[aspect-devel] Tracers and checkpointing

[Max Rudolph]

Rene,
I recompiled deal.ii with the bundled boost using cmake. With bundled
boost, aspect no longer encounters a segmentation fault when writing the
tracers!

Best,
Max

On Wed, Dec 9, 2015 at 12:22 PM, Rene Gassmoeller <rengas at gfz-potsdam.de>
wrote:

> Max,
> Let me put this on the mailing list, so that others are aware of what
> might be causing issues with checkpoints.
>
> I confirmed that checkpoints for a deal.II with 64bit indices enabled
> are not working at the moment (this was causing my issues), and will try
> to create a patch for deal.II.
>
> But that means your problem must be something different. Could you try
> using the bundled boost instead of 1.54 by setting
> DEAL_II_FORCE_BUNDLED_BOOST=ON when running cmake or ccmake?
>
> Best,
> Rene
>
>
> On 12/09/2015 02:03 PM, Max Rudolph wrote:
> > Rene,
> > It looks like the boost version is whatever is provided by Ubuntu.
> Version
> > is reported as 1.54.0
> >
> > #  BOOST_VERSION = 1.54.0
> > #            BOOST_INCLUDE_DIRS = /usr/include
> > #            BOOST_USER_INCLUDE_DIRS = /usr/include
> > #            BOOST_LIBRARIES =
> >
> /usr/lib/x86_64-linux-gnu/libboost_iostreams.so;/usr/lib/x86_64-linux-gnu/libboost_serialization.so;/usr/lib/x86_64-linux-gnu/libboost_system.so;/usr/lib/x86_64-linux-gnu/libboost_thread.so;pthread
> >
> > and 64 bit indices are turned off:
> > #      ( DEAL_II_WITH_64BIT_INDICES = OFF )
> >
> >
> > I can send you the whole detailed.log for deal.ii if that would help.
> >
> > Max
> >
> >
> >
> > On Wed, Dec 9, 2015 at 11:59 AM, Rene Gassmoeller <rengas at gfz-potsdam.de
> >
> > wrote:
> >
> >> Hi Max,
> >> I am currently looking into the checkpoint issue. I have no clear result
> >> yet, but could you check two things for me:
> >>
> >> Did you enable the 64bit_indices option when compiling deal.II? This
> >> might not work for checkpointing, but I am in the process of testing it
> >> more.
> >>
> >> Which version of boost do you use? The bundled one is 1.56 and should be
> >> ok, but we know of a bug in 1.58 that will affect checkpointing.
> >>
> >> I will let you know, when I have a better idea of what is going on.
> >>
> >> Rene
> >>
> >> On 12/07/2015 01:23 PM, Max Rudolph wrote:
> >>> I will try to get a clean-looking stack trace and open a ticket, but it
> >>> might take until tomorrow. I will do a couple of tests to see whether
> the
> >>> segfault occurs even without tracers.
> >>>
> >>> Max
> >>>
> >>> On Mon, Dec 7, 2015 at 11:20 AM, Rene Gassmoeller <
> rengas at gfz-potsdam.de
> >>>
> >>> wrote:
> >>>
> >>>> Hi Max,
> >>>> I encountered the same problem just yesterday and did not yet figure
> out
> >>>> the reason. During development of the tracers I did check the
> >>>> checkpointing (serial and parallel) and it seemed to work. It might
> >>>> actually be unrelated to the tracers, since deal.II 8.2 seems to work
> >>>> with the most recent aspect, but the current development version of
> >>>> deal.II does not (does it work when tracers are disabled?). Still the
> >>>> tracers are likely the reason. Please go ahead and open an issue to
> let
> >>>> others know as well. I will try to look into it during the next days.
> >>>>
> >>>> Best,
> >>>> Rene
> >>>>
> >>>> On 12/07/2015 12:01 PM, Max Rudolph wrote:
> >>>>> Rene,
> >>>>> When I tried to run a simple 2d calculation with tracers, aspect
> >> crashed
> >>>>> (seg faulted) when it reached a timestep for which checkpoint files
> >> would
> >>>>> be written. Is this a known issue? If not, I will open an 'issue' and
> >> try
> >>>>> to figure out what might be the problem. The machine that I
> encountered
> >>>> the
> >>>>> problem on is producing very ugly stack traces for some reason so I
> >> need
> >>>> to
> >>>>> fix this first...
> >>>>>
> >>>>> Cheers
> >>>>> Max
> >>>>>
> >>>>
> >>>
> >>
> >
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.geodynamics.org/pipermail/aspect-devel/attachments/20151209/7f3e6ac5/attachment.html>