for you too. We are willing to provide
full support to you in order to get your module working :).

Let me know if this idea of writing your own module sounds reasonable :).

-- Kapil

On Thu, Oct 20, 2011 at 4:46 PM, Hristo Asenov <nj...@ya...> wrote:

> yeah that would be really cool to try, could you point me to the code i
> need to modify in order to skip saving of the raw socket? I would also need
> to skip saving a UDP socket but I think that should be quite similar. Thanks
> alot,
>
> - Hristo
>
> --- On *Thu, 10/20/11, Kapil Arya <ka...@cc...>* wrote:
>
>
> From: Kapil Arya <ka...@cc...>
>
> Subject: Re: [Dmtcp-forum] checkpoint only certain processes
> To: "Hristo Asenov" <nj...@ya...>
> Cc: du...@ev..., dmt...@li...
> Date: Thursday, October 20, 2011, 3:39 PM
>
>
> On Thu, Oct 20, 2011 at 3:32 PM, Hristo Asenov <nj...@ya...<http://mc/compose?to=nj...@ya...>
> > wrote:
>
> well for instance process A is a routing protocol that creates a database
> from a bunch of route entries. Process B is just an interface to the outside
> world and sends everything from the TCP socket to the raw socket. I need to
> have process B under the coordinator in order to flush the data before
> checkpointing. There is no real "state" on the raw socket, only for sending
> data back and forth, and process B is not a child of process A. Im pretty
> sure process B would be able to restore the raw socket.
>
> I dont really have control of the process on the other side of the raw
> socket since that could be a router on the other side of the world. I could
> deal with checkpointing process B as long as it doesnt try to save the state
> of the raw socket.
>
>
> As far as I understand, if DMTCP can skip saving the state of the raw
> socket, there should be no problems checkpointing process B. Am I correct in
> this? If this is correct, then we can modify DMTCP to skip saving the raw
> socket. DMTCP will then disconnect the raw socket on checkpoint, assuming
> process B will later discover the disconnected state and will automatically
> recreate it.
>
> Does that sound reasonable?
>
> -- Kapil
>
>
>
> --- On *Thu, 10/20/11, du...@ev...<http://mc/compose?to=du...@ev...>
> <du...@ev... <http://mc/compose?to=du...@ev...>>* wrote:
>
>
>
>
> From: du...@ev... <http://mc/compose?to=du...@ev...> <
> du...@ev... <http://mc/compose?to=du...@ev...>>
> Subject: Re: [Dmtcp-forum] checkpoint only certain processes
> To: "Kapil Arya" <ka...@cc...<http://mc/compose?to=ka...@cc...>
> >
> Cc: "Hristo Asenov" <nj...@ya...<http://mc/compose?to=nj...@ya...>>,
> dmt...@li...<http://mc/compose?to=dmt...@li...>
> Date: Thursday, October 20, 2011, 3:00 PM
>
>
> Hi,
>
> I'd a similar issue.
> I've a process A which fork/exec a process B, I didn't want dmtcp to
> checkpoint process B.
>
> To avoid dmtcp to follow the fork/exec I get the address of the libc
> function with dlopen/dlsym, and I used them to run the process B.
> I didn't have any more issue, dmtcp only checkpoints process A.
>
>
> Regards,
> Fred
>
>
>
> > Hello Hristo,
> >
> > Currently, it is not possible to checkpoint select processes from the
> > running computation. For your particular situation, how is process B
> > created
> > and how is it related to process A? Is it possible to launch process B
> > independently so that we can disable checkpointing for it?
> >
> > Also, is process B smart enough to restore the raw socket if it sees it
> > disconnected?
> >
> > -- Kapil
> >
> > On Sun, Oct 16, 2011 at 3:57 PM, Hristo Asenov <nj...@ya...<http://mc/compose?to=nj...@ya...>>
> wrote:
> >
> >> Hello dmtc-forum,
> >>
> >> I was wondering if it was possible or whether it is a good idea to be
> >> able
> >> to checkpoint only a certain process rather than all the processes that
> >> are
> >> controlled by the coordinator. For instance I have process A that has a
> >> TCP
> >> socket set up to communicate with process B. I want to only checkpoint
> >> process A, since process B has a raw socket to the outside world which
> >> cannot be restored. I was wondering whether it was feasible to modify
> >> the
> >> code in order to implement this feature.
> >>
> >> - Hristo Asenov
> >>
> >>
> >>
> ------------------------------------------------------------------------------
> >> All the data continuously generated in your IT infrastructure contains a
> >> definitive record of customers, application performance, security
> >> threats, fraudulent activity and more. Splunk takes this data and makes
> >> sense of it. Business sense. IT sense. Common sense.
> >> http://p.sf.net/sfu/splunk-d2d-oct
> >> _______________________________________________
> >> Dmtcp-forum mailing list
> >> Dmt...@li...<http://mc/compose?to=Dmt...@li...>
> >> https://lists.sourceforge.net/lists/listinfo/dmtcp-forum
> >>
> >>
> >
> ------------------------------------------------------------------------------
> > The demand for IT networking professionals continues to grow, and the
> > demand for specialized networking skills is growing even more rapidly.
> > Take a complimentary Learning@Cisco Self-Assessment and learn
> > about Cisco certifications, training, and career opportunities.
> >
> http://p.sf.net/sfu/cisco-dev2dev_______________________________________________
> > Dmtcp-forum mailing list
> > Dmt...@li...<http://mc/compose?to=Dmt...@li...>
> > https://lists.sourceforge.net/lists/listinfo/dmtcp-forum
> >
>
>
>
>